From 1a782ed690e56bf81ec9073041c7559e1762855d Mon Sep 17 00:00:00 2001 From: Dave Houlton Date: Mon, 19 Jul 2021 16:26:33 -0600 Subject: SL-15595 update viewer autobuild to import tracy lib --- indra/llcommon/CMakeLists.txt | 2 ++ 1 file changed, 2 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index dd266630ea..f1b0506659 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -13,6 +13,7 @@ include(GoogleBreakpad) include(Copy3rdPartyLibs) include(ZLIB) include(URIPARSER) +include(Tracy) include_directories( ${EXPAT_INCLUDE_DIRS} @@ -21,6 +22,7 @@ include_directories( ${ZLIB_INCLUDE_DIRS} ${BREAKPAD_INCLUDE_DIRECTORIES} ${URIPARSER_INCLUDE_DIRS} + ${TRACY_INCLUDE_DIR} ) # add_executable(lltreeiterators lltreeiterators.cpp) -- cgit v1.3 From 7d5cd52498e3da2b2438ad82fe450c923541e798 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Tue, 27 Jul 2021 15:31:15 -0700 Subject: SL-15709: Add Tracy support to viewer --- doc/contributions.txt | 1 + indra/cmake/LLCommon.cmake | 2 +- indra/llcommon/CMakeLists.txt | 1 + indra/llcommon/linden_common.h | 2 + indra/llcommon/llfasttimer.h | 3 + indra/llcommon/llprofiler.h | 64 ++++++++++++++ indra/llcommon/llthread.cpp | 2 + indra/newview/llappviewer.cpp | 2 + indra/newview/llviewerdisplay.cpp | 81 ++++++++++-------- indra/newview/llvovolume.cpp | 175 ++++++++++++++++++++------------------ indra/newview/pipeline.cpp | 123 ++++++++++++++------------- 11 files changed, 277 insertions(+), 179 deletions(-) create mode 100644 indra/llcommon/llprofiler.h (limited to 'indra/llcommon') diff --git a/doc/contributions.txt b/doc/contributions.txt index 20980fdacd..a8deb9c2ae 100755 --- a/doc/contributions.txt +++ b/doc/contributions.txt @@ -270,6 +270,7 @@ Beq Janus SL-13583 SL-14766 SL-14927 + SL-15709 Beth Walcher Bezilon Kasei Biancaluce Robbiani diff --git a/indra/cmake/LLCommon.cmake b/indra/cmake/LLCommon.cmake index 8900419f9b..c1cfa51cfb 100644 --- a/indra/cmake/LLCommon.cmake +++ b/indra/cmake/LLCommon.cmake @@ -30,7 +30,7 @@ else (LINUX) ${BOOST_FIBER_LIBRARY} ${BOOST_CONTEXT_LIBRARY} ${BOOST_THREAD_LIBRARY} - ${BOOST_SYSTEM_LIBRARY} ) + ${BOOST_SYSTEM_LIBRARY}) endif (LINUX) set(LLCOMMON_LINK_SHARED OFF CACHE BOOL "Build the llcommon target as a static library.") diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index f1b0506659..28bf5d0c39 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -201,6 +201,7 @@ set(llcommon_HEADER_FILES llmortician.h llnametable.h llpointer.h + llprofiler.h llpounceable.h llpredicate.h llpreprocessor.h diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h index e5a913a6a9..45ac43910c 100644 --- a/indra/llcommon/linden_common.h +++ b/indra/llcommon/linden_common.h @@ -60,4 +60,6 @@ #include "llerror.h" #include "llfile.h" +#include "llprofiler.h" // must be before fast timer; needed due to LLThreads potentially needing access to tracy + #endif diff --git a/indra/llcommon/llfasttimer.h b/indra/llcommon/llfasttimer.h index dfc63d08a2..c7d5bb3761 100644 --- a/indra/llcommon/llfasttimer.h +++ b/indra/llcommon/llfasttimer.h @@ -38,7 +38,10 @@ #define LL_FAST_TIMER_ON 1 #define LL_FASTTIMER_USE_RDTSC 1 +// NOTE: Also see llprofiler.h +#if !defined(LL_PROFILER_CONFIGURATION) // defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER) #define LL_RECORD_BLOCK_TIME(timer_stat) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(timer_stat)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); +#endif // LL_PROFILER_CONFIGURATION namespace LLTrace { diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h new file mode 100644 index 0000000000..75954b9a79 --- /dev/null +++ b/indra/llcommon/llprofiler.h @@ -0,0 +1,64 @@ +/** + * @file llprofiler.h + * @brief Wrapper for Tracy and/or other profilers + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2021, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_PROFILER_H +#define LL_PROFILER_H + +#define LL_PROFILER_CONFIG_NONE 0 // No profiling +#define LL_PROFILER_CONFIG_FAST_TIMER 1 // Profiling on: Only Fast Timers +#define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy +#define LL_PROFILER_CONFIG_TRACY_FAST_TIMER 3 // Profiling on: Fast Timers + Tracy + +#if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #define TRACY_ENABLE 1 + #define TRACY_NO_BROADCAST 1 + #define TRACY_ONLY_LOCALHOST 1 + #define TRACY_ONLY_IPV4 1 + #include "Tracy.hpp" + #endif + + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #endif + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER + #define LL_PROFILER_FRAME_END + #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) + #define LL_RECORD_BLOCK_TIME(name) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #endif + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #timer_stat, true ) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #endif +#else + #define LL_PROFILER_FRAME_END + #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) +#endif // LL_PROFILER + +#endif // LL_PROFILER_H diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp index 6d531d842d..a8cc750437 100644 --- a/indra/llcommon/llthread.cpp +++ b/indra/llcommon/llthread.cpp @@ -135,6 +135,8 @@ void LLThread::threadRun() set_thread_name(-1, mName.c_str()); #endif + LL_PROFILER_SET_THREAD_NAME( mName.c_str() ); + // this is the first point at which we're actually running in the new thread mID = currentID(); diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 79f1cc0e8f..94f0b31ecd 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -1667,6 +1667,8 @@ bool LLAppViewer::doFrame() LL_INFOS() << "Exiting main_loop" << LL_ENDL; } + LL_PROFILER_FRAME_END + return ! LLApp::isRunning(); } diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp index 109dc93261..a590237440 100644 --- a/indra/newview/llviewerdisplay.cpp +++ b/indra/newview/llviewerdisplay.cpp @@ -1259,7 +1259,7 @@ bool setup_hud_matrices(const LLRect& screen_region) void render_ui(F32 zoom_factor, int subfield) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI); LLGLState::checkStates(); @@ -1274,7 +1274,7 @@ void render_ui(F32 zoom_factor, int subfield) if(LLSceneMonitor::getInstance()->needsUpdate()) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_SCENE_MON); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_SCENE_MON); gGL.pushMatrix(); gViewerWindow->setup2DRender(); LLSceneMonitor::getInstance()->compare(); @@ -1282,55 +1282,64 @@ void render_ui(F32 zoom_factor, int subfield) gGL.popMatrix(); } - // Finalize scene - gPipeline.renderFinalize(); + // Finalize scene + gPipeline.renderFinalize(); - LL_RECORD_BLOCK_TIME(FTM_RENDER_HUD); - render_hud_elements(); - render_hud_attachments(); - - LLGLSDefault gls_default; - LLGLSUIDefault gls_ui; { - gPipeline.disableLights(); - } + // SL-15709 + // NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_RENDER_HUD); + render_hud_elements(); + render_hud_attachments(); + + LLGLSDefault gls_default; + LLGLSUIDefault gls_ui; + { + gPipeline.disableLights(); + } - { - gGL.color4f(1,1,1,1); - if (gPipeline.hasRenderDebugFeatureMask(LLPipeline::RENDER_DEBUG_FEATURE_UI)) { - if (!gDisconnected) + gGL.color4f(1,1,1,1); + if (gPipeline.hasRenderDebugFeatureMask(LLPipeline::RENDER_DEBUG_FEATURE_UI)) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_3D); - render_ui_3d(); + if (!gDisconnected) + { + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_3D); + render_ui_3d(); + LLGLState::checkStates(); + } + else + { + render_disconnected_background(); + } + + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_2D); + render_ui_2d(); LLGLState::checkStates(); } - else + gGL.flush(); + { - render_disconnected_background(); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_DEBUG_TEXT); + gViewerWindow->setup2DRender(); + gViewerWindow->updateDebugText(); + gViewerWindow->drawDebugText(); } - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_2D); - render_ui_2d(); - LLGLState::checkStates(); + LLVertexBuffer::unbind(); } - gGL.flush(); + if (!gSnapshot) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_DEBUG_TEXT); - gViewerWindow->setup2DRender(); - gViewerWindow->updateDebugText(); - gViewerWindow->drawDebugText(); + set_current_modelview(saved_view); + gGL.popMatrix(); } - LLVertexBuffer::unbind(); - } - - if (!gSnapshot) - { - set_current_modelview(saved_view); - gGL.popMatrix(); - } + } // Tracy integration } static LLTrace::BlockTimerStatHandle FTM_SWAP("Swap"); diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index f063800587..f9b2285989 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -6045,123 +6045,130 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) if (group && group->hasState(LLSpatialGroup::MESH_DIRTY) && !group->hasState(LLSpatialGroup::GEOM_DIRTY)) { LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_VB); - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers + { + // SL-15709 -- NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers - group->mBuilt = 1.f; + group->mBuilt = 1.f; - S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ; + S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ; - const U32 MAX_BUFFER_COUNT = 4096; - LLVertexBuffer* locked_buffer[MAX_BUFFER_COUNT]; - - U32 buffer_count = 0; + const U32 MAX_BUFFER_COUNT = 4096; + LLVertexBuffer* locked_buffer[MAX_BUFFER_COUNT]; - for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) - { - LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); + U32 buffer_count = 0; - if (drawablep && !drawablep->isDead() && drawablep->isState(LLDrawable::REBUILD_ALL) && !drawablep->isState(LLDrawable::RIGGED) ) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { - LLVOVolume* vobj = drawablep->getVOVolume(); - if (debugLoggingEnabled("AnimatedObjectsLinkset")) - { - if (vobj->isAnimatedObject() && vobj->isRiggedMesh()) - { - std::string vobj_name = llformat("Vol%p", vobj); - F32 est_tris = vobj->getEstTrianglesMax(); - LL_DEBUGS("AnimatedObjectsLinkset") << vobj_name << " rebuildMesh, tris " << est_tris << LL_ENDL; - } - } - if (vobj->isNoLOD()) continue; - - vobj->preRebuild(); - - if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) - { - vobj->updateRelativeXform(true); - } + LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); - LLVolume* volume = vobj->getVolume(); - for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + if (drawablep && !drawablep->isDead() && drawablep->isState(LLDrawable::REBUILD_ALL) && !drawablep->isState(LLDrawable::RIGGED) ) { - LLFace* face = drawablep->getFace(i); - if (face) + LLVOVolume* vobj = drawablep->getVOVolume(); + if (debugLoggingEnabled("AnimatedObjectsLinkset")) { - LLVertexBuffer* buff = face->getVertexBuffer(); - if (buff) + if (vobj->isAnimatedObject() && vobj->isRiggedMesh()) { - llassert(!face->isState(LLFace::RIGGED)); + std::string vobj_name = llformat("Vol%p", vobj); + F32 est_tris = vobj->getEstTrianglesMax(); + LL_DEBUGS("AnimatedObjectsLinkset") << vobj_name << " rebuildMesh, tris " << est_tris << LL_ENDL; + } + } + if (vobj->isNoLOD()) continue; - if (!face->getGeometryVolume(*volume, face->getTEOffset(), - vobj->getRelativeXform(), vobj->getRelativeXformInvTrans(), face->getGeomIndex())) - { //something's gone wrong with the vertex buffer accounting, rebuild this group - group->dirtyGeom(); - gPipeline.markRebuild(group, TRUE); - } + vobj->preRebuild(); + if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + vobj->updateRelativeXform(true); + } - if (buff->isLocked() && buffer_count < MAX_BUFFER_COUNT) + LLVolume* volume = vobj->getVolume(); + for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + { + LLFace* face = drawablep->getFace(i); + if (face) + { + LLVertexBuffer* buff = face->getVertexBuffer(); + if (buff) { - locked_buffer[buffer_count++] = buff; + llassert(!face->isState(LLFace::RIGGED)); + + if (!face->getGeometryVolume(*volume, face->getTEOffset(), + vobj->getRelativeXform(), vobj->getRelativeXformInvTrans(), face->getGeomIndex())) + { //something's gone wrong with the vertex buffer accounting, rebuild this group + group->dirtyGeom(); + gPipeline.markRebuild(group, TRUE); + } + + + if (buff->isLocked() && buffer_count < MAX_BUFFER_COUNT) + { + locked_buffer[buffer_count++] = buff; + } } } } + + if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + vobj->updateRelativeXform(); + } + + drawablep->clearState(LLDrawable::REBUILD_ALL); } + } - if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + LL_RECORD_BLOCK_TIME(FTM_REBUILD_MESH_FLUSH); + for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter) { - vobj->updateRelativeXform(); + (*iter)->flush(); } - - drawablep->clearState(LLDrawable::REBUILD_ALL); + // don't forget alpha + if(group != NULL && + !group->mVertexBuffer.isNull() && + group->mVertexBuffer->isLocked()) + { + group->mVertexBuffer->flush(); + } } - } - - { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_MESH_FLUSH); - for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter) - { - (*iter)->flush(); - } - - // don't forget alpha - if(group != NULL && - !group->mVertexBuffer.isNull() && - group->mVertexBuffer->isLocked()) - { - group->mVertexBuffer->flush(); - } - } - //if not all buffers are unmapped - if(num_mapped_vertex_buffer != LLVertexBuffer::sMappedCount) - { - LL_WARNS() << "Not all mapped vertex buffers are unmapped!" << LL_ENDL ; - for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) + //if not all buffers are unmapped + if(num_mapped_vertex_buffer != LLVertexBuffer::sMappedCount) { - LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); - if(!drawablep) - { - continue; - } - for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + LL_WARNS() << "Not all mapped vertex buffers are unmapped!" << LL_ENDL ; + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { - LLFace* face = drawablep->getFace(i); - if (face) + LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); + if(!drawablep) + { + continue; + } + for (S32 i = 0; i < drawablep->getNumFaces(); ++i) { - LLVertexBuffer* buff = face->getVertexBuffer(); - if (buff && buff->isLocked()) + LLFace* face = drawablep->getFace(i); + if (face) { - buff->flush(); + LLVertexBuffer* buff = face->getVertexBuffer(); + if (buff && buff->isLocked()) + { + buff->flush(); + } } } } - } + } + + group->clearState(LLSpatialGroup::MESH_DIRTY | LLSpatialGroup::NEW_DRAWINFO); } - group->clearState(LLSpatialGroup::MESH_DIRTY | LLSpatialGroup::NEW_DRAWINFO); - } + } // Tracy integration // llassert(!group || !group->isState(LLSpatialGroup::NEW_DRAWINFO)); } diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index cd1b9c7c69..e7c2d4db39 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -4564,92 +4564,99 @@ void LLPipeline::renderGeomDeferred(LLCamera& camera) LLAppViewer::instance()->pingMainloopTimeout("Pipeline:RenderGeomDeferred"); LL_RECORD_BLOCK_TIME(FTM_RENDER_GEOMETRY); + { + // SL-15709 -- NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLS); - LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLS); - - LLGLEnable cull(GL_CULL_FACE); + LLGLEnable cull(GL_CULL_FACE); - for (pool_set_t::iterator iter = mPools.begin(); iter != mPools.end(); ++iter) - { - LLDrawPool *poolp = *iter; - if (hasRenderType(poolp->getType())) + for (pool_set_t::iterator iter = mPools.begin(); iter != mPools.end(); ++iter) { - poolp->prerender(); + LLDrawPool *poolp = *iter; + if (hasRenderType(poolp->getType())) + { + poolp->prerender(); + } } - } - LLGLEnable multisample(RenderFSAASamples > 0 ? GL_MULTISAMPLE_ARB : 0); + LLGLEnable multisample(RenderFSAASamples > 0 ? GL_MULTISAMPLE_ARB : 0); - LLVertexBuffer::unbind(); + LLVertexBuffer::unbind(); - LLGLState::checkStates(); - LLGLState::checkTextureChannels(); - LLGLState::checkClientArrays(); + LLGLState::checkStates(); + LLGLState::checkTextureChannels(); + LLGLState::checkClientArrays(); - U32 cur_type = 0; + U32 cur_type = 0; - gGL.setColorMask(true, true); + gGL.setColorMask(true, true); - pool_set_t::iterator iter1 = mPools.begin(); + pool_set_t::iterator iter1 = mPools.begin(); - while ( iter1 != mPools.end() ) - { - LLDrawPool *poolp = *iter1; + while ( iter1 != mPools.end() ) + { + LLDrawPool *poolp = *iter1; - cur_type = poolp->getType(); + cur_type = poolp->getType(); - pool_set_t::iterator iter2 = iter1; - if (hasRenderType(poolp->getType()) && poolp->getNumDeferredPasses() > 0) - { - LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLRENDER); + pool_set_t::iterator iter2 = iter1; + if (hasRenderType(poolp->getType()) && poolp->getNumDeferredPasses() > 0) + { + LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLRENDER); - gGLLastMatrix = NULL; - gGL.loadMatrix(gGLModelView); + gGLLastMatrix = NULL; + gGL.loadMatrix(gGLModelView); - for( S32 i = 0; i < poolp->getNumDeferredPasses(); i++ ) - { - LLVertexBuffer::unbind(); - poolp->beginDeferredPass(i); - for (iter2 = iter1; iter2 != mPools.end(); iter2++) + for( S32 i = 0; i < poolp->getNumDeferredPasses(); i++ ) { - LLDrawPool *p = *iter2; - if (p->getType() != cur_type) + LLVertexBuffer::unbind(); + poolp->beginDeferredPass(i); + for (iter2 = iter1; iter2 != mPools.end(); iter2++) { - break; + LLDrawPool *p = *iter2; + if (p->getType() != cur_type) + { + break; + } + + if ( !p->getSkipRenderFlag() ) { p->renderDeferred(i); } } - - if ( !p->getSkipRenderFlag() ) { p->renderDeferred(i); } - } - poolp->endDeferredPass(i); - LLVertexBuffer::unbind(); + poolp->endDeferredPass(i); + LLVertexBuffer::unbind(); - if (gDebugGL || gDebugPipeline) - { - LLGLState::checkStates(); + if (gDebugGL || gDebugPipeline) + { + LLGLState::checkStates(); + } } } - } - else - { - // Skip all pools of this type - for (iter2 = iter1; iter2 != mPools.end(); iter2++) + else { - LLDrawPool *p = *iter2; - if (p->getType() != cur_type) + // Skip all pools of this type + for (iter2 = iter1; iter2 != mPools.end(); iter2++) { - break; + LLDrawPool *p = *iter2; + if (p->getType() != cur_type) + { + break; + } } } + iter1 = iter2; + stop_glerror(); } - iter1 = iter2; - stop_glerror(); - } - gGLLastMatrix = NULL; - gGL.matrixMode(LLRender::MM_MODELVIEW); - gGL.loadMatrix(gGLModelView); + gGLLastMatrix = NULL; + gGL.matrixMode(LLRender::MM_MODELVIEW); + gGL.loadMatrix(gGLModelView); - gGL.setColorMask(true, false); + gGL.setColorMask(true, false); + + } // Tracy ZoneScoped } void LLPipeline::renderGeomPostDeferred(LLCamera& camera, bool do_occlusion) -- cgit v1.3 From 908ce721792c7b549227e1ee23ae8b7b064ad439 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Fri, 30 Jul 2021 09:36:38 -0700 Subject: SL-15709: Default to old fast timers --- indra/llcommon/llprofiler.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 75954b9a79..29331d35bf 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -32,6 +32,8 @@ #define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy #define LL_PROFILER_CONFIG_TRACY_FAST_TIMER 3 // Profiling on: Fast Timers + Tracy +#define LL_PROFILER_CONFIGURATION LL_PROFILER_CONFIG_FAST_TIMER + #if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define TRACY_ENABLE 1 -- cgit v1.3 From c443dc51e48d34a00b61468514dab8209ee214ea Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 26 Aug 2021 22:52:49 -0700 Subject: SL-15709: Cleanup --- indra/llcommon/llfasttimer.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llfasttimer.h b/indra/llcommon/llfasttimer.h index c7d5bb3761..9bd93d7240 100644 --- a/indra/llcommon/llfasttimer.h +++ b/indra/llcommon/llfasttimer.h @@ -39,7 +39,7 @@ #define LL_FASTTIMER_USE_RDTSC 1 // NOTE: Also see llprofiler.h -#if !defined(LL_PROFILER_CONFIGURATION) // defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER) +#if !defined(LL_PROFILER_CONFIGURATION) #define LL_RECORD_BLOCK_TIME(timer_stat) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(timer_stat)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); #endif // LL_PROFILER_CONFIGURATION -- cgit v1.3 From cdf2bdafd394a97b917cc0a71b2bc8531cce40c7 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 26 Aug 2021 23:58:45 -0700 Subject: SL-15709: Add Darwin support --- autobuild.xml | 14 +++++++------- indra/llcommon/llframetimer.cpp | 8 ++++++++ indra/llcommon/llprofiler.h | 2 ++ 3 files changed, 17 insertions(+), 7 deletions(-) (limited to 'indra/llcommon') diff --git a/autobuild.xml b/autobuild.xml index db5825bec4..ec2fa6578b 100644 --- a/autobuild.xml +++ b/autobuild.xml @@ -3088,9 +3088,9 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors archive hash - c7820c2c3df46ffef6b09de4a8772f86 + da7317e4a81609f624f84780f28b07de url - https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/84525/786813/tracy-v0.7.8.561601-darwin64-561601.tar.bz2 + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/86972/801630/tracy-v0.7.8.563351-darwin64-563351.tar.bz2 name darwin64 @@ -3100,11 +3100,11 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors archive hash - 503a6ea5c08e17a6709f11cb55b4204a + 47c696cd2966c5cc3c8ba6115dd1f886 hash_algorithm md5 url - https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/85310/791192/tracy-v0.7.8.562170-windows-562170.tar.bz2 + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/86973/801641/tracy-v0.7.8.563351-windows-563351.tar.bz2 name windows @@ -3114,11 +3114,11 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors archive hash - 3f61d596eee650763e3cb1d9f6869ed2 + b649ee6591e67d2341e886b3fc3484a7 hash_algorithm md5 url - https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/85309/791191/tracy-v0.7.8.562170-windows64-562170.tar.bz2 + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/86974/801642/tracy-v0.7.8.563351-windows64-563351.tar.bz2 name windows64 @@ -3129,7 +3129,7 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors source_type git version - v0.7.8.562170 + v0.7.8.563351 tut diff --git a/indra/llcommon/llframetimer.cpp b/indra/llcommon/llframetimer.cpp index 1e9920746b..e293a557c0 100644 --- a/indra/llcommon/llframetimer.cpp +++ b/indra/llcommon/llframetimer.cpp @@ -29,6 +29,14 @@ #include "llframetimer.h" +// On Windows we build a static lib and link with that +// On macOS we don't bother building a stand alone lib, just include the one source file we need for Tracy support +#if LL_DARWIN + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #include "TracyClient.cpp" + #endif // LL_PROFILER_CONFIGURATION +#endif // LL_DARWIN + // Static members //LLTimer LLFrameTimer::sInternalTimer; U64 LLFrameTimer::sStartTotalTime = totalTime(); diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 29331d35bf..4674985e06 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -32,7 +32,9 @@ #define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy #define LL_PROFILER_CONFIG_TRACY_FAST_TIMER 3 // Profiling on: Fast Timers + Tracy +#ifndef LL_PROFILER_CONFIGURATION #define LL_PROFILER_CONFIGURATION LL_PROFILER_CONFIG_FAST_TIMER +#endif #if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER -- cgit v1.3 From 3176136686adb58f4add432b017a7017a3f405a5 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Tue, 31 Aug 2021 21:41:03 -0700 Subject: SL-15709: Fix LLCommon not setting Tracy include directory and not linking to tracy.lib --- indra/cmake/LLCommon.cmake | 6 +++++- indra/llcommon/CMakeLists.txt | 1 + 2 files changed, 6 insertions(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/cmake/LLCommon.cmake b/indra/cmake/LLCommon.cmake index c1cfa51cfb..b6f310fe12 100644 --- a/indra/cmake/LLCommon.cmake +++ b/indra/cmake/LLCommon.cmake @@ -3,12 +3,14 @@ include(APR) include(Boost) include(EXPAT) +include(Tracy) include(ZLIB) set(LLCOMMON_INCLUDE_DIRS ${LIBS_OPEN_DIR}/llcommon ${APRUTIL_INCLUDE_DIR} ${APR_INCLUDE_DIR} + ${TRACY_INCLUDE_DIR} ) set(LLCOMMON_SYSTEM_INCLUDE_DIRS ${Boost_INCLUDE_DIRS} @@ -30,7 +32,9 @@ else (LINUX) ${BOOST_FIBER_LIBRARY} ${BOOST_CONTEXT_LIBRARY} ${BOOST_THREAD_LIBRARY} - ${BOOST_SYSTEM_LIBRARY}) + ${BOOST_SYSTEM_LIBRARY} + ${TRACY_LIBRARY} + ) endif (LINUX) set(LLCOMMON_LINK_SHARED OFF CACHE BOOL "Build the llcommon target as a static library.") diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 28bf5d0c39..066d0404ac 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -305,6 +305,7 @@ target_link_libraries( ${BOOST_SYSTEM_LIBRARY} ${GOOGLE_PERFTOOLS_LIBRARIES} ${URIPARSER_LIBRARIES} + ${TRACY_LIBRARY} ) if (DARWIN) -- cgit v1.3 From c37cc7c3a4888fdca132613d627d7ad90517332a Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Fri, 3 Sep 2021 17:20:22 -0700 Subject: SL-15709: Windows: Include Tracy source directly; don't use a library --- indra/cmake/LLCommon.cmake | 1 - indra/cmake/Tracy.cmake | 3 --- indra/llcommon/llframetimer.cpp | 11 ++++------- indra/llcommon/llprofiler.h | 6 ++++-- 4 files changed, 8 insertions(+), 13 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/cmake/LLCommon.cmake b/indra/cmake/LLCommon.cmake index b6f310fe12..34499aaa36 100644 --- a/indra/cmake/LLCommon.cmake +++ b/indra/cmake/LLCommon.cmake @@ -33,7 +33,6 @@ else (LINUX) ${BOOST_CONTEXT_LIBRARY} ${BOOST_THREAD_LIBRARY} ${BOOST_SYSTEM_LIBRARY} - ${TRACY_LIBRARY} ) endif (LINUX) diff --git a/indra/cmake/Tracy.cmake b/indra/cmake/Tracy.cmake index dfa3e83039..1b8c3db2e2 100644 --- a/indra/cmake/Tracy.cmake +++ b/indra/cmake/Tracy.cmake @@ -12,17 +12,14 @@ if (USE_TRACY) if (WINDOWS) MESSAGE(STATUS "Including Tracy for Windows: '${TRACY_INCLUDE_DIR}'") - set(TRACY_LIBRARY tracy) endif (WINDOWS) if (DARWIN) MESSAGE(STATUS "Including Tracy for Darwin: '${TRACY_INCLUDE_DIR}'") - set(TRACY_LIBRARY "") endif (DARWIN) if (LINUX) MESSAGE(STATUS "Including Tracy for Linux: '${TRACY_INCLUDE_DIR}'") - set(TRACY_LIBRARY "") endif (LINUX) else (USE_TRACY) # Tracy.cmake should not set LLCOMMON_INCLUDE_DIRS, let LLCommon.cmake do that diff --git a/indra/llcommon/llframetimer.cpp b/indra/llcommon/llframetimer.cpp index e293a557c0..c54029e8b4 100644 --- a/indra/llcommon/llframetimer.cpp +++ b/indra/llcommon/llframetimer.cpp @@ -29,13 +29,10 @@ #include "llframetimer.h" -// On Windows we build a static lib and link with that -// On macOS we don't bother building a stand alone lib, just include the one source file we need for Tracy support -#if LL_DARWIN - #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER - #include "TracyClient.cpp" - #endif // LL_PROFILER_CONFIGURATION -#endif // LL_DARWIN +// We don't bother building a stand alone lib; we just need to include the one source file for Tracy support +#if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #include "TracyClient.cpp" +#endif // LL_PROFILER_CONFIGURATION // Static members //LLTimer LLFrameTimer::sInternalTimer; diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 4674985e06..062c9360dd 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -39,8 +39,10 @@ #if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define TRACY_ENABLE 1 - #define TRACY_NO_BROADCAST 1 - #define TRACY_ONLY_LOCALHOST 1 +// Normally these would be enabled but we want to be able to build any viewer with Tracy enabled and run the Tracy server on another machine +// They must be undefined in order to work across multiple machines +// #define TRACY_NO_BROADCAST 1 +// #define TRACY_ONLY_LOCALHOST 1 #define TRACY_ONLY_IPV4 1 #include "Tracy.hpp" #endif -- cgit v1.3 From 2e88a3266529714efcf8ae092819a25393540c8a Mon Sep 17 00:00:00 2001 From: Dave Houlton Date: Mon, 19 Jul 2021 16:26:33 -0600 Subject: SL-15595 update viewer autobuild to import tracy lib --- autobuild.xml | 64 +++++++++++++++++++++++++++++++++++++++++++ indra/cmake/Tracy.cmake | 17 ++++++++++++ indra/llcommon/CMakeLists.txt | 2 ++ indra/newview/CMakeLists.txt | 3 ++ 4 files changed, 86 insertions(+) create mode 100644 indra/cmake/Tracy.cmake (limited to 'indra/llcommon') diff --git a/autobuild.xml b/autobuild.xml index 4768bd25c6..4eaa16e8d9 100644 --- a/autobuild.xml +++ b/autobuild.xml @@ -3067,6 +3067,70 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors version 4.10.0000.32327.5fc3fe7c.539691 + tracy + + canonical_repo + https://bitbucket.org/lindenlab/3p-tracy + copyright + Copyright (c) 2017-2021, Bartosz Taudul (wolf@nereid.pl) + description + Tracy Profiler Library + license + bsd + license_file + LICENSES/tracy_license.txt + name + tracy + platforms + + darwin64 + + archive + + hash + c7820c2c3df46ffef6b09de4a8772f86 + url + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/84525/786813/tracy-v0.7.8.561601-darwin64-561601.tar.bz2 + + name + darwin64 + + windows + + archive + + hash + 503a6ea5c08e17a6709f11cb55b4204a + hash_algorithm + md5 + url + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/85310/791192/tracy-v0.7.8.562170-windows-562170.tar.bz2 + + name + windows + + windows64 + + archive + + hash + 3f61d596eee650763e3cb1d9f6869ed2 + hash_algorithm + md5 + url + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/85309/791191/tracy-v0.7.8.562170-windows64-562170.tar.bz2 + + name + windows64 + + + source + https://bitbucket.org/lindenlab/3p-tracy + source_type + git + version + v0.7.8.562170 + tut copyright diff --git a/indra/cmake/Tracy.cmake b/indra/cmake/Tracy.cmake new file mode 100644 index 0000000000..bf09bccd4b --- /dev/null +++ b/indra/cmake/Tracy.cmake @@ -0,0 +1,17 @@ +# -*- cmake -*- +include(Prebuilt) + +set(TRACY ON CACHE BOOL "Use Tracy profiler.") + +if (TRACY) + set(TRACY_INCLUDE_DIR ${LIBS_PREBUILT_DIR}/include/tracy) + if (WINDOWS) + use_prebuilt_binary(tracy) + set(TRACY_LIBRARY tracy) + else (WINDOWS) + set(TRACY_LIBRARY "") + endif (WINDOWS) +else (TRACY) + set(TRACY_LIBRARY "") +endif (TRACY) + diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index dd266630ea..f1b0506659 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -13,6 +13,7 @@ include(GoogleBreakpad) include(Copy3rdPartyLibs) include(ZLIB) include(URIPARSER) +include(Tracy) include_directories( ${EXPAT_INCLUDE_DIRS} @@ -21,6 +22,7 @@ include_directories( ${ZLIB_INCLUDE_DIRS} ${BREAKPAD_INCLUDE_DIRECTORIES} ${URIPARSER_INCLUDE_DIRS} + ${TRACY_INCLUDE_DIR} ) # add_executable(lltreeiterators lltreeiterators.cpp) diff --git a/indra/newview/CMakeLists.txt b/indra/newview/CMakeLists.txt index 2995a006ac..fbe75af712 100644 --- a/indra/newview/CMakeLists.txt +++ b/indra/newview/CMakeLists.txt @@ -47,6 +47,7 @@ include(OpenGL) include(OpenSSL) include(PNG) include(TemplateCheck) +include(Tracy) include(UI) include(UnixInstall) include(ViewerMiscLibs) @@ -92,6 +93,7 @@ include_directories( ${LIBS_PREBUILT_DIR}/include/collada/1.4 ${LLAPPEARANCE_INCLUDE_DIRS} ${CMAKE_CURRENT_SOURCE_DIR} + ${TRACY_INCLUDE_DIR} ) include_directories(SYSTEM @@ -2066,6 +2068,7 @@ target_link_libraries(${VIEWER_BINARY_NAME} ${LLPHYSICS_LIBRARIES} ${LLPHYSICSEXTENSIONS_LIBRARIES} ${LLAPPEARANCE_LIBRARIES} + ${TRACY_LIBRARY} ) if (USE_BUGSPLAT) -- cgit v1.3 From df5127136f9f520aa31c06ddb37ac79b6a8dc16d Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Tue, 27 Jul 2021 15:31:15 -0700 Subject: SL-15709: Add Tracy support to viewer --- doc/contributions.txt | 1 + indra/cmake/LLCommon.cmake | 2 +- indra/llcommon/CMakeLists.txt | 1 + indra/llcommon/linden_common.h | 2 + indra/llcommon/llfasttimer.h | 3 + indra/llcommon/llprofiler.h | 64 ++++++++++++++ indra/llcommon/llthread.cpp | 2 + indra/newview/llappviewer.cpp | 2 + indra/newview/llviewerdisplay.cpp | 81 ++++++++++-------- indra/newview/llvovolume.cpp | 175 ++++++++++++++++++++------------------ indra/newview/pipeline.cpp | 123 ++++++++++++++------------- 11 files changed, 277 insertions(+), 179 deletions(-) create mode 100644 indra/llcommon/llprofiler.h (limited to 'indra/llcommon') diff --git a/doc/contributions.txt b/doc/contributions.txt index 20980fdacd..a8deb9c2ae 100755 --- a/doc/contributions.txt +++ b/doc/contributions.txt @@ -270,6 +270,7 @@ Beq Janus SL-13583 SL-14766 SL-14927 + SL-15709 Beth Walcher Bezilon Kasei Biancaluce Robbiani diff --git a/indra/cmake/LLCommon.cmake b/indra/cmake/LLCommon.cmake index 8900419f9b..c1cfa51cfb 100644 --- a/indra/cmake/LLCommon.cmake +++ b/indra/cmake/LLCommon.cmake @@ -30,7 +30,7 @@ else (LINUX) ${BOOST_FIBER_LIBRARY} ${BOOST_CONTEXT_LIBRARY} ${BOOST_THREAD_LIBRARY} - ${BOOST_SYSTEM_LIBRARY} ) + ${BOOST_SYSTEM_LIBRARY}) endif (LINUX) set(LLCOMMON_LINK_SHARED OFF CACHE BOOL "Build the llcommon target as a static library.") diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index f1b0506659..28bf5d0c39 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -201,6 +201,7 @@ set(llcommon_HEADER_FILES llmortician.h llnametable.h llpointer.h + llprofiler.h llpounceable.h llpredicate.h llpreprocessor.h diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h index e5a913a6a9..45ac43910c 100644 --- a/indra/llcommon/linden_common.h +++ b/indra/llcommon/linden_common.h @@ -60,4 +60,6 @@ #include "llerror.h" #include "llfile.h" +#include "llprofiler.h" // must be before fast timer; needed due to LLThreads potentially needing access to tracy + #endif diff --git a/indra/llcommon/llfasttimer.h b/indra/llcommon/llfasttimer.h index dfc63d08a2..c7d5bb3761 100644 --- a/indra/llcommon/llfasttimer.h +++ b/indra/llcommon/llfasttimer.h @@ -38,7 +38,10 @@ #define LL_FAST_TIMER_ON 1 #define LL_FASTTIMER_USE_RDTSC 1 +// NOTE: Also see llprofiler.h +#if !defined(LL_PROFILER_CONFIGURATION) // defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER) #define LL_RECORD_BLOCK_TIME(timer_stat) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(timer_stat)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); +#endif // LL_PROFILER_CONFIGURATION namespace LLTrace { diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h new file mode 100644 index 0000000000..75954b9a79 --- /dev/null +++ b/indra/llcommon/llprofiler.h @@ -0,0 +1,64 @@ +/** + * @file llprofiler.h + * @brief Wrapper for Tracy and/or other profilers + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2021, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_PROFILER_H +#define LL_PROFILER_H + +#define LL_PROFILER_CONFIG_NONE 0 // No profiling +#define LL_PROFILER_CONFIG_FAST_TIMER 1 // Profiling on: Only Fast Timers +#define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy +#define LL_PROFILER_CONFIG_TRACY_FAST_TIMER 3 // Profiling on: Fast Timers + Tracy + +#if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #define TRACY_ENABLE 1 + #define TRACY_NO_BROADCAST 1 + #define TRACY_ONLY_LOCALHOST 1 + #define TRACY_ONLY_IPV4 1 + #include "Tracy.hpp" + #endif + + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #endif + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER + #define LL_PROFILER_FRAME_END + #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) + #define LL_RECORD_BLOCK_TIME(name) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #endif + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #timer_stat, true ) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #endif +#else + #define LL_PROFILER_FRAME_END + #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) +#endif // LL_PROFILER + +#endif // LL_PROFILER_H diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp index 6d531d842d..a8cc750437 100644 --- a/indra/llcommon/llthread.cpp +++ b/indra/llcommon/llthread.cpp @@ -135,6 +135,8 @@ void LLThread::threadRun() set_thread_name(-1, mName.c_str()); #endif + LL_PROFILER_SET_THREAD_NAME( mName.c_str() ); + // this is the first point at which we're actually running in the new thread mID = currentID(); diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 79f1cc0e8f..94f0b31ecd 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -1667,6 +1667,8 @@ bool LLAppViewer::doFrame() LL_INFOS() << "Exiting main_loop" << LL_ENDL; } + LL_PROFILER_FRAME_END + return ! LLApp::isRunning(); } diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp index 109dc93261..a590237440 100644 --- a/indra/newview/llviewerdisplay.cpp +++ b/indra/newview/llviewerdisplay.cpp @@ -1259,7 +1259,7 @@ bool setup_hud_matrices(const LLRect& screen_region) void render_ui(F32 zoom_factor, int subfield) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI); LLGLState::checkStates(); @@ -1274,7 +1274,7 @@ void render_ui(F32 zoom_factor, int subfield) if(LLSceneMonitor::getInstance()->needsUpdate()) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_SCENE_MON); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_SCENE_MON); gGL.pushMatrix(); gViewerWindow->setup2DRender(); LLSceneMonitor::getInstance()->compare(); @@ -1282,55 +1282,64 @@ void render_ui(F32 zoom_factor, int subfield) gGL.popMatrix(); } - // Finalize scene - gPipeline.renderFinalize(); + // Finalize scene + gPipeline.renderFinalize(); - LL_RECORD_BLOCK_TIME(FTM_RENDER_HUD); - render_hud_elements(); - render_hud_attachments(); - - LLGLSDefault gls_default; - LLGLSUIDefault gls_ui; { - gPipeline.disableLights(); - } + // SL-15709 + // NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_RENDER_HUD); + render_hud_elements(); + render_hud_attachments(); + + LLGLSDefault gls_default; + LLGLSUIDefault gls_ui; + { + gPipeline.disableLights(); + } - { - gGL.color4f(1,1,1,1); - if (gPipeline.hasRenderDebugFeatureMask(LLPipeline::RENDER_DEBUG_FEATURE_UI)) { - if (!gDisconnected) + gGL.color4f(1,1,1,1); + if (gPipeline.hasRenderDebugFeatureMask(LLPipeline::RENDER_DEBUG_FEATURE_UI)) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_3D); - render_ui_3d(); + if (!gDisconnected) + { + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_3D); + render_ui_3d(); + LLGLState::checkStates(); + } + else + { + render_disconnected_background(); + } + + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_2D); + render_ui_2d(); LLGLState::checkStates(); } - else + gGL.flush(); + { - render_disconnected_background(); + LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_DEBUG_TEXT); + gViewerWindow->setup2DRender(); + gViewerWindow->updateDebugText(); + gViewerWindow->drawDebugText(); } - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_2D); - render_ui_2d(); - LLGLState::checkStates(); + LLVertexBuffer::unbind(); } - gGL.flush(); + if (!gSnapshot) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_DEBUG_TEXT); - gViewerWindow->setup2DRender(); - gViewerWindow->updateDebugText(); - gViewerWindow->drawDebugText(); + set_current_modelview(saved_view); + gGL.popMatrix(); } - LLVertexBuffer::unbind(); - } - - if (!gSnapshot) - { - set_current_modelview(saved_view); - gGL.popMatrix(); - } + } // Tracy integration } static LLTrace::BlockTimerStatHandle FTM_SWAP("Swap"); diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index f063800587..f9b2285989 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -6045,123 +6045,130 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) if (group && group->hasState(LLSpatialGroup::MESH_DIRTY) && !group->hasState(LLSpatialGroup::GEOM_DIRTY)) { LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_VB); - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers + { + // SL-15709 -- NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers - group->mBuilt = 1.f; + group->mBuilt = 1.f; - S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ; + S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ; - const U32 MAX_BUFFER_COUNT = 4096; - LLVertexBuffer* locked_buffer[MAX_BUFFER_COUNT]; - - U32 buffer_count = 0; + const U32 MAX_BUFFER_COUNT = 4096; + LLVertexBuffer* locked_buffer[MAX_BUFFER_COUNT]; - for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) - { - LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); + U32 buffer_count = 0; - if (drawablep && !drawablep->isDead() && drawablep->isState(LLDrawable::REBUILD_ALL) && !drawablep->isState(LLDrawable::RIGGED) ) + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { - LLVOVolume* vobj = drawablep->getVOVolume(); - if (debugLoggingEnabled("AnimatedObjectsLinkset")) - { - if (vobj->isAnimatedObject() && vobj->isRiggedMesh()) - { - std::string vobj_name = llformat("Vol%p", vobj); - F32 est_tris = vobj->getEstTrianglesMax(); - LL_DEBUGS("AnimatedObjectsLinkset") << vobj_name << " rebuildMesh, tris " << est_tris << LL_ENDL; - } - } - if (vobj->isNoLOD()) continue; - - vobj->preRebuild(); - - if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) - { - vobj->updateRelativeXform(true); - } + LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); - LLVolume* volume = vobj->getVolume(); - for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + if (drawablep && !drawablep->isDead() && drawablep->isState(LLDrawable::REBUILD_ALL) && !drawablep->isState(LLDrawable::RIGGED) ) { - LLFace* face = drawablep->getFace(i); - if (face) + LLVOVolume* vobj = drawablep->getVOVolume(); + if (debugLoggingEnabled("AnimatedObjectsLinkset")) { - LLVertexBuffer* buff = face->getVertexBuffer(); - if (buff) + if (vobj->isAnimatedObject() && vobj->isRiggedMesh()) { - llassert(!face->isState(LLFace::RIGGED)); + std::string vobj_name = llformat("Vol%p", vobj); + F32 est_tris = vobj->getEstTrianglesMax(); + LL_DEBUGS("AnimatedObjectsLinkset") << vobj_name << " rebuildMesh, tris " << est_tris << LL_ENDL; + } + } + if (vobj->isNoLOD()) continue; - if (!face->getGeometryVolume(*volume, face->getTEOffset(), - vobj->getRelativeXform(), vobj->getRelativeXformInvTrans(), face->getGeomIndex())) - { //something's gone wrong with the vertex buffer accounting, rebuild this group - group->dirtyGeom(); - gPipeline.markRebuild(group, TRUE); - } + vobj->preRebuild(); + if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + vobj->updateRelativeXform(true); + } - if (buff->isLocked() && buffer_count < MAX_BUFFER_COUNT) + LLVolume* volume = vobj->getVolume(); + for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + { + LLFace* face = drawablep->getFace(i); + if (face) + { + LLVertexBuffer* buff = face->getVertexBuffer(); + if (buff) { - locked_buffer[buffer_count++] = buff; + llassert(!face->isState(LLFace::RIGGED)); + + if (!face->getGeometryVolume(*volume, face->getTEOffset(), + vobj->getRelativeXform(), vobj->getRelativeXformInvTrans(), face->getGeomIndex())) + { //something's gone wrong with the vertex buffer accounting, rebuild this group + group->dirtyGeom(); + gPipeline.markRebuild(group, TRUE); + } + + + if (buff->isLocked() && buffer_count < MAX_BUFFER_COUNT) + { + locked_buffer[buffer_count++] = buff; + } } } } + + if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + vobj->updateRelativeXform(); + } + + drawablep->clearState(LLDrawable::REBUILD_ALL); } + } - if (drawablep->isState(LLDrawable::ANIMATED_CHILD)) + { + LL_RECORD_BLOCK_TIME(FTM_REBUILD_MESH_FLUSH); + for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter) { - vobj->updateRelativeXform(); + (*iter)->flush(); } - - drawablep->clearState(LLDrawable::REBUILD_ALL); + // don't forget alpha + if(group != NULL && + !group->mVertexBuffer.isNull() && + group->mVertexBuffer->isLocked()) + { + group->mVertexBuffer->flush(); + } } - } - - { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_MESH_FLUSH); - for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter) - { - (*iter)->flush(); - } - - // don't forget alpha - if(group != NULL && - !group->mVertexBuffer.isNull() && - group->mVertexBuffer->isLocked()) - { - group->mVertexBuffer->flush(); - } - } - //if not all buffers are unmapped - if(num_mapped_vertex_buffer != LLVertexBuffer::sMappedCount) - { - LL_WARNS() << "Not all mapped vertex buffers are unmapped!" << LL_ENDL ; - for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) + //if not all buffers are unmapped + if(num_mapped_vertex_buffer != LLVertexBuffer::sMappedCount) { - LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); - if(!drawablep) - { - continue; - } - for (S32 i = 0; i < drawablep->getNumFaces(); ++i) + LL_WARNS() << "Not all mapped vertex buffers are unmapped!" << LL_ENDL ; + for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); drawable_iter != group->getDataEnd(); ++drawable_iter) { - LLFace* face = drawablep->getFace(i); - if (face) + LLDrawable* drawablep = (LLDrawable*)(*drawable_iter)->getDrawable(); + if(!drawablep) + { + continue; + } + for (S32 i = 0; i < drawablep->getNumFaces(); ++i) { - LLVertexBuffer* buff = face->getVertexBuffer(); - if (buff && buff->isLocked()) + LLFace* face = drawablep->getFace(i); + if (face) { - buff->flush(); + LLVertexBuffer* buff = face->getVertexBuffer(); + if (buff && buff->isLocked()) + { + buff->flush(); + } } } } - } + } + + group->clearState(LLSpatialGroup::MESH_DIRTY | LLSpatialGroup::NEW_DRAWINFO); } - group->clearState(LLSpatialGroup::MESH_DIRTY | LLSpatialGroup::NEW_DRAWINFO); - } + } // Tracy integration // llassert(!group || !group->isState(LLSpatialGroup::NEW_DRAWINFO)); } diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index cd1b9c7c69..e7c2d4db39 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -4564,92 +4564,99 @@ void LLPipeline::renderGeomDeferred(LLCamera& camera) LLAppViewer::instance()->pingMainloopTimeout("Pipeline:RenderGeomDeferred"); LL_RECORD_BLOCK_TIME(FTM_RENDER_GEOMETRY); + { + // SL-15709 -- NOTE: Tracy only allows one ZoneScoped per function. + // Solutions are: + // 1. Use a new scope + // 2. Use named zones + // 3. Use transient zones + LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLS); - LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLS); - - LLGLEnable cull(GL_CULL_FACE); + LLGLEnable cull(GL_CULL_FACE); - for (pool_set_t::iterator iter = mPools.begin(); iter != mPools.end(); ++iter) - { - LLDrawPool *poolp = *iter; - if (hasRenderType(poolp->getType())) + for (pool_set_t::iterator iter = mPools.begin(); iter != mPools.end(); ++iter) { - poolp->prerender(); + LLDrawPool *poolp = *iter; + if (hasRenderType(poolp->getType())) + { + poolp->prerender(); + } } - } - LLGLEnable multisample(RenderFSAASamples > 0 ? GL_MULTISAMPLE_ARB : 0); + LLGLEnable multisample(RenderFSAASamples > 0 ? GL_MULTISAMPLE_ARB : 0); - LLVertexBuffer::unbind(); + LLVertexBuffer::unbind(); - LLGLState::checkStates(); - LLGLState::checkTextureChannels(); - LLGLState::checkClientArrays(); + LLGLState::checkStates(); + LLGLState::checkTextureChannels(); + LLGLState::checkClientArrays(); - U32 cur_type = 0; + U32 cur_type = 0; - gGL.setColorMask(true, true); + gGL.setColorMask(true, true); - pool_set_t::iterator iter1 = mPools.begin(); + pool_set_t::iterator iter1 = mPools.begin(); - while ( iter1 != mPools.end() ) - { - LLDrawPool *poolp = *iter1; + while ( iter1 != mPools.end() ) + { + LLDrawPool *poolp = *iter1; - cur_type = poolp->getType(); + cur_type = poolp->getType(); - pool_set_t::iterator iter2 = iter1; - if (hasRenderType(poolp->getType()) && poolp->getNumDeferredPasses() > 0) - { - LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLRENDER); + pool_set_t::iterator iter2 = iter1; + if (hasRenderType(poolp->getType()) && poolp->getNumDeferredPasses() > 0) + { + LL_RECORD_BLOCK_TIME(FTM_DEFERRED_POOLRENDER); - gGLLastMatrix = NULL; - gGL.loadMatrix(gGLModelView); + gGLLastMatrix = NULL; + gGL.loadMatrix(gGLModelView); - for( S32 i = 0; i < poolp->getNumDeferredPasses(); i++ ) - { - LLVertexBuffer::unbind(); - poolp->beginDeferredPass(i); - for (iter2 = iter1; iter2 != mPools.end(); iter2++) + for( S32 i = 0; i < poolp->getNumDeferredPasses(); i++ ) { - LLDrawPool *p = *iter2; - if (p->getType() != cur_type) + LLVertexBuffer::unbind(); + poolp->beginDeferredPass(i); + for (iter2 = iter1; iter2 != mPools.end(); iter2++) { - break; + LLDrawPool *p = *iter2; + if (p->getType() != cur_type) + { + break; + } + + if ( !p->getSkipRenderFlag() ) { p->renderDeferred(i); } } - - if ( !p->getSkipRenderFlag() ) { p->renderDeferred(i); } - } - poolp->endDeferredPass(i); - LLVertexBuffer::unbind(); + poolp->endDeferredPass(i); + LLVertexBuffer::unbind(); - if (gDebugGL || gDebugPipeline) - { - LLGLState::checkStates(); + if (gDebugGL || gDebugPipeline) + { + LLGLState::checkStates(); + } } } - } - else - { - // Skip all pools of this type - for (iter2 = iter1; iter2 != mPools.end(); iter2++) + else { - LLDrawPool *p = *iter2; - if (p->getType() != cur_type) + // Skip all pools of this type + for (iter2 = iter1; iter2 != mPools.end(); iter2++) { - break; + LLDrawPool *p = *iter2; + if (p->getType() != cur_type) + { + break; + } } } + iter1 = iter2; + stop_glerror(); } - iter1 = iter2; - stop_glerror(); - } - gGLLastMatrix = NULL; - gGL.matrixMode(LLRender::MM_MODELVIEW); - gGL.loadMatrix(gGLModelView); + gGLLastMatrix = NULL; + gGL.matrixMode(LLRender::MM_MODELVIEW); + gGL.loadMatrix(gGLModelView); - gGL.setColorMask(true, false); + gGL.setColorMask(true, false); + + } // Tracy ZoneScoped } void LLPipeline::renderGeomPostDeferred(LLCamera& camera, bool do_occlusion) -- cgit v1.3 From 6d74ae649c6a05403f7ddc01a41d8def694e00e1 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Fri, 30 Jul 2021 09:36:38 -0700 Subject: SL-15709: Default to old fast timers --- indra/llcommon/llprofiler.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 75954b9a79..29331d35bf 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -32,6 +32,8 @@ #define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy #define LL_PROFILER_CONFIG_TRACY_FAST_TIMER 3 // Profiling on: Fast Timers + Tracy +#define LL_PROFILER_CONFIGURATION LL_PROFILER_CONFIG_FAST_TIMER + #if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define TRACY_ENABLE 1 -- cgit v1.3 From 92e53622ea33cf1558d7079e9341038dd242c3a5 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 26 Aug 2021 22:52:49 -0700 Subject: SL-15709: Cleanup --- indra/llcommon/llfasttimer.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llfasttimer.h b/indra/llcommon/llfasttimer.h index c7d5bb3761..9bd93d7240 100644 --- a/indra/llcommon/llfasttimer.h +++ b/indra/llcommon/llfasttimer.h @@ -39,7 +39,7 @@ #define LL_FASTTIMER_USE_RDTSC 1 // NOTE: Also see llprofiler.h -#if !defined(LL_PROFILER_CONFIGURATION) // defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER) +#if !defined(LL_PROFILER_CONFIGURATION) #define LL_RECORD_BLOCK_TIME(timer_stat) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(timer_stat)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); #endif // LL_PROFILER_CONFIGURATION -- cgit v1.3 From 13ff2cba5365474fe53809968f66cc2fa20be4cc Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 26 Aug 2021 23:58:45 -0700 Subject: SL-15709: Add Darwin support --- autobuild.xml | 14 +++++++------- indra/llcommon/llframetimer.cpp | 8 ++++++++ indra/llcommon/llprofiler.h | 2 ++ 3 files changed, 17 insertions(+), 7 deletions(-) (limited to 'indra/llcommon') diff --git a/autobuild.xml b/autobuild.xml index 4eaa16e8d9..44781a358a 100644 --- a/autobuild.xml +++ b/autobuild.xml @@ -3088,9 +3088,9 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors archive hash - c7820c2c3df46ffef6b09de4a8772f86 + da7317e4a81609f624f84780f28b07de url - https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/84525/786813/tracy-v0.7.8.561601-darwin64-561601.tar.bz2 + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/86972/801630/tracy-v0.7.8.563351-darwin64-563351.tar.bz2 name darwin64 @@ -3100,11 +3100,11 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors archive hash - 503a6ea5c08e17a6709f11cb55b4204a + 47c696cd2966c5cc3c8ba6115dd1f886 hash_algorithm md5 url - https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/85310/791192/tracy-v0.7.8.562170-windows-562170.tar.bz2 + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/86973/801641/tracy-v0.7.8.563351-windows-563351.tar.bz2 name windows @@ -3114,11 +3114,11 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors archive hash - 3f61d596eee650763e3cb1d9f6869ed2 + b649ee6591e67d2341e886b3fc3484a7 hash_algorithm md5 url - https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/85309/791191/tracy-v0.7.8.562170-windows64-562170.tar.bz2 + https://automated-builds-secondlife-com.s3.amazonaws.com/ct2/86974/801642/tracy-v0.7.8.563351-windows64-563351.tar.bz2 name windows64 @@ -3129,7 +3129,7 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors source_type git version - v0.7.8.562170 + v0.7.8.563351 tut diff --git a/indra/llcommon/llframetimer.cpp b/indra/llcommon/llframetimer.cpp index 1e9920746b..e293a557c0 100644 --- a/indra/llcommon/llframetimer.cpp +++ b/indra/llcommon/llframetimer.cpp @@ -29,6 +29,14 @@ #include "llframetimer.h" +// On Windows we build a static lib and link with that +// On macOS we don't bother building a stand alone lib, just include the one source file we need for Tracy support +#if LL_DARWIN + #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #include "TracyClient.cpp" + #endif // LL_PROFILER_CONFIGURATION +#endif // LL_DARWIN + // Static members //LLTimer LLFrameTimer::sInternalTimer; U64 LLFrameTimer::sStartTotalTime = totalTime(); diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 29331d35bf..4674985e06 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -32,7 +32,9 @@ #define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy #define LL_PROFILER_CONFIG_TRACY_FAST_TIMER 3 // Profiling on: Fast Timers + Tracy +#ifndef LL_PROFILER_CONFIGURATION #define LL_PROFILER_CONFIGURATION LL_PROFILER_CONFIG_FAST_TIMER +#endif #if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER -- cgit v1.3 From 6d2cad965c63cd352f617537b73506790fd8f4a4 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Tue, 31 Aug 2021 21:41:03 -0700 Subject: SL-15709: Fix LLCommon not setting Tracy include directory and not linking to tracy.lib --- indra/cmake/LLCommon.cmake | 6 +++++- indra/llcommon/CMakeLists.txt | 1 + 2 files changed, 6 insertions(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/cmake/LLCommon.cmake b/indra/cmake/LLCommon.cmake index c1cfa51cfb..b6f310fe12 100644 --- a/indra/cmake/LLCommon.cmake +++ b/indra/cmake/LLCommon.cmake @@ -3,12 +3,14 @@ include(APR) include(Boost) include(EXPAT) +include(Tracy) include(ZLIB) set(LLCOMMON_INCLUDE_DIRS ${LIBS_OPEN_DIR}/llcommon ${APRUTIL_INCLUDE_DIR} ${APR_INCLUDE_DIR} + ${TRACY_INCLUDE_DIR} ) set(LLCOMMON_SYSTEM_INCLUDE_DIRS ${Boost_INCLUDE_DIRS} @@ -30,7 +32,9 @@ else (LINUX) ${BOOST_FIBER_LIBRARY} ${BOOST_CONTEXT_LIBRARY} ${BOOST_THREAD_LIBRARY} - ${BOOST_SYSTEM_LIBRARY}) + ${BOOST_SYSTEM_LIBRARY} + ${TRACY_LIBRARY} + ) endif (LINUX) set(LLCOMMON_LINK_SHARED OFF CACHE BOOL "Build the llcommon target as a static library.") diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 28bf5d0c39..066d0404ac 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -305,6 +305,7 @@ target_link_libraries( ${BOOST_SYSTEM_LIBRARY} ${GOOGLE_PERFTOOLS_LIBRARIES} ${URIPARSER_LIBRARIES} + ${TRACY_LIBRARY} ) if (DARWIN) -- cgit v1.3 From f2d4f83931f77282d6cdeba582def46b51c22b89 Mon Sep 17 00:00:00 2001 From: Dave Houlton Date: Wed, 15 Sep 2021 10:11:25 -0600 Subject: SL-15962 Add hooks for tracy memory profiling --- indra/llcommon/linden_common.h | 8 ++++++++ indra/llcommon/llcommon.cpp | 42 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 50 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h index 45ac43910c..b2c5be6b76 100644 --- a/indra/llcommon/linden_common.h +++ b/indra/llcommon/linden_common.h @@ -27,6 +27,14 @@ #ifndef LL_LINDEN_COMMON_H #define LL_LINDEN_COMMON_H +#include "llprofiler.h" +#if (TRACY_ENABLE) // hooks for memory profiling +void *tracy_aligned_malloc(size_t size, size_t alignment); +void tracy_aligned_free(void *memblock); +#define _aligned_malloc(X, Y) tracy_aligned_malloc((X), (Y)) +#define _aligned_free(X) tracy_aligned_free((X)) +#endif + // *NOTE: Please keep includes here to a minimum! // // Files included here are included in every library .cpp file and diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index 96be913d17..da61e7539a 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -33,6 +33,48 @@ #include "lltracethreadrecorder.h" #include "llcleanup.h" +#if (TRACY_ENABLE) +// Override new/delet for tracy memory profiling +void *operator new(size_t size) +{ + auto ptr = (malloc) (size); + if (!ptr) + { + throw std::bad_alloc(); + return nullptr; + } + TracyAlloc(ptr, size); + return ptr; +} + +void operator delete(void *ptr) noexcept +{ + TracyFree(ptr); + (free)(ptr); +} + +// C-style malloc/free can't be so easily overridden, so we define tracy versions and use +// a pre-processor #define in linden_common.h to redirect to them. The parens around the native +// functions below prevents recursive substitution by the preprocessor. +// +// Unaligned mallocs are rare in LL code but hooking them causes problems in 3p lib code (looking at +// you, Havok), so we'll only capture the aligned version. + +void *tracy_aligned_malloc(size_t size, size_t alignment) +{ + auto ptr = (_aligned_malloc) (size, alignment); + if (ptr) TracyAlloc(ptr, size); + return ptr; +} + +void tracy_aligned_free(void *memblock) +{ + TracyFree(memblock); + (_aligned_free)(memblock); +} + +#endif + //static BOOL LLCommon::sAprInitialized = FALSE; -- cgit v1.3 From 7fe2856516d9e0de0fda6ff389ad3cc977b2d309 Mon Sep 17 00:00:00 2001 From: Runitai Linden Date: Mon, 13 Sep 2021 12:41:57 -0500 Subject: SL-15975 Add Tracy-only profile macros that are no-ops when Tracy is disabled. --- indra/llcommon/llprofiler.h | 6 ++++++ indra/newview/lldrawpoolavatar.cpp | 4 ++++ 2 files changed, 10 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 062c9360dd..62e649913b 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -51,16 +51,22 @@ #define LL_PROFILER_FRAME_END FrameMark #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, name, true ); + #define LL_PROFILE_ZONE_SCOPED ZoneScoped #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER #define LL_PROFILER_FRAME_END #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) #define LL_RECORD_BLOCK_TIME(name) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_PROFILE_ZONE_NAMED(name) // LL_PROFILE_ZONE_NAMED is a no-op when Tracy is disabled + #define LL_PROFILE_ZONE_SCOPED // LL_PROFILE_ZONE_SCOPED is a no-op when Tracy is disabled #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define LL_PROFILER_FRAME_END FrameMark #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #timer_stat, true ) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_SCOPED ZoneScoped #endif #else #define LL_PROFILER_FRAME_END diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index 687b13d2c8..c04142ab47 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -1840,6 +1840,8 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( LLVolume* volume, LLVolumeFace& vol_face) { + LL_PROFILE_ZONE_SCOPED; + LLVector4a* weights = vol_face.mWeights; if (!weights) { @@ -2352,8 +2354,10 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar) //update rigged vertex buffers for (U32 type = 0; type < NUM_RIGGED_PASSES; ++type) { + LL_PROFILE_ZONE_NAMED("Pass"); for (U32 i = 0; i < mRiggedFace[type].size(); ++i) { + LL_PROFILE_ZONE_NAMED("Face"); LLFace* face = mRiggedFace[type][i]; LLDrawable* drawable = face->getDrawable(); if (!drawable) -- cgit v1.3 From 04e1962d48e36f9055f0d893fc1b7a97d9e334c6 Mon Sep 17 00:00:00 2001 From: "Brad Payne (Vir Linden)" Date: Wed, 15 Sep 2021 20:06:56 +0100 Subject: SL-15742 - python 3 support for integration test script --- indra/llcommon/tests/llprocess_test.cpp | 32 ++++++++++++++++++++------------ 1 file changed, 20 insertions(+), 12 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/tests/llprocess_test.cpp b/indra/llcommon/tests/llprocess_test.cpp index f0eafa8201..447c7f50f2 100644 --- a/indra/llcommon/tests/llprocess_test.cpp +++ b/indra/llcommon/tests/llprocess_test.cpp @@ -356,14 +356,15 @@ namespace tut // Create a script file in a temporary place. NamedTempFile script("py", + "from __future__ import print_function" EOL "import sys" EOL "import time" EOL EOL "time.sleep(2)" EOL - "print >>sys.stdout, 'stdout after wait'" EOL + "print('stdout after wait',file=sys.stdout)" EOL "sys.stdout.flush()" EOL "time.sleep(2)" EOL - "print >>sys.stderr, 'stderr after wait'" EOL + "print('stderr after wait',file=sys.stderr)" EOL "sys.stderr.flush()" EOL ); @@ -568,12 +569,12 @@ namespace tut { set_test_name("arguments"); PythonProcessLauncher py(get_test_name(), - "from __future__ import with_statement\n" + "from __future__ import with_statement, print_function\n" "import sys\n" // note nonstandard output-file arg! "with open(sys.argv[3], 'w') as f:\n" " for arg in sys.argv[1:]:\n" - " print >>f, arg\n"); + " print(arg,file=f)\n"); // We expect that PythonProcessLauncher has already appended // its own NamedTempFile to mParams.args (sys.argv[0]). py.mParams.args.add("first arg"); // sys.argv[1] @@ -857,7 +858,8 @@ namespace tut set_test_name("'bogus' test"); CaptureLog recorder; PythonProcessLauncher py(get_test_name(), - "print 'Hello world'\n"); + "from __future__ import print_function\n" + "print('Hello world')\n"); py.mParams.files.add(LLProcess::FileParam("bogus")); py.mPy = LLProcess::create(py.mParams); ensure("should have rejected 'bogus'", ! py.mPy); @@ -872,7 +874,8 @@ namespace tut // Replace this test with one or more real 'file' tests when we // implement 'file' support PythonProcessLauncher py(get_test_name(), - "print 'Hello world'\n"); + "from __future__ import print_function\n" + "print('Hello world')\n"); py.mParams.files.add(LLProcess::FileParam()); py.mParams.files.add(LLProcess::FileParam("file")); py.mPy = LLProcess::create(py.mParams); @@ -887,7 +890,8 @@ namespace tut // implement 'tpipe' support CaptureLog recorder; PythonProcessLauncher py(get_test_name(), - "print 'Hello world'\n"); + "from __future__ import print_function\n" + "print('Hello world')\n"); py.mParams.files.add(LLProcess::FileParam()); py.mParams.files.add(LLProcess::FileParam("tpipe")); py.mPy = LLProcess::create(py.mParams); @@ -904,7 +908,8 @@ namespace tut // implement 'npipe' support CaptureLog recorder; PythonProcessLauncher py(get_test_name(), - "print 'Hello world'\n"); + "from __future__ import print_function\n" + "print('Hello world')\n"); py.mParams.files.add(LLProcess::FileParam()); py.mParams.files.add(LLProcess::FileParam()); py.mParams.files.add(LLProcess::FileParam("npipe")); @@ -980,7 +985,8 @@ namespace tut { set_test_name("get*Pipe() validation"); PythonProcessLauncher py(get_test_name(), - "print 'this output is expected'\n"); + "from __future__ import print_function\n" + "print('this output is expected')\n"); py.mParams.files.add(LLProcess::FileParam("pipe")); // pipe for stdin py.mParams.files.add(LLProcess::FileParam()); // inherit stdout py.mParams.files.add(LLProcess::FileParam("pipe")); // pipe for stderr @@ -1000,14 +1006,15 @@ namespace tut { set_test_name("talk to stdin/stdout"); PythonProcessLauncher py(get_test_name(), + "from __future__ import print_function\n" "import sys, time\n" - "print 'ok'\n" + "print('ok')\n" "sys.stdout.flush()\n" "# wait for 'go' from test program\n" "go = sys.stdin.readline()\n" "if go != 'go\\n':\n" " sys.exit('expected \"go\", saw %r' % go)\n" - "print 'ack'\n"); + "print('ack')\n"); py.mParams.files.add(LLProcess::FileParam("pipe")); // stdin py.mParams.files.add(LLProcess::FileParam("pipe")); // stdout py.launch(); @@ -1118,7 +1125,8 @@ namespace tut { set_test_name("ReadPipe \"eof\" event"); PythonProcessLauncher py(get_test_name(), - "print 'Hello from Python!'\n"); + "from __future__ import print_function\n" + "print('Hello from Python!')\n"); py.mParams.files.add(LLProcess::FileParam()); // stdin py.mParams.files.add(LLProcess::FileParam("pipe")); // stdout py.launch(); -- cgit v1.3 From a4c9fb003f36955cecb0b987b5ddf9a04edd0f90 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Fri, 3 Sep 2021 17:20:22 -0700 Subject: SL-15709: Windows: Include Tracy source directly; don't use a library --- indra/cmake/LLCommon.cmake | 1 - indra/cmake/Tracy.cmake | 3 --- indra/llcommon/llframetimer.cpp | 11 ++++------- indra/llcommon/llprofiler.h | 6 ++++-- 4 files changed, 8 insertions(+), 13 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/cmake/LLCommon.cmake b/indra/cmake/LLCommon.cmake index b6f310fe12..34499aaa36 100644 --- a/indra/cmake/LLCommon.cmake +++ b/indra/cmake/LLCommon.cmake @@ -33,7 +33,6 @@ else (LINUX) ${BOOST_CONTEXT_LIBRARY} ${BOOST_THREAD_LIBRARY} ${BOOST_SYSTEM_LIBRARY} - ${TRACY_LIBRARY} ) endif (LINUX) diff --git a/indra/cmake/Tracy.cmake b/indra/cmake/Tracy.cmake index dfa3e83039..1b8c3db2e2 100644 --- a/indra/cmake/Tracy.cmake +++ b/indra/cmake/Tracy.cmake @@ -12,17 +12,14 @@ if (USE_TRACY) if (WINDOWS) MESSAGE(STATUS "Including Tracy for Windows: '${TRACY_INCLUDE_DIR}'") - set(TRACY_LIBRARY tracy) endif (WINDOWS) if (DARWIN) MESSAGE(STATUS "Including Tracy for Darwin: '${TRACY_INCLUDE_DIR}'") - set(TRACY_LIBRARY "") endif (DARWIN) if (LINUX) MESSAGE(STATUS "Including Tracy for Linux: '${TRACY_INCLUDE_DIR}'") - set(TRACY_LIBRARY "") endif (LINUX) else (USE_TRACY) # Tracy.cmake should not set LLCOMMON_INCLUDE_DIRS, let LLCommon.cmake do that diff --git a/indra/llcommon/llframetimer.cpp b/indra/llcommon/llframetimer.cpp index e293a557c0..c54029e8b4 100644 --- a/indra/llcommon/llframetimer.cpp +++ b/indra/llcommon/llframetimer.cpp @@ -29,13 +29,10 @@ #include "llframetimer.h" -// On Windows we build a static lib and link with that -// On macOS we don't bother building a stand alone lib, just include the one source file we need for Tracy support -#if LL_DARWIN - #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER - #include "TracyClient.cpp" - #endif // LL_PROFILER_CONFIGURATION -#endif // LL_DARWIN +// We don't bother building a stand alone lib; we just need to include the one source file for Tracy support +#if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER + #include "TracyClient.cpp" +#endif // LL_PROFILER_CONFIGURATION // Static members //LLTimer LLFrameTimer::sInternalTimer; diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 4674985e06..062c9360dd 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -39,8 +39,10 @@ #if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define TRACY_ENABLE 1 - #define TRACY_NO_BROADCAST 1 - #define TRACY_ONLY_LOCALHOST 1 +// Normally these would be enabled but we want to be able to build any viewer with Tracy enabled and run the Tracy server on another machine +// They must be undefined in order to work across multiple machines +// #define TRACY_NO_BROADCAST 1 +// #define TRACY_ONLY_LOCALHOST 1 #define TRACY_ONLY_IPV4 1 #include "Tracy.hpp" #endif -- cgit v1.3 From fc612fd8a0057daa7436c8d2285ccee0c634378a Mon Sep 17 00:00:00 2001 From: Runitai Linden Date: Mon, 13 Sep 2021 12:41:57 -0500 Subject: SL-15975 Add Tracy-only profile macros that are no-ops when Tracy is disabled. --- indra/llcommon/llprofiler.h | 6 ++++++ indra/newview/lldrawpoolavatar.cpp | 4 ++++ 2 files changed, 10 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 062c9360dd..62e649913b 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -51,16 +51,22 @@ #define LL_PROFILER_FRAME_END FrameMark #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, name, true ); + #define LL_PROFILE_ZONE_SCOPED ZoneScoped #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER #define LL_PROFILER_FRAME_END #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) #define LL_RECORD_BLOCK_TIME(name) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_PROFILE_ZONE_NAMED(name) // LL_PROFILE_ZONE_NAMED is a no-op when Tracy is disabled + #define LL_PROFILE_ZONE_SCOPED // LL_PROFILE_ZONE_SCOPED is a no-op when Tracy is disabled #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define LL_PROFILER_FRAME_END FrameMark #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #timer_stat, true ) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_SCOPED ZoneScoped #endif #else #define LL_PROFILER_FRAME_END diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index 687b13d2c8..c04142ab47 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -1840,6 +1840,8 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( LLVolume* volume, LLVolumeFace& vol_face) { + LL_PROFILE_ZONE_SCOPED; + LLVector4a* weights = vol_face.mWeights; if (!weights) { @@ -2352,8 +2354,10 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar) //update rigged vertex buffers for (U32 type = 0; type < NUM_RIGGED_PASSES; ++type) { + LL_PROFILE_ZONE_NAMED("Pass"); for (U32 i = 0; i < mRiggedFace[type].size(); ++i) { + LL_PROFILE_ZONE_NAMED("Face"); LLFace* face = mRiggedFace[type][i]; LLDrawable* drawable = face->getDrawable(); if (!drawable) -- cgit v1.3 From 84da92663aad221db19927de26922417e7cb45c6 Mon Sep 17 00:00:00 2001 From: Runitai Linden Date: Tue, 14 Sep 2021 20:18:58 -0500 Subject: SL-15961 Convert LLMeshRepository::mSkinMap into unordered_map and reduce number of per-frame lookups to said map. --- indra/llcommon/lluuid.h | 11 +++++++ indra/newview/lldrawpoolavatar.cpp | 65 ++++++++++++++++++-------------------- indra/newview/lldrawpoolavatar.h | 3 +- indra/newview/llmeshrepository.cpp | 40 +++++++++++------------ indra/newview/llmeshrepository.h | 3 +- 5 files changed, 65 insertions(+), 57 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/lluuid.h b/indra/llcommon/lluuid.h index fe7482ba29..86a396ab06 100644 --- a/indra/llcommon/lluuid.h +++ b/indra/llcommon/lluuid.h @@ -184,6 +184,17 @@ struct boost::hash } }; +// Adapt boost hash to std hash +namespace std +{ + template<> struct hash + { + std::size_t operator()(LLUUID const& s) const noexcept + { + return boost::hash()(s); + } + }; +} #endif diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index c04142ab47..6c4844f9ee 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -1836,7 +1836,7 @@ void LLDrawPoolAvatar::getRiggedGeometry( void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( LLVOAvatar* avatar, LLFace* face, - const LLMeshSkinInfo* skin, + const LLVOVolume* vobj, LLVolume* volume, LLVolumeFace& vol_face) { @@ -1848,14 +1848,14 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( return; } + if (!vobj || vobj->isNoLOD()) + { + return; + } + LLPointer buffer = face->getVertexBuffer(); LLDrawable* drawable = face->getDrawable(); - if (drawable->getVOVolume() && drawable->getVOVolume()->isNoLOD()) - { - return; - } - const U32 max_joints = LLSkinningUtil::getMaxJointCount(); #if USE_SEPARATE_JOINT_INDICES_AND_WEIGHTS @@ -1895,23 +1895,26 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( } #endif - // FIXME ugly const cast - LLSkinningUtil::scrubInvalidJoints(avatar, const_cast(skin)); - - U32 data_mask = face->getRiggedVertexBufferDataMask(); + U32 data_mask = face->getRiggedVertexBufferDataMask(); + const LLMeshSkinInfo* skin = nullptr; - if (!vol_face.mWeightsScrubbed) - { - LLSkinningUtil::scrubSkinWeights(weights, vol_face.mNumVertices, skin); - vol_face.mWeightsScrubbed = TRUE; - } - if (buffer.isNull() || buffer->getTypeMask() != data_mask || buffer->getNumVerts() != vol_face.mNumVertices || buffer->getNumIndices() != vol_face.mNumIndices || (drawable && drawable->isState(LLDrawable::REBUILD_ALL))) { + LL_PROFILE_ZONE_NAMED("Rigged VBO Rebuild"); + skin = vobj->getSkinInfo(); + // FIXME ugly const cast + LLSkinningUtil::scrubInvalidJoints(avatar, const_cast(skin)); + + if (!vol_face.mWeightsScrubbed) + { + LLSkinningUtil::scrubSkinWeights(weights, vol_face.mNumVertices, skin); + vol_face.mWeightsScrubbed = TRUE; + } + if (drawable && drawable->isState(LLDrawable::REBUILD_ALL)) { //rebuild EVERY face in the drawable, not just this one, to avoid missing drawable wide rebuild issues @@ -1937,18 +1940,13 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( } } - if (buffer.isNull() || - buffer->getNumVerts() != vol_face.mNumVertices || - buffer->getNumIndices() != vol_face.mNumIndices) - { - // Allocation failed - return; - } - - if (!buffer.isNull() && - sShaderLevel <= 0 && - face->mLastSkinTime < avatar->getLastSkinTime()) + if (sShaderLevel <= 0 && + face->mLastSkinTime < avatar->getLastSkinTime() && + !buffer.isNull() && + buffer->getNumVerts() == vol_face.mNumVertices && + buffer->getNumIndices() == vol_face.mNumIndices) { + LL_PROFILE_ZONE_NAMED("Software Skinning"); //perform software vertex skinning for this face LLStrider position; LLStrider normal; @@ -1965,6 +1963,11 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer( LLVector4a* norm = has_normal ? (LLVector4a*) normal.get() : NULL; + if (skin == nullptr) + { + skin = vobj->getSkinInfo(); + } + //build matrix palette LLMatrix4a mat[LL_MAX_JOINTS_PER_MESH_OBJECT]; U32 count = LLSkinningUtil::getMeshJointCount(skin); @@ -2380,16 +2383,10 @@ void LLDrawPoolAvatar::updateRiggedVertexBuffers(LLVOAvatar* avatar) continue; } - const LLMeshSkinInfo* skin = vobj->getSkinInfo(); - if (!skin) - { - continue; - } - stop_glerror(); LLVolumeFace& vol_face = volume->getVolumeFace(te); - updateRiggedFaceVertexBuffer(avatar, face, skin, volume, vol_face); + updateRiggedFaceVertexBuffer(avatar, face, vobj, volume, vol_face); } } } diff --git a/indra/newview/lldrawpoolavatar.h b/indra/newview/lldrawpoolavatar.h index 92a8538958..9b26266ced 100644 --- a/indra/newview/lldrawpoolavatar.h +++ b/indra/newview/lldrawpoolavatar.h @@ -30,6 +30,7 @@ #include "lldrawpool.h" class LLVOAvatar; +class LLVOVolume; class LLGLSLShader; class LLFace; class LLMeshSkinInfo; @@ -253,7 +254,7 @@ typedef enum void getRiggedGeometry(LLFace* face, LLPointer& buffer, U32 data_mask, const LLMeshSkinInfo* skin, LLVolume* volume, const LLVolumeFace& vol_face); void updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* facep, - const LLMeshSkinInfo* skin, + const LLVOVolume* vobj, LLVolume* volume, LLVolumeFace& vol_face); void updateRiggedVertexBuffers(LLVOAvatar* avatar); diff --git a/indra/newview/llmeshrepository.cpp b/indra/newview/llmeshrepository.cpp index 2c1c1191da..8ac64dbd15 100644 --- a/indra/newview/llmeshrepository.cpp +++ b/indra/newview/llmeshrepository.cpp @@ -4046,28 +4046,26 @@ S32 LLMeshRepository::getActualMeshLOD(const LLVolumeParams& mesh_params, S32 lo const LLMeshSkinInfo* LLMeshRepository::getSkinInfo(const LLUUID& mesh_id, const LLVOVolume* requesting_obj) { LL_RECORD_BLOCK_TIME(FTM_MESH_FETCH); + if (mesh_id.notNull()) + { + skin_map::iterator iter = mSkinMap.find(mesh_id); + if (iter != mSkinMap.end()) + { + return &(iter->second); + } - if (mesh_id.notNull()) - { - skin_map::iterator iter = mSkinMap.find(mesh_id); - if (iter != mSkinMap.end()) - { - return &(iter->second); - } - - //no skin info known about given mesh, try to fetch it - { - LLMutexLock lock(mMeshMutex); - //add volume to list of loading meshes - skin_load_map::iterator iter = mLoadingSkins.find(mesh_id); - if (iter == mLoadingSkins.end()) - { //no request pending for this skin info - mPendingSkinRequests.push(mesh_id); - } - mLoadingSkins[mesh_id].insert(requesting_obj->getID()); - } - } - + //no skin info known about given mesh, try to fetch it + { + LLMutexLock lock(mMeshMutex); + //add volume to list of loading meshes + skin_load_map::iterator iter = mLoadingSkins.find(mesh_id); + if (iter == mLoadingSkins.end()) + { //no request pending for this skin info + mPendingSkinRequests.push(mesh_id); + } + mLoadingSkins[mesh_id].insert(requesting_obj->getID()); + } + } return NULL; } diff --git a/indra/newview/llmeshrepository.h b/indra/newview/llmeshrepository.h index 81e49cb1d8..c1698194cb 100644 --- a/indra/newview/llmeshrepository.h +++ b/indra/newview/llmeshrepository.h @@ -27,6 +27,7 @@ #ifndef LL_MESH_REPOSITORY_H #define LL_MESH_REPOSITORY_H +#include #include "llassettype.h" #include "llmodel.h" #include "lluuid.h" @@ -613,7 +614,7 @@ public: typedef std::map > mesh_load_map; mesh_load_map mLoadingMeshes[4]; - typedef std::map skin_map; + typedef std::unordered_map skin_map; skin_map mSkinMap; typedef std::map decomposition_map; -- cgit v1.3 From 6c6eb59e947631275b2149208e4b46977dfa0309 Mon Sep 17 00:00:00 2001 From: Dave Houlton Date: Wed, 15 Sep 2021 10:11:25 -0600 Subject: SL-15962 Add hooks for tracy memory profiling --- indra/llcommon/linden_common.h | 8 ++++++++ indra/llcommon/llcommon.cpp | 42 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 50 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h index 45ac43910c..b2c5be6b76 100644 --- a/indra/llcommon/linden_common.h +++ b/indra/llcommon/linden_common.h @@ -27,6 +27,14 @@ #ifndef LL_LINDEN_COMMON_H #define LL_LINDEN_COMMON_H +#include "llprofiler.h" +#if (TRACY_ENABLE) // hooks for memory profiling +void *tracy_aligned_malloc(size_t size, size_t alignment); +void tracy_aligned_free(void *memblock); +#define _aligned_malloc(X, Y) tracy_aligned_malloc((X), (Y)) +#define _aligned_free(X) tracy_aligned_free((X)) +#endif + // *NOTE: Please keep includes here to a minimum! // // Files included here are included in every library .cpp file and diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index 96be913d17..da61e7539a 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -33,6 +33,48 @@ #include "lltracethreadrecorder.h" #include "llcleanup.h" +#if (TRACY_ENABLE) +// Override new/delet for tracy memory profiling +void *operator new(size_t size) +{ + auto ptr = (malloc) (size); + if (!ptr) + { + throw std::bad_alloc(); + return nullptr; + } + TracyAlloc(ptr, size); + return ptr; +} + +void operator delete(void *ptr) noexcept +{ + TracyFree(ptr); + (free)(ptr); +} + +// C-style malloc/free can't be so easily overridden, so we define tracy versions and use +// a pre-processor #define in linden_common.h to redirect to them. The parens around the native +// functions below prevents recursive substitution by the preprocessor. +// +// Unaligned mallocs are rare in LL code but hooking them causes problems in 3p lib code (looking at +// you, Havok), so we'll only capture the aligned version. + +void *tracy_aligned_malloc(size_t size, size_t alignment) +{ + auto ptr = (_aligned_malloc) (size, alignment); + if (ptr) TracyAlloc(ptr, size); + return ptr; +} + +void tracy_aligned_free(void *memblock) +{ + TracyFree(memblock); + (_aligned_free)(memblock); +} + +#endif + //static BOOL LLCommon::sAprInitialized = FALSE; -- cgit v1.3 From a35544c701b223ba08f0607c872d8afbb08114f5 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Tue, 21 Sep 2021 15:56:18 -0700 Subject: SL-16027: Add Tracy OpenGL support --- indra/llcommon/linden_common.h | 4 +--- indra/llcommon/llprofiler.h | 3 +++ indra/llrender/llglheaders.h | 19 +++++++++++++++++++ indra/llrender/llrendertarget.cpp | 2 ++ indra/llrender/llvertexbuffer.cpp | 5 +++++ indra/llwindow/llwindowwin32.cpp | 4 ++++ indra/newview/llappviewerwin32.cpp | 4 ++++ indra/newview/llface.cpp | 1 + indra/newview/llspatialpartition.cpp | 11 +++++++++-- indra/newview/pipeline.cpp | 1 + 10 files changed, 49 insertions(+), 5 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h index b2c5be6b76..a228fd22be 100644 --- a/indra/llcommon/linden_common.h +++ b/indra/llcommon/linden_common.h @@ -28,7 +28,7 @@ #define LL_LINDEN_COMMON_H #include "llprofiler.h" -#if (TRACY_ENABLE) // hooks for memory profiling +#if TRACY_ENABLE && !defined(LL_PROFILER_ENABLE_TRACY_OPENGL) // hooks for memory profiling void *tracy_aligned_malloc(size_t size, size_t alignment); void tracy_aligned_free(void *memblock); #define _aligned_malloc(X, Y) tracy_aligned_malloc((X), (Y)) @@ -68,6 +68,4 @@ void tracy_aligned_free(void *memblock); #include "llerror.h" #include "llfile.h" -#include "llprofiler.h" // must be before fast timer; needed due to LLThreads potentially needing access to tracy - #endif diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 62e649913b..62ec20fa44 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -45,6 +45,9 @@ // #define TRACY_ONLY_LOCALHOST 1 #define TRACY_ONLY_IPV4 1 #include "Tracy.hpp" + + // Mutually exclusive with detailed memory tracing + #define LL_PROFILER_ENABLE_TRACY_OPENGL 0 #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h index 6bca3623e0..e7f9315d11 100644 --- a/indra/llrender/llglheaders.h +++ b/indra/llrender/llglheaders.h @@ -812,4 +812,23 @@ extern void glGetBufferPointervARB (GLenum, GLenum, GLvoid* *); #define GL_RENDERBUFFER_FREE_MEMORY_ATI 0x87FD #endif +#if LL_PROFILER_ENABLE_TRACY_OPENGL + // Tracy uses the following: + // glGenQueries + // glGetQueryiv + // glGetQueryObjectiv + #define glGenQueries glGenQueriesARB + #define glGetQueryiv glGetQueryivARB + #define glGetQueryObjectiv glGetQueryObjectivARB + #include + + #define LL_PROFILER_GPU_ZONEC(name,color) TracyGpuZoneC(name,color); + #define LL_PROFILER_GPU_COLLECT TracyGpuCollect + #define LL_PROFILER_GPU_CONTEXT TracyGpuContext +#else + #define LL_PROFILER_GPU_ZONEC(name,color) (void)name;(void)color; + #define LL_PROFILER_GPU_COLLECT + #define LL_PROFILER_GPU_CONTEXT +#endif + #endif // LL_LLGLHEADERS_H diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp index e3c0255290..401085a00b 100644 --- a/indra/llrender/llrendertarget.cpp +++ b/indra/llrender/llrendertarget.cpp @@ -437,11 +437,13 @@ void LLRenderTarget::bindTarget() GL_COLOR_ATTACHMENT1, GL_COLOR_ATTACHMENT2, GL_COLOR_ATTACHMENT3}; + LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x4000FF ) glDrawBuffersARB(mTex.size(), drawbuffers); } if (mTex.empty()) { //no color buffer to draw to + LL_PROFILER_GPU_ZONEC( "gl.DrawBuffer", 0x0000FF ) glDrawBuffer(GL_NONE); glReadBuffer(GL_NONE); } diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 7d2b09ca4a..b69b644ebc 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -615,6 +615,7 @@ void LLVertexBuffer::drawArrays(U32 mode, const std::vector& pos, con glNormalPointer(GL_FLOAT, 0, norm[0].mV); } LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF0000 ) glDrawArrays(sGLMode[mode], 0, count); LLGLSLShader::stopProfile(count, mode); } @@ -654,6 +655,7 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto } LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x80FF80 ) glDrawElements(sGLMode[mode], num_indices, GL_UNSIGNED_SHORT, indicesp); LLGLSLShader::stopProfile(num_indices, mode); } @@ -763,6 +765,7 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi stop_glerror(); LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawRangeElements", 0xFFFF00 ) glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, idx); LLGLSLShader::stopProfile(count, mode); @@ -814,6 +817,7 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const stop_glerror(); LLGLSLShader::startProfile(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0xA0FFA0 ) glDrawElements(sGLMode[mode], count, GL_UNSIGNED_SHORT, ((U16*) getIndicesPointer()) + indices_offset); LLGLSLShader::stopProfile(count, mode); @@ -861,6 +865,7 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const stop_glerror(); LLGLSLShader::startProfile(); stop_glerror(); + LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF4040 ) glDrawArrays(sGLMode[mode], first, count); stop_glerror(); LLGLSLShader::stopProfile(count, mode); diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index b2b123f0da..1b8881cc86 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -1748,6 +1748,8 @@ const S32 max_format = (S32)num_formats - 1; return FALSE; } + LL_PROFILER_GPU_CONTEXT + if (!gGLManager.initGL()) { close(); @@ -3474,6 +3476,8 @@ BOOL LLWindowWin32::resetDisplayResolution() void LLWindowWin32::swapBuffers() { SwapBuffers(mhDC); + + LL_PROFILER_GPU_COLLECT } diff --git a/indra/newview/llappviewerwin32.cpp b/indra/newview/llappviewerwin32.cpp index 9daea515e5..fb53a7648d 100644 --- a/indra/newview/llappviewerwin32.cpp +++ b/indra/newview/llappviewerwin32.cpp @@ -323,6 +323,10 @@ int APIENTRY WINMAIN(HINSTANCE hInstance, PWSTR pCmdLine, int nCmdShow) { + // Call Tracy first thing to have it allocate memory + // https://github.com/wolfpld/tracy/issues/196 + LL_PROFILER_FRAME_END + const S32 MAX_HEAPS = 255; DWORD heap_enable_lfh_error[MAX_HEAPS]; S32 num_heaps = 0; diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 4a802ad9aa..6e55d8f66a 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -585,6 +585,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) glTexCoordPointer(2, GL_FLOAT, 8, vol_face.mTexCoords); } gGL.syncMatrices(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x00FF00 ); glDrawElements(GL_TRIANGLES, vol_face.mNumIndices, GL_UNSIGNED_SHORT, vol_face.mIndices); glDisableClientState(GL_TEXTURE_COORD_ARRAY); } diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index efa4a7fd66..0adf58a0bf 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -2700,11 +2700,17 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume) glVertexPointer(3, GL_FLOAT, 16, phys_volume->mHullPoints); gGL.diffuseColor4fv(line_color.mV); gGL.syncMatrices(); - glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + { + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x20FF20 ) + glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + } gGL.diffuseColor4fv(color.mV); glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); - glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + { + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x40FF40 ) + glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices); + } } else { @@ -3222,6 +3228,7 @@ void renderRaycast(LLDrawable* drawablep) gGL.diffuseColor4f(0,1,1,0.5f); glVertexPointer(3, GL_FLOAT, sizeof(LLVector4a), face.mPositions); gGL.syncMatrices(); + LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x60FF60 ); glDrawElements(GL_TRIANGLES, face.mNumIndices, GL_UNSIGNED_SHORT, face.mIndices); } diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index e7c2d4db39..4e28c8c493 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -11146,6 +11146,7 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) if (LLPipeline::sRenderDeferred) { GLuint buff = GL_COLOR_ATTACHMENT0; + LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x8000FF ); glDrawBuffersARB(1, &buff); } -- cgit v1.3 From b9ad51981eb992ebe77b8ffbde48b2797ff55cef Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Mon, 20 Sep 2021 08:33:41 -0700 Subject: SL-16014: Add macros for better markup in Tracy --- indra/llcommon/llprofiler.h | 50 +++++++++++++++++++++++++++++++++------------ 1 file changed, 37 insertions(+), 13 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 62ec20fa44..59e7dc02c6 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -51,25 +51,49 @@ #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY - #define LL_PROFILER_FRAME_END FrameMark - #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) - #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); - #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, name, true ); - #define LL_PROFILE_ZONE_SCOPED ZoneScoped + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneScoped // Want descriptive names; was: ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, name, true ); + #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB + #define LL_PROFILE_ZONE_SCOPED ZoneScoped + + #define LL_PROFILE_ZONE_NUM( val ) ZoneValue( val ) + #define LL_PROFILE_ZONE_TEXT( text, size ) ZoneText( text, size ) + + #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow + #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan + #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER #define LL_PROFILER_FRAME_END - #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) + #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) #define LL_RECORD_BLOCK_TIME(name) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); - #define LL_PROFILE_ZONE_NAMED(name) // LL_PROFILE_ZONE_NAMED is a no-op when Tracy is disabled - #define LL_PROFILE_ZONE_SCOPED // LL_PROFILE_ZONE_SCOPED is a no-op when Tracy is disabled + #define LL_PROFILE_ZONE_NAMED(name) // LL_PROFILE_ZONE_NAMED is a no-op when Tracy is disabled + #define LL_PROFILE_ZONE_SCOPED // LL_PROFILE_ZONE_SCOPED is a no-op when Tracy is disabled + #define LL_PRPFILE_ZONE_COLOR(name,color) // LL_RECORD_BLOCK_TIME(name) + + #define LL_PROFILE_ZONE_NUM( val ) (void)( val ); // Not supported + #define LL_PROFILE_ZONE_TEXT( text, size ) (void)( text ); void( size ); // Not supported + + #define LL_PROFILE_ZONE_ERR(name) (void)(name); // Not supported + #define LL_PROFILE_ZONE_INFO(name) (void)(name); // Not supported + #define LL_PROFILE_ZONE_WARN(name) (void)(name); // Not supported #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER - #define LL_PROFILER_FRAME_END FrameMark - #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) - #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN( ___tracy_scoped_zone, #timer_stat, true ) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); - #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); - #define LL_PROFILE_ZONE_SCOPED ZoneScoped + #define LL_PROFILER_FRAME_END FrameMark + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_RECORD_BLOCK_TIME(name) ZoneScoped const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); + #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB + #define LL_PROFILE_ZONE_SCOPED ZoneScoped + + #define LL_PROFILE_ZONE_NUM( val ) ZoneValue( val ) + #define LL_PROFILE_ZONE_TEXT( text, size ) ZoneText( text, size ) + + #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow + #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan + #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red #endif #else #define LL_PROFILER_FRAME_END -- cgit v1.3 From 51a887a51c4d86775006fea1d9522031142ec79d Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Mon, 20 Sep 2021 09:46:43 -0700 Subject: SL-16014: Add Tracy markup for LLSD --- indra/llcommon/llsdutil.cpp | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsdutil.cpp b/indra/llcommon/llsdutil.cpp index eb3a96b133..c2fe15e9b7 100644 --- a/indra/llcommon/llsdutil.cpp +++ b/indra/llcommon/llsdutil.cpp @@ -214,6 +214,8 @@ BOOL compare_llsd_with_template( const LLSD& template_llsd, LLSD& resultant_llsd) { + LL_PROFILE_ZONE_SCOPED + if ( llsd_to_test.isUndefined() && template_llsd.isDefined() ) @@ -335,6 +337,8 @@ bool filter_llsd_with_template( const LLSD & template_llsd, LLSD & resultant_llsd) { + LL_PROFILE_ZONE_SCOPED + if (llsd_to_test.isUndefined() && template_llsd.isDefined()) { resultant_llsd = template_llsd; @@ -529,6 +533,8 @@ class TypeLookup public: TypeLookup() { + LL_PROFILE_ZONE_SCOPED + for (const Data *di(boost::begin(typedata)), *dend(boost::end(typedata)); di != dend; ++di) { mMap[di->type] = di->name; @@ -537,6 +543,8 @@ public: std::string lookup(LLSD::Type type) const { + LL_PROFILE_ZONE_SCOPED + MapType::const_iterator found = mMap.find(type); if (found != mMap.end()) { @@ -587,6 +595,8 @@ static std::string match_types(LLSD::Type expect, // prototype.type() LLSD::Type actual, // type we're checking const std::string& pfx) // as for llsd_matches { + LL_PROFILE_ZONE_SCOPED + // Trivial case: if the actual type is exactly what we expect, we're good. if (actual == expect) return ""; @@ -624,6 +634,8 @@ static std::string match_types(LLSD::Type expect, // prototype.type() // see docstring in .h file std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::string& pfx) { + LL_PROFILE_ZONE_SCOPED + // An undefined prototype means that any data is valid. // An undefined slot in an array or map prototype means that any data // may fill that slot. @@ -756,6 +768,8 @@ std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::str bool llsd_equals(const LLSD& lhs, const LLSD& rhs, int bits) { + LL_PROFILE_ZONE_SCOPED + // We're comparing strict equality of LLSD representation rather than // performing any conversions. So if the types aren't equal, the LLSD // values aren't equal. @@ -864,6 +878,8 @@ namespace llsd LLSD& drill(LLSD& blob, const LLSD& rawPath) { + LL_PROFILE_ZONE_SCOPED + // Treat rawPath uniformly as an array. If it's not already an array, // store it as the only entry in one. (But let's say Undefined means an // empty array.) @@ -889,6 +905,8 @@ LLSD& drill(LLSD& blob, const LLSD& rawPath) // path entry that's bad. for (LLSD::Integer i = 0; i < path.size(); ++i) { + LL_PROFILE_ZONE_NUM( i ) + const LLSD& key{path[i]}; if (key.isString()) { @@ -917,6 +935,8 @@ LLSD& drill(LLSD& blob, const LLSD& rawPath) LLSD drill(const LLSD& blob, const LLSD& path) { + LL_PROFILE_ZONE_SCOPED + // non-const drill() does exactly what we want. Temporarily cast away // const-ness and use that. return drill(const_cast(blob), path); @@ -929,6 +949,8 @@ LLSD drill(const LLSD& blob, const LLSD& path) // filter may be include to exclude/include keys in a map. LLSD llsd_clone(LLSD value, LLSD filter) { + LL_PROFILE_ZONE_SCOPED + LLSD clone; bool has_filter(filter.isMap()); -- cgit v1.3 From 2b19015f4394de59244c78f88c5c5df172632de6 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Mon, 20 Sep 2021 09:48:16 -0700 Subject: SL-16014: Add Tracy markup for LLEventFilter --- indra/llcommon/lleventfilter.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/lleventfilter.h b/indra/llcommon/lleventfilter.h index 48c2570732..7613850fb2 100644 --- a/indra/llcommon/lleventfilter.h +++ b/indra/llcommon/lleventfilter.h @@ -429,6 +429,8 @@ public: // path, then stores it to mTarget. virtual bool post(const LLSD& event) { + LL_PROFILE_ZONE_SCOPED + // Extract the element specified by 'mPath' from 'event'. To perform a // generic type-appropriate store through mTarget, construct an // LLSDParam and store that, thus engaging LLSDParam's custom -- cgit v1.3 From 43bca9e85d7fb1e6907fbe17f527e5bc8f543411 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Wed, 22 Sep 2021 14:58:36 -0700 Subject: SL-16014: Fix typo --- indra/llcommon/llprofiler.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 59e7dc02c6..49510df913 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -71,7 +71,7 @@ #define LL_RECORD_BLOCK_TIME(name) const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); #define LL_PROFILE_ZONE_NAMED(name) // LL_PROFILE_ZONE_NAMED is a no-op when Tracy is disabled #define LL_PROFILE_ZONE_SCOPED // LL_PROFILE_ZONE_SCOPED is a no-op when Tracy is disabled - #define LL_PRPFILE_ZONE_COLOR(name,color) // LL_RECORD_BLOCK_TIME(name) + #define LL_PROFILE_ZONE_COLOR(name,color) // LL_RECORD_BLOCK_TIME(name) #define LL_PROFILE_ZONE_NUM( val ) (void)( val ); // Not supported #define LL_PROFILE_ZONE_TEXT( text, size ) (void)( text ); void( size ); // Not supported -- cgit v1.3 From 675514bdb372c25b50dd2c42b06633895c86b8ce Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 27 Sep 2021 23:56:06 +0000 Subject: SL-16093 Don't force the console window to be open on developer builds because it causes frame stalls while logging. --- indra/llcommon/llerror.cpp | 12 +++++++++++- indra/llcommon/llerror.h | 5 ++++- indra/llcommon/llerrorcontrol.h | 1 + indra/llcommon/llmemory.cpp | 2 ++ indra/llcommon/llmutex.cpp | 12 ++++++++++++ indra/llcommon/llsys.cpp | 1 + indra/llcommon/llthread.cpp | 7 +++++++ indra/llrender/llglslshader.cpp | 1 + indra/llrender/llimagegl.cpp | 5 ++++- indra/llrender/llrender.cpp | 1 + indra/newview/llappviewer.cpp | 7 +------ indra/newview/llviewerdisplay.cpp | 11 ++++++++++- indra/newview/llviewerwindow.cpp | 1 + indra/newview/llworld.cpp | 1 + indra/newview/pipeline.cpp | 2 ++ indra/test/test.cpp | 1 + 16 files changed, 60 insertions(+), 10 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llerror.cpp b/indra/llcommon/llerror.cpp index 8355df9045..f7af181927 100644 --- a/indra/llcommon/llerror.cpp +++ b/indra/llcommon/llerror.cpp @@ -109,6 +109,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED int syslogPriority = LOG_CRIT; switch (level) { case LLError::LEVEL_DEBUG: syslogPriority = LOG_DEBUG; break; @@ -166,6 +167,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED if (LLError::getAlwaysFlush()) { mFile << message << std::endl; @@ -208,6 +210,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED static std::string s_ansi_error = createANSI("31"); // red static std::string s_ansi_warn = createANSI("34"); // blue static std::string s_ansi_debug = createANSI("35"); // magenta @@ -220,7 +223,8 @@ namespace { } else { - fprintf(stderr, "%s\n", message.c_str()); + LL_PROFILE_ZONE_NAMED("fprintf"); + fprintf(stderr, "%s\n", message.c_str()); } } @@ -229,6 +233,7 @@ namespace { LL_FORCE_INLINE void writeANSI(const std::string& ansi_code, const std::string& message) { + LL_PROFILE_ZONE_SCOPED static std::string s_ansi_bold = createANSI("1"); // bold static std::string s_ansi_reset = createANSI("0"); // reset // ANSI color code escape sequence, message, and reset in one fprintf call @@ -265,6 +270,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED mBuffer->addLine(message); } @@ -291,6 +297,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED debugger_print(message); } }; @@ -1178,6 +1185,7 @@ namespace void writeToRecorders(const LLError::CallSite& site, const std::string& message) { + LL_PROFILE_ZONE_SCOPED LLError::ELevel level = site.mLevel; LLError::SettingsConfigPtr s = LLError::Settings::getInstance()->getSettingsConfig(); @@ -1311,6 +1319,7 @@ namespace LLError bool Log::shouldLog(CallSite& site) { + LL_PROFILE_ZONE_SCOPED LLMutexTrylock lock(getMutex(), 5); if (!lock.isLocked()) { @@ -1354,6 +1363,7 @@ namespace LLError void Log::flush(const std::ostringstream& out, const CallSite& site) { + LL_PROFILE_ZONE_SCOPED LLMutexTrylock lock(getMutex(),5); if (!lock.isLocked()) { diff --git a/indra/llcommon/llerror.h b/indra/llcommon/llerror.h index d439136ca8..d06c0e2132 100644 --- a/indra/llcommon/llerror.h +++ b/indra/llcommon/llerror.h @@ -35,7 +35,9 @@ #include "stdtypes.h" +#include "llprofiler.h" #include "llpreprocessor.h" + #include const int LL_ERR_NOERR = 0; @@ -348,7 +350,8 @@ typedef LLError::NoClassInfo _LL_CLASS_TO_LOG; // if (condition) LL_INFOS() << "True" << LL_ENDL; else LL_INFOS()() << "False" << LL_ENDL; #define lllog(level, once, ...) \ - do { \ + do { \ + LL_PROFILE_ZONE_NAMED("lllog"); \ const char* tags[] = {"", ##__VA_ARGS__}; \ static LLError::CallSite _site(lllog_site_args_(level, once, tags)); \ lllog_test_() diff --git a/indra/llcommon/llerrorcontrol.h b/indra/llcommon/llerrorcontrol.h index e87bb7bf35..57f10b7895 100644 --- a/indra/llcommon/llerrorcontrol.h +++ b/indra/llcommon/llerrorcontrol.h @@ -190,6 +190,7 @@ namespace LLError {} void recordMessage(LLError::ELevel level, const std::string& message) override { + LL_PROFILE_ZONE_SCOPED mCallable(level, message); } private: diff --git a/indra/llcommon/llmemory.cpp b/indra/llcommon/llmemory.cpp index ea84e4c1ea..849867586a 100644 --- a/indra/llcommon/llmemory.cpp +++ b/indra/llcommon/llmemory.cpp @@ -82,6 +82,7 @@ void LLMemory::initMaxHeapSizeGB(F32Gigabytes max_heap_size) //static void LLMemory::updateMemoryInfo() { + LL_PROFILE_ZONE_SCOPED #if LL_WINDOWS PROCESS_MEMORY_COUNTERS counters; @@ -145,6 +146,7 @@ void* LLMemory::tryToAlloc(void* address, U32 size) //static void LLMemory::logMemoryInfo(BOOL update) { + LL_PROFILE_ZONE_SCOPED if(update) { updateMemoryInfo() ; diff --git a/indra/llcommon/llmutex.cpp b/indra/llcommon/llmutex.cpp index 4d73c04d07..a49002b5dc 100644 --- a/indra/llcommon/llmutex.cpp +++ b/indra/llcommon/llmutex.cpp @@ -44,6 +44,7 @@ LLMutex::~LLMutex() void LLMutex::lock() { + LL_PROFILE_ZONE_SCOPED if(isSelfLocked()) { //redundant lock mCount++; @@ -65,6 +66,7 @@ void LLMutex::lock() void LLMutex::unlock() { + LL_PROFILE_ZONE_SCOPED if (mCount > 0) { //not the root unlock mCount--; @@ -85,6 +87,7 @@ void LLMutex::unlock() bool LLMutex::isLocked() { + LL_PROFILE_ZONE_SCOPED if (!mMutex.try_lock()) { return true; @@ -108,6 +111,7 @@ LLThread::id_t LLMutex::lockingThread() const bool LLMutex::trylock() { + LL_PROFILE_ZONE_SCOPED if(isSelfLocked()) { //redundant lock mCount++; @@ -146,17 +150,20 @@ LLCondition::~LLCondition() void LLCondition::wait() { + LL_PROFILE_ZONE_SCOPED std::unique_lock< std::mutex > lock(mMutex); mCond.wait(lock); } void LLCondition::signal() { + LL_PROFILE_ZONE_SCOPED mCond.notify_one(); } void LLCondition::broadcast() { + LL_PROFILE_ZONE_SCOPED mCond.notify_all(); } @@ -166,6 +173,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex) : mMutex(mutex), mLocked(false) { + LL_PROFILE_ZONE_SCOPED if (mMutex) mLocked = mMutex->trylock(); } @@ -174,6 +182,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms) : mMutex(mutex), mLocked(false) { + LL_PROFILE_ZONE_SCOPED if (!mMutex) return; @@ -188,6 +197,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms) LLMutexTrylock::~LLMutexTrylock() { + LL_PROFILE_ZONE_SCOPED if (mMutex && mLocked) mMutex->unlock(); } @@ -199,6 +209,7 @@ LLMutexTrylock::~LLMutexTrylock() // LLScopedLock::LLScopedLock(std::mutex* mutex) : mMutex(mutex) { + LL_PROFILE_ZONE_SCOPED if(mutex) { mutex->lock(); @@ -217,6 +228,7 @@ LLScopedLock::~LLScopedLock() void LLScopedLock::unlock() { + LL_PROFILE_ZONE_SCOPED if(mLocked) { mMutex->unlock(); diff --git a/indra/llcommon/llsys.cpp b/indra/llcommon/llsys.cpp index 4e61fb8a58..6d5d043e8d 100644 --- a/indra/llcommon/llsys.cpp +++ b/indra/llcommon/llsys.cpp @@ -861,6 +861,7 @@ LLSD LLMemoryInfo::getStatsMap() const LLMemoryInfo& LLMemoryInfo::refresh() { + LL_PROFILE_ZONE_SCOPED mStatsMap = loadStatsMap(); LL_DEBUGS("LLMemoryInfo") << "Populated mStatsMap:\n"; diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp index a8cc750437..11f5a015f1 100644 --- a/indra/llcommon/llthread.cpp +++ b/indra/llcommon/llthread.cpp @@ -333,6 +333,7 @@ bool LLThread::runCondition(void) // Stop thread execution if requested until unpaused. void LLThread::checkPause() { + LL_PROFILE_ZONE_SCOPED mDataLock->lock(); // This is in a while loop because the pthread API allows for spurious wakeups. @@ -364,17 +365,20 @@ void LLThread::setQuitting() // static LLThread::id_t LLThread::currentID() { + LL_PROFILE_ZONE_SCOPED return std::this_thread::get_id(); } // static void LLThread::yield() { + LL_PROFILE_ZONE_SCOPED std::this_thread::yield(); } void LLThread::wake() { + LL_PROFILE_ZONE_SCOPED mDataLock->lock(); if(!shouldSleep()) { @@ -385,6 +389,7 @@ void LLThread::wake() void LLThread::wakeLocked() { + LL_PROFILE_ZONE_SCOPED if(!shouldSleep()) { mRunCondition->signal(); @@ -393,11 +398,13 @@ void LLThread::wakeLocked() void LLThread::lockData() { + LL_PROFILE_ZONE_SCOPED mDataLock->lock(); } void LLThread::unlockData() { + LL_PROFILE_ZONE_SCOPED mDataLock->unlock(); } diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp index 8bd9dbf9b8..9ab38d25a9 100644 --- a/indra/llrender/llglslshader.cpp +++ b/indra/llrender/llglslshader.cpp @@ -126,6 +126,7 @@ struct LLGLSLShaderCompareTimeElapsed //static void LLGLSLShader::finishProfile(bool emit_report) { + LL_PROFILE_ZONE_SCOPED sProfileEnabled = false; if (emit_report) diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 0151d20128..276fa55e15 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -1304,7 +1304,10 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt } stop_glerror(); - glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels); + { + LL_PROFILE_ZONE_NAMED("glTexImage2D"); + glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels); + } stop_glerror(); if (use_scratch) diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 03b6aac20c..43b4441ea8 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -1848,6 +1848,7 @@ LLLightState* LLRender::getLight(U32 index) void LLRender::setAmbientLightColor(const LLColor4& color) { + LL_PROFILE_ZONE_SCOPED if (color != mAmbientLightColor) { ++mLightHash; diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index fc35590667..3d49a9eb78 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -2501,12 +2501,7 @@ bool LLAppViewer::initConfiguration() #ifndef LL_RELEASE_FOR_DOWNLOAD // provide developer build only overrides for these control variables that are not // persisted to settings.xml - LLControlVariable* c = gSavedSettings.getControl("ShowConsoleWindow"); - if (c) - { - c->setValue(true, false); - } - c = gSavedSettings.getControl("AllowMultipleViewers"); + LLControlVariable* c = gSavedSettings.getControl("AllowMultipleViewers"); if (c) { c->setValue(true, false); diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp index a590237440..33842497d1 100644 --- a/indra/newview/llviewerdisplay.cpp +++ b/indra/newview/llviewerdisplay.cpp @@ -208,9 +208,11 @@ void display_update_camera() // Write some stats to LL_INFOS() void display_stats() { + LL_PROFILE_ZONE_SCOPED F32 fps_log_freq = gSavedSettings.getF32("FPSLogFrequency"); if (fps_log_freq > 0.f && gRecentFPSTime.getElapsedTimeF32() >= fps_log_freq) { + LL_PROFILE_ZONE_NAMED("DS - FPS"); F32 fps = gRecentFrameCount / fps_log_freq; LL_INFOS() << llformat("FPS: %.02f", fps) << LL_ENDL; gRecentFrameCount = 0; @@ -219,6 +221,7 @@ void display_stats() F32 mem_log_freq = gSavedSettings.getF32("MemoryLogFrequency"); if (mem_log_freq > 0.f && gRecentMemoryTime.getElapsedTimeF32() >= mem_log_freq) { + LL_PROFILE_ZONE_NAMED("DS - Memory"); gMemoryAllocated = U64Bytes(LLMemory::getCurrentRSS()); U32Megabytes memory = gMemoryAllocated; LL_INFOS() << "MEMORY: " << memory << LL_ENDL; @@ -228,6 +231,7 @@ void display_stats() F32 asset_storage_log_freq = gSavedSettings.getF32("AssetStorageLogFrequency"); if (asset_storage_log_freq > 0.f && gAssetStorageLogTime.getElapsedTimeF32() >= asset_storage_log_freq) { + LL_PROFILE_ZONE_NAMED("DS - Asset Storage"); gAssetStorageLogTime.reset(); gAssetStorage->logAssetStorageInfo(); } @@ -630,6 +634,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) if (!gDisconnected) { + LL_PROFILE_ZONE_NAMED("display - 1"); LLAppViewer::instance()->pingMainloopTimeout("Display:Update"); if (gPipeline.hasRenderType(LLPipeline::RENDER_TYPE_HUD)) { //don't draw hud objects in this frame @@ -722,6 +727,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) LLAppViewer::instance()->pingMainloopTimeout("Display:Swap"); { + LL_PROFILE_ZONE_NAMED("display - 2") if (gResizeScreenTexture) { gResizeScreenTexture = FALSE; @@ -777,6 +783,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) //if (!for_snapshot) { + LL_PROFILE_ZONE_NAMED("display - 3") LLAppViewer::instance()->pingMainloopTimeout("Display:Imagery"); gPipeline.generateWaterReflection(*LLViewerCamera::getInstance()); gPipeline.generateHighlight(*LLViewerCamera::getInstance()); @@ -825,7 +832,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) LLImageGL::deleteDeadTextures(); stop_glerror(); }*/ - } + } LLGLState::checkStates(); LLGLState::checkClientArrays(); @@ -840,6 +847,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) // LLAppViewer::instance()->pingMainloopTimeout("Display:StateSort"); { + LL_PROFILE_ZONE_NAMED("display - 3") LLViewerCamera::sCurCameraID = LLViewerCamera::CAMERA_WORLD; gPipeline.stateSort(*LLViewerCamera::getInstance(), result); stop_glerror(); @@ -948,6 +956,7 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) if (!(LLAppViewer::instance()->logoutRequestSent() && LLAppViewer::instance()->hasSavedFinalSnapshot()) && !gRestoreGL) { + LL_PROFILE_ZONE_NAMED("display - 4") LLViewerCamera::sCurCameraID = LLViewerCamera::CAMERA_WORLD; if (gSavedSettings.getBOOL("RenderDepthPrePass") && LLGLSLShader::sNoFixedFunction) diff --git a/indra/newview/llviewerwindow.cpp b/indra/newview/llviewerwindow.cpp index 1d13a306ef..2157585364 100644 --- a/indra/newview/llviewerwindow.cpp +++ b/indra/newview/llviewerwindow.cpp @@ -5280,6 +5280,7 @@ void LLViewerWindow::setup3DRender() void LLViewerWindow::setup3DViewport(S32 x_offset, S32 y_offset) { + LL_PROFILE_ZONE_SCOPED gGLViewport[0] = mWorldViewRectRaw.mLeft + x_offset; gGLViewport[1] = mWorldViewRectRaw.mBottom + y_offset; gGLViewport[2] = mWorldViewRectRaw.getWidth(); diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp index a1a1db35d6..c7b0a2bfb4 100644 --- a/indra/newview/llworld.cpp +++ b/indra/newview/llworld.cpp @@ -1284,6 +1284,7 @@ void send_agent_pause() void send_agent_resume() { + LL_PROFILE_ZONE_SCOPED // Note: used to check for LLWorld initialization before it became a singleton. // Rather than just remove this check I'm changing it to assure that the message // system has been initialized. -MG diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index 4e28c8c493..d7acf2ec0e 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -1897,6 +1897,7 @@ void LLPipeline::createObject(LLViewerObject* vobj) void LLPipeline::resetFrameStats() { + LL_PROFILE_ZONE_SCOPED assertInitialized(); sCompiles = 0; @@ -2125,6 +2126,7 @@ void LLPipeline::grabReferences(LLCullResult& result) void LLPipeline::clearReferences() { + LL_PROFILE_ZONE_SCOPED sCull = NULL; mGroupSaveQ1.clear(); } diff --git a/indra/test/test.cpp b/indra/test/test.cpp index 87c4a8d8a3..bb48216b2b 100644 --- a/indra/test/test.cpp +++ b/indra/test/test.cpp @@ -112,6 +112,7 @@ public: virtual void recordMessage(LLError::ELevel level, const std::string& message) { + LL_PROFILE_ZONE_SCOPED mFile << message << std::endl; } -- cgit v1.3 From 548bfda290b556d3ab29cc8c2f810f4cc349c9d8 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 28 Sep 2021 16:53:04 -0400 Subject: SL-16040: operator new() must never return nullptr. --- indra/llcommon/llcommon.cpp | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index da61e7539a..92f4d569b1 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -34,14 +34,13 @@ #include "llcleanup.h" #if (TRACY_ENABLE) -// Override new/delet for tracy memory profiling +// Override new/delete for tracy memory profiling void *operator new(size_t size) { auto ptr = (malloc) (size); if (!ptr) { throw std::bad_alloc(); - return nullptr; } TracyAlloc(ptr, size); return ptr; -- cgit v1.3 From db86ec9176dcbfabe5fddb3603da4132443f8b7f Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 28 Sep 2021 22:09:02 -0400 Subject: SL-16040: _aligned_malloc() and _aligned_free() are Microsoft only. Fortunately we already have platform-independent wrappers in llmemory.h. --- indra/llcommon/llcommon.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index 92f4d569b1..5d4a623bf6 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -61,7 +61,7 @@ void operator delete(void *ptr) noexcept void *tracy_aligned_malloc(size_t size, size_t alignment) { - auto ptr = (_aligned_malloc) (size, alignment); + auto ptr = ll_aligned_malloc_fallback(size, alignment); if (ptr) TracyAlloc(ptr, size); return ptr; } @@ -69,7 +69,7 @@ void *tracy_aligned_malloc(size_t size, size_t alignment) void tracy_aligned_free(void *memblock) { TracyFree(memblock); - (_aligned_free)(memblock); + ll_aligned_free_fallback(memblock); } #endif -- cgit v1.3 From 7c9aeed97d4ba3641971b9a1a92d334ec0adbb09 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Fri, 1 Oct 2021 16:05:23 -0400 Subject: SL-16024: Enhance LLThreadSafeQueue for use with WorkQueue. First, parameterize LLThreadSafeQueue's queue type. This allows us to substitute (e.g.) a std::priority_queue for a particular instance. Use std::queue for the default queue type, changing the operations invoked on the queue type from std::deque methods to std::queue methods. Rename published methods from (e.g.) pushFront() and popBack() to simple push() and pop(), retaining legacy names as aliases. Not only are the overt Front and Back unnecessary; they're the opposite of how std::queue uses std::deque or std::list, so they only confuse the reader. Break out tryPushUntil() method. We already use that logic internally to tryPushFor(), so it's just as easy to publish it as its own entry point. Add tryPopFor() and tryPopUntil() to allow limiting the time we'll wait for a queue item to become available. --- indra/llcommon/llthreadsafequeue.h | 229 ++++++++++++++++++++++++++----------- 1 file changed, 163 insertions(+), 66 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 26e0d71d31..04f51816d7 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -1,6 +1,6 @@ /** * @file llthreadsafequeue.h - * @brief Base classes for thread, mutex and condition handling. + * @brief Queue protected with mutexes for cross-thread use * * $LicenseInfo:firstyear=2004&license=viewerlgpl$ * Second Life Viewer Source Code @@ -27,15 +27,15 @@ #ifndef LL_LLTHREADSAFEQUEUE_H #define LL_LLTHREADSAFEQUEUE_H -#include "llexception.h" -#include -#include -#include -#include "mutex.h" #include "llcoros.h" #include LLCOROS_MUTEX_HEADER #include #include LLCOROS_CONDVAR_HEADER +#include "llexception.h" +#include "mutex.h" +#include +#include +#include // // A general queue exception. @@ -66,61 +66,95 @@ public: } }; -// -// Implements a thread safe FIFO. -// -template +/** + * Implements a thread safe FIFO. + */ +// Let the default std::queue default to underlying std::deque. Override if +// desired. +template> class LLThreadSafeQueue { public: typedef ElementT value_type; - + // If the pool is set to NULL one will be allocated and managed by this // queue. LLThreadSafeQueue(U32 capacity = 1024); - - // Add an element to the front of queue (will block if the queue has + + // Add an element to the queue (will block if the queue has // reached capacity). // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. - void pushFront(ElementT const & element); - - // Try to add an element to the front of queue without blocking. Returns + void push(ElementT const& element); + // legacy name + void pushFront(ElementT const & element) { return push(element); } + + // Try to add an element to the queue without blocking. Returns // true only if the element was actually added. - bool tryPushFront(ElementT const & element); + bool tryPush(ElementT const& element); + // legacy name + bool tryPushFront(ElementT const & element) { return tryPush(element); } - // Try to add an element to the front of queue, blocking if full but with - // timeout. Returns true if the element was added. + // Try to add an element to the queue, blocking if full but with timeout + // after specified duration. Returns true if the element was added. // There are potentially two different timeouts involved: how long to try // to lock the mutex, versus how long to wait for the queue to stop being // full. Careful settings for each timeout might be orders of magnitude // apart. However, this method conflates them. template + bool tryPushFor(const std::chrono::duration& timeout, + ElementT const & element); + // legacy name + template bool tryPushFrontFor(const std::chrono::duration& timeout, - ElementT const & element); + ElementT const & element) { return tryPushFor(timeout, element); } + + // Try to add an element to the queue, blocking if full but with + // timeout at specified time_point. Returns true if the element was added. + template + bool tryPushUntil(const std::chrono::time_point& timeout, + ElementT const& element); + // no legacy name because this is a newer method - // Pop the element at the end of the queue (will block if the queue is + // Pop the element at the head of the queue (will block if the queue is // empty). // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. - ElementT popBack(void); - - // Pop an element from the end of the queue if there is one available. + ElementT pop(void); + // legacy name + ElementT popBack(void) { return pop(); } + + // Pop an element from the head of the queue if there is one available. // Returns true only if an element was popped. - bool tryPopBack(ElementT & element); - + bool tryPop(ElementT & element); + // legacy name + bool tryPopBack(ElementT & element) { return tryPop(element); } + + // Pop the element at the head of the queue, blocking if empty, with + // timeout after specified duration. Returns true if an element was popped. + template + bool tryPopFor(const std::chrono::duration& timeout, ElementT& element); + // no legacy name because this is a newer method + + // Pop the element at the head of the queue, blocking if empty, with + // timeout at specified time_point. Returns true if an element was popped. + template + bool tryPopUntil(const std::chrono::time_point& timeout, + ElementT& element); + // no legacy name because this is a newer method + // Returns the size of the queue. size_t size(); // closes the queue: - // - every subsequent pushFront() call will throw LLThreadSafeQueueInterrupt - // - every subsequent tryPushFront() call will return false - // - popBack() calls will return normally until the queue is drained, then - // every subsequent popBack() will throw LLThreadSafeQueueInterrupt - // - tryPopBack() calls will return normally until the queue is drained, - // then every subsequent tryPopBack() call will return false + // - every subsequent push() call will throw LLThreadSafeQueueInterrupt + // - every subsequent tryPush() call will return false + // - pop() calls will return normally until the queue is drained, then + // every subsequent pop() will throw LLThreadSafeQueueInterrupt + // - tryPop() calls will return normally until the queue is drained, + // then every subsequent tryPop() call will return false void close(); // detect closed state @@ -128,8 +162,9 @@ public: // inverse of isClosed() explicit operator bool(); -private: - std::deque< ElementT > mStorage; +protected: + typedef QueueT queue_type; + QueueT mStorage; U32 mCapacity; bool mClosed; @@ -142,16 +177,16 @@ private: // LLThreadSafeQueue //----------------------------------------------------------------------------- -template -LLThreadSafeQueue::LLThreadSafeQueue(U32 capacity) : +template +LLThreadSafeQueue::LLThreadSafeQueue(U32 capacity) : mCapacity(capacity), mClosed(false) { } -template -void LLThreadSafeQueue::pushFront(ElementT const & element) +template +void LLThreadSafeQueue::push(ElementT const & element) { lock_t lock1(mLock); while (true) @@ -163,7 +198,7 @@ void LLThreadSafeQueue::pushFront(ElementT const & element) if (mStorage.size() < mCapacity) { - mStorage.push_front(element); + mStorage.push(element); lock1.unlock(); mEmptyCond.notify_one(); return; @@ -175,15 +210,24 @@ void LLThreadSafeQueue::pushFront(ElementT const & element) } -template +template template -bool LLThreadSafeQueue::tryPushFrontFor(const std::chrono::duration& timeout, - ElementT const & element) +bool LLThreadSafeQueue::tryPushFor( + const std::chrono::duration& timeout, + ElementT const & element) { // Convert duration to time_point: passing the same timeout duration to // each of multiple calls is wrong. - auto endpoint = std::chrono::steady_clock::now() + timeout; + return tryPushUntil(std::chrono::steady_clock::now() + timeout, element); +} + +template +template +bool LLThreadSafeQueue::tryPushUntil( + const std::chrono::time_point& endpoint, + ElementT const& element) +{ lock_t lock1(mLock, std::defer_lock); if (!lock1.try_lock_until(endpoint)) return false; @@ -197,7 +241,7 @@ bool LLThreadSafeQueue::tryPushFrontFor(const std::chrono::duration::tryPushFrontFor(const std::chrono::duration -bool LLThreadSafeQueue::tryPushFront(ElementT const & element) +template +bool LLThreadSafeQueue::tryPush(ElementT const & element) { lock_t lock1(mLock, std::defer_lock); if (!lock1.try_lock()) @@ -228,23 +272,24 @@ bool LLThreadSafeQueue::tryPushFront(ElementT const & element) if (mStorage.size() >= mCapacity) return false; - mStorage.push_front(element); + mStorage.push(element); lock1.unlock(); mEmptyCond.notify_one(); return true; } -template -ElementT LLThreadSafeQueue::popBack(void) +template +ElementT LLThreadSafeQueue::pop(void) { lock_t lock1(mLock); while (true) { if (!mStorage.empty()) { - ElementT value = mStorage.back(); - mStorage.pop_back(); + // std::queue::front() is the element about to pop() + ElementT value = mStorage.front(); + mStorage.pop(); lock1.unlock(); mCapacityCond.notify_one(); return value; @@ -261,54 +306,106 @@ ElementT LLThreadSafeQueue::popBack(void) } -template -bool LLThreadSafeQueue::tryPopBack(ElementT & element) +template +bool LLThreadSafeQueue::tryPop(ElementT & element) { lock_t lock1(mLock, std::defer_lock); if (!lock1.try_lock()) return false; - // no need to check mClosed: tryPopBack() behavior when the queue is + // no need to check mClosed: tryPop() behavior when the queue is // closed is implemented by simple inability to push any new elements if (mStorage.empty()) return false; - element = mStorage.back(); - mStorage.pop_back(); + // std::queue::front() is the element about to pop() + element = mStorage.front(); + mStorage.pop(); lock1.unlock(); mCapacityCond.notify_one(); return true; } -template -size_t LLThreadSafeQueue::size(void) +template +template +bool LLThreadSafeQueue::tryPopFor( + const std::chrono::duration& timeout, + ElementT& element) +{ + // Convert duration to time_point: passing the same timeout duration to + // each of multiple calls is wrong. + return tryPopUntil(std::chrono::steady_clock::now() + timeout, element); +} + + +template +template +bool LLThreadSafeQueue::tryPopUntil( + const std::chrono::time_point& endpoint, + ElementT& element) +{ + lock_t lock1(mLock, std::defer_lock); + if (!lock1.try_lock_until(endpoint)) + return false; + + while (true) + { + if (!mStorage.empty()) + { + // std::queue::front() is the element about to pop() + element = mStorage.front(); + mStorage.pop(); + lock1.unlock(); + mCapacityCond.notify_one(); + return true; + } + + if (mClosed) + { + return false; + } + + // Storage empty. Wait for signal. + if (LLCoros::cv_status::timeout == mEmptyCond.wait_until(lock1, endpoint)) + { + // timed out -- formally we might recheck both conditions above + return false; + } + // If we didn't time out, we were notified for some reason. Loop back + // to check. + } +} + + +template +size_t LLThreadSafeQueue::size(void) { lock_t lock(mLock); return mStorage.size(); } -template -void LLThreadSafeQueue::close() +template +void LLThreadSafeQueue::close() { lock_t lock(mLock); mClosed = true; lock.unlock(); - // wake up any blocked popBack() calls + // wake up any blocked pop() calls mEmptyCond.notify_all(); - // wake up any blocked pushFront() calls + // wake up any blocked push() calls mCapacityCond.notify_all(); } -template -bool LLThreadSafeQueue::isClosed() +template +bool LLThreadSafeQueue::isClosed() { lock_t lock(mLock); return mClosed && mStorage.size() == 0; } -template -LLThreadSafeQueue::operator bool() +template +LLThreadSafeQueue::operator bool() { return ! isClosed(); } -- cgit v1.3 From 1b1ebdf183e50c6a751493570ee6e643c33c4eda Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Mon, 4 Oct 2021 11:48:58 -0400 Subject: SL-16024: Introduce tuple.h with tuple_cons(), tuple_cdr(). These functions allow prepending or removing an item at the left end of an arbitrary tuple -- for instance, to add a sequence key to a caller's data, then remove it again when delivering the original tuple. --- indra/llcommon/CMakeLists.txt | 2 + indra/llcommon/tests/tuple_test.cpp | 47 +++++++++++++++++++++ indra/llcommon/tuple.h | 84 +++++++++++++++++++++++++++++++++++++ 3 files changed, 133 insertions(+) create mode 100644 indra/llcommon/tests/tuple_test.cpp create mode 100644 indra/llcommon/tuple.h (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index dd266630ea..6558219462 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -254,6 +254,7 @@ set(llcommon_HEADER_FILES stdtypes.h stringize.h timer.h + tuple.h u64.h StackWalker.h ) @@ -358,6 +359,7 @@ if (LL_TESTS) LL_ADD_INTEGRATION_TEST(lluri "" "${test_libs}") LL_ADD_INTEGRATION_TEST(llunits "" "${test_libs}") LL_ADD_INTEGRATION_TEST(stringize "" "${test_libs}") + LL_ADD_INTEGRATION_TEST(tuple "" "${test_libs}") ## llexception_test.cpp isn't a regression test, and doesn't need to be run ## every build. It's to help a developer make implementation choices about diff --git a/indra/llcommon/tests/tuple_test.cpp b/indra/llcommon/tests/tuple_test.cpp new file mode 100644 index 0000000000..af94e2086c --- /dev/null +++ b/indra/llcommon/tests/tuple_test.cpp @@ -0,0 +1,47 @@ +/** + * @file tuple_test.cpp + * @author Nat Goodspeed + * @date 2021-10-04 + * @brief Test for tuple. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "tuple.h" +// STL headers +// std headers +// external library headers +// other Linden headers +#include "../test/lltut.h" + +/***************************************************************************** +* TUT +*****************************************************************************/ +namespace tut +{ + struct tuple_data + { + }; + typedef test_group tuple_group; + typedef tuple_group::object object; + tuple_group tuplegrp("tuple"); + + template<> template<> + void object::test<1>() + { + set_test_name("tuple"); + std::tuple tup{ "abc", 17 }; + std::tuple ptup{ tuple_cons(34, tup) }; + std::tuple tup2; + int i; + std::tie(i, tup2) = tuple_split(ptup); + ensure_equals("tuple_car() fail", i, 34); + ensure_equals("tuple_cdr() (0) fail", std::get<0>(tup2), "abc"); + ensure_equals("tuple_cdr() (1) fail", std::get<1>(tup2), 17); + } +} // namespace tut diff --git a/indra/llcommon/tuple.h b/indra/llcommon/tuple.h new file mode 100644 index 0000000000..bfe7e3c2ba --- /dev/null +++ b/indra/llcommon/tuple.h @@ -0,0 +1,84 @@ +/** + * @file tuple.h + * @author Nat Goodspeed + * @date 2021-10-04 + * @brief A couple tuple utilities + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_TUPLE_H) +#define LL_TUPLE_H + +#include +#include // std::remove_reference +#include // std::pair + +/** + * tuple_cons() behaves like LISP cons: it uses std::tuple_cat() to prepend a + * new item of arbitrary type to an existing std::tuple. + */ +template > +auto tuple_cons(First&& first, Tuple_&& rest) +{ + // All we need to do is make a tuple containing 'first', and let + // tuple_cat() do the hard part. + return std::tuple_cat(std::tuple(std::forward(first)), + std::forward(rest)); +} + +/** + * tuple_car() behaves like LISP car: it extracts the first item from a + * std::tuple. + */ +template > +auto tuple_car(Tuple_&& tuple) +{ + return std::get<0>(std::forward(tuple)); +} + +/** + * tuple_cdr() behaves like LISP cdr: it returns a new tuple containing + * everything BUT the first item. + */ +// derived from https://stackoverflow.com/a/24046437 +template +auto tuple_cdr_(Tuple&& tuple, const std::index_sequence) +{ + // Given an index sequence from [0..N-1), extract tuple items [1..N) + return std::make_tuple(std::get(std::forward(tuple))...); +} + +template +auto tuple_cdr(Tuple&& tuple) +{ + return tuple_cdr_( + std::forward(tuple), + // Pass helper function an index sequence one item shorter than tuple + std::make_index_sequence< + std::tuple_size< + // tuple_size doesn't like reference types + typename std::remove_reference::type + >::value - 1u> + ()); +} + +/** + * tuple_split(), the opposite of tuple_cons(), has no direct analog in LISP. + * It returns a std::pair of tuple_car(), tuple_cdr(). We could call this + * function tuple_car_cdr(), or tuple_slice() or some such. But tuple_split() + * feels more descriptive. + */ +template > +auto tuple_split(Tuple_&& tuple) +{ + // We're not really worried about forwarding multiple times a tuple that + // might contain move-only items, because the implementation above only + // applies std::get() exactly once to each item. + return std::make_pair(tuple_car(std::forward(tuple)), + tuple_cdr(std::forward(tuple))); +} + +#endif /* ! defined(LL_TUPLE_H) */ -- cgit v1.3 From ca60fbe72ce086fbdf0821043ad3be6aad06857c Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Mon, 4 Oct 2021 16:19:59 -0400 Subject: SL-16024: LLThreadSafeQueue enhancements Add LL::PriorityQueueAdapter, a wrapper for std::priority_queue to make its API more closely resemble std::queue for drop-in use as LLThreadSafeQueue's underlying QueueT container. Support move-only element types. Factor out some implementation redundancy: wrap actual push semantics as push_(), actual pop semantics as pop_(). push(), tryPush() and tryPushUntil() now call push_(); pop(), tryPop() and tryPopUntil() now call pop_(). Break out tryLock() and tryLockUntil() methods that, if they can lock, run the passed callable. Then tryPush(), tryPushUntil(), tryPop() and tryPopUntil() pass lambdas containing the meat of the original method body to tryLock() or tryLockUntil(), as appropriate. --- indra/llcommon/llthreadsafequeue.h | 358 ++++++++++++++++++++++++------------- 1 file changed, 235 insertions(+), 123 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 04f51816d7..c57520c01f 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -37,6 +37,9 @@ #include #include +/***************************************************************************** +* LLThreadSafeQueue +*****************************************************************************/ // // A general queue exception. // @@ -77,8 +80,8 @@ class LLThreadSafeQueue public: typedef ElementT value_type; - // If the pool is set to NULL one will be allocated and managed by this - // queue. + // Limiting the number of pending items prevents unbounded growth of the + // underlying queue. LLThreadSafeQueue(U32 capacity = 1024); // Add an element to the queue (will block if the queue has @@ -86,13 +89,15 @@ public: // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. - void push(ElementT const& element); + template + void push(T&& element); // legacy name void pushFront(ElementT const & element) { return push(element); } // Try to add an element to the queue without blocking. Returns // true only if the element was actually added. - bool tryPush(ElementT const& element); + template + bool tryPush(T&& element); // legacy name bool tryPushFront(ElementT const & element) { return tryPush(element); } @@ -102,9 +107,9 @@ public: // to lock the mutex, versus how long to wait for the queue to stop being // full. Careful settings for each timeout might be orders of magnitude // apart. However, this method conflates them. - template + template bool tryPushFor(const std::chrono::duration& timeout, - ElementT const & element); + T&& element); // legacy name template bool tryPushFrontFor(const std::chrono::duration& timeout, @@ -112,9 +117,9 @@ public: // Try to add an element to the queue, blocking if full but with // timeout at specified time_point. Returns true if the element was added. - template - bool tryPushUntil(const std::chrono::time_point& timeout, - ElementT const& element); + template + bool tryPushUntil(const std::chrono::time_point& until, + T&& element); // no legacy name because this is a newer method // Pop the element at the head of the queue (will block if the queue is @@ -141,7 +146,7 @@ public: // Pop the element at the head of the queue, blocking if empty, with // timeout at specified time_point. Returns true if an element was popped. template - bool tryPopUntil(const std::chrono::time_point& timeout, + bool tryPopUntil(const std::chrono::time_point& until, ElementT& element); // no legacy name because this is a newer method @@ -172,11 +177,74 @@ protected: typedef std::unique_lock lock_t; boost::fibers::condition_variable_any mCapacityCond; boost::fibers::condition_variable_any mEmptyCond; -}; -// LLThreadSafeQueue -//----------------------------------------------------------------------------- + // if we're able to lock immediately, do so and run the passed callable, + // which must accept lock_t& and return bool + template + bool tryLock(CALLABLE&& callable); + // if we're able to lock before the passed time_point, do so and run the + // passed callable, which must accept lock_t& and return bool + template + bool tryLockUntil(const std::chrono::time_point& until, + CALLABLE&& callable); + // while lock is locked, really push the passed element, if we can + template + bool push_(lock_t& lock, T&& element); + // while lock is locked, really pop the head element, if we can + template + bool pop_(lock_t& lock, ElementT& element, + PRED&& pred=[](const ElementT&){ return true; }); +}; +/***************************************************************************** +* PriorityQueueAdapter +*****************************************************************************/ +namespace LL +{ + /** + * std::priority_queue's API is almost like std::queue, intentionally of + * course, but you must access the element about to pop() as top() rather + * than as front(). Make an adapter for use with LLThreadSafeQueue. + */ + template , + typename Compare=std::less> + class PriorityQueueAdapter + { + public: + // publish all the same types + typedef std::priority_queue queue_type; + typedef typename queue_type::container_type container_type; + typedef typename queue_type::value_compare value_compare; + typedef typename queue_type::value_type value_type; + typedef typename queue_type::size_type size_type; + typedef typename queue_type::reference reference; + typedef typename queue_type::const_reference const_reference; + + // Although std::queue defines both const and non-const front() + // methods, std::priority_queue defines only const top(). + const_reference front() const { return mQ.top(); } + // std::priority_queue has no equivalent to back(), so it's good that + // LLThreadSafeQueue doesn't use it. + + // All the rest of these merely forward to the corresponding + // queue_type methods. + bool empty() const { return mQ.empty(); } + size_type size() const { return mQ.size(); } + void push(const value_type& value) { mQ.push(value); } + void push(value_type&& value) { mQ.push(std::move(value)); } + template + void emplace(Args&&... args) { mQ.emplace(std::forward(args)...); } + void pop() { mQ.pop(); } + + private: + queue_type mQ; + }; +} // namespace LL + + +/***************************************************************************** +* LLThreadSafeQueue implementation +*****************************************************************************/ template LLThreadSafeQueue::LLThreadSafeQueue(U32 capacity) : mCapacity(capacity), @@ -185,24 +253,69 @@ LLThreadSafeQueue::LLThreadSafeQueue(U32 capacity) : } -template -void LLThreadSafeQueue::push(ElementT const & element) +// if we're able to lock immediately, do so and run the passed callable, which +// must accept lock_t& and return bool +template +template +bool LLThreadSafeQueue::tryLock(CALLABLE&& callable) +{ + lock_t lock1(mLock, std::defer_lock); + if (!lock1.try_lock()) + return false; + + return std::forward(callable)(lock1); +} + + +// if we're able to lock before the passed time_point, do so and run the +// passed callable, which must accept lock_t& and return bool +template +template +bool LLThreadSafeQueue::tryLockUntil( + const std::chrono::time_point& until, + CALLABLE&& callable) +{ + lock_t lock1(mLock, std::defer_lock); + if (!lock1.try_lock_until(until)) + return false; + + return std::forward(callable)(lock1); +} + + +// while lock is locked, really push the passed element, if we can +template +template +bool LLThreadSafeQueue::push_(lock_t& lock, T&& element) +{ + if (mStorage.size() >= mCapacity) + return false; + + mStorage.push(std::forward(element)); + lock.unlock(); + // now that we've pushed, if somebody's been waiting to pop, signal them + mEmptyCond.notify_one(); + return true; +} + + +template +template +void LLThreadSafeQueue::push(T&& element) { lock_t lock1(mLock); while (true) { + // On the producer side, it doesn't matter whether the queue has been + // drained or not: the moment either end calls close(), further push() + // operations will fail. if (mClosed) { LLTHROW(LLThreadSafeQueueInterrupt()); } - if (mStorage.size() < mCapacity) - { - mStorage.push(element); - lock1.unlock(); - mEmptyCond.notify_one(); + if (push_(lock1, std::forward(element))) return; - } // Storage Full. Wait for signal. mCapacityCond.wait(lock1); @@ -210,71 +323,85 @@ void LLThreadSafeQueue::push(ElementT const & element) } +template +template +bool LLThreadSafeQueue::tryPush(T&& element) +{ + return tryLock( + [this, element=std::move(element)](lock_t& lock) + { + if (mClosed) + return false; + return push_(lock, std::move(element)); + }); +} + + template -template +template bool LLThreadSafeQueue::tryPushFor( const std::chrono::duration& timeout, - ElementT const & element) + T&& element) { // Convert duration to time_point: passing the same timeout duration to // each of multiple calls is wrong. - return tryPushUntil(std::chrono::steady_clock::now() + timeout, element); + return tryPushUntil(std::chrono::steady_clock::now() + timeout, + std::forward(element)); } template -template +template bool LLThreadSafeQueue::tryPushUntil( - const std::chrono::time_point& endpoint, - ElementT const& element) + const std::chrono::time_point& until, + T&& element) { - lock_t lock1(mLock, std::defer_lock); - if (!lock1.try_lock_until(endpoint)) - return false; - - while (true) - { - if (mClosed) + return tryLockUntil( + until, + [this, until, element=std::move(element)](lock_t& lock) { - return false; - } - - if (mStorage.size() < mCapacity) - { - mStorage.push(element); - lock1.unlock(); - mEmptyCond.notify_one(); - return true; - } - - // Storage Full. Wait for signal. - if (LLCoros::cv_status::timeout == mCapacityCond.wait_until(lock1, endpoint)) - { - // timed out -- formally we might recheck both conditions above - return false; - } - // If we didn't time out, we were notified for some reason. Loop back - // to check. - } + while (true) + { + if (mClosed) + { + return false; + } + + if (push_(lock, std::move(element))) + return true; + + // Storage Full. Wait for signal. + if (LLCoros::cv_status::timeout == mCapacityCond.wait_until(lock, until)) + { + // timed out -- formally we might recheck both conditions above + return false; + } + // If we didn't time out, we were notified for some reason. Loop back + // to check. + } + }); } -template -bool LLThreadSafeQueue::tryPush(ElementT const & element) +// while lock is locked, really pop the head element, if we can +template +template +bool LLThreadSafeQueue::pop_( + lock_t& lock, ElementT& element, PRED&& pred) { - lock_t lock1(mLock, std::defer_lock); - if (!lock1.try_lock()) - return false; - - if (mClosed) - return false; - - if (mStorage.size() >= mCapacity) + // If mStorage is empty, there's no head element. + // If there's a head element, pass it to the predicate to see if caller + // considers it ready to pop. + // Unless both are satisfied, no point in continuing. + if (mStorage.empty() || ! std::forward(pred)(mStorage.front())) return false; - mStorage.push(element); - lock1.unlock(); - mEmptyCond.notify_one(); + // std::queue::front() is the element about to pop() + element = mStorage.front(); + mStorage.pop(); + lock.unlock(); + // now that we've popped, if somebody's been waiting to push, signal them + mCapacityCond.notify_one(); return true; } @@ -285,22 +412,20 @@ ElementT LLThreadSafeQueue::pop(void) lock_t lock1(mLock); while (true) { - if (!mStorage.empty()) - { - // std::queue::front() is the element about to pop() - ElementT value = mStorage.front(); - mStorage.pop(); - lock1.unlock(); - mCapacityCond.notify_one(); - return value; - } - + // On the consumer side, we always try to pop before checking mClosed + // so we can finish draining the queue. + ElementT value; + if (pop_(lock1, value)) + return std::move(value); + + // Once the queue is empty, mClosed lets us know if there will ever be + // any more coming. if (mClosed) { LLTHROW(LLThreadSafeQueueInterrupt()); } - // Storage empty. Wait for signal. + // Storage empty, queue still open. Wait for signal. mEmptyCond.wait(lock1); } } @@ -309,21 +434,14 @@ ElementT LLThreadSafeQueue::pop(void) template bool LLThreadSafeQueue::tryPop(ElementT & element) { - lock_t lock1(mLock, std::defer_lock); - if (!lock1.try_lock()) - return false; - - // no need to check mClosed: tryPop() behavior when the queue is - // closed is implemented by simple inability to push any new elements - if (mStorage.empty()) - return false; - - // std::queue::front() is the element about to pop() - element = mStorage.front(); - mStorage.pop(); - lock1.unlock(); - mCapacityCond.notify_one(); - return true; + return tryLock( + [this, &element](lock_t& lock) + { + // no need to check mClosed: tryPop() behavior when the queue is + // closed is implemented by simple inability to push any new + // elements + return pop_(lock, element); + }); } @@ -342,39 +460,33 @@ bool LLThreadSafeQueue::tryPopFor( template template bool LLThreadSafeQueue::tryPopUntil( - const std::chrono::time_point& endpoint, + const std::chrono::time_point& until, ElementT& element) { - lock_t lock1(mLock, std::defer_lock); - if (!lock1.try_lock_until(endpoint)) - return false; - - while (true) - { - if (!mStorage.empty()) + return tryLockUntil( + until, + [this, until, &element](lock_t& lock) { - // std::queue::front() is the element about to pop() - element = mStorage.front(); - mStorage.pop(); - lock1.unlock(); - mCapacityCond.notify_one(); - return true; - } - - if (mClosed) - { - return false; - } - - // Storage empty. Wait for signal. - if (LLCoros::cv_status::timeout == mEmptyCond.wait_until(lock1, endpoint)) - { - // timed out -- formally we might recheck both conditions above - return false; - } - // If we didn't time out, we were notified for some reason. Loop back - // to check. - } + while (true) + { + if (pop_(lock, element)) + return true; + + if (mClosed) + { + return false; + } + + // Storage empty. Wait for signal. + if (LLCoros::cv_status::timeout == mEmptyCond.wait_until(lock, until)) + { + // timed out -- formally we might recheck both conditions above + return false; + } + // If we didn't time out, we were notified for some reason. Loop back + // to check. + } + }); } -- cgit v1.3 From a35e266547e4d2c8dbd6b003c64b719d91eaaf87 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Mon, 4 Oct 2021 17:21:39 -0400 Subject: SL-16024: Don't use a lambda as default arg for universal reference. Instead, break out a separate pop_() method that explicitly provides the lambda to the real pop_() implementation. --- indra/llcommon/llthreadsafequeue.h | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index c57520c01f..1dffad6b89 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -191,9 +191,11 @@ protected: template bool push_(lock_t& lock, T&& element); // while lock is locked, really pop the head element, if we can + bool pop_(lock_t& lock, ElementT& element); + // pop_() with an explicit predicate indicating whether the head element + // is ready to be popped template - bool pop_(lock_t& lock, ElementT& element, - PRED&& pred=[](const ElementT&){ return true; }); + bool pop_(lock_t& lock, ElementT& element, PRED&& pred); }; /***************************************************************************** @@ -385,6 +387,16 @@ bool LLThreadSafeQueue::tryPushUntil( // while lock is locked, really pop the head element, if we can template +bool LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) +{ + // default predicate: head element, if present, is always ready to pop + return pop_(lock, element, [](const ElementT&){ return true; }); +} + + +// pop_() with an explicit predicate indicating whether the head element +// is ready to be popped +template template bool LLThreadSafeQueue::pop_( lock_t& lock, ElementT& element, PRED&& pred) -- cgit v1.3 From 955b967623983cb50ba09f7b82e5f01f2c6bcebb Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 5 Oct 2021 17:31:53 -0400 Subject: SL-16024: Add ThreadSafeSchedule, a timestamped LLThreadSafeQueue. ThreadSafeSchedule orders its items by timestamp, which can be passed either implicitly or explicitly. The timestamp specifies earliest delivery time: an item cannot be popped until that time. Add initial tests. Tweak the LLThreadSafeQueue base class to support ThreadSafeSchedule: introduce virtual canPop() method to report whether the current head item is available to pop. The base class unconditionally says yes, ThreadSafeSchedule says it depends on whether its timestamp is still in the future. This replaces the protected pop_() overload accepting a predicate. Rather than explicitly passing a predicate through a couple levels of function call, use canPop() at the level it matters. Runtime behavior that varies depending on an object's leaf class is what virtual functions were invented for. Give pop_() a three-state enum return so pop() can distinguish between "closed and empty" (throws exception) versus "closed, not yet drained because we're not yet ready to pop the head item" (waits). Also break out protected tryPopUntil_() method, the body logic of tryPopUntil(). The public method locks the data structure, the protected method requires that its caller has already done so. Add chrono.h with a more full-featured LL::time_point_cast() function than the one found in , which only converts between time_point durations, not between time_points based on different clocks. --- indra/llcommon/CMakeLists.txt | 3 + indra/llcommon/chrono.h | 65 +++++ indra/llcommon/llthreadsafequeue.h | 121 ++++---- indra/llcommon/tests/threadsafeschedule_test.cpp | 65 +++++ indra/llcommon/threadsafeschedule.h | 334 +++++++++++++++++++++++ 5 files changed, 535 insertions(+), 53 deletions(-) create mode 100644 indra/llcommon/chrono.h create mode 100644 indra/llcommon/tests/threadsafeschedule_test.cpp create mode 100644 indra/llcommon/threadsafeschedule.h (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 6558219462..5efcfabf24 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -127,6 +127,7 @@ set(llcommon_SOURCE_FILES set(llcommon_HEADER_FILES CMakeLists.txt + chrono.h ctype_workaround.h fix_macros.h indra_constants.h @@ -253,6 +254,7 @@ set(llcommon_HEADER_FILES lockstatic.h stdtypes.h stringize.h + threadsafeschedule.h timer.h tuple.h u64.h @@ -359,6 +361,7 @@ if (LL_TESTS) LL_ADD_INTEGRATION_TEST(lluri "" "${test_libs}") LL_ADD_INTEGRATION_TEST(llunits "" "${test_libs}") LL_ADD_INTEGRATION_TEST(stringize "" "${test_libs}") + LL_ADD_INTEGRATION_TEST(threadsafeschedule "" "${test_libs}") LL_ADD_INTEGRATION_TEST(tuple "" "${test_libs}") ## llexception_test.cpp isn't a regression test, and doesn't need to be run diff --git a/indra/llcommon/chrono.h b/indra/llcommon/chrono.h new file mode 100644 index 0000000000..806e871892 --- /dev/null +++ b/indra/llcommon/chrono.h @@ -0,0 +1,65 @@ +/** + * @file chrono.h + * @author Nat Goodspeed + * @date 2021-10-05 + * @brief supplement with utility functions + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_CHRONO_H) +#define LL_CHRONO_H + +#include +#include // std::enable_if + +namespace LL +{ + +// time_point_cast() is derived from https://stackoverflow.com/a/35293183 +// without the iteration: we think errors in the ~1 microsecond range are +// probably acceptable. + +// This variant is for the optimal case when the source and dest use the same +// clock: that case is handled by std::chrono. +template ::value, + bool>::type = true> +DestTimePoint time_point_cast(const SrcTimePoint& time) +{ + return std::chrono::time_point_cast(time); +} + +// This variant is for when the source and dest use different clocks -- see +// the linked StackOverflow answer, also Howard Hinnant's, for more context. +template ::value, + bool>::type = true> +DestTimePoint time_point_cast(const SrcTimePoint& time) +{ + // The basic idea is that we must adjust the passed time_point by the + // difference between the clocks' epochs. But since time_point doesn't + // expose its epoch, we fall back on what each of them thinks is now(). + // However, since we necessarily make sequential calls to those now() + // functions, the answers differ not only by the cycles spent executing + // those calls, but by potential OS interruptions between them. Try to + // reduce that error by capturing the source clock time both before and + // after the dest clock, and splitting the difference. Of course an + // interruption between two of these now() calls without a comparable + // interruption between the other two will skew the result, but better is + // more expensive. + const auto src_before = typename SrcTimePoint::clock::now(); + const auto dest_now = typename DestTimePoint::clock::now(); + const auto src_after = typename SrcTimePoint::clock::now(); + const auto src_diff = src_after - src_before; + const auto src_now = src_before + src_diff / 2; + return dest_now + (time - src_now); +} + +} // namespace LL + +#endif /* ! defined(LL_CHRONO_H) */ diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 1dffad6b89..bd2d82d4c3 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -83,6 +83,7 @@ public: // Limiting the number of pending items prevents unbounded growth of the // underlying queue. LLThreadSafeQueue(U32 capacity = 1024); + virtual ~LLThreadSafeQueue() {} // Add an element to the queue (will block if the queue has // reached capacity). @@ -162,10 +163,10 @@ public: // then every subsequent tryPop() call will return false void close(); - // detect closed state + // producer end: are we prevented from pushing any additional items? bool isClosed(); - // inverse of isClosed() - explicit operator bool(); + // consumer end: are we done, is the queue entirely drained? + bool done(); protected: typedef QueueT queue_type; @@ -178,6 +179,11 @@ protected: boost::fibers::condition_variable_any mCapacityCond; boost::fibers::condition_variable_any mEmptyCond; + // implementation logic, suitable for passing to tryLockUntil() + template + bool tryPopUntil_(lock_t& lock, + const std::chrono::time_point& until, + ElementT& element); // if we're able to lock immediately, do so and run the passed callable, // which must accept lock_t& and return bool template @@ -191,11 +197,11 @@ protected: template bool push_(lock_t& lock, T&& element); // while lock is locked, really pop the head element, if we can - bool pop_(lock_t& lock, ElementT& element); - // pop_() with an explicit predicate indicating whether the head element - // is ready to be popped - template - bool pop_(lock_t& lock, ElementT& element, PRED&& pred); + enum pop_result { EMPTY, WAITING, POPPED }; + pop_result pop_(lock_t& lock, ElementT& element); + // Is the current head element ready to pop? We say yes; subclass can + // override as needed. + virtual bool canPop(const ElementT& head) const { return true; } }; /***************************************************************************** @@ -387,26 +393,16 @@ bool LLThreadSafeQueue::tryPushUntil( // while lock is locked, really pop the head element, if we can template -bool LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) -{ - // default predicate: head element, if present, is always ready to pop - return pop_(lock, element, [](const ElementT&){ return true; }); -} - - -// pop_() with an explicit predicate indicating whether the head element -// is ready to be popped -template -template -bool LLThreadSafeQueue::pop_( - lock_t& lock, ElementT& element, PRED&& pred) +typename LLThreadSafeQueue::pop_result +LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) { // If mStorage is empty, there's no head element. - // If there's a head element, pass it to the predicate to see if caller - // considers it ready to pop. - // Unless both are satisfied, no point in continuing. - if (mStorage.empty() || ! std::forward(pred)(mStorage.front())) - return false; + if (mStorage.empty()) + return EMPTY; + + // If there's a head element, pass it to canPop() to see if it's ready to pop. + if (! canPop(mStorage.front())) + return WAITING; // std::queue::front() is the element about to pop() element = mStorage.front(); @@ -414,7 +410,7 @@ bool LLThreadSafeQueue::pop_( lock.unlock(); // now that we've popped, if somebody's been waiting to push, signal them mCapacityCond.notify_one(); - return true; + return POPPED; } @@ -422,17 +418,20 @@ template ElementT LLThreadSafeQueue::pop(void) { lock_t lock1(mLock); + ElementT value; while (true) { // On the consumer side, we always try to pop before checking mClosed // so we can finish draining the queue. - ElementT value; - if (pop_(lock1, value)) + pop_result popped = pop_(lock1, value); + if (popped == POPPED) return std::move(value); // Once the queue is empty, mClosed lets us know if there will ever be - // any more coming. - if (mClosed) + // any more coming. If we didn't pop because WAITING, i.e. canPop() + // returned false, then even if the producer end has been closed, + // there's still at least one item to drain: wait for it. + if (popped == EMPTY && mClosed) { LLTHROW(LLThreadSafeQueueInterrupt()); } @@ -452,7 +451,7 @@ bool LLThreadSafeQueue::tryPop(ElementT & element) // no need to check mClosed: tryPop() behavior when the queue is // closed is implemented by simple inability to push any new // elements - return pop_(lock, element); + return pop_(lock, element) == POPPED; }); } @@ -479,26 +478,38 @@ bool LLThreadSafeQueue::tryPopUntil( until, [this, until, &element](lock_t& lock) { - while (true) - { - if (pop_(lock, element)) - return true; + return tryPopUntil_(lock, until, element); + }); +} - if (mClosed) - { - return false; - } - // Storage empty. Wait for signal. - if (LLCoros::cv_status::timeout == mEmptyCond.wait_until(lock, until)) - { - // timed out -- formally we might recheck both conditions above - return false; - } - // If we didn't time out, we were notified for some reason. Loop back - // to check. - } - }); +// body of tryPopUntil(), called once we have the lock +template +template +bool LLThreadSafeQueue::tryPopUntil_( + lock_t& lock, + const std::chrono::time_point& until, + ElementT& element) +{ + while (true) + { + if (pop_(lock, element) == POPPED) + return true; + + if (mClosed) + { + return false; + } + + // Storage empty. Wait for signal. + if (LLCoros::cv_status::timeout == mEmptyCond.wait_until(lock, until)) + { + // timed out -- formally we might recheck both conditions above + return false; + } + // If we didn't time out, we were notified for some reason. Loop back + // to check. + } } @@ -509,6 +520,7 @@ size_t LLThreadSafeQueue::size(void) return mStorage.size(); } + template void LLThreadSafeQueue::close() { @@ -521,17 +533,20 @@ void LLThreadSafeQueue::close() mCapacityCond.notify_all(); } + template bool LLThreadSafeQueue::isClosed() { lock_t lock(mLock); - return mClosed && mStorage.size() == 0; + return mClosed; } + template -LLThreadSafeQueue::operator bool() +bool LLThreadSafeQueue::done() { - return ! isClosed(); + lock_t lock(mLock); + return mClosed && mStorage.size() == 0; } #endif diff --git a/indra/llcommon/tests/threadsafeschedule_test.cpp b/indra/llcommon/tests/threadsafeschedule_test.cpp new file mode 100644 index 0000000000..ec0fa0c928 --- /dev/null +++ b/indra/llcommon/tests/threadsafeschedule_test.cpp @@ -0,0 +1,65 @@ +/** + * @file threadsafeschedule_test.cpp + * @author Nat Goodspeed + * @date 2021-10-04 + * @brief Test for threadsafeschedule. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "threadsafeschedule.h" +// STL headers +// std headers +#include +// external library headers +// other Linden headers +#include "../test/lltut.h" + +using namespace std::literals::chrono_literals; // ms suffix +using namespace std::literals::string_literals; // s suffix +using Queue = LL::ThreadSafeSchedule; + +/***************************************************************************** +* TUT +*****************************************************************************/ +namespace tut +{ + struct threadsafeschedule_data + { + Queue queue; + }; + typedef test_group threadsafeschedule_group; + typedef threadsafeschedule_group::object object; + threadsafeschedule_group threadsafeschedulegrp("threadsafeschedule"); + + template<> template<> + void object::test<1>() + { + set_test_name("push"); + // Simply calling push() a few times might result in indeterminate + // delivery order if the resolution of steady_clock is coarser than + // the real time required for each push() call. Explicitly increment + // the timestamp for each one -- but since we're passing explicit + // timestamps, make the queue reorder them. + queue.push(Queue::TimeTuple(Queue::Clock::now() + 20ms, "ghi")); + queue.push("abc"s); + queue.push(Queue::Clock::now() + 10ms, "def"); + queue.close(); + auto entry = queue.pop(); + ensure_equals("failed to pop first", std::get<0>(entry), "abc"s); + entry = queue.pop(); + ensure_equals("failed to pop second", std::get<0>(entry), "def"s); + ensure("queue not closed", queue.isClosed()); + ensure("queue prematurely done", ! queue.done()); + entry = queue.pop(); + ensure_equals("failed to pop third", std::get<0>(entry), "ghi"s); + bool popped = queue.tryPop(entry); + ensure("queue not empty", ! popped); + ensure("queue not done", queue.done()); + } +} // namespace tut diff --git a/indra/llcommon/threadsafeschedule.h b/indra/llcommon/threadsafeschedule.h new file mode 100644 index 0000000000..545c820f53 --- /dev/null +++ b/indra/llcommon/threadsafeschedule.h @@ -0,0 +1,334 @@ +/** + * @file threadsafeschedule.h + * @author Nat Goodspeed + * @date 2021-10-02 + * @brief ThreadSafeSchedule is an ordered queue in which every item has an + * associated timestamp. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_THREADSAFESCHEDULE_H) +#define LL_THREADSAFESCHEDULE_H + +#include "chrono.h" +#include "llexception.h" +#include "llthreadsafequeue.h" +#include "tuple.h" +#include +#include + +namespace LL +{ + namespace ThreadSafeSchedulePrivate + { + using TimePoint = std::chrono::steady_clock::time_point; + // Bundle consumer's data with a TimePoint to order items by timestamp. + template + using TimestampedTuple = std::tuple; + + // comparison functor for TimedTuples -- see TimedQueue comments + struct ReverseTupleOrder + { + template + bool operator()(const Tuple& left, const Tuple& right) const + { + return std::get<0>(left) > std::get<0>(right); + } + }; + + template + using TimedQueue = PriorityQueueAdapter< + TimestampedTuple, + // std::vector is the default storage for std::priority_queue, + // have to restate to specify comparison template parameter + std::vector>, + // std::priority_queue uses a counterintuitive comparison + // behavior: the default std::less comparator is used to present + // the *highest* value as top(). So to sort by earliest timestamp, + // we must invert by using >. + ReverseTupleOrder>; + } // namespace ThreadSafeSchedulePrivate + + /** + * ThreadSafeSchedule is an ordered LLThreadSafeQueue in which every item + * is given an associated timestamp. That is, TimePoint is implicitly + * prepended to the std::tuple with the specified types. + * + * Items are popped in increasing chronological order. Moreover, any item + * with a timestamp in the future is held back until + * std::chrono::steady_clock reaches that timestamp. + */ + template + class ThreadSafeSchedule: + public LLThreadSafeQueue, + ThreadSafeSchedulePrivate::TimedQueue> + { + public: + using DataTuple = std::tuple; + using TimeTuple = ThreadSafeSchedulePrivate::TimestampedTuple; + + private: + using super = LLThreadSafeQueue>; + using lock_t = typename super::lock_t; + using super::pop_; + using super::push_; + using super::mClosed; + using super::mEmptyCond; + using super::mCapacityCond; + + public: + using TimePoint = ThreadSafeSchedulePrivate::TimePoint; + using Clock = TimePoint::clock; + + ThreadSafeSchedule(U32 capacity=1024): + super(capacity) + {} + + /*----------------------------- push() -----------------------------*/ + /// explicitly pass TimeTuple + using super::push; + + /// pass DataTuple with implicit now + void push(const DataTuple& tuple) + { + push(tuple_cons(Clock::now(), tuple)); + } + + /// individually pass each component of the TimeTuple + void push(const TimePoint& time, Args&&... args) + { + push(TimeTuple(time, std::forward(args)...)); + } + + /// individually pass every component except the TimePoint (implies + /// now) -- could be ambiguous if the first specified template + /// parameter type is also TimePoint -- we could try to disambiguate, + /// but a simpler approach would be for the caller to explicitly + /// construct DataTuple and call that overload + void push(Args&&... args) + { + push(Clock::now(), std::forward(args)...); + } + + /*--------------------------- tryPush() ----------------------------*/ + /// explicit TimeTuple + using super::tryPush; + + /// DataTuple with implicit now + bool tryPush(const DataTuple& tuple) + { + return tryPush(tuple_cons(Clock::now(), tuple)); + } + + /// individually pass components + bool tryPush(const TimePoint& time, Args&&... args) + { + return tryPush(TimeTuple(time, std::forward(args)...)); + } + + /// individually pass components with implicit now + bool tryPush(Args&&... args) + { + return tryPush(Clock::now(), std::forward(args)...); + } + + /*-------------------------- tryPushFor() --------------------------*/ + /// explicit TimeTuple + using super::tryPushFor; + + /// DataTuple with implicit now + template + bool tryPushFor(const std::chrono::duration& timeout, + const DataTuple& tuple) + { + return tryPushFor(timeout, tuple_cons(Clock::now(), tuple)); + } + + /// individually pass components + template + bool tryPushFor(const std::chrono::duration& timeout, + const TimePoint& time, Args&&... args) + { + return tryPushFor(TimeTuple(time, std::forward(args)...)); + } + + /// individually pass components with implicit now + template + bool tryPushFor(const std::chrono::duration& timeout, + Args&&... args) + { + return tryPushFor(Clock::now(), std::forward(args)...); + } + + /*------------------------- tryPushUntil() -------------------------*/ + /// explicit TimeTuple + using super::tryPushUntil; + + /// DataTuple with implicit now + template + bool tryPushUntil(const std::chrono::time_point& until, + const DataTuple& tuple) + { + return tryPushUntil(until, tuple_cons(Clock::now(), tuple)); + } + + /// individually pass components + template + bool tryPushUntil(const std::chrono::time_point& until, + const TimePoint& time, Args&&... args) + { + return tryPushUntil(until, TimeTuple(time, std::forward(args)...)); + } + + /// individually pass components with implicit now + template + bool tryPushUntil(const std::chrono::time_point& until, + Args&&... args) + { + return tryPushUntil(until, Clock::now(), std::forward(args)...); + } + + /*----------------------------- pop() ------------------------------*/ + // Our consumer may or may not care about the timestamp associated + // with each popped item, so we allow retrieving either DataTuple or + // TimeTuple. One potential use would be to observe, and possibly + // adjust for, the time lag between the item time and the actual + // current time. + + /// pop DataTuple by value + DataTuple pop() + { + return tuple_cdr(popWithTime()); + } + + /// pop TimeTuple by value + TimeTuple popWithTime() + { + lock_t lock(super::mLock); + // We can't just sit around waiting forever, given that there may + // be items in the queue that are not yet ready but will *become* + // ready in the near future. So in fact, with this class, every + // pop() becomes a tryPopUntil(), constrained to the timestamp of + // the head item. It almost doesn't matter what we specify for the + // caller's time constraint -- all we really care about is the + // head item's timestamp. Since pop() and popWithTime() are + // defined to wait until either an item becomes available or the + // queue is closed, loop until one of those things happens. The + // constraint we pass just determines how often we'll loop while + // waiting. + TimeTuple tt; + while (true) + { + // Pick a point suitably far into the future. + TimePoint until = TimePoint::clock::now() + std::chrono::hours(24); + if (tryPopUntil_(lock, until, tt)) + return std::move(tt); + + // empty and closed: throw, just as super::pop() does + if (super::mStorage.empty() && super::mClosed) + { + LLTHROW(LLThreadSafeQueueInterrupt()); + } + // If not empty, we've still got items to drain. + // If not closed, it's worth waiting for more items. + // Either way, loop back to wait. + } + } + + // We can use tryPop(TimeTuple&) just as it stands; the only behavior + // difference is in our canPop() override method. + using super::tryPop; + + /// tryPop(DataTuple&) + bool tryPop(DataTuple& tuple) + { + TimeTuple tt; + if (! super::tryPop(tt)) + return false; + tuple = tuple_cdr(std::move(tt)); + return true; + } + + /// tryPopFor() + template + bool tryPopFor(const std::chrono::duration& timeout, Tuple& tuple) + { + // It's important to use OUR tryPopUntil() implementation, rather + // than delegating immediately to our base class. + return tryPopUntil(Clock::now() + timeout, tuple); + } + + /// tryPopUntil(TimeTuple&) + template + bool tryPopUntil(const std::chrono::time_point& until, + TimeTuple& tuple) + { + // super::tryPopUntil() wakes up when an item becomes available or + // we hit 'until', whichever comes first. Thing is, the current + // head of the queue could become ready sooner than either of + // those events, and we need to deliver it as soon as it does. + // Don't wait past the TimePoint of the head item. + // Naturally, lock the queue before peeking at mStorage. + return super::tryLockUntil( + until, + [this, until, &tuple](lock_t& lock) + { + // Use our time_point_cast to allow for 'until' that's a + // time_point type other than TimePoint. + return tryPopUntil_(lock, time_point_cast(until), tuple); + }); + } + + bool tryPopUntil_(lock_t& lock, const TimePoint& until, TimeTuple& tuple) + { + TimePoint adjusted = until; + if (! super::mStorage.empty()) + { + // use whichever is earlier: the head item's timestamp, or + // the caller's limit + adjusted = min(std::get<0>(super::mStorage.front()), adjusted); + } + // now delegate to base-class tryPopUntil_() + return super::tryPopUntil_(lock, adjusted, tuple); + } + + /// tryPopUntil(DataTuple&) + template + bool tryPopUntil(const std::chrono::time_point& until, + DataTuple& tuple) + { + TimeTuple tt; + if (! tryPopUntil(until, tt)) + return false; + tuple = tuple_cdr(std::move(tt)); + return true; + } + + /*------------------------------ etc. ------------------------------*/ + // We can't hide items that aren't yet ready because we can't traverse + // the underlying priority_queue: it has no iterators, only top(). So + // a consumer could observe size() > 0 and yet tryPop() returns false. + // Shrug, in a multi-consumer scenario that would be expected behavior. + using super::size; + // open/closed state + using super::close; + using super::isClosed; + using super::done; + + private: + // this method is called by base class pop_() every time we're + // considering whether to deliver the current head element + bool canPop(const TimeTuple& head) const override + { + // an item with a future timestamp isn't yet ready to pop + // (should we add some slop for overhead?) + return std::get<0>(head) <= Clock::now(); + } + }; + +} // namespace LL + +#endif /* ! defined(LL_THREADSAFESCHEDULE_H) */ -- cgit v1.3 From cf70766b4504f7ee745822926c526ed9c86c9339 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 6 Oct 2021 12:54:29 -0400 Subject: SL-16024: Fix ThreadSafeSchedule::tryPopFor(), tryPopUntil(). ThreadSafeSchedule::tryPopUntil() (and therefore tryPopFor()) was simply delegating to LLThreadSafeQueue::tryPopUntil(), with an adjusted timeout since we want to wake up as soon as the head item, if any, becomes ready. But then we have to loop back to retry the pop to actually deal with that head item. In addition, ThreadSafeSchedule::popWithTime() was spinning rather than properly blocking on a timed condition variable. Fixed. --- indra/llcommon/llthreadsafequeue.h | 51 +++++++++-------- indra/llcommon/tests/threadsafeschedule_test.cpp | 10 +++- indra/llcommon/threadsafeschedule.h | 72 ++++++++++++++++++------ 3 files changed, 88 insertions(+), 45 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index bd2d82d4c3..719edcd579 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -179,11 +179,12 @@ protected: boost::fibers::condition_variable_any mCapacityCond; boost::fibers::condition_variable_any mEmptyCond; + enum pop_result { EMPTY, DONE, WAITING, POPPED }; // implementation logic, suitable for passing to tryLockUntil() template - bool tryPopUntil_(lock_t& lock, - const std::chrono::time_point& until, - ElementT& element); + pop_result tryPopUntil_(lock_t& lock, + const std::chrono::time_point& until, + ElementT& element); // if we're able to lock immediately, do so and run the passed callable, // which must accept lock_t& and return bool template @@ -197,7 +198,6 @@ protected: template bool push_(lock_t& lock, T&& element); // while lock is locked, really pop the head element, if we can - enum pop_result { EMPTY, WAITING, POPPED }; pop_result pop_(lock_t& lock, ElementT& element); // Is the current head element ready to pop? We say yes; subclass can // override as needed. @@ -398,7 +398,7 @@ LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) { // If mStorage is empty, there's no head element. if (mStorage.empty()) - return EMPTY; + return mClosed? DONE : EMPTY; // If there's a head element, pass it to canPop() to see if it's ready to pop. if (! canPop(mStorage.front())) @@ -427,16 +427,16 @@ ElementT LLThreadSafeQueue::pop(void) if (popped == POPPED) return std::move(value); - // Once the queue is empty, mClosed lets us know if there will ever be - // any more coming. If we didn't pop because WAITING, i.e. canPop() - // returned false, then even if the producer end has been closed, - // there's still at least one item to drain: wait for it. - if (popped == EMPTY && mClosed) + // Once the queue is DONE, there will never be any more coming. + if (popped == DONE) { LLTHROW(LLThreadSafeQueueInterrupt()); } - // Storage empty, queue still open. Wait for signal. + // If we didn't pop because WAITING, i.e. canPop() returned false, + // then even if the producer end has been closed, there's still at + // least one item to drain: wait for it. Or we might be EMPTY, with + // the queue still open. Either way, wait for signal. mEmptyCond.wait(lock1); } } @@ -448,8 +448,8 @@ bool LLThreadSafeQueue::tryPop(ElementT & element) return tryLock( [this, &element](lock_t& lock) { - // no need to check mClosed: tryPop() behavior when the queue is - // closed is implemented by simple inability to push any new + // conflate EMPTY, DONE, WAITING: tryPop() behavior when the queue + // is closed is implemented by simple inability to push any new // elements return pop_(lock, element) == POPPED; }); @@ -478,7 +478,8 @@ bool LLThreadSafeQueue::tryPopUntil( until, [this, until, &element](lock_t& lock) { - return tryPopUntil_(lock, until, element); + // conflate EMPTY, DONE, WAITING + return tryPopUntil_(lock, until, element) == POPPED; }); } @@ -486,26 +487,28 @@ bool LLThreadSafeQueue::tryPopUntil( // body of tryPopUntil(), called once we have the lock template template -bool LLThreadSafeQueue::tryPopUntil_( +typename LLThreadSafeQueue::pop_result +LLThreadSafeQueue::tryPopUntil_( lock_t& lock, const std::chrono::time_point& until, ElementT& element) { while (true) { - if (pop_(lock, element) == POPPED) - return true; - - if (mClosed) + pop_result popped = pop_(lock, element); + if (popped == POPPED || popped == DONE) { - return false; + // If we succeeded, great! If we've drained the last item, so be + // it. Either way, break the loop and tell caller. + return popped; } - // Storage empty. Wait for signal. + // EMPTY or WAITING: wait for signal. if (LLCoros::cv_status::timeout == mEmptyCond.wait_until(lock, until)) { - // timed out -- formally we might recheck both conditions above - return false; + // timed out -- formally we might recheck + // as it is, break loop + return popped; } // If we didn't time out, we were notified for some reason. Loop back // to check. @@ -546,7 +549,7 @@ template bool LLThreadSafeQueue::done() { lock_t lock(mLock); - return mClosed && mStorage.size() == 0; + return mClosed && mStorage.empty(); } #endif diff --git a/indra/llcommon/tests/threadsafeschedule_test.cpp b/indra/llcommon/tests/threadsafeschedule_test.cpp index ec0fa0c928..af67b9f492 100644 --- a/indra/llcommon/tests/threadsafeschedule_test.cpp +++ b/indra/llcommon/tests/threadsafeschedule_test.cpp @@ -47,6 +47,8 @@ namespace tut // the timestamp for each one -- but since we're passing explicit // timestamps, make the queue reorder them. queue.push(Queue::TimeTuple(Queue::Clock::now() + 20ms, "ghi")); + // Given the various push() overloads, you have to match the type + // exactly: conversions are ambiguous. queue.push("abc"s); queue.push(Queue::Clock::now() + 10ms, "def"); queue.close(); @@ -56,9 +58,11 @@ namespace tut ensure_equals("failed to pop second", std::get<0>(entry), "def"s); ensure("queue not closed", queue.isClosed()); ensure("queue prematurely done", ! queue.done()); - entry = queue.pop(); - ensure_equals("failed to pop third", std::get<0>(entry), "ghi"s); - bool popped = queue.tryPop(entry); + std::string s; + bool popped = queue.tryPopFor(1s, s); + ensure("failed to pop third", popped); + ensure_equals("third is wrong", s, "ghi"s); + popped = queue.tryPop(s); ensure("queue not empty", ! popped); ensure("queue not done", queue.done()); } diff --git a/indra/llcommon/threadsafeschedule.h b/indra/llcommon/threadsafeschedule.h index 545c820f53..8ab4311ca1 100644 --- a/indra/llcommon/threadsafeschedule.h +++ b/indra/llcommon/threadsafeschedule.h @@ -73,11 +73,7 @@ namespace LL private: using super = LLThreadSafeQueue>; using lock_t = typename super::lock_t; - using super::pop_; - using super::push_; - using super::mClosed; - using super::mEmptyCond; - using super::mCapacityCond; + using pop_result = typename super::pop_result; public: using TimePoint = ThreadSafeSchedulePrivate::TimePoint; @@ -92,6 +88,11 @@ namespace LL using super::push; /// pass DataTuple with implicit now + // This could be ambiguous for Args with a single type. Unfortunately + // we can't enable_if an individual method with a condition based on + // the *class* template arguments, only on that method's template + // arguments. We could specialize this class for the single-Args case; + // we could minimize redundancy by breaking out a common base class... void push(const DataTuple& tuple) { push(tuple_cons(Clock::now(), tuple)); @@ -103,11 +104,11 @@ namespace LL push(TimeTuple(time, std::forward(args)...)); } - /// individually pass every component except the TimePoint (implies - /// now) -- could be ambiguous if the first specified template - /// parameter type is also TimePoint -- we could try to disambiguate, - /// but a simpler approach would be for the caller to explicitly - /// construct DataTuple and call that overload + /// individually pass every component except the TimePoint (implies now) + // This could be ambiguous if the first specified template parameter + // type is also TimePoint. We could try to disambiguate, but a simpler + // approach would be for the caller to explicitly construct DataTuple + // and call that overload. void push(Args&&... args) { push(Clock::now(), std::forward(args)...); @@ -199,6 +200,10 @@ namespace LL // current time. /// pop DataTuple by value + // It would be great to notice when sizeof...(Args) == 1 and directly + // return the first (only) value, instead of making pop()'s caller + // call std::get<0>(value). See push(DataTuple) remarks for why we + // haven't yet jumped through those hoops. DataTuple pop() { return tuple_cdr(popWithTime()); @@ -224,16 +229,17 @@ namespace LL { // Pick a point suitably far into the future. TimePoint until = TimePoint::clock::now() + std::chrono::hours(24); - if (tryPopUntil_(lock, until, tt)) + pop_result popped = tryPopUntil_(lock, until, tt); + if (popped == super::POPPED) return std::move(tt); - // empty and closed: throw, just as super::pop() does - if (super::mStorage.empty() && super::mClosed) + // DONE: throw, just as super::pop() does + if (popped == super::DONE) { LLTHROW(LLThreadSafeQueueInterrupt()); } - // If not empty, we've still got items to drain. - // If not closed, it's worth waiting for more items. + // WAITING: we've still got items to drain. + // EMPTY: not closed, so it's worth waiting for more items. // Either way, loop back to wait. } } @@ -252,6 +258,16 @@ namespace LL return true; } + /// for when Args has exactly one type + bool tryPop(typename std::tuple_element<1, TimeTuple>::type& value) + { + TimeTuple tt; + if (! super::tryPop(tt)) + return false; + value = std::get<1>(std::move(tt)); + return true; + } + /// tryPopFor() template bool tryPopFor(const std::chrono::duration& timeout, Tuple& tuple) @@ -278,11 +294,12 @@ namespace LL { // Use our time_point_cast to allow for 'until' that's a // time_point type other than TimePoint. - return tryPopUntil_(lock, time_point_cast(until), tuple); + return super::POPPED == + tryPopUntil_(lock, LL::time_point_cast(until), tuple); }); } - bool tryPopUntil_(lock_t& lock, const TimePoint& until, TimeTuple& tuple) + pop_result tryPopUntil_(lock_t& lock, const TimePoint& until, TimeTuple& tuple) { TimePoint adjusted = until; if (! super::mStorage.empty()) @@ -292,7 +309,14 @@ namespace LL adjusted = min(std::get<0>(super::mStorage.front()), adjusted); } // now delegate to base-class tryPopUntil_() - return super::tryPopUntil_(lock, adjusted, tuple); + pop_result popped; + while ((popped = super::tryPopUntil_(lock, adjusted, tuple)) == super::WAITING) + { + // If super::tryPopUntil_() returns WAITING, it means there's + // a head item, but it's not yet time. But it's worth looping + // back to recheck. + } + return popped; } /// tryPopUntil(DataTuple&) @@ -307,6 +331,18 @@ namespace LL return true; } + /// for when Args has exactly one type + template + bool tryPopUntil(const std::chrono::time_point& until, + typename std::tuple_element<1, TimeTuple>::type& value) + { + TimeTuple tt; + if (! tryPopUntil(until, tt)) + return false; + value = std::get<1>(std::move(tt)); + return true; + } + /*------------------------------ etc. ------------------------------*/ // We can't hide items that aren't yet ready because we can't traverse // the underlying priority_queue: it has no iterators, only top(). So -- cgit v1.3 From 1ef78e2afa9e8424dd5d84b2b104b31e72e9e95a Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 6 Oct 2021 15:28:58 -0400 Subject: SL-16024: Work around VS bug regarding base-class enum. --- indra/llcommon/threadsafeschedule.h | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/threadsafeschedule.h b/indra/llcommon/threadsafeschedule.h index 8ab4311ca1..0e70d30714 100644 --- a/indra/llcommon/threadsafeschedule.h +++ b/indra/llcommon/threadsafeschedule.h @@ -73,7 +73,9 @@ namespace LL private: using super = LLThreadSafeQueue>; using lock_t = typename super::lock_t; - using pop_result = typename super::pop_result; + // VS 2017 needs this due to a bug: + // https://developercommunity.visualstudio.com/t/cannot-access-protected-enumerator-of-enclosing-cl/203430 + enum pop_result { EMPTY=super::EMPTY, DONE=super::DONE, WAITING=super::WAITING, POPPED=super::POPPED }; public: using TimePoint = ThreadSafeSchedulePrivate::TimePoint; @@ -230,11 +232,11 @@ namespace LL // Pick a point suitably far into the future. TimePoint until = TimePoint::clock::now() + std::chrono::hours(24); pop_result popped = tryPopUntil_(lock, until, tt); - if (popped == super::POPPED) + if (popped == POPPED) return std::move(tt); // DONE: throw, just as super::pop() does - if (popped == super::DONE) + if (popped == DONE) { LLTHROW(LLThreadSafeQueueInterrupt()); } @@ -294,7 +296,7 @@ namespace LL { // Use our time_point_cast to allow for 'until' that's a // time_point type other than TimePoint. - return super::POPPED == + return POPPED == tryPopUntil_(lock, LL::time_point_cast(until), tuple); }); } @@ -310,7 +312,7 @@ namespace LL } // now delegate to base-class tryPopUntil_() pop_result popped; - while ((popped = super::tryPopUntil_(lock, adjusted, tuple)) == super::WAITING) + while ((popped = pop_result(super::tryPopUntil_(lock, adjusted, tuple))) == WAITING) { // If super::tryPopUntil_() returns WAITING, it means there's // a head item, but it's not yet time. But it's worth looping -- cgit v1.3 From 2cb09dd4a828756dce6180505c63851aa9875187 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 7 Oct 2021 11:53:45 -0400 Subject: SL-16024: Return shared_ptr from LLInstanceTracker::getInstance(). It feels wrong to return a dumb LLInstanceTracker subclass* from getInstance() when we use std::shared_ptr and std::weak_ptr internally. But tweak consumers to use 'auto' or LLInstanceTracker::ptr_t in case we later revisit this decision. We did add a couple get() calls where it's important to obtain a dumb pointer. --- indra/llcommon/llinstancetracker.h | 59 ++++++++++++++++++++-------- indra/llcommon/llleaplistener.cpp | 2 +- indra/llui/llnotifications.cpp | 2 +- indra/llui/llstatbar.cpp | 20 ++++------ indra/llxml/llcontrol.h | 8 ++-- indra/newview/llappviewer.cpp | 4 +- indra/newview/llbrowsernotification.cpp | 4 +- indra/newview/llfloaterwebcontent.cpp | 4 +- indra/newview/llnotificationofferhandler.cpp | 4 +- indra/newview/llviewercontrollistener.cpp | 8 ++-- indra/newview/llviewermessage.cpp | 4 +- 11 files changed, 71 insertions(+), 48 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llinstancetracker.h b/indra/llcommon/llinstancetracker.h index 402333cca7..02535a59e7 100644 --- a/indra/llcommon/llinstancetracker.h +++ b/indra/llcommon/llinstancetracker.h @@ -83,13 +83,34 @@ class LLInstanceTracker typedef llthread::LockStatic LockStatic; public: + using ptr_t = std::shared_ptr; + using weak_t = std::weak_ptr; + + /** + * Storing a dumb T* somewhere external is a bad idea, since + * LLInstanceTracker subclasses are explicitly destroyed rather than + * managed by smart pointers. It's legal to declare stack instances of an + * LLInstanceTracker subclass. But it's reasonable to store a + * std::weak_ptr, which will become invalid when the T instance is + * destroyed. + */ + weak_t getWeak() + { + return mSelf; + } + + static S32 instanceCount() + { + return LockStatic()->mMap.size(); + } + // snapshot of std::pair> pairs class snapshot { // It's very important that what we store in this snapshot are // weak_ptrs, NOT shared_ptrs. That's how we discover whether any // instance has been deleted during the lifespan of a snapshot. - typedef std::vector>> VectorType; + typedef std::vector> VectorType; // Dereferencing our iterator produces a std::shared_ptr for each // instance that still exists. Since we store weak_ptrs, that involves // two chained transformations: @@ -98,7 +119,7 @@ public: // It is very important that we filter lazily, that is, during // traversal. Any one of our stored weak_ptrs might expire during // traversal. - typedef std::pair> strong_pair; + typedef std::pair strong_pair; // Note for future reference: nat has not yet had any luck (up to // Boost 1.67) trying to use boost::transform_iterator with a hand- // coded functor, only with actual functions. In my experience, an @@ -202,17 +223,12 @@ public: iterator end() { return iterator(snapshot::end(), key_getter); } }; - static T* getInstance(const KEY& k) + static ptr_t getInstance(const KEY& k) { LockStatic lock; const InstanceMap& map(lock->mMap); typename InstanceMap::const_iterator found = map.find(k); - return (found == map.end()) ? NULL : found->second.get(); - } - - static S32 instanceCount() - { - return LockStatic()->mMap.size(); + return (found == map.end()) ? NULL : found->second; } protected: @@ -222,7 +238,9 @@ protected: // shared_ptr, so give it a no-op deleter. We store shared_ptrs in our // InstanceMap specifically so snapshot can store weak_ptrs so we can // detect deletions during traversals. - std::shared_ptr ptr(static_cast(this), [](T*){}); + ptr_t ptr(static_cast(this), [](T*){}); + // save corresponding weak_ptr for future reference + mSelf = ptr; LockStatic lock; add_(lock, key, ptr); } @@ -257,7 +275,7 @@ private: static std::string report(const char* key) { return report(std::string(key)); } // caller must instantiate LockStatic - void add_(LockStatic& lock, const KEY& key, const std::shared_ptr& ptr) + void add_(LockStatic& lock, const KEY& key, const ptr_t& ptr) { mInstanceKey = key; InstanceMap& map = lock->mMap; @@ -281,7 +299,7 @@ private: break; } } - std::shared_ptr remove_(LockStatic& lock) + ptr_t remove_(LockStatic& lock) { InstanceMap& map = lock->mMap; typename InstanceMap::iterator iter = map.find(mInstanceKey); @@ -295,6 +313,9 @@ private: } private: + // Storing a weak_ptr to self is a bit like deriving from + // std::enable_shared_from_this(), except more explicit. + weak_t mSelf; KEY mInstanceKey; }; @@ -326,6 +347,9 @@ class LLInstanceTracker typedef llthread::LockStatic LockStatic; public: + using ptr_t = std::shared_ptr; + using weak_t = std::weak_ptr; + /** * Storing a dumb T* somewhere external is a bad idea, since * LLInstanceTracker subclasses are explicitly destroyed rather than @@ -334,12 +358,15 @@ public: * std::weak_ptr, which will become invalid when the T instance is * destroyed. */ - std::weak_ptr getWeak() + weak_t getWeak() { return mSelf; } - static S32 instanceCount() { return LockStatic()->mSet.size(); } + static S32 instanceCount() + { + return LockStatic()->mSet.size(); + } // snapshot of std::shared_ptr pointers class snapshot @@ -347,7 +374,7 @@ public: // It's very important that what we store in this snapshot are // weak_ptrs, NOT shared_ptrs. That's how we discover whether any // instance has been deleted during the lifespan of a snapshot. - typedef std::vector> VectorType; + typedef std::vector VectorType; // Dereferencing our iterator produces a std::shared_ptr for each // instance that still exists. Since we store weak_ptrs, that involves // two chained transformations: @@ -453,7 +480,7 @@ protected: private: // Storing a weak_ptr to self is a bit like deriving from // std::enable_shared_from_this(), except more explicit. - std::weak_ptr mSelf; + weak_t mSelf; }; #endif diff --git a/indra/llcommon/llleaplistener.cpp b/indra/llcommon/llleaplistener.cpp index 3e6ce9092c..11bfec1b31 100644 --- a/indra/llcommon/llleaplistener.cpp +++ b/indra/llcommon/llleaplistener.cpp @@ -220,7 +220,7 @@ void LLLeapListener::getAPI(const LLSD& request) const { Response reply(LLSD(), request); - LLEventAPI* found = LLEventAPI::getInstance(request["api"]); + auto found = LLEventAPI::getInstance(request["api"]); if (found) { reply["name"] = found->getName(); diff --git a/indra/llui/llnotifications.cpp b/indra/llui/llnotifications.cpp index b791a19c2b..88eda1c172 100644 --- a/indra/llui/llnotifications.cpp +++ b/indra/llui/llnotifications.cpp @@ -1387,7 +1387,7 @@ bool LLNotifications::failedUniquenessTest(const LLSD& payload) LLNotificationChannelPtr LLNotifications::getChannel(const std::string& channelName) { - return LLNotificationChannelPtr(LLNotificationChannel::getInstance(channelName)); + return LLNotificationChannelPtr(LLNotificationChannel::getInstance(channelName).get()); } diff --git a/indra/llui/llstatbar.cpp b/indra/llui/llstatbar.cpp index 6c8e63442b..8adcd664df 100644 --- a/indra/llui/llstatbar.cpp +++ b/indra/llui/llstatbar.cpp @@ -554,29 +554,25 @@ void LLStatBar::draw() void LLStatBar::setStat(const std::string& stat_name) { using namespace LLTrace; - const StatType* count_stat; - const StatType* event_stat; - const StatType* sample_stat; - const StatType* mem_stat; - if ((count_stat = StatType::getInstance(stat_name))) + if (auto count_stat = StatType::getInstance(stat_name)) { - mStat.countStatp = count_stat; + mStat.countStatp = count_stat.get(); mStatType = STAT_COUNT; } - else if ((event_stat = StatType::getInstance(stat_name))) + else if (auto event_stat = StatType::getInstance(stat_name)) { - mStat.eventStatp = event_stat; + mStat.eventStatp = event_stat.get(); mStatType = STAT_EVENT; } - else if ((sample_stat = StatType::getInstance(stat_name))) + else if (auto sample_stat = StatType::getInstance(stat_name)) { - mStat.sampleStatp = sample_stat; + mStat.sampleStatp = sample_stat.get(); mStatType = STAT_SAMPLE; } - else if ((mem_stat = StatType::getInstance(stat_name))) + else if (auto mem_stat = StatType::getInstance(stat_name)) { - mStat.memStatp = mem_stat; + mStat.memStatp = mem_stat.get(); mStatType = STAT_MEM; } } diff --git a/indra/llxml/llcontrol.h b/indra/llxml/llcontrol.h index 19508becc3..5da13f5010 100644 --- a/indra/llxml/llcontrol.h +++ b/indra/llxml/llcontrol.h @@ -405,8 +405,8 @@ public: const T& default_value, const std::string& comment = "Declared In Code") { - mCachedControlPtr = LLControlCache::getInstance(name); - if (mCachedControlPtr.isNull()) + mCachedControlPtr = LLControlCache::getInstance(name).get(); + if (! mCachedControlPtr) { mCachedControlPtr = new LLControlCache(group, name, default_value, comment); } @@ -415,8 +415,8 @@ public: LLCachedControl(LLControlGroup& group, const std::string& name) { - mCachedControlPtr = LLControlCache::getInstance(name); - if (mCachedControlPtr.isNull()) + mCachedControlPtr = LLControlCache::getInstance(name).get(); + if (! mCachedControlPtr) { mCachedControlPtr = new LLControlCache(group, name); } diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 891722e1bd..4a9a1ad0d4 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -2336,7 +2336,7 @@ bool LLAppViewer::loadSettingsFromDirectory(const std::string& location_key, LL_INFOS("Settings") << "Attempting to load settings for the group " << file.name() << " - from location " << location_key << LL_ENDL; - LLControlGroup* settings_group = LLControlGroup::getInstance(file.name); + auto settings_group = LLControlGroup::getInstance(file.name); if(!settings_group) { LL_WARNS("Settings") << "No matching settings group for name " << file.name() << LL_ENDL; @@ -2638,7 +2638,7 @@ bool LLAppViewer::initConfiguration() group_part = name.substr(0, pos); name_part = name.substr(pos+1); LL_INFOS() << "Setting " << group_part << "." << name_part << " to " << value << LL_ENDL; - LLControlGroup* g = LLControlGroup::getInstance(group_part); + auto g = LLControlGroup::getInstance(group_part); if (g) control = g->getControl(name_part); } else diff --git a/indra/newview/llbrowsernotification.cpp b/indra/newview/llbrowsernotification.cpp index 0460bff1b4..30ac35fff7 100644 --- a/indra/newview/llbrowsernotification.cpp +++ b/indra/newview/llbrowsernotification.cpp @@ -43,14 +43,14 @@ LLBrowserNotification::LLBrowserNotification() bool LLBrowserNotification::processNotification(const LLNotificationPtr& notification) { LLUUID media_id = notification->getPayload()["media_id"].asUUID(); - LLMediaCtrl* media_instance = LLMediaCtrl::getInstance(media_id); + auto media_instance = LLMediaCtrl::getInstance(media_id); if (media_instance) { media_instance->showNotification(notification); } else if (LLViewerMediaFocus::instance().getControlsMediaID() == media_id) { - LLViewerMediaImpl* impl = LLViewerMedia::getInstance()->getMediaImplFromTextureID(media_id); + auto impl = LLViewerMedia::getInstance()->getMediaImplFromTextureID(media_id); if (impl) { impl->showNotification(notification); diff --git a/indra/newview/llfloaterwebcontent.cpp b/indra/newview/llfloaterwebcontent.cpp index 23fd6d9c8e..ceab472c55 100644 --- a/indra/newview/llfloaterwebcontent.cpp +++ b/indra/newview/llfloaterwebcontent.cpp @@ -159,7 +159,7 @@ LLFloater* LLFloaterWebContent::create( Params p) //static void LLFloaterWebContent::closeRequest(const std::string &uuid) { - LLFloaterWebContent* floaterp = instance_tracker_t::getInstance(uuid); + auto floaterp = instance_tracker_t::getInstance(uuid); if (floaterp) { floaterp->closeFloater(false); @@ -169,7 +169,7 @@ void LLFloaterWebContent::closeRequest(const std::string &uuid) //static void LLFloaterWebContent::geometryChanged(const std::string &uuid, S32 x, S32 y, S32 width, S32 height) { - LLFloaterWebContent* floaterp = instance_tracker_t::getInstance(uuid); + auto floaterp = instance_tracker_t::getInstance(uuid); if (floaterp) { floaterp->geometryChanged(x, y, width, height); diff --git a/indra/newview/llnotificationofferhandler.cpp b/indra/newview/llnotificationofferhandler.cpp index a9678b1e93..d9359d20cf 100644 --- a/indra/newview/llnotificationofferhandler.cpp +++ b/indra/newview/llnotificationofferhandler.cpp @@ -166,14 +166,14 @@ bool LLOfferHandler::processNotification(const LLNotificationPtr& notification) /*virtual*/ void LLOfferHandler::onChange(LLNotificationPtr p) { - LLToastNotifyPanel* panelp = LLToastNotifyPanel::getInstance(p->getID()); + auto panelp = LLToastNotifyPanel::getInstance(p->getID()); if (panelp) { // // HACK: if we're dealing with a notification embedded in IM, update it // otherwise remove its toast // - if (dynamic_cast(panelp)) + if (dynamic_cast(panelp.get())) { panelp->updateNotification(); } diff --git a/indra/newview/llviewercontrollistener.cpp b/indra/newview/llviewercontrollistener.cpp index 3443bb644a..8820f9ec56 100644 --- a/indra/newview/llviewercontrollistener.cpp +++ b/indra/newview/llviewercontrollistener.cpp @@ -127,7 +127,7 @@ struct Info LLEventAPI::Response response; std::string groupname; - LLControlGroup* group; + LLControlGroup::ptr_t group; std::string key; LLControlVariable* control; }; @@ -187,7 +187,7 @@ void LLViewerControlListener::groups(LLSD const & request) struct CollectVars: public LLControlGroup::ApplyFunctor { - CollectVars(LLControlGroup* g): + CollectVars(LLControlGroup::ptr_t g): mGroup(g) {} @@ -200,7 +200,7 @@ struct CollectVars: public LLControlGroup::ApplyFunctor ("comment", control->getComment())); } - LLControlGroup* mGroup; + LLControlGroup::ptr_t mGroup; LLSD vars; }; @@ -210,7 +210,7 @@ void LLViewerControlListener::vars(LLSD const & request) // control name. Response response(LLSD(), request); std::string groupname(request["group"]); - LLControlGroup* group(LLControlGroup::getInstance(groupname)); + auto group(LLControlGroup::getInstance(groupname)); if (! group) { return response.error(STRINGIZE("Unrecognized group '" << groupname << "'")); diff --git a/indra/newview/llviewermessage.cpp b/indra/newview/llviewermessage.cpp index 39c891c9c1..94d2d216b9 100644 --- a/indra/newview/llviewermessage.cpp +++ b/indra/newview/llviewermessage.cpp @@ -3978,8 +3978,8 @@ void process_sim_stats(LLMessageSystem *msg, void **user_data) F32 stat_value; msg->getU32("Stat", "StatID", stat_id, i); msg->getF32("Stat", "StatValue", stat_value, i); - LLStatViewer::SimMeasurementSampler* measurementp = LLStatViewer::SimMeasurementSampler::getInstance((ESimStatID)stat_id); - + auto measurementp = LLStatViewer::SimMeasurementSampler::getInstance((ESimStatID)stat_id); + if (measurementp ) { measurementp->sample(stat_value); -- cgit v1.3 From b554c9eaf45c83500e6b65e295cc507b9a3d537b Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 7 Oct 2021 14:00:39 -0400 Subject: SL-16024: Adapt llinstancetracker_test.cpp to getInstance() change. --- indra/llcommon/tests/llinstancetracker_test.cpp | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/tests/llinstancetracker_test.cpp b/indra/llcommon/tests/llinstancetracker_test.cpp index 9b89159625..5daa29adf4 100644 --- a/indra/llcommon/tests/llinstancetracker_test.cpp +++ b/indra/llcommon/tests/llinstancetracker_test.cpp @@ -90,19 +90,19 @@ namespace tut { Keyed one("one"); ensure_equals(Keyed::instanceCount(), 1); - Keyed* found = Keyed::getInstance("one"); - ensure("couldn't find stack Keyed", found); - ensure_equals("found wrong Keyed instance", found, &one); + auto found = Keyed::getInstance("one"); + ensure("couldn't find stack Keyed", bool(found)); + ensure_equals("found wrong Keyed instance", found.get(), &one); { boost::scoped_ptr two(new Keyed("two")); ensure_equals(Keyed::instanceCount(), 2); - Keyed* found = Keyed::getInstance("two"); - ensure("couldn't find heap Keyed", found); - ensure_equals("found wrong Keyed instance", found, two.get()); + auto found = Keyed::getInstance("two"); + ensure("couldn't find heap Keyed", bool(found)); + ensure_equals("found wrong Keyed instance", found.get(), two.get()); } ensure_equals(Keyed::instanceCount(), 1); } - Keyed* found = Keyed::getInstance("one"); + auto found = Keyed::getInstance("one"); ensure("Keyed key lives too long", ! found); ensure_equals(Keyed::instanceCount(), 0); } -- cgit v1.3 From 6e06d1db6045df2e4961243f379c4d7695a8190d Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 7 Oct 2021 15:32:19 -0400 Subject: SL-16024: Make LLCond::get() lock and return by value. Its previous behavior, returning a const reference without locking, was wrong: it could return a reference to an object in an inconsistent state if it was concurrently being modified on another thread. Locking the mutex and returning a copy by value is the correct behavior. --- indra/llcommon/llcond.h | 24 +++++++++++++++--------- 1 file changed, 15 insertions(+), 9 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcond.h b/indra/llcommon/llcond.h index e31b67d893..c08acb66a1 100644 --- a/indra/llcommon/llcond.h +++ b/indra/llcommon/llcond.h @@ -53,6 +53,8 @@ private: LLCoros::Mutex mMutex; // Use LLCoros::ConditionVariable for the same reason. LLCoros::ConditionVariable mCond; + using LockType = LLCoros::LockType; + using cv_status = LLCoros::cv_status; public: /// LLCond can be explicitly initialized with a specific value for mData if @@ -65,10 +67,14 @@ public: LLCond(const LLCond&) = delete; LLCond& operator=(const LLCond&) = delete; - /// get() returns a const reference to the stored DATA. The only way to - /// get a non-const reference -- to modify the stored DATA -- is via - /// update_one() or update_all(). - const value_type& get() const { return mData; } + /// get() returns the stored DATA by value -- so to use get(), DATA must + /// be copyable. The only way to get a non-const reference -- to modify + /// the stored DATA -- is via update_one() or update_all(). + value_type get() + { + LockType lk(mMutex); + return mData; + } /** * Pass update_one() an invocable accepting non-const (DATA&). The @@ -83,7 +89,7 @@ public: void update_one(MODIFY modify) { { // scope of lock can/should end before notify_one() - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); modify(mData); } mCond.notify_one(); @@ -102,7 +108,7 @@ public: void update_all(MODIFY modify) { { // scope of lock can/should end before notify_all() - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); modify(mData); } mCond.notify_all(); @@ -118,7 +124,7 @@ public: template void wait(Pred pred) { - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); // We must iterate explicitly since the predicate accepted by // condition_variable::wait() requires a different signature: // condition_variable::wait() calls its predicate with no arguments. @@ -205,14 +211,14 @@ private: template bool wait_until(const std::chrono::time_point& timeout_time, Pred pred) { - LLCoros::LockType lk(mMutex); + LockType lk(mMutex); // We advise the caller to pass a predicate accepting (const DATA&). // But what if they instead pass a predicate accepting non-const // (DATA&)? Such a predicate could modify mData, which would be Bad. // Forbid that. while (! pred(const_cast(mData))) { - if (LLCoros::cv_status::timeout == mCond.wait_until(lk, timeout_time)) + if (cv_status::timeout == mCond.wait_until(lk, timeout_time)) { // It's possible that wait_until() timed out AND the predicate // became true more or less simultaneously. Even though -- cgit v1.3 From 623ac79120a417ec445ce5c106a907fe46734309 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 7 Oct 2021 15:32:51 -0400 Subject: SL-16024: Add LL::WorkQueue for passing work items between threads. A typical WorkQueue has a string name, which can be used to find it to post work to it. "Work" is a nullary callable. WorkQueue is a multi-producer, multi-consumer thread-safe queue: multiple threads can service the WorkQueue, multiple threads can post work to it. Work can be scheduled in the future by submitting with a timestamp. In addition, a given work item can be scheduled to run on a recurring basis. A requesting thread servicing a WorkQueue of its own, such as the viewer's main thread, can submit work to another WorkQueue along with a callback to be passed the result (of arbitrary type) of the first work item. The callback is posted to the originating WorkQueue, permitting safe data exchange between participating threads. Methods are provided for different kinds of servicing threads. runUntilClose() is useful for a simple worker thread. runFor(duration) devotes no more than a specified time slice to that WorkQueue, e.g. for use by the main thread. --- indra/llcommon/CMakeLists.txt | 3 + indra/llcommon/tests/workqueue_test.cpp | 158 ++++++++++++++++ indra/llcommon/threadsafeschedule.h | 1 + indra/llcommon/workqueue.cpp | 114 +++++++++++ indra/llcommon/workqueue.h | 325 ++++++++++++++++++++++++++++++++ 5 files changed, 601 insertions(+) create mode 100644 indra/llcommon/tests/workqueue_test.cpp create mode 100644 indra/llcommon/workqueue.cpp create mode 100644 indra/llcommon/workqueue.h (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 5efcfabf24..a3dbb6d9d0 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -121,6 +121,7 @@ set(llcommon_SOURCE_FILES llworkerthread.cpp timing.cpp u64.cpp + workqueue.cpp StackWalker.cpp ) @@ -258,6 +259,7 @@ set(llcommon_HEADER_FILES timer.h tuple.h u64.h + workqueue.h StackWalker.h ) @@ -363,6 +365,7 @@ if (LL_TESTS) LL_ADD_INTEGRATION_TEST(stringize "" "${test_libs}") LL_ADD_INTEGRATION_TEST(threadsafeschedule "" "${test_libs}") LL_ADD_INTEGRATION_TEST(tuple "" "${test_libs}") + LL_ADD_INTEGRATION_TEST(workqueue "" "${test_libs}") ## llexception_test.cpp isn't a regression test, and doesn't need to be run ## every build. It's to help a developer make implementation choices about diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp new file mode 100644 index 0000000000..ab1cae6c14 --- /dev/null +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -0,0 +1,158 @@ +/** + * @file workqueue_test.cpp + * @author Nat Goodspeed + * @date 2021-10-07 + * @brief Test for workqueue. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "workqueue.h" +// STL headers +// std headers +#include +#include +// external library headers +// other Linden headers +#include "../test/lltut.h" +#include "llcond.h" +#include "llstring.h" +#include "stringize.h" + +using namespace LL; +using namespace std::literals::chrono_literals; // ms suffix +using namespace std::literals::string_literals; // s suffix + +/***************************************************************************** +* TUT +*****************************************************************************/ +namespace tut +{ + struct workqueue_data + { + WorkQueue queue{"queue"}; + }; + typedef test_group workqueue_group; + typedef workqueue_group::object object; + workqueue_group workqueuegrp("workqueue"); + + template<> template<> + void object::test<1>() + { + set_test_name("name"); + ensure_equals("didn't capture name", queue.getKey(), "queue"); + ensure("not findable", WorkQueue::getInstance("queue") == queue.getWeak().lock()); + WorkQueue q2; + ensure("has no name", LLStringUtil::startsWith(q2.getKey(), "WorkQueue")); + } + + template<> template<> + void object::test<2>() + { + set_test_name("post"); + bool wasRun{ false }; + // We only get away with binding a simple bool because we're running + // the work on the same thread. + queue.post([&wasRun](){ wasRun = true; }); + queue.close(); + ensure("ran too soon", ! wasRun); + queue.runUntilClose(); + ensure("didn't run", wasRun); + } + + template<> template<> + void object::test<3>() + { + set_test_name("postEvery"); + // record of runs + using Shared = std::deque; + // This is an example of how to share data between the originator of + // postEvery(work) and the work item itself, since usually a WorkQueue + // is used to dispatch work to a different thread. Neither of them + // should call any of LLCond's wait methods: you don't want to stall + // either the worker thread or the originating thread (conventionally + // main). Use LLCond or a subclass even if all you want to do is + // signal the work item that it can quit; consider LLOneShotCond. + LLCond data; + auto start = WorkQueue::TimePoint::clock::now(); + auto interval = 100ms; + queue.postEvery( + interval, + [&data, count = 0] + () mutable + { + // record the timestamp at which this instance is running + data.update_one( + [](Shared& data) + { + data.push_back(WorkQueue::TimePoint::clock::now()); + }); + // by the 3rd call, return false to stop + return (++count < 3); + }); + // no convenient way to close() our queue while we've got a + // postEvery() running, so run until we think we should have exhausted + // the iterations + queue.runFor(10*interval); + // Take a copy of the captured deque. + Shared result = data.get(); + ensure_equals("called wrong number of times", result.size(), 3); + // postEvery() assumes you want the first call to happen right away. + // Inject a fake start time that's (interval) earlier than that, to + // make our too early/too late tests uniform for all entries. + result.push_front(start - interval); + for (size_t i = 1; i < result.size(); ++i) + { + auto diff = (result[i] - result[i-1]); + try + { + ensure(STRINGIZE("call " << i << " too soon"), diff >= interval); + ensure(STRINGIZE("call " << i << " too late"), diff < interval*1.5); + } + catch (const tut::failure&) + { + auto interval_ms = interval / 1ms; + auto diff_ms = diff / 1ms; + std::cerr << "interval " << interval_ms + << "ms; diff " << diff_ms << "ms" << std::endl; + throw; + } + } + } + + template<> template<> + void object::test<4>() + { + set_test_name("postTo"); + WorkQueue main("main"); + auto qptr = WorkQueue::getInstance("queue"); + int result = 0; + main.postTo( + qptr, + [](){ return 17; }, + // Note that a postTo() *callback* can safely bind a reference to + // a variable on the invoking thread, because the callback is run + // on the invoking thread. + [&result](int i){ result = i; }); + // this should post the callback to main + qptr->runOne(); + // this should run the callback + main.runOne(); + ensure_equals("failed to run int callback", result, 17); + + std::string alpha; + // postTo() handles arbitrary return types + main.postTo( + qptr, + [](){ return "abc"s; }, + [&alpha](const std::string& s){ alpha = s; }); + qptr->runPending(); + main.runPending(); + ensure_equals("failed to run string callback", alpha, "abc"); + } +} // namespace tut diff --git a/indra/llcommon/threadsafeschedule.h b/indra/llcommon/threadsafeschedule.h index 0e70d30714..c8ad23532b 100644 --- a/indra/llcommon/threadsafeschedule.h +++ b/indra/llcommon/threadsafeschedule.h @@ -78,6 +78,7 @@ namespace LL enum pop_result { EMPTY=super::EMPTY, DONE=super::DONE, WAITING=super::WAITING, POPPED=super::POPPED }; public: + using Closed = LLThreadSafeQueueInterrupt; using TimePoint = ThreadSafeSchedulePrivate::TimePoint; using Clock = TimePoint::clock; diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp new file mode 100644 index 0000000000..15e292fb43 --- /dev/null +++ b/indra/llcommon/workqueue.cpp @@ -0,0 +1,114 @@ +/** + * @file workqueue.cpp + * @author Nat Goodspeed + * @date 2021-10-06 + * @brief Implementation for WorkQueue. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "workqueue.h" +// STL headers +// std headers +// external library headers +// other Linden headers +#include "llerror.h" +#include "llexception.h" +#include "stringize.h" + +LL::WorkQueue::WorkQueue(const std::string& name): + super(makeName(name)) +{ + // TODO: register for "LLApp" events so we can implicitly close() on + // viewer shutdown. +} + +void LL::WorkQueue::close() +{ + mQueue.close(); +} + +void LL::WorkQueue::runUntilClose() +{ + try + { + for (;;) + { + callWork(mQueue.pop()); + } + } + catch (const Queue::Closed&) + { + } +} + +bool LL::WorkQueue::runPending() +{ + for (Work work; mQueue.tryPop(work); ) + { + callWork(work); + } + return ! mQueue.done(); +} + +bool LL::WorkQueue::runOne() +{ + Work work; + if (mQueue.tryPop(work)) + { + callWork(work); + } + return ! mQueue.done(); +} + +bool LL::WorkQueue::runUntil(const TimePoint& until) +{ + // Should we subtract some slop to allow for typical Work execution time? + // How much slop? + Work work; + while (TimePoint::clock::now() < until && mQueue.tryPopUntil(until, work)) + { + callWork(work); + } + return ! mQueue.done(); +} + +std::string LL::WorkQueue::makeName(const std::string& name) +{ + if (! name.empty()) + return name; + + static thread_local U32 discriminator = 0; + return STRINGIZE("WorkQueue" << discriminator++); +} + +void LL::WorkQueue::callWork(const Queue::DataTuple& work) +{ + // ThreadSafeSchedule::pop() always delivers a tuple, even when + // there's only one data field per item, as for us. + callWork(std::get<0>(work)); +} + +void LL::WorkQueue::callWork(const Work& work) +{ + try + { + work(); + } + catch (...) + { + // No matter what goes wrong with any individual work item, the worker + // thread must go on! Log our own instance name with the exception. + LOG_UNHANDLED_EXCEPTION(getKey()); + } +} + +void LL::WorkQueue::error(const std::string& msg) +{ + LL_ERRS("WorkQueue") << msg << LL_ENDL; +} diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h new file mode 100644 index 0000000000..a52f7b0e26 --- /dev/null +++ b/indra/llcommon/workqueue.h @@ -0,0 +1,325 @@ +/** + * @file workqueue.h + * @author Nat Goodspeed + * @date 2021-09-30 + * @brief Queue used for inter-thread work passing. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_WORKQUEUE_H) +#define LL_WORKQUEUE_H + +#include "llinstancetracker.h" +#include "threadsafeschedule.h" +#include +#include // std::function +#include +#include +#include // std::pair +#include + +namespace LL +{ + /** + * A typical WorkQueue has a string name that can be used to find it. + */ + class WorkQueue: public LLInstanceTracker + { + private: + using super = LLInstanceTracker; + + public: + using Work = std::function; + + private: + using Queue = ThreadSafeSchedule; + // helper for postEvery() + template + class BackJack; + + public: + using TimePoint = Queue::TimePoint; + using TimedWork = Queue::TimeTuple; + using Closed = Queue::Closed; + + /** + * You may omit the WorkQueue name, in which case a unique name is + * synthesized; for practical purposes that makes it anonymous. + */ + WorkQueue(const std::string& name = std::string()); + + /** + * Since the point of WorkQueue is to pass work to some other worker + * thread(s) asynchronously, it's important that the WorkQueue continue + * to exist until the worker thread(s) have drained it. To communicate + * that it's time for them to quit, close() the queue. + */ + void close(); + + /*---------------------- fire and forget API -----------------------*/ + + /// fire-and-forget, but at a particular (future?) time + template + void post(const TimePoint& time, CALLABLE&& callable) + { + // Defer reifying an arbitrary CALLABLE until we hit this method. + // All other methods should accept CALLABLEs of arbitrary type to + // avoid multiple levels of std::function indirection. + mQueue.push(TimedWork(time, std::move(callable))); + } + + /// fire-and-forget + template + void post(CALLABLE&& callable) + { + // We use TimePoint::clock::now() instead of TimePoint's + // representation of the epoch because this WorkQueue may contain + // a mix of past-due TimedWork items and TimedWork items scheduled + // for the future. Sift this new item into the correct place. + post(TimePoint::clock::now(), std::move(callable)); + } + + /** + * Launch a callable returning bool that will trigger repeatedly at + * specified interval, until the callable returns false. + * + * If you need to signal that callable from outside, DO NOT bind a + * reference to a simple bool! That's not thread-safe. Instead, bind + * an LLCond variant, e.g. LLOneShotCond or LLBoolCond. + */ + template + void postEvery(const std::chrono::duration& interval, + CALLABLE&& callable); + + /*------------------------- handshake API --------------------------*/ + + /** + * Post work to another WorkQueue to be run at a specified time, + * requesting a specific callback to be run on this WorkQueue on + * completion. + * + * Returns true if able to post, false if the other WorkQueue is + * inaccessible. + */ + template + bool postTo(std::weak_ptr target, + const TimePoint& time, CALLABLE&& callable, CALLBACK&& callback) + { + // We're being asked to post to the WorkQueue at target. + // target is a weak_ptr: have to lock it to check it. + auto tptr = target.lock(); + if (! tptr) + // can't post() if the target WorkQueue has been destroyed + return false; + + // Here we believe target WorkQueue still exists. Post to it a + // lambda that packages our callable, our callback and a weak_ptr + // to this originating WorkQueue. + tptr->post( + time, + [reply = super::getWeak(), + callable = std::move(callable), + callback = std::move(callback)] + () + { + // Call the callable in any case -- but to minimize + // copying the result, immediately bind it into a reply + // lambda. The reply lambda also binds the original + // callback, so that when we, the originating WorkQueue, + // finally receive and process the reply lambda, we'll + // call the bound callback with the bound result -- on the + // same thread that originally called postTo(). + auto rlambda = + [result = callable(), + callback = std::move(callback)] + () + { callback(std::move(result)); }; + // Check if this originating WorkQueue still exists. + // Remember, the outer lambda is now running on a thread + // servicing the target WorkQueue, and real time has + // elapsed since postTo()'s tptr->post() call. + // reply is a weak_ptr: have to lock it to check it. + auto rptr = reply.lock(); + if (rptr) + { + // Only post reply lambda if the originating WorkQueue + // still exists. If not -- who would we tell? Log it? + try + { + rptr->post(std::move(rlambda)); + } + catch (const Closed&) + { + // Originating WorkQueue might still exist, but + // might be Closed. Same thing: just discard the + // callback. + } + } + }); + // looks like we were able to post() + return true; + } + + /** + * Post work to another WorkQueue, requesting a specific callback to + * be run on this WorkQueue on completion. + * + * Returns true if able to post, false if the other WorkQueue is + * inaccessible. + */ + template + bool postTo(std::weak_ptr target, + CALLABLE&& callable, CALLBACK&& callback) + { + return postTo(target, TimePoint::clock::now(), std::move(callable), std::move(callback)); + } + + /*--------------------------- worker API ---------------------------*/ + + /** + * runUntilClose() pulls TimedWork items off this WorkQueue until the + * queue is closed, at which point it returns. This would be the + * typical entry point for a simple worker thread. + */ + void runUntilClose(); + + /** + * runPending() runs all TimedWork items that are ready to run. It + * returns true if the queue remains open, false if the queue has been + * closed. This could be used by a thread whose primary purpose is to + * serve the queue, but also wants to do other things with its idle time. + */ + bool runPending(); + + /** + * runOne() runs at most one ready TimedWork item -- zero if none are + * ready. It returns true if the queue remains open, false if the + * queue has been closed. + */ + bool runOne(); + + /** + * runFor() runs a subset of ready TimedWork items, until the + * timeslice has been exceeded. It returns true if the queue remains + * open, false if the queue has been closed. This could be used by a + * busy main thread to lend a bounded few CPU cycles to this WorkQueue + * without risking the WorkQueue blowing out the length of any one + * frame. + */ + template + bool runFor(const std::chrono::duration& timeslice) + { + return runUntil(TimePoint::clock::now() + timeslice); + } + + /** + * runUntil() is just like runFor(), only with a specific end time + * instead of a timeslice duration. + */ + bool runUntil(const TimePoint& until); + + private: + static void error(const std::string& msg); + static std::string makeName(const std::string& name); + void callWork(const Queue::DataTuple& work); + void callWork(const Work& work); + Queue mQueue; + }; + + /** + * BackJack is, in effect, a hand-rolled lambda, binding a WorkQueue, a + * CALLABLE that returns bool, a TimePoint and an interval at which to + * relaunch it. As long as the callable continues returning true, BackJack + * keeps resubmitting it to the target WorkQueue. + */ + // Why is BackJack a class and not a lambda? Because, unlike a lambda, a + // class method gets its own 'this' pointer -- which we need to resubmit + // the whole BackJack callable. + template + class WorkQueue::BackJack + { + public: + // bind the desired data + BackJack(std::weak_ptr target, + const WorkQueue::TimePoint& start, + const std::chrono::duration& interval, + CALLABLE&& callable): + mTarget(target), + mStart(start), + mInterval(interval), + mCallable(std::move(callable)) + {} + + // Call by target WorkQueue -- note that although WE require a + // callable returning bool, WorkQueue wants a void callable. We + // consume the bool. + void operator()() + { + // If mCallable() throws an exception, don't catch it here: if it + // throws once, it's likely to throw every time, so it's a waste + // of time to arrange to call it again. + if (mCallable()) + { + // Modify mStart to the new start time we desire. If we simply + // added mInterval to now, we'd get actual timings of + // (mInterval + slop), where 'slop' is the latency between the + // previous mStart and the WorkQueue actually calling us. + // Instead, add mInterval to mStart so that at least we + // register our intent to fire at exact mIntervals. + mStart += mInterval; + + // We're being called at this moment by the target WorkQueue. + // Assume it still exists, rather than checking the result of + // lock(). + // Resubmit the whole *this callable: that's why we're a class + // rather than a lambda. Allow moving *this so we can carry a + // move-only callable; but naturally this statement must be + // the last time we reference this instance, which may become + // moved-from. + try + { + mTarget.lock()->post(mStart, std::move(*this)); + } + catch (const Closed&) + { + // Once this queue is closed, oh well, just stop + } + } + } + + private: + std::weak_ptr mTarget; + WorkQueue::TimePoint mStart; + std::chrono::duration mInterval; + CALLABLE mCallable; + }; + + template + void WorkQueue::postEvery(const std::chrono::duration& interval, + CALLABLE&& callable) + { + if (interval.count() <= 0) + { + // It's essential that postEvery() be called with a positive + // interval, since each call to BackJack posts another instance of + // itself at (start + interval) and we order by target time. A + // zero or negative interval would result in that BackJack + // instance going to the head of the queue every time, immediately + // ready to run. Effectively that would produce an infinite loop, + // a denial of service on this WorkQueue. + error("postEvery(interval) may not be 0"); + } + // Instantiate and post a suitable BackJack, binding a weak_ptr to + // self, the current time, the desired interval and the desired + // callable. + post( + BackJack( + getWeak(), TimePoint::clock::now(), interval, std::move(callable))); + } + +} // namespace LL + +#endif /* ! defined(LL_WORKQUEUE_H) */ -- cgit v1.3 From c585ddb75e383cdd994d0d99fed8f2de8f955e3c Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 7 Oct 2021 16:45:15 -0400 Subject: SL-16024: Defend against two threads making "anonymous" WorkQueues. Also make workqueue_test.cpp more robust. --- indra/llcommon/tests/workqueue_test.cpp | 11 ++++++----- indra/llcommon/workqueue.cpp | 18 ++++++++++++++++-- 2 files changed, 22 insertions(+), 7 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp index ab1cae6c14..d5405400fd 100644 --- a/indra/llcommon/tests/workqueue_test.cpp +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -103,12 +103,13 @@ namespace tut Shared result = data.get(); ensure_equals("called wrong number of times", result.size(), 3); // postEvery() assumes you want the first call to happen right away. - // Inject a fake start time that's (interval) earlier than that, to - // make our too early/too late tests uniform for all entries. - result.push_front(start - interval); - for (size_t i = 1; i < result.size(); ++i) + // Pretend our start time was (interval) earlier than that, to make + // our too early/too late tests uniform for all entries. + start -= interval; + for (size_t i = 0; i < result.size(); ++i) { - auto diff = (result[i] - result[i-1]); + auto diff = result[i] - start; + start += interval; try { ensure(STRINGIZE("call " << i << " too soon"), diff >= interval); diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index 15e292fb43..ffc9a97dc0 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -17,10 +17,15 @@ // std headers // external library headers // other Linden headers +#include "llcoros.h" +#include LLCOROS_MUTEX_HEADER #include "llerror.h" #include "llexception.h" #include "stringize.h" +using Mutex = LLCoros::Mutex; +using Lock = LLCoros::LockType; + LL::WorkQueue::WorkQueue(const std::string& name): super(makeName(name)) { @@ -83,8 +88,17 @@ std::string LL::WorkQueue::makeName(const std::string& name) if (! name.empty()) return name; - static thread_local U32 discriminator = 0; - return STRINGIZE("WorkQueue" << discriminator++); + static U32 discriminator = 0; + static Mutex mutex; + U32 num; + { + // Protect discriminator from concurrent access by different threads. + // It can't be thread_local, else two racing threads will come up with + // the same name. + Lock lk(mutex); + num = discriminator++; + } + return STRINGIZE("WorkQueue" << num); } void LL::WorkQueue::callWork(const Queue::DataTuple& work) -- cgit v1.3 From 54d874b1233586844f87e79ae8f211af0a1cb7a6 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Fri, 8 Oct 2021 11:52:09 -0400 Subject: SL-16024: Resolve bizarre VS compile error. Thanks Callum! It seems CALLBACK is a macro in some Microsoft header file. Bleah. --- indra/llcommon/workqueue.h | 20 ++++++++++++-------- 1 file changed, 12 insertions(+), 8 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index a52f7b0e26..b88aef989a 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -104,9 +104,13 @@ namespace LL * Returns true if able to post, false if the other WorkQueue is * inaccessible. */ - template - bool postTo(std::weak_ptr target, - const TimePoint& time, CALLABLE&& callable, CALLBACK&& callback) + // Apparently some Microsoft header file defines a macro CALLBACK? The + // natural template argument name CALLBACK produces very weird Visual + // Studio compile errors that seem utterly unrelated to this source + // code. + template + bool postTo(WorkQueue::weak_t target, + const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) { // We're being asked to post to the WorkQueue at target. // target is a weak_ptr: have to lock it to check it. @@ -170,9 +174,9 @@ namespace LL * Returns true if able to post, false if the other WorkQueue is * inaccessible. */ - template - bool postTo(std::weak_ptr target, - CALLABLE&& callable, CALLBACK&& callback) + template + bool postTo(WorkQueue::weak_t target, + CALLABLE&& callable, FOLLOWUP&& callback) { return postTo(target, TimePoint::clock::now(), std::move(callable), std::move(callback)); } @@ -243,7 +247,7 @@ namespace LL { public: // bind the desired data - BackJack(std::weak_ptr target, + BackJack(WorkQueue::weak_t target, const WorkQueue::TimePoint& start, const std::chrono::duration& interval, CALLABLE&& callable): @@ -291,7 +295,7 @@ namespace LL } private: - std::weak_ptr mTarget; + WorkQueue::weak_t mTarget; WorkQueue::TimePoint mStart; std::chrono::duration mInterval; CALLABLE mCallable; -- cgit v1.3 From d00272e0cc9974f35a46f0c313ee2c0e11cddbda Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 11 Oct 2021 16:03:40 +0000 Subject: SL-16099 Multi-threaded OpenGL usage on Windows, enable Core Profile and VAOs by default. --- indra/llcommon/llthreadsafequeue.h | 3 + indra/llimage/llimageworker.cpp | 4 + indra/llrender/llcubemap.cpp | 2 + indra/llrender/llfontfreetype.cpp | 1 + indra/llrender/llglslshader.cpp | 2 + indra/llrender/llgltexture.cpp | 2 + indra/llrender/llimagegl.cpp | 330 +++++++++++++++++++++----------- indra/llrender/llimagegl.h | 35 +++- indra/llrender/llrender.cpp | 2 +- indra/llrender/llrender.h | 2 +- indra/llwindow/llopenglview-objc.mm | 3 +- indra/llwindow/llwindow.h | 12 +- indra/llwindow/llwindowheadless.h | 3 + indra/llwindow/llwindowmacosx.cpp | 28 +++ indra/llwindow/llwindowmacosx.h | 144 +++++++------- indra/llwindow/llwindowwin32.cpp | 129 ++++++++----- indra/llwindow/llwindowwin32.h | 3 + indra/newview/app_settings/settings.xml | 4 +- indra/newview/llappviewer.cpp | 3 +- indra/newview/llnetmap.cpp | 1 + indra/newview/lltexturefetch.cpp | 4 + indra/newview/llviewertexture.cpp | 311 +++++++++++++++++++----------- indra/newview/llviewertexture.h | 8 +- indra/newview/llviewertexturelist.cpp | 36 ++++ indra/newview/llviewerwindow.cpp | 2 +- 25 files changed, 718 insertions(+), 356 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 719edcd579..06e8d8f609 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -154,6 +154,9 @@ public: // Returns the size of the queue. size_t size(); + //Returns the capacity of the queue. + U32 capacity() { return mCapacity; } + // closes the queue: // - every subsequent push() call will throw LLThreadSafeQueueInterrupt // - every subsequent tryPush() call will return false diff --git a/indra/llimage/llimageworker.cpp b/indra/llimage/llimageworker.cpp index 5f42fba866..33f8dce6ee 100644 --- a/indra/llimage/llimageworker.cpp +++ b/indra/llimage/llimageworker.cpp @@ -48,6 +48,7 @@ LLImageDecodeThread::~LLImageDecodeThread() // virtual S32 LLImageDecodeThread::update(F32 max_time_ms) { + LL_PROFILE_ZONE_SCOPED; LLMutexLock lock(mCreationMutex); for (creation_list_t::iterator iter = mCreationList.begin(); iter != mCreationList.end(); ++iter) @@ -71,6 +72,7 @@ S32 LLImageDecodeThread::update(F32 max_time_ms) LLImageDecodeThread::handle_t LLImageDecodeThread::decodeImage(LLImageFormatted* image, U32 priority, S32 discard, BOOL needs_aux, Responder* responder) { + LL_PROFILE_ZONE_SCOPED; LLMutexLock lock(mCreationMutex); handle_t handle = generateHandle(); mCreationList.push_back(creation_info(handle, image, priority, discard, needs_aux, responder)); @@ -118,6 +120,7 @@ LLImageDecodeThread::ImageRequest::~ImageRequest() // Returns true when done, whether or not decode was successful. bool LLImageDecodeThread::ImageRequest::processRequest() { + LL_PROFILE_ZONE_SCOPED; const F32 decode_time_slice = .1f; bool done = true; if (!mDecodedRaw && mFormattedImage.notNull()) @@ -164,6 +167,7 @@ bool LLImageDecodeThread::ImageRequest::processRequest() void LLImageDecodeThread::ImageRequest::finishRequest(bool completed) { + LL_PROFILE_ZONE_SCOPED; if (mResponder.notNull()) { bool success = completed && mDecodedRaw && (!mNeedsAux || mDecodedAux); diff --git a/indra/llrender/llcubemap.cpp b/indra/llrender/llcubemap.cpp index 5947bca670..d7f7b2f58e 100644 --- a/indra/llrender/llcubemap.cpp +++ b/indra/llrender/llcubemap.cpp @@ -150,6 +150,7 @@ void LLCubeMap::initRawData(const std::vector >& rawimages void LLCubeMap::initGLData() { + LL_PROFILE_ZONE_SCOPED; for (int i = 0; i < 6; i++) { mImages[i]->setSubImage(mRawImages[i], 0, 0, RESOLUTION, RESOLUTION); @@ -453,6 +454,7 @@ BOOL LLCubeMap::project(F32& v_min, F32& v_max, F32& h_min, F32& h_max, void LLCubeMap::paintIn(LLVector3 dir[4], const LLColor4U& col) { + LL_PROFILE_ZONE_SCOPED; F32 v_min, v_max, h_min, h_max; LLVector3 center = dir[0] + dir[1] + dir[2] + dir[3]; center.normVec(); diff --git a/indra/llrender/llfontfreetype.cpp b/indra/llrender/llfontfreetype.cpp index c41730ebaa..e18161e53c 100644 --- a/indra/llrender/llfontfreetype.cpp +++ b/indra/llrender/llfontfreetype.cpp @@ -460,6 +460,7 @@ LLFontGlyphInfo* LLFontFreetype::addGlyph(llwchar wch) const LLFontGlyphInfo* LLFontFreetype::addGlyphFromFont(const LLFontFreetype *fontp, llwchar wch, U32 glyph_index) const { + LL_PROFILE_ZONE_SCOPED; if (mFTFace == NULL) return NULL; diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp index 2fb3b8257d..394fcd2b2f 100644 --- a/indra/llrender/llglslshader.cpp +++ b/indra/llrender/llglslshader.cpp @@ -208,6 +208,7 @@ void LLGLSLShader::dumpStats() //static void LLGLSLShader::startProfile() { + LL_PROFILE_ZONE_SCOPED; if (sProfileEnabled && sCurBoundShaderPtr) { sCurBoundShaderPtr->placeProfileQuery(); @@ -218,6 +219,7 @@ void LLGLSLShader::startProfile() //static void LLGLSLShader::stopProfile(U32 count, U32 mode) { + LL_PROFILE_ZONE_SCOPED; if (sProfileEnabled && sCurBoundShaderPtr) { sCurBoundShaderPtr->readProfileQuery(count, mode); diff --git a/indra/llrender/llgltexture.cpp b/indra/llrender/llgltexture.cpp index ad501687ed..a279e85bae 100644 --- a/indra/llrender/llgltexture.cpp +++ b/indra/llrender/llgltexture.cpp @@ -262,6 +262,7 @@ LLTexUnit::eTextureType LLGLTexture::getTarget(void) const BOOL LLGLTexture::setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height) { + LL_PROFILE_ZONE_SCOPED; llassert(mGLTexturep.notNull()) ; return mGLTexturep->setSubImage(imageraw, x_pos, y_pos, width, height) ; @@ -269,6 +270,7 @@ BOOL LLGLTexture::setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, BOOL LLGLTexture::setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height) { + LL_PROFILE_ZONE_SCOPED; llassert(mGLTexturep.notNull()) ; return mGLTexturep->setSubImage(datap, data_width, data_height, x_pos, y_pos, width, height) ; diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 276fa55e15..aff29bd857 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -39,6 +39,7 @@ #include "llgl.h" #include "llglslshader.h" #include "llrender.h" +#include "llwindow.h" //---------------------------------------------------------------------------- const F32 MIN_TEXTURE_LIFETIME = 10.f; @@ -170,15 +171,32 @@ BOOL is_little_endian() return (*c == 0x78) ; } + +LLImageGLThread* LLImageGLThread::sInstance = nullptr; + //static -void LLImageGL::initClass(S32 num_catagories, BOOL skip_analyze_alpha /* = false */) +void LLImageGL::initClass(LLWindow* window, S32 num_catagories, BOOL skip_analyze_alpha /* = false */) { + LL_PROFILE_ZONE_SCOPED; sSkipAnalyzeAlpha = skip_analyze_alpha; + LLImageGLThread::sInstance = new LLImageGLThread(window); + LLImageGLThread::sInstance->start(); +} + +//static +void LLImageGL::updateClass() +{ + LL_PROFILE_ZONE_SCOPED; + LLImageGLThread::sInstance->executeCallbacks(); } //static void LLImageGL::cleanupClass() -{ +{ + LL_PROFILE_ZONE_SCOPED; + LLImageGLThread::sInstance->mFunctionQueue.close(); + delete LLImageGLThread::sInstance; + LLImageGLThread::sInstance = nullptr; } //static @@ -656,6 +674,7 @@ void LLImageGL::setExplicitFormat( LLGLint internal_format, LLGLenum primary_for void LLImageGL::setImage(const LLImageRaw* imageraw) { + LL_PROFILE_ZONE_SCOPED; llassert((imageraw->getWidth() == getWidth(mCurrentDiscardLevel)) && (imageraw->getHeight() == getHeight(mCurrentDiscardLevel)) && (imageraw->getComponents() == getComponents())); @@ -699,9 +718,8 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips) } llverify(gGL.getTexUnit(0)->bind(this)); - - - if (mUseMipMaps) + + if (mUseMipMaps) { if (data_hasmips) { @@ -781,7 +799,7 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips) glTexParameteri(mTarget, GL_GENERATE_MIPMAP, GL_TRUE); } - LLImageGL::setManualImage(mTarget, 0, mFormatInternal, + LLImageGL::setManualImage(mTarget, 0, mFormatInternal, w, h, mFormatPrimary, mFormatType, data_in, mAllowCompression); @@ -878,7 +896,7 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips) stop_glerror(); } - LLImageGL::setManualImage(mTarget, m, mFormatInternal, w, h, mFormatPrimary, mFormatType, cur_mip_data, mAllowCompression); + LLImageGL::setManualImage(mTarget, m, mFormatInternal, w, h, mFormatPrimary, mFormatType, cur_mip_data, mAllowCompression); if (m == 0) { analyzeAlpha(data_in, w, h); @@ -1067,6 +1085,7 @@ void LLImageGL::postAddToAtlas() BOOL LLImageGL::setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update) { + LL_PROFILE_ZONE_SCOPED; if (!width || !height) { return TRUE; @@ -1163,6 +1182,7 @@ BOOL LLImageGL::setSubImage(const U8* datap, S32 data_width, S32 data_height, S3 BOOL LLImageGL::setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update) { + LL_PROFILE_ZONE_SCOPED; return setSubImage(imageraw->getData(), imageraw->getWidth(), imageraw->getHeight(), x_pos, y_pos, width, height, force_fast_update); } @@ -1201,119 +1221,119 @@ void LLImageGL::deleteTextures(S32 numTextures, U32 *textures) // static static LLTrace::BlockTimerStatHandle FTM_SET_MANUAL_IMAGE("setManualImage"); -void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void *pixels, bool allow_compression) +void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void* pixels, bool allow_compression) { - LL_RECORD_BLOCK_TIME(FTM_SET_MANUAL_IMAGE); - bool use_scratch = false; - U32* scratch = NULL; - if (LLRender::sGLCoreProfile) - { - if (pixformat == GL_ALPHA && pixtype == GL_UNSIGNED_BYTE) - { //GL_ALPHA is deprecated, convert to RGBA - use_scratch = true; - scratch = new U32[width*height]; + LL_RECORD_BLOCK_TIME(FTM_SET_MANUAL_IMAGE); + bool use_scratch = false; + U32* scratch = NULL; + if (LLRender::sGLCoreProfile) + { + if (pixformat == GL_ALPHA && pixtype == GL_UNSIGNED_BYTE) + { //GL_ALPHA is deprecated, convert to RGBA + use_scratch = true; + scratch = new U32[width * height]; - U32 pixel_count = (U32) (width*height); - for (U32 i = 0; i < pixel_count; i++) - { - U8* pix = (U8*) &scratch[i]; - pix[0] = pix[1] = pix[2] = 0; - pix[3] = ((U8*) pixels)[i]; - } - - pixformat = GL_RGBA; - intformat = GL_RGBA8; - } + U32 pixel_count = (U32)(width * height); + for (U32 i = 0; i < pixel_count; i++) + { + U8* pix = (U8*)&scratch[i]; + pix[0] = pix[1] = pix[2] = 0; + pix[3] = ((U8*)pixels)[i]; + } - if (pixformat == GL_LUMINANCE_ALPHA && pixtype == GL_UNSIGNED_BYTE) - { //GL_LUMINANCE_ALPHA is deprecated, convert to RGBA - use_scratch = true; - scratch = new U32[width*height]; + pixformat = GL_RGBA; + intformat = GL_RGBA8; + } - U32 pixel_count = (U32) (width*height); - for (U32 i = 0; i < pixel_count; i++) - { - U8 lum = ((U8*) pixels)[i*2+0]; - U8 alpha = ((U8*) pixels)[i*2+1]; + if (pixformat == GL_LUMINANCE_ALPHA && pixtype == GL_UNSIGNED_BYTE) + { //GL_LUMINANCE_ALPHA is deprecated, convert to RGBA + use_scratch = true; + scratch = new U32[width * height]; - U8* pix = (U8*) &scratch[i]; - pix[0] = pix[1] = pix[2] = lum; - pix[3] = alpha; - } - - pixformat = GL_RGBA; - intformat = GL_RGBA8; - } + U32 pixel_count = (U32)(width * height); + for (U32 i = 0; i < pixel_count; i++) + { + U8 lum = ((U8*)pixels)[i * 2 + 0]; + U8 alpha = ((U8*)pixels)[i * 2 + 1]; - if (pixformat == GL_LUMINANCE && pixtype == GL_UNSIGNED_BYTE) - { //GL_LUMINANCE_ALPHA is deprecated, convert to RGB - use_scratch = true; - scratch = new U32[width*height]; + U8* pix = (U8*)&scratch[i]; + pix[0] = pix[1] = pix[2] = lum; + pix[3] = alpha; + } - U32 pixel_count = (U32) (width*height); - for (U32 i = 0; i < pixel_count; i++) - { - U8 lum = ((U8*) pixels)[i]; - - U8* pix = (U8*) &scratch[i]; - pix[0] = pix[1] = pix[2] = lum; - pix[3] = 255; - } - - pixformat = GL_RGBA; - intformat = GL_RGB8; - } - } + pixformat = GL_RGBA; + intformat = GL_RGBA8; + } - if (LLImageGL::sCompressTextures && allow_compression) - { - switch (intformat) - { - case GL_RGB: - case GL_RGB8: - intformat = GL_COMPRESSED_RGB; - break; - case GL_SRGB: - case GL_SRGB8: - intformat = GL_COMPRESSED_SRGB; - break; - case GL_RGBA: - case GL_RGBA8: - intformat = GL_COMPRESSED_RGBA; - break; - case GL_SRGB_ALPHA: - case GL_SRGB8_ALPHA8: - intformat = GL_COMPRESSED_SRGB_ALPHA; - break; - case GL_LUMINANCE: - case GL_LUMINANCE8: - intformat = GL_COMPRESSED_LUMINANCE; - break; - case GL_LUMINANCE_ALPHA: - case GL_LUMINANCE8_ALPHA8: - intformat = GL_COMPRESSED_LUMINANCE_ALPHA; - break; - case GL_ALPHA: - case GL_ALPHA8: - intformat = GL_COMPRESSED_ALPHA; - break; - default: - LL_WARNS() << "Could not compress format: " << std::hex << intformat << LL_ENDL; - break; - } - } + if (pixformat == GL_LUMINANCE && pixtype == GL_UNSIGNED_BYTE) + { //GL_LUMINANCE_ALPHA is deprecated, convert to RGB + use_scratch = true; + scratch = new U32[width * height]; - stop_glerror(); - { - LL_PROFILE_ZONE_NAMED("glTexImage2D"); - glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels); - } - stop_glerror(); + U32 pixel_count = (U32)(width * height); + for (U32 i = 0; i < pixel_count; i++) + { + U8 lum = ((U8*)pixels)[i]; - if (use_scratch) - { - delete [] scratch; - } + U8* pix = (U8*)&scratch[i]; + pix[0] = pix[1] = pix[2] = lum; + pix[3] = 255; + } + + pixformat = GL_RGBA; + intformat = GL_RGB8; + } + } + + if (LLImageGL::sCompressTextures && allow_compression) + { + switch (intformat) + { + case GL_RGB: + case GL_RGB8: + intformat = GL_COMPRESSED_RGB; + break; + case GL_SRGB: + case GL_SRGB8: + intformat = GL_COMPRESSED_SRGB; + break; + case GL_RGBA: + case GL_RGBA8: + intformat = GL_COMPRESSED_RGBA; + break; + case GL_SRGB_ALPHA: + case GL_SRGB8_ALPHA8: + intformat = GL_COMPRESSED_SRGB_ALPHA; + break; + case GL_LUMINANCE: + case GL_LUMINANCE8: + intformat = GL_COMPRESSED_LUMINANCE; + break; + case GL_LUMINANCE_ALPHA: + case GL_LUMINANCE8_ALPHA8: + intformat = GL_COMPRESSED_LUMINANCE_ALPHA; + break; + case GL_ALPHA: + case GL_ALPHA8: + intformat = GL_COMPRESSED_ALPHA; + break; + default: + LL_WARNS() << "Could not compress format: " << std::hex << intformat << LL_ENDL; + break; + } + } + + stop_glerror(); + { + LL_PROFILE_ZONE_NAMED("glTexImage2D"); + glTexImage2D(target, miplevel, intformat, width, height, 0, pixformat, pixtype, use_scratch ? scratch : pixels); + } + stop_glerror(); + + if (use_scratch) + { + delete[] scratch; + } } //create an empty GL texture: just create a texture name @@ -1336,6 +1356,7 @@ BOOL LLImageGL::createGLTexture() if(mTexName) { LLImageGL::deleteTextures(1, (reinterpret_cast(&mTexName))) ; + mTexName = 0; } @@ -1697,7 +1718,7 @@ void LLImageGL::destroyGLTexture() mTextureMemory = (S32Bytes)0; } - LLImageGL::deleteTextures(1, &mTexName); + LLImageGL::deleteTextures(1, &mTexName); mCurrentDiscardLevel = -1 ; //invalidate mCurrentDiscardLevel. mTexName = 0; mGLTextureCreated = FALSE ; @@ -2238,3 +2259,90 @@ void LLImageGL::resetCurTexSizebar() glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_BASE_LEVEL, 0); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, nummips); */ + +LLImageGLThread::LLImageGLThread(LLWindow* window) + : LLThread("LLImageGL"), mWindow(window) +{ + mFinished = false; + + mContext = mWindow->createSharedContext(); +} + +// post a function to be executed on the LLImageGL background thread + +bool LLImageGLThread::post(const std::function& func) +{ + try + { + if (mFunctionQueue.size() < mFunctionQueue.capacity()) + { + //NOTE: tryPushFront will return immediately if the lock is held + // desired behavior here is to push and return true unless the + // queue is full or closed + mFunctionQueue.pushFront(func); + } + else + { + return false; + } + } + catch (LLThreadSafeQueueInterrupt e) + { + return false; + } + + return true; +} + +//post a callback to be executed on the main thread + +bool LLImageGLThread::postCallback(const std::function& callback) +{ + try + { + mCallbackQueue.pushFront(callback); + } + catch (LLThreadSafeQueueInterrupt e) + { + //thread is closing, drop request + return false; + } + + return true; +} + +void LLImageGLThread::executeCallbacks() +{ + LL_PROFILE_ZONE_SCOPED; + //executed from main thread + std::function callback; + while (mCallbackQueue.tryPopBack(callback)) + { + LL_PROFILE_ZONE_NAMED("iglt - callback"); + callback(); + } +} + +void LLImageGLThread::run() +{ + mWindow->makeContextCurrent(mContext); + gGL.init(); + try + { + while (true) + { + LL_PROFILE_ZONE_SCOPED; + std::function curFunc = mFunctionQueue.popBack(); + { + LL_PROFILE_ZONE_NAMED("iglt - function") + curFunc(); + } + } + } + catch (LLThreadSafeQueueInterrupt e) + { + //queue is closed, fall out of run loop + } + gGL.shutdown(); + mWindow->destroySharedContext(mContext); +} diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h index 61ddc8d59b..8e9b483c2d 100644 --- a/indra/llrender/llimagegl.h +++ b/indra/llrender/llimagegl.h @@ -35,9 +35,11 @@ #include "llrefcount.h" #include "v2math.h" #include "llunits.h" - +#include "llthreadsafequeue.h" #include "llrender.h" class LLTextureAtlas ; +class LLWindow; + #define BYTES_TO_MEGA_BYTES(x) ((x) >> 20) #define MEGA_BYTES_TO_BYTES(x) ((x) << 20) @@ -102,7 +104,7 @@ public: void setAllowCompression(bool allow) { mAllowCompression = allow; } static void setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void *pixels, bool allow_compression = true); - + BOOL createGLTexture() ; BOOL createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename = 0, BOOL to_create = TRUE, S32 category = sMaxCategories-1); @@ -265,7 +267,8 @@ public: #endif public: - static void initClass(S32 num_catagories, BOOL skip_analyze_alpha = false); + static void initClass(LLWindow* window, S32 num_catagories, BOOL skip_analyze_alpha = false); + static void updateClass(); static void cleanupClass() ; private: @@ -301,4 +304,30 @@ public: }; +class LLImageGLThread : public LLThread +{ +public: + LLImageGLThread(LLWindow* window); + + // post a function to be executed on the LLImageGL background thread + bool post(const std::function& func); + + //post a callback to be executed on the main thread + bool postCallback(const std::function& callback); + + void executeCallbacks(); + + void run() override; + + LLThreadSafeQueue> mFunctionQueue; + LLThreadSafeQueue> mCallbackQueue; + + LLWindow* mWindow; + void* mContext; + LLAtomicBool mFinished; + + static LLImageGLThread* sInstance; +}; + + #endif // LL_LLIMAGEGL_H diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 43b4441ea8..b6711e44e3 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -36,7 +36,7 @@ #include "lltexture.h" #include "llshadermgr.h" -LLRender gGL; +thread_local LLRender gGL; // Handy copies of last good GL matrices F32 gGLModelView[16]; diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h index af8568f8a3..c08c2d6881 100644 --- a/indra/llrender/llrender.h +++ b/indra/llrender/llrender.h @@ -511,7 +511,7 @@ extern F32 gGLLastProjection[16]; extern F32 gGLProjection[16]; extern S32 gGLViewport[4]; -extern LLRender gGL; +extern thread_local LLRender gGL; // This rotation matrix moves the default OpenGL reference frame // (-Z at, Y up) to Cory's favorite reference frame (X at, Z up) diff --git a/indra/llwindow/llopenglview-objc.mm b/indra/llwindow/llopenglview-objc.mm index d2c5b11c3d..b647085b7e 100644 --- a/indra/llwindow/llopenglview-objc.mm +++ b/indra/llwindow/llopenglview-objc.mm @@ -288,7 +288,8 @@ attributedStringInfo getSegments(NSAttributedString *str) if (vsync) { - [glContext setValues:(const GLint*)1 forParameter:NSOpenGLCPSwapInterval]; + GLint value = 1; + [glContext setValues:&value forParameter:NSOpenGLCPSwapInterval]; } else { // supress this error after move to Xcode 7: // error: null passed to a callee that requires a non-null argument [-Werror,-Wnonnull] diff --git a/indra/llwindow/llwindow.h b/indra/llwindow/llwindow.h index d4d5b76937..10c0b6a424 100644 --- a/indra/llwindow/llwindow.h +++ b/indra/llwindow/llwindow.h @@ -78,7 +78,17 @@ public: BOOL setSize(LLCoordWindow size); virtual void setMinSize(U32 min_width, U32 min_height, bool enforce_immediately = true); virtual BOOL switchContext(BOOL fullscreen, const LLCoordScreen &size, BOOL disable_vsync, const LLCoordScreen * const posp = NULL) = 0; - virtual BOOL setCursorPosition(LLCoordWindow position) = 0; + + //create a new GL context that shares a namespace with this Window's main GL context and make it current on the current thread + // returns a pointer to be handed back to destroySharedConext/makeContextCurrent + virtual void* createSharedContext() = 0; + //make the given context current on the current thread + virtual void makeContextCurrent(void* context) = 0; + //destroy the given context that was retrieved by createSharedContext() + //Must be called on the same thread that called createSharedContext() + virtual void destroySharedContext(void* context) = 0; + + virtual BOOL setCursorPosition(LLCoordWindow position) = 0; virtual BOOL getCursorPosition(LLCoordWindow *position) = 0; virtual void showCursor() = 0; virtual void hideCursor() = 0; diff --git a/indra/llwindow/llwindowheadless.h b/indra/llwindow/llwindowheadless.h index c692666df1..a7ae28aa24 100644 --- a/indra/llwindow/llwindowheadless.h +++ b/indra/llwindow/llwindowheadless.h @@ -49,6 +49,9 @@ public: /*virtual*/ BOOL setSizeImpl(LLCoordScreen size) {return FALSE;}; /*virtual*/ BOOL setSizeImpl(LLCoordWindow size) {return FALSE;}; /*virtual*/ BOOL switchContext(BOOL fullscreen, const LLCoordScreen &size, BOOL disable_vsync, const LLCoordScreen * const posp = NULL) {return FALSE;}; + void* createSharedContext() { return nullptr; } + void makeContextCurrent(void*) {} + void destroySharedContext(void*) {} /*virtual*/ BOOL setCursorPosition(LLCoordWindow position) {return FALSE;}; /*virtual*/ BOOL getCursorPosition(LLCoordWindow *position) {return FALSE;}; /*virtual*/ void showCursor() {}; diff --git a/indra/llwindow/llwindowmacosx.cpp b/indra/llwindow/llwindowmacosx.cpp index 0d0607a0bb..23830dd24e 100644 --- a/indra/llwindow/llwindowmacosx.cpp +++ b/indra/llwindow/llwindowmacosx.cpp @@ -1907,6 +1907,34 @@ void LLWindowMacOSX::allowLanguageTextInput(LLPreeditor *preeditor, BOOL b) allowDirectMarkedTextInput(b, mGLView); // mLanguageTextInputAllowed and mMarkedTextAllowed should be updated at once (by Pell Smit } +class sharedContext +{ +public: + CGLContextObj mContext; +}; + +void* LLWindowMacOSX::createSharedContext() +{ + sharedContext* sc = new sharedContext(); + CGLCreateContext(mPixelFormat, mContext, &(sc->mContext)); + + return (void *)sc; +} + +void LLWindowMacOSX::makeContextCurrent(void* context) +{ + CGLSetCurrentContext(((sharedContext*)context)->mContext); +} + +void LLWindowMacOSX::destroySharedContext(void* context) +{ + sharedContext* sc = (sharedContext*)context; + + CGLDestroyContext(sc->mContext); + + delete sc; +} + void LLWindowMacOSX::interruptLanguageTextInput() { commitCurrentPreedit(mGLView); diff --git a/indra/llwindow/llwindowmacosx.h b/indra/llwindow/llwindowmacosx.h index bf45238c8d..ede2b453d5 100644 --- a/indra/llwindow/llwindowmacosx.h +++ b/indra/llwindow/llwindowmacosx.h @@ -41,85 +41,84 @@ #undef verify #undef require - class LLWindowMacOSX : public LLWindow { public: - /*virtual*/ void show(); - /*virtual*/ void hide(); - /*virtual*/ void close(); - /*virtual*/ BOOL getVisible(); - /*virtual*/ BOOL getMinimized(); - /*virtual*/ BOOL getMaximized(); - /*virtual*/ BOOL maximize(); - /*virtual*/ void minimize(); - /*virtual*/ void restore(); - /*virtual*/ BOOL getFullscreen(); - /*virtual*/ BOOL getPosition(LLCoordScreen *position); - /*virtual*/ BOOL getSize(LLCoordScreen *size); - /*virtual*/ BOOL getSize(LLCoordWindow *size); - /*virtual*/ BOOL setPosition(LLCoordScreen position); - /*virtual*/ BOOL setSizeImpl(LLCoordScreen size); - /*virtual*/ BOOL setSizeImpl(LLCoordWindow size); - /*virtual*/ BOOL switchContext(BOOL fullscreen, const LLCoordScreen &size, BOOL disable_vsync, const LLCoordScreen * const posp = NULL); - /*virtual*/ BOOL setCursorPosition(LLCoordWindow position); - /*virtual*/ BOOL getCursorPosition(LLCoordWindow *position); - /*virtual*/ void showCursor(); - /*virtual*/ void hideCursor(); - /*virtual*/ void showCursorFromMouseMove(); - /*virtual*/ void hideCursorUntilMouseMove(); - /*virtual*/ BOOL isCursorHidden(); - /*virtual*/ void updateCursor(); - /*virtual*/ ECursorType getCursor() const; - /*virtual*/ void captureMouse(); - /*virtual*/ void releaseMouse(); - /*virtual*/ void setMouseClipping( BOOL b ); - /*virtual*/ BOOL isClipboardTextAvailable(); - /*virtual*/ BOOL pasteTextFromClipboard(LLWString &dst); - /*virtual*/ BOOL copyTextToClipboard(const LLWString & src); - /*virtual*/ void flashIcon(F32 seconds); - /*virtual*/ F32 getGamma(); - /*virtual*/ BOOL setGamma(const F32 gamma); // Set the gamma - /*virtual*/ U32 getFSAASamples(); - /*virtual*/ void setFSAASamples(const U32 fsaa_samples); - /*virtual*/ BOOL restoreGamma(); // Restore original gamma table (before updating gamma) - /*virtual*/ ESwapMethod getSwapMethod() { return mSwapMethod; } - /*virtual*/ void gatherInput(); - /*virtual*/ void delayInputProcessing() {}; - /*virtual*/ void swapBuffers(); + void show() override; + void hide() override; + void close() override; + BOOL getVisible() override; + BOOL getMinimized() override; + BOOL getMaximized() override; + BOOL maximize() override; + void minimize() override; + void restore() override; + BOOL getFullscreen(); + BOOL getPosition(LLCoordScreen *position) override; + BOOL getSize(LLCoordScreen *size) override; + BOOL getSize(LLCoordWindow *size) override; + BOOL setPosition(LLCoordScreen position) override; + BOOL setSizeImpl(LLCoordScreen size) override; + BOOL setSizeImpl(LLCoordWindow size) override; + BOOL switchContext(BOOL fullscreen, const LLCoordScreen &size, BOOL disable_vsync, const LLCoordScreen * const posp = NULL) override; + BOOL setCursorPosition(LLCoordWindow position) override; + BOOL getCursorPosition(LLCoordWindow *position) override; + void showCursor() override; + void hideCursor() override; + void showCursorFromMouseMove() override; + void hideCursorUntilMouseMove() override; + BOOL isCursorHidden() override; + void updateCursor() override; + ECursorType getCursor() const override; + void captureMouse() override; + void releaseMouse() override; + void setMouseClipping( BOOL b ) override; + BOOL isClipboardTextAvailable() override; + BOOL pasteTextFromClipboard(LLWString &dst) override; + BOOL copyTextToClipboard(const LLWString & src) override; + void flashIcon(F32 seconds) override; + F32 getGamma() override; + BOOL setGamma(const F32 gamma) override; // Set the gamma + U32 getFSAASamples() override; + void setFSAASamples(const U32 fsaa_samples) override; + BOOL restoreGamma() override; // Restore original gamma table (before updating gamma) + ESwapMethod getSwapMethod() override { return mSwapMethod; } + void gatherInput() override; + void delayInputProcessing() override {}; + void swapBuffers() override; // handy coordinate space conversion routines - /*virtual*/ BOOL convertCoords(LLCoordScreen from, LLCoordWindow *to); - /*virtual*/ BOOL convertCoords(LLCoordWindow from, LLCoordScreen *to); - /*virtual*/ BOOL convertCoords(LLCoordWindow from, LLCoordGL *to); - /*virtual*/ BOOL convertCoords(LLCoordGL from, LLCoordWindow *to); - /*virtual*/ BOOL convertCoords(LLCoordScreen from, LLCoordGL *to); - /*virtual*/ BOOL convertCoords(LLCoordGL from, LLCoordScreen *to); + BOOL convertCoords(LLCoordScreen from, LLCoordWindow *to) override; + BOOL convertCoords(LLCoordWindow from, LLCoordScreen *to) override; + BOOL convertCoords(LLCoordWindow from, LLCoordGL *to) override; + BOOL convertCoords(LLCoordGL from, LLCoordWindow *to) override; + BOOL convertCoords(LLCoordScreen from, LLCoordGL *to) override; + BOOL convertCoords(LLCoordGL from, LLCoordScreen *to) override; - /*virtual*/ LLWindowResolution* getSupportedResolutions(S32 &num_resolutions); - /*virtual*/ F32 getNativeAspectRatio(); - /*virtual*/ F32 getPixelAspectRatio(); - /*virtual*/ void setNativeAspectRatio(F32 ratio) { mOverrideAspectRatio = ratio; } + LLWindowResolution* getSupportedResolutions(S32 &num_resolutions) override; + F32 getNativeAspectRatio() override; + F32 getPixelAspectRatio() override; + void setNativeAspectRatio(F32 ratio) override { mOverrideAspectRatio = ratio; } - /*virtual*/ void beforeDialog(); - /*virtual*/ void afterDialog(); + void beforeDialog() override; + void afterDialog() override; - /*virtual*/ BOOL dialogColorPicker(F32 *r, F32 *g, F32 *b); + BOOL dialogColorPicker(F32 *r, F32 *g, F32 *b) override; - /*virtual*/ void *getPlatformWindow(); - /*virtual*/ void bringToFront() {}; + void *getPlatformWindow() override; + void bringToFront() override {}; - /*virtual*/ void allowLanguageTextInput(LLPreeditor *preeditor, BOOL b); - /*virtual*/ void interruptLanguageTextInput(); - /*virtual*/ void spawnWebBrowser(const std::string& escaped_url, bool async); - /*virtual*/ F32 getSystemUISize(); + void allowLanguageTextInput(LLPreeditor *preeditor, BOOL b) override; + void interruptLanguageTextInput() override; + void spawnWebBrowser(const std::string& escaped_url, bool async) override; + F32 getSystemUISize() override; static std::vector getDisplaysResolutionList(); static std::vector getDynamicFallbackFontList(); // Provide native key event data - /*virtual*/ LLSD getNativeKeyData(); + LLSD getNativeKeyData() override; void* getWindow() { return mWindow; } LLWindowCallbacks* getCallbacks() { return mCallbacks; } @@ -132,6 +131,15 @@ public: bool allowsLanguageInput() { return mLanguageTextInputAllowed; } + //create a new GL context that shares a namespace with this Window's main GL context and make it current on the current thread + // returns a pointer to be handed back to destroySharedConext/makeContextCurrent + void* createSharedContext() override; + //make the given context current on the current thread + void makeContextCurrent(void* context) override; + //destroy the given context that was retrieved by createSharedContext() + //Must be called on the same thread that called createSharedContext() + void destroySharedContext(void* context) override; + protected: LLWindowMacOSX(LLWindowCallbacks* callbacks, const std::string& title, const std::string& name, int x, int y, int width, int height, U32 flags, @@ -141,7 +149,7 @@ protected: ~LLWindowMacOSX(); void initCursors(); - BOOL isValid(); + BOOL isValid() override; void moveWindow(const LLCoordScreen& position,const LLCoordScreen& size); @@ -157,7 +165,7 @@ protected: BOOL shouldPostQuit() { return mPostQuit; } //Satisfy MAINT-3135 and MAINT-3288 with a flag. - /*virtual */ void setOldResize(bool oldresize) {setResizeMode(oldresize, mGLView); } + /*virtual */ void setOldResize(bool oldresize) override {setResizeMode(oldresize, mGLView); } private: void restoreGLContext(); @@ -231,9 +239,9 @@ public: LLSplashScreenMacOSX(); virtual ~LLSplashScreenMacOSX(); - /*virtual*/ void showImpl(); - /*virtual*/ void updateImpl(const std::string& mesg); - /*virtual*/ void hideImpl(); + void showImpl(); + void updateImpl(const std::string& mesg); + void hideImpl(); private: WindowRef mWindow; diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index 4ce7c30bef..12d4c6c30e 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -1774,58 +1774,11 @@ const S32 max_format = (S32)num_formats - 1; mhRC = 0; if (wglCreateContextAttribsARB) { //attempt to create a specific versioned context - S32 attribs[] = - { //start at 4.2 - WGL_CONTEXT_MAJOR_VERSION_ARB, 4, - WGL_CONTEXT_MINOR_VERSION_ARB, 2, - WGL_CONTEXT_PROFILE_MASK_ARB, LLRender::sGLCoreProfile ? WGL_CONTEXT_CORE_PROFILE_BIT_ARB : WGL_CONTEXT_COMPATIBILITY_PROFILE_BIT_ARB, - WGL_CONTEXT_FLAGS_ARB, gDebugGL ? WGL_CONTEXT_DEBUG_BIT_ARB : 0, - 0 - }; - - bool done = false; - while (!done) - { - mhRC = wglCreateContextAttribsARB(mhDC, mhRC, attribs); - - if (!mhRC) - { - if (attribs[3] > 0) - { //decrement minor version - attribs[3]--; - } - else if (attribs[1] > 3) - { //decrement major version and start minor version over at 3 - attribs[1]--; - attribs[3] = 3; - } - else - { //we reached 3.0 and still failed, bail out - done = true; - } - } - else - { - LL_INFOS() << "Created OpenGL " << llformat("%d.%d", attribs[1], attribs[3]) << - (LLRender::sGLCoreProfile ? " core" : " compatibility") << " context." << LL_ENDL; - done = true; - - // force sNoFixedFunction iff we're trying to use nsight debugging which does not support many legacy API uses - - // nSight doesn't support use of legacy API funcs in the fixed function pipe - if (LLRender::sGLCoreProfile || LLRender::sNsightDebugSupport) - { - LLGLSLShader::sNoFixedFunction = true; - } - } - } - } - - if (!mhRC && !(mhRC = wglCreateContext(mhDC))) - { - close(); - OSMessageBox(mCallbacks->translateString("MBGLContextErr"), mCallbacks->translateString("MBError"), OSMB_OK); - return FALSE; + mhRC = (HGLRC) createSharedContext(); + if (!mhRC) + { + return FALSE; + } } if (!wglMakeCurrent(mhDC, mhRC)) @@ -1880,6 +1833,75 @@ const S32 max_format = (S32)num_formats - 1; return TRUE; } +void* LLWindowWin32::createSharedContext() +{ + S32 attribs[] = + { + WGL_CONTEXT_MAJOR_VERSION_ARB, 4, + WGL_CONTEXT_MINOR_VERSION_ARB, 2, + WGL_CONTEXT_PROFILE_MASK_ARB, LLRender::sGLCoreProfile ? WGL_CONTEXT_CORE_PROFILE_BIT_ARB : WGL_CONTEXT_COMPATIBILITY_PROFILE_BIT_ARB, + WGL_CONTEXT_FLAGS_ARB, gDebugGL ? WGL_CONTEXT_DEBUG_BIT_ARB : 0, + 0 + }; + + HGLRC rc = wglCreateContextAttribsARB(mhDC, mhRC, attribs); + + bool done = false; + while (!done) + { + rc = wglCreateContextAttribsARB(mhDC, mhRC, attribs); + + if (!rc) + { + if (attribs[3] > 0) + { //decrement minor version + attribs[3]--; + } + else if (attribs[1] > 3) + { //decrement major version and start minor version over at 3 + attribs[1]--; + attribs[3] = 3; + } + else + { //we reached 3.0 and still failed, bail out + done = true; + } + } + else + { + LL_INFOS() << "Created OpenGL " << llformat("%d.%d", attribs[1], attribs[3]) << + (LLRender::sGLCoreProfile ? " core" : " compatibility") << " context." << LL_ENDL; + done = true; + + // force sNoFixedFunction iff we're trying to use nsight debugging which does not support many legacy API uses + + // nSight doesn't support use of legacy API funcs in the fixed function pipe + if (LLRender::sGLCoreProfile || LLRender::sNsightDebugSupport) + { + LLGLSLShader::sNoFixedFunction = true; + } + } + } + + if (!rc && !(rc = wglCreateContext(mhDC))) + { + close(); + OSMessageBox(mCallbacks->translateString("MBGLContextErr"), mCallbacks->translateString("MBError"), OSMB_OK); + } + + return rc; +} + +void LLWindowWin32::makeContextCurrent(void* contextPtr) +{ + wglMakeCurrent(mhDC, (HGLRC) contextPtr); +} + +void LLWindowWin32::destroySharedContext(void* contextPtr) +{ + wglDeleteContext((HGLRC)contextPtr); +} + void LLWindowWin32::moveWindow( const LLCoordScreen& position, const LLCoordScreen& size ) { if( mIsMouseClipping ) @@ -3720,6 +3742,7 @@ BOOL LLWindowWin32::resetDisplayResolution() void LLWindowWin32::swapBuffers() { + LL_PROFILE_ZONE_SCOPED; ASSERT_MAIN_THREAD(); SwapBuffers(mhDC); @@ -4731,4 +4754,4 @@ void LLWindowWin32::post(const std::function& func) #else mFunctionQueue.pushFront(func); #endif -} \ No newline at end of file +} diff --git a/indra/llwindow/llwindowwin32.h b/indra/llwindow/llwindowwin32.h index 66647459b2..5f253b5df3 100644 --- a/indra/llwindow/llwindowwin32.h +++ b/indra/llwindow/llwindowwin32.h @@ -93,6 +93,9 @@ public: /*virtual*/ BOOL setSizeImpl(LLCoordScreen size); /*virtual*/ BOOL setSizeImpl(LLCoordWindow size); /*virtual*/ BOOL switchContext(BOOL fullscreen, const LLCoordScreen &size, BOOL disable_vsync, const LLCoordScreen * const posp = NULL); + void* createSharedContext() override; + void makeContextCurrent(void* context) override; + void destroySharedContext(void* context) override; /*virtual*/ BOOL setCursorPosition(LLCoordWindow position); /*virtual*/ BOOL getCursorPosition(LLCoordWindow *position); /*virtual*/ void showCursor(); diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index 05c3fc3bfe..d05d100ddf 100644 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -9814,7 +9814,7 @@ Type Boolean Value - 0 + 1 RenderGlow @@ -10599,7 +10599,7 @@ Type Boolean Value - 0 + 1 RenderUseTransformFeedback diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 722a6caa65..e46e8feb14 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -613,7 +613,7 @@ static void settings_modify() LLPipeline::sRenderDeferred = LLPipeline::sRenderTransparentWater && LLPipeline::sRenderBump && gSavedSettings.getBOOL("RenderDeferred"); LLVOSurfacePatch::sLODFactor = gSavedSettings.getF32("RenderTerrainLODFactor"); LLVOSurfacePatch::sLODFactor *= LLVOSurfacePatch::sLODFactor; //square lod factor to get exponential range of [1,4] - gDebugGL = gSavedSettings.getBOOL("RenderDebugGL") || gDebugSession; + gDebugGL = gSavedSettings.getBOOL("RenderDebugGL") || gDebugSession; gDebugPipeline = gSavedSettings.getBOOL("RenderDebugPipeline"); } @@ -4831,6 +4831,7 @@ void LLAppViewer::idle() LLNotificationsUI::LLToast::updateClass(); LLSmoothInterpolation::updateInterpolants(); LLMortician::updateClass(); + LLImageGL::updateClass(); LLFilePickerThread::clearDead(); //calls LLFilePickerThread::notify() LLDirPickerThread::clearDead(); F32 dt_raw = idle_timer.getElapsedTimeAndResetF32(); diff --git a/indra/newview/llnetmap.cpp b/indra/newview/llnetmap.cpp index 112da55682..111b45612e 100644 --- a/indra/newview/llnetmap.cpp +++ b/indra/newview/llnetmap.cpp @@ -147,6 +147,7 @@ void LLNetMap::setScale( F32 scale ) void LLNetMap::draw() { + LL_PROFILE_ZONE_SCOPED; static LLFrameTimer map_timer; static LLUIColor map_avatar_color = LLUIColorTable::instance().getColor("MapAvatarColor", LLColor4::white); static LLUIColor map_avatar_friend_color = LLUIColorTable::instance().getColor("MapAvatarFriendColor", LLColor4::white); diff --git a/indra/newview/lltexturefetch.cpp b/indra/newview/lltexturefetch.cpp index f64db7beb5..63e561147d 100644 --- a/indra/newview/lltexturefetch.cpp +++ b/indra/newview/lltexturefetch.cpp @@ -314,6 +314,7 @@ private: // Threads: Ttc virtual void completed(bool success) { + LL_PROFILE_ZONE_SCOPED; LLTextureFetchWorker* worker = mFetcher->getWorker(mID); if (worker) { @@ -338,6 +339,7 @@ private: // Threads: Ttc virtual void completed(bool success) { + LL_PROFILE_ZONE_SCOPED; LLTextureFetchWorker* worker = mFetcher->getWorker(mID); if (worker) { @@ -362,6 +364,7 @@ private: // Threads: Tid virtual void completed(bool success, LLImageRaw* raw, LLImageRaw* aux) { + LL_PROFILE_ZONE_SCOPED; LLTextureFetchWorker* worker = mFetcher->getWorker(mID); if (worker) { @@ -1139,6 +1142,7 @@ void LLTextureFetchWorker::startWork(S32 param) // Threads: Ttf bool LLTextureFetchWorker::doWork(S32 param) { + LL_PROFILE_ZONE_SCOPED; static const LLCore::HttpStatus http_not_found(HTTP_NOT_FOUND); // 404 static const LLCore::HttpStatus http_service_unavail(HTTP_SERVICE_UNAVAILABLE); // 503 static const LLCore::HttpStatus http_not_sat(HTTP_REQUESTED_RANGE_NOT_SATISFIABLE); // 416; diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index af55f68cd2..274f53a160 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -184,6 +184,7 @@ void LLViewerTextureManager::findFetchedTextures(const LLUUID& id, std::vector &output) { + LL_PROFILE_ZONE_SCOPED; std::vector fetched_output; gTextureList.findTexturesByID(id, fetched_output); std::vector::iterator iter = fetched_output.begin(); @@ -485,6 +486,7 @@ static LLTrace::BlockTimerStatHandle FTM_TEXTURE_MEMORY_CHECK("Memory Check"); //static bool LLViewerTexture::isMemoryForTextureLow() { + LL_PROFILE_ZONE_SCOPED; // Note: we need to figure out a better source for 'min' values, // what is free for low end at minimal settings is 'nothing left' // for higher end gpus at high settings. @@ -501,6 +503,7 @@ bool LLViewerTexture::isMemoryForTextureLow() //static bool LLViewerTexture::isMemoryForTextureSuficientlyFree() { + LL_PROFILE_ZONE_SCOPED; const S32Megabytes DESIRED_FREE_TEXTURE_MEMORY(50); const S32Megabytes DESIRED_FREE_MAIN_MEMORY(200); @@ -514,6 +517,7 @@ bool LLViewerTexture::isMemoryForTextureSuficientlyFree() //static void LLViewerTexture::getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &physical) { + LL_PROFILE_ZONE_SCOPED; static LLFrameTimer timer; static S32Megabytes gpu_res = S32Megabytes(S32_MAX); static S32Megabytes physical_res = S32Megabytes(S32_MAX); @@ -526,27 +530,29 @@ void LLViewerTexture::getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &p } timer.reset(); - LL_RECORD_BLOCK_TIME(FTM_TEXTURE_MEMORY_CHECK); - - if (gGLManager.mHasATIMemInfo) { - S32 meminfo[4]; - glGetIntegerv(GL_TEXTURE_FREE_MEMORY_ATI, meminfo); - gpu_res = (S32Megabytes)meminfo[0]; + LL_RECORD_BLOCK_TIME(FTM_TEXTURE_MEMORY_CHECK); - //check main memory, only works for windows. - LLMemory::updateMemoryInfo(); - physical_res = LLMemory::getAvailableMemKB(); - } - else if (gGLManager.mHasNVXMemInfo) - { - S32 free_memory; - glGetIntegerv(GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &free_memory); - gpu_res = (S32Megabytes)(free_memory / 1024); - } + if (gGLManager.mHasATIMemInfo) + { + S32 meminfo[4]; + glGetIntegerv(GL_TEXTURE_FREE_MEMORY_ATI, meminfo); + gpu_res = (S32Megabytes)meminfo[0]; - gpu = gpu_res; - physical = physical_res; + //check main memory, only works for windows. + LLMemory::updateMemoryInfo(); + physical_res = LLMemory::getAvailableMemKB(); + } + else if (gGLManager.mHasNVXMemInfo) + { + S32 free_memory; + glGetIntegerv(GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &free_memory); + gpu_res = (S32Megabytes)(free_memory / 1024); + } + + gpu = gpu_res; + physical = physical_res; + } } static LLTrace::BlockTimerStatHandle FTM_TEXTURE_UPDATE_MEDIA("Media"); @@ -555,6 +561,7 @@ static LLTrace::BlockTimerStatHandle FTM_TEXTURE_UPDATE_TEST("Test"); //static void LLViewerTexture::updateClass(const F32 velocity, const F32 angular_velocity) { + LL_PROFILE_ZONE_SCOPED; sCurrentTime = gFrameTimeSeconds; LLTexturePipelineTester* tester = (LLTexturePipelineTester*)LLMetricPerformanceTesterBasic::getTester(sTesterName); @@ -704,6 +711,7 @@ void LLViewerTexture::cleanup() void LLViewerTexture::notifyAboutCreatingTexture() { + LL_PROFILE_ZONE_SCOPED; for(U32 ch = 0; ch < LLRender::NUM_TEXTURE_CHANNELS; ++ch) { for(U32 f = 0; f < mNumFaces[ch]; f++) @@ -715,6 +723,7 @@ void LLViewerTexture::notifyAboutCreatingTexture() void LLViewerTexture::notifyAboutMissingAsset() { + LL_PROFILE_ZONE_SCOPED; for(U32 ch = 0; ch < LLRender::NUM_TEXTURE_CHANNELS; ++ch) { for(U32 f = 0; f < mNumFaces[ch]; f++) @@ -727,6 +736,7 @@ void LLViewerTexture::notifyAboutMissingAsset() // virtual void LLViewerTexture::dump() { + LL_PROFILE_ZONE_SCOPED; LLGLTexture::dump(); LL_INFOS() << "LLViewerTexture" @@ -762,6 +772,7 @@ bool LLViewerTexture::isActiveFetching() bool LLViewerTexture::bindDebugImage(const S32 stage) { + LL_PROFILE_ZONE_SCOPED; if (stage < 0) return false; bool res = true; @@ -780,6 +791,7 @@ bool LLViewerTexture::bindDebugImage(const S32 stage) bool LLViewerTexture::bindDefaultImage(S32 stage) { + LL_PROFILE_ZONE_SCOPED; if (stage < 0) return false; bool res = true; @@ -822,6 +834,7 @@ void LLViewerTexture::forceImmediateUpdate() void LLViewerTexture::addTextureStats(F32 virtual_size, BOOL needs_gltexture) const { + LL_PROFILE_ZONE_SCOPED; if(needs_gltexture) { mNeedsGLTexture = TRUE; @@ -864,6 +877,7 @@ void LLViewerTexture::setKnownDrawSize(S32 width, S32 height) //virtual void LLViewerTexture::addFace(U32 ch, LLFace* facep) { + LL_PROFILE_ZONE_SCOPED; llassert(ch < LLRender::NUM_TEXTURE_CHANNELS); if(mNumFaces[ch] >= mFaceList[ch].size()) @@ -879,6 +893,7 @@ void LLViewerTexture::addFace(U32 ch, LLFace* facep) //virtual void LLViewerTexture::removeFace(U32 ch, LLFace* facep) { + LL_PROFILE_ZONE_SCOPED; llassert(ch < LLRender::NUM_TEXTURE_CHANNELS); if(mNumFaces[ch] > 1) @@ -919,6 +934,7 @@ S32 LLViewerTexture::getNumFaces(U32 ch) const //virtual void LLViewerTexture::addVolume(U32 ch, LLVOVolume* volumep) { + LL_PROFILE_ZONE_SCOPED; if (mNumVolumes[ch] >= mVolumeList[ch].size()) { mVolumeList[ch].resize(2 * mNumVolumes[ch] + 1); @@ -932,6 +948,7 @@ void LLViewerTexture::addVolume(U32 ch, LLVOVolume* volumep) //virtual void LLViewerTexture::removeVolume(U32 ch, LLVOVolume* volumep) { + LL_PROFILE_ZONE_SCOPED; if (mNumVolumes[ch] > 1) { S32 index = volumep->getIndexInTex(ch); @@ -955,6 +972,7 @@ S32 LLViewerTexture::getNumVolumes(U32 ch) const void LLViewerTexture::reorganizeFaceList() { + LL_PROFILE_ZONE_SCOPED; static const F32 MAX_WAIT_TIME = 20.f; // seconds static const U32 MAX_EXTRA_BUFFER_SIZE = 4; @@ -978,6 +996,7 @@ void LLViewerTexture::reorganizeFaceList() void LLViewerTexture::reorganizeVolumeList() { + LL_PROFILE_ZONE_SCOPED; static const F32 MAX_WAIT_TIME = 20.f; // seconds static const U32 MAX_EXTRA_BUFFER_SIZE = 4; @@ -1180,6 +1199,7 @@ FTType LLViewerFetchedTexture::getFTType() const void LLViewerFetchedTexture::cleanup() { + LL_PROFILE_ZONE_SCOPED; for(callback_list_t::iterator iter = mLoadedCallbackList.begin(); iter != mLoadedCallbackList.end(); ) { @@ -1205,6 +1225,7 @@ void LLViewerFetchedTexture::cleanup() //access the fast cache void LLViewerFetchedTexture::loadFromFastCache() { + LL_PROFILE_ZONE_SCOPED; if(!mInFastCacheList) { return; //no need to access the fast cache. @@ -1350,6 +1371,7 @@ void LLViewerFetchedTexture::dump() // ONLY called from LLViewerFetchedTextureList void LLViewerFetchedTexture::destroyTexture() { + LL_PROFILE_ZONE_SCOPED; if(LLImageGL::sGlobalTextureMemory < sMaxDesiredTextureMem * 0.95f)//not ready to release unused memory. { return ; @@ -1366,6 +1388,7 @@ void LLViewerFetchedTexture::destroyTexture() void LLViewerFetchedTexture::addToCreateTexture() { + LL_PROFILE_ZONE_SCOPED; bool force_update = false; if (getComponents() != mRawImage->getComponents()) { @@ -1407,6 +1430,7 @@ void LLViewerFetchedTexture::addToCreateTexture() } else { + LL_PROFILE_ZONE_SCOPED; #if 1 // //if mRequestedDiscardLevel > mDesiredDiscardLevel, we assume the required image res keep going up, @@ -1451,99 +1475,100 @@ void LLViewerFetchedTexture::addToCreateTexture() } } #endif - mNeedsCreateTexture = TRUE; - gTextureList.mCreateTextureList.insert(this); - } + scheduleCreateTexture(); + } return; } // ONLY called from LLViewerTextureList -BOOL LLViewerFetchedTexture::createTexture(S32 usename/*= 0*/) +BOOL LLViewerFetchedTexture::preCreateTexture(S32 usename/*= 0*/) { - if (!mNeedsCreateTexture) - { - destroyRawImage(); - return FALSE; - } - mNeedsCreateTexture = FALSE; - if (mRawImage.isNull()) - { - LL_ERRS() << "LLViewerTexture trying to create texture with no Raw Image" << LL_ENDL; - } - if (mRawImage->isBufferInvalid()) - { - LL_WARNS() << "Can't create a texture: invalid image data" << LL_ENDL; - destroyRawImage(); - return FALSE; - } -// LL_INFOS() << llformat("IMAGE Creating (%d) [%d x %d] Bytes: %d ", -// mRawDiscardLevel, -// mRawImage->getWidth(), mRawImage->getHeight(),mRawImage->getDataSize()) -// << mID.getString() << LL_ENDL; - BOOL res = TRUE; + LL_PROFILE_ZONE_SCOPED; + if (!mNeedsCreateTexture) + { + destroyRawImage(); + return FALSE; + } + mNeedsCreateTexture = FALSE; - // store original size only for locally-sourced images - if (mUrl.compare(0, 7, "file://") == 0) - { - mOrigWidth = mRawImage->getWidth(); - mOrigHeight = mRawImage->getHeight(); + if (mRawImage.isNull()) + { + LL_ERRS() << "LLViewerTexture trying to create texture with no Raw Image" << LL_ENDL; + } + if (mRawImage->isBufferInvalid()) + { + LL_WARNS() << "Can't create a texture: invalid image data" << LL_ENDL; + destroyRawImage(); + return FALSE; + } + // LL_INFOS() << llformat("IMAGE Creating (%d) [%d x %d] Bytes: %d ", + // mRawDiscardLevel, + // mRawImage->getWidth(), mRawImage->getHeight(),mRawImage->getDataSize()) + // << mID.getString() << LL_ENDL; + BOOL res = TRUE; + + // store original size only for locally-sourced images + if (mUrl.compare(0, 7, "file://") == 0) + { + mOrigWidth = mRawImage->getWidth(); + mOrigHeight = mRawImage->getHeight(); // This is only safe because it's a local image and fetcher doesn't use raw data // from local images, but this might become unsafe in case of changes to fetcher - if (mBoostLevel == BOOST_PREVIEW) - { - mRawImage->biasedScaleToPowerOfTwo(1024); - } - else - { // leave black border, do not scale image content - mRawImage->expandToPowerOfTwo(MAX_IMAGE_SIZE, FALSE); - } - - mFullWidth = mRawImage->getWidth(); - mFullHeight = mRawImage->getHeight(); - setTexelsPerImage(); - } - else - { - mOrigWidth = mFullWidth; - mOrigHeight = mFullHeight; - } + if (mBoostLevel == BOOST_PREVIEW) + { + mRawImage->biasedScaleToPowerOfTwo(1024); + } + else + { // leave black border, do not scale image content + mRawImage->expandToPowerOfTwo(MAX_IMAGE_SIZE, FALSE); + } - bool size_okay = true; + mFullWidth = mRawImage->getWidth(); + mFullHeight = mRawImage->getHeight(); + setTexelsPerImage(); + } + else + { + mOrigWidth = mFullWidth; + mOrigHeight = mFullHeight; + } - S32 discard_level = mRawDiscardLevel; - if (mRawDiscardLevel < 0) - { - LL_DEBUGS() << "Negative raw discard level when creating image: " << mRawDiscardLevel << LL_ENDL; - discard_level = 0; - } + bool size_okay = true; - U32 raw_width = mRawImage->getWidth() << discard_level; - U32 raw_height = mRawImage->getHeight() << discard_level; + S32 discard_level = mRawDiscardLevel; + if (mRawDiscardLevel < 0) + { + LL_DEBUGS() << "Negative raw discard level when creating image: " << mRawDiscardLevel << LL_ENDL; + discard_level = 0; + } - if( raw_width > MAX_IMAGE_SIZE || raw_height > MAX_IMAGE_SIZE ) - { - LL_INFOS() << "Width or height is greater than " << MAX_IMAGE_SIZE << ": (" << raw_width << "," << raw_height << ")" << LL_ENDL; - size_okay = false; - } - - if (!LLImageGL::checkSize(mRawImage->getWidth(), mRawImage->getHeight())) - { - // A non power-of-two image was uploaded (through a non standard client) - LL_INFOS() << "Non power of two width or height: (" << mRawImage->getWidth() << "," << mRawImage->getHeight() << ")" << LL_ENDL; - size_okay = false; - } - - if( !size_okay ) - { - // An inappropriately-sized image was uploaded (through a non standard client) - // We treat these images as missing assets which causes them to - // be renderd as 'missing image' and to stop requesting data - LL_WARNS() << "!size_ok, setting as missing" << LL_ENDL; - setIsMissingAsset(); - destroyRawImage(); - return FALSE; - } + U32 raw_width = mRawImage->getWidth() << discard_level; + U32 raw_height = mRawImage->getHeight() << discard_level; + + if (raw_width > MAX_IMAGE_SIZE || raw_height > MAX_IMAGE_SIZE) + { + LL_INFOS() << "Width or height is greater than " << MAX_IMAGE_SIZE << ": (" << raw_width << "," << raw_height << ")" << LL_ENDL; + size_okay = false; + } + + if (!LLImageGL::checkSize(mRawImage->getWidth(), mRawImage->getHeight())) + { + // A non power-of-two image was uploaded (through a non standard client) + LL_INFOS() << "Non power of two width or height: (" << mRawImage->getWidth() << "," << mRawImage->getHeight() << ")" << LL_ENDL; + size_okay = false; + } + + if (!size_okay) + { + // An inappropriately-sized image was uploaded (through a non standard client) + // We treat these images as missing assets which causes them to + // be renderd as 'missing image' and to stop requesting data + LL_WARNS() << "!size_ok, setting as missing" << LL_ENDL; + setIsMissingAsset(); + destroyRawImage(); + return FALSE; + } if (mGLTexturep->getHasExplicitFormat()) { @@ -1565,19 +1590,79 @@ BOOL LLViewerFetchedTexture::createTexture(S32 usename/*= 0*/) } } - res = mGLTexturep->createGLTexture(mRawDiscardLevel, mRawImage, usename, TRUE, mBoostLevel); + return res; +} - notifyAboutCreatingTexture(); +BOOL LLViewerFetchedTexture::createTexture(S32 usename/*= 0*/) +{ + if (!mNeedsCreateTexture) + { + return FALSE; + } - setActive(); + BOOL res = mGLTexturep->createGLTexture(mRawDiscardLevel, mRawImage, usename, TRUE, mBoostLevel); + + return res; +} - if (!needsToSaveRawImage()) - { - mNeedsAux = FALSE; - destroyRawImage(); - } +void LLViewerFetchedTexture::postCreateTexture() +{ + if (!mNeedsCreateTexture) + { + return; + } - return res; + notifyAboutCreatingTexture(); + + setActive(); + + if (!needsToSaveRawImage()) + { + mNeedsAux = FALSE; + destroyRawImage(); + } + + mNeedsCreateTexture = FALSE; +} + +void LLViewerFetchedTexture::scheduleCreateTexture() +{ + ref(); + mNeedsCreateTexture = TRUE; + if (preCreateTexture()) + { + mNeedsCreateTexture = TRUE; +#if LL_WINDOWS //flip to 0 to revert to single-threaded OpenGL texture uploads + if (!LLImageGLThread::sInstance->post([this]() + { + //actually create the texture on a background thread + createTexture(); + { + LL_PROFILE_ZONE_NAMED("iglt - sync"); + if (gGLManager.mHasSync) + { + auto sync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); + glClientWaitSync(sync, 0, 0); + glDeleteSync(sync); + } + else + { + glFinish(); + } + } + LLImageGLThread::sInstance->postCallback([this]() + { + //finalize on main thread + postCreateTexture(); + unref(); + }); + })) +#endif + { + gTextureList.mCreateTextureList.insert(this); + unref(); + } + } } // Call with 0,0 to turn this feature off. @@ -1869,6 +1954,7 @@ void LLViewerFetchedTexture::setAdditionalDecodePriority(F32 priority) void LLViewerFetchedTexture::updateVirtualSize() { + LL_PROFILE_ZONE_SCOPED; if(!mMaxVirtualSizeResetCounter) { addTextureStats(0.f, FALSE);//reset @@ -1960,6 +2046,7 @@ bool LLViewerFetchedTexture::isActiveFetching() bool LLViewerFetchedTexture::updateFetch() { + LL_PROFILE_ZONE_SCOPED; static LLCachedControl textures_decode_disabled(gSavedSettings,"TextureDecodeDisabled", false); static LLCachedControl sCameraMotionThreshold(gSavedSettings,"TextureCameraMotionThreshold", 0.2); static LLCachedControl sCameraMotionBoost(gSavedSettings,"TextureCameraMotionBoost", 3); @@ -2062,7 +2149,7 @@ bool LLViewerFetchedTexture::updateFetch() } else { - mIsRawImageValid = TRUE; + mIsRawImageValid = TRUE; addToCreateTexture(); } @@ -2890,6 +2977,7 @@ void LLViewerFetchedTexture::destroyRawImage() //virtual void LLViewerFetchedTexture::switchToCachedImage() { + LL_PROFILE_ZONE_SCOPED; if(mCachedRawImage.notNull()) { mRawImage = mCachedRawImage; @@ -2901,12 +2989,12 @@ void LLViewerFetchedTexture::switchToCachedImage() mComponents = mRawImage->getComponents(); mGLTexturep->setComponents(mComponents); gTextureList.dirtyImage(this); - } + } mIsRawImageValid = TRUE; mRawDiscardLevel = mCachedRawDiscardLevel; - gTextureList.mCreateTextureList.insert(this); - mNeedsCreateTexture = TRUE; + + scheduleCreateTexture(); } } @@ -3180,6 +3268,7 @@ bool LLViewerLODTexture::isUpdateFrozen() //virtual void LLViewerLODTexture::processTextureStats() { + LL_PROFILE_ZONE_SCOPED; updateVirtualSize(); static LLCachedControl textures_fullres(gSavedSettings,"TextureLoadFullRes", false); diff --git a/indra/newview/llviewertexture.h b/indra/newview/llviewertexture.h index 69568cc825..a5a1fb2c16 100644 --- a/indra/newview/llviewertexture.h +++ b/indra/newview/llviewertexture.h @@ -321,9 +321,13 @@ public: void addToCreateTexture(); - - // ONLY call from LLViewerTextureList + //call to determine if createTexture is necessary + BOOL preCreateTexture(S32 usename = 0); + // ONLY call from LLViewerTextureList or ImageGL background thread BOOL createTexture(S32 usename = 0); + void postCreateTexture(); + void scheduleCreateTexture(); + void destroyTexture() ; virtual void processTextureStats() ; diff --git a/indra/newview/llviewertexturelist.cpp b/indra/newview/llviewertexturelist.cpp index 12495078e9..db740b69e9 100644 --- a/indra/newview/llviewertexturelist.cpp +++ b/indra/newview/llviewertexturelist.cpp @@ -114,6 +114,7 @@ void LLViewerTextureList::init() void LLViewerTextureList::doPreloadImages() { + LL_PROFILE_ZONE_SCOPED; LL_DEBUGS("ViewerImages") << "Preloading images..." << LL_ENDL; llassert_always(mInitialized) ; @@ -205,6 +206,7 @@ static std::string get_texture_list_name() void LLViewerTextureList::doPrefetchImages() { + LL_PROFILE_ZONE_SCOPED; if (LLAppViewer::instance()->getPurgeCache()) { // cache was purged, no point @@ -258,6 +260,7 @@ LLViewerTextureList::~LLViewerTextureList() void LLViewerTextureList::shutdown() { + LL_PROFILE_ZONE_SCOPED; // clear out preloads mImagePreloads.clear(); @@ -333,6 +336,7 @@ void LLViewerTextureList::shutdown() void LLViewerTextureList::dump() { + LL_PROFILE_ZONE_SCOPED; LL_INFOS() << "LLViewerTextureList::dump()" << LL_ENDL; for (image_priority_list_t::iterator it = mImageList.begin(); it != mImageList.end(); ++it) { @@ -377,6 +381,7 @@ LLViewerFetchedTexture* LLViewerTextureList::getImageFromFile(const std::string& LLGLenum primary_format, const LLUUID& force_id) { + LL_PROFILE_ZONE_SCOPED; if(!mInitialized) { return NULL ; @@ -404,6 +409,7 @@ LLViewerFetchedTexture* LLViewerTextureList::getImageFromUrl(const std::string& LLGLenum primary_format, const LLUUID& force_id) { + LL_PROFILE_ZONE_SCOPED; if(!mInitialized) { return NULL ; @@ -492,6 +498,7 @@ LLViewerFetchedTexture* LLViewerTextureList::getImage(const LLUUID &image_id, LLGLenum primary_format, LLHost request_from_host) { + LL_PROFILE_ZONE_SCOPED; if(!mInitialized) { return NULL ; @@ -554,6 +561,7 @@ LLViewerFetchedTexture* LLViewerTextureList::createImage(const LLUUID &image_id, LLGLenum primary_format, LLHost request_from_host) { + LL_PROFILE_ZONE_SCOPED; static LLCachedControl fast_cache_fetching_enabled(gSavedSettings, "FastCacheFetchEnabled", true); LLPointer imagep ; @@ -609,6 +617,7 @@ LLViewerFetchedTexture* LLViewerTextureList::createImage(const LLUUID &image_id, void LLViewerTextureList::findTexturesByID(const LLUUID &image_id, std::vector &output) { + LL_PROFILE_ZONE_SCOPED; LLTextureKey search_key(image_id, TEX_LIST_STANDARD); uuid_map_t::iterator iter = mUUIDMap.lower_bound(search_key); while (iter != mUUIDMap.end() && iter->first.textureId == image_id) @@ -634,6 +643,7 @@ LLViewerFetchedTexture *LLViewerTextureList::findImage(const LLUUID &image_id, E void LLViewerTextureList::addImageToList(LLViewerFetchedTexture *image) { + LL_PROFILE_ZONE_SCOPED; assert_main_thread(); llassert_always(mInitialized) ; llassert(image); @@ -653,6 +663,7 @@ void LLViewerTextureList::addImageToList(LLViewerFetchedTexture *image) void LLViewerTextureList::removeImageFromList(LLViewerFetchedTexture *image) { + LL_PROFILE_ZONE_SCOPED; assert_main_thread(); llassert_always(mInitialized) ; llassert(image); @@ -701,6 +712,7 @@ void LLViewerTextureList::removeImageFromList(LLViewerFetchedTexture *image) void LLViewerTextureList::addImage(LLViewerFetchedTexture *new_image, ETexListType tex_type) { + LL_PROFILE_ZONE_SCOPED; if (!new_image) { return; @@ -724,6 +736,7 @@ void LLViewerTextureList::addImage(LLViewerFetchedTexture *new_image, ETexListTy void LLViewerTextureList::deleteImage(LLViewerFetchedTexture *image) { + LL_PROFILE_ZONE_SCOPED; if( image) { if (image->hasCallbacks()) @@ -845,6 +858,7 @@ void LLViewerTextureList::updateImages(F32 max_time) void LLViewerTextureList::clearFetchingRequests() { + LL_PROFILE_ZONE_SCOPED; if (LLAppViewer::getTextureFetch()->getNumRequests() == 0) { return; @@ -862,6 +876,7 @@ void LLViewerTextureList::clearFetchingRequests() void LLViewerTextureList::updateImagesDecodePriorities() { + LL_PROFILE_ZONE_SCOPED; // Update the decode priority for N images each frame { F32 lazy_flush_timeout = 30.f; // stop decoding @@ -977,6 +992,7 @@ void LLViewerTextureList::updateImagesDecodePriorities() void LLViewerTextureList::setDebugFetching(LLViewerFetchedTexture* tex, S32 debug_level) { + LL_PROFILE_ZONE_SCOPED; if(!tex->setDebugFetching(debug_level)) { return; @@ -1025,6 +1041,7 @@ void LLViewerTextureList::setDebugFetching(LLViewerFetchedTexture* tex, S32 debu F32 LLViewerTextureList::updateImagesCreateTextures(F32 max_time) { + LL_PROFILE_ZONE_SCOPED; if (gGLManager.mIsDisabled) return 0.0f; // @@ -1041,6 +1058,7 @@ F32 LLViewerTextureList::updateImagesCreateTextures(F32 max_time) enditer = iter; LLViewerFetchedTexture *imagep = *curiter; imagep->createTexture(); + imagep->postCreateTexture(); if (create_timer.getElapsedTimeF32() > max_time) { break; @@ -1052,6 +1070,7 @@ F32 LLViewerTextureList::updateImagesCreateTextures(F32 max_time) F32 LLViewerTextureList::updateImagesLoadingFastCache(F32 max_time) { + LL_PROFILE_ZONE_SCOPED; if (gGLManager.mIsDisabled) return 0.0f; if(mFastCacheList.empty()) { @@ -1082,6 +1101,7 @@ F32 LLViewerTextureList::updateImagesLoadingFastCache(F32 max_time) void LLViewerTextureList::forceImmediateUpdate(LLViewerFetchedTexture* imagep) { + LL_PROFILE_ZONE_SCOPED; if(!imagep) { return ; @@ -1101,6 +1121,7 @@ void LLViewerTextureList::forceImmediateUpdate(LLViewerFetchedTexture* imagep) F32 LLViewerTextureList::updateImagesFetchTextures(F32 max_time) { + LL_PROFILE_ZONE_SCOPED; LLTimer image_op_timer; // Update fetch for N images each frame @@ -1176,6 +1197,7 @@ F32 LLViewerTextureList::updateImagesFetchTextures(F32 max_time) void LLViewerTextureList::updateImagesUpdateStats() { + LL_PROFILE_ZONE_SCOPED; if (mForceResetTextureStats) { for (image_priority_list_t::iterator iter = mImageList.begin(); @@ -1190,6 +1212,7 @@ void LLViewerTextureList::updateImagesUpdateStats() void LLViewerTextureList::decodeAllImages(F32 max_time) { + LL_PROFILE_ZONE_SCOPED; LLTimer timer; //loading from fast cache @@ -1259,6 +1282,7 @@ BOOL LLViewerTextureList::createUploadFile(const std::string& filename, const std::string& out_filename, const U8 codec) { + LL_PROFILE_ZONE_SCOPED; // Load the image LLPointer image = LLImageFormatted::createFromType(codec); if (image.isNull()) @@ -1312,6 +1336,7 @@ BOOL LLViewerTextureList::createUploadFile(const std::string& filename, // note: modifies the argument raw_image!!!! LLPointer LLViewerTextureList::convertToUploadFile(LLPointer raw_image) { + LL_PROFILE_ZONE_SCOPED; raw_image->biasedScaleToPowerOfTwo(LLViewerFetchedTexture::MAX_IMAGE_SIZE_DEFAULT); LLPointer compressedImage = new LLImageJ2C(); @@ -1345,6 +1370,7 @@ LLPointer LLViewerTextureList::convertToUploadFile(LLPointer U32Megabytes(1500)) ? S32Megabytes(64) : gMinVideoRam ; @@ -1354,6 +1380,7 @@ S32Megabytes LLViewerTextureList::getMinVideoRamSetting() // Returns max setting for TextureMemory (in MB) S32Megabytes LLViewerTextureList::getMaxVideoRamSetting(bool get_recommended, float mem_multiplier) { + LL_PROFILE_ZONE_SCOPED; S32Megabytes max_texmem; if (gGLManager.mVRAM != 0) { @@ -1407,6 +1434,7 @@ const S32Megabytes VIDEO_CARD_FRAMEBUFFER_MEM(12); const S32Megabytes MIN_MEM_FOR_NON_TEXTURE(512); void LLViewerTextureList::updateMaxResidentTexMem(S32Megabytes mem) { + LL_PROFILE_ZONE_SCOPED; // Initialize the image pipeline VRAM settings S32Megabytes cur_mem(gSavedSettings.getS32("TextureMemory")); F32 mem_multiplier = gSavedSettings.getF32("RenderTextureMemoryMultiple"); @@ -1647,6 +1675,7 @@ void LLUIImageList::cleanUp() LLUIImagePtr LLUIImageList::getUIImageByID(const LLUUID& image_id, S32 priority) { + LL_PROFILE_ZONE_SCOPED; // use id as image name std::string image_name = image_id.asString(); @@ -1665,6 +1694,7 @@ LLUIImagePtr LLUIImageList::getUIImageByID(const LLUUID& image_id, S32 priority) LLUIImagePtr LLUIImageList::getUIImage(const std::string& image_name, S32 priority) { + LL_PROFILE_ZONE_SCOPED; // look for existing image uuid_ui_image_map_t::iterator found_it = mUIImages.find(image_name); if (found_it != mUIImages.end()) @@ -1682,6 +1712,7 @@ LLUIImagePtr LLUIImageList::loadUIImageByName(const std::string& name, const std BOOL use_mips, const LLRect& scale_rect, const LLRect& clip_rect, LLViewerTexture::EBoostLevel boost_priority, LLUIImage::EScaleStyle scale_style) { + LL_PROFILE_ZONE_SCOPED; if (boost_priority == LLGLTexture::BOOST_NONE) { boost_priority = LLGLTexture::BOOST_UI; @@ -1694,6 +1725,7 @@ LLUIImagePtr LLUIImageList::loadUIImageByID(const LLUUID& id, BOOL use_mips, const LLRect& scale_rect, const LLRect& clip_rect, LLViewerTexture::EBoostLevel boost_priority, LLUIImage::EScaleStyle scale_style) { + LL_PROFILE_ZONE_SCOPED; if (boost_priority == LLGLTexture::BOOST_NONE) { boost_priority = LLGLTexture::BOOST_UI; @@ -1705,6 +1737,7 @@ LLUIImagePtr LLUIImageList::loadUIImageByID(const LLUUID& id, LLUIImagePtr LLUIImageList::loadUIImage(LLViewerFetchedTexture* imagep, const std::string& name, BOOL use_mips, const LLRect& scale_rect, const LLRect& clip_rect, LLUIImage::EScaleStyle scale_style) { + LL_PROFILE_ZONE_SCOPED; if (!imagep) return NULL; imagep->setAddressMode(LLTexUnit::TAM_CLAMP); @@ -1742,6 +1775,7 @@ LLUIImagePtr LLUIImageList::loadUIImage(LLViewerFetchedTexture* imagep, const st LLUIImagePtr LLUIImageList::preloadUIImage(const std::string& name, const std::string& filename, BOOL use_mips, const LLRect& scale_rect, const LLRect& clip_rect, LLUIImage::EScaleStyle scale_style) { + LL_PROFILE_ZONE_SCOPED; // look for existing image uuid_ui_image_map_t::iterator found_it = mUIImages.find(name); if (found_it != mUIImages.end()) @@ -1756,6 +1790,7 @@ LLUIImagePtr LLUIImageList::preloadUIImage(const std::string& name, const std::s //static void LLUIImageList::onUIImageLoaded( BOOL success, LLViewerFetchedTexture *src_vi, LLImageRaw* src, LLImageRaw* src_aux, S32 discard_level, BOOL final, void* user_data ) { + LL_PROFILE_ZONE_SCOPED; if(!success || !user_data) { return; @@ -1857,6 +1892,7 @@ struct UIImageDeclarations : public LLInitParam::Block bool LLUIImageList::initFromFile() { + LL_PROFILE_ZONE_SCOPED; // Look for textures.xml in all the right places. Pass // constraint=LLDir::ALL_SKINS because we want to overlay textures.xml // from all the skins directories. diff --git a/indra/newview/llviewerwindow.cpp b/indra/newview/llviewerwindow.cpp index c52a4e98d3..82ece85c1b 100644 --- a/indra/newview/llviewerwindow.cpp +++ b/indra/newview/llviewerwindow.cpp @@ -2011,7 +2011,7 @@ LLViewerWindow::LLViewerWindow(const Params& p) // Init the image list. Must happen after GL is initialized and before the images that // LLViewerWindow needs are requested. - LLImageGL::initClass(LLViewerTexture::MAX_GL_IMAGE_CATEGORY) ; + LLImageGL::initClass(mWindow, LLViewerTexture::MAX_GL_IMAGE_CATEGORY) ; gTextureList.init(); LLViewerTextureManager::init() ; gBumpImageList.init(); -- cgit v1.3 From 851767b808c3cb05d718538389ccc1ed3c95d1a1 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 14 Oct 2021 17:41:38 +0000 Subject: SL-16131 Fix for alignment warnings on Win32 builds. --- indra/llcharacter/lleditingmotion.h | 18 +- indra/llcharacter/lljoint.h | 14 +- indra/llcharacter/llkeyframestandmotion.h | 28 +- indra/llcharacter/llpose.h | 6 +- indra/llcommon/llmemory.h | 13 + indra/llprimitive/llmodel.h | 10 +- indra/llwindow/llwindow.h | 1 + indra/media_plugins/cef/CMakeLists.txt | 3 - indra/newview/llviewertexture.h | 1592 ++++++++++++++--------------- 9 files changed, 854 insertions(+), 831 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcharacter/lleditingmotion.h b/indra/llcharacter/lleditingmotion.h index 7b1c8bb059..80c1717a70 100644 --- a/indra/llcharacter/lleditingmotion.h +++ b/indra/llcharacter/lleditingmotion.h @@ -42,9 +42,11 @@ //----------------------------------------------------------------------------- // class LLEditingMotion //----------------------------------------------------------------------------- +LL_ALIGN_PREFIX(16) class LLEditingMotion : public LLMotion { + LL_ALIGN_NEW public: // Constructor LLEditingMotion(const LLUUID &id); @@ -108,6 +110,13 @@ public: //------------------------------------------------------------------------- // joint states to be animated //------------------------------------------------------------------------- + LL_ALIGN_16(LLJoint mParentJoint); + LL_ALIGN_16(LLJoint mShoulderJoint); + LL_ALIGN_16(LLJoint mElbowJoint); + LL_ALIGN_16(LLJoint mWristJoint); + LL_ALIGN_16(LLJoint mTarget); + LLJointSolverRP3 mIKSolver; + LLCharacter *mCharacter; LLVector3 mWristOffset; @@ -117,17 +126,10 @@ public: LLPointer mWristState; LLPointer mTorsoState; - LLJoint mParentJoint; - LLJoint mShoulderJoint; - LLJoint mElbowJoint; - LLJoint mWristJoint; - LLJoint mTarget; - LLJointSolverRP3 mIKSolver; - static S32 sHandPose; static S32 sHandPosePriority; LLVector3 mLastSelectPt; -}; +} LL_ALIGN_POSTFIX(16); #endif // LL_LLKEYFRAMEMOTION_H diff --git a/indra/llcharacter/lljoint.h b/indra/llcharacter/lljoint.h index ba821667c7..63d99b9209 100644 --- a/indra/llcharacter/lljoint.h +++ b/indra/llcharacter/lljoint.h @@ -86,8 +86,10 @@ inline bool operator!=(const LLVector3OverrideMap& a, const LLVector3OverrideMap //----------------------------------------------------------------------------- // class LLJoint //----------------------------------------------------------------------------- +LL_ALIGN_PREFIX(16) class LLJoint { + LL_ALIGN_NEW public: // priority levels, from highest to lowest enum JointPriority @@ -115,17 +117,17 @@ public: SUPPORT_EXTENDED }; protected: - std::string mName; + // explicit transformation members + LL_ALIGN_16(LLMatrix4a mWorldMatrix); + LLXformMatrix mXform; + + std::string mName; SupportCategory mSupport; // parent joint LLJoint *mParent; - // explicit transformation members - LLXformMatrix mXform; - LLMatrix4a mWorldMatrix; - LLVector3 mDefaultPosition; LLVector3 mDefaultScale; @@ -300,6 +302,6 @@ public: // These are used in checks of whether a pos/scale override is considered significant. bool aboveJointPosThreshold(const LLVector3& pos) const; bool aboveJointScaleThreshold(const LLVector3& scale) const; -}; +} LL_ALIGN_POSTFIX(16); #endif // LL_LLJOINT_H diff --git a/indra/llcharacter/llkeyframestandmotion.h b/indra/llcharacter/llkeyframestandmotion.h index c2634ecd6d..1aa5b187ba 100644 --- a/indra/llcharacter/llkeyframestandmotion.h +++ b/indra/llcharacter/llkeyframestandmotion.h @@ -37,9 +37,11 @@ //----------------------------------------------------------------------------- // class LLKeyframeStandMotion //----------------------------------------------------------------------------- +LL_ALIGN_PREFIX(16) class LLKeyframeStandMotion : public LLKeyframeMotion { + LL_ALIGN_NEW public: // Constructor LLKeyframeStandMotion(const LLUUID &id); @@ -69,6 +71,18 @@ public: //------------------------------------------------------------------------- // Member Data //------------------------------------------------------------------------- + LLJoint mPelvisJoint; + + LLJoint mHipLeftJoint; + LLJoint mKneeLeftJoint; + LLJoint mAnkleLeftJoint; + LLJoint mTargetLeft; + + LLJoint mHipRightJoint; + LLJoint mKneeRightJoint; + LLJoint mAnkleRightJoint; + LLJoint mTargetRight; + LLCharacter *mCharacter; BOOL mFlipFeet; @@ -83,18 +97,6 @@ public: LLPointer mKneeRightState; LLPointer mAnkleRightState; - LLJoint mPelvisJoint; - - LLJoint mHipLeftJoint; - LLJoint mKneeLeftJoint; - LLJoint mAnkleLeftJoint; - LLJoint mTargetLeft; - - LLJoint mHipRightJoint; - LLJoint mKneeRightJoint; - LLJoint mAnkleRightJoint; - LLJoint mTargetRight; - LLJointSolverRP3 mIKLeft; LLJointSolverRP3 mIKRight; @@ -110,7 +112,7 @@ public: BOOL mTrackAnkles; S32 mFrameNum; -}; +} LL_ALIGN_POSTFIX(16); #endif // LL_LLKEYFRAMESTANDMOTION_H diff --git a/indra/llcharacter/llpose.h b/indra/llcharacter/llpose.h index c004a0f3b7..1405f1e053 100644 --- a/indra/llcharacter/llpose.h +++ b/indra/llcharacter/llpose.h @@ -80,8 +80,10 @@ public: const S32 JSB_NUM_JOINT_STATES = 6; +LL_ALIGN_PREFIX(16) class LLJointStateBlender { + LL_ALIGN_NEW protected: LLPointer mJointStates[JSB_NUM_JOINT_STATES]; S32 mPriorities[JSB_NUM_JOINT_STATES]; @@ -96,8 +98,8 @@ public: void resetCachedJoint(); public: - LLJoint mJointCache; -}; + LL_ALIGN_16(LLJoint mJointCache); +} LL_ALIGN_POSTFIX(16); class LLMotion; diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h index 24f86cc11e..2704a495e0 100644 --- a/indra/llcommon/llmemory.h +++ b/indra/llcommon/llmemory.h @@ -101,6 +101,19 @@ template T* LL_NEXT_ALIGNED_ADDRESS_64(T* address) #define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16) +#define LL_ALIGN_NEW \ +public: \ + void* operator new(size_t size) \ + { \ + return ll_aligned_malloc_16(size); \ + } \ + \ + void operator delete(void* ptr) \ + { \ + ll_aligned_free_16(ptr); \ + } + + //------------------------------------------------------------------------------------------------ //------------------------------------------------------------------------------------------------ // for enable buffer overrun detection predefine LL_DEBUG_BUFFER_OVERRUN in current library diff --git a/indra/llprimitive/llmodel.h b/indra/llprimitive/llmodel.h index 96d4582b4f..cd2b6c6728 100644 --- a/indra/llprimitive/llmodel.h +++ b/indra/llprimitive/llmodel.h @@ -40,8 +40,10 @@ class domMesh; #define MAX_MODEL_FACES 8 +LL_ALIGN_PREFIX(16) class LLMeshSkinInfo { + LL_ALIGN_NEW public: LLMeshSkinInfo(); LLMeshSkinInfo(LLSD& data); @@ -55,15 +57,17 @@ public: matrix_list_t mInvBindMatrix; matrix_list_t mAlternateBindMatrix; - LLMatrix4a mBindShapeMatrix; + LL_ALIGN_16(LLMatrix4a mBindShapeMatrix); float mPelvisOffset; bool mLockScaleIfJointPosition; bool mInvalidJointsScrubbed; bool mJointNumsInitialized; -}; +} LL_ALIGN_POSTFIX(16); +LL_ALIGN_PREFIX(16) class LLModel : public LLVolume { + LL_ALIGN_NEW public: enum @@ -285,7 +289,7 @@ public: EModelStatus mStatus ; int mSubmodelID; -}; +} LL_ALIGN_POSTFIX(16); typedef std::vector > model_list; typedef std::queue > model_queue; diff --git a/indra/llwindow/llwindow.h b/indra/llwindow/llwindow.h index 10c0b6a424..0100c3bf0a 100644 --- a/indra/llwindow/llwindow.h +++ b/indra/llwindow/llwindow.h @@ -88,6 +88,7 @@ public: //Must be called on the same thread that called createSharedContext() virtual void destroySharedContext(void* context) = 0; + virtual BOOL setCursorPosition(LLCoordWindow position) = 0; virtual BOOL getCursorPosition(LLCoordWindow *position) = 0; virtual void showCursor() = 0; diff --git a/indra/media_plugins/cef/CMakeLists.txt b/indra/media_plugins/cef/CMakeLists.txt index ce6278963d..76d398576c 100644 --- a/indra/media_plugins/cef/CMakeLists.txt +++ b/indra/media_plugins/cef/CMakeLists.txt @@ -111,9 +111,6 @@ if (DARWIN) LINK_FLAGS "-exported_symbols_list ${CMAKE_CURRENT_SOURCE_DIR}/../base/media_plugin_base.exp" ) - ## turns on C++11 using Cmake - target_compile_features(media_plugin_cef PRIVATE cxx_range_for) - add_custom_command(TARGET media_plugin_cef POST_BUILD COMMAND ${CMAKE_INSTALL_NAME_TOOL} -change "@executable_path/Chromium Embedded Framework" "@executable_path/../../../../Frameworks/Chromium Embedded Framework.framework/Chromium Embedded Framework" diff --git a/indra/newview/llviewertexture.h b/indra/newview/llviewertexture.h index a5a1fb2c16..b6b4d1e41f 100644 --- a/indra/newview/llviewertexture.h +++ b/indra/newview/llviewertexture.h @@ -1,796 +1,796 @@ -/** - * @file llviewertexture.h - * @brief Object for managing images and their textures - * - * $LicenseInfo:firstyear=2000&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#ifndef LL_LLVIEWERTEXTURE_H -#define LL_LLVIEWERTEXTURE_H - -#include "llgltexture.h" -#include "lltimer.h" -#include "llframetimer.h" -#include "llhost.h" -#include "llgltypes.h" -#include "llrender.h" -#include "llmetricperformancetester.h" -#include "httpcommon.h" - -#include -#include - -extern const S32Megabytes gMinVideoRam; -extern const S32Megabytes gMaxVideoRam; - -class LLFace; -class LLImageGL ; -class LLImageRaw; -class LLViewerObject; -class LLViewerTexture; -class LLViewerFetchedTexture ; -class LLViewerMediaTexture ; -class LLTexturePipelineTester ; - - -typedef void (*loaded_callback_func)( BOOL success, LLViewerFetchedTexture *src_vi, LLImageRaw* src, LLImageRaw* src_aux, S32 discard_level, BOOL final, void* userdata ); - -class LLVFile; -class LLMessageSystem; -class LLViewerMediaImpl ; -class LLVOVolume ; -struct LLTextureKey; - -class LLLoadedCallbackEntry -{ -public: - typedef std::set< LLTextureKey > source_callback_list_t; - -public: - LLLoadedCallbackEntry(loaded_callback_func cb, - S32 discard_level, - BOOL need_imageraw, // Needs image raw for the callback - void* userdata, - source_callback_list_t* src_callback_list, - LLViewerFetchedTexture* target, - BOOL pause); - ~LLLoadedCallbackEntry(); - void removeTexture(LLViewerFetchedTexture* tex) ; - - loaded_callback_func mCallback; - S32 mLastUsedDiscard; - S32 mDesiredDiscard; - BOOL mNeedsImageRaw; - BOOL mPaused; - void* mUserData; - source_callback_list_t* mSourceCallbackList; - -public: - static void cleanUpCallbackList(LLLoadedCallbackEntry::source_callback_list_t* callback_list) ; -}; - -class LLTextureBar; - -class LLViewerTexture : public LLGLTexture -{ -public: - enum - { - LOCAL_TEXTURE, - MEDIA_TEXTURE, - DYNAMIC_TEXTURE, - FETCHED_TEXTURE, - LOD_TEXTURE, - ATLAS_TEXTURE, - INVALID_TEXTURE_TYPE - }; - - typedef std::vector ll_face_list_t; - typedef std::vector ll_volume_list_t; - - -protected: - virtual ~LLViewerTexture(); - LOG_CLASS(LLViewerTexture); - -public: - static void initClass(); - static void updateClass(const F32 velocity, const F32 angular_velocity) ; - - LLViewerTexture(BOOL usemipmaps = TRUE); - LLViewerTexture(const LLUUID& id, BOOL usemipmaps) ; - LLViewerTexture(const LLImageRaw* raw, BOOL usemipmaps) ; - LLViewerTexture(const U32 width, const U32 height, const U8 components, BOOL usemipmaps) ; - - virtual S8 getType() const; - virtual BOOL isMissingAsset() const ; - virtual void dump(); // debug info to LL_INFOS() - - virtual bool isViewerMediaTexture() const { return false; } - - /*virtual*/ bool bindDefaultImage(const S32 stage = 0) ; - /*virtual*/ bool bindDebugImage(const S32 stage = 0) ; - /*virtual*/ void forceImmediateUpdate() ; - /*virtual*/ bool isActiveFetching(); - - /*virtual*/ const LLUUID& getID() const { return mID; } - void setBoostLevel(S32 level); - S32 getBoostLevel() { return mBoostLevel; } - void setTextureListType(S32 tex_type) { mTextureListType = tex_type; } - S32 getTextureListType() { return mTextureListType; } - - void addTextureStats(F32 virtual_size, BOOL needs_gltexture = TRUE) const; - void resetTextureStats(); - void setMaxVirtualSizeResetInterval(S32 interval)const {mMaxVirtualSizeResetInterval = interval;} - void resetMaxVirtualSizeResetCounter()const {mMaxVirtualSizeResetCounter = mMaxVirtualSizeResetInterval;} - S32 getMaxVirtualSizeResetCounter() const { return mMaxVirtualSizeResetCounter; } - - virtual F32 getMaxVirtualSize() ; - - LLFrameTimer* getLastReferencedTimer() {return &mLastReferencedTimer ;} - - S32 getFullWidth() const { return mFullWidth; } - S32 getFullHeight() const { return mFullHeight; } - /*virtual*/ void setKnownDrawSize(S32 width, S32 height); - - virtual void addFace(U32 channel, LLFace* facep) ; - virtual void removeFace(U32 channel, LLFace* facep) ; - S32 getTotalNumFaces() const; - S32 getNumFaces(U32 ch) const; - const ll_face_list_t* getFaceList(U32 channel) const {llassert(channel < LLRender::NUM_TEXTURE_CHANNELS); return &mFaceList[channel];} - - virtual void addVolume(U32 channel, LLVOVolume* volumep); - virtual void removeVolume(U32 channel, LLVOVolume* volumep); - S32 getNumVolumes(U32 channel) const; - const ll_volume_list_t* getVolumeList(U32 channel) const { return &mVolumeList[channel]; } - - - virtual void setCachedRawImage(S32 discard_level, LLImageRaw* imageraw) ; - BOOL isLargeImage() ; - - void setParcelMedia(LLViewerMediaTexture* media) {mParcelMedia = media;} - BOOL hasParcelMedia() const { return mParcelMedia != NULL;} - LLViewerMediaTexture* getParcelMedia() const { return mParcelMedia;} - - /*virtual*/ void updateBindStatsForTester() ; -protected: - void cleanup() ; - void init(bool firstinit) ; - void reorganizeFaceList() ; - void reorganizeVolumeList() ; - - void notifyAboutMissingAsset(); - void notifyAboutCreatingTexture(); - -private: - friend class LLBumpImageList; - friend class LLUIImageList; - - virtual void switchToCachedImage(); - - static bool isMemoryForTextureLow() ; - static bool isMemoryForTextureSuficientlyFree(); - static void getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &physical); - -protected: - LLUUID mID; - S32 mTextureListType; // along with mID identifies where to search for this texture in TextureList - - F32 mSelectedTime; // time texture was last selected - mutable F32 mMaxVirtualSize; // The largest virtual size of the image, in pixels - how much data to we need? - mutable S32 mMaxVirtualSizeResetCounter ; - mutable S32 mMaxVirtualSizeResetInterval; - mutable F32 mAdditionalDecodePriority; // priority add to mDecodePriority. - LLFrameTimer mLastReferencedTimer; - - ll_face_list_t mFaceList[LLRender::NUM_TEXTURE_CHANNELS]; //reverse pointer pointing to the faces using this image as texture - U32 mNumFaces[LLRender::NUM_TEXTURE_CHANNELS]; - LLFrameTimer mLastFaceListUpdateTimer ; - - ll_volume_list_t mVolumeList[LLRender::NUM_VOLUME_TEXTURE_CHANNELS]; - U32 mNumVolumes[LLRender::NUM_VOLUME_TEXTURE_CHANNELS]; - LLFrameTimer mLastVolumeListUpdateTimer; - - //do not use LLPointer here. - LLViewerMediaTexture* mParcelMedia ; - - static F32 sTexelPixelRatio; -public: - static const U32 sCurrentFileVersion; - static S32 sImageCount; - static S32 sRawCount; - static S32 sAuxCount; - static LLFrameTimer sEvaluationTimer; - static F32 sDesiredDiscardBias; - static F32 sDesiredDiscardScale; - static S32Bytes sBoundTextureMemory; - static S32Bytes sTotalTextureMemory; - static S32Megabytes sMaxBoundTextureMemory; - static S32Megabytes sMaxTotalTextureMem; - static S32Bytes sMaxDesiredTextureMem ; - static S8 sCameraMovingDiscardBias; - static F32 sCameraMovingBias; - static S32 sMaxSculptRez ; - static U32 sMinLargeImageSize ; - static U32 sMaxSmallImageSize ; - static bool sFreezeImageUpdates; - static F32 sCurrentTime ; - - enum EDebugTexels - { - DEBUG_TEXELS_OFF, - DEBUG_TEXELS_CURRENT, - DEBUG_TEXELS_DESIRED, - DEBUG_TEXELS_FULL - }; - - static EDebugTexels sDebugTexelsMode; - - static LLPointer sNullImagep; // Null texture for non-textured objects. - static LLPointer sBlackImagep; // Texture to show NOTHING (pure black) - static LLPointer sCheckerBoardImagep; // Texture to show NOTHING (pure black) -}; - - -enum FTType -{ - FTT_UNKNOWN = -1, - FTT_DEFAULT = 0, // standard texture fetched by id. - FTT_SERVER_BAKE, // texture produced by appearance service and fetched from there. - FTT_HOST_BAKE, // old-style baked texture uploaded by viewer and fetched from avatar's host. - FTT_MAP_TILE, // tiles are fetched from map server directly. - FTT_LOCAL_FILE // fetch directly from a local file. -}; - -const std::string& fttype_to_string(const FTType& fttype); - -// -//textures are managed in gTextureList. -//raw image data is fetched from remote or local cache -//but the raw image this texture pointing to is fixed. -// -class LLViewerFetchedTexture : public LLViewerTexture -{ - friend class LLTextureBar; // debug info only - friend class LLTextureView; // debug info only - -protected: - /*virtual*/ ~LLViewerFetchedTexture(); -public: - LLViewerFetchedTexture(const LLUUID& id, FTType f_type, const LLHost& host = LLHost(), BOOL usemipmaps = TRUE); - LLViewerFetchedTexture(const LLImageRaw* raw, FTType f_type, BOOL usemipmaps); - LLViewerFetchedTexture(const std::string& url, FTType f_type, const LLUUID& id, BOOL usemipmaps = TRUE); - -public: - static F32 maxDecodePriority(); - - struct Compare - { - // lhs < rhs - bool operator()(const LLPointer &lhs, const LLPointer &rhs) const - { - const LLViewerFetchedTexture* lhsp = (const LLViewerFetchedTexture*)lhs; - const LLViewerFetchedTexture* rhsp = (const LLViewerFetchedTexture*)rhs; - // greater priority is "less" - const F32 lpriority = lhsp->getDecodePriority(); - const F32 rpriority = rhsp->getDecodePriority(); - if (lpriority > rpriority) // higher priority - return true; - if (lpriority < rpriority) - return false; - return lhsp < rhsp; - } - }; - -public: - /*virtual*/ S8 getType() const ; - FTType getFTType() const; - /*virtual*/ void forceImmediateUpdate() ; - /*virtual*/ void dump() ; - - // Set callbacks to get called when the image gets updated with higher - // resolution versions. - void setLoadedCallback(loaded_callback_func cb, - S32 discard_level, BOOL keep_imageraw, BOOL needs_aux, - void* userdata, LLLoadedCallbackEntry::source_callback_list_t* src_callback_list, BOOL pause = FALSE); - bool hasCallbacks() { return mLoadedCallbackList.empty() ? false : true; } - void pauseLoadedCallbacks(const LLLoadedCallbackEntry::source_callback_list_t* callback_list); - void unpauseLoadedCallbacks(const LLLoadedCallbackEntry::source_callback_list_t* callback_list); - bool doLoadedCallbacks(); - void deleteCallbackEntry(const LLLoadedCallbackEntry::source_callback_list_t* callback_list); - void clearCallbackEntryList() ; - - void addToCreateTexture(); - - //call to determine if createTexture is necessary - BOOL preCreateTexture(S32 usename = 0); - // ONLY call from LLViewerTextureList or ImageGL background thread - BOOL createTexture(S32 usename = 0); - void postCreateTexture(); - void scheduleCreateTexture(); - - void destroyTexture() ; - - virtual void processTextureStats() ; - F32 calcDecodePriority() ; - - BOOL needsAux() const { return mNeedsAux; } - - // Host we think might have this image, used for baked av textures. - void setTargetHost(LLHost host) { mTargetHost = host; } - LLHost getTargetHost() const { return mTargetHost; } - - // Set the decode priority for this image... - // DON'T CALL THIS UNLESS YOU KNOW WHAT YOU'RE DOING, it can mess up - // the priority list, and cause horrible things to happen. - void setDecodePriority(F32 priority = -1.0f); - F32 getDecodePriority() const { return mDecodePriority; }; - F32 getAdditionalDecodePriority() const { return mAdditionalDecodePriority; }; - - void setAdditionalDecodePriority(F32 priority) ; - - void updateVirtualSize() ; - - S32 getDesiredDiscardLevel() { return mDesiredDiscardLevel; } - void setMinDiscardLevel(S32 discard) { mMinDesiredDiscardLevel = llmin(mMinDesiredDiscardLevel,(S8)discard); } - - bool updateFetch(); - bool setDebugFetching(S32 debug_level); - bool isInDebug() const { return mInDebug; } - - void setUnremovable(BOOL value) { mUnremovable = value; } - bool isUnremovable() const { return mUnremovable; } - - void clearFetchedResults(); //clear all fetched results, for debug use. - - // Override the computation of discard levels if we know the exact output - // size of the image. Used for UI textures to not decode, even if we have - // more data. - /*virtual*/ void setKnownDrawSize(S32 width, S32 height); - - void setIsMissingAsset(BOOL is_missing = true); - /*virtual*/ BOOL isMissingAsset() const { return mIsMissingAsset; } - - // returns dimensions of original image for local files (before power of two scaling) - // and returns 0 for all asset system images - S32 getOriginalWidth() { return mOrigWidth; } - S32 getOriginalHeight() { return mOrigHeight; } - - BOOL isInImageList() const {return mInImageList ;} - void setInImageList(BOOL flag) {mInImageList = flag ;} - - LLFrameTimer* getLastPacketTimer() {return &mLastPacketTimer;} - - U32 getFetchPriority() const { return mFetchPriority ;} - F32 getDownloadProgress() const {return mDownloadProgress ;} - - LLImageRaw* reloadRawImage(S8 discard_level) ; - void destroyRawImage(); - bool needsToSaveRawImage(); - - const std::string& getUrl() const {return mUrl;} - //--------------- - BOOL isDeleted() ; - BOOL isInactive() ; - BOOL isDeletionCandidate(); - void setDeletionCandidate() ; - void setInactive() ; - BOOL getUseDiscard() const { return mUseMipMaps && !mDontDiscard; } - //--------------- - - void setForSculpt(); - BOOL forSculpt() const {return mForSculpt;} - BOOL isForSculptOnly() const; - - //raw image management - void checkCachedRawSculptImage() ; - LLImageRaw* getRawImage()const { return mRawImage ;} - S32 getRawImageLevel() const {return mRawDiscardLevel;} - LLImageRaw* getCachedRawImage() const { return mCachedRawImage ;} - S32 getCachedRawImageLevel() const {return mCachedRawDiscardLevel;} - BOOL isCachedRawImageReady() const {return mCachedRawImageReady ;} - BOOL isRawImageValid()const { return mIsRawImageValid ; } - void forceToSaveRawImage(S32 desired_discard = 0, F32 kept_time = 0.f) ; - void forceToRefetchTexture(S32 desired_discard = 0, F32 kept_time = 60.f); - /*virtual*/ void setCachedRawImage(S32 discard_level, LLImageRaw* imageraw) ; - void destroySavedRawImage() ; - LLImageRaw* getSavedRawImage() ; - BOOL hasSavedRawImage() const ; - F32 getElapsedLastReferencedSavedRawImageTime() const ; - BOOL isFullyLoaded() const; - - BOOL hasFetcher() const { return mHasFetcher;} - void setCanUseHTTP(bool can_use_http) {mCanUseHTTP = can_use_http;} - - void forceToDeleteRequest(); - void loadFromFastCache(); - void setInFastCacheList(bool in_list) { mInFastCacheList = in_list; } - bool isInFastCacheList() { return mInFastCacheList; } - - /*virtual*/bool isActiveFetching(); //is actively in fetching by the fetching pipeline. - -protected: - /*virtual*/ void switchToCachedImage(); - S32 getCurrentDiscardLevelForFetching() ; - -private: - void init(bool firstinit) ; - void cleanup() ; - - void saveRawImage() ; - void setCachedRawImage() ; - - //for atlas - void resetFaceAtlas() ; - void invalidateAtlas(BOOL rebuild_geom) ; - BOOL insertToAtlas() ; - -private: - BOOL mFullyLoaded; - BOOL mInDebug; - BOOL mUnremovable; - BOOL mInFastCacheList; - BOOL mForceCallbackFetch; - -protected: - std::string mLocalFileName; - - S32 mOrigWidth; - S32 mOrigHeight; - - // Override the computation of discard levels if we know the exact output size of the image. - // Used for UI textures to not decode, even if we have more data. - S32 mKnownDrawWidth; - S32 mKnownDrawHeight; - BOOL mKnownDrawSizeChanged ; - std::string mUrl; - - S32 mRequestedDiscardLevel; - F32 mRequestedDownloadPriority; - S32 mFetchState; - U32 mFetchPriority; - F32 mDownloadProgress; - F32 mFetchDeltaTime; - F32 mRequestDeltaTime; - F32 mDecodePriority; // The priority for decoding this image. - S32 mMinDiscardLevel; - S8 mDesiredDiscardLevel; // The discard level we'd LIKE to have - if we have it and there's space - S8 mMinDesiredDiscardLevel; // The minimum discard level we'd like to have - - S8 mNeedsAux; // We need to decode the auxiliary channels - S8 mHasAux; // We have aux channels - S8 mDecodingAux; // Are we decoding high components - S8 mIsRawImageValid; - S8 mHasFetcher; // We've made a fecth request - S8 mIsFetching; // Fetch request is active - bool mCanUseHTTP; //This texture can be fetched through http if true. - LLCore::HttpStatus mLastHttpGetStatus; // Result of the most recently completed http request for this texture. - - FTType mFTType; // What category of image is this - map tile, server bake, etc? - mutable S8 mIsMissingAsset; // True if we know that there is no image asset with this image id in the database. - - typedef std::list callback_list_t; - S8 mLoadedCallbackDesiredDiscardLevel; - BOOL mPauseLoadedCallBacks; - callback_list_t mLoadedCallbackList; - F32 mLastCallBackActiveTime; - - LLPointer mRawImage; - S32 mRawDiscardLevel; - - // Used ONLY for cloth meshes right now. Make SURE you know what you're - // doing if you use it for anything else! - djs - LLPointer mAuxRawImage; - - //keep a copy of mRawImage for some special purposes - //when mForceToSaveRawImage is set. - BOOL mForceToSaveRawImage ; - BOOL mSaveRawImage; - LLPointer mSavedRawImage; - S32 mSavedRawDiscardLevel; - S32 mDesiredSavedRawDiscardLevel; - F32 mLastReferencedSavedRawImageTime ; - F32 mKeptSavedRawImageTime ; - - //a small version of the copy of the raw image (<= 64 * 64) - LLPointer mCachedRawImage; - S32 mCachedRawDiscardLevel; - BOOL mCachedRawImageReady; //the rez of the mCachedRawImage reaches the upper limit. - - LLHost mTargetHost; // if invalid, just request from agent's simulator - - // Timers - LLFrameTimer mLastPacketTimer; // Time since last packet. - LLFrameTimer mStopFetchingTimer; // Time since mDecodePriority == 0.f. - - BOOL mInImageList; // TRUE if image is in list (in which case don't reset priority!) - BOOL mNeedsCreateTexture; - - BOOL mForSculpt ; //a flag if the texture is used as sculpt data. - BOOL mIsFetched ; //is loaded from remote or from cache, not generated locally. - -public: - static LLPointer sMissingAssetImagep; // Texture to show for an image asset that is not in the database - static LLPointer sWhiteImagep; // Texture to show NOTHING (whiteness) - static LLPointer sDefaultImagep; // "Default" texture for error cases, the only case of fetched texture which is generated in local. - static LLPointer sSmokeImagep; // Old "Default" translucent texture - static LLPointer sFlatNormalImagep; // Flat normal map denoting no bumpiness on a surface -}; - -// -//the image data is fetched from remote or from local cache -//the resolution of the texture is adjustable: depends on the view-dependent parameters. -// -class LLViewerLODTexture : public LLViewerFetchedTexture -{ -protected: - /*virtual*/ ~LLViewerLODTexture(){} - -public: - LLViewerLODTexture(const LLUUID& id, FTType f_type, const LLHost& host = LLHost(), BOOL usemipmaps = TRUE); - LLViewerLODTexture(const std::string& url, FTType f_type, const LLUUID& id, BOOL usemipmaps = TRUE); - - /*virtual*/ S8 getType() const; - // Process image stats to determine priority/quality requirements. - /*virtual*/ void processTextureStats(); - bool isUpdateFrozen() ; - -private: - void init(bool firstinit) ; - bool scaleDown() ; - -private: - F32 mDiscardVirtualSize; // Virtual size used to calculate desired discard - F32 mCalculatedDiscardLevel; // Last calculated discard level -}; - -// -//the image data is fetched from the media pipeline periodically -//the resolution of the texture is also adjusted by the media pipeline -// -class LLViewerMediaTexture : public LLViewerTexture -{ -protected: - /*virtual*/ ~LLViewerMediaTexture() ; - -public: - LLViewerMediaTexture(const LLUUID& id, BOOL usemipmaps = TRUE, LLImageGL* gl_image = NULL) ; - - /*virtual*/ S8 getType() const; - void reinit(BOOL usemipmaps = TRUE); - - BOOL getUseMipMaps() {return mUseMipMaps ; } - void setUseMipMaps(BOOL mipmap) ; - - void setPlaying(BOOL playing) ; - BOOL isPlaying() const {return mIsPlaying;} - void setMediaImpl() ; - - virtual bool isViewerMediaTexture() const { return true; } - - void initVirtualSize() ; - void invalidateMediaImpl() ; - - void addMediaToFace(LLFace* facep) ; - void removeMediaFromFace(LLFace* facep) ; - - /*virtual*/ void addFace(U32 ch, LLFace* facep) ; - /*virtual*/ void removeFace(U32 ch, LLFace* facep) ; - - /*virtual*/ F32 getMaxVirtualSize() ; -private: - void switchTexture(U32 ch, LLFace* facep) ; - BOOL findFaces() ; - void stopPlaying() ; - -private: - // - //an instant list, recording all faces referencing or can reference to this media texture. - //NOTE: it is NOT thread safe. - // - std::list< LLFace* > mMediaFaceList ; - - //an instant list keeping all textures which are replaced by the current media texture, - //is only used to avoid the removal of those textures from memory. - std::list< LLPointer > mTextureList ; - - LLViewerMediaImpl* mMediaImplp ; - BOOL mIsPlaying ; - U32 mUpdateVirtualSizeTime ; - -public: - static void updateClass() ; - static void cleanUpClass() ; - - static LLViewerMediaTexture* findMediaTexture(const LLUUID& media_id) ; - static void removeMediaImplFromTexture(const LLUUID& media_id) ; - -private: - typedef std::map< LLUUID, LLPointer > media_map_t ; - static media_map_t sMediaMap ; -}; - -//just an interface class, do not create instance from this class. -class LLViewerTextureManager -{ -private: - //make the constructor private to preclude creating instances from this class. - LLViewerTextureManager(){} - -public: - //texture pipeline tester - static LLTexturePipelineTester* sTesterp ; - - //returns NULL if tex is not a LLViewerFetchedTexture nor derived from LLViewerFetchedTexture. - static LLViewerFetchedTexture* staticCastToFetchedTexture(LLTexture* tex, BOOL report_error = FALSE) ; - - // - //"find-texture" just check if the texture exists, if yes, return it, otherwise return null. - // - static void findFetchedTextures(const LLUUID& id, std::vector &output); - static void findTextures(const LLUUID& id, std::vector &output); - static LLViewerFetchedTexture* findFetchedTexture(const LLUUID& id, S32 tex_type); - static LLViewerMediaTexture* findMediaTexture(const LLUUID& id) ; - - static LLViewerMediaTexture* createMediaTexture(const LLUUID& id, BOOL usemipmaps = TRUE, LLImageGL* gl_image = NULL) ; - - // - //"get-texture" will create a new texture if the texture does not exist. - // - static LLViewerMediaTexture* getMediaTexture(const LLUUID& id, BOOL usemipmaps = TRUE, LLImageGL* gl_image = NULL) ; - - static LLPointer getLocalTexture(BOOL usemipmaps = TRUE, BOOL generate_gl_tex = TRUE); - static LLPointer getLocalTexture(const LLUUID& id, BOOL usemipmaps, BOOL generate_gl_tex = TRUE) ; - static LLPointer getLocalTexture(const LLImageRaw* raw, BOOL usemipmaps) ; - static LLPointer getLocalTexture(const U32 width, const U32 height, const U8 components, BOOL usemipmaps, BOOL generate_gl_tex = TRUE) ; - - static LLViewerFetchedTexture* getFetchedTexture(const LLUUID &image_id, - FTType f_type = FTT_DEFAULT, - BOOL usemipmap = TRUE, - LLViewerTexture::EBoostLevel boost_priority = LLGLTexture::BOOST_NONE, // Get the requested level immediately upon creation. - S8 texture_type = LLViewerTexture::FETCHED_TEXTURE, - LLGLint internal_format = 0, - LLGLenum primary_format = 0, - LLHost request_from_host = LLHost() - ); - - static LLViewerFetchedTexture* getFetchedTextureFromFile(const std::string& filename, - FTType f_type = FTT_LOCAL_FILE, - BOOL usemipmap = TRUE, - LLViewerTexture::EBoostLevel boost_priority = LLGLTexture::BOOST_NONE, - S8 texture_type = LLViewerTexture::FETCHED_TEXTURE, - LLGLint internal_format = 0, - LLGLenum primary_format = 0, - const LLUUID& force_id = LLUUID::null - ); - - static LLViewerFetchedTexture* getFetchedTextureFromUrl(const std::string& url, - FTType f_type, - BOOL usemipmap = TRUE, - LLViewerTexture::EBoostLevel boost_priority = LLGLTexture::BOOST_NONE, - S8 texture_type = LLViewerTexture::FETCHED_TEXTURE, - LLGLint internal_format = 0, - LLGLenum primary_format = 0, - const LLUUID& force_id = LLUUID::null - ); - - static LLViewerFetchedTexture* getFetchedTextureFromHost(const LLUUID& image_id, FTType f_type, LLHost host) ; - - static void init() ; - static void cleanup() ; -}; -// -//this class is used for test/debug only -//it tracks the activities of the texture pipeline -//records them, and outputs them to log files -// -class LLTexturePipelineTester : public LLMetricPerformanceTesterWithSession -{ - enum - { - MIN_LARGE_IMAGE_AREA = 262144 //512 * 512 - }; -public: - LLTexturePipelineTester() ; - ~LLTexturePipelineTester() ; - - void update(); - void updateTextureBindingStats(const LLViewerTexture* imagep) ; - void updateTextureLoadingStats(const LLViewerFetchedTexture* imagep, const LLImageRaw* raw_imagep, BOOL from_cache) ; - void updateGrayTextureBinding() ; - void setStablizingTime() ; - -private: - void reset() ; - void updateStablizingTime() ; - - /*virtual*/ void outputTestRecord(LLSD* sd) ; - -private: - BOOL mPause ; -private: - BOOL mUsingDefaultTexture; //if set, some textures are still gray. - - U32Bytes mTotalBytesUsed ; //total bytes of textures bound/used for the current frame. - U32Bytes mTotalBytesUsedForLargeImage ; //total bytes of textures bound/used for the current frame for images larger than 256 * 256. - U32Bytes mLastTotalBytesUsed ; //total bytes of textures bound/used for the previous frame. - U32Bytes mLastTotalBytesUsedForLargeImage ; //total bytes of textures bound/used for the previous frame for images larger than 256 * 256. - - // - //data size - // - U32Bytes mTotalBytesLoaded ; //total bytes fetched by texture pipeline - U32Bytes mTotalBytesLoadedFromCache ; //total bytes fetched by texture pipeline from local cache - U32Bytes mTotalBytesLoadedForLargeImage ; //total bytes fetched by texture pipeline for images larger than 256 * 256. - U32Bytes mTotalBytesLoadedForSculpties ; //total bytes fetched by texture pipeline for sculpties - - // - //time - //NOTE: the error tolerances of the following timers is one frame time. - // - F32 mStartFetchingTime ; - F32 mTotalGrayTime ; //total loading time when no gray textures. - F32 mTotalStablizingTime ; //total stablizing time when texture memory overflows - F32 mStartTimeLoadingSculpties ; //the start moment of loading sculpty images. - F32 mEndTimeLoadingSculpties ; //the end moment of loading sculpty images. - F32 mStartStablizingTime ; - F32 mEndStablizingTime ; - -private: - // - //The following members are used for performance analyzing - // - class LLTextureTestSession : public LLTestSession - { - public: - LLTextureTestSession() ; - /*virtual*/ ~LLTextureTestSession() ; - - void reset() ; - - F32 mTotalFetchingTime ; - F32 mTotalGrayTime ; - F32 mTotalStablizingTime ; - F32 mStartTimeLoadingSculpties ; - F32 mTotalTimeLoadingSculpties ; - - S32 mTotalBytesLoaded ; - S32 mTotalBytesLoadedFromCache ; - S32 mTotalBytesLoadedForLargeImage ; - S32 mTotalBytesLoadedForSculpties ; - - typedef struct _texture_instant_preformance_t - { - S32 mAverageBytesUsedPerSecond ; - S32 mAverageBytesUsedForLargeImagePerSecond ; - F32 mAveragePercentageBytesUsedPerSecond ; - F32 mTime ; - }texture_instant_preformance_t ; - std::vector mInstantPerformanceList ; - S32 mInstantPerformanceListCounter ; - }; - - /*virtual*/ LLMetricPerformanceTesterWithSession::LLTestSession* loadTestSession(LLSD* log) ; - /*virtual*/ void compareTestSessions(llofstream* os) ; -}; - -#endif +/** + * @file llviewertexture.h + * @brief Object for managing images and their textures + * + * $LicenseInfo:firstyear=2000&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2010, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_LLVIEWERTEXTURE_H +#define LL_LLVIEWERTEXTURE_H + +#include "llgltexture.h" +#include "lltimer.h" +#include "llframetimer.h" +#include "llhost.h" +#include "llgltypes.h" +#include "llrender.h" +#include "llmetricperformancetester.h" +#include "httpcommon.h" + +#include +#include + +extern const S32Megabytes gMinVideoRam; +extern const S32Megabytes gMaxVideoRam; + +class LLFace; +class LLImageGL ; +class LLImageRaw; +class LLViewerObject; +class LLViewerTexture; +class LLViewerFetchedTexture ; +class LLViewerMediaTexture ; +class LLTexturePipelineTester ; + + +typedef void (*loaded_callback_func)( BOOL success, LLViewerFetchedTexture *src_vi, LLImageRaw* src, LLImageRaw* src_aux, S32 discard_level, BOOL final, void* userdata ); + +class LLVFile; +class LLMessageSystem; +class LLViewerMediaImpl ; +class LLVOVolume ; +struct LLTextureKey; + +class LLLoadedCallbackEntry +{ +public: + typedef std::set< LLTextureKey > source_callback_list_t; + +public: + LLLoadedCallbackEntry(loaded_callback_func cb, + S32 discard_level, + BOOL need_imageraw, // Needs image raw for the callback + void* userdata, + source_callback_list_t* src_callback_list, + LLViewerFetchedTexture* target, + BOOL pause); + ~LLLoadedCallbackEntry(); + void removeTexture(LLViewerFetchedTexture* tex) ; + + loaded_callback_func mCallback; + S32 mLastUsedDiscard; + S32 mDesiredDiscard; + BOOL mNeedsImageRaw; + BOOL mPaused; + void* mUserData; + source_callback_list_t* mSourceCallbackList; + +public: + static void cleanUpCallbackList(LLLoadedCallbackEntry::source_callback_list_t* callback_list) ; +}; + +class LLTextureBar; + +class LLViewerTexture : public LLGLTexture +{ +public: + enum + { + LOCAL_TEXTURE, + MEDIA_TEXTURE, + DYNAMIC_TEXTURE, + FETCHED_TEXTURE, + LOD_TEXTURE, + ATLAS_TEXTURE, + INVALID_TEXTURE_TYPE + }; + + typedef std::vector ll_face_list_t; + typedef std::vector ll_volume_list_t; + + +protected: + virtual ~LLViewerTexture(); + LOG_CLASS(LLViewerTexture); + +public: + static void initClass(); + static void updateClass(const F32 velocity, const F32 angular_velocity) ; + + LLViewerTexture(BOOL usemipmaps = TRUE); + LLViewerTexture(const LLUUID& id, BOOL usemipmaps) ; + LLViewerTexture(const LLImageRaw* raw, BOOL usemipmaps) ; + LLViewerTexture(const U32 width, const U32 height, const U8 components, BOOL usemipmaps) ; + + virtual S8 getType() const; + virtual BOOL isMissingAsset() const ; + virtual void dump(); // debug info to LL_INFOS() + + virtual bool isViewerMediaTexture() const { return false; } + + /*virtual*/ bool bindDefaultImage(const S32 stage = 0) ; + /*virtual*/ bool bindDebugImage(const S32 stage = 0) ; + /*virtual*/ void forceImmediateUpdate() ; + /*virtual*/ bool isActiveFetching(); + + /*virtual*/ const LLUUID& getID() const { return mID; } + void setBoostLevel(S32 level); + S32 getBoostLevel() { return mBoostLevel; } + void setTextureListType(S32 tex_type) { mTextureListType = tex_type; } + S32 getTextureListType() { return mTextureListType; } + + void addTextureStats(F32 virtual_size, BOOL needs_gltexture = TRUE) const; + void resetTextureStats(); + void setMaxVirtualSizeResetInterval(S32 interval)const {mMaxVirtualSizeResetInterval = interval;} + void resetMaxVirtualSizeResetCounter()const {mMaxVirtualSizeResetCounter = mMaxVirtualSizeResetInterval;} + S32 getMaxVirtualSizeResetCounter() const { return mMaxVirtualSizeResetCounter; } + + virtual F32 getMaxVirtualSize() ; + + LLFrameTimer* getLastReferencedTimer() {return &mLastReferencedTimer ;} + + S32 getFullWidth() const { return mFullWidth; } + S32 getFullHeight() const { return mFullHeight; } + /*virtual*/ void setKnownDrawSize(S32 width, S32 height); + + virtual void addFace(U32 channel, LLFace* facep) ; + virtual void removeFace(U32 channel, LLFace* facep) ; + S32 getTotalNumFaces() const; + S32 getNumFaces(U32 ch) const; + const ll_face_list_t* getFaceList(U32 channel) const {llassert(channel < LLRender::NUM_TEXTURE_CHANNELS); return &mFaceList[channel];} + + virtual void addVolume(U32 channel, LLVOVolume* volumep); + virtual void removeVolume(U32 channel, LLVOVolume* volumep); + S32 getNumVolumes(U32 channel) const; + const ll_volume_list_t* getVolumeList(U32 channel) const { return &mVolumeList[channel]; } + + + virtual void setCachedRawImage(S32 discard_level, LLImageRaw* imageraw) ; + BOOL isLargeImage() ; + + void setParcelMedia(LLViewerMediaTexture* media) {mParcelMedia = media;} + BOOL hasParcelMedia() const { return mParcelMedia != NULL;} + LLViewerMediaTexture* getParcelMedia() const { return mParcelMedia;} + + /*virtual*/ void updateBindStatsForTester() ; +protected: + void cleanup() ; + void init(bool firstinit) ; + void reorganizeFaceList() ; + void reorganizeVolumeList() ; + + void notifyAboutMissingAsset(); + void notifyAboutCreatingTexture(); + +private: + friend class LLBumpImageList; + friend class LLUIImageList; + + virtual void switchToCachedImage(); + + static bool isMemoryForTextureLow() ; + static bool isMemoryForTextureSuficientlyFree(); + static void getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &physical); + +protected: + LLUUID mID; + S32 mTextureListType; // along with mID identifies where to search for this texture in TextureList + + F32 mSelectedTime; // time texture was last selected + mutable F32 mMaxVirtualSize; // The largest virtual size of the image, in pixels - how much data to we need? + mutable S32 mMaxVirtualSizeResetCounter ; + mutable S32 mMaxVirtualSizeResetInterval; + mutable F32 mAdditionalDecodePriority; // priority add to mDecodePriority. + LLFrameTimer mLastReferencedTimer; + + ll_face_list_t mFaceList[LLRender::NUM_TEXTURE_CHANNELS]; //reverse pointer pointing to the faces using this image as texture + U32 mNumFaces[LLRender::NUM_TEXTURE_CHANNELS]; + LLFrameTimer mLastFaceListUpdateTimer ; + + ll_volume_list_t mVolumeList[LLRender::NUM_VOLUME_TEXTURE_CHANNELS]; + U32 mNumVolumes[LLRender::NUM_VOLUME_TEXTURE_CHANNELS]; + LLFrameTimer mLastVolumeListUpdateTimer; + + //do not use LLPointer here. + LLViewerMediaTexture* mParcelMedia ; + + static F32 sTexelPixelRatio; +public: + static const U32 sCurrentFileVersion; + static S32 sImageCount; + static S32 sRawCount; + static S32 sAuxCount; + static LLFrameTimer sEvaluationTimer; + static F32 sDesiredDiscardBias; + static F32 sDesiredDiscardScale; + static S32Bytes sBoundTextureMemory; + static S32Bytes sTotalTextureMemory; + static S32Megabytes sMaxBoundTextureMemory; + static S32Megabytes sMaxTotalTextureMem; + static S32Bytes sMaxDesiredTextureMem ; + static S8 sCameraMovingDiscardBias; + static F32 sCameraMovingBias; + static S32 sMaxSculptRez ; + static U32 sMinLargeImageSize ; + static U32 sMaxSmallImageSize ; + static bool sFreezeImageUpdates; + static F32 sCurrentTime ; + + enum EDebugTexels + { + DEBUG_TEXELS_OFF, + DEBUG_TEXELS_CURRENT, + DEBUG_TEXELS_DESIRED, + DEBUG_TEXELS_FULL + }; + + static EDebugTexels sDebugTexelsMode; + + static LLPointer sNullImagep; // Null texture for non-textured objects. + static LLPointer sBlackImagep; // Texture to show NOTHING (pure black) + static LLPointer sCheckerBoardImagep; // Texture to show NOTHING (pure black) +}; + + +enum FTType +{ + FTT_UNKNOWN = -1, + FTT_DEFAULT = 0, // standard texture fetched by id. + FTT_SERVER_BAKE, // texture produced by appearance service and fetched from there. + FTT_HOST_BAKE, // old-style baked texture uploaded by viewer and fetched from avatar's host. + FTT_MAP_TILE, // tiles are fetched from map server directly. + FTT_LOCAL_FILE // fetch directly from a local file. +}; + +const std::string& fttype_to_string(const FTType& fttype); + +// +//textures are managed in gTextureList. +//raw image data is fetched from remote or local cache +//but the raw image this texture pointing to is fixed. +// +class LLViewerFetchedTexture : public LLViewerTexture +{ + friend class LLTextureBar; // debug info only + friend class LLTextureView; // debug info only + +protected: + /*virtual*/ ~LLViewerFetchedTexture(); +public: + LLViewerFetchedTexture(const LLUUID& id, FTType f_type, const LLHost& host = LLHost(), BOOL usemipmaps = TRUE); + LLViewerFetchedTexture(const LLImageRaw* raw, FTType f_type, BOOL usemipmaps); + LLViewerFetchedTexture(const std::string& url, FTType f_type, const LLUUID& id, BOOL usemipmaps = TRUE); + +public: + static F32 maxDecodePriority(); + + struct Compare + { + // lhs < rhs + bool operator()(const LLPointer &lhs, const LLPointer &rhs) const + { + const LLViewerFetchedTexture* lhsp = (const LLViewerFetchedTexture*)lhs; + const LLViewerFetchedTexture* rhsp = (const LLViewerFetchedTexture*)rhs; + // greater priority is "less" + const F32 lpriority = lhsp->getDecodePriority(); + const F32 rpriority = rhsp->getDecodePriority(); + if (lpriority > rpriority) // higher priority + return true; + if (lpriority < rpriority) + return false; + return lhsp < rhsp; + } + }; + +public: + /*virtual*/ S8 getType() const ; + FTType getFTType() const; + /*virtual*/ void forceImmediateUpdate() ; + /*virtual*/ void dump() ; + + // Set callbacks to get called when the image gets updated with higher + // resolution versions. + void setLoadedCallback(loaded_callback_func cb, + S32 discard_level, BOOL keep_imageraw, BOOL needs_aux, + void* userdata, LLLoadedCallbackEntry::source_callback_list_t* src_callback_list, BOOL pause = FALSE); + bool hasCallbacks() { return mLoadedCallbackList.empty() ? false : true; } + void pauseLoadedCallbacks(const LLLoadedCallbackEntry::source_callback_list_t* callback_list); + void unpauseLoadedCallbacks(const LLLoadedCallbackEntry::source_callback_list_t* callback_list); + bool doLoadedCallbacks(); + void deleteCallbackEntry(const LLLoadedCallbackEntry::source_callback_list_t* callback_list); + void clearCallbackEntryList() ; + + void addToCreateTexture(); + + //call to determine if createTexture is necessary + BOOL preCreateTexture(S32 usename = 0); + // ONLY call from LLViewerTextureList or ImageGL background thread + BOOL createTexture(S32 usename = 0); + void postCreateTexture(); + void scheduleCreateTexture(); + + void destroyTexture() ; + + virtual void processTextureStats() ; + F32 calcDecodePriority() ; + + BOOL needsAux() const { return mNeedsAux; } + + // Host we think might have this image, used for baked av textures. + void setTargetHost(LLHost host) { mTargetHost = host; } + LLHost getTargetHost() const { return mTargetHost; } + + // Set the decode priority for this image... + // DON'T CALL THIS UNLESS YOU KNOW WHAT YOU'RE DOING, it can mess up + // the priority list, and cause horrible things to happen. + void setDecodePriority(F32 priority = -1.0f); + F32 getDecodePriority() const { return mDecodePriority; }; + F32 getAdditionalDecodePriority() const { return mAdditionalDecodePriority; }; + + void setAdditionalDecodePriority(F32 priority) ; + + void updateVirtualSize() ; + + S32 getDesiredDiscardLevel() { return mDesiredDiscardLevel; } + void setMinDiscardLevel(S32 discard) { mMinDesiredDiscardLevel = llmin(mMinDesiredDiscardLevel,(S8)discard); } + + bool updateFetch(); + bool setDebugFetching(S32 debug_level); + bool isInDebug() const { return mInDebug; } + + void setUnremovable(BOOL value) { mUnremovable = value; } + bool isUnremovable() const { return mUnremovable; } + + void clearFetchedResults(); //clear all fetched results, for debug use. + + // Override the computation of discard levels if we know the exact output + // size of the image. Used for UI textures to not decode, even if we have + // more data. + /*virtual*/ void setKnownDrawSize(S32 width, S32 height); + + void setIsMissingAsset(BOOL is_missing = true); + /*virtual*/ BOOL isMissingAsset() const { return mIsMissingAsset; } + + // returns dimensions of original image for local files (before power of two scaling) + // and returns 0 for all asset system images + S32 getOriginalWidth() { return mOrigWidth; } + S32 getOriginalHeight() { return mOrigHeight; } + + BOOL isInImageList() const {return mInImageList ;} + void setInImageList(BOOL flag) {mInImageList = flag ;} + + LLFrameTimer* getLastPacketTimer() {return &mLastPacketTimer;} + + U32 getFetchPriority() const { return mFetchPriority ;} + F32 getDownloadProgress() const {return mDownloadProgress ;} + + LLImageRaw* reloadRawImage(S8 discard_level) ; + void destroyRawImage(); + bool needsToSaveRawImage(); + + const std::string& getUrl() const {return mUrl;} + //--------------- + BOOL isDeleted() ; + BOOL isInactive() ; + BOOL isDeletionCandidate(); + void setDeletionCandidate() ; + void setInactive() ; + BOOL getUseDiscard() const { return mUseMipMaps && !mDontDiscard; } + //--------------- + + void setForSculpt(); + BOOL forSculpt() const {return mForSculpt;} + BOOL isForSculptOnly() const; + + //raw image management + void checkCachedRawSculptImage() ; + LLImageRaw* getRawImage()const { return mRawImage ;} + S32 getRawImageLevel() const {return mRawDiscardLevel;} + LLImageRaw* getCachedRawImage() const { return mCachedRawImage ;} + S32 getCachedRawImageLevel() const {return mCachedRawDiscardLevel;} + BOOL isCachedRawImageReady() const {return mCachedRawImageReady ;} + BOOL isRawImageValid()const { return mIsRawImageValid ; } + void forceToSaveRawImage(S32 desired_discard = 0, F32 kept_time = 0.f) ; + void forceToRefetchTexture(S32 desired_discard = 0, F32 kept_time = 60.f); + /*virtual*/ void setCachedRawImage(S32 discard_level, LLImageRaw* imageraw) ; + void destroySavedRawImage() ; + LLImageRaw* getSavedRawImage() ; + BOOL hasSavedRawImage() const ; + F32 getElapsedLastReferencedSavedRawImageTime() const ; + BOOL isFullyLoaded() const; + + BOOL hasFetcher() const { return mHasFetcher;} + void setCanUseHTTP(bool can_use_http) {mCanUseHTTP = can_use_http;} + + void forceToDeleteRequest(); + void loadFromFastCache(); + void setInFastCacheList(bool in_list) { mInFastCacheList = in_list; } + bool isInFastCacheList() { return mInFastCacheList; } + + /*virtual*/bool isActiveFetching(); //is actively in fetching by the fetching pipeline. + +protected: + /*virtual*/ void switchToCachedImage(); + S32 getCurrentDiscardLevelForFetching() ; + +private: + void init(bool firstinit) ; + void cleanup() ; + + void saveRawImage() ; + void setCachedRawImage() ; + + //for atlas + void resetFaceAtlas() ; + void invalidateAtlas(BOOL rebuild_geom) ; + BOOL insertToAtlas() ; + +private: + BOOL mFullyLoaded; + BOOL mInDebug; + BOOL mUnremovable; + BOOL mInFastCacheList; + BOOL mForceCallbackFetch; + +protected: + std::string mLocalFileName; + + S32 mOrigWidth; + S32 mOrigHeight; + + // Override the computation of discard levels if we know the exact output size of the image. + // Used for UI textures to not decode, even if we have more data. + S32 mKnownDrawWidth; + S32 mKnownDrawHeight; + BOOL mKnownDrawSizeChanged ; + std::string mUrl; + + S32 mRequestedDiscardLevel; + F32 mRequestedDownloadPriority; + S32 mFetchState; + U32 mFetchPriority; + F32 mDownloadProgress; + F32 mFetchDeltaTime; + F32 mRequestDeltaTime; + F32 mDecodePriority; // The priority for decoding this image. + S32 mMinDiscardLevel; + S8 mDesiredDiscardLevel; // The discard level we'd LIKE to have - if we have it and there's space + S8 mMinDesiredDiscardLevel; // The minimum discard level we'd like to have + + S8 mNeedsAux; // We need to decode the auxiliary channels + S8 mHasAux; // We have aux channels + S8 mDecodingAux; // Are we decoding high components + S8 mIsRawImageValid; + S8 mHasFetcher; // We've made a fecth request + S8 mIsFetching; // Fetch request is active + bool mCanUseHTTP; //This texture can be fetched through http if true. + LLCore::HttpStatus mLastHttpGetStatus; // Result of the most recently completed http request for this texture. + + FTType mFTType; // What category of image is this - map tile, server bake, etc? + mutable S8 mIsMissingAsset; // True if we know that there is no image asset with this image id in the database. + + typedef std::list callback_list_t; + S8 mLoadedCallbackDesiredDiscardLevel; + BOOL mPauseLoadedCallBacks; + callback_list_t mLoadedCallbackList; + F32 mLastCallBackActiveTime; + + LLPointer mRawImage; + S32 mRawDiscardLevel; + + // Used ONLY for cloth meshes right now. Make SURE you know what you're + // doing if you use it for anything else! - djs + LLPointer mAuxRawImage; + + //keep a copy of mRawImage for some special purposes + //when mForceToSaveRawImage is set. + BOOL mForceToSaveRawImage ; + BOOL mSaveRawImage; + LLPointer mSavedRawImage; + S32 mSavedRawDiscardLevel; + S32 mDesiredSavedRawDiscardLevel; + F32 mLastReferencedSavedRawImageTime ; + F32 mKeptSavedRawImageTime ; + + //a small version of the copy of the raw image (<= 64 * 64) + LLPointer mCachedRawImage; + S32 mCachedRawDiscardLevel; + BOOL mCachedRawImageReady; //the rez of the mCachedRawImage reaches the upper limit. + + LLHost mTargetHost; // if invalid, just request from agent's simulator + + // Timers + LLFrameTimer mLastPacketTimer; // Time since last packet. + LLFrameTimer mStopFetchingTimer; // Time since mDecodePriority == 0.f. + + BOOL mInImageList; // TRUE if image is in list (in which case don't reset priority!) + BOOL mNeedsCreateTexture; + + BOOL mForSculpt ; //a flag if the texture is used as sculpt data. + BOOL mIsFetched ; //is loaded from remote or from cache, not generated locally. + +public: + static LLPointer sMissingAssetImagep; // Texture to show for an image asset that is not in the database + static LLPointer sWhiteImagep; // Texture to show NOTHING (whiteness) + static LLPointer sDefaultImagep; // "Default" texture for error cases, the only case of fetched texture which is generated in local. + static LLPointer sSmokeImagep; // Old "Default" translucent texture + static LLPointer sFlatNormalImagep; // Flat normal map denoting no bumpiness on a surface +}; + +// +//the image data is fetched from remote or from local cache +//the resolution of the texture is adjustable: depends on the view-dependent parameters. +// +class LLViewerLODTexture : public LLViewerFetchedTexture +{ +protected: + /*virtual*/ ~LLViewerLODTexture(){} + +public: + LLViewerLODTexture(const LLUUID& id, FTType f_type, const LLHost& host = LLHost(), BOOL usemipmaps = TRUE); + LLViewerLODTexture(const std::string& url, FTType f_type, const LLUUID& id, BOOL usemipmaps = TRUE); + + /*virtual*/ S8 getType() const; + // Process image stats to determine priority/quality requirements. + /*virtual*/ void processTextureStats(); + bool isUpdateFrozen() ; + +private: + void init(bool firstinit) ; + bool scaleDown() ; + +private: + F32 mDiscardVirtualSize; // Virtual size used to calculate desired discard + F32 mCalculatedDiscardLevel; // Last calculated discard level +}; + +// +//the image data is fetched from the media pipeline periodically +//the resolution of the texture is also adjusted by the media pipeline +// +class LLViewerMediaTexture : public LLViewerTexture +{ +protected: + /*virtual*/ ~LLViewerMediaTexture() ; + +public: + LLViewerMediaTexture(const LLUUID& id, BOOL usemipmaps = TRUE, LLImageGL* gl_image = NULL) ; + + /*virtual*/ S8 getType() const; + void reinit(BOOL usemipmaps = TRUE); + + BOOL getUseMipMaps() {return mUseMipMaps ; } + void setUseMipMaps(BOOL mipmap) ; + + void setPlaying(BOOL playing) ; + BOOL isPlaying() const {return mIsPlaying;} + void setMediaImpl() ; + + virtual bool isViewerMediaTexture() const { return true; } + + void initVirtualSize() ; + void invalidateMediaImpl() ; + + void addMediaToFace(LLFace* facep) ; + void removeMediaFromFace(LLFace* facep) ; + + /*virtual*/ void addFace(U32 ch, LLFace* facep) ; + /*virtual*/ void removeFace(U32 ch, LLFace* facep) ; + + /*virtual*/ F32 getMaxVirtualSize() ; +private: + void switchTexture(U32 ch, LLFace* facep) ; + BOOL findFaces() ; + void stopPlaying() ; + +private: + // + //an instant list, recording all faces referencing or can reference to this media texture. + //NOTE: it is NOT thread safe. + // + std::list< LLFace* > mMediaFaceList ; + + //an instant list keeping all textures which are replaced by the current media texture, + //is only used to avoid the removal of those textures from memory. + std::list< LLPointer > mTextureList ; + + LLViewerMediaImpl* mMediaImplp ; + BOOL mIsPlaying ; + U32 mUpdateVirtualSizeTime ; + +public: + static void updateClass() ; + static void cleanUpClass() ; + + static LLViewerMediaTexture* findMediaTexture(const LLUUID& media_id) ; + static void removeMediaImplFromTexture(const LLUUID& media_id) ; + +private: + typedef std::map< LLUUID, LLPointer > media_map_t ; + static media_map_t sMediaMap ; +}; + +//just an interface class, do not create instance from this class. +class LLViewerTextureManager +{ +private: + //make the constructor private to preclude creating instances from this class. + LLViewerTextureManager(){} + +public: + //texture pipeline tester + static LLTexturePipelineTester* sTesterp ; + + //returns NULL if tex is not a LLViewerFetchedTexture nor derived from LLViewerFetchedTexture. + static LLViewerFetchedTexture* staticCastToFetchedTexture(LLTexture* tex, BOOL report_error = FALSE) ; + + // + //"find-texture" just check if the texture exists, if yes, return it, otherwise return null. + // + static void findFetchedTextures(const LLUUID& id, std::vector &output); + static void findTextures(const LLUUID& id, std::vector &output); + static LLViewerFetchedTexture* findFetchedTexture(const LLUUID& id, S32 tex_type); + static LLViewerMediaTexture* findMediaTexture(const LLUUID& id) ; + + static LLViewerMediaTexture* createMediaTexture(const LLUUID& id, BOOL usemipmaps = TRUE, LLImageGL* gl_image = NULL) ; + + // + //"get-texture" will create a new texture if the texture does not exist. + // + static LLViewerMediaTexture* getMediaTexture(const LLUUID& id, BOOL usemipmaps = TRUE, LLImageGL* gl_image = NULL) ; + + static LLPointer getLocalTexture(BOOL usemipmaps = TRUE, BOOL generate_gl_tex = TRUE); + static LLPointer getLocalTexture(const LLUUID& id, BOOL usemipmaps, BOOL generate_gl_tex = TRUE) ; + static LLPointer getLocalTexture(const LLImageRaw* raw, BOOL usemipmaps) ; + static LLPointer getLocalTexture(const U32 width, const U32 height, const U8 components, BOOL usemipmaps, BOOL generate_gl_tex = TRUE) ; + + static LLViewerFetchedTexture* getFetchedTexture(const LLUUID &image_id, + FTType f_type = FTT_DEFAULT, + BOOL usemipmap = TRUE, + LLViewerTexture::EBoostLevel boost_priority = LLGLTexture::BOOST_NONE, // Get the requested level immediately upon creation. + S8 texture_type = LLViewerTexture::FETCHED_TEXTURE, + LLGLint internal_format = 0, + LLGLenum primary_format = 0, + LLHost request_from_host = LLHost() + ); + + static LLViewerFetchedTexture* getFetchedTextureFromFile(const std::string& filename, + FTType f_type = FTT_LOCAL_FILE, + BOOL usemipmap = TRUE, + LLViewerTexture::EBoostLevel boost_priority = LLGLTexture::BOOST_NONE, + S8 texture_type = LLViewerTexture::FETCHED_TEXTURE, + LLGLint internal_format = 0, + LLGLenum primary_format = 0, + const LLUUID& force_id = LLUUID::null + ); + + static LLViewerFetchedTexture* getFetchedTextureFromUrl(const std::string& url, + FTType f_type, + BOOL usemipmap = TRUE, + LLViewerTexture::EBoostLevel boost_priority = LLGLTexture::BOOST_NONE, + S8 texture_type = LLViewerTexture::FETCHED_TEXTURE, + LLGLint internal_format = 0, + LLGLenum primary_format = 0, + const LLUUID& force_id = LLUUID::null + ); + + static LLViewerFetchedTexture* getFetchedTextureFromHost(const LLUUID& image_id, FTType f_type, LLHost host) ; + + static void init() ; + static void cleanup() ; +}; +// +//this class is used for test/debug only +//it tracks the activities of the texture pipeline +//records them, and outputs them to log files +// +class LLTexturePipelineTester : public LLMetricPerformanceTesterWithSession +{ + enum + { + MIN_LARGE_IMAGE_AREA = 262144 //512 * 512 + }; +public: + LLTexturePipelineTester() ; + ~LLTexturePipelineTester() ; + + void update(); + void updateTextureBindingStats(const LLViewerTexture* imagep) ; + void updateTextureLoadingStats(const LLViewerFetchedTexture* imagep, const LLImageRaw* raw_imagep, BOOL from_cache) ; + void updateGrayTextureBinding() ; + void setStablizingTime() ; + +private: + void reset() ; + void updateStablizingTime() ; + + /*virtual*/ void outputTestRecord(LLSD* sd) ; + +private: + BOOL mPause ; +private: + BOOL mUsingDefaultTexture; //if set, some textures are still gray. + + U32Bytes mTotalBytesUsed ; //total bytes of textures bound/used for the current frame. + U32Bytes mTotalBytesUsedForLargeImage ; //total bytes of textures bound/used for the current frame for images larger than 256 * 256. + U32Bytes mLastTotalBytesUsed ; //total bytes of textures bound/used for the previous frame. + U32Bytes mLastTotalBytesUsedForLargeImage ; //total bytes of textures bound/used for the previous frame for images larger than 256 * 256. + + // + //data size + // + U32Bytes mTotalBytesLoaded ; //total bytes fetched by texture pipeline + U32Bytes mTotalBytesLoadedFromCache ; //total bytes fetched by texture pipeline from local cache + U32Bytes mTotalBytesLoadedForLargeImage ; //total bytes fetched by texture pipeline for images larger than 256 * 256. + U32Bytes mTotalBytesLoadedForSculpties ; //total bytes fetched by texture pipeline for sculpties + + // + //time + //NOTE: the error tolerances of the following timers is one frame time. + // + F32 mStartFetchingTime ; + F32 mTotalGrayTime ; //total loading time when no gray textures. + F32 mTotalStablizingTime ; //total stablizing time when texture memory overflows + F32 mStartTimeLoadingSculpties ; //the start moment of loading sculpty images. + F32 mEndTimeLoadingSculpties ; //the end moment of loading sculpty images. + F32 mStartStablizingTime ; + F32 mEndStablizingTime ; + +private: + // + //The following members are used for performance analyzing + // + class LLTextureTestSession : public LLTestSession + { + public: + LLTextureTestSession() ; + /*virtual*/ ~LLTextureTestSession() ; + + void reset() ; + + F32 mTotalFetchingTime ; + F32 mTotalGrayTime ; + F32 mTotalStablizingTime ; + F32 mStartTimeLoadingSculpties ; + F32 mTotalTimeLoadingSculpties ; + + S32 mTotalBytesLoaded ; + S32 mTotalBytesLoadedFromCache ; + S32 mTotalBytesLoadedForLargeImage ; + S32 mTotalBytesLoadedForSculpties ; + + typedef struct _texture_instant_preformance_t + { + S32 mAverageBytesUsedPerSecond ; + S32 mAverageBytesUsedForLargeImagePerSecond ; + F32 mAveragePercentageBytesUsedPerSecond ; + F32 mTime ; + }texture_instant_preformance_t ; + std::vector mInstantPerformanceList ; + S32 mInstantPerformanceListCounter ; + }; + + /*virtual*/ LLMetricPerformanceTesterWithSession::LLTestSession* loadTestSession(LLSD* log) ; + /*virtual*/ void compareTestSessions(llofstream* os) ; +}; + +#endif -- cgit v1.3 From d2dce17803a545378407d6b7c62fdcd3007a92bc Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 19 Oct 2021 02:26:41 +0000 Subject: SL-16197 Optimize LLEnvironment handling of shader uniforms. Instrument LLSD. Enable Fast Timers when Tracy is enabled to catch Fast Timer overhead. --- indra/cmake/Tracy.cmake | 2 +- indra/llcommon/llsd.cpp | 32 +++++++++++--- indra/llcommon/llsd.h | 12 +++++- indra/llinventory/llsettingsbase.cpp | 4 ++ indra/llinventory/llsettingssky.cpp | 3 ++ indra/llrender/llglslshader.cpp | 68 +++++++++++++++++++++--------- indra/llrender/llglslshader.h | 82 ++++++++++++++++++++++++++++++++---- indra/newview/lldrawpoolbump.cpp | 1 + indra/newview/lldrawpoolterrain.cpp | 2 - indra/newview/lldrawpoolwlsky.cpp | 6 --- indra/newview/llenvironment.cpp | 39 ++++++++++++++--- indra/newview/llenvironment.h | 18 ++++++-- indra/newview/llsettingsvo.cpp | 68 +++++++++++++++--------------- indra/newview/llsettingsvo.h | 4 -- indra/newview/pipeline.cpp | 2 - 15 files changed, 247 insertions(+), 96 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/cmake/Tracy.cmake b/indra/cmake/Tracy.cmake index 1b8c3db2e2..cfff956bcf 100644 --- a/indra/cmake/Tracy.cmake +++ b/indra/cmake/Tracy.cmake @@ -7,7 +7,7 @@ if (USE_TRACY) set(TRACY_INCLUDE_DIR ${LIBS_PREBUILT_DIR}/include/tracy) # See: indra/llcommon/llprofiler.h - add_definitions(-DLL_PROFILER_CONFIGURATION=2) + add_definitions(-DLL_PROFILER_CONFIGURATION=3) use_prebuilt_binary(tracy) if (WINDOWS) diff --git a/indra/llcommon/llsd.cpp b/indra/llcommon/llsd.cpp index 57b746889d..605f6bf0e3 100644 --- a/indra/llcommon/llsd.cpp +++ b/indra/llcommon/llsd.cpp @@ -400,6 +400,7 @@ namespace ImplMap& ImplMap::makeMap(LLSD::Impl*& var) { + LL_PROFILE_ZONE_SCOPED; if (shared()) { ImplMap* i = new ImplMap(mData); @@ -414,18 +415,21 @@ namespace bool ImplMap::has(const LLSD::String& k) const { + LL_PROFILE_ZONE_SCOPED; DataMap::const_iterator i = mData.find(k); return i != mData.end(); } LLSD ImplMap::get(const LLSD::String& k) const { + LL_PROFILE_ZONE_SCOPED; DataMap::const_iterator i = mData.find(k); return (i != mData.end()) ? i->second : LLSD(); } LLSD ImplMap::getKeys() const { + LL_PROFILE_ZONE_SCOPED; LLSD keys = LLSD::emptyArray(); DataMap::const_iterator iter = mData.begin(); while (iter != mData.end()) @@ -438,11 +442,13 @@ namespace void ImplMap::insert(const LLSD::String& k, const LLSD& v) { + LL_PROFILE_ZONE_SCOPED; mData.insert(DataMap::value_type(k, v)); } void ImplMap::erase(const LLSD::String& k) { + LL_PROFILE_ZONE_SCOPED; mData.erase(k); } @@ -684,6 +690,7 @@ const LLSD::Impl& LLSD::Impl::safe(const Impl* impl) ImplMap& LLSD::Impl::makeMap(Impl*& var) { + LL_PROFILE_ZONE_SCOPED; ImplMap* im = new ImplMap; reset(var, im); return *im; @@ -887,11 +894,16 @@ LLSD& LLSD::with(const String& k, const LLSD& v) } void LLSD::erase(const String& k) { makeMap(impl).erase(k); } -LLSD& LLSD::operator[](const String& k) - { return makeMap(impl).ref(k); } +LLSD& LLSD::operator[](const String& k) +{ + LL_PROFILE_ZONE_SCOPED; + return makeMap(impl).ref(k); +} const LLSD& LLSD::operator[](const String& k) const - { return safe(impl).ref(k); } - +{ + LL_PROFILE_ZONE_SCOPED; + return safe(impl).ref(k); +} LLSD LLSD::emptyArray() { @@ -914,10 +926,16 @@ LLSD& LLSD::with(Integer i, const LLSD& v) LLSD& LLSD::append(const LLSD& v) { return makeArray(impl).append(v); } void LLSD::erase(Integer i) { makeArray(impl).erase(i); } -LLSD& LLSD::operator[](Integer i) - { return makeArray(impl).ref(i); } +LLSD& LLSD::operator[](Integer i) +{ + LL_PROFILE_ZONE_SCOPED; + return makeArray(impl).ref(i); +} const LLSD& LLSD::operator[](Integer i) const - { return safe(impl).ref(i); } +{ + LL_PROFILE_ZONE_SCOPED; + return safe(impl).ref(i); +} static const char *llsd_dump(const LLSD &llsd, bool useXMLFormat) { diff --git a/indra/llcommon/llsd.h b/indra/llcommon/llsd.h index 5b6d5545af..b8ddf21596 100644 --- a/indra/llcommon/llsd.h +++ b/indra/llcommon/llsd.h @@ -290,9 +290,17 @@ public: LLSD& with(const String&, const LLSD&); LLSD& operator[](const String&); - LLSD& operator[](const char* c) { return (*this)[String(c)]; } + LLSD& operator[](const char* c) + { + LL_PROFILE_ZONE_SCOPED; + return (*this)[String(c)]; + } const LLSD& operator[](const String&) const; - const LLSD& operator[](const char* c) const { return (*this)[String(c)]; } + const LLSD& operator[](const char* c) const + { + LL_PROFILE_ZONE_SCOPED; + return (*this)[String(c)]; + } //@} /** @name Array Values */ diff --git a/indra/llinventory/llsettingsbase.cpp b/indra/llinventory/llsettingsbase.cpp index 61b59e35aa..89a4eebf26 100644 --- a/indra/llinventory/llsettingsbase.cpp +++ b/indra/llinventory/llsettingsbase.cpp @@ -683,6 +683,7 @@ bool LLSettingsBase::Validator::verifyStringLength(LLSD &value, S32 length) //========================================================================= void LLSettingsBlender::update(const LLSettingsBase::BlendFactor& blendf) { + LL_PROFILE_ZONE_SCOPED; F64 res = setBlendFactor(blendf); llassert(res >= 0.0 && res <= 1.0); (void)res; @@ -713,6 +714,7 @@ F64 LLSettingsBlender::setBlendFactor(const LLSettingsBase::BlendFactor& blendf_ void LLSettingsBlender::triggerComplete() { + LL_PROFILE_ZONE_SCOPED; if (mTarget) mTarget->replaceSettings(mFinal->getSettings()); LLSettingsBlender::ptr_t hold = shared_from_this(); // prevents this from deleting too soon @@ -725,11 +727,13 @@ const LLSettingsBase::BlendFactor LLSettingsBlenderTimeDelta::MIN_BLEND_DELTA(FL LLSettingsBase::BlendFactor LLSettingsBlenderTimeDelta::calculateBlend(const LLSettingsBase::TrackPosition& spanpos, const LLSettingsBase::TrackPosition& spanlen) const { + LL_PROFILE_ZONE_SCOPED; return LLSettingsBase::BlendFactor(fmod((F64)spanpos, (F64)spanlen) / (F64)spanlen); } bool LLSettingsBlenderTimeDelta::applyTimeDelta(const LLSettingsBase::Seconds& timedelta) { + LL_PROFILE_ZONE_SCOPED; mTimeSpent += timedelta; if (mTimeSpent > mBlendSpan) diff --git a/indra/llinventory/llsettingssky.cpp b/indra/llinventory/llsettingssky.cpp index 81937dbda5..00c1edb55a 100644 --- a/indra/llinventory/llsettingssky.cpp +++ b/indra/llinventory/llsettingssky.cpp @@ -444,6 +444,7 @@ void LLSettingsSky::replaceWithSky(LLSettingsSky::ptr_t pother) void LLSettingsSky::blend(const LLSettingsBase::ptr_t &end, F64 blendf) { + LL_PROFILE_ZONE_SCOPED; llassert(getSettingsType() == end->getSettingsType()); LLSettingsSky::ptr_t other = PTR_NAMESPACE::dynamic_pointer_cast(end); @@ -1022,6 +1023,7 @@ LLColor3 LLSettingsSky::getLightDiffuse() const LLColor3 LLSettingsSky::getColor(const std::string& key, const LLColor3& default_value) const { + LL_PROFILE_ZONE_SCOPED; if (mSettings.has(SETTING_LEGACY_HAZE) && mSettings[SETTING_LEGACY_HAZE].has(key)) { return LLColor3(mSettings[SETTING_LEGACY_HAZE][key]); @@ -1035,6 +1037,7 @@ LLColor3 LLSettingsSky::getColor(const std::string& key, const LLColor3& default F32 LLSettingsSky::getFloat(const std::string& key, F32 default_value) const { + LL_PROFILE_ZONE_SCOPED; if (mSettings.has(SETTING_LEGACY_HAZE) && mSettings[SETTING_LEGACY_HAZE].has(key)) { return mSettings[SETTING_LEGACY_HAZE][key].asReal(); diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp index 84eac00c65..0e4753fcc6 100644 --- a/indra/llrender/llglslshader.cpp +++ b/indra/llrender/llglslshader.cpp @@ -1104,6 +1104,7 @@ S32 LLGLSLShader::disableTexture(S32 uniform, LLTexUnit::eTextureType mode, LLTe void LLGLSLShader::uniform1i(U32 index, GLint x) { + LL_PROFILE_ZONE_SCOPED if (mProgramObject) { if (mUniform.size() <= index) @@ -1114,7 +1115,7 @@ void LLGLSLShader::uniform1i(U32 index, GLint x) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); if (iter == mValue.end() || iter->second.mV[0] != x) { glUniform1iARB(mUniform[index], x); @@ -1126,6 +1127,7 @@ void LLGLSLShader::uniform1i(U32 index, GLint x) void LLGLSLShader::uniform1f(U32 index, GLfloat x) { + LL_PROFILE_ZONE_SCOPED if (mProgramObject) { if (mUniform.size() <= index) @@ -1136,7 +1138,7 @@ void LLGLSLShader::uniform1f(U32 index, GLfloat x) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); if (iter == mValue.end() || iter->second.mV[0] != x) { glUniform1fARB(mUniform[index], x); @@ -1158,7 +1160,7 @@ void LLGLSLShader::uniform2f(U32 index, GLfloat x, GLfloat y) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(x,y,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1181,7 +1183,7 @@ void LLGLSLShader::uniform3f(U32 index, GLfloat x, GLfloat y, GLfloat z) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(x,y,z,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1204,7 +1206,7 @@ void LLGLSLShader::uniform4f(U32 index, GLfloat x, GLfloat y, GLfloat z, GLfloat if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(x,y,z,w); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1227,7 +1229,7 @@ void LLGLSLShader::uniform1iv(U32 index, U32 count, const GLint* v) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(v[0],0.f,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { @@ -1250,7 +1252,7 @@ void LLGLSLShader::uniform1fv(U32 index, U32 count, const GLfloat* v) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(v[0],0.f,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { @@ -1273,7 +1275,7 @@ void LLGLSLShader::uniform2fv(U32 index, U32 count, const GLfloat* v) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(v[0],v[1],0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { @@ -1296,7 +1298,7 @@ void LLGLSLShader::uniform3fv(U32 index, U32 count, const GLfloat* v) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(v[0],v[1],v[2],0.f); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { @@ -1319,10 +1321,11 @@ void LLGLSLShader::uniform4fv(U32 index, U32 count, const GLfloat* v) if (mUniform[index] >= 0) { - std::map::iterator iter = mValue.find(mUniform[index]); + const auto& iter = mValue.find(mUniform[index]); LLVector4 vec(v[0],v[1],v[2],v[3]); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { + LL_PROFILE_ZONE_SCOPED; glUniform4fvARB(mUniform[index], count, v); mValue[mUniform[index]] = vec; } @@ -1460,7 +1463,7 @@ void LLGLSLShader::uniform1i(const LLStaticHashedString& uniform, GLint v) if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(v,0.f,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1476,7 +1479,7 @@ void LLGLSLShader::uniform2i(const LLStaticHashedString& uniform, GLint i, GLint if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(i,j,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1493,7 +1496,7 @@ void LLGLSLShader::uniform1f(const LLStaticHashedString& uniform, GLfloat v) if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(v,0.f,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1509,7 +1512,7 @@ void LLGLSLShader::uniform2f(const LLStaticHashedString& uniform, GLfloat x, GLf if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(x,y,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1526,7 +1529,7 @@ void LLGLSLShader::uniform3f(const LLStaticHashedString& uniform, GLfloat x, GLf if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(x,y,z,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec)) { @@ -1542,7 +1545,7 @@ void LLGLSLShader::uniform1fv(const LLStaticHashedString& uniform, U32 count, co if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(v[0],0.f,0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { @@ -1558,7 +1561,7 @@ void LLGLSLShader::uniform2fv(const LLStaticHashedString& uniform, U32 count, co if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(v[0],v[1],0.f,0.f); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { @@ -1574,7 +1577,7 @@ void LLGLSLShader::uniform3fv(const LLStaticHashedString& uniform, U32 count, co if (location >= 0) { - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); LLVector4 vec(v[0],v[1],v[2],0.f); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { @@ -1591,12 +1594,11 @@ void LLGLSLShader::uniform4fv(const LLStaticHashedString& uniform, U32 count, co if (location >= 0) { LLVector4 vec(v); - std::map::iterator iter = mValue.find(location); + const auto& iter = mValue.find(location); if (iter == mValue.end() || shouldChange(iter->second,vec) || count != 1) { - stop_glerror(); + LL_PROFILE_ZONE_SCOPED; glUniform4fvARB(location, count, v); - stop_glerror(); mValue[location] = vec; } } @@ -1636,3 +1638,27 @@ void LLGLSLShader::setMinimumAlpha(F32 minimum) gGL.flush(); uniform1f(LLShaderMgr::MINIMUM_ALPHA, minimum); } + +void LLShaderUniforms::apply(LLGLSLShader* shader) +{ + LL_PROFILE_ZONE_SCOPED; + for (auto& uniform : mIntegers) + { + shader->uniform1i(uniform.mUniform, uniform.mValue); + } + + for (auto& uniform : mFloats) + { + shader->uniform1f(uniform.mUniform, uniform.mValue); + } + + for (auto& uniform : mVectors) + { + shader->uniform4fv(uniform.mUniform, 1, uniform.mValue.mV); + } + + for (auto& uniform : mVector3s) + { + shader->uniform3fv(uniform.mUniform, 1, uniform.mValue.mV); + } +} diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h index 7cf6d3c941..3b23cf1b28 100644 --- a/indra/llrender/llglslshader.h +++ b/indra/llrender/llglslshader.h @@ -30,6 +30,7 @@ #include "llgl.h" #include "llrender.h" #include "llstaticstringtable.h" +#include class LLShaderFeatures { @@ -64,16 +65,79 @@ public: LLShaderFeatures(); }; +// ============= Structure for caching shader uniforms =============== +class LLGLSLShader; + +class LLShaderUniforms +{ +public: + + template + struct UniformSetting + { + S32 mUniform; + T mValue; + }; + + typedef UniformSetting IntSetting; + typedef UniformSetting FloatSetting; + typedef UniformSetting VectorSetting; + typedef UniformSetting Vector3Setting; + + void clear() + { + mIntegers.resize(0); + mFloats.resize(0); + mVectors.resize(0); + mVector3s.resize(0); + } + + void uniform1i(S32 index, S32 value) + { + mIntegers.push_back({ index, value }); + } + + void uniform1f(S32 index, F32 value) + { + mFloats.push_back({ index, value }); + } + + void uniform4fv(S32 index, const LLVector4& value) + { + mVectors.push_back({ index, value }); + } + + void uniform4fv(S32 index, const F32* value) + { + mVectors.push_back({ index, LLVector4(value) }); + } + + void uniform3fv(S32 index, const LLVector3& value) + { + mVector3s.push_back({ index, value }); + } + + void apply(LLGLSLShader* shader); + + + std::vector mIntegers; + std::vector mFloats; + std::vector mVectors; + std::vector mVector3s; +}; class LLGLSLShader { public: - enum + // enum primarily used to control application sky settings uniforms + typedef enum { - SG_DEFAULT = 0, - SG_SKY, - SG_WATER - }; + SG_DEFAULT = 0, // not sky or water specific + SG_SKY, // + SG_WATER, + SG_ANY, + SG_COUNT + } eGroup; static std::set sInstances; static bool sProfileEnabled; @@ -190,13 +254,15 @@ public: U32 mAttributeMask; //mask of which reserved attributes are set (lines up with LLVertexBuffer::getTypeMask()) std::vector mUniform; //lookup table of uniform enum to uniform location LLStaticStringTable mUniformMap; //lookup map of uniform name to uniform location - std::map mUniformNameMap; //lookup map of uniform location to uniform name - std::map mValue; //lookup map of uniform location to last known value + typedef std::unordered_map uniform_name_map_t; + typedef std::unordered_map uniform_value_map_t; + uniform_name_map_t mUniformNameMap; //lookup map of uniform location to uniform name + uniform_value_map_t mValue; //lookup map of uniform location to last known value std::vector mTexture; S32 mTotalUniformSize; S32 mActiveTextureChannels; S32 mShaderLevel; - S32 mShaderGroup; + S32 mShaderGroup; // see LLGLSLShader::eGroup BOOL mUniformsDirty; LLShaderFeatures mFeatures; std::vector< std::pair< std::string, GLenum > > mShaderFiles; diff --git a/indra/newview/lldrawpoolbump.cpp b/indra/newview/lldrawpoolbump.cpp index f316d121ab..d75884cc16 100644 --- a/indra/newview/lldrawpoolbump.cpp +++ b/indra/newview/lldrawpoolbump.cpp @@ -1062,6 +1062,7 @@ void LLBumpImageList::updateImages() // Note: the caller SHOULD NOT keep the pointer that this function returns. It may be updated as more data arrives. LLViewerTexture* LLBumpImageList::getBrightnessDarknessImage(LLViewerFetchedTexture* src_image, U8 bump_code ) { + LL_PROFILE_ZONE_SCOPED; llassert( (bump_code == BE_BRIGHTNESS) || (bump_code == BE_DARKNESS) ); LLViewerTexture* bump = NULL; diff --git a/indra/newview/lldrawpoolterrain.cpp b/indra/newview/lldrawpoolterrain.cpp index 37dc80e2b7..34a8b6b2cc 100644 --- a/indra/newview/lldrawpoolterrain.cpp +++ b/indra/newview/lldrawpoolterrain.cpp @@ -343,8 +343,6 @@ void LLDrawPoolTerrain::renderFullShader() LLSettingsWater::ptr_t pwater = LLEnvironment::instance().getCurrentWater(); - ((LLSettingsVOWater*)pwater.get())->updateShader(shader); - // // detail texture 1 // diff --git a/indra/newview/lldrawpoolwlsky.cpp b/indra/newview/lldrawpoolwlsky.cpp index 0c3d8f3098..8c8dc3f3d2 100644 --- a/indra/newview/lldrawpoolwlsky.cpp +++ b/indra/newview/lldrawpoolwlsky.cpp @@ -182,8 +182,6 @@ void LLDrawPoolWLSky::renderSkyHazeDeferred(const LLVector3& camPosLocal, F32 ca sky_shader->bindTexture(LLShaderMgr::RAINBOW_MAP, rainbow_tex); sky_shader->bindTexture(LLShaderMgr::HALO_MAP, halo_tex); - ((LLSettingsVOSky*)psky.get())->updateShader(sky_shader); - F32 moisture_level = (float)psky->getSkyMoistureLevel(); F32 droplet_radius = (float)psky->getSkyDropletRadius(); F32 ice_level = (float)psky->getSkyIceLevel(); @@ -406,8 +404,6 @@ void LLDrawPoolWLSky::renderSkyCloudsDeferred(const LLVector3& camPosLocal, F32 cloudshader->uniform1f(LLShaderMgr::CLOUD_VARIANCE, cloud_variance); cloudshader->uniform1f(LLShaderMgr::SUN_MOON_GLOW_FACTOR, psky->getSunMoonGlowFactor()); - ((LLSettingsVOSky*)psky.get())->updateShader(cloudshader); - /// Render the skydome renderDome(camPosLocal, camHeightLocal, cloudshader); @@ -462,8 +458,6 @@ void LLDrawPoolWLSky::renderSkyClouds(const LLVector3& camPosLocal, F32 camHeigh cloudshader->uniform1f(LLShaderMgr::CLOUD_VARIANCE, cloud_variance); cloudshader->uniform1f(LLShaderMgr::SUN_MOON_GLOW_FACTOR, psky->getSunMoonGlowFactor()); - ((LLSettingsVOSky*)psky.get())->updateShader(cloudshader); - /// Render the skydome renderDome(camPosLocal, camHeightLocal, cloudshader); diff --git a/indra/newview/llenvironment.cpp b/indra/newview/llenvironment.cpp index 8881d11802..69d3075928 100644 --- a/indra/newview/llenvironment.cpp +++ b/indra/newview/llenvironment.cpp @@ -1471,6 +1471,7 @@ LLEnvironment::DayInstance::ptr_t LLEnvironment::getSharedEnvironmentInstance() void LLEnvironment::updateEnvironment(LLSettingsBase::Seconds transition, bool forced) { + LL_PROFILE_ZONE_SCOPED; DayInstance::ptr_t pinstance = getSelectedEnvironmentInstance(); if ((mCurrentEnvironment != pinstance) || forced) @@ -1488,6 +1489,8 @@ void LLEnvironment::updateEnvironment(LLSettingsBase::Seconds transition, bool f { mCurrentEnvironment = pinstance; } + + updateSettingsUniforms(); } } @@ -1614,6 +1617,8 @@ void LLEnvironment::update(const LLViewerCamera * cam) stop_glerror(); + updateSettingsUniforms(); + // *TODO: potential optimization - this block may only need to be // executed some of the time. For example for water shaders only. { @@ -1648,10 +1653,16 @@ void LLEnvironment::updateCloudScroll() } // static -void LLEnvironment::updateGLVariablesForSettings(LLGLSLShader *shader, const LLSettingsBase::ptr_t &psetting) +void LLEnvironment::updateGLVariablesForSettings(LLShaderUniforms* uniforms, const LLSettingsBase::ptr_t &psetting) { LL_RECORD_BLOCK_TIME(FTM_SHADER_PARAM_UPDATE); + for (int i = 0; i < LLGLSLShader::SG_COUNT; ++i) + { + uniforms[i].clear(); + } + + LLShaderUniforms* shader = &uniforms[LLGLSLShader::SG_ANY]; //_WARNS("RIDER") << "----------------------------------------------------------------" << LL_ENDL; LLSettingsBase::parammapping_t params = psetting->getParameterMap(); for (auto &it: params) @@ -1694,7 +1705,7 @@ void LLEnvironment::updateGLVariablesForSettings(LLGLSLShader *shader, const LLS { LLVector4 vect4(value); //_WARNS("RIDER") << "pushing '" << (*it).first << "' as " << vect4 << LL_ENDL; - shader->uniform4fv(it.second.getShaderKey(), 1, vect4.mV); + shader->uniform4fv(it.second.getShaderKey(), vect4 ); break; } @@ -1707,17 +1718,30 @@ void LLEnvironment::updateGLVariablesForSettings(LLGLSLShader *shader, const LLS default: break; } - stop_glerror(); } //_WARNS("RIDER") << "----------------------------------------------------------------" << LL_ENDL; - psetting->applySpecial(shader); + psetting->applySpecial(uniforms); +} + +void LLEnvironment::updateShaderUniforms(LLGLSLShader* shader) +{ + LL_PROFILE_ZONE_SCOPED; + + // apply uniforms that should be applied to all shaders + mSkyUniforms[LLGLSLShader::SG_ANY].apply(shader); + mWaterUniforms[LLGLSLShader::SG_ANY].apply(shader); + + // apply uniforms specific to the given shader's shader group + auto group = shader->mShaderGroup; + mSkyUniforms[group].apply(shader); + mWaterUniforms[group].apply(shader); } -void LLEnvironment::updateShaderUniforms(LLGLSLShader *shader) +void LLEnvironment::updateSettingsUniforms() { - updateGLVariablesForSettings(shader, mCurrentEnvironment->getWater()); - updateGLVariablesForSettings(shader, mCurrentEnvironment->getSky()); + updateGLVariablesForSettings(mWaterUniforms, mCurrentEnvironment->getWater()); + updateGLVariablesForSettings(mSkyUniforms, mCurrentEnvironment->getSky()); } void LLEnvironment::recordEnvironment(S32 parcel_id, LLEnvironment::EnvironmentInfo::ptr_t envinfo, LLSettingsBase::Seconds transition) @@ -2618,6 +2642,7 @@ LLEnvironment::DayInstance::ptr_t LLEnvironment::DayInstance::clone() const bool LLEnvironment::DayInstance::applyTimeDelta(const LLSettingsBase::Seconds& delta) { + LL_PROFILE_ZONE_SCOPED; ptr_t keeper(shared_from_this()); // makes sure that this does not go away while it is being worked on. bool changed(false); diff --git a/indra/newview/llenvironment.h b/indra/newview/llenvironment.h index 7cbf2d25bb..3568fbcfd1 100644 --- a/indra/newview/llenvironment.h +++ b/indra/newview/llenvironment.h @@ -38,11 +38,12 @@ #include "llatmosphere.h" +#include "llglslshader.h" + #include //------------------------------------------------------------------------- class LLViewerCamera; -class LLGLSLShader; class LLParcel; //------------------------------------------------------------------------- @@ -50,8 +51,8 @@ class LLEnvironment : public LLSingleton { LLSINGLETON_C11(LLEnvironment); LOG_CLASS(LLEnvironment); - public: + static const F64Seconds TRANSITION_INSTANT; static const F64Seconds TRANSITION_FAST; static const F64Seconds TRANSITION_DEFAULT; @@ -131,9 +132,14 @@ public: void update(const LLViewerCamera * cam); - static void updateGLVariablesForSettings(LLGLSLShader *shader, const LLSettingsBase::ptr_t &psetting); + static void updateGLVariablesForSettings(LLShaderUniforms* uniforms, const LLSettingsBase::ptr_t &psetting); + + // apply current sky settings to given shader void updateShaderUniforms(LLGLSLShader *shader); + // prepare settings to be applied to shaders (call whenever settings are updated) + void updateSettingsUniforms(); + void setSelectedEnvironment(EnvSelection_t env, LLSettingsBase::Seconds transition = TRANSITION_DEFAULT, bool forced = false); EnvSelection_t getSelectedEnvironment() const { return mSelectedEnvironment; } @@ -234,6 +240,11 @@ public: void handleEnvironmentPush(LLSD &message); + //cached uniform values from LLSD values + LLShaderUniforms mWaterUniforms[LLGLSLShader::SG_COUNT]; + LLShaderUniforms mSkyUniforms[LLGLSLShader::SG_COUNT]; + // ======================================================================================= + class DayInstance: public std::enable_shared_from_this { public: @@ -288,6 +299,7 @@ public: LLSettingsDay::ptr_t mDayCycle; LLSettingsSky::ptr_t mSky; LLSettingsWater::ptr_t mWater; + S32 mSkyTrack; bool mInitialized; diff --git a/indra/newview/llsettingsvo.cpp b/indra/newview/llsettingsvo.cpp index 1e5b893cbc..6a88a8ef2c 100644 --- a/indra/newview/llsettingsvo.cpp +++ b/indra/newview/llsettingsvo.cpp @@ -637,6 +637,7 @@ LLSD LLSettingsVOSky::convertToLegacy(const LLSettingsSky::ptr_t &psky, bool isA //------------------------------------------------------------------------- void LLSettingsVOSky::updateSettings() { + LL_PROFILE_ZONE_SCOPED; LLSettingsSky::updateSettings(); LLVector3 sun_direction = getSunDirection(); LLVector3 moon_direction = getMoonDirection(); @@ -665,55 +666,55 @@ void LLSettingsVOSky::updateSettings() void LLSettingsVOSky::applySpecial(void *ptarget, bool force) { - LLGLSLShader *shader = (LLGLSLShader *)ptarget; - LLVector4 light_direction = LLEnvironment::instance().getClampedLightNorm(); - if (shader->mShaderGroup == LLGLSLShader::SG_DEFAULT) + LLShaderUniforms* shader = &((LLShaderUniforms*)ptarget)[LLGLSLShader::SG_DEFAULT]; { - shader->uniform4fv(LLViewerShaderMgr::LIGHTNORM, 1, light_direction.mV); - shader->uniform3fv(LLShaderMgr::WL_CAMPOSLOCAL, 1, LLViewerCamera::getInstance()->getOrigin().mV); + shader->uniform4fv(LLViewerShaderMgr::LIGHTNORM, light_direction); + shader->uniform3fv(LLShaderMgr::WL_CAMPOSLOCAL, LLViewerCamera::getInstance()->getOrigin()); } - else if (shader->mShaderGroup == LLGLSLShader::SG_SKY) + + shader = &((LLShaderUniforms*)ptarget)[LLGLSLShader::SG_SKY]; { - shader->uniform4fv(LLViewerShaderMgr::LIGHTNORM, 1, light_direction.mV); + shader->uniform4fv(LLViewerShaderMgr::LIGHTNORM, light_direction); - // Legacy? SETTING_CLOUD_SCROLL_RATE("cloud_scroll_rate") - LLVector4 vect_c_p_d1(mSettings[SETTING_CLOUD_POS_DENSITY1]); - LLVector4 cloud_scroll( LLEnvironment::instance().getCloudScrollDelta() ); + // Legacy? SETTING_CLOUD_SCROLL_RATE("cloud_scroll_rate") + LLVector4 vect_c_p_d1(mSettings[SETTING_CLOUD_POS_DENSITY1]); + LLVector4 cloud_scroll( LLEnvironment::instance().getCloudScrollDelta() ); - // SL-13084 EEP added support for custom cloud textures -- flip them horizontally to match the preview of Clouds > Cloud Scroll - // Keep in Sync! - // * indra\newview\llsettingsvo.cpp - // * indra\newview\app_settings\shaders\class2\windlight\cloudsV.glsl - // * indra\newview\app_settings\shaders\class1\deferred\cloudsV.glsl - cloud_scroll[0] = -cloud_scroll[0]; - vect_c_p_d1 += cloud_scroll; - shader->uniform4fv(LLShaderMgr::CLOUD_POS_DENSITY1, 1, vect_c_p_d1.mV); + // SL-13084 EEP added support for custom cloud textures -- flip them horizontally to match the preview of Clouds > Cloud Scroll + // Keep in Sync! + // * indra\newview\llsettingsvo.cpp + // * indra\newview\app_settings\shaders\class2\windlight\cloudsV.glsl + // * indra\newview\app_settings\shaders\class1\deferred\cloudsV.glsl + cloud_scroll[0] = -cloud_scroll[0]; + vect_c_p_d1 += cloud_scroll; + shader->uniform4fv(LLShaderMgr::CLOUD_POS_DENSITY1, vect_c_p_d1); - LLSettingsSky::ptr_t psky = LLEnvironment::instance().getCurrentSky(); + LLSettingsSky::ptr_t psky = LLEnvironment::instance().getCurrentSky(); - LLColor4 sunDiffuse = psky->getSunlightColor(); - LLColor4 moonDiffuse = psky->getMoonlightColor(); + LLVector4 sunDiffuse = LLVector4(psky->getSunlightColor().mV); + LLVector4 moonDiffuse = LLVector4(psky->getMoonlightColor().mV); - shader->uniform4fv(LLShaderMgr::SUNLIGHT_COLOR, 1, sunDiffuse.mV); - shader->uniform4fv(LLShaderMgr::MOONLIGHT_COLOR, 1, moonDiffuse.mV); + shader->uniform4fv(LLShaderMgr::SUNLIGHT_COLOR, sunDiffuse); + shader->uniform4fv(LLShaderMgr::MOONLIGHT_COLOR, moonDiffuse); - LLColor4 cloud_color(psky->getCloudColor(), 1.0); - shader->uniform4fv(LLShaderMgr::CLOUD_COLOR, 1, cloud_color.mV); + LLVector4 cloud_color(LLVector3(psky->getCloudColor().mV), 1.0); + shader->uniform4fv(LLShaderMgr::CLOUD_COLOR, cloud_color); } + shader = &((LLShaderUniforms*)ptarget)[LLGLSLShader::SG_ANY]; shader->uniform1f(LLShaderMgr::SCENE_LIGHT_STRENGTH, mSceneLightStrength); LLColor4 ambient(getTotalAmbient()); - shader->uniform4fv(LLShaderMgr::AMBIENT, 1, ambient.mV); + shader->uniform4fv(LLShaderMgr::AMBIENT, LLVector4(ambient.mV)); shader->uniform1i(LLShaderMgr::SUN_UP_FACTOR, getIsSunUp() ? 1 : 0); shader->uniform1f(LLShaderMgr::SUN_MOON_GLOW_FACTOR, getSunMoonGlowFactor()); shader->uniform1f(LLShaderMgr::DENSITY_MULTIPLIER, getDensityMultiplier()); shader->uniform1f(LLShaderMgr::DISTANCE_MULTIPLIER, getDistanceMultiplier()); - F32 g = getGamma(); + F32 g = getGamma(); F32 display_gamma = gSavedSettings.getF32("RenderDeferredDisplayGamma"); shader->uniform1f(LLShaderMgr::GAMMA, g); @@ -907,11 +908,11 @@ LLSD LLSettingsVOWater::convertToLegacy(const LLSettingsWater::ptr_t &pwater) //------------------------------------------------------------------------- void LLSettingsVOWater::applySpecial(void *ptarget, bool force) { - LLGLSLShader *shader = (LLGLSLShader *)ptarget; - LLEnvironment& env = LLEnvironment::instance(); - if (force || (shader->mShaderGroup == LLGLSLShader::SG_WATER)) + auto group = LLGLSLShader::SG_WATER; + LLShaderUniforms* shader = &((LLShaderUniforms*)ptarget)[group]; + { F32 water_height = env.getWaterHeight(); @@ -935,7 +936,7 @@ void LLSettingsVOWater::applySpecial(void *ptarget, bool force) LLVector4 waterPlane(enorm.v[0], enorm.v[1], enorm.v[2], -ep.dot(enorm)); - shader->uniform4fv(LLShaderMgr::WATER_WATERPLANE, 1, waterPlane.mV); + shader->uniform4fv(LLShaderMgr::WATER_WATERPLANE, waterPlane.mV); LLVector4 light_direction = env.getClampedLightNorm(); @@ -950,18 +951,19 @@ void LLSettingsVOWater::applySpecial(void *ptarget, bool force) shader->uniform1f(LLShaderMgr::WATER_FOGDENSITY, waterFogDensity); LLColor4 fog_color(env.getCurrentWater()->getWaterFogColor(), 0.0f); - shader->uniform4fv(LLShaderMgr::WATER_FOGCOLOR, 1, fog_color.mV); + shader->uniform4fv(LLShaderMgr::WATER_FOGCOLOR, fog_color.mV); F32 blend_factor = env.getCurrentWater()->getBlendFactor(); shader->uniform1f(LLShaderMgr::BLEND_FACTOR, blend_factor); // update to normal lightnorm, water shader itself will use rotated lightnorm as necessary - shader->uniform4fv(LLShaderMgr::LIGHTNORM, 1, light_direction.mV); + shader->uniform4fv(LLShaderMgr::LIGHTNORM, light_direction.mV); } } void LLSettingsVOWater::updateSettings() { + LL_PROFILE_ZONE_SCOPED; // base class clears dirty flag so as to not trigger recursive update LLSettingsBase::updateSettings(); diff --git a/indra/newview/llsettingsvo.h b/indra/newview/llsettingsvo.h index 65136ad2f5..caa3ac18d3 100644 --- a/indra/newview/llsettingsvo.h +++ b/indra/newview/llsettingsvo.h @@ -102,8 +102,6 @@ public: bool isAdvanced() const { return m_isAdvanced; } - virtual void updateShader(LLGLSLShader* shader) { applySpecial(shader, true); } - protected: LLSettingsVOSky(); @@ -136,8 +134,6 @@ public: static LLSD convertToLegacy(const ptr_t &); - virtual void updateShader(LLGLSLShader* shader) { applySpecial(shader, true); } - protected: LLSettingsVOWater(); diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index 5ef3819de4..c4976c4bbc 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -8427,8 +8427,6 @@ void LLPipeline::bindDeferredShader(LLGLSLShader& shader, LLRenderTarget* light_ LLEnvironment& environment = LLEnvironment::instance(); LLSettingsSky::ptr_t sky = environment.getCurrentSky(); - - static_cast(sky.get())->updateShader(&shader); } LLColor3 pow3f(LLColor3 v, F32 f) -- cgit v1.3 From e774bffb28a71730792931aeb1ed6a46d3cfe67b Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 21 Oct 2021 21:19:48 +0000 Subject: SL-16202 Fix for textures appearing black or flashing white due to optimization bugs. --- indra/llcommon/workqueue.cpp | 2 + indra/llrender/llimagegl.cpp | 241 ++++++++++++++++++-------------------- indra/llrender/llimagegl.h | 16 ++- indra/llrender/llrender.cpp | 14 ++- indra/llrender/llrender.h | 2 +- indra/newview/llglsandbox.cpp | 3 +- indra/newview/llviewertexture.cpp | 13 -- 7 files changed, 141 insertions(+), 150 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index ffc9a97dc0..b32357e832 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -54,6 +54,7 @@ void LL::WorkQueue::runUntilClose() bool LL::WorkQueue::runPending() { + LL_PROFILE_ZONE_SCOPED; for (Work work; mQueue.tryPop(work); ) { callWork(work); @@ -110,6 +111,7 @@ void LL::WorkQueue::callWork(const Queue::DataTuple& work) void LL::WorkQueue::callWork(const Work& work) { + LL_PROFILE_ZONE_SCOPED; try { work(); diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index aff29bd857..b5e1910242 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -683,7 +683,7 @@ void LLImageGL::setImage(const LLImageRaw* imageraw) } static LLTrace::BlockTimerStatHandle FTM_SET_IMAGE("setImage"); -BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips) +BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips, S32 usename) { LL_RECORD_BLOCK_TIME(FTM_SET_IMAGE); bool is_compressed = false; @@ -702,12 +702,11 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips) break; } - - if (mUseMipMaps) { //set has mip maps to true before binding image so tex parameters get set properly - gGL.getTexUnit(0)->unbind(mBindTarget); + gGL.getTexUnit(0)->unbind(mBindTarget); + mHasMipMaps = true; mTexOptionsDirty = true; setFilteringOption(LLTexUnit::TFO_ANISOTROPIC); @@ -717,7 +716,8 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips) mHasMipMaps = false; } - llverify(gGL.getTexUnit(0)->bind(this)); + gGL.getTexUnit(0)->bind(this, false, false, usename); + if (mUseMipMaps) { @@ -1211,7 +1211,7 @@ void LLImageGL::generateTextures(S32 numTextures, U32 *textures) } // static -void LLImageGL::deleteTextures(S32 numTextures, U32 *textures) +void LLImageGL::deleteTextures(S32 numTextures, const U32 *textures) { if (gGLManager.mInited) { @@ -1381,13 +1381,13 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S return FALSE; } - mGLTextureCreated = false ; llassert(gGLManager.mInited); stop_glerror(); if (!imageraw || imageraw->isBufferInvalid()) { LL_WARNS() << "Trying to create a texture from invalid image data" << LL_ENDL; + mGLTextureCreated = false; return FALSE; } @@ -1407,6 +1407,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S if (!setSize(w, h, imageraw->getComponents(), discard_level)) { LL_WARNS() << "Trying to create a texture with incorrect dimensions!" << LL_ENDL; + mGLTextureCreated = false; return FALSE; } @@ -1475,6 +1476,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S destroyGLTexture(); mCurrentDiscardLevel = discard_level; mLastBindTime = sLastFrameTime; + mGLTextureCreated = false; return TRUE ; } @@ -1486,104 +1488,123 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S static LLTrace::BlockTimerStatHandle FTM_CREATE_GL_TEXTURE3("createGLTexture3(data)"); BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_hasmips, S32 usename) { - LL_RECORD_BLOCK_TIME(FTM_CREATE_GL_TEXTURE3); - llassert(data_in); - stop_glerror(); - - if (discard_level < 0) - { - llassert(mCurrentDiscardLevel >= 0); - discard_level = mCurrentDiscardLevel; - } - discard_level = llclamp(discard_level, 0, (S32)mMaxDiscardLevel); + LL_RECORD_BLOCK_TIME(FTM_CREATE_GL_TEXTURE3); + llassert(data_in); + stop_glerror(); - if (mTexName != 0 && discard_level == mCurrentDiscardLevel) - { - // This will only be true if the size has not changed - return setImage(data_in, data_hasmips); - } - - U32 old_name = mTexName; -// S32 old_discard = mCurrentDiscardLevel; - - if (usename != 0) - { - mTexName = usename; - } - else - { - LLImageGL::generateTextures(1, &mTexName); - stop_glerror(); - { - llverify(gGL.getTexUnit(0)->bind(this)); - stop_glerror(); - glTexParameteri(LLTexUnit::getInternalType(mBindTarget), GL_TEXTURE_BASE_LEVEL, 0); - stop_glerror(); - glTexParameteri(LLTexUnit::getInternalType(mBindTarget), GL_TEXTURE_MAX_LEVEL, mMaxDiscardLevel-discard_level); - stop_glerror(); - } - } - if (!mTexName) - { - if (old_name) - { - sGlobalTextureMemory -= mTextureMemory; - LLImageGL::deleteTextures(1, &old_name); - disclaimMem(mTextureMemory); - stop_glerror(); - } + if (discard_level < 0) + { + llassert(mCurrentDiscardLevel >= 0); + discard_level = mCurrentDiscardLevel; + } + discard_level = llclamp(discard_level, 0, (S32)mMaxDiscardLevel); - LL_WARNS() << "LLImageGL::createGLTexture failed to make texture" << LL_ENDL; - return FALSE; - } + if (mTexName != 0 && discard_level == mCurrentDiscardLevel) + { + // This will only be true if the size has not changed + return setImage(data_in, data_hasmips); + } - if (mUseMipMaps) - { - mAutoGenMips = gGLManager.mHasMipMapGeneration; + GLuint old_texname = mTexName; + + if (usename != 0) + { + mNewTexName = usename; + } + else + { + LLImageGL::generateTextures(1, &mNewTexName); + { + gGL.getTexUnit(0)->bind(this, false, false, mNewTexName); + glTexParameteri(LLTexUnit::getInternalType(mBindTarget), GL_TEXTURE_BASE_LEVEL, 0); + glTexParameteri(LLTexUnit::getInternalType(mBindTarget), GL_TEXTURE_MAX_LEVEL, mMaxDiscardLevel - discard_level); + } + } + + if (mUseMipMaps) + { + mAutoGenMips = gGLManager.mHasMipMapGeneration; #if LL_DARWIN - // On the Mac GF2 and GF4MX drivers, auto mipmap generation doesn't work right with alpha-only textures. - if(gGLManager.mIsGF2or4MX && (mFormatInternal == GL_ALPHA8) && (mFormatPrimary == GL_ALPHA)) - { - mAutoGenMips = FALSE; - } + // On the Mac GF2 and GF4MX drivers, auto mipmap generation doesn't work right with alpha-only textures. + if (gGLManager.mIsGF2or4MX && (mFormatInternal == GL_ALPHA8) && (mFormatPrimary == GL_ALPHA)) + { + mAutoGenMips = FALSE; + } #endif - } + } - mCurrentDiscardLevel = discard_level; + mCurrentDiscardLevel = discard_level; - if (!setImage(data_in, data_hasmips)) - { - stop_glerror(); - return FALSE; - } + if (!setImage(data_in, data_hasmips, mNewTexName)) + { + return FALSE; + } - // Set texture options to our defaults. - gGL.getTexUnit(0)->setHasMipMaps(mHasMipMaps); - gGL.getTexUnit(0)->setTextureAddressMode(mAddressMode); - gGL.getTexUnit(0)->setTextureFilteringOption(mFilterOption); + // Set texture options to our defaults. + gGL.getTexUnit(0)->setHasMipMaps(mHasMipMaps); + gGL.getTexUnit(0)->setTextureAddressMode(mAddressMode); + gGL.getTexUnit(0)->setTextureFilteringOption(mFilterOption); - // things will break if we don't unbind after creation - gGL.getTexUnit(0)->unbind(mBindTarget); - stop_glerror(); + // things will break if we don't unbind after creation + gGL.getTexUnit(0)->unbind(mBindTarget); - if (old_name != 0) - { - sGlobalTextureMemory -= mTextureMemory; + if (old_texname != 0) + { + sGlobalTextureMemory -= mTextureMemory; + } - LLImageGL::deleteTextures(1, &old_name); + //if we're on the image loading thread, be sure to delete old_texname and update mTexName on the main thread + if (LLImageGLThread::sInstance != nullptr && + LLThread::currentID() == LLImageGLThread::sInstance->getID()) + { + { + LL_PROFILE_ZONE_NAMED("cglt - sync"); + if (gGLManager.mHasSync) + { + auto sync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); + glClientWaitSync(sync, 0, 0); + glDeleteSync(sync); + } + else + { + glFinish(); + } + } - stop_glerror(); - } + ref(); + LLImageGLThread::sInstance->postCallback([=]() + { + LL_PROFILE_ZONE_NAMED("cglt - delete callback"); + if (old_texname != 0) + { + LLImageGL::deleteTextures(1, &old_texname); + } + mTexName = mNewTexName; + mNewTexName = 0; + unref(); + }); + } + else + { + //not on background thread, immediately set mTexName + if (old_texname != 0) + { + LLImageGL::deleteTextures(1, &old_texname); + } + mTexName = mNewTexName; + mNewTexName = 0; + } + + disclaimMem(mTextureMemory); + mTextureMemory = (S32Bytes)getMipBytes(mCurrentDiscardLevel); + claimMem(mTextureMemory); + sGlobalTextureMemory += mTextureMemory; + mTexelsInGLTexture = getWidth() * getHeight(); - disclaimMem(mTextureMemory); - mTextureMemory = (S32Bytes)getMipBytes(discard_level); - claimMem(mTextureMemory); - sGlobalTextureMemory += mTextureMemory; - mTexelsInGLTexture = getWidth() * getHeight() ; + // mark this as bound at this point, so we don't throw it out immediately + mLastBindTime = sLastFrameTime; - // mark this as bound at this point, so we don't throw it out immediately - mLastBindTime = sLastFrameTime; - return TRUE; + return TRUE; } BOOL LLImageGL::readBackRaw(S32 discard_level, LLImageRaw* imageraw, bool compressed_ok) const @@ -2274,17 +2295,7 @@ bool LLImageGLThread::post(const std::function& func) { try { - if (mFunctionQueue.size() < mFunctionQueue.capacity()) - { - //NOTE: tryPushFront will return immediately if the lock is held - // desired behavior here is to push and return true unless the - // queue is full or closed - mFunctionQueue.pushFront(func); - } - else - { - return false; - } + mFunctionQueue.post(func); } catch (LLThreadSafeQueueInterrupt e) { @@ -2300,7 +2311,7 @@ bool LLImageGLThread::postCallback(const std::function& callback) { try { - mCallbackQueue.pushFront(callback); + mCallbackQueue.post(callback); } catch (LLThreadSafeQueueInterrupt e) { @@ -2315,34 +2326,14 @@ void LLImageGLThread::executeCallbacks() { LL_PROFILE_ZONE_SCOPED; //executed from main thread - std::function callback; - while (mCallbackQueue.tryPopBack(callback)) - { - LL_PROFILE_ZONE_NAMED("iglt - callback"); - callback(); - } + mCallbackQueue.runPending(); } void LLImageGLThread::run() { mWindow->makeContextCurrent(mContext); gGL.init(); - try - { - while (true) - { - LL_PROFILE_ZONE_SCOPED; - std::function curFunc = mFunctionQueue.popBack(); - { - LL_PROFILE_ZONE_NAMED("iglt - function") - curFunc(); - } - } - } - catch (LLThreadSafeQueueInterrupt e) - { - //queue is closed, fall out of run loop - } + mFunctionQueue.runUntilClose(); gGL.shutdown(); mWindow->destroySharedContext(mContext); } diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h index 8e9b483c2d..da626a1093 100644 --- a/indra/llrender/llimagegl.h +++ b/indra/llrender/llimagegl.h @@ -37,6 +37,8 @@ #include "llunits.h" #include "llthreadsafequeue.h" #include "llrender.h" +#include "workqueue.h" + class LLTextureAtlas ; class LLWindow; @@ -50,7 +52,7 @@ class LLImageGL : public LLRefCount, public LLTrace::MemTrackable public: // These 2 functions replace glGenTextures() and glDeleteTextures() static void generateTextures(S32 numTextures, U32 *textures); - static void deleteTextures(S32 numTextures, U32 *textures); + static void deleteTextures(S32 numTextures, const U32 *textures); static void deleteDeadTextures(); // Size calculation @@ -110,7 +112,7 @@ public: S32 category = sMaxCategories-1); BOOL createGLTexture(S32 discard_level, const U8* data, BOOL data_hasmips = FALSE, S32 usename = 0); void setImage(const LLImageRaw* imageraw); - BOOL setImage(const U8* data_in, BOOL data_hasmips = FALSE); + BOOL setImage(const U8* data_in, BOOL data_hasmips = FALSE, S32 usename = 0); BOOL setSubImage(const LLImageRaw* imageraw, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update = FALSE); BOOL setSubImage(const U8* datap, S32 data_width, S32 data_height, S32 x_pos, S32 y_pos, S32 width, S32 height, BOOL force_fast_update = FALSE); BOOL setSubImageFromFrameBuffer(S32 fb_x, S32 fb_y, S32 x_pos, S32 y_pos, S32 width, S32 height); @@ -210,8 +212,9 @@ private: bool mGLTextureCreated ; LLGLuint mTexName; + LLGLuint mNewTexName = 0; // tex name set by background thread to be applied in main thread U16 mWidth; - U16 mHeight; + U16 mHeight; S8 mCurrentDiscardLevel; S8 mDiscardLevelInAtlas; @@ -319,8 +322,11 @@ public: void run() override; - LLThreadSafeQueue> mFunctionQueue; - LLThreadSafeQueue> mCallbackQueue; + // Work Queue for background thread + LL::WorkQueue mFunctionQueue; + + // Work Queue for main thread (run from updateClass) + LL::WorkQueue mCallbackQueue; LLWindow* mWindow; void* mContext; diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 669a09d3ce..aad04beea2 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -236,6 +236,10 @@ void LLTexUnit::bindFast(LLTexture* texture) glActiveTextureARB(GL_TEXTURE0_ARB + mIndex); gGL.mCurrTextureUnitIndex = mIndex; mCurrTexture = gl_tex->getTexName(); + if (!mCurrTexture) + { + mCurrTexture = LLImageGL::sDefaultGLTexture->getTexName(); + } glBindTexture(sGLTextureType[gl_tex->getTarget()], mCurrTexture); mHasMipMaps = gl_tex->mHasMipMaps; } @@ -306,18 +310,20 @@ bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind) return true; } -bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind) +bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind, S32 usename) { stop_glerror(); if (mIndex < 0) return false; + U32 texname = usename ? usename : texture->getTexName(); + if(!texture) { LL_DEBUGS() << "NULL LLTexUnit::bind texture" << LL_ENDL; return false; } - if(!texture->getTexName()) + if(!texname) { if(LLImageGL::sDefaultGLTexture && LLImageGL::sDefaultGLTexture->getTexName()) { @@ -327,7 +333,7 @@ bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind) return false ; } - if ((mCurrTexture != texture->getTexName()) || forceBind) + if ((mCurrTexture != texname) || forceBind) { gGL.flush(); stop_glerror(); @@ -335,7 +341,7 @@ bool LLTexUnit::bind(LLImageGL* texture, bool for_rendering, bool forceBind) stop_glerror(); enable(texture->getTarget()); stop_glerror(); - mCurrTexture = texture->getTexName(); + mCurrTexture = texname; glBindTexture(sGLTextureType[texture->getTarget()], mCurrTexture); stop_glerror(); texture->updateBindStats(texture->mTextureMemory); diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h index 6e2647a16b..7f19a45410 100644 --- a/indra/llrender/llrender.h +++ b/indra/llrender/llrender.h @@ -158,7 +158,7 @@ public: // Binds the LLImageGL to this texture unit // (automatically enables the unit for the LLImageGL's texture type) - bool bind(LLImageGL* texture, bool for_rendering = false, bool forceBind = false); + bool bind(LLImageGL* texture, bool for_rendering = false, bool forceBind = false, S32 usename = 0); bool bind(LLTexture* texture, bool for_rendering = false, bool forceBind = false); // bind implementation for inner loops diff --git a/indra/newview/llglsandbox.cpp b/indra/newview/llglsandbox.cpp index 0f288e05ca..91f314c115 100644 --- a/indra/newview/llglsandbox.cpp +++ b/indra/newview/llglsandbox.cpp @@ -1090,7 +1090,7 @@ F32 gpu_benchmark() delete [] pixels; //make a dummy triangle to draw with - LLPointer buff = new LLVertexBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0, GL_STREAM_DRAW_ARB); + LLPointer buff = new LLVertexBuffer(LLVertexBuffer::MAP_VERTEX, GL_STREAM_DRAW_ARB); if (!buff->allocateBuffer(3, 0, true)) { @@ -1100,7 +1100,6 @@ F32 gpu_benchmark() } LLStrider v; - LLStrider tc; if (! buff->getVertexStrider(v)) { diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index 949e71a4c9..34847d8618 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -1637,19 +1637,6 @@ void LLViewerFetchedTexture::scheduleCreateTexture() { //actually create the texture on a background thread createTexture(); - { - LL_PROFILE_ZONE_NAMED("iglt - sync"); - if (gGLManager.mHasSync) - { - auto sync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); - glClientWaitSync(sync, 0, 0); - glDeleteSync(sync); - } - else - { - glFinish(); - } - } LLImageGLThread::sInstance->postCallback([this]() { //finalize on main thread -- cgit v1.3 From 11afa09ea3f56c0e20eb195ae1520a88602ceaca Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Fri, 22 Oct 2021 11:36:31 -0400 Subject: SL-16220: Add LL::ThreadPool class and a "General" instance. ThreadPool bundles a WorkQueue with the specified number of worker threads to service it. Each ThreadPool has a name that can be used to locate its WorkQueue. Each worker thread calls WorkQueue::runUntilClose(). ThreadPool listens on the "LLApp" LLEventPump for shutdown notification. On receiving that, it closes its WorkQueue and then join()s each of its worker threads for orderly shutdown. Add a settings.xml entry "ThreadPoolSizes", the first LLSD-valued settings entry to expect a map: pool name->size. The expectation is that usually code instantiating a particular ThreadPool will have a default size in mind, but it should check "ThreadPoolSizes" for a user override. Make idle_startup()'s STATE_SEED_CAP_GRANTED state instantiate a "General" ThreadPool. This is function-static for lazy initialization. Eliminate LLMainLoopRepeater, which is completely unreferenced. Any potential future use cases are better addressed by posting to the main loop's WorkQueue. Eliminate llappviewer.cpp's private LLDeferredTaskList class, which implemented LLAppViewer::addOnIdleCallback(). Make addOnIdleCallback() post work to the main loop's WorkQueue instead. --- indra/llcommon/CMakeLists.txt | 3 +- indra/llcommon/threadpool.cpp | 75 ++++++++++++++++++++++++++++ indra/llcommon/threadpool.h | 46 +++++++++++++++++ indra/llcommon/timing.cpp | 25 ---------- indra/llcommon/workqueue.cpp | 10 ++++ indra/llcommon/workqueue.h | 5 ++ indra/newview/CMakeLists.txt | 2 - indra/newview/app_settings/settings.xml | 14 ++++++ indra/newview/llappviewer.cpp | 47 +----------------- indra/newview/llmainlooprepeater.cpp | 88 --------------------------------- indra/newview/llmainlooprepeater.h | 64 ------------------------ indra/newview/llstartup.cpp | 18 +++++++ 12 files changed, 171 insertions(+), 226 deletions(-) create mode 100644 indra/llcommon/threadpool.cpp create mode 100644 indra/llcommon/threadpool.h delete mode 100644 indra/llcommon/timing.cpp delete mode 100644 indra/newview/llmainlooprepeater.cpp delete mode 100644 indra/newview/llmainlooprepeater.h (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index fda43dd24c..c374f1135c 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -121,8 +121,8 @@ set(llcommon_SOURCE_FILES lluriparser.cpp lluuid.cpp llworkerthread.cpp - timing.cpp u64.cpp + threadpool.cpp workqueue.cpp StackWalker.cpp ) @@ -258,6 +258,7 @@ set(llcommon_HEADER_FILES lockstatic.h stdtypes.h stringize.h + threadpool.h threadsafeschedule.h timer.h tuple.h diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp new file mode 100644 index 0000000000..aa7d4179a2 --- /dev/null +++ b/indra/llcommon/threadpool.cpp @@ -0,0 +1,75 @@ +/** + * @file threadpool.cpp + * @author Nat Goodspeed + * @date 2021-10-21 + * @brief Implementation for threadpool. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +// Precompiled header +#include "linden_common.h" +// associated header +#include "threadpool.h" +// STL headers +// std headers +// external library headers +// other Linden headers +#include "llerror.h" +#include "llevents.h" +#include "stringize.h" + +LL::ThreadPool::ThreadPool(const std::string& name, size_t threads): + mQueue(name), + mName("ThreadPool:" + name) +{ + for (size_t i = 0; i < threads; ++i) + { + std::string tname{ STRINGIZE(mName << ':' << (i+i) << '/' << threads) }; + mThreads.emplace_back(tname, [this, tname](){ run(tname); }); + } + // Listen on "LLApp", and when the app is shutting down, close the queue + // and join the workers. + LLEventPumps::instance().obtain("LLApp").listen( + mName, + [this](const LLSD& stat) + { + std::string status(stat["status"]); + if (status != "running") + { + // viewer is starting shutdown -- proclaim the end is nigh! + LL_DEBUGS("ThreadPool") << mName << " saw " << status << LL_ENDL; + close(); + } + return false; + }); +} + +LL::ThreadPool::~ThreadPool() +{ + close(); +} + +void LL::ThreadPool::close() +{ + if (! mQueue.isClosed()) + { + LL_DEBUGS("ThreadPool") << mName << " closing queue and joining threads" << LL_ENDL; + mQueue.close(); + for (auto& pair: mThreads) + { + LL_DEBUGS("ThreadPool") << mName << " waiting on thread " << pair.first << LL_ENDL; + pair.second.join(); + } + LL_DEBUGS("ThreadPool") << mName << " shutdown complete" << LL_ENDL; + } +} + +void LL::ThreadPool::run(const std::string& name) +{ + LL_DEBUGS("ThreadPool") << name << " starting" << LL_ENDL; + mQueue.runUntilClose(); + LL_DEBUGS("ThreadPool") << name << " stopping" << LL_ENDL; +} diff --git a/indra/llcommon/threadpool.h b/indra/llcommon/threadpool.h new file mode 100644 index 0000000000..8f3c8514b5 --- /dev/null +++ b/indra/llcommon/threadpool.h @@ -0,0 +1,46 @@ +/** + * @file threadpool.h + * @author Nat Goodspeed + * @date 2021-10-21 + * @brief ThreadPool configures a WorkQueue along with a pool of threads to + * service it. + * + * $LicenseInfo:firstyear=2021&license=viewerlgpl$ + * Copyright (c) 2021, Linden Research, Inc. + * $/LicenseInfo$ + */ + +#if ! defined(LL_THREADPOOL_H) +#define LL_THREADPOOL_H + +#include "workqueue.h" +#include +#include +#include // std::pair +#include + +namespace LL +{ + + class ThreadPool + { + public: + /** + * Pass ThreadPool a string name. This can be used to look up the + * relevant WorkQueue. + */ + ThreadPool(const std::string& name, size_t threads=1); + ~ThreadPool(); + void close(); + + private: + void run(const std::string& name); + + WorkQueue mQueue; + std::string mName; + std::vector> mThreads; + }; + +} // namespace LL + +#endif /* ! defined(LL_THREADPOOL_H) */ diff --git a/indra/llcommon/timing.cpp b/indra/llcommon/timing.cpp deleted file mode 100644 index c2dc695ef3..0000000000 --- a/indra/llcommon/timing.cpp +++ /dev/null @@ -1,25 +0,0 @@ -/** - * @file timing.cpp - * @brief This file will be deprecated in the future. - * - * $LicenseInfo:firstyear=2000&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index ffc9a97dc0..114aeea1f3 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -38,6 +38,16 @@ void LL::WorkQueue::close() mQueue.close(); } +bool LL::WorkQueue::isClosed() +{ + return mQueue.isClosed(); +} + +bool LL::WorkQueue::done() +{ + return mQueue.done(); +} + void LL::WorkQueue::runUntilClose() { try diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index b88aef989a..cfae2019dc 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -59,6 +59,11 @@ namespace LL */ void close(); + /// producer end: are we prevented from pushing any additional items? + bool isClosed(); + /// consumer end: are we done, is the queue entirely drained? + bool done(); + /*---------------------- fire and forget API -----------------------*/ /// fire-and-forget, but at a particular (future?) time diff --git a/indra/newview/CMakeLists.txt b/indra/newview/CMakeLists.txt index fbe75af712..bad36505d1 100644 --- a/indra/newview/CMakeLists.txt +++ b/indra/newview/CMakeLists.txt @@ -393,7 +393,6 @@ set(viewer_SOURCE_FILES llloginhandler.cpp lllogininstance.cpp llmachineid.cpp - llmainlooprepeater.cpp llmanip.cpp llmaniprotate.cpp llmanipscale.cpp @@ -1032,7 +1031,6 @@ set(viewer_HEADER_FILES llloginhandler.h lllogininstance.h llmachineid.h - llmainlooprepeater.h llmanip.h llmaniprotate.h llmanipscale.h diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index 802453d508..3c7fe174fd 100644 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -12663,6 +12663,20 @@ Value 50 + ThreadPoolSizes + + Comment + Map of size overrides for specific thread pools. + Persist + 1 + Type + LLSD + Value + + General + 4 + + ThrottleBandwidthKBPS Comment diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 7c932a3959..7c363eea5e 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -239,7 +239,6 @@ using namespace LL; // Include for security api initialization #include "llsecapi.h" #include "llmachineid.h" -#include "llmainlooprepeater.h" #include "llcleanup.h" #include "llcoproceduremanager.h" @@ -385,42 +384,6 @@ static std::string gLaunchFileOnQuit; // Used on Win32 for other apps to identify our window (eg, win_setup) const char* const VIEWER_WINDOW_CLASSNAME = "Second Life"; -//-- LLDeferredTaskList ------------------------------------------------------ - -/** - * A list of deferred tasks. - * - * We sometimes need to defer execution of some code until the viewer gets idle, - * e.g. removing an inventory item from within notifyObservers() may not work out. - * - * Tasks added to this list will be executed in the next LLAppViewer::idle() iteration. - * All tasks are executed only once. - */ -class LLDeferredTaskList: public LLSingleton -{ - LLSINGLETON_EMPTY_CTOR(LLDeferredTaskList); - LOG_CLASS(LLDeferredTaskList); - - friend class LLAppViewer; - typedef boost::signals2::signal signal_t; - - void addTask(const signal_t::slot_type& cb) - { - mSignal.connect(cb); - } - - void run() - { - if (!mSignal.empty()) - { - mSignal(); - mSignal.disconnect_all_slots(); - } - } - - signal_t mSignal; -}; - //---------------------------------------------------------------------------- // List of entries from strings.xml to always replace @@ -980,9 +943,6 @@ bool LLAppViewer::init() } LL_INFOS("InitInfo") << "Cache initialization is done." << LL_ENDL ; - // Initialize the repeater service. - LLMainLoopRepeater::instance().start(); - // // Initialize the window // @@ -2171,8 +2131,6 @@ bool LLAppViewer::cleanup() SUBSYSTEM_CLEANUP(LLProxy); LLCore::LLHttp::cleanup(); - LLMainLoopRepeater::instance().stop(); - ll_close_fail_log(); LLError::LLCallStacks::cleanup(); @@ -4437,7 +4395,7 @@ bool LLAppViewer::initCache() void LLAppViewer::addOnIdleCallback(const boost::function& cb) { - LLDeferredTaskList::instance().addTask(cb); + gMainloopWork.post(cb); } void LLAppViewer::loadKeyBindings() @@ -5211,9 +5169,6 @@ void LLAppViewer::idle() } } - // Execute deferred tasks. - LLDeferredTaskList::instance().run(); - // Service the WorkQueue we use for replies from worker threads. // Use function statics for the timeslice setting so we only have to fetch // and convert MainWorkTime once. diff --git a/indra/newview/llmainlooprepeater.cpp b/indra/newview/llmainlooprepeater.cpp deleted file mode 100644 index 6736e9a950..0000000000 --- a/indra/newview/llmainlooprepeater.cpp +++ /dev/null @@ -1,88 +0,0 @@ -/** - * @file llmachineid.cpp - * @brief retrieves unique machine ids - * - * $LicenseInfo:firstyear=2009&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#include "llviewerprecompiledheaders.h" -#include "llapr.h" -#include "llevents.h" -#include "llmainlooprepeater.h" - - - -// LLMainLoopRepeater -//----------------------------------------------------------------------------- - - -LLMainLoopRepeater::LLMainLoopRepeater(void): - mQueue(0) -{ - ; // No op. -} - - -void LLMainLoopRepeater::start(void) -{ - if(mQueue != 0) return; - - mQueue = new LLThreadSafeQueue(1024); - mMainLoopConnection = LLEventPumps::instance(). - obtain("mainloop").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMainLoop, this, _1)); - mRepeaterConnection = LLEventPumps::instance(). - obtain("mainlooprepeater").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMessage, this, _1)); -} - - -void LLMainLoopRepeater::stop(void) -{ - mMainLoopConnection.release(); - mRepeaterConnection.release(); - - delete mQueue; - mQueue = 0; -} - - -bool LLMainLoopRepeater::onMainLoop(LLSD const &) -{ - LLSD message; - while(mQueue->tryPopBack(message)) { - std::string pump = message["pump"].asString(); - if(pump.length() == 0 ) continue; // No pump. - LLEventPumps::instance().obtain(pump).post(message["payload"]); - } - return false; -} - - -bool LLMainLoopRepeater::onMessage(LLSD const & event) -{ - try { - mQueue->pushFront(event); - } catch(LLThreadSafeQueueError & e) { - LL_WARNS() << "could not repeat message (" << e.what() << ")" << - event.asString() << LL_ENDL; - } - return false; -} diff --git a/indra/newview/llmainlooprepeater.h b/indra/newview/llmainlooprepeater.h deleted file mode 100644 index 2ec3a74e4a..0000000000 --- a/indra/newview/llmainlooprepeater.h +++ /dev/null @@ -1,64 +0,0 @@ -/** - * @file llmainlooprepeater.h - * @brief a service for repeating messages on the main loop. - * - * $LicenseInfo:firstyear=2010&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#ifndef LL_LLMAINLOOPREPEATER_H -#define LL_LLMAINLOOPREPEATER_H - - -#include "llsd.h" -#include "llthreadsafequeue.h" - - -// -// A service which creates the pump 'mainlooprepeater' to which any thread can -// post a message that will be re-posted on the main loop. -// -// The posted message should contain two map elements: pump and payload. The -// pump value is a string naming the pump to which the message should be -// re-posted. The payload value is what will be posted to the designated pump. -// -class LLMainLoopRepeater: - public LLSingleton -{ - LLSINGLETON(LLMainLoopRepeater); -public: - // Start the repeater service. - void start(void); - - // Stop the repeater service. - void stop(void); - -private: - LLTempBoundListener mMainLoopConnection; - LLTempBoundListener mRepeaterConnection; - LLThreadSafeQueue * mQueue; - - bool onMainLoop(LLSD const &); - bool onMessage(LLSD const & event); -}; - - -#endif diff --git a/indra/newview/llstartup.cpp b/indra/newview/llstartup.cpp index 57c5074804..13e7fcb6e4 100644 --- a/indra/newview/llstartup.cpp +++ b/indra/newview/llstartup.cpp @@ -205,6 +205,9 @@ #include "llstacktrace.h" +#include "threadpool.h" + + #if LL_WINDOWS #include "lldxhardware.h" #endif @@ -301,6 +304,18 @@ void callback_cache_name(const LLUUID& id, const std::string& full_name, bool is // local classes // +void launchThreadPool() +{ + LLSD poolSizes{ gSavedSettings.getLLSD("ThreadPoolSizes") }; + LLSD sizeSpec{ poolSizes["General"] }; + LLSD::Integer size{ sizeSpec.isInteger()? sizeSpec.asInteger() : 3 }; + LL_DEBUGS("ThreadPool") << "Instantiating General pool with " + << size << " threads" << LL_ENDL; + // Use a function-static ThreadPool: static duration, but instantiated + // only on demand. + static LL::ThreadPool pool("General", size); +} + void update_texture_fetch() { LLAppViewer::getTextureCache()->update(1); // unpauses the texture cache thread @@ -1489,6 +1504,9 @@ bool idle_startup() gAgentCamera.resetCamera(); display_startup(); + // start up the ThreadPool we'll use for textures et al. + launchThreadPool(); + // Initialize global class data needed for surfaces (i.e. textures) LL_DEBUGS("AppInit") << "Initializing sky..." << LL_ENDL; // Initialize all of the viewer object classes for the first time (doing things like texture fetches. -- cgit v1.3 From 5553d614211998b5a10529f6b3ec68d2b25dc07a Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 22 Oct 2021 17:01:33 +0000 Subject: SL-16203 Fix for wonky handling of mouse deltas. --- indra/llcommon/llsingleton.h | 1 + indra/llwindow/llwindow.h | 3 + indra/llwindow/llwindowheadless.h | 3 + indra/llwindow/llwindowwin32.cpp | 530 ++++++++++++++----------------------- indra/llwindow/llwindowwin32.h | 13 +- indra/newview/lldrawpoolavatar.cpp | 10 +- indra/newview/llviewerwindow.cpp | 7 + 7 files changed, 230 insertions(+), 337 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 7c81d65a8b..2e43a3cbed 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -455,6 +455,7 @@ public: static DERIVED_TYPE* getInstance() { + LL_PROFILE_ZONE_SCOPED; // We know the viewer has LLSingleton dependency circularities. If you // feel strongly motivated to eliminate them, cheers and good luck. // (At that point we could consider a much simpler locking mechanism.) diff --git a/indra/llwindow/llwindow.h b/indra/llwindow/llwindow.h index 0100c3bf0a..1384ddfd82 100644 --- a/indra/llwindow/llwindow.h +++ b/indra/llwindow/llwindow.h @@ -91,6 +91,9 @@ public: virtual BOOL setCursorPosition(LLCoordWindow position) = 0; virtual BOOL getCursorPosition(LLCoordWindow *position) = 0; +#if LL_WINDOWS + virtual BOOL getCursorDelta(LLCoordCommon* delta) = 0; +#endif virtual void showCursor() = 0; virtual void hideCursor() = 0; virtual BOOL isCursorHidden() = 0; diff --git a/indra/llwindow/llwindowheadless.h b/indra/llwindow/llwindowheadless.h index a7ae28aa24..f8ba9bbed4 100644 --- a/indra/llwindow/llwindowheadless.h +++ b/indra/llwindow/llwindowheadless.h @@ -54,6 +54,9 @@ public: void destroySharedContext(void*) {} /*virtual*/ BOOL setCursorPosition(LLCoordWindow position) {return FALSE;}; /*virtual*/ BOOL getCursorPosition(LLCoordWindow *position) {return FALSE;}; +#if LL_WINDOWS + /*virtual*/ BOOL getCursorDelta(LLCoordCommon* delta) { return FALSE; } +#endif /*virtual*/ void showCursor() {}; /*virtual*/ void hideCursor() {}; /*virtual*/ void showCursorFromMouseMove() {}; diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index 12d4c6c30e..bf78bcba29 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -28,8 +28,6 @@ #if LL_WINDOWS && !LL_MESA_HEADLESS -#define LL_WINDOW_SINGLE_THREADED 0 - #include "llwindowwin32.h" // LLWindow library includes @@ -85,7 +83,7 @@ extern BOOL gDebugWindowProc; static std::thread::id sWindowThreadId; static std::thread::id sMainThreadId; -#if 1 || LL_WINDOW_SINGLE_THREADED +#if 1 // flip to zero to enable assertions for functions being called from wrong thread #define ASSERT_MAIN_THREAD() #define ASSERT_WINDOW_THREAD() #else @@ -482,9 +480,7 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, { sMainThreadId = LLThread::currentID(); mWindowThread = new LLWindowWin32Thread(this); -#if !LL_WINDOW_SINGLE_THREADED mWindowThread->start(); -#endif //MAINT-516 -- force a load of opengl32.dll just in case windows went sideways LoadLibrary(L"opengl32.dll"); @@ -492,7 +488,6 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, mIconResource = gIconResource; mOverrideAspectRatio = 0.f; mNativeAspectRatio = 0.f; - mMousePositionModified = FALSE; mInputProcessingPaused = FALSE; mPreeditor = NULL; mKeyCharCode = 0; @@ -814,6 +809,13 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, initCursors(); setCursor( UI_CURSOR_ARROW ); + mRawMouse.usUsagePage = 0x01; // HID_USAGE_PAGE_GENERIC + mRawMouse.usUsage = 0x02; // HID_USAGE_GENERIC_MOUSE + mRawMouse.dwFlags = 0; // adds mouse and also ignores legacy mouse messages + mRawMouse.hwndTarget = 0; + + RegisterRawInputDevices(&mRawMouse, 1, sizeof(mRawMouse)); + // Initialize (boot strap) the Language text input management, // based on the system's (or user's) default settings. allowLanguageTextInput(NULL, FALSE); @@ -1927,31 +1929,26 @@ BOOL LLWindowWin32::setCursorPosition(const LLCoordWindow position) { ASSERT_MAIN_THREAD(); - if (!mWindowHandle) - { - return FALSE; - } + if (!mWindowHandle) + { + return FALSE; + } - // Inform the application of the new mouse position (needed for per-frame - // hover/picking to function). - mCallbacks->handleMouseMove(this, position.convert(), (MASK)0); - - mMousePositionModified = TRUE; LLCoordScreen screen_pos(position.convert()); - - mWindowThread->post([=] + + // instantly set the cursor position from the app's point of view + mCursorPosition = position; + mLastCursorPosition = position; + + // Inform the application of the new mouse position (needed for per-frame + // hover/picking to function). + mCallbacks->handleMouseMove(this, position.convert(), (MASK)0); + + // actually set the cursor position on the window thread + mWindowThread->post([=]() { + // actually set the OS cursor position SetCursorPos(screen_pos.mX, screen_pos.mY); - // DEV-18951 VWR-8524 Camera moves wildly when alt-clicking. - // Because we have preemptively notified the application of the new - // mouse position via handleMouseMove() above, we need to clear out - // any stale mouse move events. RN/JC - MSG msg; - while (PeekMessage(&msg, NULL, WM_MOUSEMOVE, WM_MOUSEMOVE, PM_REMOVE)) - { - } - - mMousePositionModified = FALSE; }); return TRUE; @@ -1960,19 +1957,27 @@ BOOL LLWindowWin32::setCursorPosition(const LLCoordWindow position) BOOL LLWindowWin32::getCursorPosition(LLCoordWindow *position) { ASSERT_MAIN_THREAD(); - POINT cursor_point; - - if (!mWindowHandle - || !GetCursorPos(&cursor_point) - || !position) - { - return FALSE; - } + if (!position) + { + return FALSE; + } - *position = LLCoordScreen(cursor_point.x, cursor_point.y).convert(); + *position = mCursorPosition; return TRUE; } +BOOL LLWindowWin32::getCursorDelta(LLCoordCommon* delta) +{ + if (delta == nullptr) + { + return FALSE; + } + + *delta = mMouseFrameDelta; + + return TRUE; +} + void LLWindowWin32::hideCursor() { ASSERT_MAIN_THREAD(); @@ -2153,34 +2158,31 @@ void LLWindowWin32::gatherInput() LL_PROFILE_ZONE_SCOPED MSG msg; -#if LL_WINDOW_SINGLE_THREADED - int msg_count = 0; - - while ((msg_count < MAX_MESSAGE_PER_UPDATE)) { - LL_PROFILE_ZONE_NAMED("gi - loop"); - ++msg_count; - { - LL_PROFILE_ZONE_NAMED("gi - PeekMessage"); - if (!PeekMessage(&msg, NULL, 0, 0, PM_REMOVE)) - { - break; - } - } + LLMutexLock lock(&mRawMouseMutex); + mMouseFrameDelta = mRawMouseDelta; - { - LL_PROFILE_ZONE_NAMED("gi - translate"); - TranslateMessage(&msg); - } + mRawMouseDelta.mX = 0; + mRawMouseDelta.mY = 0; + } - { - LL_PROFILE_ZONE_NAMED("gi - dispatch"); - DispatchMessage(&msg); - } + if (mWindowThread->mFunctionQueue.size() > 0) + { + LL_PROFILE_ZONE_NAMED("gi - PostMessage"); + if (mWindowHandle) + { // post a nonsense user message to wake up the Window Thread in case any functions are pending + // and no windows events came through this frame + PostMessage(mWindowHandle, WM_USER + 0x0017, 0xB0B0, 0x1337); + } + } + + while (mWindowThread->mMessageQueue.tryPopBack(msg)) + { + LL_PROFILE_ZONE_NAMED("gi - message queue"); if (mInputProcessingPaused) { - break; + continue; } // For async host by name support. Really hacky. @@ -2190,45 +2192,35 @@ void LLWindowWin32::gatherInput() gAsyncMsgCallback(msg); } } -#else //multi-threaded window impl + { - if (mWindowThread->mFunctionQueue.size() > 0) + LL_PROFILE_ZONE_NAMED("gi - function queue"); + //process any pending functions + std::function curFunc; + while (mFunctionQueue.tryPopBack(curFunc)) { - LL_PROFILE_ZONE_NAMED("gi - PostMessage"); - if (mWindowHandle) - { // post a nonsense user message to wake up the Window Thread in case any functions are pending - // and no windows events came through this frame - PostMessage(mWindowHandle, WM_USER + 0x0017, 0xB0B0, 0x1337); - } + curFunc(); } - - while (mWindowThread->mMessageQueue.tryPopBack(msg)) - { - LL_PROFILE_ZONE_NAMED("gi - message queue"); - if (mInputProcessingPaused) - { - continue; - } + } - // For async host by name support. Really hacky. - if (gAsyncMsgCallback && (LL_WM_HOST_RESOLVED == msg.message)) - { - LL_PROFILE_ZONE_NAMED("gi - callback"); - gAsyncMsgCallback(msg); - } - } + // send one and only one mouse move event per frame BEFORE handling mouse button presses + if (mLastCursorPosition != mCursorPosition) + { + LL_PROFILE_ZONE_NAMED("gi - mouse move"); + mCallbacks->handleMouseMove(this, mCursorPosition.convert(), mMouseMask); } + + mLastCursorPosition = mCursorPosition; { - LL_PROFILE_ZONE_NAMED("gi - function queue"); - //process any pending functions + LL_PROFILE_ZONE_NAMED("gi - mouse queue"); + // handle mouse button presses AFTER updating mouse cursor position std::function curFunc; - while (mFunctionQueue.tryPopBack(curFunc)) + while (mMouseQueue.tryPopBack(curFunc)) { curFunc(); } } -#endif mInputProcessingPaused = FALSE; @@ -2238,11 +2230,7 @@ void LLWindowWin32::gatherInput() static LLTrace::BlockTimerStatHandle FTM_KEYHANDLER("Handle Keyboard"); static LLTrace::BlockTimerStatHandle FTM_MOUSEHANDLER("Handle Mouse"); -#if LL_WINDOW_SINGLE_THREADED -#define WINDOW_IMP_POST(x) x -#else #define WINDOW_IMP_POST(x) window_imp->post([=]() { x; }) -#endif LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_param, LPARAM l_param) { @@ -2278,10 +2266,6 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ // mouse is outside window. LLCoordWindow window_coord((S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param)); - // This doesn't work, as LOWORD returns unsigned short. - //LLCoordWindow window_coord(LOWORD(l_param), HIWORD(l_param)); - LLCoordGL gl_coord; - // pass along extended flag in mask MASK mask = (l_param >> 16 & KF_EXTENDED) ? MASK_EXTENDED : 0x0; BOOL eat_keystroke = TRUE; @@ -2665,35 +2649,19 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ LL_PROFILE_ZONE_NAMED("mwp - WM_LBUTTONDOWN"); { LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - window_imp->post([=]() + window_imp->postMouseButtonEvent([=]() { - auto glc = gl_coord; sHandleLeftMouseUp = true; - + if (LLWinImm::isAvailable() && window_imp->mPreeditor) { window_imp->interruptLanguageTextInput(); } - - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - glc = cursor_coord_window.convert(); - } - else - { - glc = window_coord.convert(); - } + MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, glc, mask); - window_imp->mCallbacks->handleMouseDown(window_imp, glc, mask); + auto gl_coord = window_imp->mCursorPosition.convert(); + window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); + window_imp->mCallbacks->handleMouseDown(window_imp, gl_coord, mask); }); return 0; @@ -2704,77 +2672,43 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ case WM_LBUTTONDBLCLK: { LL_PROFILE_ZONE_NAMED("mwp - WM_LBUTTONDBLCLK"); - //RN: ignore right button double clicks for now - //case WM_RBUTTONDBLCLK: - if (!sHandleDoubleClick) - { - sHandleDoubleClick = true; - return 0; - } - - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->post([=]() + window_imp->postMouseButtonEvent([=]() { - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - window_imp->mCallbacks->handleDoubleClick(window_imp, gl_coord, mask); + //RN: ignore right button double clicks for now + //case WM_RBUTTONDBLCLK: + if (!sHandleDoubleClick) + { + sHandleDoubleClick = true; + return; + } + MASK mask = gKeyboard->currentMask(TRUE); + + // generate move event to update mouse coordinates + window_imp->mCursorPosition = window_coord; + window_imp->mCallbacks->handleDoubleClick(window_imp, window_imp->mCursorPosition.convert(), mask); }); + return 0; } case WM_LBUTTONUP: { LL_PROFILE_ZONE_NAMED("mwp - WM_LBUTTONUP"); { - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - - if (!sHandleLeftMouseUp) - { - sHandleLeftMouseUp = true; - return 0; - } - sHandleDoubleClick = true; - window_imp->post([=]() + window_imp->postMouseButtonEvent([=]() { - auto glc = gl_coord; - - //if (gDebugClicks) - //{ - // LL_INFOS("Window") << "WndProc left button up" << LL_ENDL; - //} - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + if (!sHandleLeftMouseUp) { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - glc = cursor_coord_window.convert(); - } - else - { - glc = window_coord.convert(); + sHandleLeftMouseUp = true; + return; } + sHandleDoubleClick = true; + + MASK mask = gKeyboard->currentMask(TRUE); // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, glc, mask); - window_imp->mCallbacks->handleMouseUp(window_imp, glc, mask); + window_imp->mCursorPosition = window_coord; + window_imp->mCallbacks->handleMouseUp(window_imp, window_imp->mCursorPosition.convert(), mask); }); } return 0; @@ -2785,30 +2719,16 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ LL_PROFILE_ZONE_NAMED("mwp - WM_RBUTTONDOWN"); { LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - WINDOW_IMP_POST(window_imp->interruptLanguageTextInput()); - } - - // Because we move the cursor position in the llviewerapp, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates window_imp->post([=]() { + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + WINDOW_IMP_POST(window_imp->interruptLanguageTextInput()); + } + + MASK mask = gKeyboard->currentMask(TRUE); + // generate move event to update mouse coordinates + auto gl_coord = window_imp->mCursorPosition.convert(); window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); window_imp->mCallbacks->handleRightMouseDown(window_imp, gl_coord, mask); }); @@ -2822,28 +2742,11 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ LL_PROFILE_ZONE_NAMED("mwp - WM_RBUTTONUP"); { LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - // Because we move the cursor position in the app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleRightMouseUp(window_imp, gl_coord, mask)) - { - return 0; - } + window_imp->postMouseButtonEvent([=]() + { + MASK mask = gKeyboard->currentMask(TRUE); + window_imp->mCallbacks->handleRightMouseUp(window_imp, window_imp->mCursorPosition.convert(), mask); + }); } } break; @@ -2854,33 +2757,16 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ LL_PROFILE_ZONE_NAMED("mwp - WM_MBUTTONDOWN"); { LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - if (LLWinImm::isAvailable() && window_imp->mPreeditor) - { - window_imp->interruptLanguageTextInput(); - } + window_imp->postMouseButtonEvent([=]() + { + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + window_imp->interruptLanguageTextInput(); + } - // Because we move the cursor position in tllviewerhe app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleMiddleMouseDown(window_imp, gl_coord, mask)) - { - return 0; - } + MASK mask = gKeyboard->currentMask(TRUE); + window_imp->mCallbacks->handleMiddleMouseDown(window_imp, window_imp->mCursorPosition.convert(), mask); + }); } } break; @@ -2890,99 +2776,47 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ LL_PROFILE_ZONE_NAMED("mwp - WM_MBUTTONUP"); { LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - // Because we move the cursor position in the llviewer app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - if (window_imp->mCallbacks->handleMiddleMouseUp(window_imp, gl_coord, mask)) - { - return 0; - } + window_imp->postMouseButtonEvent([=]() + { + MASK mask = gKeyboard->currentMask(TRUE); + window_imp->mCallbacks->handleMiddleMouseUp(window_imp, window_imp->mCursorPosition.convert(), mask); + }); } } break; case WM_XBUTTONDOWN: { LL_PROFILE_ZONE_NAMED("mwp - WM_XBUTTONDOWN"); - { - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - S32 button = GET_XBUTTON_WPARAM(w_param); - if (LLWinImm::isAvailable() && window_imp->mPreeditor) + window_imp->postMouseButtonEvent([=]() { - window_imp->interruptLanguageTextInput(); - } + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + S32 button = GET_XBUTTON_WPARAM(w_param); + if (LLWinImm::isAvailable() && window_imp->mPreeditor) + { + window_imp->interruptLanguageTextInput(); + } - // Because we move the cursor position in tllviewerhe app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) - { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 - if (window_imp->mCallbacks->handleOtherMouseDown(window_imp, gl_coord, mask, button + 3)) - { - return 0; - } - } + MASK mask = gKeyboard->currentMask(TRUE); + // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 + window_imp->mCallbacks->handleOtherMouseDown(window_imp, window_imp->mCursorPosition.convert(), mask, button + 3); + }); + } break; case WM_XBUTTONUP: { LL_PROFILE_ZONE_NAMED("mwp - WM_XBUTTONUP"); - { - LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); - S32 button = GET_XBUTTON_WPARAM(w_param); - // Because we move the cursor position in the llviewer app, we need to query - // to find out where the cursor at the time the event is handled. - // If we don't do this, many clicks could get buffered up, and if the - // first click changes the cursor position, all subsequent clicks - // will occur at the wrong location. JC - if (window_imp->mMousePositionModified) + window_imp->postMouseButtonEvent([=]() { - LLCoordWindow cursor_coord_window; - window_imp->getCursorPosition(&cursor_coord_window); - gl_coord = cursor_coord_window.convert(); - } - else - { - gl_coord = window_coord.convert(); - } - MASK mask = gKeyboard->currentMask(TRUE); - // generate move event to update mouse coordinates - window_imp->mCallbacks->handleMouseMove(window_imp, gl_coord, mask); - // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 - if (window_imp->mCallbacks->handleOtherMouseUp(window_imp, gl_coord, mask, button + 3)) - { - return 0; - } - } + + LL_RECORD_BLOCK_TIME(FTM_MOUSEHANDLER); + + S32 button = GET_XBUTTON_WPARAM(w_param); + MASK mask = gKeyboard->currentMask(TRUE); + // Windows uses numbers 1 and 2 for buttons, remap to 4, 5 + window_imp->mCallbacks->handleOtherMouseUp(window_imp, window_imp->mCursorPosition.convert(), mask, button + 3); + }); } break; @@ -3022,7 +2856,8 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ // large deltas, like 480 or so. Thus we need to scroll more quickly. if (z_delta <= -WHEEL_DELTA || WHEEL_DELTA <= z_delta) { - window_imp->mCallbacks->handleScrollWheel(window_imp, -z_delta / WHEEL_DELTA); + short clicks = -z_delta / WHEEL_DELTA; + WINDOW_IMP_POST(window_imp->mCallbacks->handleScrollWheel(window_imp, clicks)); z_delta = 0; } return 0; @@ -3082,11 +2917,16 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ case WM_MOUSEMOVE: { LL_PROFILE_ZONE_NAMED("mwp - WM_MOUSEMOVE"); - if (!window_imp->mMousePositionModified) - { - MASK mask = gKeyboard->currentMask(TRUE); - WINDOW_IMP_POST(window_imp->mCallbacks->handleMouseMove(window_imp, window_coord.convert(), mask)); - } + // DO NOT use mouse event queue for move events to ensure cursor position is updated + // when button events are handled + WINDOW_IMP_POST( + { + LL_PROFILE_ZONE_NAMED("mwp - WM_MOUSEMOVE lambda"); + + MASK mask = gKeyboard->currentMask(TRUE); + window_imp->mMouseMask = mask; + window_imp->mCursorPosition = window_coord; + }); return 0; } @@ -3235,6 +3075,28 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ } break; + case WM_INPUT: + { + LL_PROFILE_ZONE_NAMED("MWP - WM_INPUT"); + + UINT dwSize = 0; + GetRawInputData((HRAWINPUT)l_param, RID_INPUT, NULL, &dwSize, sizeof(RAWINPUTHEADER)); + llassert(dwSize < 1024); + + U8 lpb[1024]; + + if (GetRawInputData((HRAWINPUT)l_param, RID_INPUT, (void*)lpb, &dwSize, sizeof(RAWINPUTHEADER)) == dwSize) + { + RAWINPUT* raw = (RAWINPUT*)lpb; + + if (raw->header.dwType == RIM_TYPEMOUSE) + { + LLMutexLock lock(&window_imp->mRawMouseMutex); + window_imp->mRawMouseDelta.mX += raw->data.mouse.lLastX; + window_imp->mRawMouseDelta.mY -= raw->data.mouse.lLastY; + } + } + } //list of messages we get often that we don't care to log about case WM_NCHITTEST: case WM_NCMOUSEMOVE: @@ -4740,18 +4602,16 @@ inline void LLWindowWin32Thread::run() void LLWindowWin32Thread::post(const std::function& func) { -#if LL_WINDOW_SINGLE_THREADED - func(); -#else mFunctionQueue.pushFront(func); -#endif } void LLWindowWin32::post(const std::function& func) { -#if LL_WINDOW_SINGLE_THREADED - func(); -#else mFunctionQueue.pushFront(func); -#endif } + +void LLWindowWin32::postMouseButtonEvent(const std::function& func) +{ + mMouseQueue.pushFront(func); +} + diff --git a/indra/llwindow/llwindowwin32.h b/indra/llwindow/llwindowwin32.h index 5f253b5df3..b44d458fc6 100644 --- a/indra/llwindow/llwindowwin32.h +++ b/indra/llwindow/llwindowwin32.h @@ -35,6 +35,7 @@ #include "lldragdropwin32.h" #include "llthread.h" #include "llthreadsafequeue.h" +#include "llmutex.h" // Hack for async host by name #define LL_WM_HOST_RESOLVED (WM_APP + 1) @@ -98,6 +99,7 @@ public: void destroySharedContext(void* context) override; /*virtual*/ BOOL setCursorPosition(LLCoordWindow position); /*virtual*/ BOOL getCursorPosition(LLCoordWindow *position); + /*virtual*/ BOOL getCursorDelta(LLCoordCommon* delta); /*virtual*/ void showCursor(); /*virtual*/ void hideCursor(); /*virtual*/ void showCursorFromMouseMove(); @@ -221,6 +223,14 @@ protected: F32 mNativeAspectRatio; HCURSOR mCursor[ UI_CURSOR_COUNT ]; // Array of all mouse cursors + LLCoordWindow mCursorPosition; // mouse cursor position, should only be mutated on main thread + LLMutex mRawMouseMutex; + RAWINPUTDEVICE mRawMouse; + LLCoordWindow mLastCursorPosition; // mouse cursor position from previous frame + LLCoordCommon mRawMouseDelta; // raw mouse delta according to window thread + LLCoordCommon mMouseFrameDelta; // how much the mouse moved between the last two calls to gatherInput + + MASK mMouseMask; static BOOL sIsClassRegistered; // has the window class been registered? @@ -231,7 +241,6 @@ protected: BOOL mCustomGammaSet; LPWSTR mIconResource; - BOOL mMousePositionModified; BOOL mInputProcessingPaused; // The following variables are for Language Text Input control. @@ -261,7 +270,9 @@ protected: LLWindowWin32Thread* mWindowThread = nullptr; LLThreadSafeQueue> mFunctionQueue; + LLThreadSafeQueue> mMouseQueue; void post(const std::function& func); + void postMouseButtonEvent(const std::function& func); friend class LLWindowManager; friend class LLWindowWin32Thread; diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index 8dd8c15b87..52d308f6bd 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -2279,7 +2279,15 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) if (normal_channel >= 0) { - gGL.getTexUnit(normal_channel)->bindFast(face->getTexture(LLRender::NORMAL_MAP)); + auto* texture = face->getTexture(LLRender::NORMAL_MAP); + if (texture) + { + gGL.getTexUnit(normal_channel)->bindFast(texture); + } + //else + //{ + // TODO handle missing normal map + //} } gGL.getTexUnit(sDiffuseChannel)->bindFast(face->getTexture(LLRender::DIFFUSE_MAP)); diff --git a/indra/newview/llviewerwindow.cpp b/indra/newview/llviewerwindow.cpp index 82ece85c1b..ce73037006 100644 --- a/indra/newview/llviewerwindow.cpp +++ b/indra/newview/llviewerwindow.cpp @@ -3781,8 +3781,15 @@ void LLViewerWindow::updateLayout() void LLViewerWindow::updateMouseDelta() { +#if LL_WINDOWS + LLCoordCommon delta; + mWindow->getCursorDelta(&delta); + S32 dx = delta.mX; + S32 dy = delta.mY; +#else S32 dx = lltrunc((F32) (mCurrentMousePoint.mX - mLastMousePoint.mX) * LLUI::getScaleFactor().mV[VX]); S32 dy = lltrunc((F32) (mCurrentMousePoint.mY - mLastMousePoint.mY) * LLUI::getScaleFactor().mV[VY]); +#endif //RN: fix for asynchronous notification of mouse leaving window not working LLCoordWindow mouse_pos; -- cgit v1.3 From d2763897f22e3d7789f97fe68000662ecd4a3548 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Fri, 22 Oct 2021 21:51:44 -0400 Subject: SL-16220: Fix thread name expression. --- indra/llcommon/threadpool.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp index aa7d4179a2..1899f9a20a 100644 --- a/indra/llcommon/threadpool.cpp +++ b/indra/llcommon/threadpool.cpp @@ -27,7 +27,7 @@ LL::ThreadPool::ThreadPool(const std::string& name, size_t threads): { for (size_t i = 0; i < threads; ++i) { - std::string tname{ STRINGIZE(mName << ':' << (i+i) << '/' << threads) }; + std::string tname{ STRINGIZE(mName << ':' << (i+1) << '/' << threads) }; mThreads.emplace_back(tname, [this, tname](){ run(tname); }); } // Listen on "LLApp", and when the app is shutting down, close the queue -- cgit v1.3 From e7b8c27741201528bf78f95c96ba820833923dab Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Mon, 25 Oct 2021 15:55:49 -0400 Subject: SL-16220: Specialize WorkQueue for callable with void return. Add a test exercising this feature. --- indra/llcommon/tests/threadsafeschedule_test.cpp | 4 +- indra/llcommon/tests/workqueue_test.cpp | 23 +++- indra/llcommon/workqueue.h | 167 +++++++++++++++-------- 3 files changed, 134 insertions(+), 60 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/tests/threadsafeschedule_test.cpp b/indra/llcommon/tests/threadsafeschedule_test.cpp index af67b9f492..c421cc7b1c 100644 --- a/indra/llcommon/tests/threadsafeschedule_test.cpp +++ b/indra/llcommon/tests/threadsafeschedule_test.cpp @@ -46,11 +46,11 @@ namespace tut // the real time required for each push() call. Explicitly increment // the timestamp for each one -- but since we're passing explicit // timestamps, make the queue reorder them. - queue.push(Queue::TimeTuple(Queue::Clock::now() + 20ms, "ghi")); + queue.push(Queue::TimeTuple(Queue::Clock::now() + 200ms, "ghi")); // Given the various push() overloads, you have to match the type // exactly: conversions are ambiguous. queue.push("abc"s); - queue.push(Queue::Clock::now() + 10ms, "def"); + queue.push(Queue::Clock::now() + 100ms, "def"); queue.close(); auto entry = queue.pop(); ensure_equals("failed to pop first", std::get<0>(entry), "abc"s); diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp index d5405400fd..b69df49d33 100644 --- a/indra/llcommon/tests/workqueue_test.cpp +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -138,7 +138,8 @@ namespace tut [](){ return 17; }, // Note that a postTo() *callback* can safely bind a reference to // a variable on the invoking thread, because the callback is run - // on the invoking thread. + // on the invoking thread. (Of course the bound variable must + // survive until the callback is called.) [&result](int i){ result = i; }); // this should post the callback to main qptr->runOne(); @@ -156,4 +157,24 @@ namespace tut main.runPending(); ensure_equals("failed to run string callback", alpha, "abc"); } + + template<> template<> + void object::test<5>() + { + set_test_name("postTo with void return"); + WorkQueue main("main"); + auto qptr = WorkQueue::getInstance("queue"); + std::string observe; + main.postTo( + qptr, + // The ONLY reason we can get away with binding a reference to + // 'observe' in our work callable is because we're directly + // calling qptr->runOne() on this same thread. It would be a + // mistake to do that if some other thread were servicing 'queue'. + [&observe](){ observe = "queue"; }, + [&observe](){ observe.append(";main"); }); + qptr->runOne(); + main.runOne(); + ensure_equals("failed to run both lambdas", observe, "queue;main"); + } } // namespace tut diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index cfae2019dc..deef3c8e84 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -115,62 +115,7 @@ namespace LL // code. template bool postTo(WorkQueue::weak_t target, - const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) - { - // We're being asked to post to the WorkQueue at target. - // target is a weak_ptr: have to lock it to check it. - auto tptr = target.lock(); - if (! tptr) - // can't post() if the target WorkQueue has been destroyed - return false; - - // Here we believe target WorkQueue still exists. Post to it a - // lambda that packages our callable, our callback and a weak_ptr - // to this originating WorkQueue. - tptr->post( - time, - [reply = super::getWeak(), - callable = std::move(callable), - callback = std::move(callback)] - () - { - // Call the callable in any case -- but to minimize - // copying the result, immediately bind it into a reply - // lambda. The reply lambda also binds the original - // callback, so that when we, the originating WorkQueue, - // finally receive and process the reply lambda, we'll - // call the bound callback with the bound result -- on the - // same thread that originally called postTo(). - auto rlambda = - [result = callable(), - callback = std::move(callback)] - () - { callback(std::move(result)); }; - // Check if this originating WorkQueue still exists. - // Remember, the outer lambda is now running on a thread - // servicing the target WorkQueue, and real time has - // elapsed since postTo()'s tptr->post() call. - // reply is a weak_ptr: have to lock it to check it. - auto rptr = reply.lock(); - if (rptr) - { - // Only post reply lambda if the originating WorkQueue - // still exists. If not -- who would we tell? Log it? - try - { - rptr->post(std::move(rlambda)); - } - catch (const Closed&) - { - // Originating WorkQueue might still exist, but - // might be Closed. Same thing: just discard the - // callback. - } - } - }); - // looks like we were able to post() - return true; - } + const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback); /** * Post work to another WorkQueue, requesting a specific callback to @@ -183,7 +128,8 @@ namespace LL bool postTo(WorkQueue::weak_t target, CALLABLE&& callable, FOLLOWUP&& callback) { - return postTo(target, TimePoint::clock::now(), std::move(callable), std::move(callback)); + return postTo(target, TimePoint::clock::now(), + std::move(callable), std::move(callback)); } /*--------------------------- worker API ---------------------------*/ @@ -231,6 +177,17 @@ namespace LL bool runUntil(const TimePoint& until); private: + template + static auto makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback); + + /// general case: arbitrary C++ return type + template + struct MakeReplyLambda; + + /// specialize for CALLABLE returning void + template + struct MakeReplyLambda; + static void error(const std::string& msg); static std::string makeName(const std::string& name); void callWork(const Queue::DataTuple& work); @@ -329,6 +286,102 @@ namespace LL getWeak(), TimePoint::clock::now(), interval, std::move(callable))); } + template + struct WorkQueue::MakeReplyLambda + { + auto operator()(CALLABLE&& callable, FOLLOWUP&& callback) + { + // Call the callable in any case -- but to minimize + // copying the result, immediately bind it into the reply + // lambda. The reply lambda also binds the original + // callback, so that when we, the originating WorkQueue, + // finally receive and process the reply lambda, we'll + // call the bound callback with the bound result -- on the + // same thread that originally called postTo(). + return + [result = std::forward(callable)(), + callback = std::move(callback)] + () + { callback(std::move(result)); }; + } + }; + + /// specialize for CALLABLE returning void + template + struct WorkQueue::MakeReplyLambda + { + auto operator()(CALLABLE&& callable, FOLLOWUP&& callback) + { + // Call the callable, which produces no result. + std::forward(callable)(); + // This reply lambda binds the original callback, so + // that when we, the originating WorkQueue, finally + // receive and process the reply lambda, we'll call + // the bound callback -- on the same thread that + // originally called postTo(). + return [callback = std::move(callback)](){ callback(); }; + } + }; + + template + auto WorkQueue::makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback) + { + return MakeReplyLambda(callable)())>() + (std::move(callable), std::move(callback)); + } + + template + bool WorkQueue::postTo(WorkQueue::weak_t target, + const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) + { + // We're being asked to post to the WorkQueue at target. + // target is a weak_ptr: have to lock it to check it. + auto tptr = target.lock(); + if (! tptr) + // can't post() if the target WorkQueue has been destroyed + return false; + + // Here we believe target WorkQueue still exists. Post to it a + // lambda that packages our callable, our callback and a weak_ptr + // to this originating WorkQueue. + tptr->post( + time, + [reply = super::getWeak(), + callable = std::move(callable), + callback = std::move(callback)] + () + { + // Make a reply lambda to repost to THIS WorkQueue. + // Delegate to makeReplyLambda() so we can partially + // specialize on void return. + auto rlambda = makeReplyLambda(std::move(callable), std::move(callback)); + // Check if this originating WorkQueue still exists. + // Remember, the outer lambda is now running on a thread + // servicing the target WorkQueue, and real time has + // elapsed since postTo()'s tptr->post() call. + // reply is a weak_ptr: have to lock it to check it. + auto rptr = reply.lock(); + if (rptr) + { + // Only post reply lambda if the originating WorkQueue + // still exists. If not -- who would we tell? Log it? + try + { + rptr->post(std::move(rlambda)); + } + catch (const Closed&) + { + // Originating WorkQueue might still exist, but + // might be Closed. Same thing: just discard the + // callback. + } + } + }); + // looks like we were able to post() + return true; + } + } // namespace LL #endif /* ! defined(LL_WORKQUEUE_H) */ -- cgit v1.3 From 647d0224d321c706ba5936905db4265becde9d8e Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 25 Oct 2021 21:11:03 +0000 Subject: SL-16243 Add Tracy timers to global new/delete overrides. --- indra/llcommon/llcommon.cpp | 24 ++++++++++++++++++++++-- indra/llcommon/llprofiler.h | 6 ++++-- indra/newview/llappviewerwin32.cpp | 3 ++- 3 files changed, 28 insertions(+), 5 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index 5d4a623bf6..abc6af2cfc 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -33,11 +33,22 @@ #include "lltracethreadrecorder.h" #include "llcleanup.h" +thread_local bool gProfilerEnabled = false; + #if (TRACY_ENABLE) // Override new/delete for tracy memory profiling void *operator new(size_t size) { - auto ptr = (malloc) (size); + void* ptr; + if (gProfilerEnabled) + { + LL_PROFILE_ZONE_SCOPED; + ptr = (malloc)(size); + } + else + { + ptr = (malloc)(size); + } if (!ptr) { throw std::bad_alloc(); @@ -49,7 +60,16 @@ void *operator new(size_t size) void operator delete(void *ptr) noexcept { TracyFree(ptr); - (free)(ptr); + + if (gProfilerEnabled) + { + LL_PROFILE_ZONE_SCOPED; + (free)(ptr); + } + else + { + (free)(ptr); + } } // C-style malloc/free can't be so easily overridden, so we define tracy versions and use diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index 49510df913..e36f693dd3 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -36,6 +36,8 @@ #define LL_PROFILER_CONFIGURATION LL_PROFILER_CONFIG_FAST_TIMER #endif +extern thread_local bool gProfilerEnabled; + #if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE) #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define TRACY_ENABLE 1 @@ -52,7 +54,7 @@ #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY #define LL_PROFILER_FRAME_END FrameMark - #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ); gProfilerEnabled = true; #define LL_RECORD_BLOCK_TIME(name) ZoneScoped // Want descriptive names; was: ZoneNamedN( ___tracy_scoped_zone, #name, true ); #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, name, true ); #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB @@ -82,7 +84,7 @@ #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define LL_PROFILER_FRAME_END FrameMark - #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ) + #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ); gProfilerEnabled = true; #define LL_RECORD_BLOCK_TIME(name) ZoneScoped const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB diff --git a/indra/newview/llappviewerwin32.cpp b/indra/newview/llappviewerwin32.cpp index 758bd73cb0..9074e6a6cf 100644 --- a/indra/newview/llappviewerwin32.cpp +++ b/indra/newview/llappviewerwin32.cpp @@ -325,7 +325,8 @@ int APIENTRY WINMAIN(HINSTANCE hInstance, { // Call Tracy first thing to have it allocate memory // https://github.com/wolfpld/tracy/issues/196 - LL_PROFILER_FRAME_END + LL_PROFILER_FRAME_END; + LL_PROFILER_SET_THREAD_NAME("App"); const S32 MAX_HEAPS = 255; DWORD heap_enable_lfh_error[MAX_HEAPS]; -- cgit v1.3 From 023d39963e850356e1af6eec7f857e2534ce8d38 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Mon, 25 Oct 2021 17:31:27 -0400 Subject: SL-16220: WorkQueue::runOn() methods submit work, wait for result. The idea is that you can call runOn(target, callable) from a (non-default) coroutine and block that coroutine until the result becomes available. As a safety check, we forbid calling runOn() from a thread's default coroutine, assuming that a given thread's default coroutine is the one servicing the relevant WorkQueue. --- indra/llcommon/workqueue.cpp | 15 +++++ indra/llcommon/workqueue.h | 150 +++++++++++++++++++++++++++++++++++++++---- 2 files changed, 154 insertions(+), 11 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index 114aeea1f3..f7ffc8233c 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -26,6 +26,11 @@ using Mutex = LLCoros::Mutex; using Lock = LLCoros::LockType; +struct NotOnDftCoro: public LLException +{ + NotOnDftCoro(const std::string& what): LLException(what) {} +}; + LL::WorkQueue::WorkQueue(const std::string& name): super(makeName(name)) { @@ -136,3 +141,13 @@ void LL::WorkQueue::error(const std::string& msg) { LL_ERRS("WorkQueue") << msg << LL_ENDL; } + +void LL::WorkQueue::checkCoroutine(const std::string& method) +{ + // By convention, the default coroutine on each thread has an empty name + // string. See also LLCoros::logname(). + if (LLCoros::getName().empty()) + { + LLTHROW(NotOnDftCoro("Do not call " + method + " from a thread's default coroutine")); + } +} diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index deef3c8e84..b17c666172 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -12,11 +12,18 @@ #if ! defined(LL_WORKQUEUE_H) #define LL_WORKQUEUE_H +#include "llcoros.h" #include "llinstancetracker.h" #include "threadsafeschedule.h" #include #include // std::function -#include +#if __cplusplus >= 201703 +#include +namespace stdopt = std; +#else +#include +namespace stdopt = boost; +#endif #include #include // std::pair #include @@ -44,6 +51,8 @@ namespace LL using TimePoint = Queue::TimePoint; using TimedWork = Queue::TimeTuple; using Closed = Queue::Closed; + template + using optional = stdopt::optional; /** * You may omit the WorkQueue name, in which case a unique name is @@ -114,7 +123,7 @@ namespace LL // Studio compile errors that seem utterly unrelated to this source // code. template - bool postTo(WorkQueue::weak_t target, + bool postTo(weak_t target, const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback); /** @@ -125,13 +134,62 @@ namespace LL * inaccessible. */ template - bool postTo(WorkQueue::weak_t target, - CALLABLE&& callable, FOLLOWUP&& callback) + bool postTo(weak_t target, CALLABLE&& callable, FOLLOWUP&& callback) { return postTo(target, TimePoint::clock::now(), std::move(callable), std::move(callback)); } + /** + * Post work to another WorkQueue to be run at a specified time, + * blocking the calling coroutine until then, returning the result to + * caller on completion. + * + * REQUIRED: + * + * * The calling thread is the thread servicing 'this' WorkQueue. + * * The calling coroutine is not the @em coroutine servicing this + * WorkQueue. We block the calling coroutine until the result is + * available. If this same coroutine is responsible for checking the + * local WorkQueue, the result will never be dequeued. In practice, + * to try to prevent mistakes, we forbid calling runOn() from a + * thread's default coroutine. + * + * Returns result if able to post, empty optional if the other + * WorkQueue is inaccessible. + * + * If the passed callable has void return, runOn() returns bool true + * if able to post, false if the other WorkQueue is inaccessible. + */ + template + auto runOn(weak_t target, const TimePoint& time, CALLABLE&& callable); + + /** + * Post work to another WorkQueue, blocking the calling coroutine + * until then, returning the result to caller on completion. + * + * REQUIRED: + * + * * The calling thread is the thread servicing 'this' WorkQueue. + * * The calling coroutine is not the @em coroutine servicing this + * WorkQueue. We block the calling coroutine until the result is + * available. If this same coroutine is responsible for checking the + * local WorkQueue, the result will never be dequeued. In practice, + * to try to prevent mistakes, we forbid calling runOn() from a + * thread's default coroutine. + * + * Returns result if able to post, empty optional if the other + * WorkQueue is inaccessible. + * + * If the passed callable has void return, runOn() returns bool true + * if able to post, false if the other WorkQueue is inaccessible. + */ + template + auto runOn(weak_t target, CALLABLE&& callable) + { + return runOn(target, TimePoint::clock::now(), std::move(callable)); + } + /*--------------------------- worker API ---------------------------*/ /** @@ -179,15 +237,21 @@ namespace LL private: template static auto makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback); - /// general case: arbitrary C++ return type template struct MakeReplyLambda; - /// specialize for CALLABLE returning void template struct MakeReplyLambda; + /// general case: arbitrary C++ return type + template + struct RunOn; + /// specialize for CALLABLE returning void + template + struct RunOn; + + static void checkCoroutine(const std::string& method); static void error(const std::string& msg); static std::string makeName(const std::string& name); void callWork(const Queue::DataTuple& work); @@ -209,8 +273,8 @@ namespace LL { public: // bind the desired data - BackJack(WorkQueue::weak_t target, - const WorkQueue::TimePoint& start, + BackJack(weak_t target, + const TimePoint& start, const std::chrono::duration& interval, CALLABLE&& callable): mTarget(target), @@ -257,8 +321,8 @@ namespace LL } private: - WorkQueue::weak_t mTarget; - WorkQueue::TimePoint mStart; + weak_t mTarget; + TimePoint mStart; std::chrono::duration mInterval; CALLABLE mCallable; }; @@ -286,6 +350,7 @@ namespace LL getWeak(), TimePoint::clock::now(), interval, std::move(callable))); } + /// general case: arbitrary C++ return type template struct WorkQueue::MakeReplyLambda { @@ -332,7 +397,7 @@ namespace LL } template - bool WorkQueue::postTo(WorkQueue::weak_t target, + bool WorkQueue::postTo(weak_t target, const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) { // We're being asked to post to the WorkQueue at target. @@ -382,6 +447,69 @@ namespace LL return true; } + /// general case: arbitrary C++ return type + template + struct WorkQueue::RunOn + { + optional operator()(WorkQueue* self, weak_t target, + const TimePoint& time, CALLABLE&& callable) + { + LLCoros::Promise promise; + if (! self->postTo( + target, + time, + std::forward(callable), + // We dare to bind a reference to Promise because it's + // specifically intended for cross-thread synchronization. + [&promise] + (RETURNTYPE&& result) + { + promise.set_value(std::forward(result)); + })) + { + // we couldn't even postTo(): return empty optional + return {}; + } + // we were able to post + auto future{ LLCoros::getFuture(promise) }; + return { future.get(); } + } + }; + + /// specialize for CALLABLE returning void + template + struct WorkQueue::RunOn + { + bool operator()(WorkQueue* self, weak_t target, + const TimePoint& time, CALLABLE&& callable) + { + LLCoros::Promise promise; + if (! self->postTo( + target, + time, + std::forward(callable), + // &promise is designed for cross-thread access + [&promise](){ promise.set_value(); })) + { + // we couldn't postTo() + return false; + } + // we were able to post + auto future{ LLCoros::getFuture(promise) }; + // block until set_value() + future.get(); + return true; + } + }; + + template + auto WorkQueue::runOn(weak_t target, const TimePoint& time, CALLABLE&& callable) + { + checkCoroutine("runOn()"); + return RunOn(callable)())>() + (this, target, time, std::forward(callable)); + } + } // namespace LL #endif /* ! defined(LL_WORKQUEUE_H) */ -- cgit v1.3 From 4e8cd9437bed90b3468b1bf12f545de16faefb67 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 26 Oct 2021 14:07:00 +0000 Subject: SL-16193 Fix for mesh selection outline not rendering correctly (and broken physics shapes display). --- indra/llcommon/llcommon.cpp | 5 +- indra/llrender/llgl.cpp | 2 + indra/llrender/llvertexbuffer.cpp | 185 ++++++++++++++--------------------- indra/llrender/llvertexbuffer.h | 3 +- indra/newview/llface.h | 4 +- indra/newview/llmodelpreview.cpp | 2 +- indra/newview/llspatialpartition.cpp | 26 +++-- 7 files changed, 95 insertions(+), 132 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index abc6af2cfc..04872564bf 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -44,6 +44,7 @@ void *operator new(size_t size) { LL_PROFILE_ZONE_SCOPED; ptr = (malloc)(size); + TracyAlloc(ptr, size); } else { @@ -53,17 +54,15 @@ void *operator new(size_t size) { throw std::bad_alloc(); } - TracyAlloc(ptr, size); return ptr; } void operator delete(void *ptr) noexcept { - TracyFree(ptr); - if (gProfilerEnabled) { LL_PROFILE_ZONE_SCOPED; + TracyFree(ptr); (free)(ptr); } else diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp index a269549c49..b66521132c 100644 --- a/indra/llrender/llgl.cpp +++ b/indra/llrender/llgl.cpp @@ -2117,6 +2117,7 @@ void LLGLState::checkClientArrays(const std::string& msg, U32 data_mask) LLGLState::LLGLState(LLGLenum state, S32 enabled) : mState(state), mWasEnabled(FALSE), mIsEnabled(FALSE) { + LL_PROFILE_ZONE_SCOPED; if (LLGLSLShader::sNoFixedFunction) { //always ignore state that's deprecated post GL 3.0 switch (state) @@ -2175,6 +2176,7 @@ void LLGLState::setEnabled(S32 enabled) LLGLState::~LLGLState() { + LL_PROFILE_ZONE_SCOPED; stop_glerror(); if (mState) { diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 103d5388d3..7e726df907 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -577,63 +577,22 @@ void LLVertexBuffer::setupClientArrays(U32 data_mask) } //static -static LLTrace::BlockTimerStatHandle FTM_VB_DRAW_ARRAYS("drawArrays"); -void LLVertexBuffer::drawArrays(U32 mode, const std::vector& pos, const std::vector& norm) +void LLVertexBuffer::drawArrays(U32 mode, const std::vector& pos) { - LL_RECORD_BLOCK_TIME(FTM_VB_DRAW_ARRAYS); - llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL); - gGL.syncMatrices(); - - U32 count = pos.size(); - - llassert(norm.size() >= pos.size()); - llassert(count > 0); - - if( count == 0 ) - { - LL_WARNS() << "Called drawArrays with 0 vertices" << LL_ENDL; - return; - } - - if( norm.size() < pos.size() ) - { - LL_WARNS() << "Called drawArrays with #" << norm.size() << " normals and #" << pos.size() << " vertices" << LL_ENDL; - return; - } - - unbind(); - - setupClientArrays(MAP_VERTEX | MAP_NORMAL); - - LLGLSLShader* shader = LLGLSLShader::sCurBoundShaderPtr; - - if (shader) - { - S32 loc = LLVertexBuffer::TYPE_VERTEX; - if (loc > -1) - { - glVertexAttribPointerARB(loc, 3, GL_FLOAT, GL_FALSE, 0, pos[0].mV); - } - loc = LLVertexBuffer::TYPE_NORMAL; - if (loc > -1) - { - glVertexAttribPointerARB(loc, 3, GL_FLOAT, GL_FALSE, 0, norm[0].mV); - } - } - else - { - glVertexPointer(3, GL_FLOAT, 0, pos[0].mV); - glNormalPointer(GL_FLOAT, 0, norm[0].mV); - } - LLGLSLShader::startProfile(); - LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF0000 ) - glDrawArrays(sGLMode[mode], 0, count); - LLGLSLShader::stopProfile(count, mode); + LL_PROFILE_ZONE_SCOPED; + gGL.begin(mode); + for (auto& v : pos) + { + gGL.vertex3fv(v.mV); + } + gGL.end(); + gGL.flush(); } //static void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVector2* tc, S32 num_indices, const U16* indicesp) { + LL_PROFILE_ZONE_SCOPED; llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL); gGL.syncMatrices(); @@ -646,29 +605,27 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto unbind(); - setupClientArrays(mask); + gGL.begin(mode); - if (LLGLSLShader::sNoFixedFunction) - { - S32 loc = LLVertexBuffer::TYPE_VERTEX; - glVertexAttribPointerARB(loc, 3, GL_FLOAT, GL_FALSE, 16, pos); - - if (tc) - { - loc = LLVertexBuffer::TYPE_TEXCOORD0; - glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, 0, tc); - } - } - else - { - glTexCoordPointer(2, GL_FLOAT, 0, tc); - glVertexPointer(3, GL_FLOAT, 16, pos); - } - - LLGLSLShader::startProfile(); - LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x80FF80 ) - glDrawElements(sGLMode[mode], num_indices, GL_UNSIGNED_SHORT, indicesp); - LLGLSLShader::stopProfile(num_indices, mode); + if (tc != nullptr) + { + for (int i = 0; i < num_indices; ++i) + { + U16 idx = indicesp[i]; + gGL.texCoord2fv(tc[idx].mV); + gGL.vertex3fv(pos[idx].getF32ptr()); + } + } + else + { + for (int i = 0; i < num_indices; ++i) + { + U16 idx = indicesp[i]; + gGL.vertex3fv(pos[idx].getF32ptr()); + } + } + gGL.end(); + gGL.flush(); } void LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_offset) const @@ -851,51 +808,51 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const static LLTrace::BlockTimerStatHandle FTM_GL_DRAW_ARRAYS("GL draw arrays"); void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const { - llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL); - mMappable = false; - gGL.syncMatrices(); - - llassert(mNumVerts >= 0); - if (first >= (U32) mNumVerts || - first + count > (U32) mNumVerts) - { - LL_ERRS() << "Bad vertex buffer draw range: [" << first << ", " << first+count << "]" << LL_ENDL; - } + llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL); + mMappable = false; + gGL.syncMatrices(); - if (mGLArray) - { - if (mGLArray != sGLRenderArray) - { - LL_ERRS() << "Wrong vertex array bound." << LL_ENDL; - } - } - else - { - if (mGLBuffer != sGLRenderBuffer || useVBOs() != sVBOActive) - { - LL_ERRS() << "Wrong vertex buffer bound." << LL_ENDL; - } - } + llassert(mNumVerts >= 0); + if (first >= (U32)mNumVerts || + first + count > (U32)mNumVerts) + { + LL_ERRS() << "Bad vertex buffer draw range: [" << first << ", " << first + count << "]" << LL_ENDL; + } - if (mode >= LLRender::NUM_MODES) - { - LL_ERRS() << "Invalid draw mode: " << mode << LL_ENDL; - return; - } + if (mGLArray) + { + if (mGLArray != sGLRenderArray) + { + LL_ERRS() << "Wrong vertex array bound." << LL_ENDL; + } + } + else + { + if (mGLBuffer != sGLRenderBuffer || useVBOs() != sVBOActive) + { + LL_ERRS() << "Wrong vertex buffer bound." << LL_ENDL; + } + } - { - LL_RECORD_BLOCK_TIME(FTM_GL_DRAW_ARRAYS); - stop_glerror(); - LLGLSLShader::startProfile(); - stop_glerror(); - LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF4040 ) - glDrawArrays(sGLMode[mode], first, count); - stop_glerror(); - LLGLSLShader::stopProfile(count, mode); - } + if (mode >= LLRender::NUM_MODES) + { + LL_ERRS() << "Invalid draw mode: " << mode << LL_ENDL; + return; + } - stop_glerror(); - placeFence(); + { + LL_RECORD_BLOCK_TIME(FTM_GL_DRAW_ARRAYS); + stop_glerror(); + LLGLSLShader::startProfile(); + stop_glerror(); + LL_PROFILER_GPU_ZONEC("gl.DrawArrays", 0xFF4040) + glDrawArrays(sGLMode[mode], first, count); + stop_glerror(); + LLGLSLShader::stopProfile(count, mode); + } + + stop_glerror(); + placeFence(); } //static diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h index 51ed85510e..fad474a143 100644 --- a/indra/llrender/llvertexbuffer.h +++ b/indra/llrender/llvertexbuffer.h @@ -146,8 +146,7 @@ public: static void initClass(bool use_vbo, bool no_vbo_mapping); static void cleanupClass(); static void setupClientArrays(U32 data_mask); - static void pushPositions(U32 mode, const LLVector4a* pos, U32 count); - static void drawArrays(U32 mode, const std::vector& pos, const std::vector& norm); + static void drawArrays(U32 mode, const std::vector& pos); static void drawElements(U32 mode, const LLVector4a* pos, const LLVector2* tc, S32 num_indices, const U16* indicesp); static void unbind(); //unbind any bound vertex buffer diff --git a/indra/newview/llface.h b/indra/newview/llface.h index 3611539ff8..9dd365e3dc 100644 --- a/indra/newview/llface.h +++ b/indra/newview/llface.h @@ -271,10 +271,10 @@ private: LLColor4 mFaceColor; // overrides material color if state |= USE_FACE_COLOR U16 mGeomCount; // vertex count for this face - U16 mGeomIndex; // index into draw pool + U16 mGeomIndex; // starting index into mVertexBuffer's vertex array U8 mTextureIndex; // index of texture channel to use for pseudo-atlasing U32 mIndicesCount; - U32 mIndicesIndex; // index into draw pool for indices (yeah, I know!) + U32 mIndicesIndex; // index into mVertexBuffer's index array S32 mIndexInTex[LLRender::NUM_TEXTURE_CHANNELS]; LLXformMatrix* mXform; diff --git a/indra/newview/llmodelpreview.cpp b/indra/newview/llmodelpreview.cpp index 01bddd781d..4fce6735e0 100644 --- a/indra/newview/llmodelpreview.cpp +++ b/indra/newview/llmodelpreview.cpp @@ -3126,7 +3126,7 @@ BOOL LLModelPreview::render() } gGL.diffuseColor4ubv(hull_colors[i].mV); - LLVertexBuffer::drawArrays(LLRender::TRIANGLES, physics.mMesh[i].mPositions, physics.mMesh[i].mNormals); + LLVertexBuffer::drawArrays(LLRender::TRIANGLES, physics.mMesh[i].mPositions); if (explode > 0.f) { diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index 253b6b9953..5cac9cd7ae 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -2386,7 +2386,12 @@ void renderMeshBaseHull(LLVOVolume* volume, U32 data_mask, LLColor4& color, LLCo if (!decomp->mBaseHullMesh.empty()) { gGL.diffuseColor4fv(color.mV); - LLVertexBuffer::drawArrays(LLRender::TRIANGLES, decomp->mBaseHullMesh.mPositions, decomp->mBaseHullMesh.mNormals); + LLVertexBuffer::drawArrays(LLRender::TRIANGLES, decomp->mBaseHullMesh.mPositions); + + glPolygonMode(GL_FRONT_AND_BACK, GL_LINE); + gGL.diffuseColor4fv(line_color.mV); + LLVertexBuffer::drawArrays(LLRender::TRIANGLES, decomp->mBaseHullMesh.mPositions); + glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); } else { @@ -2406,13 +2411,11 @@ void renderMeshBaseHull(LLVOVolume* volume, U32 data_mask, LLColor4& color, LLCo void render_hull(LLModel::PhysicsMesh& mesh, const LLColor4& color, const LLColor4& line_color) { gGL.diffuseColor4fv(color.mV); - LLVertexBuffer::drawArrays(LLRender::TRIANGLES, mesh.mPositions, mesh.mNormals); - LLGLEnable offset(GL_POLYGON_OFFSET_LINE); + LLVertexBuffer::drawArrays(LLRender::TRIANGLES, mesh.mPositions); glPolygonMode(GL_FRONT_AND_BACK, GL_LINE); - glPolygonOffset(3.f, 3.f); glLineWidth(3.f); gGL.diffuseColor4fv(line_color.mV); - LLVertexBuffer::drawArrays(LLRender::TRIANGLES, mesh.mPositions, mesh.mNormals); + LLVertexBuffer::drawArrays(LLRender::TRIANGLES, mesh.mPositions); glLineWidth(1.f); glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); } @@ -2467,6 +2470,9 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume) gGL.pushMatrix(); gGL.multMatrix((F32*) volume->getRelativeXform().mMatrix); + LLGLEnable(GL_POLYGON_OFFSET_LINE); + glPolygonOffset(3.f, 3.f); + if (type == LLPhysicsShapeBuilderUtil::PhysicsShapeSpecification::USER_MESH) { LLUUID mesh_id = volume->getVolume()->getParams().getSculptID(); @@ -2494,12 +2500,12 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume) { //decomp has physics mesh, render that mesh gGL.diffuseColor4fv(color.mV); - LLVertexBuffer::drawArrays(LLRender::TRIANGLES, decomp->mPhysicsShapeMesh.mPositions, decomp->mPhysicsShapeMesh.mNormals); + LLVertexBuffer::drawArrays(LLRender::TRIANGLES, decomp->mPhysicsShapeMesh.mPositions); glPolygonMode(GL_FRONT_AND_BACK, GL_LINE); gGL.diffuseColor4fv(line_color.mV); - LLVertexBuffer::drawArrays(LLRender::TRIANGLES, decomp->mPhysicsShapeMesh.mPositions, decomp->mPhysicsShapeMesh.mNormals); - glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); + LLVertexBuffer::drawArrays(LLRender::TRIANGLES, decomp->mPhysicsShapeMesh.mPositions); + glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); } else { //no mesh or decomposition, render base hull @@ -2626,8 +2632,8 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume) LLVertexBuffer::unbind(); llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShader != 0); - - LLVertexBuffer::drawElements(LLRender::TRIANGLES, phys_volume->mHullPoints, NULL, phys_volume->mNumHullIndices, phys_volume->mHullIndices); + + LLVertexBuffer::drawElements(LLRender::TRIANGLES, phys_volume->mHullPoints, NULL, phys_volume->mNumHullIndices, phys_volume->mHullIndices); gGL.diffuseColor4fv(color.mV); glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); -- cgit v1.3 From c907d067f41930bd6a4bbef9903febfab1090982 Mon Sep 17 00:00:00 2001 From: Runitai Linden Date: Tue, 26 Oct 2021 09:23:17 -0500 Subject: SL-16243 Followup -- fix for inconsistently calling TracyAlloc/TracyFree --- indra/llcommon/llcommon.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index 04872564bf..25a809dad2 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -44,7 +44,6 @@ void *operator new(size_t size) { LL_PROFILE_ZONE_SCOPED; ptr = (malloc)(size); - TracyAlloc(ptr, size); } else { @@ -54,15 +53,16 @@ void *operator new(size_t size) { throw std::bad_alloc(); } + TracyAlloc(ptr, size); return ptr; } void operator delete(void *ptr) noexcept { + TracyFree(ptr); if (gProfilerEnabled) { LL_PROFILE_ZONE_SCOPED; - TracyFree(ptr); (free)(ptr); } else -- cgit v1.3 From e6eebea8da545350f6684c191c633dd2fbc6f6f1 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 26 Oct 2021 11:49:53 -0400 Subject: SL-16220: Change WorkQueue::runOn() to waitForResult(). In addition to the name making the blocking explicit, we changed the signature: instead of specifying a target WorkQueue on which to run, waitForResult() runs the passed callable on its own WorkQueue. Why is that? Because, unlike postTo(), we do not require a handshake between two different WorkQueues. postTo() allows running arbitrary callback code, setting variables or whatever, on the originating WorkQueue (presumably on the originating thread). waitForResult() synchronizes using Promise/Future, which are explicitly designed for cross-thread communication. We need not call set_value() on the originating thread, so we don't need a postTo() callback lambda. --- indra/llcommon/workqueue.cpp | 7 +-- indra/llcommon/workqueue.h | 145 ++++++++++++++++++------------------------- 2 files changed, 62 insertions(+), 90 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index f7ffc8233c..ac3086aac5 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -26,11 +26,6 @@ using Mutex = LLCoros::Mutex; using Lock = LLCoros::LockType; -struct NotOnDftCoro: public LLException -{ - NotOnDftCoro(const std::string& what): LLException(what) {} -}; - LL::WorkQueue::WorkQueue(const std::string& name): super(makeName(name)) { @@ -148,6 +143,6 @@ void LL::WorkQueue::checkCoroutine(const std::string& method) // string. See also LLCoros::logname(). if (LLCoros::getName().empty()) { - LLTHROW(NotOnDftCoro("Do not call " + method + " from a thread's default coroutine")); + LLTHROW(Error("Do not call " + method + " from a thread's default coroutine")); } } diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index b17c666172..869f5d9a82 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -13,20 +13,13 @@ #define LL_WORKQUEUE_H #include "llcoros.h" +#include "llexception.h" #include "llinstancetracker.h" #include "threadsafeschedule.h" #include +#include // std::current_exception #include // std::function -#if __cplusplus >= 201703 -#include -namespace stdopt = std; -#else -#include -namespace stdopt = boost; -#endif #include -#include // std::pair -#include namespace LL { @@ -51,8 +44,11 @@ namespace LL using TimePoint = Queue::TimePoint; using TimedWork = Queue::TimeTuple; using Closed = Queue::Closed; - template - using optional = stdopt::optional; + + struct Error: public LLException + { + Error(const std::string& what): LLException(what) {} + }; /** * You may omit the WorkQueue name, in which case a unique name is @@ -145,49 +141,25 @@ namespace LL * blocking the calling coroutine until then, returning the result to * caller on completion. * - * REQUIRED: - * - * * The calling thread is the thread servicing 'this' WorkQueue. - * * The calling coroutine is not the @em coroutine servicing this - * WorkQueue. We block the calling coroutine until the result is - * available. If this same coroutine is responsible for checking the - * local WorkQueue, the result will never be dequeued. In practice, - * to try to prevent mistakes, we forbid calling runOn() from a - * thread's default coroutine. - * - * Returns result if able to post, empty optional if the other - * WorkQueue is inaccessible. - * - * If the passed callable has void return, runOn() returns bool true - * if able to post, false if the other WorkQueue is inaccessible. + * In general, we assume that each thread's default coroutine is busy + * servicing its WorkQueue or whatever. To try to prevent mistakes, we + * forbid calling waitForResult() from a thread's default coroutine. */ template - auto runOn(weak_t target, const TimePoint& time, CALLABLE&& callable); + auto waitForResult(const TimePoint& time, CALLABLE&& callable); /** * Post work to another WorkQueue, blocking the calling coroutine * until then, returning the result to caller on completion. * - * REQUIRED: - * - * * The calling thread is the thread servicing 'this' WorkQueue. - * * The calling coroutine is not the @em coroutine servicing this - * WorkQueue. We block the calling coroutine until the result is - * available. If this same coroutine is responsible for checking the - * local WorkQueue, the result will never be dequeued. In practice, - * to try to prevent mistakes, we forbid calling runOn() from a - * thread's default coroutine. - * - * Returns result if able to post, empty optional if the other - * WorkQueue is inaccessible. - * - * If the passed callable has void return, runOn() returns bool true - * if able to post, false if the other WorkQueue is inaccessible. + * In general, we assume that each thread's default coroutine is busy + * servicing its WorkQueue or whatever. To try to prevent mistakes, we + * forbid calling waitForResult() from a thread's default coroutine. */ template - auto runOn(weak_t target, CALLABLE&& callable) + auto waitForResult(CALLABLE&& callable) { - return runOn(target, TimePoint::clock::now(), std::move(callable)); + return waitForResult(TimePoint::clock::now(), std::move(callable)); } /*--------------------------- worker API ---------------------------*/ @@ -246,10 +218,10 @@ namespace LL /// general case: arbitrary C++ return type template - struct RunOn; + struct WaitForResult; /// specialize for CALLABLE returning void template - struct RunOn; + struct WaitForResult; static void checkCoroutine(const std::string& method); static void error(const std::string& msg); @@ -449,65 +421,70 @@ namespace LL /// general case: arbitrary C++ return type template - struct WorkQueue::RunOn + struct WorkQueue::WaitForResult { - optional operator()(WorkQueue* self, weak_t target, - const TimePoint& time, CALLABLE&& callable) + auto operator()(WorkQueue* self, const TimePoint& time, CALLABLE&& callable) { LLCoros::Promise promise; - if (! self->postTo( - target, - time, - std::forward(callable), - // We dare to bind a reference to Promise because it's - // specifically intended for cross-thread synchronization. - [&promise] - (RETURNTYPE&& result) + self->post( + time, + // We dare to bind a reference to Promise because it's + // specifically designed for cross-thread communication. + [&promise, callable = std::move(callable)]() + { + try { - promise.set_value(std::forward(result)); - })) - { - // we couldn't even postTo(): return empty optional - return {}; - } - // we were able to post + // call the caller's callable and trigger promise with result + promise.set_value(callable()); + } + catch (...) + { + promise.set_exception(std::current_exception()); + } + }); auto future{ LLCoros::getFuture(promise) }; - return { future.get(); } + // now, on the calling thread, wait for that result + LLCoros::TempStatus st("waiting for WorkQueue::waitForResult()"); + return future.get(); } }; /// specialize for CALLABLE returning void template - struct WorkQueue::RunOn + struct WorkQueue::WaitForResult { - bool operator()(WorkQueue* self, weak_t target, - const TimePoint& time, CALLABLE&& callable) + void operator()(WorkQueue* self, const TimePoint& time, CALLABLE&& callable) { LLCoros::Promise promise; - if (! self->postTo( - target, - time, - std::forward(callable), - // &promise is designed for cross-thread access - [&promise](){ promise.set_value(); })) - { - // we couldn't postTo() - return false; - } - // we were able to post + self->post( + time, + // &promise is designed for cross-thread access + [&promise, callable = std::move(callable)]() + { + try + { + callable(); + promise.set_value(); + } + catch (...) + { + promise.set_exception(std::current_exception()); + } + }); auto future{ LLCoros::getFuture(promise) }; // block until set_value() + LLCoros::TempStatus st("waiting for void WorkQueue::waitForResult()"); future.get(); - return true; } }; template - auto WorkQueue::runOn(weak_t target, const TimePoint& time, CALLABLE&& callable) + auto WorkQueue::waitForResult(const TimePoint& time, CALLABLE&& callable) { - checkCoroutine("runOn()"); - return RunOn(callable)())>() - (this, target, time, std::forward(callable)); + checkCoroutine("waitForResult()"); + // derive callable's return type so we can specialize for void + return WaitForResult(callable)())>() + (this, time, std::forward(callable)); } } // namespace LL -- cgit v1.3 From f06765cba868679492934452354d16f9f3af9ade Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 26 Oct 2021 12:29:49 -0400 Subject: SL-16220: Make WorkQueue::postTo() return exception to caller. postTo() sets up two-way communication: the caller asks to run work on some other WorkQueue, expecting an eventual callback on the originating WorkQueue. That permits us to transport any exception thrown by the work callable back to rethrow on the originating WorkQueue. --- indra/llcommon/workqueue.h | 93 +++++++++++++++++++++++++++++++--------------- 1 file changed, 64 insertions(+), 29 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index 869f5d9a82..42f5d78ba3 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -136,6 +136,25 @@ namespace LL std::move(callable), std::move(callback)); } + /** + * Post work to be run at a specified time to another WorkQueue, which + * may or may not still exist and be open. Return true if we were able + * to post. + */ + template + static bool postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable); + + /** + * Post work to another WorkQueue, which may or may not still exist + * and be open. Return true if we were able to post. + */ + template + static bool postMaybe(weak_t target, CALLABLE&& callable) + { + return postMaybe(target, TimePoint::clock::now(), + std::forward(callable)); + } + /** * Post work to another WorkQueue to be run at a specified time, * blocking the calling coroutine until then, returning the result to @@ -351,12 +370,8 @@ namespace LL { // Call the callable, which produces no result. std::forward(callable)(); - // This reply lambda binds the original callback, so - // that when we, the originating WorkQueue, finally - // receive and process the reply lambda, we'll call - // the bound callback -- on the same thread that - // originally called postTo(). - return [callback = std::move(callback)](){ callback(); }; + // Our completion callback is simply the caller's callback. + return std::move(callback); } }; @@ -389,36 +404,56 @@ namespace LL callback = std::move(callback)] () { - // Make a reply lambda to repost to THIS WorkQueue. - // Delegate to makeReplyLambda() so we can partially - // specialize on void return. - auto rlambda = makeReplyLambda(std::move(callable), std::move(callback)); - // Check if this originating WorkQueue still exists. - // Remember, the outer lambda is now running on a thread - // servicing the target WorkQueue, and real time has - // elapsed since postTo()'s tptr->post() call. - // reply is a weak_ptr: have to lock it to check it. - auto rptr = reply.lock(); - if (rptr) + // Use postMaybe() below in case this originating WorkQueue + // has been closed or destroyed. Remember, the outer lambda is + // now running on a thread servicing the target WorkQueue, and + // real time has elapsed since postTo()'s tptr->post() call. + try { - // Only post reply lambda if the originating WorkQueue - // still exists. If not -- who would we tell? Log it? - try - { - rptr->post(std::move(rlambda)); - } - catch (const Closed&) - { - // Originating WorkQueue might still exist, but - // might be Closed. Same thing: just discard the - // callback. - } + // Make a reply lambda to repost to THIS WorkQueue. + // Delegate to makeReplyLambda() so we can partially + // specialize on void return. + postMaybe(reply, makeReplyLambda(std::move(callable), std::move(callback))); + } + catch (...) + { + // Either variant of makeReplyLambda() is responsible for + // calling the caller's callable. If that throws, return + // the exception to the originating thread. + postMaybe( + reply, + // Bind the current exception to transport back to the + // originating WorkQueue. Once there, rethrow it. + [exc = std::current_exception()](){ std::rethrow_exception(exc); }); } }); + // looks like we were able to post() return true; } + template + bool WorkQueue::postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable) + { + // target is a weak_ptr: have to lock it to check it + auto tptr = target.lock(); + if (tptr) + { + try + { + tptr->post(time, std::forward(callable)); + // we were able to post() + return true; + } + catch (const Closed&) + { + // target WorkQueue still exists, but is Closed + } + } + // either target no longer exists, or its WorkQueue is Closed + return false; + } + /// general case: arbitrary C++ return type template struct WorkQueue::WaitForResult -- cgit v1.3 From af5c5a994b90a27e16ef6f2f5044e096269e4217 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 27 Oct 2021 13:01:37 -0400 Subject: SL-16207: Update llstring.h handling of different string types. In llpreprocessor.h, consider the case of clang on Windows: #define LL_WCHAR_T_NATIVE there as well as for the Microsoft compiler with /Zc:wchar_t switch. In stdtypes.h, inject a LLWCHAR_IS_WCHAR_T symbol to allow the preprocessor to make decisions about when the types are identical. llstring.h's conversion logic deals with three types of wide strings (LLWString, std::wstring and utf16string) based on three types of wide char (llwchar, wchar_t and U16, respectively). Sometimes they're three distinct types, sometimes wchar_t is identical to llwchar and sometimes wchar_t is identical to U16. Rationalize the three cases using ll_convert_u16_alias() and new ll_convert_wstr_alias() macros. stringize.h was directly calling wstring_to_utf8str() and utf8str_to_wstring(), which was producing errors with VS 2019 clang since there isn't actually a wstring_to_utf8str(std::wstring) overload. Use ll_convert() instead, since that redirects to the relevant ll_convert_wide_to_string() function. (And now you see why we've been trying to migrate to the uniform ll_convert() wrapper!) Similarly, call ll_convert() instead of a two-step conversion from utf8str_to_wstring(), producing LLWString, then a character-by-character copy from LLWString to std::wstring. That isn't even correct: on Windows, we should be encoding from UTF32 to UTF16. --- indra/llcommon/llpreprocessor.h | 4 ++- indra/llcommon/llstring.h | 65 ++++++++++++++++++++++------------------- indra/llcommon/stdtypes.h | 7 +++++ indra/llcommon/stringize.h | 13 ++++----- 4 files changed, 50 insertions(+), 39 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llpreprocessor.h b/indra/llcommon/llpreprocessor.h index b17a8e761a..dc586b0008 100644 --- a/indra/llcommon/llpreprocessor.h +++ b/indra/llcommon/llpreprocessor.h @@ -171,7 +171,9 @@ #define LL_DLLIMPORT #endif // LL_WINDOWS -#if ! defined(LL_WINDOWS) +#if __clang__ || ! defined(LL_WINDOWS) +// Only on Windows, and only with the Microsoft compiler (vs. clang) is +// wchar_t potentially not a distinct type. #define LL_WCHAR_T_NATIVE 1 #else // LL_WINDOWS // https://docs.microsoft.com/en-us/cpp/preprocessor/predefined-macros diff --git a/indra/llcommon/llstring.h b/indra/llcommon/llstring.h index 4263122f36..89e95ef40a 100644 --- a/indra/llcommon/llstring.h +++ b/indra/llcommon/llstring.h @@ -535,6 +535,11 @@ struct ll_convert_impl \ TO operator()(const FROM& in) const { return EXPR; } \ } +// If all we're doing is copying characters, pass this as EXPR. Since it +// expands into the 'return EXPR' slot in the ll_convert_impl specialization +// above, it implies TO{ in.begin(), in.end() }. +#define LL_CONVERT_COPY_CHARS { in.begin(), in.end() } + // Make the incoming string a utf8 string. Replaces any unknown glyph // with the UNKNOWN_CHARACTER. Once any unknown glyph is found, the rest // of the data may not be recovered. @@ -571,30 +576,31 @@ LL_COMMON_API std::string rawstr_to_utf8(const std::string& raw); // LL_WCHAR_T_NATIVE. typedef std::basic_string llutf16string; -#if ! defined(LL_WCHAR_T_NATIVE) -// wchar_t is identical to U16, and std::wstring is identical to llutf16string. -// Defining an ll_convert alias involving llutf16string would collide with the -// comparable preferred alias involving std::wstring. (In this scenario, if -// you pass llutf16string, it will engage the std::wstring specialization.) -#define ll_convert_u16_alias(TO, FROM, EXPR) // nothing -#else // defined(LL_WCHAR_T_NATIVE) -// wchar_t is a distinct native type, so llutf16string is also a distinct -// type, and there IS a point to converting separately to/from llutf16string. -// (But why? Windows APIs are still defined in terms of wchar_t, and -// in this scenario llutf16string won't work for them!) -#define ll_convert_u16_alias(TO, FROM, EXPR) ll_convert_alias(TO, FROM, EXPR) - -#if LL_WINDOWS -// LL_WCHAR_T_NATIVE is defined on non-Windows systems because, in fact, -// wchar_t is native. Everywhere but Windows, we use it for llwchar (see -// stdtypes.h). That makes LLWString identical to std::wstring, so these -// aliases for std::wstring would collide with those for LLWString. Only -// define on Windows, where converting between std::wstring and llutf16string -// means copying chars. -ll_convert_alias(llutf16string, std::wstring, llutf16string(in.begin(), in.end())); -ll_convert_alias(std::wstring, llutf16string, std::wstring(in.begin(), in.end())); -#endif // LL_WINDOWS -#endif // defined(LL_WCHAR_T_NATIVE) +// Considering wchar_t, llwchar and U16, there are three relevant cases: +#if LLWCHAR_IS_WCHAR_T // every which way but Windows +// llwchar is identical to wchar_t, LLWString is identical to std::wstring. +// U16 is distinct, llutf16string is distinct (though pretty useless). +// Given conversions to/from LLWString and to/from llutf16string, conversions +// involving std::wstring would collide. +#define ll_convert_wstr_alias(TO, FROM, EXPR) // nothing +// but we can define conversions involving llutf16string without collisions +#define ll_convert_u16_alias(TO, FROM, EXPR) ll_convert_alias(TO, FROM, EXPR) + +#elif defined(LL_WCHAR_T_NATIVE) // Windows, either clang or MS /Zc:wchar_t +// llwchar (32-bit), wchar_t (16-bit) and U16 are all different types. +// Conversions to/from LLWString, to/from std::wstring and to/from llutf16string +// can all be defined. +#define ll_convert_wstr_alias(TO, FROM, EXPR) ll_convert_alias(TO, FROM, EXPR) +#define ll_convert_u16_alias(TO, FROM, EXPR) ll_convert_alias(TO, FROM, EXPR) + +#else // ! LL_WCHAR_T_NATIVE: Windows with MS /Zc:wchar_t- +// wchar_t is identical to U16, std::wstring is identical to llutf16string. +// Given conversions to/from LLWString and to/from std::wstring, conversions +// involving llutf16string would collide. +#define ll_convert_u16_alias(TO, FROM, EXPR) // nothing +// but we can define conversions involving std::wstring without collisions +#define ll_convert_wstr_alias(TO, FROM, EXPR) ll_convert_alias(TO, FROM, EXPR) +#endif LL_COMMON_API LLWString utf16str_to_wstring(const llutf16string &utf16str, S32 len); LL_COMMON_API LLWString utf16str_to_wstring(const llutf16string &utf16str); @@ -625,9 +631,8 @@ LL_COMMON_API std::string utf16str_to_utf8str(const llutf16string &utf16str, S32 LL_COMMON_API std::string utf16str_to_utf8str(const llutf16string &utf16str); ll_convert_u16_alias(std::string, llutf16string, utf16str_to_utf8str(in)); -#if LL_WINDOWS +// an older alias for utf16str_to_utf8str(llutf16string) inline std::string wstring_to_utf8str(const llutf16string &utf16str) { return utf16str_to_utf8str(utf16str);} -#endif // Length of this UTF32 string in bytes when transformed to UTF8 LL_COMMON_API S32 wstring_utf8_length(const LLWString& wstr); @@ -715,7 +720,7 @@ inline std::string ll_convert_wide_to_string(const std::wstring& in) { return ll_convert_wide_to_string(in.c_str()); } -ll_convert_alias(std::string, std::wstring, ll_convert_wide_to_string(in)); +ll_convert_wstr_alias(std::string, std::wstring, ll_convert_wide_to_string(in)); /** * Converts a string to wide string. @@ -724,19 +729,19 @@ LL_COMMON_API std::wstring ll_convert_string_to_wide(const std::string& in, unsigned int code_page); LL_COMMON_API std::wstring ll_convert_string_to_wide(const std::string& in); // default CP_UTF8 -ll_convert_alias(std::wstring, std::string, ll_convert_string_to_wide(in)); +ll_convert_wstr_alias(std::wstring, std::string, ll_convert_string_to_wide(in)); /** * Convert a Windows wide string to our LLWString */ LL_COMMON_API LLWString ll_convert_wide_to_wstring(const std::wstring& in); -ll_convert_alias(LLWString, std::wstring, ll_convert_wide_to_wstring(in)); +ll_convert_wstr_alias(LLWString, std::wstring, ll_convert_wide_to_wstring(in)); /** * Convert LLWString to Windows wide string */ LL_COMMON_API std::wstring ll_convert_wstring_to_wide(const LLWString& in); -ll_convert_alias(std::wstring, LLWString, ll_convert_wstring_to_wide(in)); +ll_convert_wstr_alias(std::wstring, LLWString, ll_convert_wstring_to_wide(in)); /** * Converts incoming string into utf8 string diff --git a/indra/llcommon/stdtypes.h b/indra/llcommon/stdtypes.h index 887f6ab733..b07805b628 100644 --- a/indra/llcommon/stdtypes.h +++ b/indra/llcommon/stdtypes.h @@ -42,10 +42,17 @@ typedef unsigned int U32; // Windows wchar_t is 16-bit, whichever way /Zc:wchar_t is set. In effect, // Windows wchar_t is always a typedef, either for unsigned short or __wchar_t. // (__wchar_t, available either way, is Microsoft's native 2-byte wchar_t type.) +// The version of clang available with VS 2019 also defines wchar_t as __wchar_t +// which is also 16 bits. // In any case, llwchar should be a UTF-32 type. typedef U32 llwchar; #else typedef wchar_t llwchar; +// What we'd actually want is a simple module-scope 'if constexpr' to test +// std::is_same::value and use that to define, or not +// define, string conversion specializations. Since we don't have that, we'll +// have to rely on #if instead. Sorry, Dr. Stroustrup. +#define LLWCHAR_IS_WCHAR_T 1 #endif #if LL_WINDOWS diff --git a/indra/llcommon/stringize.h b/indra/llcommon/stringize.h index 38dd198ad3..31a114f167 100644 --- a/indra/llcommon/stringize.h +++ b/indra/llcommon/stringize.h @@ -52,7 +52,7 @@ std::basic_string gstringize(const T& item) */ inline std::string stringize(const std::wstring& item) { - return wstring_to_utf8str(item); + return ll_convert(item); } /** @@ -72,8 +72,7 @@ inline std::wstring wstringize(const std::string& item) { // utf8str_to_wstring() returns LLWString, which isn't necessarily the // same as std::wstring - LLWString s(utf8str_to_wstring(item)); - return std::wstring(s.begin(), s.end()); + return ll_convert(item); } /** @@ -146,11 +145,9 @@ void destringize_f(std::basic_string const & str, Functor const & f) * std::istringstream in(str); * in >> item1 >> item2 >> item3 ... ; * @endcode - * @NOTE - once we get generic lambdas, we shouldn't need DEWSTRINGIZE() any - * more since DESTRINGIZE() should do the right thing with a std::wstring. But - * until then, the lambda we pass must accept the right std::basic_istream. */ -#define DESTRINGIZE(STR, EXPRESSION) (destringize_f((STR), [&](std::istream& in){in >> EXPRESSION;})) -#define DEWSTRINGIZE(STR, EXPRESSION) (destringize_f((STR), [&](std::wistream& in){in >> EXPRESSION;})) +#define DESTRINGIZE(STR, EXPRESSION) (destringize_f((STR), [&](auto& in){in >> EXPRESSION;})) +// legacy name, just use DESTRINGIZE() going forward +#define DEWSTRINGIZE(STR, EXPRESSION) DESTRINGIZE(STR, EXPRESSION) #endif /* ! defined(LL_STRINGIZE_H) */ -- cgit v1.3 From 8b16ecb9cfb4917fe38e4e5b0e4f40a23dd4ffbf Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 27 Oct 2021 15:31:54 -0400 Subject: SL-16220: Add tests for WorkQueue::waitForResult(), void & non-void. --- indra/llcommon/tests/workqueue_test.cpp | 49 +++++++++++++++++++++++++++++++++ indra/llcommon/workqueue.h | 38 ++++++++++++------------- 2 files changed, 68 insertions(+), 19 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp index b69df49d33..bea3ad911b 100644 --- a/indra/llcommon/tests/workqueue_test.cpp +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -20,7 +20,10 @@ // external library headers // other Linden headers #include "../test/lltut.h" +#include "../test/catch_and_store_what_in.h" #include "llcond.h" +#include "llcoros.h" +#include "lleventcoro.h" #include "llstring.h" #include "stringize.h" @@ -177,4 +180,50 @@ namespace tut main.runOne(); ensure_equals("failed to run both lambdas", observe, "queue;main"); } + + template<> template<> + void object::test<6>() + { + set_test_name("waitForResult"); + std::string stored; + // Try to call waitForResult() on this thread's main coroutine. It + // should throw because the main coroutine must service the queue. + auto what{ catch_what( + [this, &stored](){ stored = queue.waitForResult( + [](){ return "should throw"; }); }) }; + ensure("lambda should not have run", stored.empty()); + ensure_not("waitForResult() should have thrown", what.empty()); + ensure(STRINGIZE("should mention waitForResult: " << what), + what.find("waitForResult") != std::string::npos); + + // Call waitForResult() on a coroutine, with a string result. + LLCoros::instance().launch( + "waitForResult string", + [this, &stored]() + { stored = queue.waitForResult( + [](){ return "string result"; }); }); + llcoro::suspend(); + // Nothing will have happened yet because, even if the coroutine did + // run immediately, all it did was to queue the inner lambda on + // 'queue'. Service it. + queue.runOne(); + llcoro::suspend(); + ensure_equals("bad waitForResult return", stored, "string result"); + + // Call waitForResult() on a coroutine, with a void callable. + stored.clear(); + bool done = false; + LLCoros::instance().launch( + "waitForResult void", + [this, &stored, &done]() + { + queue.waitForResult([&stored](){ stored = "ran"; }); + done = true; + }); + llcoro::suspend(); + queue.runOne(); + llcoro::suspend(); + ensure_equals("didn't run coroutine", stored, "ran"); + ensure("void waitForResult() didn't return", done); + } } // namespace tut diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index 42f5d78ba3..7dbc735c6d 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -92,6 +92,25 @@ namespace LL post(TimePoint::clock::now(), std::move(callable)); } + /** + * Post work to be run at a specified time to another WorkQueue, which + * may or may not still exist and be open. Return true if we were able + * to post. + */ + template + static bool postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable); + + /** + * Post work to another WorkQueue, which may or may not still exist + * and be open. Return true if we were able to post. + */ + template + static bool postMaybe(weak_t target, CALLABLE&& callable) + { + return postMaybe(target, TimePoint::clock::now(), + std::forward(callable)); + } + /** * Launch a callable returning bool that will trigger repeatedly at * specified interval, until the callable returns false. @@ -136,25 +155,6 @@ namespace LL std::move(callable), std::move(callback)); } - /** - * Post work to be run at a specified time to another WorkQueue, which - * may or may not still exist and be open. Return true if we were able - * to post. - */ - template - static bool postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable); - - /** - * Post work to another WorkQueue, which may or may not still exist - * and be open. Return true if we were able to post. - */ - template - static bool postMaybe(weak_t target, CALLABLE&& callable) - { - return postMaybe(target, TimePoint::clock::now(), - std::forward(callable)); - } - /** * Post work to another WorkQueue to be run at a specified time, * blocking the calling coroutine until then, returning the result to -- cgit v1.3 From 8d20480c5f77fe1fab8149d3cda79bdd61e77656 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 28 Oct 2021 18:06:21 +0000 Subject: SL-16148 SL-16244 SL-16270 SL-16253 Remove most BlockTimers, remove LLMemTracked, introduce alignas, hook most/all reamining allocs, disable synchronous occlusion, and convert frequently accessed LLSingletons to LLSimpleton --- indra/llappearance/llavatarappearance.cpp | 2 +- indra/llappearance/lldriverparam.h | 114 +++---- indra/llappearance/llpolymorph.cpp | 4 +- indra/llappearance/llpolymorph.h | 29 +- indra/llappearance/llpolyskeletaldistortion.cpp | 4 +- indra/llappearance/llpolyskeletaldistortion.h | 30 +- indra/llappearance/lltexlayer.cpp | 21 +- indra/llappearance/lltexlayerparams.cpp | 3 +- indra/llappearance/lltexlayerparams.h | 31 +- indra/llcharacter/llcharacter.cpp | 8 +- indra/llcharacter/lleditingmotion.cpp | 1 + indra/llcharacter/llhandmotion.cpp | 1 + indra/llcharacter/llheadrotmotion.cpp | 2 + indra/llcharacter/llkeyframefallmotion.cpp | 1 + indra/llcharacter/llkeyframemotion.cpp | 1 + indra/llcharacter/llkeyframemotionparam.cpp | 1 + indra/llcharacter/llkeyframewalkmotion.cpp | 3 + indra/llcharacter/llmotioncontroller.cpp | 10 +- indra/llcharacter/lltargetingmotion.cpp | 1 + indra/llcommon/lldate.cpp | 6 +- indra/llcommon/llfasttimer.cpp | 49 +-- indra/llcommon/llmemory.h | 77 +++-- indra/llcommon/llprofiler.h | 8 +- indra/llcommon/llsdparam.cpp | 2 - indra/llcommon/llsdparam.h | 3 +- indra/llcommon/llsingleton.h | 26 ++ indra/llcommon/llstring.cpp | 7 +- indra/llcommon/llsys.cpp | 4 +- indra/llcommon/lltrace.cpp | 1 + indra/llcommon/lltrace.h | 148 +------- indra/llcommon/lltraceaccumulators.cpp | 14 +- indra/llcommon/lltraceaccumulators.h | 19 +- indra/llcommon/lltracerecording.cpp | 109 ++++-- indra/llcommon/lltracerecording.h | 21 ++ indra/llcommon/lltracethreadrecorder.cpp | 4 +- indra/llimage/llimage.cpp | 9 +- indra/llimage/llimage.h | 3 +- indra/llimage/llimagej2c.cpp | 1 - indra/llimage/llimagejpeg.cpp | 8 - indra/llimage/tests/llimageworker_test.cpp | 3 +- indra/llinventory/llinventory.cpp | 24 +- indra/llinventory/llinventory.h | 2 +- indra/llinventory/llsettingsdaycycle.cpp | 3 - indra/llinventory/llsettingssky.cpp | 7 +- indra/llinventory/llsettingswater.cpp | 8 - indra/llkdu/tests/llimagej2ckdu_test.cpp | 3 +- indra/llmath/lloctree.h | 13 +- indra/llmath/llrigginginfo.h | 29 +- indra/llmath/llvector4a.h | 6 +- indra/llmath/llvolumeoctree.h | 27 +- indra/llmessage/llfiltersd2xmlrpc.cpp | 14 +- indra/llmessage/llhttpnode.cpp | 4 + indra/llmessage/lliohttpserver.cpp | 20 +- indra/llmessage/lliosocket.cpp | 10 +- indra/llmessage/llioutil.cpp | 6 +- indra/llmessage/llpumpio.cpp | 11 +- indra/llmessage/lltemplatemessagereader.cpp | 9 +- indra/llrender/llfontbitmapcache.cpp | 18 +- indra/llrender/llfontbitmapcache.h | 2 +- indra/llrender/llfontfreetype.cpp | 9 +- indra/llrender/llfontfreetype.h | 2 +- indra/llrender/llfontgl.cpp | 4 +- indra/llrender/llgl.cpp | 1 + indra/llrender/llimagegl.cpp | 43 +-- indra/llrender/llimagegl.h | 2 +- indra/llrender/llrender2dutils.cpp | 8 +- indra/llrender/lltexture.h | 3 +- indra/llrender/llvertexbuffer.cpp | 72 ++-- indra/llrender/llvertexbuffer.h | 5 +- indra/llui/llfloater.cpp | 17 +- indra/llui/llfolderview.cpp | 10 +- indra/llui/llfolderviewmodel.h | 6 +- indra/llui/lllayoutstack.cpp | 4 +- indra/llui/llpanel.cpp | 4 +- indra/llui/llscrolllistctrl.cpp | 7 +- indra/llui/lltextbase.cpp | 16 +- indra/llui/lluictrl.cpp | 17 +- indra/llui/lluictrlfactory.cpp | 15 +- indra/llui/lluictrlfactory.h | 16 +- indra/llui/llview.cpp | 13 +- indra/llui/llview.h | 3 +- indra/llui/llviewereventrecorder.cpp | 2 + indra/llui/llviewereventrecorder.h | 10 +- indra/llui/llviewmodel.cpp | 20 +- indra/llui/llviewmodel.h | 3 +- indra/newview/app_settings/settings.xml | 13 +- indra/newview/llappviewer.cpp | 28 +- .../newview/lldonotdisturbnotificationstorage.cpp | 4 +- indra/newview/lldrawable.cpp | 41 +-- indra/newview/lldrawable.h | 9 +- indra/newview/lldrawpoolalpha.cpp | 19 +- indra/newview/lldrawpoolavatar.cpp | 378 ++++++++++----------- indra/newview/lldrawpoolbump.cpp | 28 +- indra/newview/lldrawpoolsimple.cpp | 9 +- indra/newview/lldrawpoolterrain.cpp | 1 + indra/newview/lldrawpooltree.cpp | 7 +- indra/newview/lldynamictexture.h | 10 +- indra/newview/llenvironment.cpp | 6 +- indra/newview/llenvironment.h | 11 +- indra/newview/llface.cpp | 135 ++------ indra/newview/llface.h | 6 +- indra/newview/llfasttimerview.cpp | 19 +- indra/newview/llflexibleobject.cpp | 13 +- indra/newview/llgroupmgr.cpp | 14 +- indra/newview/llinventoryfilter.cpp | 4 +- indra/newview/llinventoryitemslist.cpp | 4 +- indra/newview/llinventorypanel.cpp | 6 +- indra/newview/llmaterialmgr.cpp | 4 +- indra/newview/llpersistentnotificationstorage.cpp | 8 +- indra/newview/llphysicsmotion.cpp | 1 + indra/newview/llscenemonitor.cpp | 10 - indra/newview/llscreenchannel.cpp | 3 +- indra/newview/llscripteditor.cpp | 6 +- indra/newview/llselectmgr.cpp | 4 +- indra/newview/llselectmgr.h | 8 +- indra/newview/llspatialpartition.cpp | 70 ++-- indra/newview/llspatialpartition.h | 4 +- indra/newview/llviewercamera.cpp | 5 +- indra/newview/llviewercamera.h | 22 +- indra/newview/llviewerdisplay.cpp | 16 +- indra/newview/llviewerjointmesh.cpp | 6 +- indra/newview/llviewermessage.cpp | 12 +- indra/newview/llviewerobject.cpp | 10 +- indra/newview/llviewerobject.h | 3 +- indra/newview/llviewerobjectlist.cpp | 22 +- indra/newview/llvieweroctree.cpp | 65 +--- indra/newview/llvieweroctree.h | 7 +- indra/newview/llviewertexture.cpp | 17 +- indra/newview/llviewertexture.h | 2 +- indra/newview/llviewertexturelist.cpp | 87 ++--- indra/newview/llvoavatar.cpp | 41 +-- indra/newview/llvoavatar.h | 11 +- indra/newview/llvocache.cpp | 10 +- indra/newview/llvocache.h | 6 +- indra/newview/llvograss.cpp | 4 +- indra/newview/llvopartgroup.cpp | 12 +- indra/newview/llvosky.cpp | 19 +- indra/newview/llvosurfacepatch.cpp | 6 +- indra/newview/llvotree.cpp | 4 +- indra/newview/llvovolume.cpp | 92 ++--- indra/newview/llvowater.cpp | 4 +- indra/newview/llvowlsky.cpp | 4 +- indra/newview/llworld.cpp | 11 +- indra/newview/llworld.h | 7 +- indra/newview/pipeline.cpp | 265 +++++---------- indra/newview/pipeline.h | 3 - 146 files changed, 1066 insertions(+), 1907 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llappearance/llavatarappearance.cpp b/indra/llappearance/llavatarappearance.cpp index 90dfa04f28..2d6d2a10d2 100644 --- a/indra/llappearance/llavatarappearance.cpp +++ b/indra/llappearance/llavatarappearance.cpp @@ -1590,7 +1590,7 @@ BOOL LLAvatarAppearance::allocateCollisionVolumes( U32 num ) delete_and_clear_array(mCollisionVolumes); mNumCollisionVolumes = 0; - mCollisionVolumes = new(std::nothrow) LLAvatarJointCollisionVolume[num]; + mCollisionVolumes = new LLAvatarJointCollisionVolume[num]; if (!mCollisionVolumes) { LL_WARNS() << "Failed to allocate collision volumes" << LL_ENDL; diff --git a/indra/llappearance/lldriverparam.h b/indra/llappearance/lldriverparam.h index f278dcc2e2..a6261b507b 100644 --- a/indra/llappearance/lldriverparam.h +++ b/indra/llappearance/lldriverparam.h @@ -77,73 +77,63 @@ protected: //----------------------------------------------------------------------------- -LL_ALIGN_PREFIX(16) -class LLDriverParam : public LLViewerVisualParam +class alignas(16) LLDriverParam : public LLViewerVisualParam { + LL_ALIGN_NEW private: - // Hide the default constructor. Force construction with LLAvatarAppearance. - LLDriverParam() {} + // Hide the default constructor. Force construction with LLAvatarAppearance. + LLDriverParam() {} public: - LLDriverParam(LLAvatarAppearance *appearance, LLWearable* wearable = NULL); - ~LLDriverParam(); - - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - - // Special: These functions are overridden by child classes - LLDriverParamInfo* getInfo() const { return (LLDriverParamInfo*)mInfo; } - // This sets mInfo and calls initialization functions - BOOL setInfo(LLDriverParamInfo *info); - - LLAvatarAppearance* getAvatarAppearance() { return mAvatarAppearance; } - const LLAvatarAppearance* getAvatarAppearance() const { return mAvatarAppearance; } - - void updateCrossDrivenParams(LLWearableType::EType driven_type); - - /*virtual*/ LLViewerVisualParam* cloneParam(LLWearable* wearable) const; - - // LLVisualParam Virtual functions - /*virtual*/ void apply( ESex sex ) {} // apply is called separately for each driven param. - /*virtual*/ void setWeight(F32 weight); - /*virtual*/ void setAnimationTarget( F32 target_value); - /*virtual*/ void stopAnimating(); - /*virtual*/ BOOL linkDrivenParams(visual_param_mapper mapper, BOOL only_cross_params); - /*virtual*/ void resetDrivenParams(); - - // LLViewerVisualParam Virtual functions - /*virtual*/ F32 getTotalDistortion(); - /*virtual*/ const LLVector4a& getAvgDistortion(); - /*virtual*/ F32 getMaxDistortion(); - /*virtual*/ LLVector4a getVertexDistortion(S32 index, LLPolyMesh *poly_mesh); - /*virtual*/ const LLVector4a* getFirstDistortion(U32 *index, LLPolyMesh **poly_mesh); - /*virtual*/ const LLVector4a* getNextDistortion(U32 *index, LLPolyMesh **poly_mesh); - - S32 getDrivenParamsCount() const; - const LLViewerVisualParam* getDrivenParam(S32 index) const; - - typedef std::vector entry_list_t; - entry_list_t& getDrivenList() { return mDriven; } + LLDriverParam(LLAvatarAppearance* appearance, LLWearable* wearable = NULL); + ~LLDriverParam(); + + // Special: These functions are overridden by child classes + LLDriverParamInfo* getInfo() const { return (LLDriverParamInfo*)mInfo; } + // This sets mInfo and calls initialization functions + BOOL setInfo(LLDriverParamInfo* info); + + LLAvatarAppearance* getAvatarAppearance() { return mAvatarAppearance; } + const LLAvatarAppearance* getAvatarAppearance() const { return mAvatarAppearance; } + + void updateCrossDrivenParams(LLWearableType::EType driven_type); + + /*virtual*/ LLViewerVisualParam* cloneParam(LLWearable* wearable) const; + + // LLVisualParam Virtual functions + /*virtual*/ void apply(ESex sex) {} // apply is called separately for each driven param. + /*virtual*/ void setWeight(F32 weight); + /*virtual*/ void setAnimationTarget(F32 target_value); + /*virtual*/ void stopAnimating(); + /*virtual*/ BOOL linkDrivenParams(visual_param_mapper mapper, BOOL only_cross_params); + /*virtual*/ void resetDrivenParams(); + + // LLViewerVisualParam Virtual functions + /*virtual*/ F32 getTotalDistortion(); + /*virtual*/ const LLVector4a& getAvgDistortion(); + /*virtual*/ F32 getMaxDistortion(); + /*virtual*/ LLVector4a getVertexDistortion(S32 index, LLPolyMesh* poly_mesh); + /*virtual*/ const LLVector4a* getFirstDistortion(U32* index, LLPolyMesh** poly_mesh); + /*virtual*/ const LLVector4a* getNextDistortion(U32* index, LLPolyMesh** poly_mesh); + + S32 getDrivenParamsCount() const; + const LLViewerVisualParam* getDrivenParam(S32 index) const; + + typedef std::vector entry_list_t; + entry_list_t& getDrivenList() { return mDriven; } void setDrivenList(entry_list_t& driven_list) { mDriven = driven_list; } protected: - LLDriverParam(const LLDriverParam& pOther); - F32 getDrivenWeight(const LLDrivenEntry* driven, F32 input_weight); - void setDrivenWeight(LLDrivenEntry *driven, F32 driven_weight); - - - LL_ALIGN_16(LLVector4a mDefaultVec); // temp holder - entry_list_t mDriven; - LLViewerVisualParam* mCurrentDistortionParam; - // Backlink only; don't make this an LLPointer. - LLAvatarAppearance* mAvatarAppearance; - LLWearable* mWearablep; -} LL_ALIGN_POSTFIX(16); + LLDriverParam(const LLDriverParam& pOther); + F32 getDrivenWeight(const LLDrivenEntry* driven, F32 input_weight); + void setDrivenWeight(LLDrivenEntry* driven, F32 driven_weight); + + + LL_ALIGN_16(LLVector4a mDefaultVec); // temp holder + entry_list_t mDriven; + LLViewerVisualParam* mCurrentDistortionParam; + // Backlink only; don't make this an LLPointer. + LLAvatarAppearance* mAvatarAppearance; + LLWearable* mWearablep; +}; #endif // LL_LLDRIVERPARAM_H diff --git a/indra/llappearance/llpolymorph.cpp b/indra/llappearance/llpolymorph.cpp index ce7010984a..16b5f1e204 100644 --- a/indra/llappearance/llpolymorph.cpp +++ b/indra/llappearance/llpolymorph.cpp @@ -539,8 +539,6 @@ F32 LLPolyMorphTarget::getMaxDistortion() //----------------------------------------------------------------------------- // apply() //----------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_APPLY_MORPH_TARGET("Apply Morph"); - void LLPolyMorphTarget::apply( ESex avatar_sex ) { if (!mMorphData || mNumMorphMasksPending > 0) @@ -548,7 +546,7 @@ void LLPolyMorphTarget::apply( ESex avatar_sex ) return; } - LL_RECORD_BLOCK_TIME(FTM_APPLY_MORPH_TARGET); + LL_PROFILE_ZONE_SCOPED; mLastSex = avatar_sex; diff --git a/indra/llappearance/llpolymorph.h b/indra/llappearance/llpolymorph.h index c6133cd831..29cd373636 100644 --- a/indra/llappearance/llpolymorph.h +++ b/indra/llappearance/llpolymorph.h @@ -41,24 +41,14 @@ class LLWearable; //----------------------------------------------------------------------------- // LLPolyMorphData() //----------------------------------------------------------------------------- -LL_ALIGN_PREFIX(16) -class LLPolyMorphData +class alignas(16) LLPolyMorphData { + LL_ALIGN_NEW public: LLPolyMorphData(const std::string& morph_name); ~LLPolyMorphData(); LLPolyMorphData(const LLPolyMorphData &rhs); - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - BOOL loadBinary(LLFILE* fp, LLPolyMeshSharedData *mesh); const std::string& getName() { return mName; } @@ -76,7 +66,7 @@ public: F32 mTotalDistortion; // vertex distortion summed over entire morph F32 mMaxDistortion; // maximum single vertex distortion in a given morph - LL_ALIGN_16(LLVector4a mAvgDistortion); // average vertex distortion, to infer directionality of the morph + LLVector4a mAvgDistortion; // average vertex distortion, to infer directionality of the morph LLPolyMeshSharedData* mMesh; private: @@ -154,8 +144,9 @@ protected: // These morph targets must be topologically consistent with a given Polymesh // (share face sets) //----------------------------------------------------------------------------- -class LLPolyMorphTarget : public LLViewerVisualParam +class alignas(16) LLPolyMorphTarget : public LLViewerVisualParam { + LL_ALIGN_NEW public: LLPolyMorphTarget(LLPolyMesh *poly_mesh); ~LLPolyMorphTarget(); @@ -184,16 +175,6 @@ public: void applyVolumeChanges(F32 delta_weight); // SL-315 - for resetSkeleton() - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - protected: LLPolyMorphTarget(const LLPolyMorphTarget& pOther); diff --git a/indra/llappearance/llpolyskeletaldistortion.cpp b/indra/llappearance/llpolyskeletaldistortion.cpp index ae38c25dbf..360f17508f 100644 --- a/indra/llappearance/llpolyskeletaldistortion.cpp +++ b/indra/llappearance/llpolyskeletaldistortion.cpp @@ -188,11 +188,9 @@ BOOL LLPolySkeletalDistortion::setInfo(LLPolySkeletalDistortionInfo *info) //----------------------------------------------------------------------------- // apply() //----------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_POLYSKELETAL_DISTORTION_APPLY("Skeletal Distortion"); - void LLPolySkeletalDistortion::apply( ESex avatar_sex ) { - LL_RECORD_BLOCK_TIME(FTM_POLYSKELETAL_DISTORTION_APPLY); + LL_PROFILE_ZONE_SCOPED; F32 effective_weight = ( getSex() & avatar_sex ) ? mCurWeight : getDefaultWeight(); diff --git a/indra/llappearance/llpolyskeletaldistortion.h b/indra/llappearance/llpolyskeletaldistortion.h index ab1a132d19..585d85f055 100644 --- a/indra/llappearance/llpolyskeletaldistortion.h +++ b/indra/llappearance/llpolyskeletaldistortion.h @@ -62,9 +62,9 @@ struct LLPolySkeletalBoneInfo BOOL mHasPositionDeformation; }; -LL_ALIGN_PREFIX(16) -class LLPolySkeletalDistortionInfo : public LLViewerVisualParamInfo +class alignas(16) LLPolySkeletalDistortionInfo : public LLViewerVisualParamInfo { + LL_ALIGN_NEW friend class LLPolySkeletalDistortion; public: @@ -73,19 +73,6 @@ public: /*virtual*/ BOOL parseXml(LLXmlTreeNode* node); - - - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - - protected: typedef std::vector bone_info_list_t; bone_info_list_t mBoneInfoList; @@ -95,19 +82,10 @@ protected: // LLPolySkeletalDeformation // A set of joint scale data for deforming the avatar mesh //----------------------------------------------------------------------------- -class LLPolySkeletalDistortion : public LLViewerVisualParam +class alignas(16) LLPolySkeletalDistortion : public LLViewerVisualParam { + LL_ALIGN_NEW public: - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - LLPolySkeletalDistortion(LLAvatarAppearance *avatarp); ~LLPolySkeletalDistortion(); diff --git a/indra/llappearance/lltexlayer.cpp b/indra/llappearance/lltexlayer.cpp index e5039141de..26e0ae9086 100644 --- a/indra/llappearance/lltexlayer.cpp +++ b/indra/llappearance/lltexlayer.cpp @@ -522,10 +522,9 @@ const LLTexLayerSetBuffer* LLTexLayerSet::getComposite() const return mComposite; } -static LLTrace::BlockTimerStatHandle FTM_GATHER_MORPH_MASK_ALPHA("gatherMorphMaskAlpha"); void LLTexLayerSet::gatherMorphMaskAlpha(U8 *data, S32 origin_x, S32 origin_y, S32 width, S32 height, LLRenderTarget* bound_target) { - LL_RECORD_BLOCK_TIME(FTM_GATHER_MORPH_MASK_ALPHA); + LL_PROFILE_ZONE_SCOPED; memset(data, 255, width * height); for( layer_list_t::iterator iter = mLayerList.begin(); iter != mLayerList.end(); iter++ ) @@ -538,10 +537,9 @@ void LLTexLayerSet::gatherMorphMaskAlpha(U8 *data, S32 origin_x, S32 origin_y, S renderAlphaMaskTextures(origin_x, origin_y, width, height, bound_target, true); } -static LLTrace::BlockTimerStatHandle FTM_RENDER_ALPHA_MASK_TEXTURES("renderAlphaMaskTextures"); void LLTexLayerSet::renderAlphaMaskTextures(S32 x, S32 y, S32 width, S32 height, LLRenderTarget* bound_target, bool forceClear) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_MASK_TEXTURES); + LL_PROFILE_ZONE_SCOPED; const LLTexLayerSetInfo *info = getInfo(); bool use_shaders = LLGLSLShader::sNoFixedFunction; @@ -1433,7 +1431,6 @@ BOOL LLTexLayer::blendAlphaTexture(S32 x, S32 y, S32 width, S32 height) addAlphaMask(data, originX, originY, width, height, bound_target); } -static LLTrace::BlockTimerStatHandle FTM_RENDER_MORPH_MASKS("renderMorphMasks"); void LLTexLayer::renderMorphMasks(S32 x, S32 y, S32 width, S32 height, const LLColor4 &layer_color, LLRenderTarget* bound_target, bool force_render) { if (!force_render && !hasMorph()) @@ -1441,7 +1438,7 @@ void LLTexLayer::renderMorphMasks(S32 x, S32 y, S32 width, S32 height, const LLC LL_DEBUGS() << "skipping renderMorphMasks for " << getUUID() << LL_ENDL; return; } - LL_RECORD_BLOCK_TIME(FTM_RENDER_MORPH_MASKS); + LL_PROFILE_ZONE_SCOPED; BOOL success = TRUE; llassert( !mParamAlphaList.empty() ); @@ -1639,10 +1636,9 @@ void LLTexLayer::renderMorphMasks(S32 x, S32 y, S32 width, S32 height, const LLC } } -static LLTrace::BlockTimerStatHandle FTM_ADD_ALPHA_MASK("addAlphaMask"); void LLTexLayer::addAlphaMask(U8 *data, S32 originX, S32 originY, S32 width, S32 height, LLRenderTarget* bound_target) { - LL_RECORD_BLOCK_TIME(FTM_ADD_ALPHA_MASK); + LL_PROFILE_ZONE_SCOPED; S32 size = width * height; const U8* alphaData = getAlphaData(); if (!alphaData && hasAlphaParams()) @@ -1983,10 +1979,9 @@ void LLTexLayerStaticImageList::deleteCachedImages() // Returns an LLImageTGA that contains the encoded data from a tga file named file_name. // Caches the result to speed identical subsequent requests. -static LLTrace::BlockTimerStatHandle FTM_LOAD_STATIC_TGA("getImageTGA"); LLImageTGA* LLTexLayerStaticImageList::getImageTGA(const std::string& file_name) { - LL_RECORD_BLOCK_TIME(FTM_LOAD_STATIC_TGA); + LL_PROFILE_ZONE_SCOPED; const char *namekey = mImageNames.addString(file_name); image_tga_map_t::const_iterator iter = mStaticImageListTGA.find(namekey); if( iter != mStaticImageListTGA.end() ) @@ -2013,10 +2008,9 @@ LLImageTGA* LLTexLayerStaticImageList::getImageTGA(const std::string& file_name) // Returns a GL Image (without a backing ImageRaw) that contains the decoded data from a tga file named file_name. // Caches the result to speed identical subsequent requests. -static LLTrace::BlockTimerStatHandle FTM_LOAD_STATIC_TEXTURE("getTexture"); LLGLTexture* LLTexLayerStaticImageList::getTexture(const std::string& file_name, BOOL is_mask) { - LL_RECORD_BLOCK_TIME(FTM_LOAD_STATIC_TEXTURE); + LL_PROFILE_ZONE_SCOPED; LLPointer tex; const char *namekey = mImageNames.addString(file_name); @@ -2063,10 +2057,9 @@ LLGLTexture* LLTexLayerStaticImageList::getTexture(const std::string& file_name, // Reads a .tga file, decodes it, and puts the decoded data in image_raw. // Returns TRUE if successful. -static LLTrace::BlockTimerStatHandle FTM_LOAD_IMAGE_RAW("loadImageRaw"); BOOL LLTexLayerStaticImageList::loadImageRaw(const std::string& file_name, LLImageRaw* image_raw) { - LL_RECORD_BLOCK_TIME(FTM_LOAD_IMAGE_RAW); + LL_PROFILE_ZONE_SCOPED; BOOL success = FALSE; std::string path; path = gDirUtilp->getExpandedFilename(LL_PATH_CHARACTER,file_name); diff --git a/indra/llappearance/lltexlayerparams.cpp b/indra/llappearance/lltexlayerparams.cpp index ff682d6906..ce5c7142d5 100644 --- a/indra/llappearance/lltexlayerparams.cpp +++ b/indra/llappearance/lltexlayerparams.cpp @@ -261,10 +261,9 @@ BOOL LLTexLayerParamAlpha::getSkip() const } -static LLTrace::BlockTimerStatHandle FTM_TEX_LAYER_PARAM_ALPHA("alpha render"); BOOL LLTexLayerParamAlpha::render(S32 x, S32 y, S32 width, S32 height) { - LL_RECORD_BLOCK_TIME(FTM_TEX_LAYER_PARAM_ALPHA); + LL_PROFILE_ZONE_SCOPED; BOOL success = TRUE; if (!mTexLayer) diff --git a/indra/llappearance/lltexlayerparams.h b/indra/llappearance/lltexlayerparams.h index 0cb2dedbff..e2440998b3 100644 --- a/indra/llappearance/lltexlayerparams.h +++ b/indra/llappearance/lltexlayerparams.h @@ -63,23 +63,14 @@ protected: // //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ LL_ALIGN_PREFIX(16) -class LLTexLayerParamAlpha : public LLTexLayerParam +class alignas(16) LLTexLayerParamAlpha : public LLTexLayerParam { + LL_ALIGN_NEW public: LLTexLayerParamAlpha( LLTexLayerInterface* layer ); LLTexLayerParamAlpha( LLAvatarAppearance* appearance ); /*virtual*/ ~LLTexLayerParamAlpha(); - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - /*virtual*/ LLViewerVisualParam* cloneParam(LLWearable* wearable = NULL) const; // LLVisualParam Virtual functions @@ -146,9 +137,9 @@ private: // //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -LL_ALIGN_PREFIX(16) -class LLTexLayerParamColor : public LLTexLayerParam +class alignas(16) LLTexLayerParamColor : public LLTexLayerParam { + LL_ALIGN_NEW public: enum EColorOperation { @@ -161,16 +152,6 @@ public: LLTexLayerParamColor( LLTexLayerInterface* layer ); LLTexLayerParamColor( LLAvatarAppearance* appearance ); - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - /* virtual */ ~LLTexLayerParamColor(); /*virtual*/ LLViewerVisualParam* cloneParam(LLWearable* wearable = NULL) const; @@ -198,8 +179,8 @@ protected: virtual void onGlobalColorChanged() {} private: - LL_ALIGN_16(LLVector4a mAvgDistortionVec); -} LL_ALIGN_POSTFIX(16); + LLVector4a mAvgDistortionVec; +}; class LLTexLayerParamColorInfo : public LLViewerVisualParamInfo { diff --git a/indra/llcharacter/llcharacter.cpp b/indra/llcharacter/llcharacter.cpp index b764ef0c7e..8874df32f5 100644 --- a/indra/llcharacter/llcharacter.cpp +++ b/indra/llcharacter/llcharacter.cpp @@ -188,20 +188,15 @@ void LLCharacter::requestStopMotion( LLMotion* motion) //----------------------------------------------------------------------------- // updateMotions() //----------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_UPDATE_ANIMATION("Update Animation"); -static LLTrace::BlockTimerStatHandle FTM_UPDATE_HIDDEN_ANIMATION("Update Hidden Anim"); -static LLTrace::BlockTimerStatHandle FTM_UPDATE_MOTIONS("Update Motions"); - void LLCharacter::updateMotions(e_update_t update_type) { + LL_PROFILE_ZONE_SCOPED; if (update_type == HIDDEN_UPDATE) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_HIDDEN_ANIMATION); mMotionController.updateMotionsMinimal(); } else { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_ANIMATION); // unpause if the number of outstanding pause requests has dropped to the initial one if (mMotionController.isPaused() && mPauseRequest->getNumRefs() == 1) { @@ -209,7 +204,6 @@ void LLCharacter::updateMotions(e_update_t update_type) } bool force_update = (update_type == FORCE_UPDATE); { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_MOTIONS); mMotionController.updateMotions(force_update); } } diff --git a/indra/llcharacter/lleditingmotion.cpp b/indra/llcharacter/lleditingmotion.cpp index ddf89f30f2..c5757163d9 100644 --- a/indra/llcharacter/lleditingmotion.cpp +++ b/indra/llcharacter/lleditingmotion.cpp @@ -163,6 +163,7 @@ BOOL LLEditingMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLEditingMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; LLVector3 focus_pt; LLVector3* pointAtPt = (LLVector3*)mCharacter->getAnimationData("PointAtPoint"); diff --git a/indra/llcharacter/llhandmotion.cpp b/indra/llcharacter/llhandmotion.cpp index b3bf5a9a91..ceba956214 100644 --- a/indra/llcharacter/llhandmotion.cpp +++ b/indra/llcharacter/llhandmotion.cpp @@ -121,6 +121,7 @@ BOOL LLHandMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLHandMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; eHandPose *requestedHandPose; F32 timeDelta = time - mLastTime; diff --git a/indra/llcharacter/llheadrotmotion.cpp b/indra/llcharacter/llheadrotmotion.cpp index e91de7a11d..fdf97266a3 100644 --- a/indra/llcharacter/llheadrotmotion.cpp +++ b/indra/llcharacter/llheadrotmotion.cpp @@ -175,6 +175,7 @@ BOOL LLHeadRotMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLHeadRotMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; LLQuaternion targetHeadRotWorld; LLQuaternion currentRootRotWorld = mRootJoint->getWorldRotation(); LLQuaternion currentInvRootRotWorld = ~currentRootRotWorld; @@ -458,6 +459,7 @@ void LLEyeMotion::adjustEyeTarget(LLVector3* targetPos, LLJointState& left_eye_s //----------------------------------------------------------------------------- BOOL LLEyeMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; //calculate jitter if (mEyeJitterTimer.getElapsedTimeF32() > mEyeJitterTime) { diff --git a/indra/llcharacter/llkeyframefallmotion.cpp b/indra/llcharacter/llkeyframefallmotion.cpp index 60ab2e9929..ac53bcd768 100644 --- a/indra/llcharacter/llkeyframefallmotion.cpp +++ b/indra/llcharacter/llkeyframefallmotion.cpp @@ -116,6 +116,7 @@ BOOL LLKeyframeFallMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLKeyframeFallMotion::onUpdate(F32 activeTime, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; BOOL result = LLKeyframeMotion::onUpdate(activeTime, joint_mask); F32 slerp_amt = clamp_rescale(activeTime / getDuration(), 0.5f, 0.75f, 0.f, 1.f); diff --git a/indra/llcharacter/llkeyframemotion.cpp b/indra/llcharacter/llkeyframemotion.cpp index cde38c8091..429c479d0d 100644 --- a/indra/llcharacter/llkeyframemotion.cpp +++ b/indra/llcharacter/llkeyframemotion.cpp @@ -683,6 +683,7 @@ BOOL LLKeyframeMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLKeyframeMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; // llassert(time >= 0.f); // This will fire time = llmax(0.f, time); diff --git a/indra/llcharacter/llkeyframemotionparam.cpp b/indra/llcharacter/llkeyframemotionparam.cpp index 6ed18bc445..aba1c5db39 100644 --- a/indra/llcharacter/llkeyframemotionparam.cpp +++ b/indra/llcharacter/llkeyframemotionparam.cpp @@ -158,6 +158,7 @@ BOOL LLKeyframeMotionParam::onActivate() //----------------------------------------------------------------------------- BOOL LLKeyframeMotionParam::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; F32 weightFactor = 1.f / (F32)mParameterizedMotions.size(); // zero out all pose weights diff --git a/indra/llcharacter/llkeyframewalkmotion.cpp b/indra/llcharacter/llkeyframewalkmotion.cpp index f180702385..298b37e60c 100644 --- a/indra/llcharacter/llkeyframewalkmotion.cpp +++ b/indra/llcharacter/llkeyframewalkmotion.cpp @@ -105,6 +105,7 @@ void LLKeyframeWalkMotion::onDeactivate() //----------------------------------------------------------------------------- BOOL LLKeyframeWalkMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; // compute time since last update F32 deltaTime = time - mRealTimeLast; @@ -198,6 +199,7 @@ BOOL LLWalkAdjustMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLWalkAdjustMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; // delta_time is guaranteed to be non zero F32 delta_time = llclamp(time - mLastTime, TIME_EPSILON, MAX_TIME_DELTA); mLastTime = time; @@ -373,6 +375,7 @@ BOOL LLFlyAdjustMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLFlyAdjustMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; LLVector3 ang_vel = mCharacter->getCharacterAngularVelocity() * mCharacter->getTimeDilation(); F32 speed = mCharacter->getCharacterVelocity().magVec(); diff --git a/indra/llcharacter/llmotioncontroller.cpp b/indra/llcharacter/llmotioncontroller.cpp index c48d02b652..d4546ce901 100644 --- a/indra/llcharacter/llmotioncontroller.cpp +++ b/indra/llcharacter/llmotioncontroller.cpp @@ -503,6 +503,7 @@ void LLMotionController::resetJointSignatures() //----------------------------------------------------------------------------- void LLMotionController::updateIdleMotion(LLMotion* motionp) { + LL_PROFILE_ZONE_SCOPED; if (motionp->isStopped() && mAnimTime > motionp->getStopTime() + motionp->getEaseOutDuration()) { deactivateMotionInstance(motionp); @@ -541,6 +542,7 @@ void LLMotionController::updateIdleMotion(LLMotion* motionp) //----------------------------------------------------------------------------- void LLMotionController::updateIdleActiveMotions() { + LL_PROFILE_ZONE_SCOPED; for (motion_list_t::iterator iter = mActiveMotions.begin(); iter != mActiveMotions.end(); ) { @@ -553,10 +555,9 @@ void LLMotionController::updateIdleActiveMotions() //----------------------------------------------------------------------------- // updateMotionsByType() //----------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_MOTION_ON_UPDATE("Motion onUpdate"); - void LLMotionController::updateMotionsByType(LLMotion::LLMotionBlendType anim_type) { + LL_PROFILE_ZONE_SCOPED; BOOL update_result = TRUE; U8 last_joint_signature[LL_CHARACTER_MAX_ANIMATED_JOINTS]; @@ -712,7 +713,6 @@ void LLMotionController::updateMotionsByType(LLMotion::LLMotionBlendType anim_ty // perform motion update { - LL_RECORD_BLOCK_TIME(FTM_MOTION_ON_UPDATE); update_result = motionp->onUpdate(mAnimTime - motionp->mActivationTimestamp, last_joint_signature); } } @@ -768,6 +768,7 @@ void LLMotionController::updateMotionsByType(LLMotion::LLMotionBlendType anim_ty //----------------------------------------------------------------------------- void LLMotionController::updateLoadingMotions() { + LL_PROFILE_ZONE_SCOPED; // query pending motions for completion for (motion_set_t::iterator iter = mLoadingMotions.begin(); iter != mLoadingMotions.end(); ) @@ -815,6 +816,7 @@ void LLMotionController::updateLoadingMotions() //----------------------------------------------------------------------------- void LLMotionController::updateMotions(bool force_update) { + LL_PROFILE_ZONE_SCOPED; // SL-763: "Distant animated objects run at super fast speed" // The use_quantum optimization or possibly the associated code in setTimeStamp() // does not work as implemented. @@ -907,6 +909,7 @@ void LLMotionController::updateMotions(bool force_update) //----------------------------------------------------------------------------- void LLMotionController::updateMotionsMinimal() { + LL_PROFILE_ZONE_SCOPED; // Always update mPrevTimerElapsed mPrevTimerElapsed = mTimer.getElapsedTimeF32(); @@ -924,6 +927,7 @@ void LLMotionController::updateMotionsMinimal() //----------------------------------------------------------------------------- BOOL LLMotionController::activateMotionInstance(LLMotion *motion, F32 time) { + LL_PROFILE_ZONE_SCOPED; // It's not clear why the getWeight() line seems to be crashing this, but // hopefully this fixes it. if (motion == NULL || motion->getPose() == NULL) diff --git a/indra/llcharacter/lltargetingmotion.cpp b/indra/llcharacter/lltargetingmotion.cpp index 69681e4197..ec75212a40 100644 --- a/indra/llcharacter/lltargetingmotion.cpp +++ b/indra/llcharacter/lltargetingmotion.cpp @@ -103,6 +103,7 @@ BOOL LLTargetingMotion::onActivate() //----------------------------------------------------------------------------- BOOL LLTargetingMotion::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; F32 slerp_amt = LLSmoothInterpolation::getInterpolant(TORSO_TARGET_HALF_LIFE); LLVector3 target; diff --git a/indra/llcommon/lldate.cpp b/indra/llcommon/lldate.cpp index 7a2a0869f4..2ddcf40895 100644 --- a/indra/llcommon/lldate.cpp +++ b/indra/llcommon/lldate.cpp @@ -86,11 +86,9 @@ std::string LLDate::asRFC1123() const return toHTTPDateString (std::string ("%A, %d %b %Y %H:%M:%S GMT")); } -LLTrace::BlockTimerStatHandle FT_DATE_FORMAT("Date Format"); - std::string LLDate::toHTTPDateString (std::string fmt) const { - LL_RECORD_BLOCK_TIME(FT_DATE_FORMAT); + LL_PROFILE_ZONE_SCOPED; time_t locSeconds = (time_t) mSecondsSinceEpoch; struct tm * gmt = gmtime (&locSeconds); @@ -99,7 +97,7 @@ std::string LLDate::toHTTPDateString (std::string fmt) const std::string LLDate::toHTTPDateString (tm * gmt, std::string fmt) { - LL_RECORD_BLOCK_TIME(FT_DATE_FORMAT); + LL_PROFILE_ZONE_SCOPED; // avoid calling setlocale() unnecessarily - it's expensive. static std::string prev_locale = ""; diff --git a/indra/llcommon/llfasttimer.cpp b/indra/llcommon/llfasttimer.cpp index 5b6a7b82f8..d38946004f 100644 --- a/indra/llcommon/llfasttimer.cpp +++ b/indra/llcommon/llfasttimer.cpp @@ -191,29 +191,30 @@ TimeBlockTreeNode& BlockTimerStatHandle::getTreeNode() const } + void BlockTimer::bootstrapTimerTree() { - for (auto& base : BlockTimerStatHandle::instance_snapshot()) - { - // because of indirect derivation from LLInstanceTracker, have to downcast - BlockTimerStatHandle& timer = static_cast(base); - if (&timer == &BlockTimer::getRootTimeBlock()) continue; - - // bootstrap tree construction by attaching to last timer to be on stack - // when this timer was called - if (timer.getParent() == &BlockTimer::getRootTimeBlock()) - { - TimeBlockAccumulator& accumulator = timer.getCurrentAccumulator(); - - if (accumulator.mLastCaller) - { - timer.setParent(accumulator.mLastCaller); - accumulator.mParent = accumulator.mLastCaller; - } - // no need to push up tree on first use, flag can be set spuriously - accumulator.mMoveUpTree = false; - } - } + for (auto& base : BlockTimerStatHandle::instance_snapshot()) + { + // because of indirect derivation from LLInstanceTracker, have to downcast + BlockTimerStatHandle& timer = static_cast(base); + if (&timer == &BlockTimer::getRootTimeBlock()) continue; + + // bootstrap tree construction by attaching to last timer to be on stack + // when this timer was called + if (timer.getParent() == &BlockTimer::getRootTimeBlock()) + { + TimeBlockAccumulator& accumulator = timer.getCurrentAccumulator(); + + if (accumulator.mLastCaller) + { + timer.setParent(accumulator.mLastCaller); + accumulator.mParent = accumulator.mLastCaller; + } + // no need to push up tree on first use, flag can be set spuriously + accumulator.mMoveUpTree = false; + } + } } // bump timers up tree if they have been flagged as being in the wrong place @@ -221,6 +222,7 @@ void BlockTimer::bootstrapTimerTree() // this preserves partial order derived from current frame's observations void BlockTimer::incrementalUpdateTimerTree() { + LL_PROFILE_ZONE_SCOPED; for(block_timer_tree_df_post_iterator_t it = begin_block_timer_tree_df_post(BlockTimer::getRootTimeBlock()); it != end_block_timer_tree_df_post(); ++it) @@ -260,7 +262,8 @@ void BlockTimer::incrementalUpdateTimerTree() void BlockTimer::updateTimes() - { +{ + LL_PROFILE_ZONE_SCOPED; // walk up stack of active timers and accumulate current time while leaving timing structures active BlockTimerStackRecord* stack_record = LLThreadLocalSingletonPointer::getInstance(); if (!stack_record) return; @@ -271,7 +274,7 @@ void BlockTimer::updateTimes() while(cur_timer && cur_timer->mParentTimerData.mActiveTimer != cur_timer) // root defined by parent pointing to self - { + { U64 cumulative_time_delta = cur_time - cur_timer->mStartTime; cur_timer->mStartTime = cur_time; diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h index 2704a495e0..41023b4ba4 100644 --- a/indra/llcommon/llmemory.h +++ b/indra/llcommon/llmemory.h @@ -109,6 +109,16 @@ public: \ } \ \ void operator delete(void* ptr) \ + { \ + ll_aligned_free_16(ptr); \ + } \ + \ + void* operator new[](size_t size) \ + { \ + return ll_aligned_malloc_16(size); \ + } \ + \ + void operator delete[](void* ptr) \ { \ ll_aligned_free_16(ptr); \ } @@ -126,8 +136,9 @@ public: \ #else inline void* ll_aligned_malloc_fallback( size_t size, int align ) { + LL_PROFILE_ZONE_SCOPED; #if defined(LL_WINDOWS) - return _aligned_malloc(size, align); + void* ret = _aligned_malloc(size, align); #else char* aligned = NULL; void* mem = malloc( size + (align - 1) + sizeof(void*) ); @@ -138,12 +149,16 @@ public: \ ((void**)aligned)[-1] = mem; } - return aligned; + void* ret = aligned; #endif + LL_PROFILE_ALLOC(ret, size); + return ret; } inline void ll_aligned_free_fallback( void* ptr ) { + LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_FREE(ptr); #if defined(LL_WINDOWS) _aligned_free(ptr); #else @@ -159,21 +174,24 @@ public: \ inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed with ll_aligned_free_16(). { + LL_PROFILE_ZONE_SCOPED; #if defined(LL_WINDOWS) - return _aligned_malloc(size, 16); + void* ret = _aligned_malloc(size, 16); #elif defined(LL_DARWIN) - return malloc(size); // default osx malloc is 16 byte aligned. + void* ret = malloc(size); // default osx malloc is 16 byte aligned. #else - void *rtn; - if (LL_LIKELY(0 == posix_memalign(&rtn, 16, size))) - return rtn; - else // bad alignment requested, or out of memory - return NULL; + void *ret; + if (0 != posix_memalign(&ret, 16, size)) + return nullptr; #endif + LL_PROFILE_ALLOC(ret, size); + return ret; } inline void ll_aligned_free_16(void *p) { + LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_FREE(p); #if defined(LL_WINDOWS) _aligned_free(p); #elif defined(LL_DARWIN) @@ -185,10 +203,12 @@ inline void ll_aligned_free_16(void *p) inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // returned hunk MUST be freed with ll_aligned_free_16(). { + LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_FREE(ptr); #if defined(LL_WINDOWS) - return _aligned_realloc(ptr, size, 16); + void* ret = _aligned_realloc(ptr, size, 16); #elif defined(LL_DARWIN) - return realloc(ptr,size); // default osx malloc is 16 byte aligned. + void* ret = realloc(ptr,size); // default osx malloc is 16 byte aligned. #else //FIXME: memcpy is SLOW void* ret = ll_aligned_malloc_16(size); @@ -201,27 +221,31 @@ inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // r } ll_aligned_free_16(ptr); } - return ret; #endif + LL_PROFILE_ALLOC(ptr, size); + return ret; } inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed with ll_aligned_free_32(). { + LL_PROFILE_ZONE_SCOPED; #if defined(LL_WINDOWS) - return _aligned_malloc(size, 32); + void* ret = _aligned_malloc(size, 32); #elif defined(LL_DARWIN) - return ll_aligned_malloc_fallback( size, 32 ); + void* ret = ll_aligned_malloc_fallback( size, 32 ); #else - void *rtn; - if (LL_LIKELY(0 == posix_memalign(&rtn, 32, size))) - return rtn; - else // bad alignment requested, or out of memory - return NULL; + void *ret; + if (0 != posix_memalign(&ret, 32, size)) + return nullptr; #endif + LL_PROFILE_ALLOC(ret, size); + return ret; } inline void ll_aligned_free_32(void *p) { + LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_FREE(p); #if defined(LL_WINDOWS) _aligned_free(p); #elif defined(LL_DARWIN) @@ -235,29 +259,35 @@ inline void ll_aligned_free_32(void *p) template LL_FORCE_INLINE void* ll_aligned_malloc(size_t size) { + LL_PROFILE_ZONE_SCOPED; + void* ret; if (LL_DEFAULT_HEAP_ALIGN % ALIGNMENT == 0) { - return malloc(size); + ret = malloc(size); + LL_PROFILE_ALLOC(ret, size); } else if (ALIGNMENT == 16) { - return ll_aligned_malloc_16(size); + ret = ll_aligned_malloc_16(size); } else if (ALIGNMENT == 32) { - return ll_aligned_malloc_32(size); + ret = ll_aligned_malloc_32(size); } else { - return ll_aligned_malloc_fallback(size, ALIGNMENT); + ret = ll_aligned_malloc_fallback(size, ALIGNMENT); } + return ret; } template LL_FORCE_INLINE void ll_aligned_free(void* ptr) { + LL_PROFILE_ZONE_SCOPED; if (ALIGNMENT == LL_DEFAULT_HEAP_ALIGN) { + LL_PROFILE_FREE(ptr); free(ptr); } else if (ALIGNMENT == 16) @@ -279,6 +309,7 @@ LL_FORCE_INLINE void ll_aligned_free(void* ptr) // inline void ll_memcpy_nonaliased_aligned_16(char* __restrict dst, const char* __restrict src, size_t bytes) { + LL_PROFILE_ZONE_SCOPED; assert(src != NULL); assert(dst != NULL); assert(bytes > 0); diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index e36f693dd3..ca60d23248 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -66,6 +66,8 @@ extern thread_local bool gProfilerEnabled; #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red + #define LL_PROFILE_ALLOC(ptr, size) TracyAlloc(ptr, size) + #define LL_PROFILE_FREE(ptr) TracyFree(ptr) #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_FAST_TIMER #define LL_PROFILER_FRAME_END @@ -81,11 +83,13 @@ extern thread_local bool gProfilerEnabled; #define LL_PROFILE_ZONE_ERR(name) (void)(name); // Not supported #define LL_PROFILE_ZONE_INFO(name) (void)(name); // Not supported #define LL_PROFILE_ZONE_WARN(name) (void)(name); // Not supported + #define LL_PROFILE_ALLOC(ptr, size) (void)(ptr); (void)(size); + #define LL_PROFILE_FREE(ptr) (void)(ptr); #endif #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER #define LL_PROFILER_FRAME_END FrameMark #define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ); gProfilerEnabled = true; - #define LL_RECORD_BLOCK_TIME(name) ZoneScoped const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); + #define LL_RECORD_BLOCK_TIME(name) ZoneNamedN(___tracy_scoped_zone, #name, true); const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__); #define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true ); #define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB #define LL_PROFILE_ZONE_SCOPED ZoneScoped @@ -96,6 +100,8 @@ extern thread_local bool gProfilerEnabled; #define LL_PROFILE_ZONE_ERR(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0XFF0000 ) // RGB yellow #define LL_PROFILE_ZONE_INFO(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0X00FFFF ) // RGB cyan #define LL_PROFILE_ZONE_WARN(name) LL_PROFILE_ZONE_NAMED_COLOR( name, 0x0FFFF00 ) // RGB red + #define LL_PROFILE_ALLOC(ptr, size) TracyAlloc(ptr, size) + #define LL_PROFILE_FREE(ptr) TracyFree(ptr) #endif #else #define LL_PROFILER_FRAME_END diff --git a/indra/llcommon/llsdparam.cpp b/indra/llcommon/llsdparam.cpp index 2e7b46f885..af4ccf25fd 100644 --- a/indra/llcommon/llsdparam.cpp +++ b/indra/llcommon/llsdparam.cpp @@ -37,8 +37,6 @@ static LLInitParam::Parser::parser_write_func_map_t sWriteFuncs; static LLInitParam::Parser::parser_inspect_func_map_t sInspectFuncs; static const LLSD NO_VALUE_MARKER; -LLTrace::BlockTimerStatHandle FTM_SD_PARAM_ADAPTOR("LLSD to LLInitParam conversion"); - // // LLParamSDParser // diff --git a/indra/llcommon/llsdparam.h b/indra/llcommon/llsdparam.h index 93910b70ae..82a623a8a0 100644 --- a/indra/llcommon/llsdparam.h +++ b/indra/llcommon/llsdparam.h @@ -110,7 +110,6 @@ private: }; -extern LL_COMMON_API LLTrace::BlockTimerStatHandle FTM_SD_PARAM_ADAPTOR; template class LLSDParamAdapter : public T { @@ -118,7 +117,7 @@ public: LLSDParamAdapter() {} LLSDParamAdapter(const LLSD& sd) { - LL_RECORD_BLOCK_TIME(FTM_SD_PARAM_ADAPTOR); + LL_PROFILE_ZONE_SCOPED; LLParamSDParser parser; // don't spam for implicit parsing of LLSD, as we want to allow arbitrary freeform data and ignore most of it bool parse_silently = true; diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 2e43a3cbed..10a8ecfedb 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -839,4 +839,30 @@ private: \ /* LLSINGLETON() is carefully implemented to permit exactly this */ \ LLSINGLETON_C11(DERIVED_CLASS) {} +// Relatively unsafe singleton implementation that is much faster +// and simpler than LLSingleton, but has no dependency tracking +// or inherent thread safety and requires manual invocation of +// createInstance before first use. +template +class LLSimpleton +{ +public: + static T* sInstance; + + static void createInstance() + { + llassert(sInstance == nullptr); + sInstance = new T(); + } + + static inline T* getInstance() { return sInstance; } + static inline T& instance() { return *getInstance(); } + static inline bool instanceExists() { return sInstance != nullptr; } + + static void deleteSingleton() { + delete sInstance; + sInstance = nullptr; + } +}; + #endif diff --git a/indra/llcommon/llstring.cpp b/indra/llcommon/llstring.cpp index 0290eea143..f6f9f97809 100644 --- a/indra/llcommon/llstring.cpp +++ b/indra/llcommon/llstring.cpp @@ -37,9 +37,6 @@ #include // for WideCharToMultiByte #endif -LLTrace::BlockTimerStatHandle FT_STRING_FORMAT("String Format"); - - std::string ll_safe_string(const char* in) { if(in) return std::string(in); @@ -1356,7 +1353,7 @@ bool LLStringUtil::formatDatetime(std::string& replacement, std::string token, template<> S32 LLStringUtil::format(std::string& s, const format_map_t& substitutions) { - LL_RECORD_BLOCK_TIME(FT_STRING_FORMAT); + LL_PROFILE_ZONE_SCOPED; S32 res = 0; std::string output; @@ -1429,7 +1426,7 @@ S32 LLStringUtil::format(std::string& s, const format_map_t& substitutions) template<> S32 LLStringUtil::format(std::string& s, const LLSD& substitutions) { - LL_RECORD_BLOCK_TIME(FT_STRING_FORMAT); + LL_PROFILE_ZONE_SCOPED; S32 res = 0; if (!substitutions.isMap()) diff --git a/indra/llcommon/llsys.cpp b/indra/llcommon/llsys.cpp index 6d5d043e8d..306ef05b6d 100644 --- a/indra/llcommon/llsys.cpp +++ b/indra/llcommon/llsys.cpp @@ -871,11 +871,9 @@ LLMemoryInfo& LLMemoryInfo::refresh() return *this; } -static LLTrace::BlockTimerStatHandle FTM_MEMINFO_LOAD_STATS("MemInfo Load Stats"); - LLSD LLMemoryInfo::loadStatsMap() { - LL_RECORD_BLOCK_TIME(FTM_MEMINFO_LOAD_STATS); + LL_PROFILE_ZONE_SCOPED; // This implementation is derived from stream() code (as of 2011-06-29). Stats stats; diff --git a/indra/llcommon/lltrace.cpp b/indra/llcommon/lltrace.cpp index 54079a4689..f59b207ded 100644 --- a/indra/llcommon/lltrace.cpp +++ b/indra/llcommon/lltrace.cpp @@ -61,6 +61,7 @@ TimeBlockTreeNode::TimeBlockTreeNode() void TimeBlockTreeNode::setParent( BlockTimerStatHandle* parent ) { + LL_PROFILE_ZONE_SCOPED; llassert_always(parent != mBlock); llassert_always(parent != NULL); diff --git a/indra/llcommon/lltrace.h b/indra/llcommon/lltrace.h index 0d0cd6f581..4051c558a4 100644 --- a/indra/llcommon/lltrace.h +++ b/indra/llcommon/lltrace.h @@ -227,6 +227,7 @@ public: void setName(const char* name) { + LL_PROFILE_ZONE_SCOPED; mName = name; setKey(name); } @@ -234,12 +235,14 @@ public: /*virtual*/ const char* getUnitLabel() const { return "KB"; } StatType& allocations() - { + { + LL_PROFILE_ZONE_SCOPED; return static_cast&>(*(StatType*)this); } StatType& deallocations() - { + { + LL_PROFILE_ZONE_SCOPED; return static_cast&>(*(StatType*)this); } }; @@ -261,6 +264,7 @@ struct MeasureMem { static size_t measureFootprint(const T& value) { + LL_PROFILE_ZONE_SCOPED; return sizeof(T) + value.getMemFootprint(); } }; @@ -270,6 +274,7 @@ struct MeasureMem { static size_t measureFootprint(const T& value) { + LL_PROFILE_ZONE_SCOPED; return U32Bytes(value).value(); } }; @@ -279,6 +284,7 @@ struct MeasureMem { static size_t measureFootprint(const T* value) { + LL_PROFILE_ZONE_SCOPED; if (!value) { return 0; @@ -323,6 +329,7 @@ struct MeasureMem, IS_MEM_TRACKABLE, IS_BYTES> { static size_t measureFootprint(const std::basic_string& value) { + LL_PROFILE_ZONE_SCOPED; return value.capacity() * sizeof(T); } }; @@ -331,6 +338,7 @@ struct MeasureMem, IS_MEM_TRACKABLE, IS_BYTES> template inline void claim_alloc(MemStatHandle& measurement, const T& value) { + LL_PROFILE_ZONE_SCOPED; #if LL_TRACE_ENABLED S32 size = MeasureMem::measureFootprint(value); if(size == 0) return; @@ -343,6 +351,7 @@ inline void claim_alloc(MemStatHandle& measurement, const T& value) template inline void disclaim_alloc(MemStatHandle& measurement, const T& value) { + LL_PROFILE_ZONE_SCOPED; #if LL_TRACE_ENABLED S32 size = MeasureMem::measureFootprint(value); if(size == 0) return; @@ -352,141 +361,6 @@ inline void disclaim_alloc(MemStatHandle& measurement, const T& value) #endif } -template -class MemTrackableNonVirtual -{ -public: - typedef void mem_trackable_tag_t; - - MemTrackableNonVirtual(const char* name) -#if LL_TRACE_ENABLED - : mMemFootprint(0) -#endif - { -#if LL_TRACE_ENABLED - static bool name_initialized = false; - if (!name_initialized) - { - name_initialized = true; - sMemStat.setName(name); - } -#endif - } - -#if LL_TRACE_ENABLED - ~MemTrackableNonVirtual() - { - disclaimMem(mMemFootprint); - } - - static MemStatHandle& getMemStatHandle() - { - return sMemStat; - } - - S32 getMemFootprint() const { return mMemFootprint; } -#endif - - void* operator new(size_t size) - { -#if LL_TRACE_ENABLED - claim_alloc(sMemStat, size); -#endif - return ll_aligned_malloc(size); - } - - template - static void* aligned_new(size_t size) - { -#if LL_TRACE_ENABLED - claim_alloc(sMemStat, size); -#endif - return ll_aligned_malloc(size); - } - - void operator delete(void* ptr, size_t size) - { -#if LL_TRACE_ENABLED - disclaim_alloc(sMemStat, size); -#endif - ll_aligned_free(ptr); - } - - template - static void aligned_delete(void* ptr, size_t size) - { -#if LL_TRACE_ENABLED - disclaim_alloc(sMemStat, size); -#endif - ll_aligned_free(ptr); - } - - void* operator new [](size_t size) - { -#if LL_TRACE_ENABLED - claim_alloc(sMemStat, size); -#endif - return ll_aligned_malloc(size); - } - - void operator delete[](void* ptr, size_t size) - { -#if LL_TRACE_ENABLED - disclaim_alloc(sMemStat, size); -#endif - ll_aligned_free(ptr); - } - - // claim memory associated with other objects/data as our own, adding to our calculated footprint - template - void claimMem(const CLAIM_T& value) const - { -#if LL_TRACE_ENABLED - S32 size = MeasureMem::measureFootprint(value); - claim_alloc(sMemStat, size); - mMemFootprint += size; -#endif - } - - // remove memory we had claimed from our calculated footprint - template - void disclaimMem(const CLAIM_T& value) const - { -#if LL_TRACE_ENABLED - S32 size = MeasureMem::measureFootprint(value); - disclaim_alloc(sMemStat, size); - mMemFootprint -= size; -#endif - } - -private: -#if LL_TRACE_ENABLED - // use signed values so that we can temporarily go negative - // and reconcile in destructor - // NB: this assumes that no single class is responsible for > 2GB of allocations - mutable S32 mMemFootprint; - - static MemStatHandle sMemStat; -#endif - -}; - -#if LL_TRACE_ENABLED -template -MemStatHandle MemTrackableNonVirtual::sMemStat(typeid(MemTrackableNonVirtual).name()); -#endif - -template -class MemTrackable : public MemTrackableNonVirtual -{ -public: - MemTrackable(const char* name) - : MemTrackableNonVirtual(name) - {} - - virtual ~MemTrackable() - {} -}; } #endif // LL_LLTRACE_H diff --git a/indra/llcommon/lltraceaccumulators.cpp b/indra/llcommon/lltraceaccumulators.cpp index b1c23c6fb7..8e9aaee0e6 100644 --- a/indra/llcommon/lltraceaccumulators.cpp +++ b/indra/llcommon/lltraceaccumulators.cpp @@ -41,6 +41,7 @@ extern MemStatHandle gTraceMemStat; AccumulatorBufferGroup::AccumulatorBufferGroup() { + LL_PROFILE_ZONE_SCOPED; claim_alloc(gTraceMemStat, mCounts.capacity() * sizeof(CountAccumulator)); claim_alloc(gTraceMemStat, mSamples.capacity() * sizeof(SampleAccumulator)); claim_alloc(gTraceMemStat, mEvents.capacity() * sizeof(EventAccumulator)); @@ -55,6 +56,7 @@ AccumulatorBufferGroup::AccumulatorBufferGroup(const AccumulatorBufferGroup& oth mStackTimers(other.mStackTimers), mMemStats(other.mMemStats) { + LL_PROFILE_ZONE_SCOPED; claim_alloc(gTraceMemStat, mCounts.capacity() * sizeof(CountAccumulator)); claim_alloc(gTraceMemStat, mSamples.capacity() * sizeof(SampleAccumulator)); claim_alloc(gTraceMemStat, mEvents.capacity() * sizeof(EventAccumulator)); @@ -64,6 +66,7 @@ AccumulatorBufferGroup::AccumulatorBufferGroup(const AccumulatorBufferGroup& oth AccumulatorBufferGroup::~AccumulatorBufferGroup() { + LL_PROFILE_ZONE_SCOPED; disclaim_alloc(gTraceMemStat, mCounts.capacity() * sizeof(CountAccumulator)); disclaim_alloc(gTraceMemStat, mSamples.capacity() * sizeof(SampleAccumulator)); disclaim_alloc(gTraceMemStat, mEvents.capacity() * sizeof(EventAccumulator)); @@ -73,6 +76,7 @@ AccumulatorBufferGroup::~AccumulatorBufferGroup() void AccumulatorBufferGroup::handOffTo(AccumulatorBufferGroup& other) { + LL_PROFILE_ZONE_SCOPED; other.mCounts.reset(&mCounts); other.mSamples.reset(&mSamples); other.mEvents.reset(&mEvents); @@ -82,6 +86,7 @@ void AccumulatorBufferGroup::handOffTo(AccumulatorBufferGroup& other) void AccumulatorBufferGroup::makeCurrent() { + LL_PROFILE_ZONE_SCOPED; mCounts.makeCurrent(); mSamples.makeCurrent(); mEvents.makeCurrent(); @@ -104,6 +109,7 @@ void AccumulatorBufferGroup::makeCurrent() //static void AccumulatorBufferGroup::clearCurrent() { + LL_PROFILE_ZONE_SCOPED; AccumulatorBuffer::clearCurrent(); AccumulatorBuffer::clearCurrent(); AccumulatorBuffer::clearCurrent(); @@ -118,6 +124,7 @@ bool AccumulatorBufferGroup::isCurrent() const void AccumulatorBufferGroup::append( const AccumulatorBufferGroup& other ) { + LL_PROFILE_ZONE_SCOPED; mCounts.addSamples(other.mCounts, SEQUENTIAL); mSamples.addSamples(other.mSamples, SEQUENTIAL); mEvents.addSamples(other.mEvents, SEQUENTIAL); @@ -127,6 +134,7 @@ void AccumulatorBufferGroup::append( const AccumulatorBufferGroup& other ) void AccumulatorBufferGroup::merge( const AccumulatorBufferGroup& other) { + LL_PROFILE_ZONE_SCOPED; mCounts.addSamples(other.mCounts, NON_SEQUENTIAL); mSamples.addSamples(other.mSamples, NON_SEQUENTIAL); mEvents.addSamples(other.mEvents, NON_SEQUENTIAL); @@ -137,6 +145,7 @@ void AccumulatorBufferGroup::merge( const AccumulatorBufferGroup& other) void AccumulatorBufferGroup::reset(AccumulatorBufferGroup* other) { + LL_PROFILE_ZONE_SCOPED; mCounts.reset(other ? &other->mCounts : NULL); mSamples.reset(other ? &other->mSamples : NULL); mEvents.reset(other ? &other->mEvents : NULL); @@ -146,6 +155,7 @@ void AccumulatorBufferGroup::reset(AccumulatorBufferGroup* other) void AccumulatorBufferGroup::sync() { + LL_PROFILE_ZONE_SCOPED; if (isCurrent()) { F64SecondsImplicit time_stamp = LLTimer::getTotalSeconds(); @@ -190,7 +200,7 @@ F64 SampleAccumulator::mergeSumsOfSquares(const SampleAccumulator& a, const Samp void SampleAccumulator::addSamples( const SampleAccumulator& other, EBufferAppendType append_type ) { - if (append_type == NON_SEQUENTIAL) + if (append_type == NON_SEQUENTIAL) { return; } @@ -289,7 +299,7 @@ void EventAccumulator::addSamples( const EventAccumulator& other, EBufferAppendT void EventAccumulator::reset( const EventAccumulator* other ) { - mNumSamples = 0; + mNumSamples = 0; mSum = 0; mMin = F32(NaN); mMax = F32(NaN); diff --git a/indra/llcommon/lltraceaccumulators.h b/indra/llcommon/lltraceaccumulators.h index 8eb5338a2a..b183fcd14a 100644 --- a/indra/llcommon/lltraceaccumulators.h +++ b/indra/llcommon/lltraceaccumulators.h @@ -66,6 +66,7 @@ namespace LLTrace : mStorageSize(0), mStorage(NULL) { + LL_PROFILE_ZONE_SCOPED; const AccumulatorBuffer& other = *getDefaultBuffer(); resize(sNextStorageSlot); for (S32 i = 0; i < sNextStorageSlot; i++) @@ -76,6 +77,7 @@ namespace LLTrace ~AccumulatorBuffer() { + LL_PROFILE_ZONE_SCOPED; if (isCurrent()) { LLThreadLocalSingletonPointer::setInstance(NULL); @@ -98,6 +100,7 @@ namespace LLTrace : mStorageSize(0), mStorage(NULL) { + LL_PROFILE_ZONE_SCOPED; resize(sNextStorageSlot); for (S32 i = 0; i < sNextStorageSlot; i++) { @@ -107,6 +110,7 @@ namespace LLTrace void addSamples(const AccumulatorBuffer& other, EBufferAppendType append_type) { + LL_PROFILE_ZONE_SCOPED; llassert(mStorageSize >= sNextStorageSlot && other.mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -116,6 +120,7 @@ namespace LLTrace void copyFrom(const AccumulatorBuffer& other) { + LL_PROFILE_ZONE_SCOPED; llassert(mStorageSize >= sNextStorageSlot && other.mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -125,6 +130,7 @@ namespace LLTrace void reset(const AccumulatorBuffer* other = NULL) { + LL_PROFILE_ZONE_SCOPED; llassert(mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -134,6 +140,7 @@ namespace LLTrace void sync(F64SecondsImplicit time_stamp) { + LL_PROFILE_ZONE_SCOPED; llassert(mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -153,12 +160,13 @@ namespace LLTrace static void clearCurrent() { - LLThreadLocalSingletonPointer::setInstance(NULL); + LLThreadLocalSingletonPointer::setInstance(NULL); } // NOTE: this is not thread-safe. We assume that slots are reserved in the main thread before any child threads are spawned size_t reserveSlot() { + LL_PROFILE_ZONE_SCOPED; size_t next_slot = sNextStorageSlot++; if (next_slot >= mStorageSize) { @@ -172,6 +180,7 @@ namespace LLTrace void resize(size_t new_size) { + LL_PROFILE_ZONE_SCOPED; if (new_size <= mStorageSize) return; ACCUMULATOR* old_storage = mStorage; @@ -212,6 +221,7 @@ namespace LLTrace static self_t* getDefaultBuffer() { + LL_PROFILE_ZONE_SCOPED; static bool sInitialized = false; if (!sInitialized) { @@ -326,6 +336,7 @@ namespace LLTrace void sample(F64 value) { + LL_PROFILE_ZONE_SCOPED; F64SecondsImplicit time_stamp = LLTimer::getTotalSeconds(); // store effect of last value @@ -444,9 +455,9 @@ namespace LLTrace S32 mNumSamples; }; - class TimeBlockAccumulator + class alignas(32) TimeBlockAccumulator { - public: + public: typedef F64Seconds value_t; static F64Seconds getDefaultValue() { return F64Seconds(0); } @@ -539,6 +550,7 @@ namespace LLTrace void addSamples(const MemAccumulator& other, EBufferAppendType append_type) { + LL_PROFILE_ZONE_SCOPED; mAllocations.addSamples(other.mAllocations, append_type); mDeallocations.addSamples(other.mDeallocations, append_type); @@ -557,6 +569,7 @@ namespace LLTrace void reset(const MemAccumulator* other) { + LL_PROFILE_ZONE_SCOPED; mSize.reset(other ? &other->mSize : NULL); mAllocations.reset(other ? &other->mAllocations : NULL); mDeallocations.reset(other ? &other->mDeallocations : NULL); diff --git a/indra/llcommon/lltracerecording.cpp b/indra/llcommon/lltracerecording.cpp index 3094b627a2..c72a64d086 100644 --- a/indra/llcommon/lltracerecording.cpp +++ b/indra/llcommon/lltracerecording.cpp @@ -50,6 +50,7 @@ Recording::Recording(EPlayState state) : mElapsedSeconds(0), mActiveBuffers(NULL) { + LL_PROFILE_ZONE_SCOPED; claim_alloc(gTraceMemStat, this); mBuffers = new AccumulatorBufferGroup(); claim_alloc(gTraceMemStat, mBuffers); @@ -59,12 +60,14 @@ Recording::Recording(EPlayState state) Recording::Recording( const Recording& other ) : mActiveBuffers(NULL) { + LL_PROFILE_ZONE_SCOPED; claim_alloc(gTraceMemStat, this); *this = other; } Recording& Recording::operator = (const Recording& other) { + LL_PROFILE_ZONE_SCOPED; // this will allow us to seamlessly start without affecting any data we've acquired from other setPlayState(PAUSED); @@ -85,6 +88,7 @@ Recording& Recording::operator = (const Recording& other) Recording::~Recording() { + LL_PROFILE_ZONE_SCOPED; disclaim_alloc(gTraceMemStat, this); disclaim_alloc(gTraceMemStat, mBuffers); @@ -103,6 +107,7 @@ void Recording::update() #if LL_TRACE_ENABLED if (isStarted()) { + LL_PROFILE_ZONE_SCOPED; mElapsedSeconds += mSamplingTimer.getElapsedTimeF64(); // must have @@ -123,6 +128,7 @@ void Recording::update() void Recording::handleReset() { + LL_PROFILE_ZONE_SCOPED; #if LL_TRACE_ENABLED mBuffers.write()->reset(); @@ -133,6 +139,7 @@ void Recording::handleReset() void Recording::handleStart() { + LL_PROFILE_ZONE_SCOPED; #if LL_TRACE_ENABLED mSamplingTimer.reset(); mBuffers.setStayUnique(true); @@ -144,6 +151,7 @@ void Recording::handleStart() void Recording::handleStop() { + LL_PROFILE_ZONE_SCOPED; #if LL_TRACE_ENABLED mElapsedSeconds += mSamplingTimer.getElapsedTimeF64(); // must have thread recorder running on this thread @@ -273,7 +281,7 @@ F64Kilobytes Recording::getMean(const StatType& stat) F64Kilobytes Recording::getMax(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return F64Bytes(llmax(accumulator.mSize.getMax(), active_accumulator && active_accumulator->mSize.hasValue() ? active_accumulator->mSize.getMax() : F32_MIN)); @@ -281,7 +289,7 @@ F64Kilobytes Recording::getMax(const StatType& stat) F64Kilobytes Recording::getStandardDeviation(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; if (active_accumulator && active_accumulator->hasValue()) @@ -297,7 +305,7 @@ F64Kilobytes Recording::getStandardDeviation(const StatType& sta F64Kilobytes Recording::getLastValue(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return F64Bytes(active_accumulator ? active_accumulator->mSize.getLastValue() : accumulator.mSize.getLastValue()); @@ -305,7 +313,7 @@ F64Kilobytes Recording::getLastValue(const StatType& stat) bool Recording::hasValue(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return accumulator.mAllocations.hasValue() || (active_accumulator ? active_accumulator->mAllocations.hasValue() : false); @@ -313,7 +321,7 @@ bool Recording::hasValue(const StatType& stat) F64Kilobytes Recording::getSum(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return F64Bytes(accumulator.mAllocations.getSum() + (active_accumulator ? active_accumulator->mAllocations.getSum() : 0)); @@ -321,7 +329,7 @@ F64Kilobytes Recording::getSum(const StatType& F64Kilobytes Recording::getPerSec(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return F64Bytes((accumulator.mAllocations.getSum() + (active_accumulator ? active_accumulator->mAllocations.getSum() : 0)) / mElapsedSeconds.value()); @@ -329,7 +337,7 @@ F64Kilobytes Recording::getPerSec(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return accumulator.mAllocations.getSampleCount() + (active_accumulator ? active_accumulator->mAllocations.getSampleCount() : 0); @@ -337,7 +345,7 @@ S32 Recording::getSampleCount(const StatType& s bool Recording::hasValue(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return accumulator.mDeallocations.hasValue() || (active_accumulator ? active_accumulator->mDeallocations.hasValue() : false); @@ -346,7 +354,7 @@ bool Recording::hasValue(const StatType& stat F64Kilobytes Recording::getSum(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return F64Bytes(accumulator.mDeallocations.getSum() + (active_accumulator ? active_accumulator->mDeallocations.getSum() : 0)); @@ -354,7 +362,7 @@ F64Kilobytes Recording::getSum(const StatType F64Kilobytes Recording::getPerSec(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return F64Bytes((accumulator.mDeallocations.getSum() + (active_accumulator ? active_accumulator->mDeallocations.getSum() : 0)) / mElapsedSeconds.value()); @@ -362,7 +370,7 @@ F64Kilobytes Recording::getPerSec(const StatType& stat) { - update(); + update(); const MemAccumulator& accumulator = mBuffers->mMemStats[stat.getIndex()]; const MemAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mMemStats[stat.getIndex()] : NULL; return accumulator.mDeallocations.getSampleCount() + (active_accumulator ? active_accumulator->mDeallocations.getSampleCount() : 0); @@ -370,7 +378,7 @@ S32 Recording::getSampleCount(const StatType& bool Recording::hasValue(const StatType& stat) { - update(); + update(); const CountAccumulator& accumulator = mBuffers->mCounts[stat.getIndex()]; const CountAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mCounts[stat.getIndex()] : NULL; return accumulator.hasValue() || (active_accumulator ? active_accumulator->hasValue() : false); @@ -378,7 +386,7 @@ bool Recording::hasValue(const StatType& stat) F64 Recording::getSum(const StatType& stat) { - update(); + update(); const CountAccumulator& accumulator = mBuffers->mCounts[stat.getIndex()]; const CountAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mCounts[stat.getIndex()] : NULL; return accumulator.getSum() + (active_accumulator ? active_accumulator->getSum() : 0); @@ -386,7 +394,7 @@ F64 Recording::getSum(const StatType& stat) F64 Recording::getPerSec( const StatType& stat ) { - update(); + update(); const CountAccumulator& accumulator = mBuffers->mCounts[stat.getIndex()]; const CountAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mCounts[stat.getIndex()] : NULL; F64 sum = accumulator.getSum() + (active_accumulator ? active_accumulator->getSum() : 0); @@ -395,7 +403,7 @@ F64 Recording::getPerSec( const StatType& stat ) S32 Recording::getSampleCount( const StatType& stat ) { - update(); + update(); const CountAccumulator& accumulator = mBuffers->mCounts[stat.getIndex()]; const CountAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mCounts[stat.getIndex()] : NULL; return accumulator.getSampleCount() + (active_accumulator ? active_accumulator->getSampleCount() : 0); @@ -403,7 +411,7 @@ S32 Recording::getSampleCount( const StatType& stat ) bool Recording::hasValue(const StatType& stat) { - update(); + update(); const SampleAccumulator& accumulator = mBuffers->mSamples[stat.getIndex()]; const SampleAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mSamples[stat.getIndex()] : NULL; return accumulator.hasValue() || (active_accumulator && active_accumulator->hasValue()); @@ -411,7 +419,7 @@ bool Recording::hasValue(const StatType& stat) F64 Recording::getMin( const StatType& stat ) { - update(); + update(); const SampleAccumulator& accumulator = mBuffers->mSamples[stat.getIndex()]; const SampleAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mSamples[stat.getIndex()] : NULL; return llmin(accumulator.getMin(), active_accumulator && active_accumulator->hasValue() ? active_accumulator->getMin() : F32_MAX); @@ -419,7 +427,7 @@ F64 Recording::getMin( const StatType& stat ) F64 Recording::getMax( const StatType& stat ) { - update(); + update(); const SampleAccumulator& accumulator = mBuffers->mSamples[stat.getIndex()]; const SampleAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mSamples[stat.getIndex()] : NULL; return llmax(accumulator.getMax(), active_accumulator && active_accumulator->hasValue() ? active_accumulator->getMax() : F32_MIN); @@ -427,7 +435,7 @@ F64 Recording::getMax( const StatType& stat ) F64 Recording::getMean( const StatType& stat ) { - update(); + update(); const SampleAccumulator& accumulator = mBuffers->mSamples[stat.getIndex()]; const SampleAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mSamples[stat.getIndex()] : NULL; if (active_accumulator && active_accumulator->hasValue()) @@ -448,7 +456,7 @@ F64 Recording::getMean( const StatType& stat ) F64 Recording::getStandardDeviation( const StatType& stat ) { - update(); + update(); const SampleAccumulator& accumulator = mBuffers->mSamples[stat.getIndex()]; const SampleAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mSamples[stat.getIndex()] : NULL; @@ -465,7 +473,7 @@ F64 Recording::getStandardDeviation( const StatType& stat ) F64 Recording::getLastValue( const StatType& stat ) { - update(); + update(); const SampleAccumulator& accumulator = mBuffers->mSamples[stat.getIndex()]; const SampleAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mSamples[stat.getIndex()] : NULL; return (active_accumulator && active_accumulator->hasValue() ? active_accumulator->getLastValue() : accumulator.getLastValue()); @@ -473,7 +481,7 @@ F64 Recording::getLastValue( const StatType& stat ) S32 Recording::getSampleCount( const StatType& stat ) { - update(); + update(); const SampleAccumulator& accumulator = mBuffers->mSamples[stat.getIndex()]; const SampleAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mSamples[stat.getIndex()] : NULL; return accumulator.getSampleCount() + (active_accumulator && active_accumulator->hasValue() ? active_accumulator->getSampleCount() : 0); @@ -481,7 +489,7 @@ S32 Recording::getSampleCount( const StatType& stat ) bool Recording::hasValue(const StatType& stat) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; return accumulator.hasValue() || (active_accumulator && active_accumulator->hasValue()); @@ -489,7 +497,7 @@ bool Recording::hasValue(const StatType& stat) F64 Recording::getSum( const StatType& stat) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; return (F64)(accumulator.getSum() + (active_accumulator && active_accumulator->hasValue() ? active_accumulator->getSum() : 0)); @@ -497,7 +505,7 @@ F64 Recording::getSum( const StatType& stat) F64 Recording::getMin( const StatType& stat ) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; return llmin(accumulator.getMin(), active_accumulator && active_accumulator->hasValue() ? active_accumulator->getMin() : F32_MAX); @@ -505,7 +513,7 @@ F64 Recording::getMin( const StatType& stat ) F64 Recording::getMax( const StatType& stat ) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; return llmax(accumulator.getMax(), active_accumulator && active_accumulator->hasValue() ? active_accumulator->getMax() : F32_MIN); @@ -513,7 +521,7 @@ F64 Recording::getMax( const StatType& stat ) F64 Recording::getMean( const StatType& stat ) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; if (active_accumulator && active_accumulator->hasValue()) @@ -534,7 +542,7 @@ F64 Recording::getMean( const StatType& stat ) F64 Recording::getStandardDeviation( const StatType& stat ) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; @@ -551,7 +559,7 @@ F64 Recording::getStandardDeviation( const StatType& stat ) F64 Recording::getLastValue( const StatType& stat ) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; return active_accumulator ? active_accumulator->getLastValue() : accumulator.getLastValue(); @@ -559,7 +567,7 @@ F64 Recording::getLastValue( const StatType& stat ) S32 Recording::getSampleCount( const StatType& stat ) { - update(); + update(); const EventAccumulator& accumulator = mBuffers->mEvents[stat.getIndex()]; const EventAccumulator* active_accumulator = mActiveBuffers ? &mActiveBuffers->mEvents[stat.getIndex()] : NULL; return accumulator.getSampleCount() + (active_accumulator ? active_accumulator->getSampleCount() : 0); @@ -575,17 +583,20 @@ PeriodicRecording::PeriodicRecording( S32 num_periods, EPlayState state) mNumRecordedPeriods(0), mRecordingPeriods(num_periods ? num_periods : 1) { + LL_PROFILE_ZONE_SCOPED; setPlayState(state); claim_alloc(gTraceMemStat, this); } PeriodicRecording::~PeriodicRecording() { + LL_PROFILE_ZONE_SCOPED; disclaim_alloc(gTraceMemStat, this); } void PeriodicRecording::nextPeriod() { + LL_PROFILE_ZONE_SCOPED; if (mAutoResize) { mRecordingPeriods.push_back(Recording()); @@ -600,6 +611,7 @@ void PeriodicRecording::nextPeriod() void PeriodicRecording::appendRecording(Recording& recording) { + LL_PROFILE_ZONE_SCOPED; getCurRecording().appendRecording(recording); nextPeriod(); } @@ -607,6 +619,7 @@ void PeriodicRecording::appendRecording(Recording& recording) void PeriodicRecording::appendPeriodicRecording( PeriodicRecording& other ) { + LL_PROFILE_ZONE_SCOPED; if (other.mRecordingPeriods.empty()) return; getCurRecording().update(); @@ -680,6 +693,7 @@ void PeriodicRecording::appendPeriodicRecording( PeriodicRecording& other ) F64Seconds PeriodicRecording::getDuration() const { + LL_PROFILE_ZONE_SCOPED; F64Seconds duration; S32 num_periods = mRecordingPeriods.size(); for (S32 i = 1; i <= num_periods; i++) @@ -693,6 +707,7 @@ F64Seconds PeriodicRecording::getDuration() const LLTrace::Recording PeriodicRecording::snapshotCurRecording() const { + LL_PROFILE_ZONE_SCOPED; Recording recording_copy(getCurRecording()); recording_copy.stop(); return recording_copy; @@ -735,16 +750,19 @@ const Recording& PeriodicRecording::getPrevRecording( S32 offset ) const void PeriodicRecording::handleStart() { + LL_PROFILE_ZONE_SCOPED; getCurRecording().start(); } void PeriodicRecording::handleStop() { + LL_PROFILE_ZONE_SCOPED; getCurRecording().pause(); } void PeriodicRecording::handleReset() { + LL_PROFILE_ZONE_SCOPED; getCurRecording().stop(); if (mAutoResize) @@ -768,11 +786,13 @@ void PeriodicRecording::handleReset() void PeriodicRecording::handleSplitTo(PeriodicRecording& other) { + LL_PROFILE_ZONE_SCOPED; getCurRecording().splitTo(other.getCurRecording()); } F64 PeriodicRecording::getPeriodMin( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -794,6 +814,7 @@ F64 PeriodicRecording::getPeriodMin( const StatType& stat, S32 F64 PeriodicRecording::getPeriodMax( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -816,6 +837,7 @@ F64 PeriodicRecording::getPeriodMax( const StatType& stat, S32 // calculates means using aggregates per period F64 PeriodicRecording::getPeriodMean( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 mean = 0; @@ -839,6 +861,7 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S3 F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 period_mean = getPeriodMean(stat, num_periods); @@ -863,6 +886,7 @@ F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -884,6 +908,7 @@ F64 PeriodicRecording::getPeriodMin( const StatType& stat, S3 F64 PeriodicRecording::getPeriodMax(const StatType& stat, S32 num_periods /*= S32_MAX*/) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -906,6 +931,7 @@ F64 PeriodicRecording::getPeriodMax(const StatType& stat, S32 F64 PeriodicRecording::getPeriodMean( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); S32 valid_period_count = 0; @@ -928,6 +954,7 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 period_mean = getPeriodMean(stat, num_periods); @@ -953,6 +980,7 @@ F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes min_val(std::numeric_limits::max()); @@ -972,6 +1000,7 @@ F64Kilobytes PeriodicRecording::getPeriodMin(const MemStatHandle& stat, S32 num_ F64Kilobytes PeriodicRecording::getPeriodMax(const StatType& stat, S32 num_periods /*= S32_MAX*/) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes max_val(0.0); @@ -991,6 +1020,7 @@ F64Kilobytes PeriodicRecording::getPeriodMax(const MemStatHandle& stat, S32 num_ F64Kilobytes PeriodicRecording::getPeriodMean( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes mean(0); @@ -1011,6 +1041,7 @@ F64Kilobytes PeriodicRecording::getPeriodMean(const MemStatHandle& stat, S32 num F64Kilobytes PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes period_mean = getPeriodMean(stat, num_periods); @@ -1044,6 +1075,7 @@ F64Kilobytes PeriodicRecording::getPeriodStandardDeviation(const MemStatHandle& void ExtendableRecording::extend() { + LL_PROFILE_ZONE_SCOPED; // push the data back to accepted recording mAcceptedRecording.appendRecording(mPotentialRecording); // flush data, so we can start from scratch @@ -1052,22 +1084,26 @@ void ExtendableRecording::extend() void ExtendableRecording::handleStart() { + LL_PROFILE_ZONE_SCOPED; mPotentialRecording.start(); } void ExtendableRecording::handleStop() { + LL_PROFILE_ZONE_SCOPED; mPotentialRecording.pause(); } void ExtendableRecording::handleReset() { + LL_PROFILE_ZONE_SCOPED; mAcceptedRecording.reset(); mPotentialRecording.reset(); } void ExtendableRecording::handleSplitTo(ExtendableRecording& other) { + LL_PROFILE_ZONE_SCOPED; mPotentialRecording.splitTo(other.mPotentialRecording); } @@ -1084,6 +1120,7 @@ ExtendablePeriodicRecording::ExtendablePeriodicRecording() void ExtendablePeriodicRecording::extend() { + LL_PROFILE_ZONE_SCOPED; // push the data back to accepted recording mAcceptedRecording.appendPeriodicRecording(mPotentialRecording); // flush data, so we can start from scratch @@ -1093,22 +1130,26 @@ void ExtendablePeriodicRecording::extend() void ExtendablePeriodicRecording::handleStart() { + LL_PROFILE_ZONE_SCOPED; mPotentialRecording.start(); } void ExtendablePeriodicRecording::handleStop() { + LL_PROFILE_ZONE_SCOPED; mPotentialRecording.pause(); } void ExtendablePeriodicRecording::handleReset() { + LL_PROFILE_ZONE_SCOPED; mAcceptedRecording.reset(); mPotentialRecording.reset(); } void ExtendablePeriodicRecording::handleSplitTo(ExtendablePeriodicRecording& other) { + LL_PROFILE_ZONE_SCOPED; mPotentialRecording.splitTo(other.mPotentialRecording); } @@ -1123,6 +1164,7 @@ PeriodicRecording& get_frame_recording() void LLStopWatchControlsMixinCommon::start() { + LL_PROFILE_ZONE_SCOPED; switch (mPlayState) { case STOPPED: @@ -1144,6 +1186,7 @@ void LLStopWatchControlsMixinCommon::start() void LLStopWatchControlsMixinCommon::stop() { + LL_PROFILE_ZONE_SCOPED; switch (mPlayState) { case STOPPED: @@ -1163,6 +1206,7 @@ void LLStopWatchControlsMixinCommon::stop() void LLStopWatchControlsMixinCommon::pause() { + LL_PROFILE_ZONE_SCOPED; switch (mPlayState) { case STOPPED: @@ -1182,6 +1226,7 @@ void LLStopWatchControlsMixinCommon::pause() void LLStopWatchControlsMixinCommon::unpause() { + LL_PROFILE_ZONE_SCOPED; switch (mPlayState) { case STOPPED: @@ -1201,6 +1246,7 @@ void LLStopWatchControlsMixinCommon::unpause() void LLStopWatchControlsMixinCommon::resume() { + LL_PROFILE_ZONE_SCOPED; switch (mPlayState) { case STOPPED: @@ -1221,6 +1267,7 @@ void LLStopWatchControlsMixinCommon::resume() void LLStopWatchControlsMixinCommon::restart() { + LL_PROFILE_ZONE_SCOPED; switch (mPlayState) { case STOPPED: @@ -1244,11 +1291,13 @@ void LLStopWatchControlsMixinCommon::restart() void LLStopWatchControlsMixinCommon::reset() { + LL_PROFILE_ZONE_SCOPED; handleReset(); } void LLStopWatchControlsMixinCommon::setPlayState( EPlayState state ) { + LL_PROFILE_ZONE_SCOPED; switch(state) { case STOPPED: diff --git a/indra/llcommon/lltracerecording.h b/indra/llcommon/lltracerecording.h index d0b4a842a6..6715104613 100644 --- a/indra/llcommon/lltracerecording.h +++ b/indra/llcommon/lltracerecording.h @@ -355,6 +355,7 @@ namespace LLTrace template S32 getSampleCount(const StatType& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); S32 num_samples = 0; @@ -374,6 +375,7 @@ namespace LLTrace template typename T::value_t getPeriodMin(const StatType& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -396,6 +398,7 @@ namespace LLTrace template T getPeriodMin(const CountStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return T(getPeriodMin(static_cast&>(stat), num_periods)); } @@ -403,6 +406,7 @@ namespace LLTrace template T getPeriodMin(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return T(getPeriodMin(static_cast&>(stat), num_periods)); } @@ -410,6 +414,7 @@ namespace LLTrace template T getPeriodMin(const EventStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return T(getPeriodMin(static_cast&>(stat), num_periods)); } @@ -419,6 +424,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMinPerSec(const StatType& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); typename RelatedTypes::fractional_t min_val(std::numeric_limits::max()); @@ -433,6 +439,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMinPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodMinPerSec(static_cast&>(stat), num_periods)); } @@ -444,6 +451,7 @@ namespace LLTrace template typename T::value_t getPeriodMax(const StatType& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -466,6 +474,7 @@ namespace LLTrace template T getPeriodMax(const CountStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return T(getPeriodMax(static_cast&>(stat), num_periods)); } @@ -473,6 +482,7 @@ namespace LLTrace template T getPeriodMax(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return T(getPeriodMax(static_cast&>(stat), num_periods)); } @@ -480,6 +490,7 @@ namespace LLTrace template T getPeriodMax(const EventStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return T(getPeriodMax(static_cast&>(stat), num_periods)); } @@ -489,6 +500,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMaxPerSec(const StatType& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 max_val = std::numeric_limits::min(); @@ -503,6 +515,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMaxPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodMaxPerSec(static_cast&>(stat), num_periods)); } @@ -514,6 +527,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMean(const StatType& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); typename RelatedTypes::fractional_t mean(0); @@ -534,12 +548,14 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMean(const CountStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodMean(static_cast&>(stat), num_periods)); } F64 getPeriodMean(const StatType& stat, S32 num_periods = S32_MAX); template typename RelatedTypes::fractional_t getPeriodMean(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodMean(static_cast&>(stat), num_periods)); } @@ -547,6 +563,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMean(const EventStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodMean(static_cast&>(stat), num_periods)); } @@ -556,6 +573,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMeanPerSec(const StatType& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; num_periods = llmin(num_periods, getNumRecordedPeriods()); typename RelatedTypes::fractional_t mean = 0; @@ -577,6 +595,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMeanPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodMeanPerSec(static_cast&>(stat), num_periods)); } @@ -589,6 +608,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodStandardDeviation(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodStandardDeviation(static_cast&>(stat), num_periods)); } @@ -596,6 +616,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodStandardDeviation(const EventStatHandle& stat, S32 num_periods = S32_MAX) { + LL_PROFILE_ZONE_SCOPED; return typename RelatedTypes::fractional_t(getPeriodStandardDeviation(static_cast&>(stat), num_periods)); } diff --git a/indra/llcommon/lltracethreadrecorder.cpp b/indra/llcommon/lltracethreadrecorder.cpp index 025dc57044..7ae1e72784 100644 --- a/indra/llcommon/lltracethreadrecorder.cpp +++ b/indra/llcommon/lltracethreadrecorder.cpp @@ -274,12 +274,10 @@ void ThreadRecorder::pushToParent() } -static LLTrace::BlockTimerStatHandle FTM_PULL_TRACE_DATA_FROM_CHILDREN("Pull child thread trace data"); - void ThreadRecorder::pullFromChildren() { #if LL_TRACE_ENABLED - LL_RECORD_BLOCK_TIME(FTM_PULL_TRACE_DATA_FROM_CHILDREN); + LL_PROFILE_ZONE_SCOPED; if (mActiveRecordings.empty()) return; { LLMutexLock lock(&mChildListMutex); diff --git a/indra/llimage/llimage.cpp b/indra/llimage/llimage.cpp index aed8943439..5c49ec02ea 100644 --- a/indra/llimage/llimage.cpp +++ b/indra/llimage/llimage.cpp @@ -623,8 +623,7 @@ void LLImage::setLastError(const std::string& message) //--------------------------------------------------------------------------- LLImageBase::LLImageBase() -: LLTrace::MemTrackable("LLImage"), - mData(NULL), +: mData(NULL), mDataSize(0), mWidth(0), mHeight(0), @@ -673,7 +672,6 @@ void LLImageBase::sanityCheck() void LLImageBase::deleteData() { ll_aligned_free_16(mData); - disclaimMem(mDataSize); mDataSize = 0; mData = NULL; } @@ -731,7 +729,6 @@ U8* LLImageBase::allocateData(S32 size) } } mDataSize = size; - claimMem(mDataSize); return mData; } @@ -752,9 +749,7 @@ U8* LLImageBase::reallocateData(S32 size) ll_aligned_free_16(mData) ; } mData = new_datap; - disclaimMem(mDataSize); mDataSize = size; - claimMem(mDataSize); mBadBufferAllocation = false; return mData; } @@ -2258,9 +2253,7 @@ void LLImageBase::setDataAndSize(U8 *data, S32 size) { ll_assert_aligned(data, 16); mData = data; - disclaimMem(mDataSize); mDataSize = size; - claimMem(mDataSize); } //static diff --git a/indra/llimage/llimage.h b/indra/llimage/llimage.h index f66b1666d7..354926ee58 100644 --- a/indra/llimage/llimage.h +++ b/indra/llimage/llimage.h @@ -112,8 +112,7 @@ protected: // Image base class class LLImageBase -: public LLThreadSafeRefCount, - public LLTrace::MemTrackable +: public LLThreadSafeRefCount { protected: virtual ~LLImageBase(); diff --git a/indra/llimage/llimagej2c.cpp b/indra/llimage/llimagej2c.cpp index 4bff21610f..e1809dbe59 100644 --- a/indra/llimage/llimagej2c.cpp +++ b/indra/llimage/llimagej2c.cpp @@ -60,7 +60,6 @@ LLImageJ2C::LLImageJ2C() : LLImageFormatted(IMG_CODEC_J2C), mAreaUsedForDataSizeCalcs(0) { mImpl.reset(fallbackCreateLLImageJ2CImpl()); - claimMem(mImpl); // Clear data size table for( S32 i = 0; i <= MAX_DISCARD_LEVEL; i++) diff --git a/indra/llimage/llimagejpeg.cpp b/indra/llimage/llimagejpeg.cpp index 62638fa16c..32a5472ec8 100644 --- a/indra/llimage/llimagejpeg.cpp +++ b/indra/llimage/llimagejpeg.cpp @@ -393,9 +393,7 @@ boolean LLImageJPEG::encodeEmptyOutputBuffer( j_compress_ptr cinfo ) cinfo->dest->next_output_byte = self->mOutputBuffer + self->mOutputBufferSize; cinfo->dest->free_in_buffer = self->mOutputBufferSize; - self->disclaimMem(self->mOutputBufferSize); self->mOutputBufferSize = new_buffer_size; - self->claimMem(new_buffer_size); return true; } @@ -501,13 +499,10 @@ bool LLImageJPEG::encode( const LLImageRaw* raw_image, F32 encode_time ) // Allocate a temporary buffer big enough to hold the entire compressed image (and then some) // (Note: we make it bigger in emptyOutputBuffer() if we need to) delete[] mOutputBuffer; - disclaimMem(mOutputBufferSize); mOutputBufferSize = getWidth() * getHeight() * getComponents() + 1024; - claimMem(mOutputBufferSize); mOutputBuffer = new(std::nothrow) U8[ mOutputBufferSize ]; if (mOutputBuffer == NULL) { - disclaimMem(mOutputBufferSize); mOutputBufferSize = 0; setLastError("Failed to allocate output buffer"); return false; @@ -547,7 +542,6 @@ bool LLImageJPEG::encode( const LLImageRaw* raw_image, F32 encode_time ) jpeg_destroy_compress(&cinfo); delete[] mOutputBuffer; mOutputBuffer = NULL; - disclaimMem(mOutputBufferSize); mOutputBufferSize = 0; return false; } @@ -650,7 +644,6 @@ bool LLImageJPEG::encode( const LLImageRaw* raw_image, F32 encode_time ) // After finish_compress, we can release the temp output buffer. delete[] mOutputBuffer; mOutputBuffer = NULL; - disclaimMem(mOutputBufferSize); mOutputBufferSize = 0; //////////////////////////////////////// @@ -663,7 +656,6 @@ bool LLImageJPEG::encode( const LLImageRaw* raw_image, F32 encode_time ) jpeg_destroy_compress(&cinfo); delete[] mOutputBuffer; mOutputBuffer = NULL; - disclaimMem(mOutputBufferSize); mOutputBufferSize = 0; return false; } diff --git a/indra/llimage/tests/llimageworker_test.cpp b/indra/llimage/tests/llimageworker_test.cpp index 51c5c63556..9011ac615c 100644 --- a/indra/llimage/tests/llimageworker_test.cpp +++ b/indra/llimage/tests/llimageworker_test.cpp @@ -45,8 +45,7 @@ // * A simulator for a class can be implemented here. Please comment and document thoroughly. LLImageBase::LLImageBase() -: LLTrace::MemTrackable("LLImageBase"), -mData(NULL), +: mData(NULL), mDataSize(0), mWidth(0), mHeight(0), diff --git a/indra/llinventory/llinventory.cpp b/indra/llinventory/llinventory.cpp index 18bc1b5a91..81261f0767 100644 --- a/indra/llinventory/llinventory.cpp +++ b/indra/llinventory/llinventory.cpp @@ -74,20 +74,17 @@ LLInventoryObject::LLInventoryObject(const LLUUID& uuid, const LLUUID& parent_uuid, LLAssetType::EType type, const std::string& name) -: LLTrace::MemTrackable("LLInventoryObject"), - mUUID(uuid), +: mUUID(uuid), mParentUUID(parent_uuid), mType(type), mName(name), mCreationDate(0) { - claimMem(mName); correctInventoryName(mName); } LLInventoryObject::LLInventoryObject() -: LLTrace::MemTrackable("LLInventoryObject"), - mType(LLAssetType::AT_NONE), +: mType(LLAssetType::AT_NONE), mCreationDate(0) { } @@ -101,9 +98,7 @@ void LLInventoryObject::copyObject(const LLInventoryObject* other) mUUID = other->mUUID; mParentUUID = other->mParentUUID; mType = other->mType; - disclaimMem(mName); mName = other->mName; - claimMem(mName); } const LLUUID& LLInventoryObject::getUUID() const @@ -156,9 +151,7 @@ void LLInventoryObject::rename(const std::string& n) correctInventoryName(new_name); if( !new_name.empty() && new_name != mName ) { - disclaimMem(mName); mName = new_name; - claimMem(mName); } } @@ -311,7 +304,6 @@ LLInventoryItem::LLInventoryItem(const LLUUID& uuid, LLStringUtil::replaceNonstandardASCII(mDescription, ' '); LLStringUtil::replaceChar(mDescription, '|', ' '); - claimMem(mDescription); mPermissions.initMasks(inv_type); } @@ -344,9 +336,7 @@ void LLInventoryItem::copyItem(const LLInventoryItem* other) copyObject(other); mPermissions = other->mPermissions; mAssetUUID = other->mAssetUUID; - disclaimMem(mDescription); mDescription = other->mDescription; - claimMem(mDescription); mSaleInfo = other->mSaleInfo; mInventoryType = other->mInventoryType; mFlags = other->mFlags; @@ -426,9 +416,7 @@ void LLInventoryItem::setDescription(const std::string& d) LLInventoryItem::correctInventoryDescription(new_desc); if( new_desc != mDescription ) { - disclaimMem(mDescription); mDescription = new_desc; - claimMem(mDescription); } } @@ -708,10 +696,8 @@ BOOL LLInventoryItem::importLegacyStream(std::istream& input_stream) valuestr[0] = '\000'; } - disclaimMem(mDescription); mDescription.assign(valuestr); LLStringUtil::replaceNonstandardASCII(mDescription, ' '); - claimMem(mDescription); /* TODO -- ask Ian about this code const char *donkey = mDescription.c_str(); if (donkey[0] == '|') @@ -840,11 +826,9 @@ void LLInventoryItem::asLLSD( LLSD& sd ) const sd[INV_CREATION_DATE_LABEL] = (S32) mCreationDate; } -LLTrace::BlockTimerStatHandle FTM_INVENTORY_SD_DESERIALIZE("Inventory SD Deserialize"); - bool LLInventoryItem::fromLLSD(const LLSD& sd, bool is_new) { - LL_RECORD_BLOCK_TIME(FTM_INVENTORY_SD_DESERIALIZE); + LL_PROFILE_ZONE_SCOPED; if (is_new) { // If we're adding LLSD to an existing object, need avoid @@ -961,10 +945,8 @@ bool LLInventoryItem::fromLLSD(const LLSD& sd, bool is_new) w = INV_DESC_LABEL; if (sd.has(w)) { - disclaimMem(mDescription); mDescription = sd[w].asString(); LLStringUtil::replaceNonstandardASCII(mDescription, ' '); - claimMem(mDescription); } w = INV_CREATION_DATE_LABEL; if (sd.has(w)) diff --git a/indra/llinventory/llinventory.h b/indra/llinventory/llinventory.h index 0f336a072f..7d9f9704f1 100644 --- a/indra/llinventory/llinventory.h +++ b/indra/llinventory/llinventory.h @@ -44,7 +44,7 @@ class LLMessageSystem; // Base class for anything in the user's inventory. Handles the common code // between items and categories. //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -class LLInventoryObject : public LLRefCount, public LLTrace::MemTrackable +class LLInventoryObject : public LLRefCount { public: typedef std::list > object_list_t; diff --git a/indra/llinventory/llsettingsdaycycle.cpp b/indra/llinventory/llsettingsdaycycle.cpp index a687fd840d..b390ea5b1e 100644 --- a/indra/llinventory/llsettingsdaycycle.cpp +++ b/indra/llinventory/llsettingsdaycycle.cpp @@ -41,9 +41,6 @@ //========================================================================= namespace { - LLTrace::BlockTimerStatHandle FTM_BLEND_WATERVALUES("Blending Water Environment Day"); - LLTrace::BlockTimerStatHandle FTM_UPDATE_WATERVALUES("Update Water Environment Day"); - template inline T get_wrapping_distance(T begin, T end) { diff --git a/indra/llinventory/llsettingssky.cpp b/indra/llinventory/llsettingssky.cpp index 82c67a1066..23f9c8e9d7 100644 --- a/indra/llinventory/llsettingssky.cpp +++ b/indra/llinventory/llsettingssky.cpp @@ -66,11 +66,6 @@ namespace { } } -static LLTrace::BlockTimerStatHandle FTM_BLEND_SKYVALUES("Blending Sky Environment"); -static LLTrace::BlockTimerStatHandle FTM_RECALCULATE_SKYVALUES("Recalculate Sky"); -static LLTrace::BlockTimerStatHandle FTM_RECALCULATE_BODIES("Recalculate Heavenly Bodies"); -static LLTrace::BlockTimerStatHandle FTM_RECALCULATE_LIGHTING("Recalculate Lighting"); - //========================================================================= const std::string LLSettingsSky::SETTING_AMBIENT("ambient"); const std::string LLSettingsSky::SETTING_BLUE_DENSITY("blue_density"); @@ -940,7 +935,7 @@ LLSD LLSettingsSky::translateLegacySettings(const LLSD& legacy) void LLSettingsSky::updateSettings() { - LL_RECORD_BLOCK_TIME(FTM_RECALCULATE_SKYVALUES); + LL_PROFILE_ZONE_SCOPED; // base class clears dirty flag so as to not trigger recursive update LLSettingsBase::updateSettings(); diff --git a/indra/llinventory/llsettingswater.cpp b/indra/llinventory/llsettingswater.cpp index 0eb95dcd89..16cd3343e4 100644 --- a/indra/llinventory/llsettingswater.cpp +++ b/indra/llinventory/llsettingswater.cpp @@ -33,14 +33,6 @@ #include "v3colorutil.h" #include "indra_constants.h" -//========================================================================= -namespace -{ - LLTrace::BlockTimerStatHandle FTM_BLEND_WATERVALUES("Blending Water Environment"); - LLTrace::BlockTimerStatHandle FTM_UPDATE_WATERVALUES("Update Water Environment"); -} - -//========================================================================= const std::string LLSettingsWater::SETTING_BLUR_MULTIPLIER("blur_multiplier"); const std::string LLSettingsWater::SETTING_FOG_COLOR("water_fog_color"); const std::string LLSettingsWater::SETTING_FOG_DENSITY("water_fog_density"); diff --git a/indra/llkdu/tests/llimagej2ckdu_test.cpp b/indra/llkdu/tests/llimagej2ckdu_test.cpp index ee7b14be85..16213b7f45 100644 --- a/indra/llkdu/tests/llimagej2ckdu_test.cpp +++ b/indra/llkdu/tests/llimagej2ckdu_test.cpp @@ -63,8 +63,7 @@ U8* LLImageRaw::reallocateData(S32 ) { return NULL; } bool LLImageRaw::resize(U16, U16, S8) { return true; } // this method always returns true... LLImageBase::LLImageBase() -: LLTrace::MemTrackable("LLImageBase"), -mData(NULL), +: mData(NULL), mDataSize(0), mWidth(0), mHeight(0), diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index 0e2f62f9db..8c4a1304b4 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -74,8 +74,9 @@ public: }; template -class LLOctreeNode : public LLTreeNode +class alignas(16) LLOctreeNode : public LLTreeNode { + LL_ALIGN_NEW public: typedef LLOctreeTraveler oct_traveler; @@ -91,16 +92,6 @@ public: typedef LLOctreeNode oct_node; typedef LLOctreeListener oct_listener; - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - LLOctreeNode( const LLVector4a& center, const LLVector4a& size, BaseType* parent, diff --git a/indra/llmath/llrigginginfo.h b/indra/llmath/llrigginginfo.h index b3d6bc2d19..059c6ae082 100644 --- a/indra/llmath/llrigginginfo.h +++ b/indra/llmath/llrigginginfo.h @@ -34,9 +34,9 @@ // Extents are in joint space // isRiggedTo is based on the state of all currently associated rigged meshes -LL_ALIGN_PREFIX(16) -class LLJointRiggingInfo +class alignas(16) LLJointRiggingInfo { + LL_ALIGN_NEW public: LLJointRiggingInfo(); bool isRiggedTo() const; @@ -45,31 +45,10 @@ public: const LLVector4a *getRiggedExtents() const; void merge(const LLJointRiggingInfo& other); - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - - void* operator new[](size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete[](void* ptr) - { - ll_aligned_free_16(ptr); - } - - private: - LL_ALIGN_16(LLVector4a mRiggedExtents[2]); + LLVector4a mRiggedExtents[2]; bool mIsRiggedTo; -} LL_ALIGN_POSTFIX(16); +}; // For storing all the rigging info associated with a given avatar or // object, keyed by joint_num. diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h index 5a02928374..53c8f604f6 100644 --- a/indra/llmath/llvector4a.h +++ b/indra/llmath/llvector4a.h @@ -47,9 +47,9 @@ class LLRotation; // LLVector3/LLVector4. ///////////////////////////////// -LL_ALIGN_PREFIX(16) -class LLVector4a +class alignas(16) LLVector4a { + LL_ALIGN_NEW public: /////////////////////////////////// @@ -323,7 +323,7 @@ public: private: LLQuad mQ; -} LL_ALIGN_POSTFIX(16); +}; inline void update_min_max(LLVector4a& min, LLVector4a& max, const LLVector4a& p) { diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h index 13150028d8..b2bc440368 100644 --- a/indra/llmath/llvolumeoctree.h +++ b/indra/llmath/llvolumeoctree.h @@ -34,19 +34,10 @@ #include "llvolume.h" #include "llvector4a.h" -class LLVolumeTriangle : public LLRefCount +class alignas(16) LLVolumeTriangle : public LLRefCount { + LL_ALIGN_NEW public: - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - LLVolumeTriangle() { mBinIndex = -1; @@ -86,20 +77,10 @@ public: }; -class LLVolumeOctreeListener : public LLOctreeListener +class alignas(16) LLVolumeOctreeListener : public LLOctreeListener { + LL_ALIGN_NEW public: - - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } - LLVolumeOctreeListener(LLOctreeNode* node); ~LLVolumeOctreeListener(); diff --git a/indra/llmessage/llfiltersd2xmlrpc.cpp b/indra/llmessage/llfiltersd2xmlrpc.cpp index d3e195789b..0abdafbdfc 100644 --- a/indra/llmessage/llfiltersd2xmlrpc.cpp +++ b/indra/llmessage/llfiltersd2xmlrpc.cpp @@ -309,7 +309,6 @@ LLFilterSD2XMLRPCResponse::~LLFilterSD2XMLRPCResponse() } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_SD2XMLRPC_RESPONSE("SD2XMLRPC Response"); // virtual LLIOPipe::EStatus LLFilterSD2XMLRPCResponse::process_impl( const LLChannelDescriptors& channels, @@ -318,7 +317,7 @@ LLIOPipe::EStatus LLFilterSD2XMLRPCResponse::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_SD2XMLRPC_RESPONSE); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; // This pipe does not work if it does not have everyting. This @@ -386,8 +385,6 @@ LLFilterSD2XMLRPCRequest::~LLFilterSD2XMLRPCRequest() { } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_SD2XMLRPC_REQUEST("S22XMLRPC Request"); - // virtual LLIOPipe::EStatus LLFilterSD2XMLRPCRequest::process_impl( const LLChannelDescriptors& channels, @@ -396,7 +393,7 @@ LLIOPipe::EStatus LLFilterSD2XMLRPCRequest::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_SD2XMLRPC_REQUEST); + LL_PROFILE_ZONE_SCOPED; // This pipe does not work if it does not have everyting. This // could be addressed by making a stream parser for llsd which // handled partial information. @@ -593,8 +590,6 @@ LLFilterXMLRPCResponse2LLSD::~LLFilterXMLRPCResponse2LLSD() { } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_XMLRPC2LLSD_RESPONSE("XMLRPC2LLSD Response"); - LLIOPipe::EStatus LLFilterXMLRPCResponse2LLSD::process_impl( const LLChannelDescriptors& channels, buffer_ptr_t& buffer, @@ -602,7 +597,7 @@ LLIOPipe::EStatus LLFilterXMLRPCResponse2LLSD::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_XMLRPC2LLSD_RESPONSE); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; if(!eos) return STATUS_BREAK; @@ -679,7 +674,6 @@ LLFilterXMLRPCRequest2LLSD::~LLFilterXMLRPCRequest2LLSD() { } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_XMLRPC2LLSD_REQUEST("XMLRPC2LLSD Request"); LLIOPipe::EStatus LLFilterXMLRPCRequest2LLSD::process_impl( const LLChannelDescriptors& channels, buffer_ptr_t& buffer, @@ -687,7 +681,7 @@ LLIOPipe::EStatus LLFilterXMLRPCRequest2LLSD::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_XMLRPC2LLSD_REQUEST); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; if(!eos) return STATUS_BREAK; if(!buffer) return STATUS_ERROR; diff --git a/indra/llmessage/llhttpnode.cpp b/indra/llmessage/llhttpnode.cpp index 6fd17c9154..6e9598a0a3 100644 --- a/indra/llmessage/llhttpnode.cpp +++ b/indra/llmessage/llhttpnode.cpp @@ -121,6 +121,7 @@ LLSD LLHTTPNode::simplePost(const LLSD& input) const // virtual void LLHTTPNode::get(LLHTTPNode::ResponsePtr response, const LLSD& context) const { + LL_PROFILE_ZONE_SCOPED; try { response->result(simpleGet()); @@ -134,6 +135,7 @@ void LLHTTPNode::get(LLHTTPNode::ResponsePtr response, const LLSD& context) cons // virtual void LLHTTPNode::put(LLHTTPNode::ResponsePtr response, const LLSD& context, const LLSD& input) const { + LL_PROFILE_ZONE_SCOPED; try { response->result(simplePut(input)); @@ -147,6 +149,7 @@ void LLHTTPNode::put(LLHTTPNode::ResponsePtr response, const LLSD& context, cons // virtual void LLHTTPNode::post(LLHTTPNode::ResponsePtr response, const LLSD& context, const LLSD& input) const { + LL_PROFILE_ZONE_SCOPED; try { response->result(simplePost(input)); @@ -160,6 +163,7 @@ void LLHTTPNode::post(LLHTTPNode::ResponsePtr response, const LLSD& context, con // virtual void LLHTTPNode::del(LLHTTPNode::ResponsePtr response, const LLSD& context) const { + LL_PROFILE_ZONE_SCOPED; try { response->result(simpleDel(context)); diff --git a/indra/llmessage/lliohttpserver.cpp b/indra/llmessage/lliohttpserver.cpp index d9042fa8b0..c707c7ad09 100644 --- a/indra/llmessage/lliohttpserver.cpp +++ b/indra/llmessage/lliohttpserver.cpp @@ -132,12 +132,6 @@ private: LLSD mHeaders; }; -static LLTrace::BlockTimerStatHandle FTM_PROCESS_HTTP_PIPE("HTTP Pipe"); -static LLTrace::BlockTimerStatHandle FTM_PROCESS_HTTP_GET("HTTP Get"); -static LLTrace::BlockTimerStatHandle FTM_PROCESS_HTTP_PUT("HTTP Put"); -static LLTrace::BlockTimerStatHandle FTM_PROCESS_HTTP_POST("HTTP Post"); -static LLTrace::BlockTimerStatHandle FTM_PROCESS_HTTP_DELETE("HTTP Delete"); - LLIOPipe::EStatus LLHTTPPipe::process_impl( const LLChannelDescriptors& channels, buffer_ptr_t& buffer, @@ -145,7 +139,7 @@ LLIOPipe::EStatus LLHTTPPipe::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_HTTP_PIPE); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; LL_DEBUGS() << "LLSDHTTPServer::process_impl" << LL_ENDL; @@ -174,12 +168,10 @@ LLIOPipe::EStatus LLHTTPPipe::process_impl( std::string verb = context[CONTEXT_REQUEST][CONTEXT_VERB]; if(verb == HTTP_VERB_GET) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_HTTP_GET); mNode.get(LLHTTPNode::ResponsePtr(mResponse), context); } else if(verb == HTTP_VERB_PUT) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_HTTP_PUT); LLSD input; if (mNode.getContentType() == LLHTTPNode::CONTENT_TYPE_LLSD) { @@ -195,7 +187,6 @@ LLIOPipe::EStatus LLHTTPPipe::process_impl( } else if(verb == HTTP_VERB_POST) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_HTTP_POST); LLSD input; if (mNode.getContentType() == LLHTTPNode::CONTENT_TYPE_LLSD) { @@ -211,7 +202,6 @@ LLIOPipe::EStatus LLHTTPPipe::process_impl( } else if(verb == HTTP_VERB_DELETE) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_HTTP_DELETE); mNode.del(LLHTTPNode::ResponsePtr(mResponse), context); } else if(verb == HTTP_VERB_OPTIONS) @@ -455,8 +445,6 @@ protected: * LLHTTPResponseHeader */ -static LLTrace::BlockTimerStatHandle FTM_PROCESS_HTTP_HEADER("HTTP Header"); - // virtual LLIOPipe::EStatus LLHTTPResponseHeader::process_impl( const LLChannelDescriptors& channels, @@ -465,7 +453,7 @@ LLIOPipe::EStatus LLHTTPResponseHeader::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_HTTP_HEADER); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; if(eos) { @@ -655,8 +643,6 @@ void LLHTTPResponder::markBad( << "\n\n"; } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_HTTP_RESPONDER("HTTP Responder"); - // virtual LLIOPipe::EStatus LLHTTPResponder::process_impl( const LLChannelDescriptors& channels, @@ -665,7 +651,7 @@ LLIOPipe::EStatus LLHTTPResponder::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_HTTP_RESPONDER); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; LLIOPipe::EStatus status = STATUS_OK; diff --git a/indra/llmessage/lliosocket.cpp b/indra/llmessage/lliosocket.cpp index a9cc71c365..321d7286eb 100644 --- a/indra/llmessage/lliosocket.cpp +++ b/indra/llmessage/lliosocket.cpp @@ -300,8 +300,6 @@ LLIOSocketReader::~LLIOSocketReader() //LL_DEBUGS() << "Destroying LLIOSocketReader" << LL_ENDL; } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_SOCKET_READER("Socket Reader"); - // virtual LLIOPipe::EStatus LLIOSocketReader::process_impl( const LLChannelDescriptors& channels, @@ -310,7 +308,7 @@ LLIOPipe::EStatus LLIOSocketReader::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_SOCKET_READER); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; if(!mSource) return STATUS_PRECONDITION_NOT_MET; if(!mInitialized) @@ -400,7 +398,6 @@ LLIOSocketWriter::~LLIOSocketWriter() //LL_DEBUGS() << "Destroying LLIOSocketWriter" << LL_ENDL; } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_SOCKET_WRITER("Socket Writer"); // virtual LLIOPipe::EStatus LLIOSocketWriter::process_impl( const LLChannelDescriptors& channels, @@ -409,7 +406,7 @@ LLIOPipe::EStatus LLIOSocketWriter::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_SOCKET_WRITER); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; if(!mDestination) return STATUS_PRECONDITION_NOT_MET; if(!mInitialized) @@ -556,7 +553,6 @@ void LLIOServerSocket::setResponseTimeout(F32 timeout_secs) mResponseTimeout = timeout_secs; } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_SERVER_SOCKET("Server Socket"); // virtual LLIOPipe::EStatus LLIOServerSocket::process_impl( const LLChannelDescriptors& channels, @@ -565,7 +561,7 @@ LLIOPipe::EStatus LLIOServerSocket::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_SERVER_SOCKET); + LL_PROFILE_ZONE_SCOPED; PUMP_DEBUG; if(!pump) { diff --git a/indra/llmessage/llioutil.cpp b/indra/llmessage/llioutil.cpp index b8443c0600..850bc2a616 100644 --- a/indra/llmessage/llioutil.cpp +++ b/indra/llmessage/llioutil.cpp @@ -45,7 +45,6 @@ LLIOPipe::EStatus LLIOFlush::process_impl( } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_SLEEP("IO Sleep"); /** * @class LLIOSleep */ @@ -56,7 +55,7 @@ LLIOPipe::EStatus LLIOSleep::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_SLEEP); + LL_PROFILE_ZONE_SCOPED; if(mSeconds > 0.0) { if(pump) pump->sleepChain(mSeconds); @@ -66,7 +65,6 @@ LLIOPipe::EStatus LLIOSleep::process_impl( return STATUS_DONE; } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_ADD_CHAIN("Add Chain"); /** * @class LLIOAddChain */ @@ -77,7 +75,7 @@ LLIOPipe::EStatus LLIOAddChain::process_impl( LLSD& context, LLPumpIO* pump) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_ADD_CHAIN); + LL_PROFILE_ZONE_SCOPED; pump->addChain(mChain, mTimeout); return STATUS_DONE; } diff --git a/indra/llmessage/llpumpio.cpp b/indra/llmessage/llpumpio.cpp index a2524e9804..35365665f6 100644 --- a/indra/llmessage/llpumpio.cpp +++ b/indra/llmessage/llpumpio.cpp @@ -416,9 +416,6 @@ void LLPumpIO::pump() pump(DEFAULT_POLL_TIMEOUT); } -static LLTrace::BlockTimerStatHandle FTM_PUMP_IO("Pump IO"); -static LLTrace::BlockTimerStatHandle FTM_PUMP_POLL("Pump Poll"); - LLPumpIO::current_chain_t LLPumpIO::removeRunningChain(LLPumpIO::current_chain_t& run_chain) { std::for_each( @@ -431,7 +428,7 @@ LLPumpIO::current_chain_t LLPumpIO::removeRunningChain(LLPumpIO::current_chain_t //timeout is in microseconds void LLPumpIO::pump(const S32& poll_timeout) { - LL_RECORD_BLOCK_TIME(FTM_PUMP_IO); + LL_PROFILE_ZONE_SCOPED; //LL_INFOS() << "LLPumpIO::pump()" << LL_ENDL; // Run any pending runners. @@ -509,7 +506,7 @@ void LLPumpIO::pump(const S32& poll_timeout) S32 count = 0; S32 client_id = 0; { - LL_RECORD_BLOCK_TIME(FTM_PUMP_POLL); + LL_PROFILE_ZONE_SCOPED; apr_pollset_poll(mPollset, poll_timeout, &count, &poll_fd); } PUMP_DEBUG; @@ -737,10 +734,9 @@ bool LLPumpIO::respond( return true; } -static LLTrace::BlockTimerStatHandle FTM_PUMP_CALLBACK_CHAIN("Chain"); - void LLPumpIO::callback() { + LL_PROFILE_ZONE_SCOPED; //LL_INFOS() << "LLPumpIO::callback()" << LL_ENDL; if(true) { @@ -756,7 +752,6 @@ void LLPumpIO::callback() callbacks_t::iterator end = mCallbacks.end(); for(; it != end; ++it) { - LL_RECORD_BLOCK_TIME(FTM_PUMP_CALLBACK_CHAIN); // it's always the first and last time for respone chains (*it).mHead = (*it).mChainLinks.begin(); (*it).mInit = true; diff --git a/indra/llmessage/lltemplatemessagereader.cpp b/indra/llmessage/lltemplatemessagereader.cpp index 6d5ad0ba08..32f79f0546 100644 --- a/indra/llmessage/lltemplatemessagereader.cpp +++ b/indra/llmessage/lltemplatemessagereader.cpp @@ -533,6 +533,8 @@ static LLTrace::BlockTimerStatHandle FTM_PROCESS_MESSAGES("Process Messages"); // decode a given message BOOL LLTemplateMessageReader::decodeData(const U8* buffer, const LLHost& sender ) { + LL_RECORD_BLOCK_TIME(FTM_PROCESS_MESSAGES); + llassert( mReceiveSize >= 0 ); llassert( mCurrentRMessageTemplate); llassert( !mCurrentRMessageData ); @@ -707,12 +709,9 @@ BOOL LLTemplateMessageReader::decodeData(const U8* buffer, const LLHost& sender decode_timer.reset(); } + if( !mCurrentRMessageTemplate->callHandlerFunc(gMessageSystem) ) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_MESSAGES); - if( !mCurrentRMessageTemplate->callHandlerFunc(gMessageSystem) ) - { - LL_WARNS() << "Message from " << sender << " with no handler function received: " << mCurrentRMessageTemplate->mName << LL_ENDL; - } + LL_WARNS() << "Message from " << sender << " with no handler function received: " << mCurrentRMessageTemplate->mName << LL_ENDL; } if(LLMessageReader::getTimeDecodes() || gMessageSystem->getTimingCallback()) diff --git a/indra/llrender/llfontbitmapcache.cpp b/indra/llrender/llfontbitmapcache.cpp index f128636ab2..c71e24c83a 100644 --- a/indra/llrender/llfontbitmapcache.cpp +++ b/indra/llrender/llfontbitmapcache.cpp @@ -30,8 +30,7 @@ #include "llfontbitmapcache.h" LLFontBitmapCache::LLFontBitmapCache() -: LLTrace::MemTrackable("LLFontBitmapCache"), - mNumComponents(0), +: mNumComponents(0), mBitmapWidth(0), mBitmapHeight(0), mBitmapNum(-1), @@ -124,9 +123,6 @@ BOOL LLFontBitmapCache::nextOpenPos(S32 width, S32 &pos_x, S32 &pos_y, S32& bitm image_gl->createGLTexture(0, image_raw); gGL.getTexUnit(0)->bind(image_gl); image_gl->setFilteringOption(LLTexUnit::TFO_POINT); // was setMipFilterNearest(TRUE, TRUE); - - claimMem(image_raw); - claimMem(image_gl); } else { @@ -156,20 +152,8 @@ void LLFontBitmapCache::destroyGL() void LLFontBitmapCache::reset() { - for (std::vector >::iterator it = mImageRawVec.begin(), end_it = mImageRawVec.end(); - it != end_it; - ++it) - { - disclaimMem(**it); - } mImageRawVec.clear(); - for (std::vector >::iterator it = mImageGLVec.begin(), end_it = mImageGLVec.end(); - it != end_it; - ++it) - { - disclaimMem(**it); - } mImageGLVec.clear(); mBitmapWidth = 0; diff --git a/indra/llrender/llfontbitmapcache.h b/indra/llrender/llfontbitmapcache.h index 75df3a94a7..7de3a6b56f 100644 --- a/indra/llrender/llfontbitmapcache.h +++ b/indra/llrender/llfontbitmapcache.h @@ -32,7 +32,7 @@ // Maintain a collection of bitmaps containing rendered glyphs. // Generalizes the single-bitmap logic from LLFontFreetype and LLFontGL. -class LLFontBitmapCache : public LLTrace::MemTrackable +class LLFontBitmapCache { public: LLFontBitmapCache(); diff --git a/indra/llrender/llfontfreetype.cpp b/indra/llrender/llfontfreetype.cpp index e18161e53c..e964d1586f 100644 --- a/indra/llrender/llfontfreetype.cpp +++ b/indra/llrender/llfontfreetype.cpp @@ -104,8 +104,7 @@ LLFontGlyphInfo::LLFontGlyphInfo(U32 index) } LLFontFreetype::LLFontFreetype() -: LLTrace::MemTrackable("LLFontFreetype"), - mFontBitmapCachep(new LLFontBitmapCache), +: mFontBitmapCachep(new LLFontBitmapCache), mAscender(0.f), mDescender(0.f), mLineHeight(0.f), @@ -222,8 +221,6 @@ BOOL LLFontFreetype::loadFace(const std::string& filename, F32 point_size, F32 v S32 max_char_height = ll_round(0.5f + (y_max - y_min)); mFontBitmapCachep->init(components, max_char_width, max_char_height); - claimMem(mFontBitmapCachep); - if (!mFTFace->charmap) { @@ -238,7 +235,6 @@ BOOL LLFontFreetype::loadFace(const std::string& filename, F32 point_size, F32 v } mName = filename; - claimMem(mName); mPointSize = point_size; mStyle = LLFontGL::NORMAL; @@ -586,7 +582,6 @@ void LLFontFreetype::insertGlyphInfo(llwchar wch, LLFontGlyphInfo* gi) const } else { - claimMem(gi); mCharGlyphInfoMap[wch] = gi; } } @@ -632,11 +627,9 @@ void LLFontFreetype::resetBitmapCache() it != end_it; ++it) { - disclaimMem(it->second); delete it->second; } mCharGlyphInfoMap.clear(); - disclaimMem(mFontBitmapCachep); mFontBitmapCachep->reset(); // Adding default glyph is skipped for fallback fonts here as well as in loadFace(). diff --git a/indra/llrender/llfontfreetype.h b/indra/llrender/llfontfreetype.h index 1afe84e770..f61f169987 100644 --- a/indra/llrender/llfontfreetype.h +++ b/indra/llrender/llfontfreetype.h @@ -76,7 +76,7 @@ struct LLFontGlyphInfo extern LLFontManager *gFontManagerp; -class LLFontFreetype : public LLRefCount, public LLTrace::MemTrackable +class LLFontFreetype : public LLRefCount { public: LLFontFreetype(); diff --git a/indra/llrender/llfontgl.cpp b/indra/llrender/llfontgl.cpp index 266399d212..7f734e41f3 100644 --- a/indra/llrender/llfontgl.cpp +++ b/indra/llrender/llfontgl.cpp @@ -109,8 +109,6 @@ S32 LLFontGL::getNumFaces(const std::string& filename) return mFontFreetype->getNumFaces(filename); } -static LLTrace::BlockTimerStatHandle FTM_RENDER_FONTS("Fonts"); - S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, const LLRect& rect, const LLColor4 &color, HAlign halign, VAlign valign, U8 style, ShadowType shadow, S32 max_chars, F32* right_x, BOOL use_ellipses) const { @@ -147,7 +145,7 @@ S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, const LLRectf& rec S32 LLFontGL::render(const LLWString &wstr, S32 begin_offset, F32 x, F32 y, const LLColor4 &color, HAlign halign, VAlign valign, U8 style, ShadowType shadow, S32 max_chars, S32 max_pixels, F32* right_x, BOOL use_ellipses) const { - LL_RECORD_BLOCK_TIME(FTM_RENDER_FONTS); + LL_PROFILE_ZONE_SCOPED; if(!sDisplayFont) //do not display texts { diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp index b66521132c..4f9aa5f979 100644 --- a/indra/llrender/llgl.cpp +++ b/indra/llrender/llgl.cpp @@ -2434,6 +2434,7 @@ void LLGLNamePool::cleanup() GLuint LLGLNamePool::allocate() { + LL_PROFILE_ZONE_SCOPED; #if LL_GL_NAME_POOLING for (name_list_t::iterator iter = mNameList.begin(); iter != mNameList.end(); ++iter) { diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index b5e1910242..09b1c71f02 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -276,11 +276,10 @@ S32 LLImageGL::dataFormatComponents(S32 dataformat) //---------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_IMAGE_UPDATE_STATS("Image Stats"); // static void LLImageGL::updateStats(F32 current_time) { - LL_RECORD_BLOCK_TIME(FTM_IMAGE_UPDATE_STATS); + LL_PROFILE_ZONE_SCOPED; sLastFrameTime = current_time; sBoundTextureMemory = sCurBoundTextureMemory; sCurBoundTextureMemory = S32Bytes(0); @@ -313,10 +312,8 @@ void LLImageGL::destroyGL(BOOL save_state) if (save_state && glimage->isGLTextureCreated() && glimage->mComponents) { glimage->mSaveData = new LLImageRaw; - glimage->claimMem(glimage->mSaveData); if(!glimage->readBackRaw(glimage->mCurrentDiscardLevel, glimage->mSaveData, false)) //necessary, keep it. { - glimage->disclaimMem(glimage->mSaveData); glimage->mSaveData = NULL ; } } @@ -390,8 +387,7 @@ BOOL LLImageGL::create(LLPointer& dest, const LLImageRaw* imageraw, B //---------------------------------------------------------------------------- LLImageGL::LLImageGL(BOOL usemipmaps) -: LLTrace::MemTrackable("LLImageGL"), - mSaveData(0), mExternalTexture(FALSE) +: mSaveData(0), mExternalTexture(FALSE) { init(usemipmaps); setSize(0, 0, 0); @@ -400,8 +396,7 @@ LLImageGL::LLImageGL(BOOL usemipmaps) } LLImageGL::LLImageGL(U32 width, U32 height, U8 components, BOOL usemipmaps) -: LLTrace::MemTrackable("LLImageGL"), - mSaveData(0), mExternalTexture(FALSE) +: mSaveData(0), mExternalTexture(FALSE) { llassert( components <= 4 ); init(usemipmaps); @@ -411,8 +406,7 @@ LLImageGL::LLImageGL(U32 width, U32 height, U8 components, BOOL usemipmaps) } LLImageGL::LLImageGL(const LLImageRaw* imageraw, BOOL usemipmaps) -: LLTrace::MemTrackable("LLImageGL"), - mSaveData(0), mExternalTexture(FALSE) +: mSaveData(0), mExternalTexture(FALSE) { init(usemipmaps); setSize(0, 0, 0); @@ -430,7 +424,6 @@ LLImageGL::LLImageGL( LLGLenum formatPrimary, LLGLenum formatType, LLTexUnit::eTextureAddressMode addressMode) - : LLTrace::MemTrackable("LLImageGL"), mSaveData(0), mExternalTexture(TRUE) { init(false); mTexName = texName; @@ -682,10 +675,9 @@ void LLImageGL::setImage(const LLImageRaw* imageraw) setImage(rawdata, FALSE); } -static LLTrace::BlockTimerStatHandle FTM_SET_IMAGE("setImage"); BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips, S32 usename) { - LL_RECORD_BLOCK_TIME(FTM_SET_IMAGE); + LL_PROFILE_ZONE_SCOPED; bool is_compressed = false; switch (mFormatPrimary) @@ -746,8 +738,6 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips, S32 usename) } else { -// LL_RECORD_BLOCK_TIME(FTM_TEMP4); - if(mFormatSwapBytes) { glPixelStorei(GL_UNPACK_SWAP_BYTES, 1); @@ -778,8 +768,6 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips, S32 usename) { stop_glerror(); { -// LL_RECORD_BLOCK_TIME(FTM_TEMP4); - if(mFormatSwapBytes) { glPixelStorei(GL_UNPACK_SWAP_BYTES, 1); @@ -889,7 +877,6 @@ BOOL LLImageGL::setImage(const U8* data_in, BOOL data_hasmips, S32 usename) llassert(w > 0 && h > 0 && cur_mip_data); (void)cur_mip_data; { -// LL_RECORD_BLOCK_TIME(FTM_TEMP4); if(mFormatSwapBytes) { glPixelStorei(GL_UNPACK_SWAP_BYTES, 1); @@ -1203,10 +1190,9 @@ BOOL LLImageGL::setSubImageFromFrameBuffer(S32 fb_x, S32 fb_y, S32 x_pos, S32 y_ } // static -static LLTrace::BlockTimerStatHandle FTM_GENERATE_TEXTURES("generate textures"); void LLImageGL::generateTextures(S32 numTextures, U32 *textures) { - LL_RECORD_BLOCK_TIME(FTM_GENERATE_TEXTURES); + LL_PROFILE_ZONE_SCOPED; glGenTextures(numTextures, textures); } @@ -1220,10 +1206,9 @@ void LLImageGL::deleteTextures(S32 numTextures, const U32 *textures) } // static -static LLTrace::BlockTimerStatHandle FTM_SET_MANUAL_IMAGE("setManualImage"); void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 width, S32 height, U32 pixformat, U32 pixtype, const void* pixels, bool allow_compression) { - LL_RECORD_BLOCK_TIME(FTM_SET_MANUAL_IMAGE); + LL_PROFILE_ZONE_SCOPED; bool use_scratch = false; U32* scratch = NULL; if (LLRender::sGLCoreProfile) @@ -1338,10 +1323,9 @@ void LLImageGL::setManualImage(U32 target, S32 miplevel, S32 intformat, S32 widt //create an empty GL texture: just create a texture name //the texture is assiciate with some image by calling glTexImage outside LLImageGL -static LLTrace::BlockTimerStatHandle FTM_CREATE_GL_TEXTURE1("createGLTexture()"); BOOL LLImageGL::createGLTexture() { - LL_RECORD_BLOCK_TIME(FTM_CREATE_GL_TEXTURE1); + LL_PROFILE_ZONE_SCOPED; if (gGLManager.mIsDisabled) { LL_WARNS() << "Trying to create a texture while GL is disabled!" << LL_ENDL; @@ -1371,10 +1355,9 @@ BOOL LLImageGL::createGLTexture() return TRUE ; } -static LLTrace::BlockTimerStatHandle FTM_CREATE_GL_TEXTURE2("createGLTexture(raw)"); BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S32 usename/*=0*/, BOOL to_create, S32 category) { - LL_RECORD_BLOCK_TIME(FTM_CREATE_GL_TEXTURE2); + LL_PROFILE_ZONE_SCOPED; if (gGLManager.mIsDisabled) { LL_WARNS() << "Trying to create a texture while GL is disabled!" << LL_ENDL; @@ -1485,10 +1468,9 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const LLImageRaw* imageraw, S return createGLTexture(discard_level, rawdata, FALSE, usename); } -static LLTrace::BlockTimerStatHandle FTM_CREATE_GL_TEXTURE3("createGLTexture3(data)"); BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_hasmips, S32 usename) { - LL_RECORD_BLOCK_TIME(FTM_CREATE_GL_TEXTURE3); + LL_PROFILE_ZONE_SCOPED; llassert(data_in); stop_glerror(); @@ -1595,9 +1577,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_ mNewTexName = 0; } - disclaimMem(mTextureMemory); mTextureMemory = (S32Bytes)getMipBytes(mCurrentDiscardLevel); - claimMem(mTextureMemory); sGlobalTextureMemory += mTextureMemory; mTexelsInGLTexture = getWidth() * getHeight(); @@ -1735,7 +1715,6 @@ void LLImageGL::destroyGLTexture() if(mTextureMemory != S32Bytes(0)) { sGlobalTextureMemory -= mTextureMemory; - disclaimMem(mTextureMemory); mTextureMemory = (S32Bytes)0; } @@ -2088,7 +2067,6 @@ U32 LLImageGL::createPickMask(S32 pWidth, S32 pHeight) U32 size = pick_width * pick_height; size = (size + 7) / 8; // pixelcount-to-bits mPickMask = new U8[size]; - claimMem(size); mPickMaskWidth = pick_width - 1; mPickMaskHeight = pick_height - 1; @@ -2103,7 +2081,6 @@ void LLImageGL::freePickMask() // pickmask validity depends on old image size, delete it if (mPickMask != NULL) { - disclaimMem((mPickMaskWidth * mPickMaskHeight + 7) / 8); delete [] mPickMask; } mPickMask = NULL; diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h index da626a1093..95b60bd0bd 100644 --- a/indra/llrender/llimagegl.h +++ b/indra/llrender/llimagegl.h @@ -46,7 +46,7 @@ class LLWindow; #define MEGA_BYTES_TO_BYTES(x) ((x) << 20) //============================================================================ -class LLImageGL : public LLRefCount, public LLTrace::MemTrackable +class LLImageGL : public LLRefCount { friend class LLTexUnit; public: diff --git a/indra/llrender/llrender2dutils.cpp b/indra/llrender/llrender2dutils.cpp index dd34f3e383..5a942996be 100644 --- a/indra/llrender/llrender2dutils.cpp +++ b/indra/llrender/llrender2dutils.cpp @@ -1083,8 +1083,6 @@ void gl_rect_2d_simple( S32 width, S32 height ) gGL.end(); } -static LLTrace::BlockTimerStatHandle FTM_RENDER_SEGMENTED_RECT ("Render segmented rectangle"); - void gl_segmented_rect_2d_tex(const S32 left, const S32 top, const S32 right, @@ -1094,7 +1092,7 @@ void gl_segmented_rect_2d_tex(const S32 left, const S32 border_size, const U32 edges) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_SEGMENTED_RECT); + LL_PROFILE_ZONE_SCOPED; S32 width = llabs(right - left); S32 height = llabs(top - bottom); @@ -1253,7 +1251,7 @@ void gl_segmented_rect_2d_fragment_tex(const LLRect& rect, const F32 end_fragment, const U32 edges) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_SEGMENTED_RECT); + LL_PROFILE_ZONE_SCOPED; const S32 left = rect.mLeft; const S32 right = rect.mRight; const S32 top = rect.mTop; @@ -1440,7 +1438,7 @@ void gl_segmented_rect_2d_fragment_tex(const LLRect& rect, void gl_segmented_rect_3d_tex(const LLRectf& clip_rect, const LLRectf& center_uv_rect, const LLRectf& center_draw_rect, const LLVector3& width_vec, const LLVector3& height_vec) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_SEGMENTED_RECT); + LL_PROFILE_ZONE_SCOPED; gGL.begin(LLRender::QUADS); { diff --git a/indra/llrender/lltexture.h b/indra/llrender/lltexture.h index 256d85ce5a..e890a5a30b 100644 --- a/indra/llrender/lltexture.h +++ b/indra/llrender/lltexture.h @@ -42,7 +42,7 @@ class LLFontGL ; // //this is an abstract class as the parent for the class LLGLTexture // -class LLTexture : public virtual LLRefCount, public LLTrace::MemTrackable +class LLTexture : public virtual LLRefCount { friend class LLTexUnit ; friend class LLFontGL ; @@ -52,7 +52,6 @@ protected: public: LLTexture() - : LLTrace::MemTrackable("LLTexture") {} // diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 7e726df907..5ea07ddcb1 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -805,13 +805,15 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const placeFence(); } -static LLTrace::BlockTimerStatHandle FTM_GL_DRAW_ARRAYS("GL draw arrays"); + void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const { + LL_PROFILE_ZONE_SCOPED; llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL); mMappable = false; gGL.syncMatrices(); +#ifndef LL_RELEASE_FOR_DOWNLOAD llassert(mNumVerts >= 0); if (first >= (U32)mNumVerts || first + count > (U32)mNumVerts) @@ -839,16 +841,11 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const LL_ERRS() << "Invalid draw mode: " << mode << LL_ENDL; return; } +#endif { - LL_RECORD_BLOCK_TIME(FTM_GL_DRAW_ARRAYS); - stop_glerror(); - LLGLSLShader::startProfile(); - stop_glerror(); LL_PROFILER_GPU_ZONEC("gl.DrawArrays", 0xFF4040) glDrawArrays(sGLMode[mode], first, count); - stop_glerror(); - LLGLSLShader::stopProfile(count, mode); } stop_glerror(); @@ -949,8 +946,7 @@ S32 LLVertexBuffer::determineUsage(S32 usage) } LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) -: LLTrace::MemTrackable("LLVertexBuffer"), - LLRefCount(), +: LLRefCount(), mNumVerts(0), mNumIndices(0), @@ -1095,9 +1091,7 @@ void LLVertexBuffer::waitFence() const void LLVertexBuffer::genBuffer(U32 size) { - disclaimMem(mSize); mSize = vbo_block_size(size); - claimMem(mSize); if (mUsage == GL_STREAM_DRAW_ARB) { @@ -1193,9 +1187,7 @@ bool LLVertexBuffer::createGLBuffer(U32 size) static int gl_buffer_idx = 0; mGLBuffer = ++gl_buffer_idx; mMappedData = (U8*)ll_aligned_malloc_16(size); - disclaimMem(mSize); mSize = size; - claimMem(mSize); } if (!mMappedData) @@ -1365,8 +1357,6 @@ bool LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) return success; } -static LLTrace::BlockTimerStatHandle FTM_SETUP_VERTEX_ARRAY("Setup VAO"); - void LLVertexBuffer::setupVertexArray() { if (!mGLArray) @@ -1374,7 +1364,7 @@ void LLVertexBuffer::setupVertexArray() return; } - LL_RECORD_BLOCK_TIME(FTM_SETUP_VERTEX_ARRAY); + LL_PROFILE_ZONE_SCOPED; #if GL_ARB_vertex_array_object glBindVertexArray(mGLArray); #endif @@ -1547,12 +1537,11 @@ bool expand_region(LLVertexBuffer::MappedRegion& region, S32 index, S32 count) return true; } -static LLTrace::BlockTimerStatHandle FTM_VBO_MAP_BUFFER_RANGE("VBO Map Range"); -static LLTrace::BlockTimerStatHandle FTM_VBO_MAP_BUFFER("VBO Map"); // Map for data access volatile U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range) { + LL_PROFILE_ZONE_SCOPED; bindGLBuffer(true); if (mFinal) { @@ -1619,7 +1608,6 @@ volatile U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, boo if (map_range) { #ifdef GL_ARB_map_buffer_range - LL_RECORD_BLOCK_TIME(FTM_VBO_MAP_BUFFER_RANGE); S32 offset = mOffsets[type] + sTypeSize[type]*index; S32 length = (sTypeSize[type]*count+0xF) & ~0xF; src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, offset, length, @@ -1643,7 +1631,6 @@ volatile U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, boo } } - LL_RECORD_BLOCK_TIME(FTM_VBO_MAP_BUFFER); src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, 0, mSize, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT); @@ -1729,11 +1716,9 @@ volatile U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, boo } -static LLTrace::BlockTimerStatHandle FTM_VBO_MAP_INDEX_RANGE("IBO Map Range"); -static LLTrace::BlockTimerStatHandle FTM_VBO_MAP_INDEX("IBO Map"); - volatile U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range) { + LL_PROFILE_ZONE_SCOPED; bindGLIndices(true); if (mFinal) { @@ -1808,7 +1793,6 @@ volatile U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range if (map_range) { #ifdef GL_ARB_map_buffer_range - LL_RECORD_BLOCK_TIME(FTM_VBO_MAP_INDEX_RANGE); S32 offset = sizeof(U16)*index; S32 length = sizeof(U16)*count; src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, @@ -1820,7 +1804,6 @@ volatile U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range else { #ifdef GL_ARB_map_buffer_range - LL_RECORD_BLOCK_TIME(FTM_VBO_MAP_INDEX); src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, sizeof(U16)*mNumIndices, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT); @@ -1844,7 +1827,6 @@ volatile U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range } else { - LL_RECORD_BLOCK_TIME(FTM_VBO_MAP_INDEX); map_range = false; src = (U8*) glMapBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB, GL_WRITE_ONLY_ARB); } @@ -1895,13 +1877,6 @@ volatile U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range } } -static LLTrace::BlockTimerStatHandle FTM_VBO_UNMAP("VBO Unmap"); -static LLTrace::BlockTimerStatHandle FTM_VBO_FLUSH_RANGE("Flush VBO Range"); - - -static LLTrace::BlockTimerStatHandle FTM_IBO_UNMAP("IBO Unmap"); -static LLTrace::BlockTimerStatHandle FTM_IBO_FLUSH_RANGE("Flush IBO Range"); - void LLVertexBuffer::unmapBuffer() { if (!useVBOs()) @@ -1910,10 +1885,10 @@ void LLVertexBuffer::unmapBuffer() } bool updated_all = false; - + LL_PROFILE_ZONE_SCOPED; if (mMappedData && mVertexLocked) { - LL_RECORD_BLOCK_TIME(FTM_VBO_UNMAP); + LL_PROFILE_ZONE_NAMED("unmapBuffer - vertex"); bindGLBuffer(true); updated_all = mIndexLocked; //both vertex and index buffers done updating @@ -1960,7 +1935,7 @@ void LLVertexBuffer::unmapBuffer() { if (!mMappedVertexRegions.empty()) { - stop_glerror(); + LL_PROFILE_ZONE_NAMED("unmapBuffer - flush vertex"); for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) { const MappedRegion& region = mMappedVertexRegions[i]; @@ -1968,18 +1943,16 @@ void LLVertexBuffer::unmapBuffer() S32 length = sTypeSize[region.mType]*region.mCount; if (gGLManager.mHasMapBufferRange) { - LL_RECORD_BLOCK_TIME(FTM_VBO_FLUSH_RANGE); #ifdef GL_ARB_map_buffer_range glFlushMappedBufferRange(GL_ARRAY_BUFFER_ARB, offset, length); #endif } else if (gGLManager.mHasFlushBufferRange) - { + { #ifndef LL_MESA_HEADLESS glFlushMappedBufferRangeAPPLE(GL_ARRAY_BUFFER_ARB, offset, length); #endif } - stop_glerror(); } mMappedVertexRegions.clear(); @@ -1998,7 +1971,7 @@ void LLVertexBuffer::unmapBuffer() if (mMappedIndexData && mIndexLocked) { - LL_RECORD_BLOCK_TIME(FTM_IBO_UNMAP); + LL_PROFILE_ZONE_NAMED("unmapBuffer - index"); bindGLIndices(); if(!mMappable) { @@ -2044,12 +2017,12 @@ void LLVertexBuffer::unmapBuffer() { for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) { + LL_PROFILE_ZONE_NAMED("unmapBuffer - flush index"); const MappedRegion& region = mMappedIndexRegions[i]; S32 offset = region.mIndex >= 0 ? sizeof(U16)*region.mIndex : 0; S32 length = sizeof(U16)*region.mCount; if (gGLManager.mHasMapBufferRange) { - LL_RECORD_BLOCK_TIME(FTM_IBO_FLUSH_RANGE); #ifdef GL_ARB_map_buffer_range glFlushMappedBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length); #endif @@ -2068,9 +2041,8 @@ void LLVertexBuffer::unmapBuffer() mMappedIndexRegions.clear(); } } - stop_glerror(); - glUnmapBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB); - stop_glerror(); + + glUnmapBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB); mMappedIndexData = NULL; } @@ -2193,13 +2165,12 @@ bool LLVertexBuffer::getClothWeightStrider(LLStrider& strider, S32 in //---------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_BIND_GL_ARRAY("Bind Array"); bool LLVertexBuffer::bindGLArray() { if (mGLArray && sGLRenderArray != mGLArray) { { - LL_RECORD_BLOCK_TIME(FTM_BIND_GL_ARRAY); + LL_PROFILE_ZONE_SCOPED; #if GL_ARB_vertex_array_object glBindVertexArray(mGLArray); #endif @@ -2216,8 +2187,6 @@ bool LLVertexBuffer::bindGLArray() return false; } -static LLTrace::BlockTimerStatHandle FTM_BIND_GL_BUFFER("Bind Buffer"); - bool LLVertexBuffer::bindGLBuffer(bool force_bind) { bindGLArray(); @@ -2226,8 +2195,7 @@ bool LLVertexBuffer::bindGLBuffer(bool force_bind) if (useVBOs() && (force_bind || (mGLBuffer && (mGLBuffer != sGLRenderBuffer || !sVBOActive)))) { - //LL_RECORD_BLOCK_TIME(FTM_BIND_GL_BUFFER); - + LL_PROFILE_ZONE_SCOPED; glBindBufferARB(GL_ARRAY_BUFFER_ARB, mGLBuffer); sGLRenderBuffer = mGLBuffer; sBindCount++; @@ -2256,16 +2224,14 @@ bool LLVertexBuffer::bindGLBufferFast() return false; } -static LLTrace::BlockTimerStatHandle FTM_BIND_GL_INDICES("Bind Indices"); - bool LLVertexBuffer::bindGLIndices(bool force_bind) { + LL_PROFILE_ZONE_SCOPED; bindGLArray(); bool ret = false; if (useVBOs() && (force_bind || (mGLIndices && (mGLIndices != sGLRenderIndices || !sIBOActive)))) { - LL_RECORD_BLOCK_TIME(FTM_BIND_GL_INDICES); /*if (sMapped) { LL_ERRS() << "VBO bound while another VBO mapped!" << LL_ENDL; diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h index fad474a143..1b400b3aad 100644 --- a/indra/llrender/llvertexbuffer.h +++ b/indra/llrender/llvertexbuffer.h @@ -98,7 +98,7 @@ public: //============================================================================ // base class class LLPrivateMemoryPool; -class LLVertexBuffer : public LLRefCount, public LLTrace::MemTrackable +class LLVertexBuffer : public LLRefCount { public: class MappedRegion @@ -113,8 +113,7 @@ public: }; LLVertexBuffer(const LLVertexBuffer& rhs) - : LLTrace::MemTrackable("LLVertexBuffer"), - mUsage(rhs.mUsage) + : mUsage(rhs.mUsage) { *this = rhs; } diff --git a/indra/llui/llfloater.cpp b/indra/llui/llfloater.cpp index 0e42922543..03efd09689 100644 --- a/indra/llui/llfloater.cpp +++ b/indra/llui/llfloater.cpp @@ -3268,11 +3268,9 @@ boost::signals2::connection LLFloater::setCloseCallback( const commit_signal_t:: return mCloseSignal.connect(cb); } -LLTrace::BlockTimerStatHandle POST_BUILD("Floater Post Build"); -static LLTrace::BlockTimerStatHandle FTM_EXTERNAL_FLOATER_LOAD("Load Extern Floater Reference"); - bool LLFloater::initFloaterXML(LLXMLNodePtr node, LLView *parent, const std::string& filename, LLXMLNodePtr output_node) { + LL_PROFILE_ZONE_SCOPED; Params default_params(LLUICtrlFactory::getDefaultParams()); Params params(default_params); @@ -3299,7 +3297,6 @@ bool LLFloater::initFloaterXML(LLXMLNodePtr node, LLView *parent, const std::str LLUICtrlFactory::instance().pushFileName(xml_filename); - LL_RECORD_BLOCK_TIME(FTM_EXTERNAL_FLOATER_LOAD); if (!LLUICtrlFactory::getLayeredXMLNode(xml_filename, referenced_xml)) { LL_WARNS() << "Couldn't parse panel from: " << xml_filename << LL_ENDL; @@ -3375,12 +3372,8 @@ bool LLFloater::initFloaterXML(LLXMLNodePtr node, LLView *parent, const std::str } BOOL result; - { - LL_RECORD_BLOCK_TIME(POST_BUILD); - - result = postBuild(); - } - + result = postBuild(); + if (!result) { LL_ERRS() << "Failed to construct floater " << getName() << LL_ENDL; @@ -3424,11 +3417,9 @@ bool LLFloater::isVisible(const LLFloater* floater) return floater && floater->getVisible(); } -static LLTrace::BlockTimerStatHandle FTM_BUILD_FLOATERS("Build Floaters"); - bool LLFloater::buildFromFile(const std::string& filename) { - LL_RECORD_BLOCK_TIME(FTM_BUILD_FLOATERS); + LL_PROFILE_ZONE_SCOPED; LLXMLNodePtr root; if (!LLUICtrlFactory::getLayeredXMLNode(filename, root)) diff --git a/indra/llui/llfolderview.cpp b/indra/llui/llfolderview.cpp index 622c9edba7..0996e82bf7 100644 --- a/indra/llui/llfolderview.cpp +++ b/indra/llui/llfolderview.cpp @@ -189,7 +189,6 @@ LLFolderView::LLFolderView(const Params& p) mViewModel(p.view_model), mGroupedItemModel(p.grouped_item_model) { - claimMem(mViewModel); LLPanel* panel = p.parent_panel; mParentPanel = panel->getHandle(); mViewModel->setFolderView(this); @@ -337,11 +336,9 @@ S32 LLFolderView::arrange( S32* unused_width, S32* unused_height ) return ll_round(mTargetHeight); } -static LLTrace::BlockTimerStatHandle FTM_FILTER("Filter Folder View"); - void LLFolderView::filter( LLFolderViewFilter& filter ) { - LL_RECORD_BLOCK_TIME(FTM_FILTER); + LL_PROFILE_ZONE_SCOPED; static LLCachedControl time_visible(*LLUI::getInstance()->mSettingGroups["config"], "FilterItemsMaxTimePerFrameVisible", 10); static LLCachedControl time_invisible(*LLUI::getInstance()->mSettingGroups["config"], "FilterItemsMaxTimePerFrameUnvisible", 1); filter.resetTime(llclamp((mParentPanel.get()->getVisible() ? time_visible() : time_invisible()), 1, 100)); @@ -503,10 +500,9 @@ BOOL LLFolderView::changeSelection(LLFolderViewItem* selection, BOOL selected) return rv; } -static LLTrace::BlockTimerStatHandle FTM_SANITIZE_SELECTION("Sanitize Selection"); void LLFolderView::sanitizeSelection() { - LL_RECORD_BLOCK_TIME(FTM_SANITIZE_SELECTION); + LL_PROFILE_ZONE_SCOPED; // store off current item in case it is automatically deselected // and we want to preserve context LLFolderViewItem* original_selected_item = getCurSelectedItem(); @@ -1621,7 +1617,6 @@ void LLFolderView::setShowSingleSelection(bool show) } } -static LLTrace::BlockTimerStatHandle FTM_AUTO_SELECT("Open and Select"); static LLTrace::BlockTimerStatHandle FTM_INVENTORY("Inventory"); // Main idle routine @@ -1657,7 +1652,6 @@ void LLFolderView::update() // automatically show matching items, and select first one if we had a selection if (mNeedsAutoSelect) { - LL_RECORD_BLOCK_TIME(FTM_AUTO_SELECT); // select new item only if a filtered item not currently selected and there was a selection LLFolderViewItem* selected_itemp = mSelectedItems.empty() ? NULL : mSelectedItems.back(); if (!mAutoSelectOverride && selected_itemp && !selected_itemp->getViewModelItem()->potentiallyVisible()) diff --git a/indra/llui/llfolderviewmodel.h b/indra/llui/llfolderviewmodel.h index e62b2779dd..093e213be3 100644 --- a/indra/llui/llfolderviewmodel.h +++ b/indra/llui/llfolderviewmodel.h @@ -108,11 +108,10 @@ public: virtual S32 getFirstRequiredGeneration() const = 0; }; -class LLFolderViewModelInterface : public LLTrace::MemTrackable +class LLFolderViewModelInterface { public: LLFolderViewModelInterface() - : LLTrace::MemTrackable("LLFolderViewModelInterface") {} virtual ~LLFolderViewModelInterface() {} @@ -133,11 +132,10 @@ public: // This is an abstract base class that users of the folderview classes // would use to bridge the folder view with the underlying data -class LLFolderViewModelItem : public LLRefCount, public LLTrace::MemTrackable +class LLFolderViewModelItem : public LLRefCount { public: LLFolderViewModelItem() - : LLTrace::MemTrackable("LLFolderViewModelItem") {} virtual ~LLFolderViewModelItem() { } diff --git a/indra/llui/lllayoutstack.cpp b/indra/llui/lllayoutstack.cpp index 29a156e933..528d2e70ad 100644 --- a/indra/llui/lllayoutstack.cpp +++ b/indra/llui/lllayoutstack.cpp @@ -340,8 +340,6 @@ void LLLayoutStack::collapsePanel(LLPanel* panel, BOOL collapsed) mNeedsLayout = true; } -static LLTrace::BlockTimerStatHandle FTM_UPDATE_LAYOUT("Update LayoutStacks"); - class LLImagePanel : public LLPanel { public: @@ -369,7 +367,7 @@ private: void LLLayoutStack::updateLayout() { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_LAYOUT); + LL_PROFILE_ZONE_SCOPED; if (!mNeedsLayout) return; diff --git a/indra/llui/llpanel.cpp b/indra/llui/llpanel.cpp index 00da0f5fec..f770920c4a 100644 --- a/indra/llui/llpanel.cpp +++ b/indra/llui/llpanel.cpp @@ -800,14 +800,12 @@ boost::signals2::connection LLPanel::setVisibleCallback( const commit_signal_t:: return mVisibleSignal->connect(cb); } -static LLTrace::BlockTimerStatHandle FTM_BUILD_PANELS("Build Panels"); - //----------------------------------------------------------------------------- // buildPanel() //----------------------------------------------------------------------------- BOOL LLPanel::buildFromFile(const std::string& filename, const LLPanel::Params& default_params) { - LL_RECORD_BLOCK_TIME(FTM_BUILD_PANELS); + LL_PROFILE_ZONE_SCOPED; BOOL didPost = FALSE; LLXMLNodePtr root; diff --git a/indra/llui/llscrolllistctrl.cpp b/indra/llui/llscrolllistctrl.cpp index de644185fd..c70085b72f 100644 --- a/indra/llui/llscrolllistctrl.cpp +++ b/indra/llui/llscrolllistctrl.cpp @@ -3045,10 +3045,9 @@ LLScrollListColumn* LLScrollListCtrl::getColumn(const std::string& name) return NULL; } -LLTrace::BlockTimerStatHandle FTM_ADD_SCROLLLIST_ELEMENT("Add Scroll List Item"); LLScrollListItem* LLScrollListCtrl::addElement(const LLSD& element, EAddPosition pos, void* userdata) { - LL_RECORD_BLOCK_TIME(FTM_ADD_SCROLLLIST_ELEMENT); + LL_PROFILE_ZONE_SCOPED; LLScrollListItem::Params item_params; LLParamSDParser parser; parser.readSD(element, item_params); @@ -3058,14 +3057,14 @@ LLScrollListItem* LLScrollListCtrl::addElement(const LLSD& element, EAddPosition LLScrollListItem* LLScrollListCtrl::addRow(const LLScrollListItem::Params& item_p, EAddPosition pos) { - LL_RECORD_BLOCK_TIME(FTM_ADD_SCROLLLIST_ELEMENT); + LL_PROFILE_ZONE_SCOPED; LLScrollListItem *new_item = new LLScrollListItem(item_p); return addRow(new_item, item_p, pos); } LLScrollListItem* LLScrollListCtrl::addRow(LLScrollListItem *new_item, const LLScrollListItem::Params& item_p, EAddPosition pos) { - LL_RECORD_BLOCK_TIME(FTM_ADD_SCROLLLIST_ELEMENT); + LL_PROFILE_ZONE_SCOPED; if (!item_p.validateBlock() || !new_item) return NULL; new_item->setNumColumns(mColumns.size()); diff --git a/indra/llui/lltextbase.cpp b/indra/llui/lltextbase.cpp index 20bea7fe24..4274c81645 100644 --- a/indra/llui/lltextbase.cpp +++ b/indra/llui/lltextbase.cpp @@ -1484,11 +1484,9 @@ S32 LLTextBase::getLeftOffset(S32 width) } } - -static LLTrace::BlockTimerStatHandle FTM_TEXT_REFLOW ("Text Reflow"); void LLTextBase::reflow() { - LL_RECORD_BLOCK_TIME(FTM_TEXT_REFLOW); + LL_PROFILE_ZONE_SCOPED; updateSegments(); @@ -1833,10 +1831,9 @@ void LLTextBase::removeDocumentChild(LLView* view) } -static LLTrace::BlockTimerStatHandle FTM_UPDATE_TEXT_SEGMENTS("Update Text Segments"); void LLTextBase::updateSegments() { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_TEXT_SEGMENTS); + LL_PROFILE_ZONE_SCOPED; createDefaultSegment(); } @@ -2096,19 +2093,16 @@ static LLUIImagePtr image_from_icon_name(const std::string& icon_name) } } -static LLTrace::BlockTimerStatHandle FTM_PARSE_HTML("Parse HTML"); - - void LLTextBase::appendTextImpl(const std::string &new_text, const LLStyle::Params& input_params) { + LL_PROFILE_ZONE_SCOPED; LLStyle::Params style_params(input_params); style_params.fillFrom(getStyleParams()); S32 part = (S32)LLTextParser::WHOLE; if (mParseHTML && !style_params.is_link) // Don't search for URLs inside a link segment (STORM-358). { - LL_RECORD_BLOCK_TIME(FTM_PARSE_HTML); S32 start=0,end=0; LLUrlMatch match; std::string text = new_text; @@ -2202,11 +2196,9 @@ void LLTextBase::setLastSegmentToolTip(const std::string &tooltip) } } -static LLTrace::BlockTimerStatHandle FTM_APPEND_TEXT("Append Text"); - void LLTextBase::appendText(const std::string &new_text, bool prepend_newline, const LLStyle::Params& input_params) { - LL_RECORD_BLOCK_TIME(FTM_APPEND_TEXT); + LL_PROFILE_ZONE_SCOPED; if (new_text.empty()) return; diff --git a/indra/llui/lluictrl.cpp b/indra/llui/lluictrl.cpp index 5924542a19..51391bb5e8 100644 --- a/indra/llui/lluictrl.cpp +++ b/indra/llui/lluictrl.cpp @@ -121,7 +121,6 @@ LLUICtrl::LLUICtrl(const LLUICtrl::Params& p, const LLViewModelPtr& viewmodel) mDoubleClickSignal(NULL), mTransparencyType(TT_DEFAULT) { - claimMem(viewmodel.get()); } void LLUICtrl::initFromParams(const Params& p) @@ -476,6 +475,7 @@ LLViewModel* LLUICtrl::getViewModel() const //virtual BOOL LLUICtrl::postBuild() { + LL_PROFILE_ZONE_SCOPED; // // Find all of the children that want to be in front and move them to the front // @@ -781,12 +781,9 @@ BOOL LLUICtrl::getIsChrome() const } - -LLTrace::BlockTimerStatHandle FTM_FOCUS_FIRST_ITEM("Focus First Item"); - BOOL LLUICtrl::focusFirstItem(BOOL prefer_text_fields, BOOL focus_flash) { - LL_RECORD_BLOCK_TIME(FTM_FOCUS_FIRST_ITEM); + LL_PROFILE_ZONE_SCOPED; // try to select default tab group child LLViewQuery query = getTabOrderQuery(); child_list_t result = query(this); @@ -1005,7 +1002,6 @@ boost::signals2::connection LLUICtrl::setCommitCallback( boost::function cb ) { if (!mValidateSignal) mValidateSignal = new enable_signal_t(); - claimMem(mValidateSignal); return mValidateSignal->connect(boost::bind(cb, _2)); } @@ -1070,7 +1066,6 @@ boost::signals2::connection LLUICtrl::setValidateCallback(const EnableCallbackPa boost::signals2::connection LLUICtrl::setCommitCallback( const commit_signal_t::slot_type& cb ) { if (!mCommitSignal) mCommitSignal = new commit_signal_t(); - claimMem(mCommitSignal); return mCommitSignal->connect(cb); } @@ -1078,7 +1073,6 @@ boost::signals2::connection LLUICtrl::setCommitCallback( const commit_signal_t:: boost::signals2::connection LLUICtrl::setValidateCallback( const enable_signal_t::slot_type& cb ) { if (!mValidateSignal) mValidateSignal = new enable_signal_t(); - claimMem(mValidateSignal); return mValidateSignal->connect(cb); } @@ -1086,7 +1080,6 @@ boost::signals2::connection LLUICtrl::setValidateCallback( const enable_signal_t boost::signals2::connection LLUICtrl::setMouseEnterCallback( const commit_signal_t::slot_type& cb ) { if (!mMouseEnterSignal) mMouseEnterSignal = new commit_signal_t(); - claimMem(mMouseEnterSignal); return mMouseEnterSignal->connect(cb); } @@ -1094,7 +1087,6 @@ boost::signals2::connection LLUICtrl::setMouseEnterCallback( const commit_signal boost::signals2::connection LLUICtrl::setMouseLeaveCallback( const commit_signal_t::slot_type& cb ) { if (!mMouseLeaveSignal) mMouseLeaveSignal = new commit_signal_t(); - claimMem(mMouseLeaveSignal); return mMouseLeaveSignal->connect(cb); } @@ -1102,7 +1094,6 @@ boost::signals2::connection LLUICtrl::setMouseLeaveCallback( const commit_signal boost::signals2::connection LLUICtrl::setMouseDownCallback( const mouse_signal_t::slot_type& cb ) { if (!mMouseDownSignal) mMouseDownSignal = new mouse_signal_t(); - claimMem(mMouseDownSignal); return mMouseDownSignal->connect(cb); } @@ -1110,7 +1101,6 @@ boost::signals2::connection LLUICtrl::setMouseDownCallback( const mouse_signal_t boost::signals2::connection LLUICtrl::setMouseUpCallback( const mouse_signal_t::slot_type& cb ) { if (!mMouseUpSignal) mMouseUpSignal = new mouse_signal_t(); - claimMem(mMouseUpSignal); return mMouseUpSignal->connect(cb); } @@ -1118,7 +1108,6 @@ boost::signals2::connection LLUICtrl::setMouseUpCallback( const mouse_signal_t:: boost::signals2::connection LLUICtrl::setRightMouseDownCallback( const mouse_signal_t::slot_type& cb ) { if (!mRightMouseDownSignal) mRightMouseDownSignal = new mouse_signal_t(); - claimMem(mRightMouseDownSignal); return mRightMouseDownSignal->connect(cb); } @@ -1126,7 +1115,6 @@ boost::signals2::connection LLUICtrl::setRightMouseDownCallback( const mouse_sig boost::signals2::connection LLUICtrl::setRightMouseUpCallback( const mouse_signal_t::slot_type& cb ) { if (!mRightMouseUpSignal) mRightMouseUpSignal = new mouse_signal_t(); - claimMem(mRightMouseUpSignal); return mRightMouseUpSignal->connect(cb); } @@ -1134,7 +1122,6 @@ boost::signals2::connection LLUICtrl::setRightMouseUpCallback( const mouse_signa boost::signals2::connection LLUICtrl::setDoubleClickCallback( const mouse_signal_t::slot_type& cb ) { if (!mDoubleClickSignal) mDoubleClickSignal = new mouse_signal_t(); - claimMem(mDoubleClickSignal); return mDoubleClickSignal->connect(cb); } diff --git a/indra/llui/lluictrlfactory.cpp b/indra/llui/lluictrlfactory.cpp index fdefae01b1..2d0c0ea8aa 100644 --- a/indra/llui/lluictrlfactory.cpp +++ b/indra/llui/lluictrlfactory.cpp @@ -44,10 +44,6 @@ // this library includes #include "llpanel.h" -LLTrace::BlockTimerStatHandle FTM_WIDGET_CONSTRUCTION("Widget Construction"); -LLTrace::BlockTimerStatHandle FTM_INIT_FROM_PARAMS("Widget InitFromParams"); -LLTrace::BlockTimerStatHandle FTM_WIDGET_SETUP("Widget Setup"); - //----------------------------------------------------------------------------- // UI Ctrl class for padding @@ -117,12 +113,10 @@ void LLUICtrlFactory::loadWidgetTemplate(const std::string& widget_tag, LLInitPa } } -static LLTrace::BlockTimerStatHandle FTM_CREATE_CHILDREN("Create XUI Children"); - //static void LLUICtrlFactory::createChildren(LLView* viewp, LLXMLNodePtr node, const widget_registry_t& registry, LLXMLNodePtr output_node) { - LL_RECORD_BLOCK_TIME(FTM_CREATE_CHILDREN); + LL_PROFILE_ZONE_SCOPED; if (node.isNull()) return; for (LLXMLNodePtr child_node = node->getFirstChild(); child_node.notNull(); child_node = child_node->getNextSibling()) @@ -159,14 +153,13 @@ void LLUICtrlFactory::createChildren(LLView* viewp, LLXMLNodePtr node, const wid } -static LLTrace::BlockTimerStatHandle FTM_XML_PARSE("XML Reading/Parsing"); //----------------------------------------------------------------------------- // getLayeredXMLNode() //----------------------------------------------------------------------------- bool LLUICtrlFactory::getLayeredXMLNode(const std::string &xui_filename, LLXMLNodePtr& root, LLDir::ESkinConstraint constraint) { - LL_RECORD_BLOCK_TIME(FTM_XML_PARSE); + LL_PROFILE_ZONE_SCOPED; std::vector paths = gDirUtilp->findSkinnedFilenames(LLDir::XUI, xui_filename, constraint); @@ -191,11 +184,9 @@ S32 LLUICtrlFactory::saveToXML(LLView* viewp, const std::string& filename) //----------------------------------------------------------------------------- //----------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_CREATE_FROM_XML("Create child widget"); - LLView *LLUICtrlFactory::createFromXML(LLXMLNodePtr node, LLView* parent, const std::string& filename, const widget_registry_t& registry, LLXMLNodePtr output_node) { - LL_RECORD_BLOCK_TIME(FTM_CREATE_FROM_XML); + LL_PROFILE_ZONE_SCOPED; std::string ctrl_type = node->getName()->mString; LLStringUtil::toLower(ctrl_type); diff --git a/indra/llui/lluictrlfactory.h b/indra/llui/lluictrlfactory.h index 135ed57a4f..3f24a3f1a5 100644 --- a/indra/llui/lluictrlfactory.h +++ b/indra/llui/lluictrlfactory.h @@ -79,10 +79,6 @@ class LLWidgetNameRegistry // LLSINGLETON(LLDefaultParamBlockRegistry); //}; -extern LLTrace::BlockTimerStatHandle FTM_WIDGET_SETUP; -extern LLTrace::BlockTimerStatHandle FTM_WIDGET_CONSTRUCTION; -extern LLTrace::BlockTimerStatHandle FTM_INIT_FROM_PARAMS; - // Build time optimization, generate this once in .cpp file #ifndef LLUICTRLFACTORY_CPP extern template class LLUICtrlFactory* LLSingleton::getInstance(); @@ -213,6 +209,7 @@ private: template static T* createWidgetImpl(const typename T::Params& params, LLView* parent = NULL) { + LL_PROFILE_ZONE_SCOPED; T* widget = NULL; if (!params.validateBlock()) @@ -221,12 +218,9 @@ private: //return NULL; } - { LL_RECORD_BLOCK_TIME(FTM_WIDGET_CONSTRUCTION); - widget = new T(params); - } - { LL_RECORD_BLOCK_TIME(FTM_INIT_FROM_PARAMS); - widget->initFromParams(params); - } + widget = new T(params); + + widget->initFromParams(params); if (parent) { @@ -239,7 +233,7 @@ private: template static T* defaultBuilder(LLXMLNodePtr node, LLView *parent, LLXMLNodePtr output_node) { - LL_RECORD_BLOCK_TIME(FTM_WIDGET_SETUP); + LL_PROFILE_ZONE_SCOPED; typename T::Params params(getDefaultParams()); diff --git a/indra/llui/llview.cpp b/indra/llui/llview.cpp index b942be2a4a..bd0b9d3db2 100644 --- a/indra/llui/llview.cpp +++ b/indra/llui/llview.cpp @@ -140,8 +140,7 @@ LLView::Params::Params() } LLView::LLView(const LLView::Params& p) -: LLTrace::MemTrackable("LLView"), - mVisible(p.visible), +: mVisible(p.visible), mInDraw(false), mName(p.name), mParentView(NULL), @@ -1597,15 +1596,11 @@ LLView* LLView::getChildView(const std::string& name, BOOL recurse) const return getChild(name, recurse); } -static LLTrace::BlockTimerStatHandle FTM_FIND_VIEWS("Find Widgets"); - LLView* LLView::findChildView(const std::string& name, BOOL recurse) const { - LL_RECORD_BLOCK_TIME(FTM_FIND_VIEWS); - //richard: should we allow empty names? - //if(name.empty()) - // return NULL; - // Look for direct children *first* + LL_PROFILE_ZONE_SCOPED; + + // Look for direct children *first* BOOST_FOREACH(LLView* childp, mChildList) { llassert(childp); diff --git a/indra/llui/llview.h b/indra/llui/llview.h index c60dcf3344..bec45df78a 100644 --- a/indra/llui/llview.h +++ b/indra/llui/llview.h @@ -100,8 +100,7 @@ class LLView : public LLMouseHandler, // handles mouse events public LLFocusableElement, // handles keyboard events public LLMortician, // lazy deletion - public LLHandleProvider, // passes out weak references to self - public LLTrace::MemTrackable // track memory usage + public LLHandleProvider // passes out weak references to self { public: diff --git a/indra/llui/llviewereventrecorder.cpp b/indra/llui/llviewereventrecorder.cpp index cb000aef74..5a44ec947a 100644 --- a/indra/llui/llviewereventrecorder.cpp +++ b/indra/llui/llviewereventrecorder.cpp @@ -28,6 +28,8 @@ #include "llui.h" #include "llleap.h" +LLViewerEventRecorder* LLSimpleton::sInstance = nullptr; + LLViewerEventRecorder::LLViewerEventRecorder() { clear(UNDEFINED); diff --git a/indra/llui/llviewereventrecorder.h b/indra/llui/llviewereventrecorder.h index d1059d55de..d2c0780361 100644 --- a/indra/llui/llviewereventrecorder.h +++ b/indra/llui/llviewereventrecorder.h @@ -42,12 +42,12 @@ #include "llsingleton.h" // includes llerror which we need here so we can skip the include here -class LLViewerEventRecorder : public LLSingleton +class LLViewerEventRecorder : public LLSimpleton { - LLSINGLETON(LLViewerEventRecorder); - ~LLViewerEventRecorder(); - - public: +public: + LLViewerEventRecorder(); + ~LLViewerEventRecorder(); + void updateMouseEventInfo(S32 local_x,S32 local_y, S32 global_x, S32 global_y, std::string mName); void setMouseLocalCoords(S32 x,S32 y); void setMouseGlobalCoords(S32 x,S32 y); diff --git a/indra/llui/llviewmodel.cpp b/indra/llui/llviewmodel.cpp index 282addf692..a400eb70c0 100644 --- a/indra/llui/llviewmodel.cpp +++ b/indra/llui/llviewmodel.cpp @@ -37,15 +37,13 @@ /// LLViewModel::LLViewModel() -: LLTrace::MemTrackable("LLViewModel"), - mDirty(false) +: mDirty(false) { } /// Instantiate an LLViewModel with an existing data value LLViewModel::LLViewModel(const LLSD& value) -: LLTrace::MemTrackable("LLViewModel"), - mDirty(false) +: mDirty(false) { setValue(value); } @@ -82,15 +80,9 @@ LLTextViewModel::LLTextViewModel(const LLSD& value) void LLTextViewModel::setValue(const LLSD& value) { // approximate LLSD storage usage - disclaimMem(mDisplay.size()); LLViewModel::setValue(value); - disclaimMem(mDisplay); mDisplay = utf8str_to_wstring(value.asString()); - claimMem(mDisplay); - // approximate LLSD storage usage - claimMem(mDisplay.size()); - // mDisplay and mValue agree mUpdateFromDisplay = false; } @@ -101,12 +93,8 @@ void LLTextViewModel::setDisplay(const LLWString& value) // and do the utf8str_to_wstring() to get the corresponding mDisplay // value. But a text editor might want to edit the display string // directly, then convert back to UTF8 on commit. - disclaimMem(mDisplay.size()); - disclaimMem(mDisplay); - mDisplay = value; - claimMem(mDisplay); - claimMem(mDisplay.size()); - mDirty = true; + mDisplay = value; + mDirty = true; // Don't immediately convert to UTF8 -- do it lazily -- we expect many // more setDisplay() calls than getValue() calls. Just flag that it needs // doing. diff --git a/indra/llui/llviewmodel.h b/indra/llui/llviewmodel.h index 49d7c322a3..e7dceb6c31 100644 --- a/indra/llui/llviewmodel.h +++ b/indra/llui/llviewmodel.h @@ -62,8 +62,7 @@ typedef LLPointer LLListViewModelPtr; * last referencing widget is destroyed. */ class LLViewModel -: public LLRefCount, - public LLTrace::MemTrackable +: public LLRefCount { public: LLViewModel(); diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index aeeba58a68..291f0f7d95 100644 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -14104,18 +14104,7 @@ Value 1 - RenderSynchronousOcclusion - - Comment - Don't let occlusion queries get more than one frame behind (block until they complete). - Persist - 1 - Type - Boolean - Value - 1 - - RenderDelayVBUpdate + RenderDelayVBUpdate Comment Delay vertex buffer updates until just before rendering diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 400a6a722b..8b4fcfccd9 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -977,6 +977,9 @@ bool LLAppViewer::init() // Initialize the repeater service. LLMainLoopRepeater::instance().start(); + // Initialize event recorder + LLViewerEventRecorder::createInstance(); + // // Initialize the window // @@ -1316,6 +1319,13 @@ bool LLAppViewer::init() // Load User's bindings loadKeyBindings(); + //LLSimpleton creations + LLEnvironment::createInstance(); + LLEnvironment::getInstance()->initSingleton(); + LLWorld::createInstance(); + LLSelectMgr::createInstance(); + LLViewerCamera::createInstance(); + return true; } @@ -1410,10 +1420,14 @@ bool LLAppViewer::doFrame() LLSD newFrame; { - LL_PROFILE_ZONE_NAMED( "df blocktimer" ) - LLTrace::BlockTimer::processTimes(); - LLTrace::get_frame_recording().nextPeriod(); - LLTrace::BlockTimer::logStats(); + LL_PROFILE_ZONE_NAMED("df LLTrace"); + if (LLFloaterReg::instanceVisible("block_timers")) + { + LLTrace::BlockTimer::processTimes(); + } + + LLTrace::get_frame_recording().nextPeriod(); + LLTrace::BlockTimer::logStats(); } LLTrace::get_thread_recorder()->pullFromChildren(); @@ -2184,6 +2198,10 @@ bool LLAppViewer::cleanup() LLError::LLCallStacks::cleanup(); + LLEnvironment::deleteSingleton(); + LLSelectMgr::deleteSingleton(); + LLViewerEventRecorder::deleteSingleton(); + // It's not at first obvious where, in this long sequence, a generic cleanup // call OUGHT to go. So let's say this: as we migrate cleanup from // explicit hand-placed calls into the generic mechanism, eventually @@ -2195,6 +2213,7 @@ bool LLAppViewer::cleanup() // deleteSingleton() methods. LLSingletonBase::deleteAll(); + LL_INFOS() << "Goodbye!" << LL_ENDL; removeDumpDir(); @@ -5641,6 +5660,7 @@ void LLAppViewer::disconnectViewer() LLWorld::getInstance()->destroyClass(); } LLVOCache::deleteSingleton(); + LLViewerCamera::deleteSingleton(); // call all self-registered classes LLDestroyClassList::instance().fireCallbacks(); diff --git a/indra/newview/lldonotdisturbnotificationstorage.cpp b/indra/newview/lldonotdisturbnotificationstorage.cpp index cb5f9c8a2c..7d4961c598 100644 --- a/indra/newview/lldonotdisturbnotificationstorage.cpp +++ b/indra/newview/lldonotdisturbnotificationstorage.cpp @@ -96,11 +96,9 @@ void LLDoNotDisturbNotificationStorage::resetDirty() mDirty = false; } -static LLTrace::BlockTimerStatHandle FTM_SAVE_DND_NOTIFICATIONS("Save DND Notifications"); - void LLDoNotDisturbNotificationStorage::saveNotifications() { - LL_RECORD_BLOCK_TIME(FTM_SAVE_DND_NOTIFICATIONS); + LL_PROFILE_ZONE_SCOPED; LLNotificationChannelPtr channelPtr = getCommunicationChannel(); const LLCommunicationChannel *commChannel = dynamic_cast(channelPtr.get()); diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp index 495e06b6f7..502ebbd4b1 100644 --- a/indra/newview/lldrawable.cpp +++ b/indra/newview/lldrawable.cpp @@ -58,8 +58,6 @@ const F32 MIN_INTERPOLATE_DISTANCE_SQUARED = 0.001f * 0.001f; const F32 MAX_INTERPOLATE_DISTANCE_SQUARED = 10.f * 10.f; const F32 OBJECT_DAMPING_TIME_CONSTANT = 0.06f; -static LLTrace::BlockTimerStatHandle FTM_CULL_REBOUND("Cull Rebound"); - extern bool gShiftFrame; @@ -93,7 +91,6 @@ void LLDrawable::incrementVisible() LLDrawable::LLDrawable(LLViewerObject *vobj, bool new_entry) : LLViewerOctreeEntryData(LLViewerOctreeEntry::LLDRAWABLE), - LLTrace::MemTrackable("LLDrawable"), mVObjp(vobj) { init(new_entry); @@ -263,19 +260,13 @@ BOOL LLDrawable::isLight() const } } -static LLTrace::BlockTimerStatHandle FTM_CLEANUP_DRAWABLE("Cleanup Drawable"); -static LLTrace::BlockTimerStatHandle FTM_DEREF_DRAWABLE("Deref"); -static LLTrace::BlockTimerStatHandle FTM_DELETE_FACES("Faces"); - void LLDrawable::cleanupReferences() { - LL_RECORD_BLOCK_TIME(FTM_CLEANUP_DRAWABLE); + LL_PROFILE_ZONE_SCOPED; - { - LL_RECORD_BLOCK_TIME(FTM_DELETE_FACES); - std::for_each(mFaces.begin(), mFaces.end(), DeletePointer()); - mFaces.clear(); - } + + std::for_each(mFaces.begin(), mFaces.end(), DeletePointer()); + mFaces.clear(); gObjectList.removeDrawable(this); @@ -283,12 +274,9 @@ void LLDrawable::cleanupReferences() removeFromOctree(); - { - LL_RECORD_BLOCK_TIME(FTM_DEREF_DRAWABLE); - // Cleanup references to other objects - mVObjp = NULL; - mParent = NULL; - } + // Cleanup references to other objects + mVObjp = NULL; + mParent = NULL; } void LLDrawable::removeFromOctree() @@ -333,15 +321,12 @@ S32 LLDrawable::findReferences(LLDrawable *drawablep) return count; } -static LLTrace::BlockTimerStatHandle FTM_ALLOCATE_FACE("Allocate Face"); - LLFace* LLDrawable::addFace(LLFacePool *poolp, LLViewerTexture *texturep) { LL_PROFILE_ZONE_SCOPED LLFace *face; { - LL_RECORD_BLOCK_TIME(FTM_ALLOCATE_FACE); face = new LLFace(this, mVObjp); } @@ -370,11 +355,8 @@ LLFace* LLDrawable::addFace(const LLTextureEntry *te, LLViewerTexture *texturep) LLFace *face; - { - LL_RECORD_BLOCK_TIME(FTM_ALLOCATE_FACE); - face = new LLFace(this, mVObjp); - } - + face = new LLFace(this, mVObjp); + face->setTEOffset(mFaces.size()); face->setTexture(texturep); face->setPoolType(gPipeline.getPoolTypeFromTE(te, texturep)); @@ -1333,10 +1315,7 @@ void LLSpatialBridge::updateSpatialExtents() LLSpatialGroup* root = (LLSpatialGroup*) mOctree->getListener(0); - { - LL_RECORD_BLOCK_TIME(FTM_CULL_REBOUND); - root->rebound(); - } + root->rebound(); const LLVector4a* root_bounds = root->getBounds(); LLVector4a offset; diff --git a/indra/newview/lldrawable.h b/indra/newview/lldrawable.h index 14d782d6f2..6002e3e0dd 100644 --- a/indra/newview/lldrawable.h +++ b/indra/newview/lldrawable.h @@ -59,14 +59,13 @@ const U32 SILHOUETTE_HIGHLIGHT = 0; // All data for new renderer goes into this class. LL_ALIGN_PREFIX(16) -class LLDrawable -: public LLViewerOctreeEntryData, - public LLTrace::MemTrackable +class LLDrawable + : public LLViewerOctreeEntryData { + LL_ALIGN_NEW; public: LLDrawable(const LLDrawable& rhs) - : LLTrace::MemTrackable("LLDrawable"), - LLViewerOctreeEntryData(rhs) + : LLViewerOctreeEntryData(rhs) { *this = rhs; } diff --git a/indra/newview/lldrawpoolalpha.cpp b/indra/newview/lldrawpoolalpha.cpp index 369d7a6bb8..34f9bfe35d 100644 --- a/indra/newview/lldrawpoolalpha.cpp +++ b/indra/newview/lldrawpoolalpha.cpp @@ -53,10 +53,6 @@ BOOL LLDrawPoolAlpha::sShowDebugAlpha = FALSE; static BOOL deferred_render = FALSE; -static LLTrace::BlockTimerStatHandle FTM_RENDER_ALPHA_SETUP("Alpha Setup"); -static LLTrace::BlockTimerStatHandle FTM_RENDER_ALPHA_GROUP_LOOP("Alpha Group"); -static LLTrace::BlockTimerStatHandle FTM_RENDER_ALPHA_DEFERRED("Alpha Deferred"); - LLDrawPoolAlpha::LLDrawPoolAlpha(U32 type) : LLRenderPass(type), current_shader(NULL), target_shader(NULL), simple_shader(NULL), fullbright_shader(NULL), emissive_shader(NULL), @@ -98,7 +94,7 @@ S32 LLDrawPoolAlpha::getNumPostDeferredPasses() void LLDrawPoolAlpha::beginPostDeferredPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_DEFERRED); + LL_PROFILE_ZONE_SCOPED; F32 gamma = gSavedSettings.getF32("RenderDeferredDisplayGamma"); @@ -153,7 +149,7 @@ void LLDrawPoolAlpha::beginPostDeferredPass(S32 pass) void LLDrawPoolAlpha::endPostDeferredPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_DEFERRED); + LL_PROFILE_ZONE_SCOPED; if (pass == 1 && !LLPipeline::sImpostorRender) { @@ -168,13 +164,13 @@ void LLDrawPoolAlpha::endPostDeferredPass(S32 pass) void LLDrawPoolAlpha::renderPostDeferred(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_DEFERRED); + LL_PROFILE_ZONE_SCOPED; render(pass); } void LLDrawPoolAlpha::beginRenderPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_SETUP); + LL_PROFILE_ZONE_SCOPED; simple_shader = (LLPipeline::sImpostorRender) ? &gObjectSimpleImpostorProgram : (LLPipeline::sUnderWaterRender) ? &gObjectSimpleWaterProgram : &gObjectSimpleProgram; @@ -225,7 +221,7 @@ void LLDrawPoolAlpha::beginRenderPass(S32 pass) void LLDrawPoolAlpha::endRenderPass( S32 pass ) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_SETUP); + LL_PROFILE_ZONE_SCOPED; LLRenderPass::endRenderPass(pass); if(gPipeline.canUseWindLightShaders()) @@ -522,6 +518,7 @@ void LLDrawPoolAlpha::renderEmissives(U32 mask, std::vector& emissi void LLDrawPoolAlpha::renderAlpha(U32 mask, S32 pass) { + LL_PROFILE_ZONE_SCOPED; BOOL batch_fullbrights = gSavedSettings.getBOOL("RenderAlphaBatchFullbrights"); BOOL batch_emissives = gSavedSettings.getBOOL("RenderAlphaBatchEmissives"); BOOL initialized_lighting = FALSE; @@ -529,6 +526,7 @@ void LLDrawPoolAlpha::renderAlpha(U32 mask, S32 pass) for (LLCullResult::sg_iterator i = gPipeline.beginAlphaGroups(); i != gPipeline.endAlphaGroups(); ++i) { + LL_PROFILE_ZONE_NAMED("renderAlpha - group"); LLSpatialGroup* group = *i; llassert(group); llassert(group->getSpatialPartition()); @@ -546,9 +544,6 @@ void LLDrawPoolAlpha::renderAlpha(U32 mask, S32 pass) bool draw_glow_for_this_partition = mShaderLevel > 0; // no shaders = no glow. - - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_GROUP_LOOP); - bool disable_cull = is_particle_or_hud_particle; LLGLDisable cull(disable_cull ? GL_CULL_FACE : 0); diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index 52d308f6bd..5b51e9db24 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -103,8 +103,6 @@ S32 normal_channel = -1; S32 specular_channel = -1; S32 cube_channel = -1; -static LLTrace::BlockTimerStatHandle FTM_SHADOW_AVATAR("Avatar Shadow"); - LLDrawPoolAvatar::LLDrawPoolAvatar(U32 type) : LLFacePool(type) { @@ -196,7 +194,7 @@ LLMatrix4& LLDrawPoolAvatar::getModelView() void LLDrawPoolAvatar::beginDeferredPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_CHARACTERS); + LL_PROFILE_ZONE_SCOPED; sSkipTransparent = TRUE; is_deferred_render = true; @@ -231,7 +229,7 @@ void LLDrawPoolAvatar::beginDeferredPass(S32 pass) void LLDrawPoolAvatar::endDeferredPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_CHARACTERS); + LL_PROFILE_ZONE_SCOPED; sSkipTransparent = FALSE; is_deferred_render = false; @@ -454,229 +452,220 @@ S32 LLDrawPoolAvatar::getNumShadowPasses() void LLDrawPoolAvatar::beginShadowPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); + LL_PROFILE_ZONE_SCOPED; + + if (pass == SHADOW_PASS_AVATAR_OPAQUE) { - LL_PROFILE_ZONE_SCOPED; + sVertexProgram = &gDeferredAvatarShadowProgram; - if (pass == SHADOW_PASS_AVATAR_OPAQUE) + if ((sShaderLevel > 0)) // for hardware blending { - sVertexProgram = &gDeferredAvatarShadowProgram; - - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } - - gGL.diffuseColor4f(1, 1, 1, 1); + sRenderingSkinned = TRUE; + sVertexProgram->bind(); } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) - { - sVertexProgram = &gDeferredAvatarAlphaShadowProgram; - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) - { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } - - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) + { + sVertexProgram = &gDeferredAvatarAlphaShadowProgram; - gGL.diffuseColor4f(1, 1, 1, 1); - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) { - sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram; + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) - { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) + { + sVertexProgram = &gDeferredAvatarAlphaMaskShadowProgram; - gGL.diffuseColor4f(1, 1, 1, 1); - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) { - sVertexProgram = &gDeferredAttachmentAlphaShadowProgram; + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) - { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) + { + sVertexProgram = &gDeferredAttachmentAlphaShadowProgram; - gGL.diffuseColor4f(1, 1, 1, 1); - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) { - sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram; + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } - // bind diffuse tex so we can reference the alpha channel... - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) - { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + if ((sShaderLevel > 0)) // for hardware blending + { + sRenderingSkinned = TRUE; + sVertexProgram->bind(); + } - if ((sShaderLevel > 0)) // for hardware blending - { - sRenderingSkinned = TRUE; - sVertexProgram->bind(); - } + gGL.diffuseColor4f(1, 1, 1, 1); + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) + { + sVertexProgram = &gDeferredAttachmentAlphaMaskShadowProgram; - gGL.diffuseColor4f(1, 1, 1, 1); + // bind diffuse tex so we can reference the alpha channel... + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); } - else // SHADOW_PASS_ATTACHMENT_OPAQUE + + if ((sShaderLevel > 0)) // for hardware blending { - sVertexProgram = &gDeferredAttachmentShadowProgram; - S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); - sDiffuseChannel = 0; - if (loc != -1) - { - sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); - } + sRenderingSkinned = TRUE; sVertexProgram->bind(); } + + gGL.diffuseColor4f(1, 1, 1, 1); + } + else // SHADOW_PASS_ATTACHMENT_OPAQUE + { + sVertexProgram = &gDeferredAttachmentShadowProgram; + S32 loc = sVertexProgram->getUniformLocation(LLViewerShaderMgr::DIFFUSE_MAP); + sDiffuseChannel = 0; + if (loc != -1) + { + sDiffuseChannel = sVertexProgram->enableTexture(LLViewerShaderMgr::DIFFUSE_MAP); + } + sVertexProgram->bind(); } } void LLDrawPoolAvatar::endShadowPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); - { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED; - if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE) - { - LLVertexBuffer::unbind(); - } + if (pass == SHADOW_PASS_ATTACHMENT_OPAQUE) + { + LLVertexBuffer::unbind(); + } - if (sShaderLevel > 0) - { - sVertexProgram->unbind(); - } - sVertexProgram = NULL; - sRenderingSkinned = FALSE; - LLDrawPoolAvatar::sShadowPass = -1; + if (sShaderLevel > 0) + { + sVertexProgram->unbind(); } + sVertexProgram = NULL; + sRenderingSkinned = FALSE; + LLDrawPoolAvatar::sShadowPass = -1; } void LLDrawPoolAvatar::renderShadow(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_SHADOW_AVATAR); - { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED; - if (mDrawFace.empty()) - { - return; - } + if (mDrawFace.empty()) + { + return; + } - const LLFace *facep = mDrawFace[0]; - if (!facep->getDrawable()) - { - return; - } - LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); + const LLFace *facep = mDrawFace[0]; + if (!facep->getDrawable()) + { + return; + } + LLVOAvatar *avatarp = (LLVOAvatar *)facep->getDrawable()->getVObj().get(); - if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull()) - { - return; - } - LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance(); - BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor(); - if (impostor || (oa == LLVOAvatar::AOA_INVISIBLE)) - { - // No shadows for impostored (including jellydolled) or invisible avs. - return; - } + if (avatarp->isDead() || avatarp->isUIAvatar() || avatarp->mDrawable.isNull()) + { + return; + } + LLVOAvatar::AvatarOverallAppearance oa = avatarp->getOverallAppearance(); + BOOL impostor = !LLPipeline::sImpostorRender && avatarp->isImpostor(); + if (impostor || (oa == LLVOAvatar::AOA_INVISIBLE)) + { + // No shadows for impostored (including jellydolled) or invisible avs. + return; + } - LLDrawPoolAvatar::sShadowPass = pass; + LLDrawPoolAvatar::sShadowPass = pass; - if (pass == SHADOW_PASS_AVATAR_OPAQUE) - { - LLDrawPoolAvatar::sSkipTransparent = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipTransparent = false; - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) - { - LLDrawPoolAvatar::sSkipOpaque = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) - { - LLDrawPoolAvatar::sSkipOpaque = true; - avatarp->renderSkinned(); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha - { - LLDrawPoolAvatar::sSkipOpaque = true; - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA); - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE); - renderRigged(avatarp, RIGGED_ALPHA); - renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA); - renderRigged(avatarp, RIGGED_GLOW); - renderRigged(avatarp, RIGGED_SPECMAP_BLEND); - renderRigged(avatarp, RIGGED_NORMMAP_BLEND); - renderRigged(avatarp, RIGGED_NORMSPEC_BLEND); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask - { - LLDrawPoolAvatar::sSkipOpaque = true; - renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK); - renderRigged(avatarp, RIGGED_NORMMAP_MASK); - renderRigged(avatarp, RIGGED_SPECMAP_MASK); - renderRigged(avatarp, RIGGED_NORMSPEC_MASK); - renderRigged(avatarp, RIGGED_GLOW); - LLDrawPoolAvatar::sSkipOpaque = false; - } - else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE - { - LLDrawPoolAvatar::sSkipTransparent = true; - renderRigged(avatarp, RIGGED_MATERIAL); - renderRigged(avatarp, RIGGED_SPECMAP); - renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE); - renderRigged(avatarp, RIGGED_NORMMAP); - renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE); - renderRigged(avatarp, RIGGED_NORMSPEC); - renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE); - renderRigged(avatarp, RIGGED_SIMPLE); - renderRigged(avatarp, RIGGED_FULLBRIGHT); - renderRigged(avatarp, RIGGED_SHINY); - renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY); - renderRigged(avatarp, RIGGED_GLOW); - renderRigged(avatarp, RIGGED_DEFERRED_BUMP); - renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE); - LLDrawPoolAvatar::sSkipTransparent = false; - } + if (pass == SHADOW_PASS_AVATAR_OPAQUE) + { + LLDrawPoolAvatar::sSkipTransparent = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipTransparent = false; + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_BLEND) + { + LLDrawPoolAvatar::sSkipOpaque = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_AVATAR_ALPHA_MASK) + { + LLDrawPoolAvatar::sSkipOpaque = true; + avatarp->renderSkinned(); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_BLEND) // rigged alpha + { + LLDrawPoolAvatar::sSkipOpaque = true; + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA); + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_EMISSIVE); + renderRigged(avatarp, RIGGED_ALPHA); + renderRigged(avatarp, RIGGED_FULLBRIGHT_ALPHA); + renderRigged(avatarp, RIGGED_GLOW); + renderRigged(avatarp, RIGGED_SPECMAP_BLEND); + renderRigged(avatarp, RIGGED_NORMMAP_BLEND); + renderRigged(avatarp, RIGGED_NORMSPEC_BLEND); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else if (pass == SHADOW_PASS_ATTACHMENT_ALPHA_MASK) // rigged alpha mask + { + LLDrawPoolAvatar::sSkipOpaque = true; + renderRigged(avatarp, RIGGED_MATERIAL_ALPHA_MASK); + renderRigged(avatarp, RIGGED_NORMMAP_MASK); + renderRigged(avatarp, RIGGED_SPECMAP_MASK); + renderRigged(avatarp, RIGGED_NORMSPEC_MASK); + renderRigged(avatarp, RIGGED_GLOW); + LLDrawPoolAvatar::sSkipOpaque = false; + } + else // rigged opaque (SHADOW_PASS_ATTACHMENT_OPAQUE + { + LLDrawPoolAvatar::sSkipTransparent = true; + renderRigged(avatarp, RIGGED_MATERIAL); + renderRigged(avatarp, RIGGED_SPECMAP); + renderRigged(avatarp, RIGGED_SPECMAP_EMISSIVE); + renderRigged(avatarp, RIGGED_NORMMAP); + renderRigged(avatarp, RIGGED_NORMMAP_EMISSIVE); + renderRigged(avatarp, RIGGED_NORMSPEC); + renderRigged(avatarp, RIGGED_NORMSPEC_EMISSIVE); + renderRigged(avatarp, RIGGED_SIMPLE); + renderRigged(avatarp, RIGGED_FULLBRIGHT); + renderRigged(avatarp, RIGGED_SHINY); + renderRigged(avatarp, RIGGED_FULLBRIGHT_SHINY); + renderRigged(avatarp, RIGGED_GLOW); + renderRigged(avatarp, RIGGED_DEFERRED_BUMP); + renderRigged(avatarp, RIGGED_DEFERRED_SIMPLE); + LLDrawPoolAvatar::sSkipTransparent = false; } } @@ -712,7 +701,7 @@ S32 LLDrawPoolAvatar::getNumDeferredPasses() void LLDrawPoolAvatar::render(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_CHARACTERS); + LL_PROFILE_ZONE_SCOPED; if (LLPipeline::sImpostorRender) { renderAvatars(NULL, pass+2); @@ -724,7 +713,7 @@ void LLDrawPoolAvatar::render(S32 pass) void LLDrawPoolAvatar::beginRenderPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_CHARACTERS); + LL_PROFILE_ZONE_SCOPED; //reset vertex buffer mappings LLVertexBuffer::unbind(); @@ -775,7 +764,7 @@ void LLDrawPoolAvatar::beginRenderPass(S32 pass) void LLDrawPoolAvatar::endRenderPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_CHARACTERS); + LL_PROFILE_ZONE_SCOPED; if (LLPipeline::sImpostorRender) { @@ -1542,13 +1531,8 @@ void LLDrawPoolAvatar::endDeferredSkinned() gGL.getTexUnit(0)->activate(); } -static LLTrace::BlockTimerStatHandle FTM_RENDER_AVATARS("renderAvatars"); - - void LLDrawPoolAvatar::renderAvatars(LLVOAvatar* single_avatar, S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_AVATARS); - if (pass == -1) { for (S32 i = 1; i < getNumPasses(); i++) @@ -1588,6 +1572,8 @@ void LLDrawPoolAvatar::renderAvatars(LLVOAvatar* single_avatar, S32 pass) return; } + LL_RECORD_BLOCK_TIME(FTM_RENDER_CHARACTERS); + if (!single_avatar && !avatarp->isFullyLoaded() ) { if (pass==0 && (!gPipeline.hasRenderType(LLPipeline::RENDER_TYPE_PARTICLES) || LLViewerPartSim::getMaxPartCount() <= 0)) diff --git a/indra/newview/lldrawpoolbump.cpp b/indra/newview/lldrawpoolbump.cpp index b08fbcbd89..8f3b0c99b4 100644 --- a/indra/newview/lldrawpoolbump.cpp +++ b/indra/newview/lldrawpoolbump.cpp @@ -1116,11 +1116,10 @@ LLViewerTexture* LLBumpImageList::getBrightnessDarknessImage(LLViewerFetchedText } -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_STANDARD_LOADED("Bump Standard Callback"); - // static void LLBumpImageList::onSourceBrightnessLoaded( BOOL success, LLViewerFetchedTexture *src_vi, LLImageRaw* src, LLImageRaw* aux_src, S32 discard_level, BOOL final, void* userdata ) { + LL_PROFILE_ZONE_SCOPED; LLUUID* source_asset_id = (LLUUID*)userdata; LLBumpImageList::onSourceLoaded( success, src_vi, src, *source_asset_id, BE_BRIGHTNESS ); if( final ) @@ -1140,22 +1139,17 @@ void LLBumpImageList::onSourceDarknessLoaded( BOOL success, LLViewerFetchedTextu } } -static LLTrace::BlockTimerStatHandle FTM_BUMP_GEN_NORMAL("Generate Normal Map"); -static LLTrace::BlockTimerStatHandle FTM_BUMP_CREATE_TEXTURE("Create GL Normal Map"); - void LLBumpImageList::onSourceStandardLoaded( BOOL success, LLViewerFetchedTexture* src_vi, LLImageRaw* src, LLImageRaw* aux_src, S32 discard_level, BOOL final, void* userdata) { if (success && LLPipeline::sRenderDeferred) { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_STANDARD_LOADED); + LL_PROFILE_ZONE_SCOPED; LLPointer nrm_image = new LLImageRaw(src->getWidth(), src->getHeight(), 4); { - LL_RECORD_BLOCK_TIME(FTM_BUMP_GEN_NORMAL); generateNormalMapFromAlpha(src, nrm_image); } src_vi->setExplicitFormat(GL_RGBA, GL_RGBA); { - LL_RECORD_BLOCK_TIME(FTM_BUMP_CREATE_TEXTURE); src_vi->createGLTexture(src_vi->getDiscardLevel(), nrm_image); } } @@ -1216,28 +1210,18 @@ void LLBumpImageList::generateNormalMapFromAlpha(LLImageRaw* src, LLImageRaw* nr } } - -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_LOADED("Bump Source Loaded"); -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_ENTRIES_UPDATE("Entries Update"); -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_MIN_MAX("Min/Max"); -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_RGB2LUM("RGB to Luminance"); -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_RESCALE("Rescale"); -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_GEN_NORMAL("Generate Normal"); -static LLTrace::BlockTimerStatHandle FTM_BUMP_SOURCE_CREATE("Bump Source Create"); - // static void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLImageRaw* src, LLUUID& source_asset_id, EBumpEffect bump_code ) { if( success ) { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_LOADED); + LL_PROFILE_ZONE_SCOPED; bump_image_map_t& entries_list(bump_code == BE_BRIGHTNESS ? gBumpImageList.mBrightnessEntries : gBumpImageList.mDarknessEntries ); bump_image_map_t::iterator iter = entries_list.find(source_asset_id); { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_ENTRIES_UPDATE); if (iter == entries_list.end() || iter->second.isNull() || iter->second->getWidth() != src->getWidth() || @@ -1280,7 +1264,6 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI case 1: case 2: { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_MIN_MAX); if( src_data_size == dst_data_size * src_components ) { for( S32 i = 0, j=0; i < dst_data_size; i++, j+= src_components ) @@ -1306,7 +1289,6 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI case 3: case 4: { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_RGB2LUM); if( src_data_size == dst_data_size * src_components ) { for( S32 i = 0, j=0; i < dst_data_size; i++, j+= src_components ) @@ -1339,7 +1321,6 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI if( maximum > minimum ) { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_RESCALE); U8 bias_and_scale_lut[256]; F32 twice_one_over_range = 2.f / (maximum - minimum); S32 i; @@ -1375,7 +1356,6 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI if (!LLPipeline::sRenderDeferred) { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_CREATE); bump->setExplicitFormat(GL_ALPHA8, GL_ALPHA); bump->createGLTexture(0, dst_image); } @@ -1386,13 +1366,11 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI bump->getGLTexture()->setAllowCompression(false); { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_CREATE); bump->setExplicitFormat(GL_RGBA8, GL_ALPHA); bump->createGLTexture(0, dst_image); } { - LL_RECORD_BLOCK_TIME(FTM_BUMP_SOURCE_GEN_NORMAL); gPipeline.mScreen.bindTarget(); LLGLDepthTest depth(GL_FALSE); diff --git a/indra/newview/lldrawpoolsimple.cpp b/indra/newview/lldrawpoolsimple.cpp index 843288cfb0..320160d10d 100644 --- a/indra/newview/lldrawpoolsimple.cpp +++ b/indra/newview/lldrawpoolsimple.cpp @@ -57,8 +57,6 @@ void LLDrawPoolGlow::beginPostDeferredPass(S32 pass) } } -static LLTrace::BlockTimerStatHandle FTM_RENDER_GLOW_PUSH("Glow Push"); - void LLDrawPoolGlow::renderPostDeferred(S32 pass) { LL_RECORD_BLOCK_TIME(FTM_RENDER_GLOW); @@ -73,10 +71,7 @@ void LLDrawPoolGlow::renderPostDeferred(S32 pass) LLGLDepthTest depth(GL_TRUE, GL_FALSE); gGL.setColorMask(false, true); - { - LL_RECORD_BLOCK_TIME(FTM_RENDER_GLOW_PUSH); - pushBatches(LLRenderPass::PASS_GLOW, getVertexDataMask() | LLVertexBuffer::MAP_TEXTURE_INDEX, TRUE, TRUE); - } + pushBatches(LLRenderPass::PASS_GLOW, getVertexDataMask() | LLVertexBuffer::MAP_TEXTURE_INDEX, TRUE, TRUE); gGL.setColorMask(true, false); gGL.setSceneBlendType(LLRender::BT_ALPHA); @@ -309,7 +304,7 @@ void LLDrawPoolAlphaMask::endRenderPass(S32 pass) void LLDrawPoolAlphaMask::render(S32 pass) { LLGLDisable blend(GL_BLEND); - LL_RECORD_BLOCK_TIME(FTM_RENDER_ALPHA_MASK); + LL_PROFILE_ZONE_SCOPED; if (mShaderLevel > 0) { diff --git a/indra/newview/lldrawpoolterrain.cpp b/indra/newview/lldrawpoolterrain.cpp index 34a8b6b2cc..9cea8f5460 100644 --- a/indra/newview/lldrawpoolterrain.cpp +++ b/indra/newview/lldrawpoolterrain.cpp @@ -920,6 +920,7 @@ void LLDrawPoolTerrain::renderOwnership() void LLDrawPoolTerrain::dirtyTextures(const std::set& textures) { + LL_PROFILE_ZONE_SCOPED; LLViewerFetchedTexture* tex = LLViewerTextureManager::staticCastToFetchedTexture(mTexturep) ; if (tex && textures.find(tex) != textures.end()) { diff --git a/indra/newview/lldrawpooltree.cpp b/indra/newview/lldrawpooltree.cpp index a1ff020068..202f648e3f 100644 --- a/indra/newview/lldrawpooltree.cpp +++ b/indra/newview/lldrawpooltree.cpp @@ -42,7 +42,6 @@ S32 LLDrawPoolTree::sDiffTex = 0; static LLGLSLShader* shader = NULL; -static LLTrace::BlockTimerStatHandle FTM_SHADOW_TREE("Tree Shadow"); LLDrawPoolTree::LLDrawPoolTree(LLViewerTexture *texturep) : LLFacePool(POOL_TREE), @@ -84,7 +83,7 @@ void LLDrawPoolTree::beginRenderPass(S32 pass) void LLDrawPoolTree::render(S32 pass) { - LL_RECORD_BLOCK_TIME(LLPipeline::sShadowRender ? FTM_SHADOW_TREE : FTM_RENDER_TREES); + LL_PROFILE_ZONE_SCOPED; if (mDrawFace.empty()) { @@ -169,7 +168,7 @@ void LLDrawPoolTree::endDeferredPass(S32 pass) //============================================ void LLDrawPoolTree::beginShadowPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_SHADOW_TREE); + LL_PROFILE_ZONE_SCOPED; glPolygonOffset(gSavedSettings.getF32("RenderDeferredTreeShadowOffset"), gSavedSettings.getF32("RenderDeferredTreeShadowBias")); @@ -188,7 +187,7 @@ void LLDrawPoolTree::renderShadow(S32 pass) void LLDrawPoolTree::endShadowPass(S32 pass) { - LL_RECORD_BLOCK_TIME(FTM_SHADOW_TREE); + LL_PROFILE_ZONE_SCOPED; glPolygonOffset(gSavedSettings.getF32("RenderDeferredSpotShadowOffset"), gSavedSettings.getF32("RenderDeferredSpotShadowBias")); diff --git a/indra/newview/lldynamictexture.h b/indra/newview/lldynamictexture.h index 4bd74a8425..caedf928c3 100644 --- a/indra/newview/lldynamictexture.h +++ b/indra/newview/lldynamictexture.h @@ -35,16 +35,8 @@ class LLViewerDynamicTexture : public LLViewerTexture { + LL_ALIGN_NEW public: - void* operator new(size_t size) - { - return LLTrace::MemTrackable::aligned_new<16>(size); - } - - void operator delete(void* ptr, size_t size) - { - LLTrace::MemTrackable::aligned_delete<16>(ptr, size); - } enum { diff --git a/indra/newview/llenvironment.cpp b/indra/newview/llenvironment.cpp index 69d3075928..dba24b3d02 100644 --- a/indra/newview/llenvironment.cpp +++ b/indra/newview/llenvironment.cpp @@ -105,7 +105,6 @@ namespace //--------------------------------------------------------------------- LLTrace::BlockTimerStatHandle FTM_ENVIRONMENT_UPDATE("Update Environment Tick"); - LLTrace::BlockTimerStatHandle FTM_SHADER_PARAM_UPDATE("Update Shader Parameters"); LLSettingsBase::Seconds DEFAULT_UPDATE_THRESHOLD(10.0); const LLSettingsBase::Seconds MINIMUM_SPANLENGTH(0.01f); @@ -825,7 +824,7 @@ std::string env_selection_to_string(LLEnvironment::EnvSelection_t sel) #undef RTNENUM } - +LLEnvironment* LLSimpleton::sInstance = nullptr; //------------------------------------------------------------------------- LLEnvironment::LLEnvironment(): mCloudScrollDelta(), @@ -879,6 +878,7 @@ void LLEnvironment::cleanupSingleton() LLEnvironment::~LLEnvironment() { + cleanupSingleton(); } bool LLEnvironment::canEdit() const @@ -1655,7 +1655,7 @@ void LLEnvironment::updateCloudScroll() // static void LLEnvironment::updateGLVariablesForSettings(LLShaderUniforms* uniforms, const LLSettingsBase::ptr_t &psetting) { - LL_RECORD_BLOCK_TIME(FTM_SHADER_PARAM_UPDATE); + LL_PROFILE_ZONE_SCOPED; for (int i = 0; i < LLGLSLShader::SG_COUNT; ++i) { diff --git a/indra/newview/llenvironment.h b/indra/newview/llenvironment.h index 3568fbcfd1..0ec06402f8 100644 --- a/indra/newview/llenvironment.h +++ b/indra/newview/llenvironment.h @@ -47,11 +47,11 @@ class LLViewerCamera; class LLParcel; //------------------------------------------------------------------------- -class LLEnvironment : public LLSingleton +class LLEnvironment : public LLSimpleton { - LLSINGLETON_C11(LLEnvironment); LOG_CLASS(LLEnvironment); public: + LLEnvironment(); static const F64Seconds TRANSITION_INSTANT; static const F64Seconds TRANSITION_FAST; @@ -115,7 +115,7 @@ public: typedef std::array altitude_list_t; typedef std::vector altitudes_vect_t; - virtual ~LLEnvironment(); + ~LLEnvironment(); bool canEdit() const; bool isExtendedEnvironmentEnabled() const; @@ -337,9 +337,10 @@ public: DayInstance::ptr_t getSelectedEnvironmentInstance(); DayInstance::ptr_t getSharedEnvironmentInstance(); + void initSingleton(); + protected: - virtual void initSingleton() override; - virtual void cleanupSingleton() override; + void cleanupSingleton(); private: diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 34448a780d..88b958d24a 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -127,6 +127,7 @@ void planarProjection(LLVector2 &tc, const LLVector4a& normal, void LLFace::init(LLDrawable* drawablep, LLViewerObject* objp) { + LL_PROFILE_ZONE_SCOPED; mLastUpdateTime = gFrameTimeSeconds; mLastMoveTime = 0.f; mLastSkinTime = gFrameTimeSeconds; @@ -1206,12 +1207,10 @@ bool LLFace::canRenderAsMask() } -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_VOLUME("Volume VB Cache"); - //static void LLFace::cacheFaceInVRAM(const LLVolumeFace& vf) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_VOLUME); + LL_PROFILE_ZONE_SCOPED; U32 mask = LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0 | LLVertexBuffer::MAP_TANGENT | LLVertexBuffer::MAP_NORMAL; @@ -1273,41 +1272,13 @@ void push_for_transform(LLVertexBuffer* buff, U32 source_count, U32 dest_count) } } -static LLTrace::BlockTimerStatHandle FTM_FACE_GET_GEOM("Face Geom"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_POSITION("Position"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_NORMAL("Normal"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_TEXTURE("Texture"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_COLOR("Color"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_EMISSIVE("Emissive"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_WEIGHTS("Weights"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_TANGENT("Binormal"); - -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_FEEDBACK("Face Feedback"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_FEEDBACK_POSITION("Feedback Position"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_FEEDBACK_NORMAL("Feedback Normal"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_FEEDBACK_TEXTURE("Feedback Texture"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_FEEDBACK_COLOR("Feedback Color"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_FEEDBACK_EMISSIVE("Feedback Emissive"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_FEEDBACK_BINORMAL("Feedback Binormal"); - -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_INDEX("Index"); -static LLTrace::BlockTimerStatHandle FTM_FACE_GEOM_INDEX_TAIL("Tail"); -static LLTrace::BlockTimerStatHandle FTM_FACE_POSITION_STORE("Pos"); -static LLTrace::BlockTimerStatHandle FTM_FACE_TEXTURE_INDEX_STORE("TexIdx"); -static LLTrace::BlockTimerStatHandle FTM_FACE_POSITION_PAD("Pad"); -static LLTrace::BlockTimerStatHandle FTM_FACE_TEX_DEFAULT("Default"); -static LLTrace::BlockTimerStatHandle FTM_FACE_TEX_QUICK("Quick"); -static LLTrace::BlockTimerStatHandle FTM_FACE_TEX_QUICK_NO_XFORM("No Xform"); -static LLTrace::BlockTimerStatHandle FTM_FACE_TEX_QUICK_XFORM("Xform"); -static LLTrace::BlockTimerStatHandle FTM_FACE_TEX_QUICK_PLANAR("Quick Planar"); - BOOL LLFace::getGeometryVolume(const LLVolume& volume, const S32 &f, const LLMatrix4& mat_vert_in, const LLMatrix3& mat_norm_in, const U16 &index_offset, bool force_rebuild) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GET_GEOM); + LL_PROFILE_ZONE_SCOPED; llassert(verify()); if (volume.getNumVolumeFaces() <= f) { @@ -1448,7 +1419,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, // INDICES if (full_rebuild) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_INDEX); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - indices"); mVertexBuffer->getIndexStrider(indicesp, mIndicesIndex, mIndicesCount, map_range); volatile __m128i* dst = (__m128i*) indicesp.get(); @@ -1464,7 +1435,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_INDEX_TAIL); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - indices tail"); U16* idx = (U16*) dst; for (S32 i = end*8; i < num_indices; ++i) @@ -1527,7 +1498,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, !volume.isUnique()) //source volume is NOT flexi { //use transform feedback to pack vertex buffer //gGLDebugLoggingEnabled = TRUE; - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_FEEDBACK); + + LL_PROFILE_ZONE_NAMED("getGeometryVolume - transform feedback"); LLGLEnable discard(GL_RASTERIZER_DISCARD); LLVertexBuffer* buff = (LLVertexBuffer*) vf.mVertexBuffer.get(); @@ -1545,7 +1517,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_pos) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_FEEDBACK_POSITION); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tf position"); gTransformPositionProgram.bind(); mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_VERTEX, mGeomIndex, mGeomCount); @@ -1570,7 +1542,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_color) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_FEEDBACK_COLOR); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tf color"); gTransformColorProgram.bind(); mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_COLOR, mGeomIndex, mGeomCount); @@ -1586,7 +1558,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_emissive) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_FEEDBACK_EMISSIVE); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tf emissive"); gTransformColorProgram.bind(); mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_EMISSIVE, mGeomIndex, mGeomCount); @@ -1607,7 +1579,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_normal) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_FEEDBACK_NORMAL); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tf normal"); gTransformNormalProgram.bind(); mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_NORMAL, mGeomIndex, mGeomCount); @@ -1620,7 +1592,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_tangent) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_TANGENT); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tf tangent"); gTransformTangentProgram.bind(); mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_TANGENT, mGeomIndex, mGeomCount); @@ -1633,7 +1605,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_tcoord) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_FEEDBACK_TEXTURE); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tf tcoord"); gTransformTexCoordProgram.bind(); mVertexBuffer->bindForFeedback(0, LLVertexBuffer::TYPE_TEXCOORD0, mGeomIndex, mGeomCount); @@ -1672,7 +1644,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_tcoord) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_TEXTURE); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tcoord"); //bump setup LLVector4a binormal_dir( -sin_ang, cos_ang, 0.f ); @@ -1795,18 +1767,18 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (texgen != LLTextureEntry::TEX_GEN_PLANAR) { - LL_RECORD_BLOCK_TIME(FTM_FACE_TEX_QUICK); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - texgen"); if (!do_tex_mat) { if (!do_xform) { - LL_RECORD_BLOCK_TIME(FTM_FACE_TEX_QUICK_NO_XFORM); + LL_PROFILE_ZONE_NAMED("ggv - texgen 1"); S32 tc_size = (num_vertices*2*sizeof(F32)+0xF) & ~0xF; LLVector4a::memcpyNonAliased16((F32*) tex_coords0.get(), (F32*) vf.mTexCoords, tc_size); } else { - LL_RECORD_BLOCK_TIME(FTM_FACE_TEX_QUICK_XFORM); + LL_PROFILE_ZONE_NAMED("ggv - texgen 2"); F32* dst = (F32*) tex_coords0.get(); LLVector4a* src = (LLVector4a*) vf.mTexCoords; @@ -1846,9 +1818,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, for (S32 i = 0; i < num_vertices; i++) { LLVector2 tc(vf.mTexCoords[i]); - //LLVector4a& norm = vf.mNormals[i]; - //LLVector4a& center = *(vf.mCenter); - + LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f); tmp = tmp * *mTextureMatrix; tc.mV[0] = tmp.mV[0]; @@ -1859,7 +1829,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } else { //no bump, tex gen planar - LL_RECORD_BLOCK_TIME(FTM_FACE_TEX_QUICK_PLANAR); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - texgen planar"); if (do_tex_mat) { for (S32 i = 0; i < num_vertices; i++) @@ -1904,7 +1874,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, } else { //bump mapped or has material, just do the whole expensive loop - LL_RECORD_BLOCK_TIME(FTM_FACE_TEX_DEFAULT); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - texgen default"); std::vector bump_tc; @@ -2062,7 +2032,6 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLVector4a* end = src+num_vertices; //LLVector4a* end_64 = end-4; - //LL_RECORD_TIME_BLOCK(FTM_FACE_GEOM_POSITION); llassert(num_vertices > 0); mVertexBuffer->getVertexStrider(vert, mGeomIndex, mGeomCount, map_range); @@ -2099,53 +2068,19 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLVector4a tmp; - { - //LL_RECORD_TIME_BLOCK(FTM_FACE_POSITION_STORE); - - /*if (num_vertices > 4) - { //more than 64 bytes - while (src < end_64) - { - _mm_prefetch((char*)src + 64, _MM_HINT_T0); - _mm_prefetch((char*)dst + 64, _MM_HINT_T0); - - mat_vert.affineTransform(*src, res0); - tmp.setSelectWithMask(mask, texIdx, res0); - tmp.store4a((F32*) dst); - - mat_vert.affineTransform(*(src+1), res1); - tmp.setSelectWithMask(mask, texIdx, res1); - tmp.store4a((F32*) dst+4); - - mat_vert.affineTransform(*(src+2), res2); - tmp.setSelectWithMask(mask, texIdx, res2); - tmp.store4a((F32*) dst+8); - - mat_vert.affineTransform(*(src+3), res3); - tmp.setSelectWithMask(mask, texIdx, res3); - tmp.store4a((F32*) dst+12); - - dst += 16; - src += 4; - } - }*/ - - while (src < end) - { - mat_vert.affineTransform(*src++, res0); - tmp.setSelectWithMask(mask, texIdx, res0); - tmp.store4a((F32*) dst); - dst += 4; - } + + while (src < end) + { + mat_vert.affineTransform(*src++, res0); + tmp.setSelectWithMask(mask, texIdx, res0); + tmp.store4a((F32*) dst); + dst += 4; } - + + while (dst < end_f32) { - //LL_RECORD_TIME_BLOCK(FTM_FACE_POSITION_PAD); - while (dst < end_f32) - { - res0.store4a((F32*) dst); - dst += 4; - } + res0.store4a((F32*) dst); + dst += 4; } if (map_range) @@ -2179,7 +2114,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_tangent) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_TANGENT); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - tangent"); mVertexBuffer->getTangentStrider(tangent, mGeomIndex, mGeomCount, map_range); F32* tangents = (F32*) tangent.get(); @@ -2212,7 +2147,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_weights && vf.mWeights) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_WEIGHTS); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - weight"); mVertexBuffer->getWeight4Strider(wght, mGeomIndex, mGeomCount, map_range); F32* weights = (F32*) wght.get(); LLVector4a::memcpyNonAliased16(weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32)); @@ -2224,7 +2159,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_color && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_COLOR) ) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_COLOR); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - color"); mVertexBuffer->getColorStrider(colors, mGeomIndex, mGeomCount, map_range); LLVector4a src; @@ -2255,7 +2190,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_emissive) { - LL_RECORD_BLOCK_TIME(FTM_FACE_GEOM_EMISSIVE); + LL_PROFILE_ZONE_NAMED("getGeometryVolume - emissive"); LLStrider emissive; mVertexBuffer->getEmissiveStrider(emissive, mGeomIndex, mGeomCount, map_range); diff --git a/indra/newview/llface.h b/indra/newview/llface.h index 9dd365e3dc..2e76c974fa 100644 --- a/indra/newview/llface.h +++ b/indra/newview/llface.h @@ -54,11 +54,11 @@ const F32 MIN_ALPHA_SIZE = 1024.f; const F32 MIN_TEX_ANIM_SIZE = 512.f; const U8 FACE_DO_NOT_BATCH_TEXTURES = 255; -class LLFace : public LLTrace::MemTrackableNonVirtual +class alignas(16) LLFace { + LL_ALIGN_NEW public: LLFace(const LLFace& rhs) - : LLTrace::MemTrackableNonVirtual("LLFace") { *this = rhs; } @@ -85,8 +85,8 @@ public: public: LLFace(LLDrawable* drawablep, LLViewerObject* objp) - : LLTrace::MemTrackableNonVirtual("LLFace") { + LL_PROFILE_ZONE_SCOPED; init(drawablep, objp); } ~LLFace() { destroy(); } diff --git a/indra/newview/llfasttimerview.cpp b/indra/newview/llfasttimerview.cpp index 239d162101..1605e4133d 100644 --- a/indra/newview/llfasttimerview.cpp +++ b/indra/newview/llfasttimerview.cpp @@ -716,7 +716,6 @@ void LLFastTimerView::exportCharts(const std::string& base, const std::string& t //====================================== buffer.clear(); - gGL.color3fv(base_col.mV); U32 count = 0; U32 total_count = base_execution.size(); @@ -1019,11 +1018,9 @@ void LLFastTimerView::printLineStats() } } -static LLTrace::BlockTimerStatHandle FTM_DRAW_LINE_GRAPH("Draw line graph"); - void LLFastTimerView::drawLineGraph() { - LL_RECORD_BLOCK_TIME(FTM_DRAW_LINE_GRAPH); + LL_PROFILE_ZONE_SCOPED; //draw line graph history gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE); LLLocalClipRect clip(mGraphRect); @@ -1062,6 +1059,7 @@ void LLFastTimerView::drawLineGraph() F32Seconds cur_max(0); U32 cur_max_calls = 0; + for(block_timer_tree_df_iterator_t it = LLTrace::begin_block_timer_tree_df(FTM_FRAME); it != LLTrace::end_block_timer_tree_df(); ++it) @@ -1096,6 +1094,7 @@ void LLFastTimerView::drawLineGraph() F32 call_scale_factor = (F32)mGraphRect.getHeight() / (F32)max_calls; F32 time_scale_factor = (F32)mGraphRect.getHeight() / max_time.value(); F32 hz_scale_factor = (F32) mGraphRect.getHeight() / (1.f / max_time.value()); + for (U32 j = mRecording.getNumRecordedPeriods(); j > 0; j--) @@ -1103,7 +1102,7 @@ void LLFastTimerView::drawLineGraph() LLTrace::Recording& recording = mRecording.getPrevRecording(j); F32Seconds time = llmax(recording.getSum(*idp), F64Seconds(0.000001)); U32 calls = recording.getSum(idp->callCount()); - + if (is_hover_timer) { //normalize to highlighted timer @@ -1450,6 +1449,7 @@ void LLFastTimerView::updateTotalTime() void LLFastTimerView::drawBars() { + LL_PROFILE_ZONE_SCOPED; LLLocalClipRect clip(mBarRect); S32 bar_height = mBarRect.getHeight() / (MAX_VISIBLE_HISTORY + 2); @@ -1527,11 +1527,9 @@ void LLFastTimerView::drawBars() gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE); } -static LLTrace::BlockTimerStatHandle FTM_UPDATE_TIMER_BAR_WIDTHS("Update timer bar widths"); - F32Seconds LLFastTimerView::updateTimerBarWidths(LLTrace::BlockTimerStatHandle* time_block, TimerBarRow& row, S32 history_index, U32& bar_index) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_TIMER_BAR_WIDTHS); + LL_PROFILE_ZONE_SCOPED; const F32Seconds self_time = history_index == -1 ? mRecording.getPeriodMean(time_block->selfTime(), RUNNING_AVERAGE_WIDTH) : mRecording.getPrevRecording(history_index).getSum(time_block->selfTime()); @@ -1555,11 +1553,9 @@ F32Seconds LLFastTimerView::updateTimerBarWidths(LLTrace::BlockTimerStatHandle* return full_time; } -static LLTrace::BlockTimerStatHandle FTM_UPDATE_TIMER_BAR_FRACTIONS("Update timer bar fractions"); - S32 LLFastTimerView::updateTimerBarOffsets(LLTrace::BlockTimerStatHandle* time_block, TimerBarRow& row, S32 timer_bar_index) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_TIMER_BAR_FRACTIONS); + LL_PROFILE_ZONE_SCOPED; TimerBar& timer_bar = row.mBars[timer_bar_index]; const F32Seconds bar_time = timer_bar.mTotalTime - timer_bar.mSelfTime; @@ -1620,6 +1616,7 @@ S32 LLFastTimerView::updateTimerBarOffsets(LLTrace::BlockTimerStatHandle* time_b S32 LLFastTimerView::drawBar(LLRect bar_rect, TimerBarRow& row, S32 image_width, S32 image_height, bool hovered, bool visible, S32 bar_index) { + LL_PROFILE_ZONE_SCOPED; TimerBar& timer_bar = row.mBars[bar_index]; LLTrace::BlockTimerStatHandle* time_block = timer_bar.mTimeBlock; diff --git a/indra/newview/llflexibleobject.cpp b/indra/newview/llflexibleobject.cpp index e075a311c2..0b0567b687 100644 --- a/indra/newview/llflexibleobject.cpp +++ b/indra/newview/llflexibleobject.cpp @@ -47,9 +47,6 @@ static const F32 SEC_PER_FLEXI_FRAME = 1.f / 60.f; // 60 flexi updates per secon /*static*/ F32 LLVolumeImplFlexible::sUpdateFactor = 1.0f; std::vector LLVolumeImplFlexible::sInstanceList; -static LLTrace::BlockTimerStatHandle FTM_FLEXIBLE_REBUILD("Rebuild"); -static LLTrace::BlockTimerStatHandle FTM_DO_FLEXIBLE_UPDATE("Flexible Update"); - // LLFlexibleObjectData::pack/unpack now in llprimitive.cpp //----------------------------------------------- @@ -95,7 +92,7 @@ LLVolumeImplFlexible::~LLVolumeImplFlexible() //static void LLVolumeImplFlexible::updateClass() { - LL_RECORD_BLOCK_TIME(FTM_DO_FLEXIBLE_UPDATE); + LL_PROFILE_ZONE_SCOPED; U64 virtual_frame_num = LLTimer::getElapsedSeconds() / SEC_PER_FLEXI_FRAME; for (std::vector::iterator iter = sInstanceList.begin(); @@ -429,7 +426,7 @@ inline S32 log2(S32 x) void LLVolumeImplFlexible::doFlexibleUpdate() { - LL_RECORD_BLOCK_TIME(FTM_DO_FLEXIBLE_UPDATE); + LL_PROFILE_ZONE_SCOPED; LLVolume* volume = mVO->getVolume(); LLPath *path = &volume->getPath(); if ((mSimulateRes == 0 || !mInitialized) && mVO->mDrawable->isVisible()) @@ -720,13 +717,12 @@ void LLVolumeImplFlexible::doFlexibleUpdate() mLastSegmentRotation = parentSegmentRotation; } -static LLTrace::BlockTimerStatHandle FTM_FLEXI_PREBUILD("Flexi Prebuild"); void LLVolumeImplFlexible::preRebuild() { if (!mUpdated) { - LL_RECORD_BLOCK_TIME(FTM_FLEXI_PREBUILD); + LL_PROFILE_ZONE_SCOPED; doFlexibleRebuild(false); } } @@ -752,6 +748,7 @@ void LLVolumeImplFlexible::onSetScale(const LLVector3& scale, BOOL damped) BOOL LLVolumeImplFlexible::doUpdateGeometry(LLDrawable *drawable) { + LL_PROFILE_ZONE_SCOPED; LLVOVolume *volume = (LLVOVolume*)mVO; if (mVO->isAttachment()) @@ -789,7 +786,6 @@ BOOL LLVolumeImplFlexible::doUpdateGeometry(LLDrawable *drawable) if (mRenderRes > -1) { - LL_RECORD_BLOCK_TIME(FTM_DO_FLEXIBLE_UPDATE); doFlexibleUpdate(); } @@ -809,7 +805,6 @@ BOOL LLVolumeImplFlexible::doUpdateGeometry(LLDrawable *drawable) volume->mDrawable->setState(LLDrawable::REBUILD_VOLUME); volume->dirtySpatialGroup(); { - LL_RECORD_BLOCK_TIME(FTM_FLEXIBLE_REBUILD); doFlexibleRebuild(volume->mVolumeChanged); } volume->genBBoxes(isVolumeGlobal()); diff --git a/indra/newview/llgroupmgr.cpp b/indra/newview/llgroupmgr.cpp index 32f88b49ac..7f65153879 100644 --- a/indra/newview/llgroupmgr.cpp +++ b/indra/newview/llgroupmgr.cpp @@ -944,12 +944,10 @@ static void formatDateString(std::string &date_string) } } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_GROUP_MEMBERS_REPLY("Process Group Members"); - // static void LLGroupMgr::processGroupMembersReply(LLMessageSystem* msg, void** data) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_GROUP_MEMBERS_REPLY); + LL_PROFILE_ZONE_SCOPED; LL_DEBUGS("GrpMgr") << "LLGroupMgr::processGroupMembersReply" << LL_ENDL; LLUUID agent_id; @@ -1054,12 +1052,10 @@ void LLGroupMgr::processGroupMembersReply(LLMessageSystem* msg, void** data) LLGroupMgr::getInstance()->notifyObservers(GC_MEMBER_DATA); } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_GROUP_PROPERTIES_REPLY("Process Group Properties"); - //static void LLGroupMgr::processGroupPropertiesReply(LLMessageSystem* msg, void** data) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_GROUP_PROPERTIES_REPLY); + LL_PROFILE_ZONE_SCOPED; LL_DEBUGS("GrpMgr") << "LLGroupMgr::processGroupPropertiesReply" << LL_ENDL; if (!msg) @@ -1139,11 +1135,10 @@ void LLGroupMgr::processGroupPropertiesReply(LLMessageSystem* msg, void** data) LLGroupMgr::getInstance()->notifyObservers(GC_PROPERTIES); } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_GROUP_ROLE_DATA_REPLY("Process Group Role Data"); // static void LLGroupMgr::processGroupRoleDataReply(LLMessageSystem* msg, void** data) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_GROUP_ROLE_DATA_REPLY); + LL_PROFILE_ZONE_SCOPED; LL_DEBUGS("GrpMgr") << "LLGroupMgr::processGroupRoleDataReply" << LL_ENDL; LLUUID agent_id; @@ -1227,11 +1222,10 @@ void LLGroupMgr::processGroupRoleDataReply(LLMessageSystem* msg, void** data) LLGroupMgr::getInstance()->notifyObservers(GC_ROLE_DATA); } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_GROUP_ROLE_MEMBERS_REPLY("Process Group Role Members"); // static void LLGroupMgr::processGroupRoleMembersReply(LLMessageSystem* msg, void** data) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_GROUP_ROLE_MEMBERS_REPLY); + LL_PROFILE_ZONE_SCOPED; LL_DEBUGS("GrpMgr") << "LLGroupMgr::processGroupRoleMembersReply" << LL_ENDL; LLUUID agent_id; diff --git a/indra/newview/llinventoryfilter.cpp b/indra/newview/llinventoryfilter.cpp index 411311bbea..84018655d0 100644 --- a/indra/newview/llinventoryfilter.cpp +++ b/indra/newview/llinventoryfilter.cpp @@ -47,8 +47,6 @@ #include "llclipboard.h" #include "lltrans.h" -LLTrace::BlockTimerStatHandle FT_FILTER_CLIPBOARD("Filter Clipboard"); - LLInventoryFilter::FilterOps::FilterOps(const Params& p) : mFilterObjectTypes(p.object_types), mFilterCategoryTypes(p.category_types), @@ -505,7 +503,7 @@ bool LLInventoryFilter::checkAgainstClipboard(const LLUUID& object_id) const { if (LLClipboard::instance().isCutMode()) { - LL_RECORD_BLOCK_TIME(FT_FILTER_CLIPBOARD); + LL_PROFILE_ZONE_SCOPED; LLUUID current_id = object_id; LLInventoryObject *current_object = gInventory.getObject(object_id); while (current_id.notNull() && current_object) diff --git a/indra/newview/llinventoryitemslist.cpp b/indra/newview/llinventoryitemslist.cpp index 1dc1aa395e..23129f7d44 100644 --- a/indra/newview/llinventoryitemslist.cpp +++ b/indra/newview/llinventoryitemslist.cpp @@ -133,11 +133,9 @@ void LLInventoryItemsList::idle(void* user_data) } } -LLTrace::BlockTimerStatHandle FTM_INVENTORY_ITEMS_REFRESH("Inventory List Refresh"); - void LLInventoryItemsList::refresh() { - LL_RECORD_BLOCK_TIME(FTM_INVENTORY_ITEMS_REFRESH); + LL_PROFILE_ZONE_SCOPED; switch (mRefreshState) { diff --git a/indra/newview/llinventorypanel.cpp b/indra/newview/llinventorypanel.cpp index 3608f9e23f..9b7e0c7824 100644 --- a/indra/newview/llinventorypanel.cpp +++ b/indra/newview/llinventorypanel.cpp @@ -673,10 +673,9 @@ void LLInventoryPanel::itemChanged(const LLUUID& item_id, U32 mask, const LLInve } // Called when something changed in the global model (new item, item coming through the wire, rename, move, etc...) (CHUI-849) -static LLTrace::BlockTimerStatHandle FTM_REFRESH("Inventory Refresh"); void LLInventoryPanel::modelChanged(U32 mask) { - LL_RECORD_BLOCK_TIME(FTM_REFRESH); + LL_PROFILE_ZONE_SCOPED; if (mViewsInitialized != VIEWS_INITIALIZED) return; @@ -1691,10 +1690,9 @@ void LLInventoryPanel::removeItemID(const LLUUID& id) } } -LLTrace::BlockTimerStatHandle FTM_GET_ITEM_BY_ID("Get FolderViewItem by ID"); LLFolderViewItem* LLInventoryPanel::getItemByID(const LLUUID& id) { - LL_RECORD_BLOCK_TIME(FTM_GET_ITEM_BY_ID); + LL_PROFILE_ZONE_SCOPED; std::map::iterator map_it; map_it = mItemMap.find(id); diff --git a/indra/newview/llmaterialmgr.cpp b/indra/newview/llmaterialmgr.cpp index 52b9fb40ae..11aa607393 100644 --- a/indra/newview/llmaterialmgr.cpp +++ b/indra/newview/llmaterialmgr.cpp @@ -580,11 +580,9 @@ void LLMaterialMgr::onPutResponse(bool success, const LLSD& content) } } -static LLTrace::BlockTimerStatHandle FTM_MATERIALS_IDLE("Idle Materials"); - void LLMaterialMgr::onIdle(void*) { - LL_RECORD_BLOCK_TIME(FTM_MATERIALS_IDLE); + LL_PROFILE_ZONE_SCOPED; LLMaterialMgr* instancep = LLMaterialMgr::getInstance(); diff --git a/indra/newview/llpersistentnotificationstorage.cpp b/indra/newview/llpersistentnotificationstorage.cpp index f95ab9928d..20cf4df56b 100644 --- a/indra/newview/llpersistentnotificationstorage.cpp +++ b/indra/newview/llpersistentnotificationstorage.cpp @@ -47,11 +47,9 @@ LLPersistentNotificationStorage::~LLPersistentNotificationStorage() { } -static LLTrace::BlockTimerStatHandle FTM_SAVE_NOTIFICATIONS("Save Notifications"); - void LLPersistentNotificationStorage::saveNotifications() { - LL_RECORD_BLOCK_TIME(FTM_SAVE_NOTIFICATIONS); + LL_PROFILE_ZONE_SCOPED; boost::intrusive_ptr history_channel = boost::dynamic_pointer_cast(LLNotifications::instance().getChannel("Persistent")); if (!history_channel) @@ -90,11 +88,9 @@ void LLPersistentNotificationStorage::saveNotifications() writeNotifications(output); } -static LLTrace::BlockTimerStatHandle FTM_LOAD_NOTIFICATIONS("Load Notifications"); - void LLPersistentNotificationStorage::loadNotifications() { - LL_RECORD_BLOCK_TIME(FTM_LOAD_NOTIFICATIONS); + LL_PROFILE_ZONE_SCOPED; LL_INFOS("LLPersistentNotificationStorage") << "start loading notifications" << LL_ENDL; diff --git a/indra/newview/llphysicsmotion.cpp b/indra/newview/llphysicsmotion.cpp index f48ce680fd..30ca7ae539 100644 --- a/indra/newview/llphysicsmotion.cpp +++ b/indra/newview/llphysicsmotion.cpp @@ -453,6 +453,7 @@ F32 LLPhysicsMotion::calculateAcceleration_local(const F32 velocity_local, const BOOL LLPhysicsMotionController::onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; // Skip if disabled globally. if (!gSavedSettings.getBOOL("AvatarPhysics")) { diff --git a/indra/newview/llscenemonitor.cpp b/indra/newview/llscenemonitor.cpp index f9baf5fbd3..177bc84cee 100644 --- a/indra/newview/llscenemonitor.cpp +++ b/indra/newview/llscenemonitor.cpp @@ -332,9 +332,6 @@ bool LLSceneMonitor::needsUpdate() const return mDiffState == NEED_DIFF; } -static LLTrace::BlockTimerStatHandle FTM_GENERATE_SCENE_LOAD_DITHER_TEXTURE("Generate Scene Load Dither Texture"); -static LLTrace::BlockTimerStatHandle FTM_SCENE_LOAD_IMAGE_DIFF("Scene Load Image Diff"); - static LLStaticHashedString sDitherScale("dither_scale"); static LLStaticHashedString sDitherScaleS("dither_scale_s"); static LLStaticHashedString sDitherScaleT("dither_scale_t"); @@ -356,14 +353,12 @@ void LLSceneMonitor::compare() return; } - LL_RECORD_BLOCK_TIME(FTM_SCENE_LOAD_IMAGE_DIFF); mDiffState = EXECUTE_DIFF; S32 width = gViewerWindow->getWindowWidthRaw(); S32 height = gViewerWindow->getWindowHeightRaw(); if(!mDiff) { - LL_RECORD_BLOCK_TIME(FTM_GENERATE_SCENE_LOAD_DITHER_TEXTURE); mDiff = new LLRenderTarget(); mDiff->allocate(width, height, GL_RGBA, false, false, LLTexUnit::TT_TEXTURE, true); @@ -371,7 +366,6 @@ void LLSceneMonitor::compare() } else if(mDiff->getWidth() != width || mDiff->getHeight() != height) { - LL_RECORD_BLOCK_TIME(FTM_GENERATE_SCENE_LOAD_DITHER_TEXTURE); mDiff->resize(width, height); generateDitheringTexture(width, height); } @@ -427,8 +421,6 @@ void LLSceneMonitor::calcDiffAggregate() { #ifdef LL_WINDOWS - LL_RECORD_BLOCK_TIME(FTM_SCENE_LOAD_IMAGE_DIFF); - if(mDiffState != EXECUTE_DIFF && !mDebugViewerVisible) { return; @@ -481,8 +473,6 @@ void LLSceneMonitor::calcDiffAggregate() static LLTrace::EventStatHandle<> sFramePixelDiff("FramePixelDifference"); void LLSceneMonitor::fetchQueryResult() { - LL_RECORD_BLOCK_TIME(FTM_SCENE_LOAD_IMAGE_DIFF); - // also throttle timing here, to avoid going below sample time due to phasing with frame capture static LLCachedControl scene_load_sample_time_control(gSavedSettings, "SceneLoadingMonitorSampleTime"); F32Seconds scene_load_sample_time = (F32Seconds)scene_load_sample_time_control(); diff --git a/indra/newview/llscreenchannel.cpp b/indra/newview/llscreenchannel.cpp index ca48c9d58c..17f2970f99 100644 --- a/indra/newview/llscreenchannel.cpp +++ b/indra/newview/llscreenchannel.cpp @@ -49,10 +49,9 @@ using namespace LLNotificationsUI; bool LLScreenChannel::mWasStartUpToastShown = false; -LLTrace::BlockTimerStatHandle FTM_GET_CHANNEL_RECT("Calculate Notification Channel Region"); LLRect LLScreenChannelBase::getChannelRect() { - LL_RECORD_BLOCK_TIME(FTM_GET_CHANNEL_RECT); + LL_PROFILE_ZONE_SCOPED; if (mFloaterSnapRegion == NULL) { diff --git a/indra/newview/llscripteditor.cpp b/indra/newview/llscripteditor.cpp index cd3a4dfd11..c6bb2f19dd 100644 --- a/indra/newview/llscripteditor.cpp +++ b/indra/newview/llscripteditor.cpp @@ -138,11 +138,9 @@ void LLScriptEditor::initKeywords() mKeywords.initialize(LLSyntaxIdLSL::getInstance()->getKeywordsXML()); } -LLTrace::BlockTimerStatHandle FTM_SYNTAX_HIGHLIGHTING("Syntax Highlighting"); - void LLScriptEditor::loadKeywords() { - LL_RECORD_BLOCK_TIME(FTM_SYNTAX_HIGHLIGHTING); + LL_PROFILE_ZONE_SCOPED; mKeywords.processTokens(); segment_vec_t segment_list; @@ -160,7 +158,7 @@ void LLScriptEditor::updateSegments() { if (mReflowIndex < S32_MAX && mKeywords.isLoaded() && mParseOnTheFly) { - LL_RECORD_BLOCK_TIME(FTM_SYNTAX_HIGHLIGHTING); + LL_PROFILE_ZONE_SCOPED; // HACK: No non-ascii keywords for now segment_vec_t segment_list; mKeywords.findSegments(&segment_list, getWText(), mDefaultColor.get(), *this); diff --git a/indra/newview/llselectmgr.cpp b/indra/newview/llselectmgr.cpp index 50884762a8..53247031b4 100644 --- a/indra/newview/llselectmgr.cpp +++ b/indra/newview/llselectmgr.cpp @@ -97,6 +97,8 @@ #include "llglheaders.h" #include "llinventoryobserver.h" +LLSelectMgr* LLSimpleton::sInstance = nullptr; + LLViewerObject* getSelectedParentObject(LLViewerObject *object) ; // // Consts @@ -209,8 +211,6 @@ void LLSelectMgr::cleanupGlobals() LLSelectMgr::getInstance()->clearSelections(); } -// Build time optimization, generate this function once here -template class LLSelectMgr* LLSingleton::getInstance(); //----------------------------------------------------------------------------- // LLSelectMgr() //----------------------------------------------------------------------------- diff --git a/indra/newview/llselectmgr.h b/indra/newview/llselectmgr.h index 57fdfce152..0fd2e74090 100644 --- a/indra/newview/llselectmgr.h +++ b/indra/newview/llselectmgr.h @@ -407,11 +407,8 @@ private: LLObjectSelectionHandle mSelectedObjects; }; -class LLSelectMgr : public LLEditMenuHandler, public LLSingleton +class LLSelectMgr : public LLEditMenuHandler, public LLSimpleton { - LLSINGLETON(LLSelectMgr); - ~LLSelectMgr(); - public: static BOOL sRectSelectInclusive; // do we need to surround an object to pick it? static BOOL sRenderHiddenSelections; // do we show selection silhouettes that are occluded? @@ -437,6 +434,9 @@ public: LLCachedControl mDebugSelectMgr; public: + LLSelectMgr(); + ~LLSelectMgr(); + static void cleanupGlobals(); // LLEditMenuHandler interface diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index 5cac9cd7ae..30b7124550 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -55,9 +55,6 @@ #include "llviewershadermgr.h" #include "llcontrolavatar.h" -static LLTrace::BlockTimerStatHandle FTM_FRUSTUM_CULL("Frustum Culling"); -static LLTrace::BlockTimerStatHandle FTM_CULL_REBOUND("Cull Rebound Partition"); - extern bool gShiftFrame; static U32 sZombieGroups = 0; @@ -409,11 +406,6 @@ void LLSpatialGroup::rebuildMesh() } } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_VBO("VBO Rebuilt"); -static LLTrace::BlockTimerStatHandle FTM_ADD_GEOMETRY_COUNT("Add Geometry"); -static LLTrace::BlockTimerStatHandle FTM_CREATE_VB("Create VB"); -static LLTrace::BlockTimerStatHandle FTM_GET_GEOMETRY("Get Geometry"); - void LLSpatialPartition::rebuildGeom(LLSpatialGroup* group) { if (group->isDead() || !group->hasState(LLSpatialGroup::GEOM_DIRTY)) @@ -427,7 +419,7 @@ void LLSpatialPartition::rebuildGeom(LLSpatialGroup* group) group->mLastUpdateViewAngle = group->mViewAngle; } - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VBO); + LL_PROFILE_ZONE_SCOPED; group->clearDrawMap(); @@ -435,15 +427,12 @@ void LLSpatialPartition::rebuildGeom(LLSpatialGroup* group) U32 index_count = 0; U32 vertex_count = 0; - { - LL_RECORD_BLOCK_TIME(FTM_ADD_GEOMETRY_COUNT); - addGeometryCount(group, vertex_count, index_count); - } - + addGeometryCount(group, vertex_count, index_count); + if (vertex_count > 0 && index_count > 0) { //create vertex buffer containing volume geometry for this node { - LL_RECORD_BLOCK_TIME(FTM_CREATE_VB); + group->mBuilt = 1.f; if (group->mVertexBuffer.isNull() || !group->mVertexBuffer->isWriteable() || @@ -458,7 +447,6 @@ void LLSpatialPartition::rebuildGeom(LLSpatialGroup* group) group->mVertexBuffer = NULL; group->mBufferMap.clear(); } - stop_glerror(); } else { @@ -471,13 +459,11 @@ void LLSpatialPartition::rebuildGeom(LLSpatialGroup* group) group->mVertexBuffer = NULL; group->mBufferMap.clear(); } - stop_glerror(); } } if (group->mVertexBuffer) { - LL_RECORD_BLOCK_TIME(FTM_GET_GEOMETRY); getGeometry(group); } } @@ -1489,12 +1475,12 @@ void LLSpatialPartition::resetVertexBuffers() BOOL LLSpatialPartition::getVisibleExtents(LLCamera& camera, LLVector3& visMin, LLVector3& visMax) { + LL_PROFILE_ZONE_SCOPED; LLVector4a visMina, visMaxa; visMina.load3(visMin.mV); visMaxa.load3(visMax.mV); { - LL_RECORD_BLOCK_TIME(FTM_CULL_REBOUND); LLSpatialGroup* group = (LLSpatialGroup*) mOctree->getListener(0); group->rebound(); } @@ -1516,11 +1502,11 @@ BOOL LLSpatialPartition::visibleObjectsInFrustum(LLCamera& camera) S32 LLSpatialPartition::cull(LLCamera &camera, std::vector* results, BOOL for_select) { + LL_PROFILE_ZONE_SCOPED; #if LL_OCTREE_PARANOIA_CHECK ((LLSpatialGroup*)mOctree->getListener(0))->checkStates(); #endif { - LL_RECORD_BLOCK_TIME(FTM_CULL_REBOUND); LLSpatialGroup* group = (LLSpatialGroup*) mOctree->getListener(0); group->rebound(); } @@ -1537,37 +1523,32 @@ S32 LLSpatialPartition::cull(LLCamera &camera, std::vector* result S32 LLSpatialPartition::cull(LLCamera &camera, bool do_occlusion) { + LL_PROFILE_ZONE_SCOPED; #if LL_OCTREE_PARANOIA_CHECK ((LLSpatialGroup*)mOctree->getListener(0))->checkStates(); #endif - { - LL_RECORD_BLOCK_TIME(FTM_CULL_REBOUND); - LLSpatialGroup* group = (LLSpatialGroup*) mOctree->getListener(0); - group->rebound(); - } + LLSpatialGroup* group = (LLSpatialGroup*) mOctree->getListener(0); + group->rebound(); #if LL_OCTREE_PARANOIA_CHECK ((LLSpatialGroup*)mOctree->getListener(0))->validate(); #endif - if (LLPipeline::sShadowRender) - { - LL_RECORD_BLOCK_TIME(FTM_FRUSTUM_CULL); - LLOctreeCullShadow culler(&camera); - culler.traverse(mOctree); - } - else if (mInfiniteFarClip || !LLPipeline::sUseFarClip) - { - LL_RECORD_BLOCK_TIME(FTM_FRUSTUM_CULL); - LLOctreeCullNoFarClip culler(&camera); - culler.traverse(mOctree); - } - else - { - LL_RECORD_BLOCK_TIME(FTM_FRUSTUM_CULL); - LLOctreeCull culler(&camera); - culler.traverse(mOctree); - } + if (LLPipeline::sShadowRender) + { + LLOctreeCullShadow culler(&camera); + culler.traverse(mOctree); + } + else if (mInfiniteFarClip || !LLPipeline::sUseFarClip) + { + LLOctreeCullNoFarClip culler(&camera); + culler.traverse(mOctree); + } + else + { + LLOctreeCull culler(&camera); + culler.traverse(mOctree); + } return 0; } @@ -4048,8 +4029,7 @@ LLDrawInfo::LLDrawInfo(U16 start, U16 end, U32 count, U32 offset, LLViewerTexture* texture, LLVertexBuffer* buffer, bool selected, BOOL fullbright, U8 bump, BOOL particle, F32 part_size) -: LLTrace::MemTrackableNonVirtual("LLDrawInfo"), - mVertexBuffer(buffer), +: mVertexBuffer(buffer), mTexture(texture), mTextureMatrix(NULL), mModelMatrix(NULL), diff --git a/indra/newview/llspatialpartition.h b/indra/newview/llspatialpartition.h index 6ef82fac9c..8cc50e71b1 100644 --- a/indra/newview/llspatialpartition.h +++ b/indra/newview/llspatialpartition.h @@ -56,14 +56,14 @@ class LLViewerRegion; void pushVerts(LLFace* face, U32 mask); -class LLDrawInfo : public LLRefCount, public LLTrace::MemTrackableNonVirtual +class LLDrawInfo : public LLRefCount { + LL_ALIGN_NEW; protected: ~LLDrawInfo(); public: LLDrawInfo(const LLDrawInfo& rhs) - : LLTrace::MemTrackableNonVirtual("LLDrawInfo") { *this = rhs; } diff --git a/indra/newview/llviewercamera.cpp b/indra/newview/llviewercamera.cpp index 27a87ee1a0..5ebce115f6 100644 --- a/indra/newview/llviewercamera.cpp +++ b/indra/newview/llviewercamera.cpp @@ -54,6 +54,8 @@ // System includes #include // for setprecision +LLViewerCamera* LLSimpleton::sInstance = nullptr; + LLTrace::CountStatHandle<> LLViewerCamera::sVelocityStat("camera_velocity"); LLTrace::CountStatHandle<> LLViewerCamera::sAngularVelocityStat("camera_angular_velocity"); @@ -81,9 +83,6 @@ glh::matrix4f gl_pick_matrix(GLfloat x, GLfloat y, GLfloat width, GLfloat height return glh::matrix4f(m); } -// Build time optimization, generate this once in .cpp file -template class LLViewerCamera* LLSingleton::getInstance(); - LLViewerCamera::LLViewerCamera() : LLCamera() { calcProjection(getFar()); diff --git a/indra/newview/llviewercamera.h b/indra/newview/llviewercamera.h index fb07a3fb2d..549778a841 100644 --- a/indra/newview/llviewercamera.h +++ b/indra/newview/llviewercamera.h @@ -38,25 +38,11 @@ class LLViewerObject; const BOOL FOR_SELECTION = TRUE; const BOOL NOT_FOR_SELECTION = FALSE; -// Build time optimization, generate this once in .cpp file -#ifndef LLVIEWERCAMERA_CPP -extern template class LLViewerCamera* LLSingleton::getInstance(); -#endif - -LL_ALIGN_PREFIX(16) -class LLViewerCamera : public LLCamera, public LLSingleton +class alignas(16) LLViewerCamera : public LLCamera, public LLSimpleton { - LLSINGLETON(LLViewerCamera); + LL_ALIGN_NEW public: - void* operator new(size_t size) - { - return ll_aligned_malloc_16(size); - } - - void operator delete(void* ptr) - { - ll_aligned_free_16(ptr); - } + LLViewerCamera(); typedef enum { @@ -141,7 +127,7 @@ protected: S16 mZoomSubregion; public: -} LL_ALIGN_POSTFIX(16); +}; #endif // LL_LLVIEWERCAMERA_H diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp index 1236695e4f..6368286f6e 100644 --- a/indra/newview/llviewerdisplay.cpp +++ b/indra/newview/llviewerdisplay.cpp @@ -809,13 +809,9 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) { LL_RECORD_BLOCK_TIME(FTM_IMAGE_UPDATE_CLASS); - LLTrace::CountStatHandle<>* velocity_stat = LLViewerCamera::getVelocityStat(); - LLTrace::CountStatHandle<>* angular_velocity_stat = LLViewerCamera::getAngularVelocityStat(); - LLViewerTexture::updateClass(LLTrace::get_frame_recording().getPeriodMeanPerSec(*velocity_stat), - LLTrace::get_frame_recording().getPeriodMeanPerSec(*angular_velocity_stat)); + LLViewerTexture::updateClass(); } - { LL_RECORD_BLOCK_TIME(FTM_IMAGE_UPDATE_BUMP); gBumpImageList.updateImages(); // must be called before gTextureList version so that it's textures are thrown out first. @@ -1285,7 +1281,6 @@ void render_ui(F32 zoom_factor, int subfield) if(LLSceneMonitor::getInstance()->needsUpdate()) { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_SCENE_MON); gGL.pushMatrix(); gViewerWindow->setup2DRender(); LLSceneMonitor::getInstance()->compare(); @@ -1334,12 +1329,9 @@ void render_ui(F32 zoom_factor, int subfield) } gGL.flush(); - { - LL_RECORD_BLOCK_TIME(FTM_RENDER_UI_DEBUG_TEXT); - gViewerWindow->setup2DRender(); - gViewerWindow->updateDebugText(); - gViewerWindow->drawDebugText(); - } + gViewerWindow->setup2DRender(); + gViewerWindow->updateDebugText(); + gViewerWindow->drawDebugText(); LLVertexBuffer::unbind(); } diff --git a/indra/newview/llviewerjointmesh.cpp b/indra/newview/llviewerjointmesh.cpp index 63ad708e59..3e385a46e4 100644 --- a/indra/newview/llviewerjointmesh.cpp +++ b/indra/newview/llviewerjointmesh.cpp @@ -362,7 +362,6 @@ void LLViewerJointMesh::updateFaceSizes(U32 &num_vertices, U32& num_indices, F32 //----------------------------------------------------------------------------- // updateFaceData() //----------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_AVATAR_FACE("Avatar Face"); void LLViewerJointMesh::updateFaceData(LLFace *face, F32 pixel_area, BOOL damp_wind, bool terse_update) { @@ -383,9 +382,8 @@ void LLViewerJointMesh::updateFaceData(LLFace *face, F32 pixel_area, BOOL damp_w // since mMesh is being copied into mVertexBuffer every frame return; } - - - LL_RECORD_BLOCK_TIME(FTM_AVATAR_FACE); + + LL_PROFILE_ZONE_SCOPED; LLStrider verticesp; LLStrider normalsp; diff --git a/indra/newview/llviewermessage.cpp b/indra/newview/llviewermessage.cpp index 5e99d13206..52a65abf9b 100644 --- a/indra/newview/llviewermessage.cpp +++ b/indra/newview/llviewermessage.cpp @@ -2219,11 +2219,9 @@ protected: } }; -static LLTrace::BlockTimerStatHandle FTM_PROCESS_IMPROVED_IM("Process IM"); - void process_improved_im(LLMessageSystem *msg, void **user_data) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_IMPROVED_IM); + LL_PROFILE_ZONE_SCOPED; LLUUID from_id; BOOL from_group; @@ -3267,10 +3265,9 @@ const F32 THRESHOLD_HEAD_ROT_QDOT = 0.9997f; // ~= 2.5 degrees -- if its less th const F32 MAX_HEAD_ROT_QDOT = 0.99999f; // ~= 0.5 degrees -- if its greater than this then no need to update head_rot // between these values we delay the updates (but no more than one second) -static LLTrace::BlockTimerStatHandle FTM_AGENT_UPDATE_SEND("Send Message"); - void send_agent_update(BOOL force_send, BOOL send_reliable) { + LL_PROFILE_ZONE_SCOPED; if (gAgent.getTeleportState() != LLAgent::TELEPORT_NONE) { // We don't care if they want to send an agent update, they're not allowed to until the simulator @@ -3451,7 +3448,6 @@ void send_agent_update(BOOL force_send, BOOL send_reliable) } */ - LL_RECORD_BLOCK_TIME(FTM_AGENT_UPDATE_SEND); // Build the message msg->newMessageFast(_PREHASH_AgentUpdate); msg->nextBlockFast(_PREHASH_AgentData); @@ -3701,11 +3697,9 @@ void process_terse_object_update_improved(LLMessageSystem *mesgsys, void **user_ } } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_OBJECTS("Process Kill Objects"); - void process_kill_object(LLMessageSystem *mesgsys, void **user_data) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_OBJECTS); + LL_PROFILE_ZONE_SCOPED; LLUUID id; diff --git a/indra/newview/llviewerobject.cpp b/indra/newview/llviewerobject.cpp index 0832415e1e..c671aec224 100644 --- a/indra/newview/llviewerobject.cpp +++ b/indra/newview/llviewerobject.cpp @@ -146,16 +146,14 @@ const S32 MAX_OBJECT_BINARY_DATA_SIZE = 60 + 16; const F64 INVENTORY_UPDATE_WAIT_TIME_DESYNC = 5; // seconds const F64 INVENTORY_UPDATE_WAIT_TIME_OUTDATED = 1; -static LLTrace::BlockTimerStatHandle FTM_CREATE_OBJECT("Create Object"); - // static LLViewerObject *LLViewerObject::createObject(const LLUUID &id, const LLPCode pcode, LLViewerRegion *regionp, S32 flags) { + LL_PROFILE_ZONE_SCOPED; LL_DEBUGS("ObjectUpdate") << "creating " << id << LL_ENDL; dumpStack("ObjectUpdateStack"); LLViewerObject *res = NULL; - LL_RECORD_BLOCK_TIME(FTM_CREATE_OBJECT); if (gNonInteractive && pcode != LL_PCODE_LEGACY_AVATAR @@ -252,8 +250,7 @@ LLViewerObject *LLViewerObject::createObject(const LLUUID &id, const LLPCode pco } LLViewerObject::LLViewerObject(const LLUUID &id, const LLPCode pcode, LLViewerRegion *regionp, BOOL is_global) -: LLTrace::MemTrackable("LLViewerObject"), - LLPrimitive(), +: LLPrimitive(), mChildList(), mID(id), mLocalID(0), @@ -2517,9 +2514,6 @@ void LLViewerObject::loadFlags(U32 flags) void LLViewerObject::idleUpdate(LLAgent &agent, const F64 &frame_time) { - //static LLTrace::BlockTimerStatHandle ftm("Viewer Object"); - //LL_RECORD_BLOCK_TIME(ftm); - if (!mDead) { if (!mStatic && sVelocityInterpolate && !isSelected()) diff --git a/indra/newview/llviewerobject.h b/indra/newview/llviewerobject.h index 250c4ac328..bef8e3e7e3 100644 --- a/indra/newview/llviewerobject.h +++ b/indra/newview/llviewerobject.h @@ -110,8 +110,7 @@ struct PotentialReturnableObject class LLViewerObject : public LLPrimitive, public LLRefCount, - public LLGLUpdate, - public LLTrace::MemTrackable + public LLGLUpdate { protected: virtual ~LLViewerObject(); // use unref() diff --git a/indra/newview/llviewerobjectlist.cpp b/indra/newview/llviewerobjectlist.cpp index 097b5e3645..971a355a65 100644 --- a/indra/newview/llviewerobjectlist.cpp +++ b/indra/newview/llviewerobjectlist.cpp @@ -849,8 +849,6 @@ void LLViewerObjectList::updateApparentAngles(LLAgent &agent) LLVOAvatar::cullAvatarsByPixelArea(); } -static LLTrace::BlockTimerStatHandle FTM_IDLE_COPY("Idle Copy"); - void LLViewerObjectList::update(LLAgent &agent) { LL_PROFILE_ZONE_SCOPED @@ -906,8 +904,6 @@ void LLViewerObjectList::update(LLAgent &agent) U32 idle_count = 0; { - LL_RECORD_BLOCK_TIME(FTM_IDLE_COPY); - for (std::vector >::iterator active_iter = mActiveObjects.begin(); active_iter != mActiveObjects.end(); active_iter++) { @@ -1352,11 +1348,9 @@ void LLViewerObjectList::cleanupReferences(LLViewerObject *objectp) } } -static LLTrace::BlockTimerStatHandle FTM_REMOVE_DRAWABLE("Remove Drawable"); - void LLViewerObjectList::removeDrawable(LLDrawable* drawablep) { - LL_RECORD_BLOCK_TIME(FTM_REMOVE_DRAWABLE); + LL_PROFILE_ZONE_SCOPED; if (!drawablep) { @@ -1645,12 +1639,9 @@ void LLViewerObjectList::onPhysicsFlagsFetchFailure(const LLUUID& object_id) mPendingPhysicsFlags.erase(object_id); } -static LLTrace::BlockTimerStatHandle FTM_SHIFT_OBJECTS("Shift Objects"); -static LLTrace::BlockTimerStatHandle FTM_PIPELINE_SHIFT("Pipeline Shift"); -static LLTrace::BlockTimerStatHandle FTM_REGION_SHIFT("Region Shift"); - void LLViewerObjectList::shiftObjects(const LLVector3 &offset) { + LL_PROFILE_ZONE_SCOPED; // This is called when we shift our origin when we cross region boundaries... // We need to update many object caches, I'll document this more as I dig through the code // cleaning things out... @@ -1660,7 +1651,6 @@ void LLViewerObjectList::shiftObjects(const LLVector3 &offset) return; } - LL_RECORD_BLOCK_TIME(FTM_SHIFT_OBJECTS); LLViewerObject *objectp; for (vobj_list_t::iterator iter = mObjects.begin(); iter != mObjects.end(); ++iter) @@ -1678,16 +1668,10 @@ void LLViewerObjectList::shiftObjects(const LLVector3 &offset) } } - { - LL_RECORD_BLOCK_TIME(FTM_PIPELINE_SHIFT); gPipeline.shiftObjects(offset); - } - - { - LL_RECORD_BLOCK_TIME(FTM_REGION_SHIFT); + LLWorld::getInstance()->shiftRegions(offset); } -} void LLViewerObjectList::repartitionObjects() { diff --git a/indra/newview/llvieweroctree.cpp b/indra/newview/llvieweroctree.cpp index 6365df09e1..868cf75d11 100644 --- a/indra/newview/llvieweroctree.cpp +++ b/indra/newview/llvieweroctree.cpp @@ -231,8 +231,7 @@ S32 AABBSphereIntersectR2(const LLVector4a& min, const LLVector4a& max, const LL //class LLViewerOctreeEntry definitions //----------------------------------------------------------------------------------- LLViewerOctreeEntry::LLViewerOctreeEntry() -: LLTrace::MemTrackable("LLViewerOctreeEntry"), - mGroup(NULL), +: mGroup(NULL), mBinRadius(0.f), mBinIndex(-1), mVisible(0) @@ -458,8 +457,7 @@ LLViewerOctreeGroup::~LLViewerOctreeGroup() } LLViewerOctreeGroup::LLViewerOctreeGroup(OctreeNode* node) -: LLTrace::MemTrackable("LLViewerOctreeGroup"), - mOctreeNode(node), +: mOctreeNode(node), mAnyVisible(0), mState(CLEAN) { @@ -545,6 +543,7 @@ void LLViewerOctreeGroup::unbound() //virtual void LLViewerOctreeGroup::rebound() { + LL_PROFILE_ZONE_SCOPED; if (!isDirty()) { return; @@ -1039,11 +1038,9 @@ void LLOcclusionCullingGroup::clearOcclusionState(U32 state, S32 mode) } } -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_READBACK("Readback Occlusion"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_WAIT("Occlusion Wait"); - BOOL LLOcclusionCullingGroup::earlyFail(LLCamera* camera, const LLVector4a* bounds) { + LL_PROFILE_ZONE_SCOPED; if (camera->getOrigin().isExactlyZero()) { return FALSE; @@ -1094,7 +1091,7 @@ void LLOcclusionCullingGroup::checkOcclusion() { if (LLPipeline::sUseOcclusion > 1) { - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_READBACK); + LL_PROFILE_ZONE_SCOPED; LLOcclusionCullingGroup* parent = (LLOcclusionCullingGroup*)getParent(); if (parent && parent->isOcclusionState(LLOcclusionCullingGroup::OCCLUDED)) { //if the parent has been marked as occluded, the child is implicitly occluded @@ -1106,19 +1103,8 @@ void LLOcclusionCullingGroup::checkOcclusion() GLuint available = 0; if (mOcclusionQuery[LLViewerCamera::sCurCameraID]) { + LL_PROFILE_ZONE_NAMED("co - query available") glGetQueryObjectuivARB(mOcclusionQuery[LLViewerCamera::sCurCameraID], GL_QUERY_RESULT_AVAILABLE_ARB, &available); - - static LLCachedControl wait_for_query(gSavedSettings, "RenderSynchronousOcclusion", true); - - if (wait_for_query && mOcclusionIssued[LLViewerCamera::sCurCameraID] < gFrameCount) - { //query was issued last frame, wait until it's available - S32 max_loop = 1024; - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_WAIT); - while (!available && max_loop-- > 0) - { - glGetQueryObjectuivARB(mOcclusionQuery[LLViewerCamera::sCurCameraID], GL_QUERY_RESULT_AVAILABLE_ARB, &available); - } - } } else { @@ -1130,6 +1116,7 @@ void LLOcclusionCullingGroup::checkOcclusion() GLuint res = 1; if (!isOcclusionState(DISCARD_QUERY) && mOcclusionQuery[LLViewerCamera::sCurCameraID]) { + LL_PROFILE_ZONE_NAMED("co - query result") glGetQueryObjectuivARB(mOcclusionQuery[LLViewerCamera::sCurCameraID], GL_QUERY_RESULT_ARB, &res); #if LL_TRACK_PENDING_OCCLUSION_QUERIES sPendingQueries.erase(mOcclusionQuery[LLViewerCamera::sCurCameraID]); @@ -1173,19 +1160,9 @@ void LLOcclusionCullingGroup::checkOcclusion() } } -static LLTrace::BlockTimerStatHandle FTM_PUSH_OCCLUSION_VERTS("Push Occlusion"); -static LLTrace::BlockTimerStatHandle FTM_SET_OCCLUSION_STATE("Occlusion State"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_EARLY_FAIL("Occlusion Early Fail"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_ALLOCATE("Allocate"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_BUILD("Build"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_BEGIN_QUERY("Begin Query"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_END_QUERY("End Query"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_SET_BUFFER("Set Buffer"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_DRAW_WATER("Draw Water"); -static LLTrace::BlockTimerStatHandle FTM_OCCLUSION_DRAW("Draw"); - void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* shift) { + LL_PROFILE_ZONE_SCOPED; if (mSpatialPartition->isOcclusionEnabled() && LLPipeline::sUseOcclusion > 1) { //move mBounds to the agent space if necessary @@ -1206,7 +1183,7 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh // Don't cull hole/edge water, unless we have the GL_ARB_depth_clamp extension if (earlyFail(camera, bounds)) { - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_EARLY_FAIL); + LL_PROFILE_ZONE_NAMED("doOcclusion - early fail"); setOcclusionState(LLOcclusionCullingGroup::DISCARD_QUERY); assert_states_valid(this); clearOcclusionState(LLOcclusionCullingGroup::OCCLUDED, LLOcclusionCullingGroup::STATE_MODE_DIFF); @@ -1217,11 +1194,10 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh if (!isOcclusionState(QUERY_PENDING) || isOcclusionState(DISCARD_QUERY)) { { //no query pending, or previous query to be discarded - LL_RECORD_BLOCK_TIME(FTM_RENDER_OCCLUSION); + LL_PROFILE_ZONE_NAMED("doOcclusion - render"); if (!mOcclusionQuery[LLViewerCamera::sCurCameraID]) { - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_ALLOCATE); mOcclusionQuery[LLViewerCamera::sCurCameraID] = getNewOcclusionQueryObjectName(); } @@ -1246,15 +1222,12 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh add(sOcclusionQueries, 1); { - LL_RECORD_BLOCK_TIME(FTM_PUSH_OCCLUSION_VERTS); + LL_PROFILE_ZONE_NAMED("doOcclusion - push"); //store which frame this query was issued on mOcclusionIssued[LLViewerCamera::sCurCameraID] = gFrameCount; - { - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_BEGIN_QUERY); - glBeginQueryARB(mode, mOcclusionQuery[LLViewerCamera::sCurCameraID]); - } + glBeginQueryARB(mode, mOcclusionQuery[LLViewerCamera::sCurCameraID]); LLGLSLShader* shader = LLGLSLShader::sCurBoundShaderPtr; llassert(shader); @@ -1266,7 +1239,7 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh if (!use_depth_clamp && mSpatialPartition->mDrawableType == LLDrawPool::POOL_VOIDWATER) { - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_DRAW_WATER); + LL_PROFILE_ZONE_NAMED("doOcclusion - draw water"); LLGLSquashToFarClip squash; if (camera->getOrigin().isExactlyZero()) @@ -1281,7 +1254,7 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh } else { - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_DRAW); + LL_PROFILE_ZONE_NAMED("doOcclusion - draw"); if (camera->getOrigin().isExactlyZero()) { //origin is invalid, draw entire box gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, 0); @@ -1292,17 +1265,13 @@ void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* sh gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, get_box_fan_indices(camera, bounds[0])); } } - - - { - LL_RECORD_BLOCK_TIME(FTM_OCCLUSION_END_QUERY); - glEndQueryARB(mode); - } + + glEndQueryARB(mode); } } { - LL_RECORD_BLOCK_TIME(FTM_SET_OCCLUSION_STATE); + LL_PROFILE_ZONE_NAMED("doOcclusion - set state"); setOcclusionState(LLOcclusionCullingGroup::QUERY_PENDING); clearOcclusionState(LLOcclusionCullingGroup::DISCARD_QUERY); } diff --git a/indra/newview/llvieweroctree.h b/indra/newview/llvieweroctree.h index 219ec7e8da..11ba7e4f1e 100644 --- a/indra/newview/llvieweroctree.h +++ b/indra/newview/llvieweroctree.h @@ -71,8 +71,9 @@ S32 AABBSphereIntersectR2(const LLVector3& min, const LLVector3& max, const LLVe //defines data needed for octree of an entry //LL_ALIGN_PREFIX(16) -class LLViewerOctreeEntry : public LLRefCount, public LLTrace::MemTrackable +class LLViewerOctreeEntry : public LLRefCount { + LL_ALIGN_NEW friend class LLViewerOctreeEntryData; public: @@ -178,8 +179,9 @@ protected: //defines an octree group for an octree node, which contains multiple entries. //LL_ALIGN_PREFIX(16) class LLViewerOctreeGroup -: public LLOctreeListener, public LLTrace::MemTrackable +: public LLOctreeListener { + LL_ALIGN_NEW friend class LLViewerOctreeCull; protected: virtual ~LLViewerOctreeGroup(); @@ -201,7 +203,6 @@ public: LLViewerOctreeGroup(OctreeNode* node); LLViewerOctreeGroup(const LLViewerOctreeGroup& rhs) - : LLTrace::MemTrackable("LLViewerOctreeGroup") { *this = rhs; } diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index 34847d8618..fbc5830a5c 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -481,8 +481,6 @@ const F32 GPU_MEMORY_CHECK_WAIT_TIME = 1.0f; F32 texmem_lower_bound_scale = 0.85f; F32 texmem_middle_bound_scale = 0.925f; -static LLTrace::BlockTimerStatHandle FTM_TEXTURE_MEMORY_CHECK("Memory Check"); - //static bool LLViewerTexture::isMemoryForTextureLow() { @@ -531,8 +529,6 @@ void LLViewerTexture::getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &p timer.reset(); { - LL_RECORD_BLOCK_TIME(FTM_TEXTURE_MEMORY_CHECK); - if (gGLManager.mHasATIMemInfo) { S32 meminfo[4]; @@ -555,11 +551,8 @@ void LLViewerTexture::getGPUMemoryForTextures(S32Megabytes &gpu, S32Megabytes &p } } -static LLTrace::BlockTimerStatHandle FTM_TEXTURE_UPDATE_MEDIA("Media"); -static LLTrace::BlockTimerStatHandle FTM_TEXTURE_UPDATE_TEST("Test"); - //static -void LLViewerTexture::updateClass(const F32 velocity, const F32 angular_velocity) +void LLViewerTexture::updateClass() { LL_PROFILE_ZONE_SCOPED; sCurrentTime = gFrameTimeSeconds; @@ -567,14 +560,10 @@ void LLViewerTexture::updateClass(const F32 velocity, const F32 angular_velocity LLTexturePipelineTester* tester = (LLTexturePipelineTester*)LLMetricPerformanceTesterBasic::getTester(sTesterName); if (tester) { - LL_RECORD_BLOCK_TIME(FTM_TEXTURE_UPDATE_TEST); tester->update(); } - { - LL_RECORD_BLOCK_TIME(FTM_TEXTURE_UPDATE_MEDIA); - LLViewerMediaTexture::updateClass(); - } + LLViewerMediaTexture::updateClass(); sBoundTextureMemory = LLImageGL::sBoundTextureMemory; sTotalTextureMemory = LLImageGL::sGlobalTextureMemory; @@ -2614,6 +2603,7 @@ void LLViewerFetchedTexture::pauseLoadedCallbacks(const LLLoadedCallbackEntry::s bool LLViewerFetchedTexture::doLoadedCallbacks() { + LL_PROFILE_ZONE_SCOPED; static const F32 MAX_INACTIVE_TIME = 900.f ; //seconds static const F32 MAX_IDLE_WAIT_TIME = 5.f ; //seconds @@ -3419,6 +3409,7 @@ bool LLViewerLODTexture::scaleDown() //static void LLViewerMediaTexture::updateClass() { + LL_PROFILE_ZONE_SCOPED; static const F32 MAX_INACTIVE_TIME = 30.f; #if 0 diff --git a/indra/newview/llviewertexture.h b/indra/newview/llviewertexture.h index a5a1fb2c16..f9f1bfef44 100644 --- a/indra/newview/llviewertexture.h +++ b/indra/newview/llviewertexture.h @@ -114,7 +114,7 @@ protected: public: static void initClass(); - static void updateClass(const F32 velocity, const F32 angular_velocity) ; + static void updateClass(); LLViewerTexture(BOOL usemipmaps = TRUE); LLViewerTexture(const LLUUID& id, BOOL usemipmaps) ; diff --git a/indra/newview/llviewertexturelist.cpp b/indra/newview/llviewertexturelist.cpp index db740b69e9..fe26cd67a4 100644 --- a/indra/newview/llviewertexturelist.cpp +++ b/indra/newview/llviewertexturelist.cpp @@ -68,7 +68,6 @@ void (*LLViewerTextureList::sUUIDCallback)(void **, const LLUUID&) = NULL; S32 LLViewerTextureList::sNumImages = 0; LLViewerTextureList gTextureList; -static LLTrace::BlockTimerStatHandle FTM_PROCESS_IMAGES("Process Images"); ETexListType get_element_type(S32 priority) { @@ -761,18 +760,10 @@ void LLViewerTextureList::dirtyImage(LLViewerFetchedTexture *image) } //////////////////////////////////////////////////////////////////////////// -static LLTrace::BlockTimerStatHandle FTM_IMAGE_MARK_DIRTY("Dirty Images"); -static LLTrace::BlockTimerStatHandle FTM_IMAGE_UPDATE_PRIORITIES("Prioritize"); -static LLTrace::BlockTimerStatHandle FTM_IMAGE_CALLBACKS("Callbacks"); -static LLTrace::BlockTimerStatHandle FTM_IMAGE_FETCH("Fetch"); -static LLTrace::BlockTimerStatHandle FTM_FAST_CACHE_IMAGE_FETCH("Fast Cache Fetch"); -static LLTrace::BlockTimerStatHandle FTM_IMAGE_CREATE("Create"); -static LLTrace::BlockTimerStatHandle FTM_IMAGE_STATS("Stats"); -static LLTrace::BlockTimerStatHandle FTM_UPDATE_TEXTURES("Update Textures"); void LLViewerTextureList::updateImages(F32 max_time) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_TEXTURES); + LL_PROFILE_ZONE_SCOPED; static BOOL cleared = FALSE; if(gTeleportDisplay) { @@ -798,62 +789,44 @@ void LLViewerTextureList::updateImages(F32 max_time) sample(FORMATTED_MEM, F64Bytes(LLImageFormatted::sGlobalFormattedMemory)); } - { - //loading from fast cache - LL_RECORD_BLOCK_TIME(FTM_FAST_CACHE_IMAGE_FETCH); - max_time -= updateImagesLoadingFastCache(max_time); - } - - { - LL_RECORD_BLOCK_TIME(FTM_IMAGE_UPDATE_PRIORITIES); - updateImagesDecodePriorities(); - } - - F32 total_max_time = max_time; - - { - LL_RECORD_BLOCK_TIME(FTM_IMAGE_FETCH); - max_time -= updateImagesFetchTextures(max_time); - } + //loading from fast cache + max_time -= updateImagesLoadingFastCache(max_time); - { - LL_RECORD_BLOCK_TIME(FTM_IMAGE_CREATE); - max_time = llmax(max_time, total_max_time*.50f); // at least 50% of max_time - max_time -= updateImagesCreateTextures(max_time); - } + updateImagesDecodePriorities(); + + F32 total_max_time = max_time; + + max_time -= updateImagesFetchTextures(max_time); + + max_time = llmax(max_time, total_max_time*.50f); // at least 50% of max_time + max_time -= updateImagesCreateTextures(max_time); if (!mDirtyTextureList.empty()) { - LL_RECORD_BLOCK_TIME(FTM_IMAGE_MARK_DIRTY); gPipeline.dirtyPoolObjectTextures(mDirtyTextureList); mDirtyTextureList.clear(); } + bool didone = false; + for (image_list_t::iterator iter = mCallbackList.begin(); + iter != mCallbackList.end(); ) { - LL_RECORD_BLOCK_TIME(FTM_IMAGE_CALLBACKS); - bool didone = false; - for (image_list_t::iterator iter = mCallbackList.begin(); - iter != mCallbackList.end(); ) + //trigger loaded callbacks on local textures immediately + LLViewerFetchedTexture* image = *iter++; + if (!image->getUrl().empty()) { - //trigger loaded callbacks on local textures immediately - LLViewerFetchedTexture* image = *iter++; - if (!image->getUrl().empty()) - { - // Do stuff to handle callbacks, update priorities, etc. - didone = image->doLoadedCallbacks(); - } - else if (!didone) - { - // Do stuff to handle callbacks, update priorities, etc. - didone = image->doLoadedCallbacks(); - } + // Do stuff to handle callbacks, update priorities, etc. + didone = image->doLoadedCallbacks(); + } + else if (!didone) + { + // Do stuff to handle callbacks, update priorities, etc. + didone = image->doLoadedCallbacks(); } } + - { - LL_RECORD_BLOCK_TIME(FTM_IMAGE_STATS); - updateImagesUpdateStats(); - } + updateImagesUpdateStats(); } void LLViewerTextureList::clearFetchingRequests() @@ -1497,8 +1470,8 @@ void LLViewerTextureList::receiveImageHeader(LLMessageSystem *msg, void **user_d { static LLCachedControl log_texture_traffic(gSavedSettings,"LogTextureNetworkTraffic", false) ; - LL_RECORD_BLOCK_TIME(FTM_PROCESS_IMAGES); - + LL_PROFILE_ZONE_SCOPED; + // Receive image header, copy into image object and decompresses // if this is a one-packet image. @@ -1569,7 +1542,7 @@ void LLViewerTextureList::receiveImagePacket(LLMessageSystem *msg, void **user_d { static LLCachedControl log_texture_traffic(gSavedSettings,"LogTextureNetworkTraffic", false) ; - LL_RECORD_BLOCK_TIME(FTM_PROCESS_IMAGES); + LL_PROFILE_ZONE_SCOPED; // Receives image packet, copy into image object, // checks if all packets received, decompresses if so. @@ -1642,7 +1615,7 @@ void LLViewerTextureList::receiveImagePacket(LLMessageSystem *msg, void **user_d // static void LLViewerTextureList::processImageNotInDatabase(LLMessageSystem *msg,void **user_data) { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_IMAGES); + LL_PROFILE_ZONE_SCOPED; LLUUID image_id; msg->getUUIDFast(_PREHASH_ImageID, _PREHASH_ID, image_id); diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp index a588d05ff7..4a179146f8 100644 --- a/indra/newview/llvoavatar.cpp +++ b/indra/newview/llvoavatar.cpp @@ -328,6 +328,7 @@ public: // must return FALSE when the motion is completed. virtual BOOL onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; F32 nx[2]; nx[0]=time*TORSO_NOISE_SPEED; nx[1]=0.0f; @@ -448,6 +449,7 @@ public: // must return FALSE when the motion is completed. virtual BOOL onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; mBreatheRate = 1.f; F32 breathe_amt = (sinf(mBreatheRate * time) * BREATHE_ROT_MOTION_STRENGTH); @@ -549,6 +551,7 @@ public: // must return FALSE when the motion is completed. virtual BOOL onUpdate(F32 time, U8* joint_mask) { + LL_PROFILE_ZONE_SCOPED; mPelvisState->setPosition(LLVector3::zero); return TRUE; @@ -1318,11 +1321,9 @@ void LLVOAvatar::updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax) } -static LLTrace::BlockTimerStatHandle FTM_AVATAR_EXTENT_UPDATE("Av Upd Extent"); - void LLVOAvatar::calculateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax) { - LL_RECORD_BLOCK_TIME(FTM_AVATAR_EXTENT_UPDATE); + LL_PROFILE_ZONE_SCOPED; S32 box_detail = gSavedSettings.getS32("AvatarBoundingBoxComplexity"); if (getOverallAppearance() != AOA_NORMAL) @@ -2491,10 +2492,6 @@ S32 LLVOAvatar::setTETexture(const U8 te, const LLUUID& uuid) return setTETextureCore(te, image); } -static LLTrace::BlockTimerStatHandle FTM_AVATAR_UPDATE("Avatar Update"); -static LLTrace::BlockTimerStatHandle FTM_AVATAR_UPDATE_COMPLEXITY("Avatar Update Complexity"); -static LLTrace::BlockTimerStatHandle FTM_JOINT_UPDATE("Update Joints"); - //------------------------------------------------------------------------ // LLVOAvatar::dumpAnimationState() //------------------------------------------------------------------------ @@ -2527,7 +2524,7 @@ void LLVOAvatar::dumpAnimationState() //------------------------------------------------------------------------ void LLVOAvatar::idleUpdate(LLAgent &agent, const F64 &time) { - LL_RECORD_BLOCK_TIME(FTM_AVATAR_UPDATE); + LL_PROFILE_ZONE_SCOPED; if (isDead()) { @@ -2563,8 +2560,6 @@ void LLVOAvatar::idleUpdate(LLAgent &agent, const F64 &time) // force asynchronous drawable update if(mDrawable.notNull()) { - LL_RECORD_BLOCK_TIME(FTM_JOINT_UPDATE); - if (isSitting() && getParent()) { LLViewerObject *root_object = (LLViewerObject*)getRoot(); @@ -2664,9 +2659,8 @@ void LLVOAvatar::idleUpdate(LLAgent &agent, const F64 &time) if ((LLFrameTimer::getFrameCount() + mID.mData[0]) % compl_upd_freq == 0) { - LL_RECORD_BLOCK_TIME(FTM_AVATAR_UPDATE_COMPLEXITY); - idleUpdateRenderComplexity(); -} + idleUpdateRenderComplexity(); + } idleUpdateDebugInfo(); } @@ -2779,10 +2773,9 @@ void LLVOAvatar::idleUpdateVoiceVisualizer(bool voice_enabled) }//if ( voiceEnabled ) } -static LLTrace::BlockTimerStatHandle FTM_ATTACHMENT_UPDATE("Update Attachments"); - void LLVOAvatar::idleUpdateMisc(bool detailed_update) { + LL_PROFILE_ZONE_SCOPED; if (LLVOAvatar::sJointDebug) { LL_INFOS() << getFullname() << ": joint touches: " << LLJoint::sNumTouches << " updates: " << LLJoint::sNumUpdates << LL_ENDL; @@ -2796,7 +2789,6 @@ void LLVOAvatar::idleUpdateMisc(bool detailed_update) // update attachments positions if (detailed_update) { - LL_RECORD_BLOCK_TIME(FTM_ATTACHMENT_UPDATE); for (attachment_map_t::iterator iter = mAttachmentPoints.begin(); iter != mAttachmentPoints.end(); ++iter) @@ -7132,10 +7124,9 @@ void LLVOAvatar::updateGL() //----------------------------------------------------------------------------- // updateGeometry() //----------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_UPDATE_AVATAR("Update Avatar"); BOOL LLVOAvatar::updateGeometry(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_AVATAR); + LL_PROFILE_ZONE_SCOPED; if (!(gPipeline.hasRenderType(mIsControlAvatar ? LLPipeline::RENDER_TYPE_CONTROL_AV : LLPipeline::RENDER_TYPE_AVATAR))) { return TRUE; @@ -10193,6 +10184,7 @@ void showRigInfoTabExtents(LLVOAvatar *avatar, LLJointRiggingInfoTab& tab, S32& void LLVOAvatar::getAssociatedVolumes(std::vector& volumes) { + LL_PROFILE_ZONE_SCOPED; for ( LLVOAvatar::attachment_map_t::iterator iter = mAttachmentPoints.begin(); iter != mAttachmentPoints.end(); ++iter ) { LLViewerJointAttachment* attachment = iter->second; @@ -10250,27 +10242,19 @@ void LLVOAvatar::getAssociatedVolumes(std::vector& volumes) } } -static LLTrace::BlockTimerStatHandle FTM_AVATAR_RIGGING_INFO_UPDATE("Av Upd Rig Info"); -static LLTrace::BlockTimerStatHandle FTM_AVATAR_RIGGING_KEY_UPDATE("Av Upd Rig Key"); -static LLTrace::BlockTimerStatHandle FTM_AVATAR_RIGGING_AVOL_UPDATE("Av Upd Avol"); - // virtual void LLVOAvatar::updateRiggingInfo() { - LL_RECORD_BLOCK_TIME(FTM_AVATAR_RIGGING_INFO_UPDATE); + LL_PROFILE_ZONE_SCOPED; LL_DEBUGS("RigSpammish") << getFullname() << " updating rig tab" << LL_ENDL; std::vector volumes; - { - LL_RECORD_BLOCK_TIME(FTM_AVATAR_RIGGING_AVOL_UPDATE); - getAssociatedVolumes(volumes); - } + getAssociatedVolumes(volumes); std::map curr_rigging_info_key; { - LL_RECORD_BLOCK_TIME(FTM_AVATAR_RIGGING_KEY_UPDATE); // Get current rigging info key for (std::vector::iterator it = volumes.begin(); it != volumes.end(); ++it) { @@ -10432,6 +10416,7 @@ void LLVOAvatar::updateImpostorRendering(U32 newMaxNonImpostorsValue) void LLVOAvatar::idleUpdateRenderComplexity() { + LL_PROFILE_ZONE_SCOPED; if (isControlAvatar()) { LLControlAvatar *cav = dynamic_cast(this); diff --git a/indra/newview/llvoavatar.h b/indra/newview/llvoavatar.h index 39adaab8ca..7042406091 100644 --- a/indra/newview/llvoavatar.h +++ b/indra/newview/llvoavatar.h @@ -87,6 +87,7 @@ class LLVOAvatar : public LLViewerObject, public boost::signals2::trackable { + LL_ALIGN_NEW; LOG_CLASS(LLVOAvatar); public: @@ -99,16 +100,6 @@ public: **/ public: - void* operator new(size_t size) - { - return LLTrace::MemTrackable::aligned_new<16>(size); - } - - void operator delete(void* ptr, size_t size) - { - LLTrace::MemTrackable::aligned_delete<16>(ptr, size); - } - LLVOAvatar(const LLUUID &id, const LLPCode pcode, LLViewerRegion *regionp); virtual void markDead(); static void initClass(); // Initialize data that's only init'd once per class. diff --git a/indra/newview/llvocache.cpp b/indra/newview/llvocache.cpp index 5ebc65405f..e10a9f9bcb 100644 --- a/indra/newview/llvocache.cpp +++ b/indra/newview/llvocache.cpp @@ -63,8 +63,7 @@ BOOL check_write(LLAPRFile* apr_file, void* src, S32 n_bytes) //--------------------------------------------------------------------------- LLVOCacheEntry::LLVOCacheEntry(U32 local_id, U32 crc, LLDataPackerBinaryBuffer &dp) -: LLTrace::MemTrackable("LLVOCacheEntry"), - LLViewerOctreeEntryData(LLViewerOctreeEntry::LLVOCACHEENTRY), +: LLViewerOctreeEntryData(LLViewerOctreeEntry::LLVOCACHEENTRY), mLocalID(local_id), mCRC(crc), mUpdateFlags(-1), @@ -83,8 +82,7 @@ LLVOCacheEntry::LLVOCacheEntry(U32 local_id, U32 crc, LLDataPackerBinaryBuffer & } LLVOCacheEntry::LLVOCacheEntry() -: LLTrace::MemTrackable("LLVOCacheEntry"), - LLViewerOctreeEntryData(LLViewerOctreeEntry::LLVOCACHEENTRY), +: LLViewerOctreeEntryData(LLViewerOctreeEntry::LLVOCACHEENTRY), mLocalID(0), mCRC(0), mUpdateFlags(-1), @@ -102,8 +100,7 @@ LLVOCacheEntry::LLVOCacheEntry() } LLVOCacheEntry::LLVOCacheEntry(LLAPRFile* apr_file) -: LLTrace::MemTrackable("LLVOCacheEntry"), - LLViewerOctreeEntryData(LLViewerOctreeEntry::LLVOCACHEENTRY), +: LLViewerOctreeEntryData(LLViewerOctreeEntry::LLVOCACHEENTRY), mBuffer(NULL), mUpdateFlags(-1), mState(INACTIVE), @@ -619,7 +616,6 @@ void LLVOCacheGroup::handleChildAddition(const OctreeNode* parent, OctreeNode* c } LLVOCachePartition::LLVOCachePartition(LLViewerRegion* regionp) -: LLTrace::MemTrackable("LLVOCachePartition") { mLODPeriod = 16; mRegionp = regionp; diff --git a/indra/newview/llvocache.h b/indra/newview/llvocache.h index dd6afd6b85..c510ff77fc 100644 --- a/indra/newview/llvocache.h +++ b/indra/newview/llvocache.h @@ -38,9 +38,9 @@ class LLCamera; class LLVOCacheEntry -: public LLViewerOctreeEntryData, - public LLTrace::MemTrackable +: public LLViewerOctreeEntryData { + LL_ALIGN_NEW public: enum { @@ -185,7 +185,7 @@ protected: virtual ~LLVOCacheGroup(); }; -class LLVOCachePartition : public LLViewerOctreePartition, public LLTrace::MemTrackable +class LLVOCachePartition : public LLViewerOctreePartition { public: LLVOCachePartition(LLViewerRegion* regionp); diff --git a/indra/newview/llvograss.cpp b/indra/newview/llvograss.cpp index 345e87eea8..9a41eedb54 100644 --- a/indra/newview/llvograss.cpp +++ b/indra/newview/llvograss.cpp @@ -661,11 +661,9 @@ void LLGrassPartition::addGeometryCount(LLSpatialGroup* group, U32& vertex_count } } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_GRASS_VB("Grass VB"); - void LLGrassPartition::getGeometry(LLSpatialGroup* group) { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_GRASS_VB); + LL_PROFILE_ZONE_SCOPED; std::sort(mFaceList.begin(), mFaceList.end(), LLFace::CompareDistanceGreater()); diff --git a/indra/newview/llvopartgroup.cpp b/indra/newview/llvopartgroup.cpp index b31afca61d..068e8a131d 100644 --- a/indra/newview/llvopartgroup.cpp +++ b/indra/newview/llvopartgroup.cpp @@ -306,10 +306,9 @@ LLVector3 LLVOPartGroup::getCameraPosition() const return gAgentCamera.getCameraPositionAgent(); } -static LLTrace::BlockTimerStatHandle FTM_UPDATE_PARTICLES("Update Particles"); BOOL LLVOPartGroup::updateGeometry(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_PARTICLES); + LL_PROFILE_ZONE_SCOPED; dirtySpatialGroup(); @@ -754,10 +753,9 @@ LLHUDParticlePartition::LLHUDParticlePartition(LLViewerRegion* regionp) : mPartitionType = LLViewerRegion::PARTITION_HUD_PARTICLE; } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_PARTICLE_VBO("Particle VBO"); - void LLParticlePartition::rebuildGeom(LLSpatialGroup* group) { + LL_PROFILE_ZONE_SCOPED; if (group->isDead() || !group->hasState(LLSpatialGroup::GEOM_DIRTY)) { return; @@ -769,8 +767,6 @@ void LLParticlePartition::rebuildGeom(LLSpatialGroup* group) group->mLastUpdateViewAngle = group->mViewAngle; } - LL_RECORD_BLOCK_TIME(FTM_REBUILD_PARTICLE_VBO); - group->clearDrawMap(); //get geometry count @@ -843,11 +839,9 @@ void LLParticlePartition::addGeometryCount(LLSpatialGroup* group, U32& vertex_co } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_PARTICLE_GEOM("Particle Geom"); - void LLParticlePartition::getGeometry(LLSpatialGroup* group) { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_PARTICLE_GEOM); + LL_PROFILE_ZONE_SCOPED; std::sort(mFaceList.begin(), mFaceList.end(), LLFace::CompareDistanceGreater()); diff --git a/indra/newview/llvosky.cpp b/indra/newview/llvosky.cpp index edf8c40bd3..1e546861b9 100644 --- a/indra/newview/llvosky.cpp +++ b/indra/newview/llvosky.cpp @@ -79,11 +79,6 @@ namespace const LLVector2 TEX10 = LLVector2(1.f, 0.f); const LLVector2 TEX11 = LLVector2(1.f, 1.f); - LLTrace::BlockTimerStatHandle FTM_VOSKY_UPDATETIMER("VOSky Update Timer Tick"); - LLTrace::BlockTimerStatHandle FTM_VOSKY_CALC("VOSky Update Calculations"); - LLTrace::BlockTimerStatHandle FTM_VOSKY_CREATETEXTURES("VOSky Update Textures"); - LLTrace::BlockTimerStatHandle FTM_VOSKY_UPDATEFORCED("VOSky Update Forced"); - F32Seconds UPDATE_EXPRY(0.25f); const F32 UPDATE_MIN_DELTA_THRESHOLD = 0.0005f; @@ -518,6 +513,7 @@ void LLVOSky::cacheEnvironment(LLSettingsSky::ptr_t psky,AtmosphericsVars& atmos void LLVOSky::calc() { + LL_PROFILE_ZONE_SCOPED; LLSettingsSky::ptr_t psky = LLEnvironment::instance().getCurrentSky(); cacheEnvironment(psky,m_atmosphericsVars); @@ -686,6 +682,8 @@ bool LLVOSky::updateSky() return TRUE; } + LL_PROFILE_ZONE_SCOPED; + static S32 next_frame = 0; mNeedUpdate = mForceUpdate; @@ -706,7 +704,6 @@ bool LLVOSky::updateSky() if (mCubeMapUpdateStage < 0) { - LL_RECORD_BLOCK_TIME(FTM_VOSKY_CALC); calc(); bool same_atmospherics = approximatelyEqual(m_lastAtmosphericsVars, m_atmosphericsVars, UPDATE_MIN_DELTA_THRESHOLD); @@ -723,7 +720,7 @@ bool LLVOSky::updateSky() } else if (mCubeMapUpdateStage == NUM_CUBEMAP_FACES) { - LL_RECORD_BLOCK_TIME(FTM_VOSKY_UPDATEFORCED); + LL_PROFILE_ZONE_NAMED("updateSky - forced"); LLSkyTex::stepCurrent(); bool is_alm_wl_sky = gPipeline.canUseWindLightShaders(); @@ -783,8 +780,8 @@ bool LLVOSky::updateSky() } // run 0 to 5 faces, each face in own frame else if (mCubeMapUpdateStage >= 0 && mCubeMapUpdateStage < NUM_CUBEMAP_FACES) - { - LL_RECORD_BLOCK_TIME(FTM_VOSKY_CREATETEXTURES); + { + LL_PROFILE_ZONE_NAMED("updateSky - create"); S32 side = mCubeMapUpdateStage; // CPU hungry part, createSkyTexture() is math heavy // Prior to EEP it was mostly per tile, but since EPP it is per face. @@ -974,11 +971,9 @@ void LLVOSky::setBloomTextures(const LLUUID& bloom_texture, const LLUUID& bloom_ } } -static LLTrace::BlockTimerStatHandle FTM_GEO_SKY("Sky Geometry"); - BOOL LLVOSky::updateGeometry(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_GEO_SKY); + LL_PROFILE_ZONE_SCOPED; if (mFace[FACE_REFLECTION] == NULL) { LLDrawPoolWater *poolp = (LLDrawPoolWater*) gPipeline.getPool(LLDrawPool::POOL_WATER); diff --git a/indra/newview/llvosurfacepatch.cpp b/indra/newview/llvosurfacepatch.cpp index b5560d3d79..4d25e8c7bd 100644 --- a/indra/newview/llvosurfacepatch.cpp +++ b/indra/newview/llvosurfacepatch.cpp @@ -212,7 +212,6 @@ LLDrawable *LLVOSurfacePatch::createDrawable(LLPipeline *pipeline) return mDrawable; } -static LLTrace::BlockTimerStatHandle FTM_UPDATE_TERRAIN("Update Terrain"); void LLVOSurfacePatch::updateGL() { @@ -225,7 +224,7 @@ void LLVOSurfacePatch::updateGL() BOOL LLVOSurfacePatch::updateGeometry(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_TERRAIN); + LL_PROFILE_ZONE_SCOPED; dirtySpatialGroup(TRUE); @@ -1071,10 +1070,9 @@ LLVertexBuffer* LLTerrainPartition::createVertexBuffer(U32 type_mask, U32 usage) return new LLVertexBufferTerrain(); } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_TERRAIN_VB("Terrain VB"); void LLTerrainPartition::getGeometry(LLSpatialGroup* group) { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_TERRAIN_VB); + LL_PROFILE_ZONE_SCOPED; LLVertexBuffer* buffer = group->mVertexBuffer; diff --git a/indra/newview/llvotree.cpp b/indra/newview/llvotree.cpp index 41099cb570..493162b47b 100644 --- a/indra/newview/llvotree.cpp +++ b/indra/newview/llvotree.cpp @@ -493,11 +493,9 @@ LLDrawable* LLVOTree::createDrawable(LLPipeline *pipeline) const S32 LEAF_INDICES = 24; const S32 LEAF_VERTICES = 16; -static LLTrace::BlockTimerStatHandle FTM_UPDATE_TREE("Update Tree"); - BOOL LLVOTree::updateGeometry(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_TREE); + LL_PROFILE_ZONE_SCOPED; if(mTrunkLOD >= sMAX_NUM_TREE_LOD_LEVELS) //do not display the tree. { diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index b86935b081..be390e44a3 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -103,10 +103,6 @@ S32 LLVOVolume::mRenderComplexity_current = 0; LLPointer LLVOVolume::sObjectMediaClient = NULL; LLPointer LLVOVolume::sObjectMediaNavigateClient = NULL; -static LLTrace::BlockTimerStatHandle FTM_GEN_TRIANGLES("Generate Triangles"); -static LLTrace::BlockTimerStatHandle FTM_GEN_VOLUME("Generate Volumes"); -static LLTrace::BlockTimerStatHandle FTM_VOLUME_TEXTURES("Volume Textures"); - extern BOOL gGLDebugLoggingEnabled; // Implementation class of LLMediaDataClientObject. See llmediadataclient.h @@ -692,7 +688,7 @@ BOOL LLVOVolume::isVisible() const void LLVOVolume::updateTextureVirtualSize(bool forced) { - LL_RECORD_BLOCK_TIME(FTM_VOLUME_TEXTURES); + LL_PROFILE_ZONE_SCOPED; // Update the pixel area of all faces if (mDrawable.isNull()) @@ -971,6 +967,7 @@ LLDrawable *LLVOVolume::createDrawable(LLPipeline *pipeline) BOOL LLVOVolume::setVolume(const LLVolumeParams ¶ms_in, const S32 detail, bool unique_volume) { + LL_PROFILE_ZONE_SCOPED; LLVolumeParams volume_params = params_in; S32 last_lod = mVolumep.notNull() ? LLVolumeLODGroup::getVolumeDetailFromScale(mVolumep->getDetail()) : -1; @@ -1598,6 +1595,7 @@ BOOL LLVOVolume::setParent(LLViewerObject* parent) // NOTE: regenFaces() MUST be followed by genTriangles()! void LLVOVolume::regenFaces() { + LL_PROFILE_ZONE_SCOPED; // remove existing faces BOOL count_changed = mNumFaces != getNumTEs(); @@ -1645,6 +1643,7 @@ void LLVOVolume::regenFaces() BOOL LLVOVolume::genBBoxes(BOOL force_global) { + LL_PROFILE_ZONE_SCOPED; BOOL res = TRUE; LLVector4a min,max; @@ -1855,12 +1854,9 @@ void LLVOVolume::updateRelativeXform(bool force_identity) } } -static LLTrace::BlockTimerStatHandle FTM_GEN_FLEX("Generate Flexies"); -static LLTrace::BlockTimerStatHandle FTM_UPDATE_PRIMITIVES("Update Primitives"); -static LLTrace::BlockTimerStatHandle FTM_UPDATE_RIGGED_VOLUME("Update Rigged"); - bool LLVOVolume::lodOrSculptChanged(LLDrawable *drawable, BOOL &compiled) { + LL_PROFILE_ZONE_SCOPED; bool regen_faces = false; LLVolume *old_volumep, *new_volumep; @@ -1873,7 +1869,6 @@ bool LLVOVolume::lodOrSculptChanged(LLDrawable *drawable, BOOL &compiled) old_volumep = NULL; { - LL_RECORD_BLOCK_TIME(FTM_GEN_VOLUME); const LLVolumeParams &volume_params = getVolume()->getParams(); setVolume(volume_params, 0); } @@ -1901,7 +1896,6 @@ bool LLVOVolume::lodOrSculptChanged(LLDrawable *drawable, BOOL &compiled) drawable->setState(LLDrawable::REBUILD_VOLUME); // for face->genVolumeTriangles() { - LL_RECORD_BLOCK_TIME(FTM_GEN_TRIANGLES); regen_faces = new_num_faces != old_num_faces || mNumFaces != (S32)getNumTEs(); if (regen_faces) { @@ -1926,14 +1920,11 @@ bool LLVOVolume::lodOrSculptChanged(LLDrawable *drawable, BOOL &compiled) BOOL LLVOVolume::updateGeometry(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_PRIMITIVES); + LL_PROFILE_ZONE_SCOPED; if (mDrawable->isState(LLDrawable::REBUILD_RIGGED)) { - { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_RIGGED_VOLUME); - updateRiggedVolume(); - } + updateRiggedVolume(); genBBoxes(FALSE); mDrawable->clearState(LLDrawable::REBUILD_RIGGED); } @@ -1942,7 +1933,6 @@ BOOL LLVOVolume::updateGeometry(LLDrawable *drawable) { BOOL res; { - LL_RECORD_BLOCK_TIME(FTM_GEN_FLEX); res = mVolumeImpl->doUpdateGeometry(drawable); } updateFaceFlags(); @@ -1982,7 +1972,6 @@ BOOL LLVOVolume::updateGeometry(LLDrawable *drawable) } if (!was_regen_faces) { - LL_RECORD_BLOCK_TIME(FTM_GEN_TRIANGLES); regenFaces(); } @@ -2005,7 +1994,6 @@ BOOL LLVOVolume::updateGeometry(LLDrawable *drawable) { compiled = TRUE; // All it did was move or we changed the texture coordinate offset - LL_RECORD_BLOCK_TIME(FTM_GEN_TRIANGLES); genBBoxes(FALSE); } @@ -3717,11 +3705,9 @@ void LLVOVolume::afterReparent() } //---------------------------------------------------------------------------- -static LLTrace::BlockTimerStatHandle FTM_VOVOL_RIGGING_INFO("VOVol Rigging Info"); - void LLVOVolume::updateRiggingInfo() { - LL_RECORD_BLOCK_TIME(FTM_VOVOL_RIGGING_INFO); + LL_PROFILE_ZONE_SCOPED; if (isRiggedMesh()) { const LLMeshSkinInfo* skin = getSkinInfo(); @@ -4726,6 +4712,7 @@ void LLVOVolume::clearRiggedVolume() void LLVOVolume::updateRiggedVolume(bool force_update) { + LL_PROFILE_ZONE_SCOPED; //Update mRiggedVolume to match current animation frame of avatar. //Also update position/size in octree. @@ -4761,11 +4748,9 @@ void LLVOVolume::updateRiggedVolume(bool force_update) mRiggedVolume->update(skin, avatar, volume); } -static LLTrace::BlockTimerStatHandle FTM_SKIN_RIGGED("Skin"); -static LLTrace::BlockTimerStatHandle FTM_RIGGED_OCTREE("Octree"); - void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, const LLVolume* volume) { + LL_PROFILE_ZONE_SCOPED; bool copy = false; if (volume->getNumVolumeFaces() != getNumVolumeFaces()) { @@ -4829,8 +4814,6 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons if (pos && dst_face.mExtents) { - LL_RECORD_BLOCK_TIME(FTM_SKIN_RIGGED); - U32 max_joints = LLSkinningUtil::getMaxJointCount(); rigged_vert_count += dst_face.mNumVertices; rigged_face_count++; @@ -4901,8 +4884,7 @@ void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, cons } { - LL_RECORD_BLOCK_TIME(FTM_RIGGED_OCTREE); - delete dst_face.mOctree; + delete dst_face.mOctree; dst_face.mOctree = NULL; LLVector4a size; @@ -5066,11 +5048,9 @@ void LLVolumeGeometryManager::freeFaces() sAlphaFaces = NULL; } -static LLTrace::BlockTimerStatHandle FTM_REGISTER_FACE("Register Face"); - void LLVolumeGeometryManager::registerFace(LLSpatialGroup* group, LLFace* facep, U32 type) { - LL_RECORD_BLOCK_TIME(FTM_REGISTER_FACE); + LL_PROFILE_ZONE_SCOPED; if ( type == LLRenderPass::PASS_ALPHA && facep->getTextureEntry()->getMaterialParams().notNull() && !facep->getVertexBuffer()->hasDataType(LLVertexBuffer::TYPE_TANGENT) @@ -5302,10 +5282,6 @@ void LLVolumeGeometryManager::getGeometry(LLSpatialGroup* group) } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_VOLUME_VB("Volume VB"); -static LLTrace::BlockTimerStatHandle FTM_REBUILD_VOLUME_FACE_LIST("Build Face List"); -static LLTrace::BlockTimerStatHandle FTM_REBUILD_VOLUME_GEN_DRAW_INFO("Gen Draw Info"); - static LLDrawPoolAvatar* get_avatar_drawpool(LLViewerObject* vobj) { LLVOAvatar* avatar = vobj->getAvatar(); @@ -5411,6 +5387,7 @@ void handleRenderAutoMuteByteLimitChanged(const LLSD& new_value) void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) { + LL_PROFILE_ZONE_SCOPED; if (group->changeLOD()) { group->mLastUpdateDistance = group->mDistance; @@ -5427,8 +5404,6 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) return; } - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_VB); - group->mBuilt = 1.f; LLSpatialBridge* bridge = group->getSpatialPartition()->asBridge(); @@ -5485,7 +5460,7 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) #endif { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_FACE_LIST); + LL_PROFILE_ZONE_NAMED("rebuildGeom - face list"); //get all the faces into a list for (LLSpatialGroup::element_iter drawable_iter = group->getDataBegin(); @@ -6036,23 +6011,16 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) mFaceList.clear(); } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_MESH_FLUSH("Flush Mesh"); - void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) { + LL_PROFILE_ZONE_SCOPED; llassert(group); if (group && group->hasState(LLSpatialGroup::MESH_DIRTY) && !group->hasState(LLSpatialGroup::GEOM_DIRTY)) { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_VB); { - // SL-15709 -- NOTE: Tracy only allows one ZoneScoped per function. - // Solutions are: - // 1. Use a new scope - // 2. Use named zones - // 3. Use transient zones - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); //make sure getgeometryvolume shows up in the right place in timers + LL_PROFILE_ZONE_NAMED("rebuildMesh - gen draw info"); - group->mBuilt = 1.f; + group->mBuilt = 1.f; S32 num_mapped_vertex_buffer = LLVertexBuffer::sMappedCount ; @@ -6123,7 +6091,7 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) } { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_MESH_FLUSH); + LL_PROFILE_ZONE_NAMED("rebuildMesh - flush"); for (LLVertexBuffer** iter = locked_buffer, ** end_iter = locked_buffer+buffer_count; iter != end_iter; ++iter) { (*iter)->flush(); @@ -6166,10 +6134,7 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) group->clearState(LLSpatialGroup::MESH_DIRTY | LLSpatialGroup::NEW_DRAWINFO); } - - } // Tracy integration - -// llassert(!group || !group->isState(LLSpatialGroup::NEW_DRAWINFO)); + } } struct CompareBatchBreakerModified @@ -6202,19 +6167,10 @@ struct CompareBatchBreakerModified } }; -static LLTrace::BlockTimerStatHandle FTM_GEN_DRAW_INFO_SORT("Draw Info Face Sort"); -static LLTrace::BlockTimerStatHandle FTM_GEN_DRAW_INFO_FACE_SIZE("Face Sizing"); -static LLTrace::BlockTimerStatHandle FTM_GEN_DRAW_INFO_ALLOCATE("Allocate VB"); -static LLTrace::BlockTimerStatHandle FTM_GEN_DRAW_INFO_FIND_VB("Find VB"); -static LLTrace::BlockTimerStatHandle FTM_GEN_DRAW_INFO_RESIZE_VB("Resize VB"); - - - - U32 LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, LLFace** faces, U32 face_count, BOOL distance_sort, BOOL batch_textures, BOOL no_materials) { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_VOLUME_GEN_DRAW_INFO); + LL_PROFILE_ZONE_SCOPED; U32 geometryBytes = 0; U32 buffer_usage = group->mBufferUsage; @@ -6246,7 +6202,7 @@ U32 LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, LLFace max_vertices = llmin(max_vertices, (U32) 65535); { - LL_RECORD_BLOCK_TIME(FTM_GEN_DRAW_INFO_SORT); + LL_PROFILE_ZONE_NAMED("genDrawInfo - sort"); if (!distance_sort) { //sort faces by things that break batches @@ -6332,7 +6288,7 @@ U32 LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, LLFace U32 texture_count = 0; { - LL_RECORD_BLOCK_TIME(FTM_GEN_DRAW_INFO_FACE_SIZE); + LL_PROFILE_ZONE_NAMED("genDrawInfo - face size"); if (batch_textures) { U8 cur_tex = 0; @@ -6455,7 +6411,7 @@ U32 LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, LLFace LLPointer buffer; { - LL_RECORD_BLOCK_TIME(FTM_GEN_DRAW_INFO_ALLOCATE); + LL_PROFILE_ZONE_NAMED("genDrawInfo - allocate"); buffer = createVertexBuffer(mask, buffer_usage); if(!buffer->allocateBuffer(geom_count, index_count, TRUE)) { @@ -6843,6 +6799,8 @@ U32 LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, LLFace void LLGeometryManager::addGeometryCount(LLSpatialGroup* group, U32 &vertex_count, U32 &index_count) { + LL_PROFILE_ZONE_SCOPED; + //initialize to default usage for this partition U32 usage = group->getSpatialPartition()->mBufferUsage; diff --git a/indra/newview/llvowater.cpp b/indra/newview/llvowater.cpp index 12def24a0d..efe6aa158e 100644 --- a/indra/newview/llvowater.cpp +++ b/indra/newview/llvowater.cpp @@ -114,11 +114,9 @@ LLDrawable *LLVOWater::createDrawable(LLPipeline *pipeline) return mDrawable; } -static LLTrace::BlockTimerStatHandle FTM_UPDATE_WATER("Update Water"); - BOOL LLVOWater::updateGeometry(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_WATER); + LL_PROFILE_ZONE_SCOPED; LLFace *face; if (drawable->getNumFaces() < 1) diff --git a/indra/newview/llvowlsky.cpp b/indra/newview/llvowlsky.cpp index d428cb1568..5a8126fa38 100644 --- a/indra/newview/llvowlsky.cpp +++ b/indra/newview/llvowlsky.cpp @@ -141,11 +141,9 @@ void LLVOWLSky::restoreGL() gPipeline.markRebuild(mDrawable, LLDrawable::REBUILD_ALL, TRUE); } -static LLTrace::BlockTimerStatHandle FTM_GEO_SKY("Windlight Sky Geometry"); - BOOL LLVOWLSky::updateGeometry(LLDrawable * drawable) { - LL_RECORD_BLOCK_TIME(FTM_GEO_SKY); + LL_PROFILE_ZONE_SCOPED; LLStrider vertices; LLStrider texCoords; LLStrider indices; diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp index c7b0a2bfb4..0a8457eb2c 100644 --- a/indra/newview/llworld.cpp +++ b/indra/newview/llworld.cpp @@ -62,6 +62,8 @@ #include +LLWorld* LLSimpleton::sInstance = nullptr; + // // Globals // @@ -135,6 +137,7 @@ void LLWorld::destroyClass() LLDrawable::incrementVisible(); LLSceneMonitor::deleteSingleton(); + LLWorld::deleteSingleton(); } @@ -1077,6 +1080,7 @@ void LLWorld::updateWaterObjects() void LLWorld::shiftRegions(const LLVector3& offset) { + LL_PROFILE_ZONE_SCOPED; for (region_list_t::const_iterator i = getRegionList().begin(); i != getRegionList().end(); ++i) { LLViewerRegion* region = *i; @@ -1147,11 +1151,9 @@ void LLWorld::disconnectRegions() } } -static LLTrace::BlockTimerStatHandle FTM_ENABLE_SIMULATOR("Enable Sim"); - void process_enable_simulator(LLMessageSystem *msg, void **user_data) { - LL_RECORD_BLOCK_TIME(FTM_ENABLE_SIMULATOR); + LL_PROFILE_ZONE_SCOPED; // enable the appropriate circuit for this simulator and // add its values into the gSimulator structure U64 handle; @@ -1217,12 +1219,11 @@ public: } }; -static LLTrace::BlockTimerStatHandle FTM_DISABLE_REGION("Disable Region"); // disable the circuit to this simulator // Called in response to "DisableSimulator" message. void process_disable_simulator(LLMessageSystem *mesgsys, void **user_data) { - LL_RECORD_BLOCK_TIME(FTM_DISABLE_REGION); + LL_PROFILE_ZONE_SCOPED; LLHost host = mesgsys->getSender(); diff --git a/indra/newview/llworld.h b/indra/newview/llworld.h index 98552bc4d1..69f2df4203 100644 --- a/indra/newview/llworld.h +++ b/indra/newview/llworld.h @@ -59,11 +59,12 @@ class LLVOAvatar; // as simulators are connected to, viewer_regions are popped off the stack and connected as required // as simulators are removed, they are pushed back onto the stack -class LLWorld : public LLSingleton +class LLWorld : public LLSimpleton { - LLSINGLETON(LLWorld); public: - void destroyClass(); + LLWorld(); + + void destroyClass(); LLViewerRegion* addRegion(const U64 ®ion_handle, const LLHost &host); // safe to call if already present, does the "right thing" if diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index 7aa05fb22f..6ed6e20b03 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -238,7 +238,6 @@ const LLMatrix4* gGLLastMatrix = NULL; LLTrace::BlockTimerStatHandle FTM_RENDER_GEOMETRY("Render Geometry"); LLTrace::BlockTimerStatHandle FTM_RENDER_GRASS("Grass"); LLTrace::BlockTimerStatHandle FTM_RENDER_INVISIBLE("Invisible"); -LLTrace::BlockTimerStatHandle FTM_RENDER_OCCLUSION("Occlusion"); LLTrace::BlockTimerStatHandle FTM_RENDER_SHINY("Shiny"); LLTrace::BlockTimerStatHandle FTM_RENDER_SIMPLE("Simple"); LLTrace::BlockTimerStatHandle FTM_RENDER_TERRAIN("Terrain"); @@ -253,14 +252,12 @@ LLTrace::BlockTimerStatHandle FTM_RENDER_MATERIALS("Render Materials"); LLTrace::BlockTimerStatHandle FTM_RENDER_FULLBRIGHT("Fullbright"); LLTrace::BlockTimerStatHandle FTM_RENDER_GLOW("Glow"); LLTrace::BlockTimerStatHandle FTM_GEO_UPDATE("Geo Update"); -LLTrace::BlockTimerStatHandle FTM_PIPELINE_CREATE("Pipeline Create"); LLTrace::BlockTimerStatHandle FTM_POOLRENDER("RenderPool"); LLTrace::BlockTimerStatHandle FTM_POOLS("Pools"); LLTrace::BlockTimerStatHandle FTM_DEFERRED_POOLRENDER("RenderPool (Deferred)"); LLTrace::BlockTimerStatHandle FTM_DEFERRED_POOLS("Pools (Deferred)"); LLTrace::BlockTimerStatHandle FTM_POST_DEFERRED_POOLRENDER("RenderPool (Post)"); LLTrace::BlockTimerStatHandle FTM_POST_DEFERRED_POOLS("Pools (Post)"); -LLTrace::BlockTimerStatHandle FTM_RENDER_BLOOM_FBO("First FBO"); LLTrace::BlockTimerStatHandle FTM_STATESORT("Sort Draw State"); LLTrace::BlockTimerStatHandle FTM_PIPELINE("Pipeline"); LLTrace::BlockTimerStatHandle FTM_CLIENT_COPY("Client Copy"); @@ -269,11 +266,8 @@ LLTrace::BlockTimerStatHandle FTM_RENDER_DEFERRED("Deferred Shading"); LLTrace::BlockTimerStatHandle FTM_RENDER_UI_HUD("HUD"); LLTrace::BlockTimerStatHandle FTM_RENDER_UI_3D("3D"); LLTrace::BlockTimerStatHandle FTM_RENDER_UI_2D("2D"); -LLTrace::BlockTimerStatHandle FTM_RENDER_UI_DEBUG_TEXT("Debug Text"); -LLTrace::BlockTimerStatHandle FTM_RENDER_UI_SCENE_MON("Scene Mon"); static LLTrace::BlockTimerStatHandle FTM_STATESORT_DRAWABLE("Sort Drawables"); -static LLTrace::BlockTimerStatHandle FTM_STATESORT_POSTSORT("Post Sort"); static LLStaticHashedString sTint("tint"); static LLStaticHashedString sAmbiance("ambiance"); @@ -727,8 +721,6 @@ void LLPipeline::destroyGL() } } -static LLTrace::BlockTimerStatHandle FTM_RESIZE_SCREEN_TEXTURE("Resize Screen Texture"); - void LLPipeline::requestResizeScreenTexture() { gResizeScreenTexture = TRUE; @@ -748,7 +740,6 @@ void LLPipeline::resizeShadowTexture() void LLPipeline::resizeScreenTexture() { - LL_RECORD_BLOCK_TIME(FTM_RESIZE_SCREEN_TEXTURE); if (gPipeline.canUseVertexShaders() && assertInitialized()) { GLuint resX = gViewerWindow->getWorldViewWidthRaw(); @@ -1518,6 +1509,7 @@ public: // Called when a texture changes # of channels (causes faces to move to alpha pool) void LLPipeline::dirtyPoolObjectTextures(const std::set& textures) { + LL_PROFILE_ZONE_SCOPED; assertInitialized(); // *TODO: This is inefficient and causes frame spikes; need a better way to do this @@ -1729,15 +1721,9 @@ void LLPipeline::allocDrawable(LLViewerObject *vobj) } -static LLTrace::BlockTimerStatHandle FTM_UNLINK("Unlink"); -static LLTrace::BlockTimerStatHandle FTM_REMOVE_FROM_MOVE_LIST("Movelist"); -static LLTrace::BlockTimerStatHandle FTM_REMOVE_FROM_SPATIAL_PARTITION("Spatial Partition"); -static LLTrace::BlockTimerStatHandle FTM_REMOVE_FROM_LIGHT_SET("Light Set"); -static LLTrace::BlockTimerStatHandle FTM_REMOVE_FROM_HIGHLIGHT_SET("Highlight Set"); - void LLPipeline::unlinkDrawable(LLDrawable *drawable) { - LL_RECORD_BLOCK_TIME(FTM_UNLINK); + LL_PROFILE_ZONE_SCOPED; assertInitialized(); @@ -1746,7 +1732,6 @@ void LLPipeline::unlinkDrawable(LLDrawable *drawable) // Based on flags, remove the drawable from the queues that it's on. if (drawablep->isState(LLDrawable::ON_MOVE_LIST)) { - LL_RECORD_BLOCK_TIME(FTM_REMOVE_FROM_MOVE_LIST); LLDrawable::drawable_vector_t::iterator iter = std::find(mMovedList.begin(), mMovedList.end(), drawablep); if (iter != mMovedList.end()) { @@ -1756,7 +1741,6 @@ void LLPipeline::unlinkDrawable(LLDrawable *drawable) if (drawablep->getSpatialGroup()) { - LL_RECORD_BLOCK_TIME(FTM_REMOVE_FROM_SPATIAL_PARTITION); if (!drawablep->getSpatialGroup()->getSpatialPartition()->remove(drawablep, drawablep->getSpatialGroup())) { #ifdef LL_RELEASE_FOR_DOWNLOAD @@ -1767,30 +1751,24 @@ void LLPipeline::unlinkDrawable(LLDrawable *drawable) } } - { - LL_RECORD_BLOCK_TIME(FTM_REMOVE_FROM_LIGHT_SET); - mLights.erase(drawablep); + mLights.erase(drawablep); - for (light_set_t::iterator iter = mNearbyLights.begin(); - iter != mNearbyLights.end(); iter++) + for (light_set_t::iterator iter = mNearbyLights.begin(); + iter != mNearbyLights.end(); iter++) + { + if (iter->drawable == drawablep) { - if (iter->drawable == drawablep) - { - mNearbyLights.erase(iter); - break; - } + mNearbyLights.erase(iter); + break; } } - { - LL_RECORD_BLOCK_TIME(FTM_REMOVE_FROM_HIGHLIGHT_SET); - HighlightItem item(drawablep); - mHighlightSet.erase(item); + HighlightItem item(drawablep); + mHighlightSet.erase(item); - if (mHighlightObject == drawablep) - { - mHighlightObject = NULL; - } + if (mHighlightObject == drawablep) + { + mHighlightObject = NULL; } for (U32 i = 0; i < 2; ++i) @@ -1805,14 +1783,12 @@ void LLPipeline::unlinkDrawable(LLDrawable *drawable) mTargetShadowSpotLight[i] = NULL; } } - - } //static void LLPipeline::removeMutedAVsLights(LLVOAvatar* muted_avatar) { - LL_RECORD_BLOCK_TIME(FTM_REMOVE_FROM_LIGHT_SET); + LL_PROFILE_ZONE_SCOPED; for (light_set_t::iterator iter = gPipeline.mNearbyLights.begin(); iter != gPipeline.mNearbyLights.end(); iter++) { @@ -1840,7 +1816,7 @@ U32 LLPipeline::addObject(LLViewerObject *vobj) void LLPipeline::createObjects(F32 max_dtime) { - LL_RECORD_BLOCK_TIME(FTM_PIPELINE_CREATE); + LL_PROFILE_ZONE_SCOPED; LLTimer update_timer; @@ -1864,6 +1840,7 @@ void LLPipeline::createObjects(F32 max_dtime) void LLPipeline::createObject(LLViewerObject* vobj) { + LL_PROFILE_ZONE_SCOPED; LLDrawable* drawablep = vobj->mDrawable; if (!drawablep) @@ -2010,14 +1987,9 @@ void LLPipeline::updateMovedList(LLDrawable::drawable_vector_t& moved_list) } } -static LLTrace::BlockTimerStatHandle FTM_OCTREE_BALANCE("Balance Octree"); -static LLTrace::BlockTimerStatHandle FTM_UPDATE_MOVE("Update Move"); -static LLTrace::BlockTimerStatHandle FTM_RETEXTURE("Retexture"); -static LLTrace::BlockTimerStatHandle FTM_MOVED_LIST("Moved List"); - void LLPipeline::updateMove() { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_MOVE); + LL_PROFILE_ZONE_SCOPED; if (FreezeTime) { @@ -2026,49 +1998,38 @@ void LLPipeline::updateMove() assertInitialized(); + for (LLDrawable::drawable_set_t::iterator iter = mRetexturedList.begin(); + iter != mRetexturedList.end(); ++iter) { - LL_RECORD_BLOCK_TIME(FTM_RETEXTURE); - - for (LLDrawable::drawable_set_t::iterator iter = mRetexturedList.begin(); - iter != mRetexturedList.end(); ++iter) + LLDrawable* drawablep = *iter; + if (drawablep && !drawablep->isDead()) { - LLDrawable* drawablep = *iter; - if (drawablep && !drawablep->isDead()) - { - drawablep->updateTexture(); - } + drawablep->updateTexture(); } - mRetexturedList.clear(); } + mRetexturedList.clear(); - { - LL_RECORD_BLOCK_TIME(FTM_MOVED_LIST); - updateMovedList(mMovedList); - } + updateMovedList(mMovedList); //balance octrees + for (LLWorld::region_list_t::const_iterator iter = LLWorld::getInstance()->getRegionList().begin(); + iter != LLWorld::getInstance()->getRegionList().end(); ++iter) { - LL_RECORD_BLOCK_TIME(FTM_OCTREE_BALANCE); - - for (LLWorld::region_list_t::const_iterator iter = LLWorld::getInstance()->getRegionList().begin(); - iter != LLWorld::getInstance()->getRegionList().end(); ++iter) + LLViewerRegion* region = *iter; + for (U32 i = 0; i < LLViewerRegion::NUM_PARTITIONS; i++) { - LLViewerRegion* region = *iter; - for (U32 i = 0; i < LLViewerRegion::NUM_PARTITIONS; i++) + LLSpatialPartition* part = region->getSpatialPartition(i); + if (part) { - LLSpatialPartition* part = region->getSpatialPartition(i); - if (part) - { - part->mOctree->balance(); - } + part->mOctree->balance(); } + } - //balance the VO Cache tree - LLVOCachePartition* vo_part = region->getVOCachePartition(); - if(vo_part) - { - vo_part->mOctree->balance(); - } + //balance the VO Cache tree + LLVOCachePartition* vo_part = region->getVOCachePartition(); + if(vo_part) + { + vo_part->mOctree->balance(); } } } @@ -2752,14 +2713,10 @@ bool LLPipeline::updateDrawableGeom(LLDrawable* drawablep, bool priority) return update_complete; } -static LLTrace::BlockTimerStatHandle FTM_SEED_VBO_POOLS("Seed VBO Pool"); - -static LLTrace::BlockTimerStatHandle FTM_UPDATE_GL("Update GL"); - void LLPipeline::updateGL() { + LL_PROFILE_ZONE_SCOPED; { - LL_RECORD_BLOCK_TIME(FTM_UPDATE_GL); while (!LLGLUpdate::sGLQ.empty()) { LLGLUpdate* glu = LLGLUpdate::sGLQ.front(); @@ -2770,15 +2727,13 @@ void LLPipeline::updateGL() } { //seed VBO Pools - LL_RECORD_BLOCK_TIME(FTM_SEED_VBO_POOLS); LLVertexBuffer::seedPools(); } } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_PRIORITY_GROUPS("Rebuild Priority Groups"); - void LLPipeline::clearRebuildGroups() { + LL_PROFILE_ZONE_SCOPED; LLSpatialGroup::sg_vector_t hudGroups; mGroupQ1Locked = true; @@ -2883,7 +2838,7 @@ void LLPipeline::clearRebuildDrawables() void LLPipeline::rebuildPriorityGroups() { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_PRIORITY_GROUPS); + LL_PROFILE_ZONE_SCOPED; LLTimer update_timer; assertInitialized(); @@ -2905,8 +2860,6 @@ void LLPipeline::rebuildPriorityGroups() } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_GROUPS("Rebuild Groups"); - void LLPipeline::rebuildGroups() { if (mGroupQ2.empty()) @@ -2914,7 +2867,7 @@ void LLPipeline::rebuildGroups() return; } - LL_RECORD_BLOCK_TIME(FTM_REBUILD_GROUPS); + LL_PROFILE_ZONE_SCOPED; mGroupQ2Locked = true; // Iterate through some drawables on the non-priority build queue S32 size = (S32) mGroupQ2.size(); @@ -3160,12 +3113,9 @@ void LLPipeline::markShift(LLDrawable *drawablep) } } -static LLTrace::BlockTimerStatHandle FTM_SHIFT_DRAWABLE("Shift Drawable"); -static LLTrace::BlockTimerStatHandle FTM_SHIFT_OCTREE("Shift Octree"); -static LLTrace::BlockTimerStatHandle FTM_SHIFT_HUD("Shift HUD"); - void LLPipeline::shiftObjects(const LLVector3 &offset) { + LL_PROFILE_ZONE_SCOPED; assertInitialized(); glClear(GL_DEPTH_BUFFER_BIT); @@ -3174,46 +3124,36 @@ void LLPipeline::shiftObjects(const LLVector3 &offset) LLVector4a offseta; offseta.load3(offset.mV); + for (LLDrawable::drawable_vector_t::iterator iter = mShiftList.begin(); + iter != mShiftList.end(); iter++) { - LL_RECORD_BLOCK_TIME(FTM_SHIFT_DRAWABLE); - - for (LLDrawable::drawable_vector_t::iterator iter = mShiftList.begin(); - iter != mShiftList.end(); iter++) + LLDrawable *drawablep = *iter; + if (drawablep->isDead()) { - LLDrawable *drawablep = *iter; - if (drawablep->isDead()) - { - continue; - } - drawablep->shiftPos(offseta); - drawablep->clearState(LLDrawable::ON_SHIFT_LIST); - } - mShiftList.resize(0); + continue; + } + drawablep->shiftPos(offseta); + drawablep->clearState(LLDrawable::ON_SHIFT_LIST); } - + mShiftList.resize(0); + for (LLWorld::region_list_t::const_iterator iter = LLWorld::getInstance()->getRegionList().begin(); + iter != LLWorld::getInstance()->getRegionList().end(); ++iter) { - LL_RECORD_BLOCK_TIME(FTM_SHIFT_OCTREE); - for (LLWorld::region_list_t::const_iterator iter = LLWorld::getInstance()->getRegionList().begin(); - iter != LLWorld::getInstance()->getRegionList().end(); ++iter) + LLViewerRegion* region = *iter; + for (U32 i = 0; i < LLViewerRegion::NUM_PARTITIONS; i++) { - LLViewerRegion* region = *iter; - for (U32 i = 0; i < LLViewerRegion::NUM_PARTITIONS; i++) + LLSpatialPartition* part = region->getSpatialPartition(i); + if (part) { - LLSpatialPartition* part = region->getSpatialPartition(i); - if (part) - { - part->shift(offseta); - } + part->shift(offseta); } } } - { - LL_RECORD_BLOCK_TIME(FTM_SHIFT_HUD); - LLHUDText::shiftAll(offset); - LLHUDNameTag::shiftAll(offset); - } + LLHUDText::shiftAll(offset); + LLHUDNameTag::shiftAll(offset); + display_update_camera(); } @@ -3244,10 +3184,9 @@ void LLPipeline::markPartitionMove(LLDrawable* drawable) } } -static LLTrace::BlockTimerStatHandle FTM_PROCESS_PARTITIONQ("PartitionQ"); void LLPipeline::processPartitionQ() { - LL_RECORD_BLOCK_TIME(FTM_PROCESS_PARTITIONQ); + LL_PROFILE_ZONE_SCOPED; for (LLDrawable::drawable_list_t::iterator iter = mPartitionQ.begin(); iter != mPartitionQ.end(); ++iter) { LLDrawable* drawable = *iter; @@ -3347,8 +3286,6 @@ void LLPipeline::markRebuild(LLDrawable *drawablep, LLDrawable::EDrawableFlags f } } -static LLTrace::BlockTimerStatHandle FTM_RESET_DRAWORDER("Reset Draw Order"); - void LLPipeline::stateSort(LLCamera& camera, LLCullResult &result) { if (hasAnyRenderType(LLPipeline::RENDER_TYPE_AVATAR, @@ -3362,7 +3299,6 @@ void LLPipeline::stateSort(LLCamera& camera, LLCullResult &result) LLPipeline::END_RENDER_TYPES)) { //clear faces from face pools - LL_RECORD_BLOCK_TIME(FTM_RESET_DRAWORDER); gPipeline.resetDrawOrders(); } @@ -3811,7 +3747,7 @@ void LLPipeline::touchTextures(LLDrawInfo* info) void LLPipeline::postSort(LLCamera& camera) { - LL_RECORD_BLOCK_TIME(FTM_STATESORT_POSTSORT); + LL_PROFILE_ZONE_SCOPED; assertInitialized(); @@ -4837,6 +4773,7 @@ void LLPipeline::renderGeomShadow(LLCamera& camera) void LLPipeline::addTrianglesDrawn(S32 index_count, U32 render_type) { + LL_PROFILE_ZONE_SCOPED; assertInitialized(); S32 count = 0; if (render_type == LLRender::TRIANGLE_STRIP) @@ -5617,11 +5554,9 @@ void LLPipeline::renderDebug() } } -static LLTrace::BlockTimerStatHandle FTM_REBUILD_POOLS("Rebuild Pools"); - void LLPipeline::rebuildPools() { - LL_RECORD_BLOCK_TIME(FTM_REBUILD_POOLS); + LL_PROFILE_ZONE_SCOPED; assertInitialized(); @@ -5965,6 +5900,7 @@ void LLPipeline::removeFromQuickLookup( LLDrawPool* poolp ) void LLPipeline::resetDrawOrders() { + LL_PROFILE_ZONE_SCOPED; assertInitialized(); // Iterate through all of the draw pools and rebuild them. for (pool_set_t::iterator iter = mPools.begin(); iter != mPools.end(); ++iter) @@ -7380,8 +7316,6 @@ void LLPipeline::resetVertexBuffers() mResetVertexBuffers = true; } -static LLTrace::BlockTimerStatHandle FTM_RESET_VB("Reset VB"); - void LLPipeline::doResetVertexBuffers(bool forced) { if (!mResetVertexBuffers) @@ -7403,7 +7337,7 @@ void LLPipeline::doResetVertexBuffers(bool forced) } } - LL_RECORD_BLOCK_TIME(FTM_RESET_VB); + LL_PROFILE_ZONE_SCOPED; mResetVertexBuffers = false; mCubeVB = NULL; @@ -7603,11 +7537,8 @@ void LLPipeline::renderFinalize() if (sRenderGlow) { - { - LL_RECORD_BLOCK_TIME(FTM_RENDER_BLOOM_FBO); - mGlow[2].bindTarget(); - mGlow[2].clear(); - } + mGlow[2].bindTarget(); + mGlow[2].clear(); gGlowExtractProgram.bind(); F32 minLum = llmax((F32) RenderGlowMinLuminance, 0.0f); @@ -7673,11 +7604,8 @@ void LLPipeline::renderFinalize() for (S32 i = 0; i < kernel; i++) { - { - LL_RECORD_BLOCK_TIME(FTM_RENDER_BLOOM_FBO); - mGlow[i % 2].bindTarget(); - mGlow[i % 2].clear(); - } + mGlow[i % 2].bindTarget(); + mGlow[i % 2].clear(); if (i == 0) { @@ -8194,11 +8122,9 @@ void LLPipeline::renderFinalize() LLGLState::checkTextureChannels(); } -static LLTrace::BlockTimerStatHandle FTM_BIND_DEFERRED("Bind Deferred"); - void LLPipeline::bindDeferredShader(LLGLSLShader& shader, LLRenderTarget* light_target) { - LL_RECORD_BLOCK_TIME(FTM_BIND_DEFERRED); + LL_PROFILE_ZONE_SCOPED; LLRenderTarget* deferred_target = &mDeferredScreen; LLRenderTarget* deferred_depth_target = &mDeferredDepth; @@ -8458,16 +8384,7 @@ LLVector4 pow4fsrgb(LLVector4 v, F32 f) return v; } -static LLTrace::BlockTimerStatHandle FTM_GI_TRACE("Trace"); -static LLTrace::BlockTimerStatHandle FTM_GI_GATHER("Gather"); -static LLTrace::BlockTimerStatHandle FTM_SUN_SHADOW("Shadow Map"); -static LLTrace::BlockTimerStatHandle FTM_SOFTEN_SHADOW("Shadow Soften"); -static LLTrace::BlockTimerStatHandle FTM_EDGE_DETECTION("Find Edges"); -static LLTrace::BlockTimerStatHandle FTM_LOCAL_LIGHTS("Local Lights"); -static LLTrace::BlockTimerStatHandle FTM_ATMOSPHERICS("Atmospherics"); -static LLTrace::BlockTimerStatHandle FTM_FULLSCREEN_LIGHTS("Fullscreen Lights"); -static LLTrace::BlockTimerStatHandle FTM_PROJECTORS("Projectors"); -static LLTrace::BlockTimerStatHandle FTM_POST("Post"); +static LLTrace::BlockTimerStatHandle FTM_DEFERRED_LIGHTING("Deferred Lighting"); void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) { @@ -8475,6 +8392,7 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) { return; } + LL_RECORD_BLOCK_TIME(FTM_DEFERRED_LIGHTING); LLRenderTarget *deferred_target = &mDeferredScreen; LLRenderTarget *deferred_depth_target = &mDeferredDepth; @@ -8547,7 +8465,7 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) { deferred_light_target->bindTarget(); { // paint shadow/SSAO light map (direct lighting lightmap) - LL_RECORD_BLOCK_TIME(FTM_SUN_SHADOW); + LL_PROFILE_ZONE_NAMED("renderDeferredLighting - sun shadow"); bindDeferredShader(gDeferredSunProgram, deferred_light_target); mDeferredVB->setBuffer(LLVertexBuffer::MAP_VERTEX); glClearColor(1, 1, 1, 1); @@ -8593,7 +8511,7 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) if (RenderDeferredSSAO) { // soften direct lighting lightmap - LL_RECORD_BLOCK_TIME(FTM_SOFTEN_SHADOW); + LL_PROFILE_ZONE_NAMED("renderDeferredLighting - soften shadow"); // blur lightmap screen_target->bindTarget(); glClearColor(1, 1, 1, 1); @@ -8671,7 +8589,7 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) { // apply sunlight contribution LLGLSLShader &soften_shader = LLPipeline::sUnderWaterRender ? gDeferredSoftenWaterProgram : gDeferredSoftenProgram; - LL_RECORD_BLOCK_TIME(FTM_ATMOSPHERICS); + LL_PROFILE_ZONE_NAMED("renderDeferredLighting - atmospherics"); bindDeferredShader(soften_shader); LLEnvironment &environment = LLEnvironment::instance(); @@ -8737,6 +8655,7 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) LLVertexBuffer::unbind(); { + LL_PROFILE_ZONE_NAMED("renderDeferredLighting - local lights"); bindDeferredShader(gDeferredLightProgram); if (mCubeVB.isNull()) @@ -8807,7 +8726,6 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) continue; } - LL_RECORD_BLOCK_TIME(FTM_LOCAL_LIGHTS); gDeferredLightProgram.uniform3fv(LLShaderMgr::LIGHT_CENTER, 1, c); gDeferredLightProgram.uniform1f(LLShaderMgr::LIGHT_SIZE, s); gDeferredLightProgram.uniform3fv(LLShaderMgr::DIFFUSE_COLOR, 1, col.mV); @@ -8843,6 +8761,7 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) if (!spot_lights.empty()) { + LL_PROFILE_ZONE_NAMED("renderDeferredLighting - projectors"); LLGLDepthTest depth(GL_TRUE, GL_FALSE); bindDeferredShader(gDeferredSpotLightProgram); @@ -8852,7 +8771,6 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) for (LLDrawable::drawable_list_t::iterator iter = spot_lights.begin(); iter != spot_lights.end(); ++iter) { - LL_RECORD_BLOCK_TIME(FTM_PROJECTORS); LLDrawable *drawablep = *iter; LLVOVolume *volume = drawablep->getVOVolume(); @@ -8888,6 +8806,7 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) vert[2].set(3, 1, 0); { + LL_PROFILE_ZONE_NAMED("renderDeferredLighting - fullscreen lights"); LLGLDepthTest depth(GL_FALSE); // full screen blit @@ -8907,7 +8826,6 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) while (!fullscreen_lights.empty()) { - LL_RECORD_BLOCK_TIME(FTM_FULLSCREEN_LIGHTS); light[count] = fullscreen_lights.front(); fullscreen_lights.pop_front(); col[count] = light_colors.front(); @@ -8939,7 +8857,6 @@ void LLPipeline::renderDeferredLighting(LLRenderTarget *screen_target) for (LLDrawable::drawable_list_t::iterator iter = fullscreen_spot_lights.begin(); iter != fullscreen_spot_lights.end(); ++iter) { - LL_RECORD_BLOCK_TIME(FTM_PROJECTORS); LLDrawable *drawablep = *iter; LLVOVolume *volume = drawablep->getVOVolume(); LLVector3 center = drawablep->getPositionAgent(); @@ -9849,10 +9766,9 @@ void LLPipeline::renderShadow(glh::matrix4f& view, glh::matrix4f& proj, LLCamera LLPipeline::sShadowRender = false; } -static LLTrace::BlockTimerStatHandle FTM_VISIBLE_CLOUD("Visible Cloud"); bool LLPipeline::getVisiblePointCloud(LLCamera& camera, LLVector3& min, LLVector3& max, std::vector& fp, LLVector3 light_dir) { - LL_RECORD_BLOCK_TIME(FTM_VISIBLE_CLOUD); + LL_PROFILE_ZONE_SCOPED; //get point cloud of intersection of frust and min, max if (getVisibleExtents(camera, min, max)) @@ -10109,9 +10025,6 @@ LLRenderTarget* LLPipeline::getShadowTarget(U32 i) } static LLTrace::BlockTimerStatHandle FTM_GEN_SUN_SHADOW("Gen Sun Shadow"); -static LLTrace::BlockTimerStatHandle FTM_GEN_SUN_SHADOW_SETUP("Sun Shadow Setup"); -static LLTrace::BlockTimerStatHandle FTM_GEN_SUN_SHADOW_RENDER_DIRECTIONAL("Render Dir"); -static LLTrace::BlockTimerStatHandle FTM_GEN_SUN_SHADOW_SPOT_SETUP("Spot Shadow Setup"); static LLTrace::BlockTimerStatHandle FTM_GEN_SUN_SHADOW_SPOT_RENDER("Spot Shadow Render"); void LLPipeline::generateSunShadow(LLCamera& camera) @@ -10907,14 +10820,11 @@ void LLPipeline::renderGroups(LLRenderPass* pass, U32 type, U32 mask, bool textu } } -static LLTrace::BlockTimerStatHandle FTM_IMPOSTOR_MARK_VISIBLE("Impostor Mark Visible"); -static LLTrace::BlockTimerStatHandle FTM_IMPOSTOR_SETUP("Impostor Setup"); -static LLTrace::BlockTimerStatHandle FTM_IMPOSTOR_BACKGROUND("Impostor Background"); -static LLTrace::BlockTimerStatHandle FTM_IMPOSTOR_ALLOCATE("Impostor Allocate"); -static LLTrace::BlockTimerStatHandle FTM_IMPOSTOR_RESIZE("Impostor Resize"); +static LLTrace::BlockTimerStatHandle FTM_GENERATE_IMPOSTOR("Generate Impostor"); void LLPipeline::generateImpostor(LLVOAvatar* avatar) { + LL_RECORD_BLOCK_TIME(FTM_GENERATE_IMPOSTOR); LLGLState::checkStates(); LLGLState::checkTextureChannels(); LLGLState::checkClientArrays(); @@ -10989,7 +10899,6 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) LLViewerCamera* viewer_camera = LLViewerCamera::getInstance(); { - LL_RECORD_BLOCK_TIME(FTM_IMPOSTOR_MARK_VISIBLE); markVisible(avatar->mDrawable, *viewer_camera); LLVOAvatar::attachment_map_t::iterator iter; @@ -11018,7 +10927,6 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) U32 resX = 0; { - LL_RECORD_BLOCK_TIME(FTM_IMPOSTOR_SETUP); const LLVector4a* ext = avatar->mDrawable->getSpatialExtents(); LLVector3 pos(avatar->getRenderPosition()+avatar->getImpostorOffset()); @@ -11075,9 +10983,6 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) if (!avatar->mImpostor.isComplete()) { - LL_RECORD_BLOCK_TIME(FTM_IMPOSTOR_ALLOCATE); - - if (LLPipeline::sRenderDeferred) { avatar->mImpostor.allocate(resX,resY,GL_SRGB8_ALPHA8,TRUE,FALSE); @@ -11094,7 +10999,6 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) } else if(resX != avatar->mImpostor.getWidth() || resY != avatar->mImpostor.getHeight()) { - LL_RECORD_BLOCK_TIME(FTM_IMPOSTOR_RESIZE); avatar->mImpostor.resize(resX,resY); } @@ -11156,7 +11060,6 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) LLDrawPoolAvatar::sMinimumAlpha = old_alpha; { //create alpha mask based on depth buffer (grey out if muted) - LL_RECORD_BLOCK_TIME(FTM_IMPOSTOR_BACKGROUND); if (LLPipeline::sRenderDeferred) { GLuint buff = GL_COLOR_ATTACHMENT0; diff --git a/indra/newview/pipeline.h b/indra/newview/pipeline.h index 8ffbddca21..b87a726647 100644 --- a/indra/newview/pipeline.h +++ b/indra/newview/pipeline.h @@ -66,7 +66,6 @@ bool setup_hud_matrices(const LLRect& screen_region); // specify portion of scre extern LLTrace::BlockTimerStatHandle FTM_RENDER_GEOMETRY; extern LLTrace::BlockTimerStatHandle FTM_RENDER_GRASS; extern LLTrace::BlockTimerStatHandle FTM_RENDER_INVISIBLE; -extern LLTrace::BlockTimerStatHandle FTM_RENDER_OCCLUSION; extern LLTrace::BlockTimerStatHandle FTM_RENDER_SHINY; extern LLTrace::BlockTimerStatHandle FTM_RENDER_SIMPLE; extern LLTrace::BlockTimerStatHandle FTM_RENDER_TERRAIN; @@ -87,8 +86,6 @@ extern LLTrace::BlockTimerStatHandle FTM_CLIENT_COPY; extern LLTrace::BlockTimerStatHandle FTM_RENDER_UI_HUD; extern LLTrace::BlockTimerStatHandle FTM_RENDER_UI_3D; extern LLTrace::BlockTimerStatHandle FTM_RENDER_UI_2D; -extern LLTrace::BlockTimerStatHandle FTM_RENDER_UI_DEBUG_TEXT; -extern LLTrace::BlockTimerStatHandle FTM_RENDER_UI_SCENE_MON; class LLPipeline { -- cgit v1.3 From 3faba7515c757ca3183522bd017c0f76d9c4581c Mon Sep 17 00:00:00 2001 From: Mnikolenko Productengine Date: Mon, 1 Nov 2021 19:38:55 +0200 Subject: SL-16237 FIXED Viewer hangs on login --- indra/llcommon/workqueue.h | 6 ++++++ indra/llrender/llimagegl.cpp | 17 ++++++++++++++++- indra/llrender/llimagegl.h | 2 ++ 3 files changed, 24 insertions(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index b88aef989a..5ec790da79 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -94,6 +94,12 @@ namespace LL void postEvery(const std::chrono::duration& interval, CALLABLE&& callable); + template + bool tryPost(CALLABLE&& callable) + { + return mQueue.tryPush(TimedWork(TimePoint::clock::now(), std::move(callable))); + } + /*------------------------- handshake API --------------------------*/ /** diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 09b1c71f02..cbc5392882 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -2288,7 +2288,10 @@ bool LLImageGLThread::postCallback(const std::function& callback) { try { - mCallbackQueue.post(callback); + if (!mCallbackQueue.tryPost(callback)) + { + mPendingCallbackQ.push(callback); + } } catch (LLThreadSafeQueueInterrupt e) { @@ -2304,6 +2307,18 @@ void LLImageGLThread::executeCallbacks() LL_PROFILE_ZONE_SCOPED; //executed from main thread mCallbackQueue.runPending(); + + while (!mPendingCallbackQ.empty()) + { + if (mCallbackQueue.tryPost(mPendingCallbackQ.front())) + { + mPendingCallbackQ.pop(); + } + else + { + break; + } + } } void LLImageGLThread::run() diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h index 95b60bd0bd..8264e4a5f2 100644 --- a/indra/llrender/llimagegl.h +++ b/indra/llrender/llimagegl.h @@ -332,6 +332,8 @@ public: void* mContext; LLAtomicBool mFinished; + std::queue> mPendingCallbackQ; + static LLImageGLThread* sInstance; }; -- cgit v1.3 From 10692ab4a4f999e1ee302675e4ffb84f37a52643 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 2 Nov 2021 10:35:34 -0400 Subject: SL-16207: Create uniform overload sets for wide-string conversions. Use new ll_convert_forms() macro in llstring.h to declare, for each wide-string conversion function of interest, four overloads. The real one, the nontrivial one, is (const char*, size_t len), implemented in llstring.cpp. Then (const string&, size_t len), (const char*) and (const string&) are each trivially implemented with an inline call to (const char*, size_t len). Notably, we change all S32 len parameters to size_t. Using S32 is old skool. Tweak each nontrivial implementation in llstring.cpp to accept (const char*, size_t len) instead of (const string&) with or without explicit length. Eliminate from llstring.cpp trivial overloads (deriving length from either a const char* or from a string), since those are now inline in the header. Of course three of those overloads will be unified once we enable C++17 and change each relevant parameter to std::string_view, but we're not yet there. Meanwhile, this suite of overloads minimizes, to the best of our ability, new string allocations solely for parameter passing. And use of a macro means we need only change the macro once we get std::string_view. We take this step because some use cases require (const char*), some require (const string&, size_t len), others (const char*, size_t len) ... We were missing some key overloads, and had to work around them by instantiating new string objects (necessitating both allocation and character copying) just to pass the desired parameter. Using the macro ensures this consistent set of overloads for every wide-string conversion function. Additionally, knowing that the ugly-name overloads exist, ll_convert_forms() implicitly defines corresponding ll_convert() overloads. Streamline declarations of utf16str_to_wstring(), wstring_to_utf16str(), utf8str_to_utf16str(), utf16str_to_utf8str(), utf8str_to_wstring(), wstring_to_utf8str(), ll_convert_wide_to_wstring() and ll_convert_wstring_to_wide() using ll_convert_forms(). Use corresponding new ll_convert_cp_forms() macro to declare consistent overloads for conversion functions accepting an optional unsigned int code_page parameter. We used to delegate to the .cpp file the implementation of each overload accepting code_page so llstring.h need not include the Windows header defining the CP_UTF8 default; this is more simply accomplished by introducing a small ll_wstring_default_code_page() function to retrieve it from the .cpp file. That lets us specify the code_page parameter as optional, using that function as its default value. Use ll_convert_cp_forms() to streamline declarations of ll_convert_wide_to_string() and ll_convert_string_to_wide(). Introduce real implementations of ll_convert_wide_to_wstring() and ll_convert_wstring_to_wide(). The previous implementations merely copied individual characters, which is wrong: when we convert UTF16LE to UTF32, we can and should fold multi-character UTF16LE encodings to the corresponding single UTF32 character. The real implemenations leverage our awareness that both llutf16string and Windows std::wstring (either variant) use UTF16LE encoding, so we can reuse the corresponding llutf16string conversions. Introduce generic ll_convert_length() function, specialized as either std::strlen() or std::wcslen() depending on parameter type. (Even if std::wcslen() is derived from classic C, why doesn't the C++ standard library define a std::strlen(const wchar_t*) overload to call it?) Fix ll_convert_alias()'s ll_convert_impl specialization's operator() to accept boost::call_traits::param_type, so we can pass (e.g.) const std::wstring& but also const wchar_t* instead of const wchar_t*&. --- indra/llcommon/llstring.cpp | 95 +++++++++-------------------- indra/llcommon/llstring.h | 143 ++++++++++++++++++++++++++------------------ 2 files changed, 113 insertions(+), 125 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llstring.cpp b/indra/llcommon/llstring.cpp index 0290eea143..5f426e5dca 100644 --- a/indra/llcommon/llstring.cpp +++ b/indra/llcommon/llstring.cpp @@ -215,7 +215,7 @@ S32 utf16chars_to_wchar(const U16* inchars, llwchar* outchar) return inchars - base; } -llutf16string wstring_to_utf16str(const LLWString &utf32str, S32 len) +llutf16string wstring_to_utf16str(const llwchar* utf32str, size_t len) { llutf16string out; @@ -237,27 +237,19 @@ llutf16string wstring_to_utf16str(const LLWString &utf32str, S32 len) return out; } -llutf16string wstring_to_utf16str(const LLWString &utf32str) +llutf16string utf8str_to_utf16str( const char* utf8str, size_t len ) { - const S32 len = (S32)utf32str.length(); - return wstring_to_utf16str(utf32str, len); -} - -llutf16string utf8str_to_utf16str ( const std::string& utf8str ) -{ - LLWString wstr = utf8str_to_wstring ( utf8str ); + LLWString wstr = utf8str_to_wstring ( utf8str, len ); return wstring_to_utf16str ( wstr ); } - -LLWString utf16str_to_wstring(const llutf16string &utf16str, S32 len) +LLWString utf16str_to_wstring(const U16* utf16str, size_t len) { LLWString wout; - if((len <= 0) || utf16str.empty()) return wout; + if (len == 0) return wout; S32 i = 0; - // craziness to make gcc happy (llutf16string.c_str() is tweaked on linux): - const U16* chars16 = &(*(utf16str.begin())); + const U16* chars16 = utf16str; while (i < len) { llwchar cur_char; @@ -267,12 +259,6 @@ LLWString utf16str_to_wstring(const llutf16string &utf16str, S32 len) return wout; } -LLWString utf16str_to_wstring(const llutf16string &utf16str) -{ - const S32 len = (S32)utf16str.length(); - return utf16str_to_wstring(utf16str, len); -} - // Length in llwchar (UTF-32) of the first len units (16 bits) of the given UTF-16 string. S32 utf16str_wstring_length(const llutf16string &utf16str, const S32 utf16_len) { @@ -392,8 +378,7 @@ S32 wstring_utf8_length(const LLWString& wstr) return len; } - -LLWString utf8str_to_wstring(const std::string& utf8str, S32 len) +LLWString utf8str_to_wstring(const char* utf8str, size_t len) { LLWString wout; @@ -481,13 +466,7 @@ LLWString utf8str_to_wstring(const std::string& utf8str, S32 len) return wout; } -LLWString utf8str_to_wstring(const std::string& utf8str) -{ - const S32 len = (S32)utf8str.length(); - return utf8str_to_wstring(utf8str, len); -} - -std::string wstring_to_utf8str(const LLWString& utf32str, S32 len) +std::string wstring_to_utf8str(const llwchar* utf32str, size_t len) { std::string out; @@ -503,20 +482,9 @@ std::string wstring_to_utf8str(const LLWString& utf32str, S32 len) return out; } -std::string wstring_to_utf8str(const LLWString& utf32str) -{ - const S32 len = (S32)utf32str.length(); - return wstring_to_utf8str(utf32str, len); -} - -std::string utf16str_to_utf8str(const llutf16string& utf16str) -{ - return wstring_to_utf8str(utf16str_to_wstring(utf16str)); -} - -std::string utf16str_to_utf8str(const llutf16string& utf16str, S32 len) +std::string utf16str_to_utf8str(const U16* utf16str, size_t len) { - return wstring_to_utf8str(utf16str_to_wstring(utf16str, len), len); + return wstring_to_utf8str(utf16str_to_wstring(utf16str, len)); } std::string utf8str_trim(const std::string& utf8str) @@ -657,17 +625,16 @@ std::string utf8str_removeCRLF(const std::string& utf8str) } #if LL_WINDOWS -std::string ll_convert_wide_to_string(const wchar_t* in) +unsigned int ll_wstring_default_code_page() { - return ll_convert_wide_to_string(in, CP_UTF8); + return CP_UTF8; } -std::string ll_convert_wide_to_string(const wchar_t* in, unsigned int code_page) +std::string ll_convert_wide_to_string(const wchar_t* in, size_t len_in, unsigned int code_page) { std::string out; if(in) { - int len_in = wcslen(in); int len_out = WideCharToMultiByte( code_page, 0, @@ -699,12 +666,7 @@ std::string ll_convert_wide_to_string(const wchar_t* in, unsigned int code_page) return out; } -std::wstring ll_convert_string_to_wide(const std::string& in) -{ - return ll_convert_string_to_wide(in, CP_UTF8); -} - -std::wstring ll_convert_string_to_wide(const std::string& in, unsigned int code_page) +std::wstring ll_convert_string_to_wide(const char* in, size_t len, unsigned int code_page) { // From review: // We can preallocate a wide char buffer that is the same length (in wchar_t elements) as the utf8 input, @@ -716,10 +678,10 @@ std::wstring ll_convert_string_to_wide(const std::string& in, unsigned int code_ // reserve an output buffer that will be destroyed on exit, with a place // to put NULL terminator - std::vector w_out(in.length() + 1); + std::vector w_out(len + 1); memset(&w_out[0], 0, w_out.size()); - int real_output_str_len = MultiByteToWideChar(code_page, 0, in.c_str(), in.length(), + int real_output_str_len = MultiByteToWideChar(code_page, 0, in, len, &w_out[0], w_out.size() - 1); //looks like MultiByteToWideChar didn't add null terminator to converted string, see EXT-4858. @@ -729,22 +691,23 @@ std::wstring ll_convert_string_to_wide(const std::string& in, unsigned int code_ return {&w_out[0]}; } -LLWString ll_convert_wide_to_wstring(const std::wstring& in) +LLWString ll_convert_wide_to_wstring(const wchar_t* in, size_t len) { - // This function, like its converse, is a placeholder, encapsulating a - // guilty little hack: the only "official" way nat has found to convert - // between std::wstring (16 bits on Windows) and LLWString (UTF-32) is - // by using iconv, which we've avoided so far. It kinda sorta works to - // just copy individual characters... - // The point is that if/when we DO introduce some more official way to - // perform such conversions, we should only have to call it here. - return { in.begin(), in.end() }; + // Whether or not std::wstring and llutf16string are distinct types, they + // both hold UTF-16LE characters. (See header file comments.) Pretend this + // wchar_t* sequence is really a U16* sequence and use the conversion we + // define above. + return utf16str_to_wstring(reinterpret_cast(in), len); } -std::wstring ll_convert_wstring_to_wide(const LLWString& in) +std::wstring ll_convert_wstring_to_wide(const llwchar* in, size_t len) { - // See comments in ll_convert_wide_to_wstring() - return { in.begin(), in.end() }; + // first, convert to llutf16string, for which we have a real implementation + auto utf16str{ wstring_to_utf16str(in, len) }; + // then, because each U16 char must be UTF-16LE encoded, pretend the U16* + // string pointer is a wchar_t* and instantiate a std::wstring of the same + // length. + return { reinterpret_cast(utf16str.c_str()), utf16str.length() }; } std::string ll_convert_string_to_utf8_string(const std::string& in) diff --git a/indra/llcommon/llstring.h b/indra/llcommon/llstring.h index 89e95ef40a..a0598e8a11 100644 --- a/indra/llcommon/llstring.h +++ b/indra/llcommon/llstring.h @@ -27,9 +27,11 @@ #ifndef LL_LLSTRING_H #define LL_LLSTRING_H +#include #include #include #include +#include // std::wcslen() //#include #include #include @@ -532,14 +534,59 @@ struct ll_convert_impl template<> \ struct ll_convert_impl \ { \ - TO operator()(const FROM& in) const { return EXPR; } \ + /* param_type optimally passes both char* and string */ \ + TO operator()(typename boost::call_traits::param_type in) const { return EXPR; } \ } -// If all we're doing is copying characters, pass this as EXPR. Since it -// expands into the 'return EXPR' slot in the ll_convert_impl specialization -// above, it implies TO{ in.begin(), in.end() }. +// If all we're doing is copying characters, pass this to ll_convert_alias as +// EXPR. Since it expands into the 'return EXPR' slot in the ll_convert_impl +// specialization above, it implies TO{ in.begin(), in.end() }. #define LL_CONVERT_COPY_CHARS { in.begin(), in.end() } +// Generic name for strlen() / wcslen() - the default implementation should +// (!) work with U16 and llwchar, but we don't intend to engage it. +template +size_t ll_convert_length(const CHARTYPE* zstr) +{ + const CHARTYPE* zp; + // classic C string scan + for (zp = zstr; *zp; ++zp) + ; + return (zp - zstr); +} + +// specialize where we have a library function; may use intrinsic operations +template <> +inline size_t ll_convert_length(const wchar_t* zstr) { return std::wcslen(zstr); } +template <> +inline size_t ll_convert_length (const char* zstr) { return std::strlen(zstr); } + +// ll_convert_forms() is short for a bunch of boilerplate. It defines +// longname(const char*, len), longname(const char*), longname(const string&) +// and longname(const string&, len) so calls written pre-ll_convert() will +// work. Most of these overloads will be unified once we turn on C++17 and can +// use std::string_view. +// It also uses aliasmacro to ensure that both ll_convert(const char*) +// and ll_convert(const string&) will work. +#define ll_convert_forms(aliasmacro, OUTSTR, INSTR, longname) \ +LL_COMMON_API OUTSTR longname(const INSTR::value_type* in, size_t len); \ +inline auto longname(const INSTR& in, size_t len) \ +{ \ + return longname(in.c_str(), len); \ +} \ +inline auto longname(const INSTR::value_type* in) \ +{ \ + return longname(in, ll_convert_length(in)); \ +} \ +inline auto longname(const INSTR& in) \ +{ \ + return longname(in.c_str(), in.length()); \ +} \ +/* string param */ \ +aliasmacro(OUTSTR, INSTR, longname(in)); \ +/* char* param */ \ +aliasmacro(OUTSTR, const INSTR::value_type*, longname(in)) + // Make the incoming string a utf8 string. Replaces any unknown glyph // with the UNKNOWN_CHARACTER. Once any unknown glyph is found, the rest // of the data may not be recovered. @@ -602,34 +649,18 @@ typedef std::basic_string llutf16string; #define ll_convert_wstr_alias(TO, FROM, EXPR) ll_convert_alias(TO, FROM, EXPR) #endif -LL_COMMON_API LLWString utf16str_to_wstring(const llutf16string &utf16str, S32 len); -LL_COMMON_API LLWString utf16str_to_wstring(const llutf16string &utf16str); -ll_convert_u16_alias(LLWString, llutf16string, utf16str_to_wstring(in)); - -LL_COMMON_API llutf16string wstring_to_utf16str(const LLWString &utf32str, S32 len); -LL_COMMON_API llutf16string wstring_to_utf16str(const LLWString &utf32str); -ll_convert_u16_alias(llutf16string, LLWString, wstring_to_utf16str(in)); +ll_convert_forms(ll_convert_u16_alias, LLWString, llutf16string, utf16str_to_wstring); +ll_convert_forms(ll_convert_u16_alias, llutf16string, LLWString, wstring_to_utf16str); +ll_convert_forms(ll_convert_u16_alias, llutf16string, std::string, utf8str_to_utf16str); +ll_convert_forms(ll_convert_alias, LLWString, std::string, utf8str_to_wstring); -LL_COMMON_API llutf16string utf8str_to_utf16str ( const std::string& utf8str, S32 len); -LL_COMMON_API llutf16string utf8str_to_utf16str ( const std::string& utf8str ); -ll_convert_u16_alias(llutf16string, std::string, utf8str_to_utf16str(in)); - -LL_COMMON_API LLWString utf8str_to_wstring(const std::string &utf8str, S32 len); -LL_COMMON_API LLWString utf8str_to_wstring(const std::string &utf8str); // Same function, better name. JC inline LLWString utf8string_to_wstring(const std::string& utf8_string) { return utf8str_to_wstring(utf8_string); } -// best name of all -ll_convert_alias(LLWString, std::string, utf8string_to_wstring(in)); -// LL_COMMON_API S32 wchar_to_utf8chars(llwchar inchar, char* outchars); -LL_COMMON_API std::string wstring_to_utf8str(const LLWString &utf32str, S32 len); -LL_COMMON_API std::string wstring_to_utf8str(const LLWString &utf32str); -ll_convert_alias(std::string, LLWString, wstring_to_utf8str(in)); -LL_COMMON_API std::string utf16str_to_utf8str(const llutf16string &utf16str, S32 len); -LL_COMMON_API std::string utf16str_to_utf8str(const llutf16string &utf16str); -ll_convert_u16_alias(std::string, llutf16string, utf16str_to_utf8str(in)); +ll_convert_forms(ll_convert_alias, std::string, LLWString, wstring_to_utf8str); +ll_convert_forms(ll_convert_u16_alias, std::string, llutf16string, utf16str_to_utf8str); // an older alias for utf16str_to_utf8str(llutf16string) inline std::string wstring_to_utf8str(const llutf16string &utf16str) { return utf16str_to_utf8str(utf16str);} @@ -706,42 +737,36 @@ LL_COMMON_API std::string utf8str_removeCRLF(const std::string& utf8str); //@{ /** - * @brief Convert a wide string to std::string + * @brief Convert a wide string to/from std::string + * Convert a Windows wide string to/from our LLWString * * This replaces the unsafe W2A macro from ATL. */ -LL_COMMON_API std::string ll_convert_wide_to_string(const wchar_t* in, unsigned int code_page); -LL_COMMON_API std::string ll_convert_wide_to_string(const wchar_t* in); // default CP_UTF8 -inline std::string ll_convert_wide_to_string(const std::wstring& in, unsigned int code_page) -{ - return ll_convert_wide_to_string(in.c_str(), code_page); -} -inline std::string ll_convert_wide_to_string(const std::wstring& in) -{ - return ll_convert_wide_to_string(in.c_str()); -} -ll_convert_wstr_alias(std::string, std::wstring, ll_convert_wide_to_string(in)); - -/** - * Converts a string to wide string. - */ -LL_COMMON_API std::wstring ll_convert_string_to_wide(const std::string& in, - unsigned int code_page); -LL_COMMON_API std::wstring ll_convert_string_to_wide(const std::string& in); - // default CP_UTF8 -ll_convert_wstr_alias(std::wstring, std::string, ll_convert_string_to_wide(in)); - -/** - * Convert a Windows wide string to our LLWString - */ -LL_COMMON_API LLWString ll_convert_wide_to_wstring(const std::wstring& in); -ll_convert_wstr_alias(LLWString, std::wstring, ll_convert_wide_to_wstring(in)); - -/** - * Convert LLWString to Windows wide string - */ -LL_COMMON_API std::wstring ll_convert_wstring_to_wide(const LLWString& in); -ll_convert_wstr_alias(std::wstring, LLWString, ll_convert_wstring_to_wide(in)); +// Avoid requiring this header to #include the Windows header file declaring +// our actual default code_page by delegating this function to our .cpp file. +LL_COMMON_API unsigned int ll_wstring_default_code_page(); + +// This is like ll_convert_forms(), with the added complexity of a code page +// parameter that may or may not be passed. +#define ll_convert_cp_forms(aliasmacro, OUTSTR, INSTR, longname) \ +LL_COMMON_API OUTSTR longname( \ + const INSTR::value_type* in, \ + size_t len=ll_convert_length(in), \ + unsigned int code_page=ll_wstring_default_code_page()); \ +inline auto longname( \ + const INSTR& in, \ + size_t len=in.length(), \ + unsigned int code_page=ll_wstring_default_code_page()) \ +{ \ + return longname(in.c_str(), len, code_page); \ +} \ +aliasmacro(OUTSTR, INSTR, longname(in)); \ +aliasmacro(OUTSTR, const INSTR::value_type*, longname(in)) + +ll_convert_cp_forms(ll_convert_wstr_alias, std::string, std::wstring, ll_convert_wide_to_string); +ll_convert_cp_forms(ll_convert_wstr_alias, std::wstring, std::string, ll_convert_string_to_wide); + ll_convert_forms(ll_convert_wstr_alias, LLWString, std::wstring, ll_convert_wide_to_wstring); + ll_convert_forms(ll_convert_wstr_alias, std::wstring, LLWString, ll_convert_wstring_to_wide); /** * Converts incoming string into utf8 string -- cgit v1.3 From 95958bc8b2a5340bef93996f2ff0c04956bfb743 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 2 Nov 2021 11:31:47 -0400 Subject: SL-16207: Guess Microsoft compiler isn't smart about default params? clang allows us to specify, as a default function parameter, an expression involving a preceding parameter, e.g. (char* ptr, size_t len=strlen(ptr)). The Microsoft compiler produces errors, requiring more overloads to address that. Also #undef llstring.h's declaration helper macros at the bottom of the file. Once we've used them to declare stuff, they need not (should not) be visible to the consuming source file. --- indra/llcommon/llstring.h | 26 ++++++++++++++++++++++++-- 1 file changed, 24 insertions(+), 2 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llstring.h b/indra/llcommon/llstring.h index a0598e8a11..54e3f9ee63 100644 --- a/indra/llcommon/llstring.h +++ b/indra/llcommon/llstring.h @@ -749,17 +749,29 @@ LL_COMMON_API unsigned int ll_wstring_default_code_page(); // This is like ll_convert_forms(), with the added complexity of a code page // parameter that may or may not be passed. #define ll_convert_cp_forms(aliasmacro, OUTSTR, INSTR, longname) \ +/* declare the only nontrivial implementation (in .cpp file) */ \ LL_COMMON_API OUTSTR longname( \ const INSTR::value_type* in, \ - size_t len=ll_convert_length(in), \ + size_t len, \ unsigned int code_page=ll_wstring_default_code_page()); \ +/* if passed only a char pointer, scan for nul terminator */ \ +inline auto longname(const INSTR::value_type* in) \ +{ \ + return longname(in, ll_convert_length(in)); \ +} \ +/* if passed string and length, extract its char pointer */ \ inline auto longname( \ const INSTR& in, \ - size_t len=in.length(), \ + size_t len, \ unsigned int code_page=ll_wstring_default_code_page()) \ { \ return longname(in.c_str(), len, code_page); \ } \ +/* if passed only a string object, no scan, pass known length */ \ +inline auto longname(const INSTR& in) \ +{ \ + return longname(in.c_str(), in.length()); \ +} \ aliasmacro(OUTSTR, INSTR, longname(in)); \ aliasmacro(OUTSTR, const INSTR::value_type*, longname(in)) @@ -1967,4 +1979,14 @@ void LLStringUtilBase::truncate(string_type& string, size_type count) string.resize(count < cur_size ? count : cur_size); } +// The good thing about *declaration* macros, vs. usage macros, is that now +// we're done with them: we don't need them to bleed into the consuming source +// file. +#undef ll_convert_alias +#undef ll_convert_u16_alias +#undef ll_convert_wstr_alias +#undef LL_CONVERT_COPY_CHARS +#undef ll_convert_forms +#undef ll_convert_cp_forms + #endif // LL_STRING_H -- cgit v1.3 From a33718ee4ca4edbbc4c4034b29ec4c8d102f3a7e Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 2 Nov 2021 17:27:59 -0400 Subject: SL-16207: Fix bug in ll_convert_string_to_utf8_string(). That function wants to pass a code_page to ll_convert_string_to_wide(), but the code_page parameter was being mistaken for the length parameter, leading to access violations. --- indra/llcommon/llstring.cpp | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llstring.cpp b/indra/llcommon/llstring.cpp index 5f426e5dca..03f706f5a5 100644 --- a/indra/llcommon/llstring.cpp +++ b/indra/llcommon/llstring.cpp @@ -712,10 +712,11 @@ std::wstring ll_convert_wstring_to_wide(const llwchar* in, size_t len) std::string ll_convert_string_to_utf8_string(const std::string& in) { - auto w_mesg = ll_convert_string_to_wide(in, CP_ACP); - std::string out_utf8(ll_convert_wide_to_string(w_mesg.c_str(), CP_UTF8)); - - return out_utf8; + // If you pass code_page, you must also pass length, otherwise the code + // page parameter will be mistaken for length. + auto w_mesg = ll_convert_string_to_wide(in, in.length(), CP_ACP); + // CP_UTF8 is default -- see ll_wstring_default_code_page() above. + return ll_convert_wide_to_string(w_mesg); } namespace -- cgit v1.3 From 8458ad8890cf0a11804996210d7bcfbdaa3eec2e Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 4 Nov 2021 16:40:05 -0400 Subject: SL-16202: Instantiate LLSimpleton::sInstance generically instead of requiring a separate declaration for each subclass. The previous way produces errors in clang. --- indra/llcommon/llsingleton.h | 3 +++ indra/llui/llviewereventrecorder.cpp | 2 -- indra/newview/llenvironment.cpp | 1 - indra/newview/llselectmgr.cpp | 2 -- indra/newview/llviewercamera.cpp | 2 -- indra/newview/llworld.cpp | 2 -- 6 files changed, 3 insertions(+), 9 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 10a8ecfedb..24d01812c9 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -865,4 +865,7 @@ public: } }; +template +T* LLSimpleton::sInstance{ nullptr }; + #endif diff --git a/indra/llui/llviewereventrecorder.cpp b/indra/llui/llviewereventrecorder.cpp index 5a44ec947a..cb000aef74 100644 --- a/indra/llui/llviewereventrecorder.cpp +++ b/indra/llui/llviewereventrecorder.cpp @@ -28,8 +28,6 @@ #include "llui.h" #include "llleap.h" -LLViewerEventRecorder* LLSimpleton::sInstance = nullptr; - LLViewerEventRecorder::LLViewerEventRecorder() { clear(UNDEFINED); diff --git a/indra/newview/llenvironment.cpp b/indra/newview/llenvironment.cpp index dba24b3d02..1a66f10b8f 100644 --- a/indra/newview/llenvironment.cpp +++ b/indra/newview/llenvironment.cpp @@ -824,7 +824,6 @@ std::string env_selection_to_string(LLEnvironment::EnvSelection_t sel) #undef RTNENUM } -LLEnvironment* LLSimpleton::sInstance = nullptr; //------------------------------------------------------------------------- LLEnvironment::LLEnvironment(): mCloudScrollDelta(), diff --git a/indra/newview/llselectmgr.cpp b/indra/newview/llselectmgr.cpp index 53247031b4..bc00c518e9 100644 --- a/indra/newview/llselectmgr.cpp +++ b/indra/newview/llselectmgr.cpp @@ -97,8 +97,6 @@ #include "llglheaders.h" #include "llinventoryobserver.h" -LLSelectMgr* LLSimpleton::sInstance = nullptr; - LLViewerObject* getSelectedParentObject(LLViewerObject *object) ; // // Consts diff --git a/indra/newview/llviewercamera.cpp b/indra/newview/llviewercamera.cpp index 5ebce115f6..5d8e80cc41 100644 --- a/indra/newview/llviewercamera.cpp +++ b/indra/newview/llviewercamera.cpp @@ -54,8 +54,6 @@ // System includes #include // for setprecision -LLViewerCamera* LLSimpleton::sInstance = nullptr; - LLTrace::CountStatHandle<> LLViewerCamera::sVelocityStat("camera_velocity"); LLTrace::CountStatHandle<> LLViewerCamera::sAngularVelocityStat("camera_angular_velocity"); diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp index 0a8457eb2c..5f62908009 100644 --- a/indra/newview/llworld.cpp +++ b/indra/newview/llworld.cpp @@ -62,8 +62,6 @@ #include -LLWorld* LLSimpleton::sInstance = nullptr; - // // Globals // -- cgit v1.3 From 89f2169e9d2c03ed92810689563ca110886abf16 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 4 Nov 2021 16:43:11 -0400 Subject: SL-16202: Add postIfOpen() methods to WorkQueue, LLThreadSafeQueue. postIfOpen() provides a no-exception alternative to post(), which blocks if full but throws if closed. postIfOpen() likewise blocks if full, but returns true if able to post and false if the queue was closed. --- indra/llcommon/llthreadsafequeue.h | 30 ++++++++++++++++++++++-------- indra/llcommon/workqueue.h | 29 ++++++++++++++++++++++++++--- 2 files changed, 48 insertions(+), 11 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 06e8d8f609..5c934791fe 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -85,8 +85,8 @@ public: LLThreadSafeQueue(U32 capacity = 1024); virtual ~LLThreadSafeQueue() {} - // Add an element to the queue (will block if the queue has - // reached capacity). + // Add an element to the queue (will block if the queue has reached + // capacity). // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. @@ -95,6 +95,11 @@ public: // legacy name void pushFront(ElementT const & element) { return push(element); } + // Add an element to the queue (will block if the queue has reached + // capacity). Return false if the queue is closed before push is possible. + template + bool pushIfOpen(T&& element); + // Try to add an element to the queue without blocking. Returns // true only if the element was actually added. template @@ -311,8 +316,8 @@ bool LLThreadSafeQueue::push_(lock_t& lock, T&& element) template -template -void LLThreadSafeQueue::push(T&& element) +template +bool LLThreadSafeQueue::pushIfOpen(T&& element) { lock_t lock1(mLock); while (true) @@ -321,12 +326,10 @@ void LLThreadSafeQueue::push(T&& element) // drained or not: the moment either end calls close(), further push() // operations will fail. if (mClosed) - { - LLTHROW(LLThreadSafeQueueInterrupt()); - } + return false; if (push_(lock1, std::forward(element))) - return; + return true; // Storage Full. Wait for signal. mCapacityCond.wait(lock1); @@ -334,6 +337,17 @@ void LLThreadSafeQueue::push(T&& element) } +template +template +void LLThreadSafeQueue::push(T&& element) +{ + if (! pushIfOpen(std::forward(element))) + { + LLTHROW(LLThreadSafeQueueInterrupt()); + } +} + + template template bool LLThreadSafeQueue::tryPush(T&& element) diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index 76d31f32a6..d0e3f870fe 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -75,9 +75,10 @@ namespace LL template void post(const TimePoint& time, CALLABLE&& callable) { - // Defer reifying an arbitrary CALLABLE until we hit this method. - // All other methods should accept CALLABLEs of arbitrary type to - // avoid multiple levels of std::function indirection. + // Defer reifying an arbitrary CALLABLE until we hit this or + // postIfOpen(). All other methods should accept CALLABLEs of + // arbitrary type to avoid multiple levels of std::function + // indirection. mQueue.push(TimedWork(time, std::move(callable))); } @@ -92,6 +93,28 @@ namespace LL post(TimePoint::clock::now(), std::move(callable)); } + /** + * post work for a particular time, unless the queue is closed before + * we can post + */ + template + bool postIfOpen(const TimePoint& time, CALLABLE&& callable) + { + // Defer reifying an arbitrary CALLABLE until we hit this or + // post(). All other methods should accept CALLABLEs of arbitrary + // type to avoid multiple levels of std::function indirection. + return mQueue.pushIfOpen(TimedWork(time, std::move(callable))); + } + + /** + * post work, unless the queue is closed before we can post + */ + template + bool postIfOpen(CALLABLE&& callable) + { + return postIfOpen(TimePoint::clock::now(), std::move(callable)); + } + /** * Post work to be run at a specified time to another WorkQueue, which * may or may not still exist and be open. Return true if we were able -- cgit v1.3 From 834e7ca088b5f417235327cd290b42459c733594 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 4 Nov 2021 17:18:57 -0400 Subject: SL-16202: Use large WorkQueue size limits for mainloop and General. Give ThreadPool and WorkQueue the ability to override default ThreadSafeSchedule capacity. Instantiate "mainloop" WorkQueue and "General" ThreadPool with very large capacity because we never want to have to block trying to push to either. --- indra/llcommon/threadpool.cpp | 4 ++-- indra/llcommon/threadpool.h | 2 +- indra/llcommon/workqueue.cpp | 5 +++-- indra/llcommon/workqueue.h | 2 +- indra/newview/llappviewer.cpp | 4 +++- indra/newview/llstartup.cpp | 4 +++- 6 files changed, 13 insertions(+), 8 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp index 1899f9a20a..e4fa0eccf3 100644 --- a/indra/llcommon/threadpool.cpp +++ b/indra/llcommon/threadpool.cpp @@ -21,8 +21,8 @@ #include "llevents.h" #include "stringize.h" -LL::ThreadPool::ThreadPool(const std::string& name, size_t threads): - mQueue(name), +LL::ThreadPool::ThreadPool(const std::string& name, size_t threads, size_t capacity): + mQueue(name, capacity), mName("ThreadPool:" + name) { for (size_t i = 0; i < threads; ++i) diff --git a/indra/llcommon/threadpool.h b/indra/llcommon/threadpool.h index 8f3c8514b5..6e3858508b 100644 --- a/indra/llcommon/threadpool.h +++ b/indra/llcommon/threadpool.h @@ -29,7 +29,7 @@ namespace LL * Pass ThreadPool a string name. This can be used to look up the * relevant WorkQueue. */ - ThreadPool(const std::string& name, size_t threads=1); + ThreadPool(const std::string& name, size_t threads=1, size_t capacity=1024); ~ThreadPool(); void close(); diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index 9808757b0a..14ae4c4ab8 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -26,8 +26,9 @@ using Mutex = LLCoros::Mutex; using Lock = LLCoros::LockType; -LL::WorkQueue::WorkQueue(const std::string& name): - super(makeName(name)) +LL::WorkQueue::WorkQueue(const std::string& name, size_t capacity): + super(makeName(name)), + mQueue(capacity) { // TODO: register for "LLApp" events so we can implicitly close() on // viewer shutdown. diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index d0e3f870fe..5987883829 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -54,7 +54,7 @@ namespace LL * You may omit the WorkQueue name, in which case a unique name is * synthesized; for practical purposes that makes it anonymous. */ - WorkQueue(const std::string& name = std::string()); + WorkQueue(const std::string& name = std::string(), size_t capacity=1024); /** * Since the point of WorkQueue is to pass work to some other worker diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index ea2e3a4007..02b4dd57f1 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -367,7 +367,9 @@ BOOL gLogoutInProgress = FALSE; BOOL gSimulateMemLeak = FALSE; -WorkQueue gMainloopWork("mainloop"); +// We don't want anyone, especially threads working on the graphics pipeline, +// to have to block due to this WorkQueue being full. +WorkQueue gMainloopWork("mainloop", 1024*1024); //////////////////////////////////////////////////////////// // Internal globals... that should be removed. diff --git a/indra/newview/llstartup.cpp b/indra/newview/llstartup.cpp index 13e7fcb6e4..9a4149948c 100644 --- a/indra/newview/llstartup.cpp +++ b/indra/newview/llstartup.cpp @@ -313,7 +313,9 @@ void launchThreadPool() << size << " threads" << LL_ENDL; // Use a function-static ThreadPool: static duration, but instantiated // only on demand. - static LL::ThreadPool pool("General", size); + // We don't want anyone, especially the main thread, to have to block + // due to this ThreadPool being full. + static LL::ThreadPool pool("General", size, 1024*1024); } void update_texture_fetch() -- cgit v1.3 From ff5496239bffadaca111b1e4380a01447f85843a Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Fri, 5 Nov 2021 12:33:31 -0400 Subject: SL-16202: Use WorkQueue::postTo() for texture create/post handshake. That is, when LLViewerFetchedTexture::scheduleCreateTexture() wants to call createTexture() on the LLImageGLThread, but postCreateTexture() on the main thread, use the "mainloop" WorkQueue to set up the handshake. Give ThreadPool a public virtual run() method so a subclass can override with desired behavior. This necessitates a virtual destructor. Add accessors for embedded WorkQueue (for post calls), ThreadPool name and width (in threads). Allow LLSimpleton::createInstance() to forward arguments to the subject constructor. Make LLImageGLThread an LLSimpleton - that abstraction didn't yet exist at the time LLImageGLThread was coded. Also derive from ThreadPool rather than LLThread. Make it a single-thread "pool" with a very large queue capacity. --- indra/llcommon/llsingleton.h | 20 +++++++++++--------- indra/llcommon/threadpool.cpp | 7 ++++++- indra/llcommon/threadpool.h | 18 +++++++++++++++++- indra/llrender/llimagegl.cpp | 22 +++++++++++----------- indra/llrender/llimagegl.h | 11 ++++------- indra/newview/llviewertexture.cpp | 34 +++++++++++++++++++++++----------- indra/newview/llviewertexture.h | 4 ++++ 7 files changed, 76 insertions(+), 40 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 24d01812c9..fdd5bdfea9 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -847,22 +847,24 @@ template class LLSimpleton { public: - static T* sInstance; - - static void createInstance() - { + template + static void createInstance(ARGS&&... args) + { llassert(sInstance == nullptr); - sInstance = new T(); + sInstance = new T(std::forward(args)...); } - + static inline T* getInstance() { return sInstance; } static inline T& instance() { return *getInstance(); } static inline bool instanceExists() { return sInstance != nullptr; } - static void deleteSingleton() { - delete sInstance; - sInstance = nullptr; + static void deleteSingleton() { + delete sInstance; + sInstance = nullptr; } + +private: + static T* sInstance; }; template diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp index e4fa0eccf3..cf25cc838e 100644 --- a/indra/llcommon/threadpool.cpp +++ b/indra/llcommon/threadpool.cpp @@ -70,6 +70,11 @@ void LL::ThreadPool::close() void LL::ThreadPool::run(const std::string& name) { LL_DEBUGS("ThreadPool") << name << " starting" << LL_ENDL; - mQueue.runUntilClose(); + run(); LL_DEBUGS("ThreadPool") << name << " stopping" << LL_ENDL; } + +void LL::ThreadPool::run() +{ + mQueue.runUntilClose(); +} diff --git a/indra/llcommon/threadpool.h b/indra/llcommon/threadpool.h index 6e3858508b..1ca24aec58 100644 --- a/indra/llcommon/threadpool.h +++ b/indra/llcommon/threadpool.h @@ -30,9 +30,25 @@ namespace LL * relevant WorkQueue. */ ThreadPool(const std::string& name, size_t threads=1, size_t capacity=1024); - ~ThreadPool(); + virtual ~ThreadPool(); + + /** + * ThreadPool listens for application shutdown messages on the "LLApp" + * LLEventPump. Call close() to shut down this ThreadPool early. + */ void close(); + std::string getName() const { return mName; } + size_t getWidth() const { return mThreads.size(); } + /// obtain a non-const reference to the WorkQueue to post work to it + WorkQueue& getQueue() { return mQueue; } + + /** + * Override run() if you need special processing. The default run() + * implementation simply calls WorkQueue::runUntilClose(). + */ + virtual void run(); + private: void run(const std::string& name); diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 71c48801ac..1b6920fe3b 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -172,24 +172,19 @@ BOOL is_little_endian() return (*c == 0x78) ; } -LLImageGLThread* LLImageGLThread::sInstance = nullptr; - //static void LLImageGL::initClass(LLWindow* window, S32 num_catagories, BOOL skip_analyze_alpha /* = false */) { LL_PROFILE_ZONE_SCOPED; sSkipAnalyzeAlpha = skip_analyze_alpha; - LLImageGLThread::sInstance = new LLImageGLThread(window); - LLImageGLThread::sInstance->start(); + LLImageGLThread::createInstance(window); } //static void LLImageGL::cleanupClass() { LL_PROFILE_ZONE_SCOPED; - LLImageGLThread::sInstance->mFunctionQueue.close(); - delete LLImageGLThread::sInstance; - LLImageGLThread::sInstance = nullptr; + LLImageGLThread::deleteSingleton(); } //static @@ -1532,8 +1527,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_ } //if we're on the image loading thread, be sure to delete old_texname and update mTexName on the main thread - if (LLImageGLThread::sInstance != nullptr && - LLThread::currentID() == LLImageGLThread::sInstance->getID()) + if (! on_main_thread()) { { LL_PROFILE_ZONE_NAMED("cglt - sync"); @@ -2257,7 +2251,11 @@ void LLImageGL::resetCurTexSizebar() */ LLImageGLThread::LLImageGLThread(LLWindow* window) - : LLThread("LLImageGL"), mWindow(window) + // We want exactly one thread, but a very large capacity: we never want + // anyone, especially inner-loop render code, to have to block on post() + // because we're full. + : ThreadPool("LLImageGL", 1, 1024*1024) + , mWindow(window) { mFinished = false; @@ -2266,9 +2264,11 @@ LLImageGLThread::LLImageGLThread(LLWindow* window) void LLImageGLThread::run() { + // We must perform setup on this thread before actually servicing our + // WorkQueue, likewise cleanup afterwards. mWindow->makeContextCurrent(mContext); gGL.init(); - mFunctionQueue.runUntilClose(); + ThreadPool::run(); gGL.shutdown(); mWindow->destroySharedContext(mContext); } diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h index b9de481aae..27496def1d 100644 --- a/indra/llrender/llimagegl.h +++ b/indra/llrender/llimagegl.h @@ -37,6 +37,7 @@ #include "llunits.h" #include "llthreadsafequeue.h" #include "llrender.h" +#include "threadpool.h" #include "workqueue.h" class LLTextureAtlas ; @@ -307,7 +308,7 @@ public: }; -class LLImageGLThread : public LLThread +class LLImageGLThread : public LLSimpleton, LL::ThreadPool { public: LLImageGLThread(LLWindow* window); @@ -316,19 +317,15 @@ public: template bool post(CALLABLE&& func) { - return mFunctionQueue.postIfOpen(std::forward(func)); + return getQueue().postIfOpen(std::forward(func)); } void run() override; - // Work Queue for background thread - LL::WorkQueue mFunctionQueue; - +private: LLWindow* mWindow; void* mContext; LLAtomicBool mFinished; - - static LLImageGLThread* sInstance; }; diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index 9f3819f7d1..498e4ef8bc 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -679,6 +679,9 @@ void LLViewerTexture::init(bool firstinit) mVolumeList[LLRender::LIGHT_TEX].clear(); mVolumeList[LLRender::SCULPT_TEX].clear(); + + mMainQueue = LL::WorkQueue::getInstance("mainloop"); + mImageQueue = LL::WorkQueue::getInstance("LLImageGL"); } //virtual @@ -1622,17 +1625,26 @@ void LLViewerFetchedTexture::scheduleCreateTexture() { mNeedsCreateTexture = TRUE; #if LL_WINDOWS //flip to 0 to revert to single-threaded OpenGL texture uploads - if (!LLImageGLThread::sInstance->post([this]() - { - //actually create the texture on a background thread - createTexture(); - LL::WorkQueue::getInstance("mainloop")->post([this]() - { - //finalize on main thread - postCreateTexture(); - unref(); - }); - })) + auto mainq = mMainQueue.lock(); + if (mainq) + { + mainq->postTo( + mImageQueue, + // work to be done on LLImageGL worker thread + [this]() + { + //actually create the texture on a background thread + createTexture(); + }, + // callback to be run on main thread + [this]() + { + //finalize on main thread + postCreateTexture(); + unref(); + }); + } + else #endif { gTextureList.mCreateTextureList.insert(this); diff --git a/indra/newview/llviewertexture.h b/indra/newview/llviewertexture.h index f9f1bfef44..4cd4c7cd39 100644 --- a/indra/newview/llviewertexture.h +++ b/indra/newview/llviewertexture.h @@ -35,6 +35,7 @@ #include "llrender.h" #include "llmetricperformancetester.h" #include "httpcommon.h" +#include "workqueue.h" #include #include @@ -213,6 +214,9 @@ protected: //do not use LLPointer here. LLViewerMediaTexture* mParcelMedia ; + LL::WorkQueue::weak_t mMainQueue; + LL::WorkQueue::weak_t mImageQueue; + static F32 sTexelPixelRatio; public: static const U32 sCurrentFileVersion; -- cgit v1.3 From ca0b9a3753fa3b42d4ac8183adcf30d957f55016 Mon Sep 17 00:00:00 2001 From: "Brad Payne (Vir Linden)" Date: Tue, 9 Nov 2021 20:25:25 +0000 Subject: SL-16329 - track frame time and jitter (as average deviation frame to frame) in stats window --- indra/llcommon/lltracerecording.cpp | 27 +++++++++++++++++++- indra/llcommon/lltracerecording.h | 29 ++++++++++++++++++++++ indra/llui/llstatbar.cpp | 17 +++++++++++-- indra/llui/llstatbar.h | 6 +++-- indra/newview/llviewerstats.cpp | 11 +++++--- indra/newview/llviewerstats.h | 4 +-- .../newview/skins/default/xui/en/floater_stats.xml | 19 ++++++++++++++ 7 files changed, 103 insertions(+), 10 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/lltracerecording.cpp b/indra/llcommon/lltracerecording.cpp index c72a64d086..5ce1b337fe 100644 --- a/indra/llcommon/lltracerecording.cpp +++ b/indra/llcommon/lltracerecording.cpp @@ -858,7 +858,6 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S3 : NaN; } - F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { LL_PROFILE_ZONE_SCOPED; @@ -952,6 +951,32 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S : NaN; } +F64 PeriodicRecording::getPeriodMedian( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) +{ + LL_PROFILE_ZONE_SCOPED; + num_periods = llmin(num_periods, getNumRecordedPeriods()); + + std::vector buf; + for (S32 i = 1; i <= num_periods; i++) + { + Recording& recording = getPrevRecording(i); + if (recording.getDuration() > (F32Seconds)0.f) + { + if (recording.hasValue(stat)) + { + buf.push_back(recording.getMean(stat)); + } + } + } + if (buf.size()==0) + { + return 0.0f; + } + std::sort(buf.begin(), buf.end()); + + return F64((buf.size() % 2 == 0) ? (buf[buf.size() / 2 - 1] + buf[buf.size() / 2]) / 2 : buf[buf.size() / 2]); +} + F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { LL_PROFILE_ZONE_SCOPED; diff --git a/indra/llcommon/lltracerecording.h b/indra/llcommon/lltracerecording.h index 6715104613..1f3d37336a 100644 --- a/indra/llcommon/lltracerecording.h +++ b/indra/llcommon/lltracerecording.h @@ -599,6 +599,35 @@ namespace LLTrace return typename RelatedTypes::fractional_t(getPeriodMeanPerSec(static_cast&>(stat), num_periods)); } + F64 getPeriodMedian( const StatType& stat, S32 num_periods = S32_MAX); + + template + typename RelatedTypes::fractional_t getPeriodMedianPerSec(const StatType& stat, S32 num_periods = S32_MAX) + { + LL_PROFILE_ZONE_SCOPED; + num_periods = llmin(num_periods, getNumRecordedPeriods()); + + std::vector ::fractional_t> buf; + for (S32 i = 1; i <= num_periods; i++) + { + Recording& recording = getPrevRecording(i); + if (recording.getDuration() > (F32Seconds)0.f) + { + buf.push_back(recording.getPerSec(stat)); + } + } + std::sort(buf.begin(), buf.end()); + + return typename RelatedTypes::fractional_t((buf.size() % 2 == 0) ? (buf[buf.size() / 2 - 1] + buf[buf.size() / 2]) / 2 : buf[buf.size() / 2]); + } + + template + typename RelatedTypes::fractional_t getPeriodMedianPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) + { + LL_PROFILE_ZONE_SCOPED; + return typename RelatedTypes::fractional_t(getPeriodMedianPerSec(static_cast&>(stat), num_periods)); + } + // // PERIODIC STANDARD DEVIATION // diff --git a/indra/llui/llstatbar.cpp b/indra/llui/llstatbar.cpp index 8adcd664df..2449100952 100644 --- a/indra/llui/llstatbar.cpp +++ b/indra/llui/llstatbar.cpp @@ -160,6 +160,7 @@ LLStatBar::Params::Params() tick_spacing("tick_spacing", 0.f), decimal_digits("decimal_digits", 3), show_bar("show_bar", false), + show_median("show_median", false), show_history("show_history", false), scale_range("scale_range", true), num_frames("num_frames", 200), @@ -186,6 +187,7 @@ LLStatBar::LLStatBar(const Params& p) mNumShortHistoryFrames(p.num_frames_short), mMaxHeight(p.max_height), mDisplayBar(p.show_bar), + mShowMedian(p.show_median), mDisplayHistory(p.show_history), mOrientation(p.orientation), mAutoScaleMax(!p.bar_max.isProvided()), @@ -318,7 +320,14 @@ void LLStatBar::draw() min = frame_recording.getPeriodMinPerSec(count_stat, num_frames); max = frame_recording.getPeriodMaxPerSec(count_stat, num_frames); mean = frame_recording.getPeriodMeanPerSec(count_stat, num_frames); - display_value = mean; + if (mShowMedian) + { + display_value = frame_recording.getPeriodMedianPerSec(count_stat, num_frames); + } + else + { + display_value = mean; + } } break; case STAT_EVENT: @@ -344,7 +353,11 @@ void LLStatBar::draw() mean = frame_recording.getPeriodMean(sample_stat, num_frames); num_rapid_changes = calc_num_rapid_changes(frame_recording, sample_stat, RAPID_CHANGE_WINDOW); - if (num_rapid_changes / RAPID_CHANGE_WINDOW.value() > MAX_RAPID_CHANGES_PER_SEC) + if (mShowMedian) + { + display_value = frame_recording.getPeriodMedian(sample_stat, num_frames); + } + else if (num_rapid_changes / RAPID_CHANGE_WINDOW.value() > MAX_RAPID_CHANGES_PER_SEC) { display_value = mean; } diff --git a/indra/llui/llstatbar.h b/indra/llui/llstatbar.h index 1ff4c67fc5..6b481ca68f 100644 --- a/indra/llui/llstatbar.h +++ b/indra/llui/llstatbar.h @@ -44,9 +44,10 @@ public: bar_max, tick_spacing; - Optional show_bar, + Optional show_bar, show_history, - scale_range; + scale_range, + show_median; // default is mean Optional decimal_digits, num_frames, @@ -112,6 +113,7 @@ private: bool mDisplayBar, // Display the bar graph. mDisplayHistory, + mShowMedian, mAutoScaleMax, mAutoScaleMin; }; diff --git a/indra/newview/llviewerstats.cpp b/indra/newview/llviewerstats.cpp index 314c1a1f1e..ac8a657fb2 100644 --- a/indra/newview/llviewerstats.cpp +++ b/indra/newview/llviewerstats.cpp @@ -183,8 +183,9 @@ SimMeasurement SIM_UNACKED_BYTES("simtotalunackedbytes", "", LL_S SimMeasurement SIM_PHYSICS_MEM("physicsmemoryallocated", "", LL_SIM_STAT_SIMPHYSICSMEMORY); LLTrace::SampleStatHandle FRAMETIME_JITTER("frametimejitter", "Average delta between successive frame times"), - FRAMETIME_SLEW("frametimeslew", "Average delta between frame time and mean"), - SIM_PING("simpingstat"); + FRAMETIME_SLEW("frametimeslew", "Average delta between frame time and mean"), + FRAMETIME("frametime", "Measured frame time"), + SIM_PING("simpingstat"); LLTrace::EventStatHandle > AGENT_POSITION_SNAP("agentpositionsnap", "agent position corrections"); @@ -261,8 +262,12 @@ void LLViewerStats::updateFrameStats(const F64Seconds time_diff) // new "stutter" meter add(LLStatViewer::FRAMETIME_DOUBLED, time_diff >= 2.0 * mLastTimeDiff ? 1 : 0); + sample(LLStatViewer::FRAMETIME, time_diff); + // old stats that were never really used - sample(LLStatViewer::FRAMETIME_JITTER, F64Milliseconds (mLastTimeDiff - time_diff)); + F64Seconds jit = (F64Seconds) std::fabs((mLastTimeDiff - time_diff)); + LL_INFOS() << "times " << mLastTimeDiff << ", " << time_diff << " jit " << jit << LL_ENDL; + sample(LLStatViewer::FRAMETIME_JITTER, jit); F32Seconds average_frametime = gRenderStartTime.getElapsedTimeF32() / (F32)gFrameCount; sample(LLStatViewer::FRAMETIME_SLEW, F64Milliseconds (average_frametime - time_diff)); diff --git a/indra/newview/llviewerstats.h b/indra/newview/llviewerstats.h index 04870e0c26..ac8eccc0ca 100644 --- a/indra/newview/llviewerstats.h +++ b/indra/newview/llviewerstats.h @@ -218,8 +218,8 @@ extern SimMeasurement SIM_PHYSICS_MEM; extern LLTrace::SampleStatHandle FRAMETIME_JITTER, - FRAMETIME_SLEW, - SIM_PING; + FRAMETIME_SLEW, + SIM_PING; extern LLTrace::EventStatHandle > AGENT_POSITION_SNAP; diff --git a/indra/newview/skins/default/xui/en/floater_stats.xml b/indra/newview/skins/default/xui/en/floater_stats.xml index e4f735740b..6f84930c75 100644 --- a/indra/newview/skins/default/xui/en/floater_stats.xml +++ b/indra/newview/skins/default/xui/en/floater_stats.xml @@ -35,6 +35,25 @@ decimal_digits="1" show_bar="true" show_history="true"/> + + + Date: Wed, 10 Nov 2021 10:13:38 -0500 Subject: SL-16094: Add WorkQueue::size() method to support changeset 08336bb. We want to skip calling PostMessage() to bump the window thread out of GetMessage() in any frame with no work functions pending for that thread. That test depends on being able to sense the size() of the queue. Having converted to WorkQueue, we need that queue to support size(). --- indra/llcommon/workqueue.cpp | 5 +++++ indra/llcommon/workqueue.h | 15 +++++++++++++++ 2 files changed, 20 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index 14ae4c4ab8..633594ceea 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -39,6 +39,11 @@ void LL::WorkQueue::close() mQueue.close(); } +size_t LL::WorkQueue::size() +{ + return mQueue.size(); +} + bool LL::WorkQueue::isClosed() { return mQueue.isClosed(); diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index 5987883829..c25d787425 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -64,6 +64,21 @@ namespace LL */ void close(); + /** + * WorkQueue supports multiple producers and multiple consumers. In + * the general case it's misleading to test size(), since any other + * thread might change it the nanosecond the lock is released. On that + * basis, some might argue against publishing a size() method at all. + * + * But there are two specific cases in which a test based on size() + * might be reasonable: + * + * * If you're the only producer, noticing that size() == 0 is + * meaningful. + * * If you're the only consumer, noticing that size() > 0 is + * meaningful. + */ + size_t size(); /// producer end: are we prevented from pushing any additional items? bool isClosed(); /// consumer end: are we done, is the queue entirely drained? -- cgit v1.3 From 75110629de7786d667ea7c90b025f97c22650316 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 11 Nov 2021 10:23:16 -0500 Subject: SL-16094: Stylish braces! --- indra/llcommon/llsingleton.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index fdd5bdfea9..6042c0906c 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -858,7 +858,8 @@ public: static inline T& instance() { return *getInstance(); } static inline bool instanceExists() { return sInstance != nullptr; } - static void deleteSingleton() { + static void deleteSingleton() + { delete sInstance; sInstance = nullptr; } -- cgit v1.3 From 029b41c0419e975bbb28454538b46dc69ce5d2ba Mon Sep 17 00:00:00 2001 From: Dave Houlton Date: Mon, 15 Nov 2021 09:25:35 -0700 Subject: Revert "SL-16220: Merge branch 'origin/DRTVWR-546' into glthread" This reverts commit 5188a26a8521251dda07ac0140bb129f28417e49, reversing changes made to 819088563e13f1d75e048311fbaf0df4a79b7e19. --- indra/llcommon/CMakeLists.txt | 3 +- indra/llcommon/llsingleton.h | 24 +- indra/llcommon/llthreadsafequeue.h | 30 +- indra/llcommon/tests/threadsafeschedule_test.cpp | 4 +- indra/llcommon/tests/workqueue_test.cpp | 72 +--- indra/llcommon/threadpool.cpp | 80 ----- indra/llcommon/threadpool.h | 62 ---- indra/llcommon/timing.cpp | 25 ++ indra/llcommon/workqueue.cpp | 30 +- indra/llcommon/workqueue.h | 378 ++++---------------- indra/llrender/llimagegl.cpp | 91 ++++- indra/llrender/llimagegl.h | 27 +- indra/llui/CMakeLists.txt | 6 +- indra/llui/llviewereventrecorder.cpp | 2 + indra/llwindow/llwindowwin32.cpp | 426 ++++++++++++++++------- indra/llwindow/llwindowwin32.h | 48 ++- indra/newview/CMakeLists.txt | 3 +- indra/newview/app_settings/settings.xml | 25 -- indra/newview/llappviewer.cpp | 66 ++-- indra/newview/llenvironment.cpp | 1 + indra/newview/llmainlooprepeater.cpp | 88 +++++ indra/newview/llmainlooprepeater.h | 64 ++++ indra/newview/llselectmgr.cpp | 2 + indra/newview/llstartup.cpp | 20 -- indra/newview/llviewercamera.cpp | 2 + indra/newview/llviewertexture.cpp | 34 +- indra/newview/llviewertexture.h | 4 - indra/newview/llworld.cpp | 2 + 28 files changed, 770 insertions(+), 849 deletions(-) delete mode 100644 indra/llcommon/threadpool.cpp delete mode 100644 indra/llcommon/threadpool.h create mode 100644 indra/llcommon/timing.cpp create mode 100644 indra/newview/llmainlooprepeater.cpp create mode 100644 indra/newview/llmainlooprepeater.h (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 78d6ea3090..ad6d3a5049 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -119,8 +119,8 @@ set(llcommon_SOURCE_FILES lluriparser.cpp lluuid.cpp llworkerthread.cpp + timing.cpp u64.cpp - threadpool.cpp workqueue.cpp StackWalker.cpp ) @@ -256,7 +256,6 @@ set(llcommon_HEADER_FILES lockstatic.h stdtypes.h stringize.h - threadpool.h threadsafeschedule.h timer.h tuple.h diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 6042c0906c..10a8ecfedb 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -847,28 +847,22 @@ template class LLSimpleton { public: - template - static void createInstance(ARGS&&... args) - { + static T* sInstance; + + static void createInstance() + { llassert(sInstance == nullptr); - sInstance = new T(std::forward(args)...); + sInstance = new T(); } - + static inline T* getInstance() { return sInstance; } static inline T& instance() { return *getInstance(); } static inline bool instanceExists() { return sInstance != nullptr; } - static void deleteSingleton() - { - delete sInstance; - sInstance = nullptr; + static void deleteSingleton() { + delete sInstance; + sInstance = nullptr; } - -private: - static T* sInstance; }; -template -T* LLSimpleton::sInstance{ nullptr }; - #endif diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 5c934791fe..06e8d8f609 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -85,8 +85,8 @@ public: LLThreadSafeQueue(U32 capacity = 1024); virtual ~LLThreadSafeQueue() {} - // Add an element to the queue (will block if the queue has reached - // capacity). + // Add an element to the queue (will block if the queue has + // reached capacity). // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. @@ -95,11 +95,6 @@ public: // legacy name void pushFront(ElementT const & element) { return push(element); } - // Add an element to the queue (will block if the queue has reached - // capacity). Return false if the queue is closed before push is possible. - template - bool pushIfOpen(T&& element); - // Try to add an element to the queue without blocking. Returns // true only if the element was actually added. template @@ -316,8 +311,8 @@ bool LLThreadSafeQueue::push_(lock_t& lock, T&& element) template -template -bool LLThreadSafeQueue::pushIfOpen(T&& element) +template +void LLThreadSafeQueue::push(T&& element) { lock_t lock1(mLock); while (true) @@ -326,10 +321,12 @@ bool LLThreadSafeQueue::pushIfOpen(T&& element) // drained or not: the moment either end calls close(), further push() // operations will fail. if (mClosed) - return false; + { + LLTHROW(LLThreadSafeQueueInterrupt()); + } if (push_(lock1, std::forward(element))) - return true; + return; // Storage Full. Wait for signal. mCapacityCond.wait(lock1); @@ -337,17 +334,6 @@ bool LLThreadSafeQueue::pushIfOpen(T&& element) } -template -template -void LLThreadSafeQueue::push(T&& element) -{ - if (! pushIfOpen(std::forward(element))) - { - LLTHROW(LLThreadSafeQueueInterrupt()); - } -} - - template template bool LLThreadSafeQueue::tryPush(T&& element) diff --git a/indra/llcommon/tests/threadsafeschedule_test.cpp b/indra/llcommon/tests/threadsafeschedule_test.cpp index c421cc7b1c..af67b9f492 100644 --- a/indra/llcommon/tests/threadsafeschedule_test.cpp +++ b/indra/llcommon/tests/threadsafeschedule_test.cpp @@ -46,11 +46,11 @@ namespace tut // the real time required for each push() call. Explicitly increment // the timestamp for each one -- but since we're passing explicit // timestamps, make the queue reorder them. - queue.push(Queue::TimeTuple(Queue::Clock::now() + 200ms, "ghi")); + queue.push(Queue::TimeTuple(Queue::Clock::now() + 20ms, "ghi")); // Given the various push() overloads, you have to match the type // exactly: conversions are ambiguous. queue.push("abc"s); - queue.push(Queue::Clock::now() + 100ms, "def"); + queue.push(Queue::Clock::now() + 10ms, "def"); queue.close(); auto entry = queue.pop(); ensure_equals("failed to pop first", std::get<0>(entry), "abc"s); diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp index bea3ad911b..d5405400fd 100644 --- a/indra/llcommon/tests/workqueue_test.cpp +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -20,10 +20,7 @@ // external library headers // other Linden headers #include "../test/lltut.h" -#include "../test/catch_and_store_what_in.h" #include "llcond.h" -#include "llcoros.h" -#include "lleventcoro.h" #include "llstring.h" #include "stringize.h" @@ -141,8 +138,7 @@ namespace tut [](){ return 17; }, // Note that a postTo() *callback* can safely bind a reference to // a variable on the invoking thread, because the callback is run - // on the invoking thread. (Of course the bound variable must - // survive until the callback is called.) + // on the invoking thread. [&result](int i){ result = i; }); // this should post the callback to main qptr->runOne(); @@ -160,70 +156,4 @@ namespace tut main.runPending(); ensure_equals("failed to run string callback", alpha, "abc"); } - - template<> template<> - void object::test<5>() - { - set_test_name("postTo with void return"); - WorkQueue main("main"); - auto qptr = WorkQueue::getInstance("queue"); - std::string observe; - main.postTo( - qptr, - // The ONLY reason we can get away with binding a reference to - // 'observe' in our work callable is because we're directly - // calling qptr->runOne() on this same thread. It would be a - // mistake to do that if some other thread were servicing 'queue'. - [&observe](){ observe = "queue"; }, - [&observe](){ observe.append(";main"); }); - qptr->runOne(); - main.runOne(); - ensure_equals("failed to run both lambdas", observe, "queue;main"); - } - - template<> template<> - void object::test<6>() - { - set_test_name("waitForResult"); - std::string stored; - // Try to call waitForResult() on this thread's main coroutine. It - // should throw because the main coroutine must service the queue. - auto what{ catch_what( - [this, &stored](){ stored = queue.waitForResult( - [](){ return "should throw"; }); }) }; - ensure("lambda should not have run", stored.empty()); - ensure_not("waitForResult() should have thrown", what.empty()); - ensure(STRINGIZE("should mention waitForResult: " << what), - what.find("waitForResult") != std::string::npos); - - // Call waitForResult() on a coroutine, with a string result. - LLCoros::instance().launch( - "waitForResult string", - [this, &stored]() - { stored = queue.waitForResult( - [](){ return "string result"; }); }); - llcoro::suspend(); - // Nothing will have happened yet because, even if the coroutine did - // run immediately, all it did was to queue the inner lambda on - // 'queue'. Service it. - queue.runOne(); - llcoro::suspend(); - ensure_equals("bad waitForResult return", stored, "string result"); - - // Call waitForResult() on a coroutine, with a void callable. - stored.clear(); - bool done = false; - LLCoros::instance().launch( - "waitForResult void", - [this, &stored, &done]() - { - queue.waitForResult([&stored](){ stored = "ran"; }); - done = true; - }); - llcoro::suspend(); - queue.runOne(); - llcoro::suspend(); - ensure_equals("didn't run coroutine", stored, "ran"); - ensure("void waitForResult() didn't return", done); - } } // namespace tut diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp deleted file mode 100644 index cf25cc838e..0000000000 --- a/indra/llcommon/threadpool.cpp +++ /dev/null @@ -1,80 +0,0 @@ -/** - * @file threadpool.cpp - * @author Nat Goodspeed - * @date 2021-10-21 - * @brief Implementation for threadpool. - * - * $LicenseInfo:firstyear=2021&license=viewerlgpl$ - * Copyright (c) 2021, Linden Research, Inc. - * $/LicenseInfo$ - */ - -// Precompiled header -#include "linden_common.h" -// associated header -#include "threadpool.h" -// STL headers -// std headers -// external library headers -// other Linden headers -#include "llerror.h" -#include "llevents.h" -#include "stringize.h" - -LL::ThreadPool::ThreadPool(const std::string& name, size_t threads, size_t capacity): - mQueue(name, capacity), - mName("ThreadPool:" + name) -{ - for (size_t i = 0; i < threads; ++i) - { - std::string tname{ STRINGIZE(mName << ':' << (i+1) << '/' << threads) }; - mThreads.emplace_back(tname, [this, tname](){ run(tname); }); - } - // Listen on "LLApp", and when the app is shutting down, close the queue - // and join the workers. - LLEventPumps::instance().obtain("LLApp").listen( - mName, - [this](const LLSD& stat) - { - std::string status(stat["status"]); - if (status != "running") - { - // viewer is starting shutdown -- proclaim the end is nigh! - LL_DEBUGS("ThreadPool") << mName << " saw " << status << LL_ENDL; - close(); - } - return false; - }); -} - -LL::ThreadPool::~ThreadPool() -{ - close(); -} - -void LL::ThreadPool::close() -{ - if (! mQueue.isClosed()) - { - LL_DEBUGS("ThreadPool") << mName << " closing queue and joining threads" << LL_ENDL; - mQueue.close(); - for (auto& pair: mThreads) - { - LL_DEBUGS("ThreadPool") << mName << " waiting on thread " << pair.first << LL_ENDL; - pair.second.join(); - } - LL_DEBUGS("ThreadPool") << mName << " shutdown complete" << LL_ENDL; - } -} - -void LL::ThreadPool::run(const std::string& name) -{ - LL_DEBUGS("ThreadPool") << name << " starting" << LL_ENDL; - run(); - LL_DEBUGS("ThreadPool") << name << " stopping" << LL_ENDL; -} - -void LL::ThreadPool::run() -{ - mQueue.runUntilClose(); -} diff --git a/indra/llcommon/threadpool.h b/indra/llcommon/threadpool.h deleted file mode 100644 index 1ca24aec58..0000000000 --- a/indra/llcommon/threadpool.h +++ /dev/null @@ -1,62 +0,0 @@ -/** - * @file threadpool.h - * @author Nat Goodspeed - * @date 2021-10-21 - * @brief ThreadPool configures a WorkQueue along with a pool of threads to - * service it. - * - * $LicenseInfo:firstyear=2021&license=viewerlgpl$ - * Copyright (c) 2021, Linden Research, Inc. - * $/LicenseInfo$ - */ - -#if ! defined(LL_THREADPOOL_H) -#define LL_THREADPOOL_H - -#include "workqueue.h" -#include -#include -#include // std::pair -#include - -namespace LL -{ - - class ThreadPool - { - public: - /** - * Pass ThreadPool a string name. This can be used to look up the - * relevant WorkQueue. - */ - ThreadPool(const std::string& name, size_t threads=1, size_t capacity=1024); - virtual ~ThreadPool(); - - /** - * ThreadPool listens for application shutdown messages on the "LLApp" - * LLEventPump. Call close() to shut down this ThreadPool early. - */ - void close(); - - std::string getName() const { return mName; } - size_t getWidth() const { return mThreads.size(); } - /// obtain a non-const reference to the WorkQueue to post work to it - WorkQueue& getQueue() { return mQueue; } - - /** - * Override run() if you need special processing. The default run() - * implementation simply calls WorkQueue::runUntilClose(). - */ - virtual void run(); - - private: - void run(const std::string& name); - - WorkQueue mQueue; - std::string mName; - std::vector> mThreads; - }; - -} // namespace LL - -#endif /* ! defined(LL_THREADPOOL_H) */ diff --git a/indra/llcommon/timing.cpp b/indra/llcommon/timing.cpp new file mode 100644 index 0000000000..c2dc695ef3 --- /dev/null +++ b/indra/llcommon/timing.cpp @@ -0,0 +1,25 @@ +/** + * @file timing.cpp + * @brief This file will be deprecated in the future. + * + * $LicenseInfo:firstyear=2000&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2010, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index 633594ceea..b32357e832 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -26,9 +26,8 @@ using Mutex = LLCoros::Mutex; using Lock = LLCoros::LockType; -LL::WorkQueue::WorkQueue(const std::string& name, size_t capacity): - super(makeName(name)), - mQueue(capacity) +LL::WorkQueue::WorkQueue(const std::string& name): + super(makeName(name)) { // TODO: register for "LLApp" events so we can implicitly close() on // viewer shutdown. @@ -39,21 +38,6 @@ void LL::WorkQueue::close() mQueue.close(); } -size_t LL::WorkQueue::size() -{ - return mQueue.size(); -} - -bool LL::WorkQueue::isClosed() -{ - return mQueue.isClosed(); -} - -bool LL::WorkQueue::done() -{ - return mQueue.done(); -} - void LL::WorkQueue::runUntilClose() { try @@ -144,13 +128,3 @@ void LL::WorkQueue::error(const std::string& msg) { LL_ERRS("WorkQueue") << msg << LL_ENDL; } - -void LL::WorkQueue::checkCoroutine(const std::string& method) -{ - // By convention, the default coroutine on each thread has an empty name - // string. See also LLCoros::logname(). - if (LLCoros::getName().empty()) - { - LLTHROW(Error("Do not call " + method + " from a thread's default coroutine")); - } -} diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index c25d787425..5ec790da79 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -12,14 +12,14 @@ #if ! defined(LL_WORKQUEUE_H) #define LL_WORKQUEUE_H -#include "llcoros.h" -#include "llexception.h" #include "llinstancetracker.h" #include "threadsafeschedule.h" #include -#include // std::current_exception #include // std::function +#include #include +#include // std::pair +#include namespace LL { @@ -45,16 +45,11 @@ namespace LL using TimedWork = Queue::TimeTuple; using Closed = Queue::Closed; - struct Error: public LLException - { - Error(const std::string& what): LLException(what) {} - }; - /** * You may omit the WorkQueue name, in which case a unique name is * synthesized; for practical purposes that makes it anonymous. */ - WorkQueue(const std::string& name = std::string(), size_t capacity=1024); + WorkQueue(const std::string& name = std::string()); /** * Since the point of WorkQueue is to pass work to some other worker @@ -64,36 +59,15 @@ namespace LL */ void close(); - /** - * WorkQueue supports multiple producers and multiple consumers. In - * the general case it's misleading to test size(), since any other - * thread might change it the nanosecond the lock is released. On that - * basis, some might argue against publishing a size() method at all. - * - * But there are two specific cases in which a test based on size() - * might be reasonable: - * - * * If you're the only producer, noticing that size() == 0 is - * meaningful. - * * If you're the only consumer, noticing that size() > 0 is - * meaningful. - */ - size_t size(); - /// producer end: are we prevented from pushing any additional items? - bool isClosed(); - /// consumer end: are we done, is the queue entirely drained? - bool done(); - /*---------------------- fire and forget API -----------------------*/ /// fire-and-forget, but at a particular (future?) time template void post(const TimePoint& time, CALLABLE&& callable) { - // Defer reifying an arbitrary CALLABLE until we hit this or - // postIfOpen(). All other methods should accept CALLABLEs of - // arbitrary type to avoid multiple levels of std::function - // indirection. + // Defer reifying an arbitrary CALLABLE until we hit this method. + // All other methods should accept CALLABLEs of arbitrary type to + // avoid multiple levels of std::function indirection. mQueue.push(TimedWork(time, std::move(callable))); } @@ -108,47 +82,6 @@ namespace LL post(TimePoint::clock::now(), std::move(callable)); } - /** - * post work for a particular time, unless the queue is closed before - * we can post - */ - template - bool postIfOpen(const TimePoint& time, CALLABLE&& callable) - { - // Defer reifying an arbitrary CALLABLE until we hit this or - // post(). All other methods should accept CALLABLEs of arbitrary - // type to avoid multiple levels of std::function indirection. - return mQueue.pushIfOpen(TimedWork(time, std::move(callable))); - } - - /** - * post work, unless the queue is closed before we can post - */ - template - bool postIfOpen(CALLABLE&& callable) - { - return postIfOpen(TimePoint::clock::now(), std::move(callable)); - } - - /** - * Post work to be run at a specified time to another WorkQueue, which - * may or may not still exist and be open. Return true if we were able - * to post. - */ - template - static bool postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable); - - /** - * Post work to another WorkQueue, which may or may not still exist - * and be open. Return true if we were able to post. - */ - template - static bool postMaybe(weak_t target, CALLABLE&& callable) - { - return postMaybe(target, TimePoint::clock::now(), - std::forward(callable)); - } - /** * Launch a callable returning bool that will trigger repeatedly at * specified interval, until the callable returns false. @@ -182,8 +115,63 @@ namespace LL // Studio compile errors that seem utterly unrelated to this source // code. template - bool postTo(weak_t target, - const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback); + bool postTo(WorkQueue::weak_t target, + const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) + { + // We're being asked to post to the WorkQueue at target. + // target is a weak_ptr: have to lock it to check it. + auto tptr = target.lock(); + if (! tptr) + // can't post() if the target WorkQueue has been destroyed + return false; + + // Here we believe target WorkQueue still exists. Post to it a + // lambda that packages our callable, our callback and a weak_ptr + // to this originating WorkQueue. + tptr->post( + time, + [reply = super::getWeak(), + callable = std::move(callable), + callback = std::move(callback)] + () + { + // Call the callable in any case -- but to minimize + // copying the result, immediately bind it into a reply + // lambda. The reply lambda also binds the original + // callback, so that when we, the originating WorkQueue, + // finally receive and process the reply lambda, we'll + // call the bound callback with the bound result -- on the + // same thread that originally called postTo(). + auto rlambda = + [result = callable(), + callback = std::move(callback)] + () + { callback(std::move(result)); }; + // Check if this originating WorkQueue still exists. + // Remember, the outer lambda is now running on a thread + // servicing the target WorkQueue, and real time has + // elapsed since postTo()'s tptr->post() call. + // reply is a weak_ptr: have to lock it to check it. + auto rptr = reply.lock(); + if (rptr) + { + // Only post reply lambda if the originating WorkQueue + // still exists. If not -- who would we tell? Log it? + try + { + rptr->post(std::move(rlambda)); + } + catch (const Closed&) + { + // Originating WorkQueue might still exist, but + // might be Closed. Same thing: just discard the + // callback. + } + } + }); + // looks like we were able to post() + return true; + } /** * Post work to another WorkQueue, requesting a specific callback to @@ -193,36 +181,10 @@ namespace LL * inaccessible. */ template - bool postTo(weak_t target, CALLABLE&& callable, FOLLOWUP&& callback) + bool postTo(WorkQueue::weak_t target, + CALLABLE&& callable, FOLLOWUP&& callback) { - return postTo(target, TimePoint::clock::now(), - std::move(callable), std::move(callback)); - } - - /** - * Post work to another WorkQueue to be run at a specified time, - * blocking the calling coroutine until then, returning the result to - * caller on completion. - * - * In general, we assume that each thread's default coroutine is busy - * servicing its WorkQueue or whatever. To try to prevent mistakes, we - * forbid calling waitForResult() from a thread's default coroutine. - */ - template - auto waitForResult(const TimePoint& time, CALLABLE&& callable); - - /** - * Post work to another WorkQueue, blocking the calling coroutine - * until then, returning the result to caller on completion. - * - * In general, we assume that each thread's default coroutine is busy - * servicing its WorkQueue or whatever. To try to prevent mistakes, we - * forbid calling waitForResult() from a thread's default coroutine. - */ - template - auto waitForResult(CALLABLE&& callable) - { - return waitForResult(TimePoint::clock::now(), std::move(callable)); + return postTo(target, TimePoint::clock::now(), std::move(callable), std::move(callback)); } /*--------------------------- worker API ---------------------------*/ @@ -270,23 +232,6 @@ namespace LL bool runUntil(const TimePoint& until); private: - template - static auto makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback); - /// general case: arbitrary C++ return type - template - struct MakeReplyLambda; - /// specialize for CALLABLE returning void - template - struct MakeReplyLambda; - - /// general case: arbitrary C++ return type - template - struct WaitForResult; - /// specialize for CALLABLE returning void - template - struct WaitForResult; - - static void checkCoroutine(const std::string& method); static void error(const std::string& msg); static std::string makeName(const std::string& name); void callWork(const Queue::DataTuple& work); @@ -308,8 +253,8 @@ namespace LL { public: // bind the desired data - BackJack(weak_t target, - const TimePoint& start, + BackJack(WorkQueue::weak_t target, + const WorkQueue::TimePoint& start, const std::chrono::duration& interval, CALLABLE&& callable): mTarget(target), @@ -356,8 +301,8 @@ namespace LL } private: - weak_t mTarget; - TimePoint mStart; + WorkQueue::weak_t mTarget; + WorkQueue::TimePoint mStart; std::chrono::duration mInterval; CALLABLE mCallable; }; @@ -385,187 +330,6 @@ namespace LL getWeak(), TimePoint::clock::now(), interval, std::move(callable))); } - /// general case: arbitrary C++ return type - template - struct WorkQueue::MakeReplyLambda - { - auto operator()(CALLABLE&& callable, FOLLOWUP&& callback) - { - // Call the callable in any case -- but to minimize - // copying the result, immediately bind it into the reply - // lambda. The reply lambda also binds the original - // callback, so that when we, the originating WorkQueue, - // finally receive and process the reply lambda, we'll - // call the bound callback with the bound result -- on the - // same thread that originally called postTo(). - return - [result = std::forward(callable)(), - callback = std::move(callback)] - () - { callback(std::move(result)); }; - } - }; - - /// specialize for CALLABLE returning void - template - struct WorkQueue::MakeReplyLambda - { - auto operator()(CALLABLE&& callable, FOLLOWUP&& callback) - { - // Call the callable, which produces no result. - std::forward(callable)(); - // Our completion callback is simply the caller's callback. - return std::move(callback); - } - }; - - template - auto WorkQueue::makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback) - { - return MakeReplyLambda(callable)())>() - (std::move(callable), std::move(callback)); - } - - template - bool WorkQueue::postTo(weak_t target, - const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) - { - // We're being asked to post to the WorkQueue at target. - // target is a weak_ptr: have to lock it to check it. - auto tptr = target.lock(); - if (! tptr) - // can't post() if the target WorkQueue has been destroyed - return false; - - // Here we believe target WorkQueue still exists. Post to it a - // lambda that packages our callable, our callback and a weak_ptr - // to this originating WorkQueue. - tptr->post( - time, - [reply = super::getWeak(), - callable = std::move(callable), - callback = std::move(callback)] - () - { - // Use postMaybe() below in case this originating WorkQueue - // has been closed or destroyed. Remember, the outer lambda is - // now running on a thread servicing the target WorkQueue, and - // real time has elapsed since postTo()'s tptr->post() call. - try - { - // Make a reply lambda to repost to THIS WorkQueue. - // Delegate to makeReplyLambda() so we can partially - // specialize on void return. - postMaybe(reply, makeReplyLambda(std::move(callable), std::move(callback))); - } - catch (...) - { - // Either variant of makeReplyLambda() is responsible for - // calling the caller's callable. If that throws, return - // the exception to the originating thread. - postMaybe( - reply, - // Bind the current exception to transport back to the - // originating WorkQueue. Once there, rethrow it. - [exc = std::current_exception()](){ std::rethrow_exception(exc); }); - } - }); - - // looks like we were able to post() - return true; - } - - template - bool WorkQueue::postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable) - { - // target is a weak_ptr: have to lock it to check it - auto tptr = target.lock(); - if (tptr) - { - try - { - tptr->post(time, std::forward(callable)); - // we were able to post() - return true; - } - catch (const Closed&) - { - // target WorkQueue still exists, but is Closed - } - } - // either target no longer exists, or its WorkQueue is Closed - return false; - } - - /// general case: arbitrary C++ return type - template - struct WorkQueue::WaitForResult - { - auto operator()(WorkQueue* self, const TimePoint& time, CALLABLE&& callable) - { - LLCoros::Promise promise; - self->post( - time, - // We dare to bind a reference to Promise because it's - // specifically designed for cross-thread communication. - [&promise, callable = std::move(callable)]() - { - try - { - // call the caller's callable and trigger promise with result - promise.set_value(callable()); - } - catch (...) - { - promise.set_exception(std::current_exception()); - } - }); - auto future{ LLCoros::getFuture(promise) }; - // now, on the calling thread, wait for that result - LLCoros::TempStatus st("waiting for WorkQueue::waitForResult()"); - return future.get(); - } - }; - - /// specialize for CALLABLE returning void - template - struct WorkQueue::WaitForResult - { - void operator()(WorkQueue* self, const TimePoint& time, CALLABLE&& callable) - { - LLCoros::Promise promise; - self->post( - time, - // &promise is designed for cross-thread access - [&promise, callable = std::move(callable)]() - { - try - { - callable(); - promise.set_value(); - } - catch (...) - { - promise.set_exception(std::current_exception()); - } - }); - auto future{ LLCoros::getFuture(promise) }; - // block until set_value() - LLCoros::TempStatus st("waiting for void WorkQueue::waitForResult()"); - future.get(); - } - }; - - template - auto WorkQueue::waitForResult(const TimePoint& time, CALLABLE&& callable) - { - checkCoroutine("waitForResult()"); - // derive callable's return type so we can specialize for void - return WaitForResult(callable)())>() - (this, time, std::forward(callable)); - } - } // namespace LL #endif /* ! defined(LL_WORKQUEUE_H) */ diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 1b6920fe3b..cbc5392882 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -172,19 +172,31 @@ BOOL is_little_endian() return (*c == 0x78) ; } +LLImageGLThread* LLImageGLThread::sInstance = nullptr; + //static void LLImageGL::initClass(LLWindow* window, S32 num_catagories, BOOL skip_analyze_alpha /* = false */) { LL_PROFILE_ZONE_SCOPED; sSkipAnalyzeAlpha = skip_analyze_alpha; - LLImageGLThread::createInstance(window); + LLImageGLThread::sInstance = new LLImageGLThread(window); + LLImageGLThread::sInstance->start(); +} + +//static +void LLImageGL::updateClass() +{ + LL_PROFILE_ZONE_SCOPED; + LLImageGLThread::sInstance->executeCallbacks(); } //static void LLImageGL::cleanupClass() { LL_PROFILE_ZONE_SCOPED; - LLImageGLThread::deleteSingleton(); + LLImageGLThread::sInstance->mFunctionQueue.close(); + delete LLImageGLThread::sInstance; + LLImageGLThread::sInstance = nullptr; } //static @@ -492,9 +504,6 @@ void LLImageGL::init(BOOL usemipmaps) #endif mCategory = -1; - - // Sometimes we have to post work for the main thread. - mMainQueue = LL::WorkQueue::getInstance("mainloop"); } void LLImageGL::cleanup() @@ -1527,7 +1536,8 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_ } //if we're on the image loading thread, be sure to delete old_texname and update mTexName on the main thread - if (! on_main_thread()) + if (LLImageGLThread::sInstance != nullptr && + LLThread::currentID() == LLImageGLThread::sInstance->getID()) { { LL_PROFILE_ZONE_NAMED("cglt - sync"); @@ -1544,9 +1554,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_ } ref(); - LL::WorkQueue::postMaybe( - mMainQueue, - [=]() + LLImageGLThread::sInstance->postCallback([=]() { LL_PROFILE_ZONE_NAMED("cglt - delete callback"); if (old_texname != 0) @@ -2251,24 +2259,73 @@ void LLImageGL::resetCurTexSizebar() */ LLImageGLThread::LLImageGLThread(LLWindow* window) - // We want exactly one thread, but a very large capacity: we never want - // anyone, especially inner-loop render code, to have to block on post() - // because we're full. - : ThreadPool("LLImageGL", 1, 1024*1024) - , mWindow(window) + : LLThread("LLImageGL"), mWindow(window) { mFinished = false; mContext = mWindow->createSharedContext(); } +// post a function to be executed on the LLImageGL background thread + +bool LLImageGLThread::post(const std::function& func) +{ + try + { + mFunctionQueue.post(func); + } + catch (LLThreadSafeQueueInterrupt e) + { + return false; + } + + return true; +} + +//post a callback to be executed on the main thread + +bool LLImageGLThread::postCallback(const std::function& callback) +{ + try + { + if (!mCallbackQueue.tryPost(callback)) + { + mPendingCallbackQ.push(callback); + } + } + catch (LLThreadSafeQueueInterrupt e) + { + //thread is closing, drop request + return false; + } + + return true; +} + +void LLImageGLThread::executeCallbacks() +{ + LL_PROFILE_ZONE_SCOPED; + //executed from main thread + mCallbackQueue.runPending(); + + while (!mPendingCallbackQ.empty()) + { + if (mCallbackQueue.tryPost(mPendingCallbackQ.front())) + { + mPendingCallbackQ.pop(); + } + else + { + break; + } + } +} + void LLImageGLThread::run() { - // We must perform setup on this thread before actually servicing our - // WorkQueue, likewise cleanup afterwards. mWindow->makeContextCurrent(mContext); gGL.init(); - ThreadPool::run(); + mFunctionQueue.runUntilClose(); gGL.shutdown(); mWindow->destroySharedContext(mContext); } diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h index 27496def1d..8264e4a5f2 100644 --- a/indra/llrender/llimagegl.h +++ b/indra/llrender/llimagegl.h @@ -37,7 +37,6 @@ #include "llunits.h" #include "llthreadsafequeue.h" #include "llrender.h" -#include "threadpool.h" #include "workqueue.h" class LLTextureAtlas ; @@ -199,7 +198,6 @@ private: void freePickMask(); LLPointer mSaveData; // used for destroyGL/restoreGL - LL::WorkQueue::weak_t mMainQueue; U8* mPickMask; //downsampled bitmap approximation of alpha channel. NULL if no alpha channel U16 mPickMaskWidth; U16 mPickMaskHeight; @@ -273,6 +271,7 @@ public: public: static void initClass(LLWindow* window, S32 num_catagories, BOOL skip_analyze_alpha = false); + static void updateClass(); static void cleanupClass() ; private: @@ -308,24 +307,34 @@ public: }; -class LLImageGLThread : public LLSimpleton, LL::ThreadPool +class LLImageGLThread : public LLThread { public: LLImageGLThread(LLWindow* window); // post a function to be executed on the LLImageGL background thread - template - bool post(CALLABLE&& func) - { - return getQueue().postIfOpen(std::forward(func)); - } + bool post(const std::function& func); + + //post a callback to be executed on the main thread + bool postCallback(const std::function& callback); + + void executeCallbacks(); void run() override; -private: + // Work Queue for background thread + LL::WorkQueue mFunctionQueue; + + // Work Queue for main thread (run from updateClass) + LL::WorkQueue mCallbackQueue; + LLWindow* mWindow; void* mContext; LLAtomicBool mFinished; + + std::queue> mPendingCallbackQ; + + static LLImageGLThread* sInstance; }; diff --git a/indra/llui/CMakeLists.txt b/indra/llui/CMakeLists.txt index 55c1655d7b..f781ff4110 100644 --- a/indra/llui/CMakeLists.txt +++ b/indra/llui/CMakeLists.txt @@ -308,10 +308,6 @@ if(LL_TESTS) ${BOOST_FIBER_LIBRARY} ${BOOST_CONTEXT_LIBRARY} ${BOOST_SYSTEM_LIBRARY} ${WINDOWS_LIBRARIES}) if(NOT LINUX) - if(WINDOWS) - LL_ADD_INTEGRATION_TEST(llurlentry llurlentry.cpp "imm32;${test_libs}") - else(WINDOWS) - LL_ADD_INTEGRATION_TEST(llurlentry llurlentry.cpp "${test_libs}") - endif(WINDOWS) + LL_ADD_INTEGRATION_TEST(llurlentry llurlentry.cpp "${test_libs}") endif(NOT LINUX) endif(LL_TESTS) diff --git a/indra/llui/llviewereventrecorder.cpp b/indra/llui/llviewereventrecorder.cpp index cb000aef74..5a44ec947a 100644 --- a/indra/llui/llviewereventrecorder.cpp +++ b/indra/llui/llviewereventrecorder.cpp @@ -28,6 +28,8 @@ #include "llui.h" #include "llleap.h" +LLViewerEventRecorder* LLSimpleton::sInstance = nullptr; + LLViewerEventRecorder::LLViewerEventRecorder() { clear(UNDEFINED); diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index 3f3dd43daf..e52624d66a 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -55,8 +55,6 @@ #include #include #include -#include -#include // std::pair // Require DirectInput version 8 #define DIRECTINPUT_VERSION 0x0800 @@ -176,19 +174,23 @@ DWORD LLWindowWin32::sWinIMESentenceMode = IME_SMODE_AUTOMATIC; LLCoordWindow LLWindowWin32::sWinIMEWindowPosition(-1,-1); // The following class LLWinImm delegates Windows IMM APIs. -// It was originally introduced to support US Windows XP, on which we needed -// to dynamically load IMM32.DLL and use GetProcAddress to resolve its entry -// points. Now that that's moot, we retain this wrapper only for hooks for -// metrics. +// We need this because some language versions of Windows, +// e.g., US version of Windows XP, doesn't install IMM32.DLL +// as a default, and we can't link against imm32.lib statically. +// I believe DLL loading of this type is best suited to do +// in a static initialization of a class. What I'm not sure is +// whether it follows the Linden Conding Standard... +// See http://wiki.secondlife.com/wiki/Coding_standards#Static_Members class LLWinImm { public: - static bool isAvailable() { return true; } + static bool isAvailable() { return sTheInstance.mHImmDll != NULL; } public: // Wrappers for IMM API. static BOOL isIME(HKL hkl); + static HWND getDefaultIMEWnd(HWND hwnd); static HIMC getContext(HWND hwnd); static BOOL releaseContext(HWND hwnd, HIMC himc); static BOOL getOpenStatus(HIMC himc); @@ -202,96 +204,236 @@ public: static BOOL setCompositionFont(HIMC himc, LPLOGFONTW logfont); static BOOL setCandidateWindow(HIMC himc, LPCANDIDATEFORM candidate_form); static BOOL notifyIME(HIMC himc, DWORD action, DWORD index, DWORD value); + +private: + LLWinImm(); + ~LLWinImm(); + +private: + // Pointers to IMM API. + BOOL (WINAPI *mImmIsIME)(HKL); + HWND (WINAPI *mImmGetDefaultIMEWnd)(HWND); + HIMC (WINAPI *mImmGetContext)(HWND); + BOOL (WINAPI *mImmReleaseContext)(HWND, HIMC); + BOOL (WINAPI *mImmGetOpenStatus)(HIMC); + BOOL (WINAPI *mImmSetOpenStatus)(HIMC, BOOL); + BOOL (WINAPI *mImmGetConversionStatus)(HIMC, LPDWORD, LPDWORD); + BOOL (WINAPI *mImmSetConversionStatus)(HIMC, DWORD, DWORD); + BOOL (WINAPI *mImmGetCompostitionWindow)(HIMC, LPCOMPOSITIONFORM); + BOOL (WINAPI *mImmSetCompostitionWindow)(HIMC, LPCOMPOSITIONFORM); + LONG (WINAPI *mImmGetCompositionString)(HIMC, DWORD, LPVOID, DWORD); + BOOL (WINAPI *mImmSetCompositionString)(HIMC, DWORD, LPVOID, DWORD, LPVOID, DWORD); + BOOL (WINAPI *mImmSetCompositionFont)(HIMC, LPLOGFONTW); + BOOL (WINAPI *mImmSetCandidateWindow)(HIMC, LPCANDIDATEFORM); + BOOL (WINAPI *mImmNotifyIME)(HIMC, DWORD, DWORD, DWORD); + +private: + HMODULE mHImmDll; + static LLWinImm sTheInstance; }; +LLWinImm LLWinImm::sTheInstance; + +LLWinImm::LLWinImm() : mHImmDll(NULL) +{ + // Check system metrics + if ( !GetSystemMetrics( SM_IMMENABLED ) ) + return; + + mHImmDll = LoadLibraryA("Imm32"); + if (mHImmDll != NULL) + { + mImmIsIME = (BOOL (WINAPI *)(HKL)) GetProcAddress(mHImmDll, "ImmIsIME"); + mImmGetDefaultIMEWnd = (HWND (WINAPI *)(HWND)) GetProcAddress(mHImmDll, "ImmGetDefaultIMEWnd"); + mImmGetContext = (HIMC (WINAPI *)(HWND)) GetProcAddress(mHImmDll, "ImmGetContext"); + mImmReleaseContext = (BOOL (WINAPI *)(HWND, HIMC)) GetProcAddress(mHImmDll, "ImmReleaseContext"); + mImmGetOpenStatus = (BOOL (WINAPI *)(HIMC)) GetProcAddress(mHImmDll, "ImmGetOpenStatus"); + mImmSetOpenStatus = (BOOL (WINAPI *)(HIMC, BOOL)) GetProcAddress(mHImmDll, "ImmSetOpenStatus"); + mImmGetConversionStatus = (BOOL (WINAPI *)(HIMC, LPDWORD, LPDWORD)) GetProcAddress(mHImmDll, "ImmGetConversionStatus"); + mImmSetConversionStatus = (BOOL (WINAPI *)(HIMC, DWORD, DWORD)) GetProcAddress(mHImmDll, "ImmSetConversionStatus"); + mImmGetCompostitionWindow = (BOOL (WINAPI *)(HIMC, LPCOMPOSITIONFORM)) GetProcAddress(mHImmDll, "ImmGetCompositionWindow"); + mImmSetCompostitionWindow = (BOOL (WINAPI *)(HIMC, LPCOMPOSITIONFORM)) GetProcAddress(mHImmDll, "ImmSetCompositionWindow"); + mImmGetCompositionString= (LONG (WINAPI *)(HIMC, DWORD, LPVOID, DWORD)) GetProcAddress(mHImmDll, "ImmGetCompositionStringW"); + mImmSetCompositionString= (BOOL (WINAPI *)(HIMC, DWORD, LPVOID, DWORD, LPVOID, DWORD)) GetProcAddress(mHImmDll, "ImmSetCompositionStringW"); + mImmSetCompositionFont = (BOOL (WINAPI *)(HIMC, LPLOGFONTW)) GetProcAddress(mHImmDll, "ImmSetCompositionFontW"); + mImmSetCandidateWindow = (BOOL (WINAPI *)(HIMC, LPCANDIDATEFORM)) GetProcAddress(mHImmDll, "ImmSetCandidateWindow"); + mImmNotifyIME = (BOOL (WINAPI *)(HIMC, DWORD, DWORD, DWORD)) GetProcAddress(mHImmDll, "ImmNotifyIME"); + + if (mImmIsIME == NULL || + mImmGetDefaultIMEWnd == NULL || + mImmGetContext == NULL || + mImmReleaseContext == NULL || + mImmGetOpenStatus == NULL || + mImmSetOpenStatus == NULL || + mImmGetConversionStatus == NULL || + mImmSetConversionStatus == NULL || + mImmGetCompostitionWindow == NULL || + mImmSetCompostitionWindow == NULL || + mImmGetCompositionString == NULL || + mImmSetCompositionString == NULL || + mImmSetCompositionFont == NULL || + mImmSetCandidateWindow == NULL || + mImmNotifyIME == NULL) + { + // If any of the above API entires are not found, we can't use IMM API. + // So, turn off the IMM support. We should log some warning message in + // the case, since it is very unusual; these APIs are available from + // the beginning, and all versions of IMM32.DLL should have them all. + // Unfortunately, this code may be executed before initialization of + // the logging channel (LL_WARNS()), and we can't do it here... Yes, this + // is one of disadvantages to use static constraction to DLL loading. + FreeLibrary(mHImmDll); + mHImmDll = NULL; + + // If we unload the library, make sure all the function pointers are cleared + mImmIsIME = NULL; + mImmGetDefaultIMEWnd = NULL; + mImmGetContext = NULL; + mImmReleaseContext = NULL; + mImmGetOpenStatus = NULL; + mImmSetOpenStatus = NULL; + mImmGetConversionStatus = NULL; + mImmSetConversionStatus = NULL; + mImmGetCompostitionWindow = NULL; + mImmSetCompostitionWindow = NULL; + mImmGetCompositionString = NULL; + mImmSetCompositionString = NULL; + mImmSetCompositionFont = NULL; + mImmSetCandidateWindow = NULL; + mImmNotifyIME = NULL; + } + } +} + + // static BOOL LLWinImm::isIME(HKL hkl) { - return ImmIsIME(hkl); + if ( sTheInstance.mImmIsIME ) + return sTheInstance.mImmIsIME(hkl); + return FALSE; } // static HIMC LLWinImm::getContext(HWND hwnd) { - return ImmGetContext(hwnd); + if ( sTheInstance.mImmGetContext ) + return sTheInstance.mImmGetContext(hwnd); + return 0; } //static BOOL LLWinImm::releaseContext(HWND hwnd, HIMC himc) { - return ImmReleaseContext(hwnd, himc); + if ( sTheInstance.mImmIsIME ) + return sTheInstance.mImmReleaseContext(hwnd, himc); + return FALSE; } // static BOOL LLWinImm::getOpenStatus(HIMC himc) { - return ImmGetOpenStatus(himc); + if ( sTheInstance.mImmGetOpenStatus ) + return sTheInstance.mImmGetOpenStatus(himc); + return FALSE; } // static BOOL LLWinImm::setOpenStatus(HIMC himc, BOOL status) { - return ImmSetOpenStatus(himc, status); + if ( sTheInstance.mImmSetOpenStatus ) + return sTheInstance.mImmSetOpenStatus(himc, status); + return FALSE; } // static BOOL LLWinImm::getConversionStatus(HIMC himc, LPDWORD conversion, LPDWORD sentence) { - return ImmGetConversionStatus(himc, conversion, sentence); + if ( sTheInstance.mImmGetConversionStatus ) + return sTheInstance.mImmGetConversionStatus(himc, conversion, sentence); + return FALSE; } // static BOOL LLWinImm::setConversionStatus(HIMC himc, DWORD conversion, DWORD sentence) { - return ImmSetConversionStatus(himc, conversion, sentence); + if ( sTheInstance.mImmSetConversionStatus ) + return sTheInstance.mImmSetConversionStatus(himc, conversion, sentence); + return FALSE; } // static BOOL LLWinImm::getCompositionWindow(HIMC himc, LPCOMPOSITIONFORM form) { - return ImmGetCompositionWindow(himc, form); + if ( sTheInstance.mImmGetCompostitionWindow ) + return sTheInstance.mImmGetCompostitionWindow(himc, form); + return FALSE; } // static BOOL LLWinImm::setCompositionWindow(HIMC himc, LPCOMPOSITIONFORM form) { - return ImmSetCompositionWindow(himc, form); + if ( sTheInstance.mImmSetCompostitionWindow ) + return sTheInstance.mImmSetCompostitionWindow(himc, form); + return FALSE; } // static LONG LLWinImm::getCompositionString(HIMC himc, DWORD index, LPVOID data, DWORD length) { - return ImmGetCompositionString(himc, index, data, length); + if ( sTheInstance.mImmGetCompositionString ) + return sTheInstance.mImmGetCompositionString(himc, index, data, length); + return FALSE; } // static BOOL LLWinImm::setCompositionString(HIMC himc, DWORD index, LPVOID pComp, DWORD compLength, LPVOID pRead, DWORD readLength) { - return ImmSetCompositionString(himc, index, pComp, compLength, pRead, readLength); + if ( sTheInstance.mImmSetCompositionString ) + return sTheInstance.mImmSetCompositionString(himc, index, pComp, compLength, pRead, readLength); + return FALSE; } // static BOOL LLWinImm::setCompositionFont(HIMC himc, LPLOGFONTW pFont) { - return ImmSetCompositionFont(himc, pFont); + if ( sTheInstance.mImmSetCompositionFont ) + return sTheInstance.mImmSetCompositionFont(himc, pFont); + return FALSE; } // static BOOL LLWinImm::setCandidateWindow(HIMC himc, LPCANDIDATEFORM form) { - return ImmSetCandidateWindow(himc, form); + if ( sTheInstance.mImmSetCandidateWindow ) + return sTheInstance.mImmSetCandidateWindow(himc, form); + return FALSE; } // static BOOL LLWinImm::notifyIME(HIMC himc, DWORD action, DWORD index, DWORD value) { - return ImmNotifyIME(himc, action, index, value); + if ( sTheInstance.mImmNotifyIME ) + return sTheInstance.mImmNotifyIME(himc, action, index, value); + return FALSE; } + +// ---------------------------------------------------------------------------------------- +LLWinImm::~LLWinImm() +{ + if (mHImmDll != NULL) + { + FreeLibrary(mHImmDll); + mHImmDll = NULL; + } +} + + class LLMonitorInfo { public: @@ -326,32 +468,6 @@ private: static LLMonitorInfo sMonitorInfo; -// Thread that owns the Window Handle -// This whole struct is private to LLWindowWin32, which needs to mess with its -// members, which is why it's a struct rather than a class. In effect, we make -// the containing class a friend. -struct LLWindowWin32::LLWindowWin32Thread : public LL::ThreadPool -{ - static const int MAX_QUEUE_SIZE = 2048; - - LLThreadSafeQueue mMessageQueue; - - LLWindowWin32Thread(); - - void run() override; - - template - void post(CALLABLE&& func) - { - getQueue().post(std::forward(func)); - } - - // call PeekMessage and pull enqueue messages for later processing - void gatherInput(); - HWND mWindowHandle = NULL; - HDC mhDC = 0; -}; - LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, const std::string& title, const std::string& name, S32 x, S32 y, S32 width, @@ -363,7 +479,8 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, : LLWindow(callbacks, fullscreen, flags) { sMainThreadId = LLThread::currentID(); - mWindowThread = new LLWindowWin32Thread(); + mWindowThread = new LLWindowWin32Thread(this); + mWindowThread->start(); //MAINT-516 -- force a load of opengl32.dll just in case windows went sideways LoadLibrary(L"opengl32.dll"); @@ -434,6 +551,7 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, // Make an instance of our window then define the window class mhInstance = GetModuleHandle(NULL); + mWndProc = NULL; // Init Direct Input - needed for joystick / Spacemouse @@ -857,13 +975,17 @@ void LLWindowWin32::close() // Something killed the window while we were busy destroying gl or handle somehow got broken LL_WARNS("Window") << "Failed to destroy Window, invalid handle!" << LL_ENDL; } + mWindowHandle = NULL; + mWindowThread->mFinished = true; }); - // Even though the above lambda might not yet have run, we've already - // bound mWindowHandle into it by value, which should suffice for the - // operations we're asking. That's the last time WE should touch it. - mWindowHandle = NULL; - mWindowThread->close(); + + while (!mWindowThread->isStopped()) + { + //nudge window thread + PostMessage(mWindowHandle, WM_USER + 0x0017, 0xB0B0, 0x1337); + std::this_thread::sleep_for(std::chrono::milliseconds(1)); + } } BOOL LLWindowWin32::isValid() @@ -1156,7 +1278,51 @@ BOOL LLWindowWin32::switchContext(BOOL fullscreen, const LLCoordScreen& size, BO << " Fullscreen: " << mFullscreen << LL_ENDL; - recreateWindow(window_rect, dw_ex_style, dw_style); + auto oldHandle = mWindowHandle; + + //zero out mWindowHandle and mhDC before destroying window so window thread falls back to peekmessage + mWindowHandle = 0; + mhDC = 0; + + if (oldHandle && !destroy_window_handler(oldHandle)) + { + LL_WARNS("Window") << "Failed to properly close window before recreating it!" << LL_ENDL; + } + + mWindowHandle = NULL; + mhDC = 0; + + mWindowThread->post( + [this, window_rect, dw_ex_style, dw_style]() + { + mWindowHandle = CreateWindowEx(dw_ex_style, + mWindowClassName, + mWindowTitle, + WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, + window_rect.left, // x pos + window_rect.top, // y pos + window_rect.right - window_rect.left, // width + window_rect.bottom - window_rect.top, // height + NULL, + NULL, + mhInstance, + NULL); + + if (mWindowHandle) + { + mhDC = GetDC(mWindowHandle); + } + } + ); + + // HACK wait for above handle to become populated + // TODO: use a future + int count = 1024; + while (!mhDC && count > 0) + { + Sleep(10); + --count; + } if (mWindowHandle) { @@ -1484,7 +1650,48 @@ const S32 max_format = (S32)num_formats - 1; mhDC = 0; // Zero The Device Context } - recreateWindow(window_rect, dw_ex_style, dw_style); + auto oldHandle = mWindowHandle; + mWindowHandle = NULL; + mhDC = 0; + + // Destroy The Window + if (oldHandle && !destroy_window_handler(oldHandle)) + { + LL_WARNS("Window") << "Failed to properly close window!" << LL_ENDL; + } + + mWindowThread->post( + [this, window_rect, dw_ex_style, dw_style]() + { + mWindowHandle = CreateWindowEx(dw_ex_style, + mWindowClassName, + mWindowTitle, + WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, + window_rect.left, // x pos + window_rect.top, // y pos + window_rect.right - window_rect.left, // width + window_rect.bottom - window_rect.top, // height + NULL, + NULL, + mhInstance, + NULL); + + if (mWindowHandle) + { + mhDC = GetDC(mWindowHandle); + } + } + ); + + // HACK wait for above handle to become populated + // TODO: use a future + int count = 1024; + while (!mhDC && count > 0) + { + PostMessage(oldHandle, WM_USER + 8, 0x1717, 0x3b3b); + Sleep(10); + --count; + } if (mWindowHandle) { @@ -1621,64 +1828,6 @@ const S32 max_format = (S32)num_formats - 1; return TRUE; } -void LLWindowWin32::recreateWindow(RECT window_rect, DWORD dw_ex_style, DWORD dw_style) -{ - auto oldHandle = mWindowHandle; - - // zero out mWindowHandle and mhDC before destroying window so window - // thread falls back to peekmessage - mWindowHandle = 0; - mhDC = 0; - - if (oldHandle && !destroy_window_handler(oldHandle)) - { - LL_WARNS("Window") << "Failed to properly close window before recreating it!" << LL_ENDL; - } - - std::promise> promise; - mWindowThread->post( - [this, window_rect, dw_ex_style, dw_style, &promise]() - { - auto handle = CreateWindowEx(dw_ex_style, - mWindowClassName, - mWindowTitle, - WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, - window_rect.left, // x pos - window_rect.top, // y pos - window_rect.right - window_rect.left, // width - window_rect.bottom - window_rect.top, // height - NULL, - NULL, - mhInstance, - NULL); - - if (! handle) - { - // Failed to create window: clear the variables. This - // assignment is valid because we're running on mWindowThread. - mWindowThread->mWindowHandle = NULL; - mWindowThread->mhDC = 0; - } - else - { - // Update mWindowThread's own mWindowHandle and mhDC. - mWindowThread->mWindowHandle = handle; - mWindowThread->mhDC = GetDC(handle); - } - - // It's important to wake up the future either way. - promise.set_value(std::make_pair(mWindowThread->mWindowHandle, mWindowThread->mhDC)); - } - ); - - auto future = promise.get_future(); - // This blocks until mWindowThread processes CreateWindowEx() and calls - // promise.set_value(). - auto pair = future.get(); - mWindowHandle = pair.first; - mhDC = pair.second; -} - void* LLWindowWin32::createSharedContext() { S32 attribs[] = @@ -2032,14 +2181,12 @@ void LLWindowWin32::gatherInput() } - if (mWindowThread->getQueue().size()) + if (mWindowThread->mFunctionQueue.size() > 0) { LL_PROFILE_ZONE_NAMED("gi - PostMessage"); if (mWindowHandle) - { - // post a nonsense user message to wake up the Window Thread in - // case any functions are pending and no windows events came - // through this frame + { // post a nonsense user message to wake up the Window Thread in case any functions are pending + // and no windows events came through this frame PostMessage(mWindowHandle, WM_USER + 0x0017, 0xB0B0, 0x1337); } } @@ -2129,6 +2276,17 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ if (NULL != window_imp) { + // Has user provided their own window callback? + if (NULL != window_imp->mWndProc) + { + LL_PROFILE_ZONE_NAMED("mwp - WndProc"); + if (!window_imp->mWndProc(h_wnd, u_msg, w_param, l_param)) + { + // user has handled window message + return 0; + } + } + // Juggle to make sure we can get negative positions for when // mouse is outside window. LLCoordWindow window_coord((S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param)); @@ -4409,32 +4567,35 @@ std::vector LLWindowWin32::getDynamicFallbackFontList() #endif // LL_WINDOWS -inline LLWindowWin32::LLWindowWin32Thread::LLWindowWin32Thread() - : ThreadPool("Window Thread", 1, MAX_QUEUE_SIZE) +inline LLWindowWin32Thread::LLWindowWin32Thread(LLWindowWin32* window) + : LLThread("Window Thread"), + mWindow(window), + mFunctionQueue(MAX_QUEUE_SIZE) { + } -void LLWindowWin32::LLWindowWin32Thread::run() +inline void LLWindowWin32Thread::run() { - sWindowThreadId = std::this_thread::get_id(); - while (! getQueue().done()) + sWindowThreadId = getID(); + while (!mFinished) { LL_PROFILE_ZONE_SCOPED; - if (mWindowHandle != 0) + if (mWindow && mWindow->mWindowHandle != 0) { MSG msg; BOOL status; - if (mhDC == 0) + if (mWindow->mhDC == 0) { LL_PROFILE_ZONE_NAMED("w32t - PeekMessage"); - status = PeekMessage(&msg, mWindowHandle, 0, 0, PM_REMOVE); + status = PeekMessage(&msg, mWindow->mWindowHandle, 0, 0, PM_REMOVE); } else { LL_PROFILE_ZONE_NAMED("w32t - GetMessage"); - status = GetMessage(&msg, mWindowHandle, 0, 0); + status = GetMessage(&msg, mWindow->mWindowHandle, 0, 0); } if (status > 0) { @@ -4448,7 +4609,11 @@ void LLWindowWin32::LLWindowWin32Thread::run() { LL_PROFILE_ZONE_NAMED("w32t - Function Queue"); //process any pending functions - getQueue().runPending(); + std::function curFunc; + while (mFunctionQueue.tryPopBack(curFunc)) + { + curFunc(); + } } #if 0 @@ -4460,6 +4625,11 @@ void LLWindowWin32::LLWindowWin32Thread::run() } } +void LLWindowWin32Thread::post(const std::function& func) +{ + mFunctionQueue.pushFront(func); +} + void LLWindowWin32::post(const std::function& func) { mFunctionQueue.pushFront(func); diff --git a/indra/llwindow/llwindowwin32.h b/indra/llwindow/llwindowwin32.h index 5966061177..d082080807 100644 --- a/indra/llwindow/llwindowwin32.h +++ b/indra/llwindow/llwindowwin32.h @@ -36,12 +36,44 @@ #include "llthread.h" #include "llthreadsafequeue.h" #include "llmutex.h" -#include "workqueue.h" // Hack for async host by name #define LL_WM_HOST_RESOLVED (WM_APP + 1) typedef void (*LLW32MsgCallback)(const MSG &msg); +class LLWindowWin32; + +// Thread that owns the Window Handle +class LLWindowWin32Thread : public LLThread +{ +public: + class Message + { + public: + LRESULT mMsg; + }; + + static const int MAX_QUEUE_SIZE = 2048; + + LLThreadSafeQueue mMessageQueue; + LLThreadSafeQueue> mFunctionQueue; + + bool mFinished = false; + + LLWindowWin32Thread(LLWindowWin32* window); + + void run() override; + + void post(const std::function& func); + +private: + + // call PeekMessage and pull enqueue messages for later processing + void gatherInput(); + LLWindowWin32* mWindow = nullptr; + +}; + class LLWindowWin32 : public LLWindow { public: @@ -186,6 +218,7 @@ protected: HGLRC mhRC = 0; // OpenGL rendering context HDC mhDC = 0; // Windows Device context handle HINSTANCE mhInstance; // handle to application instance + WNDPROC mWndProc; // user-installable window proc RECT mOldMouseClip; // Screen rect to which the mouse cursor was globally constrained before we changed it in clipMouse() WPARAM mLastSizeWParam; F32 mOverrideAspectRatio; @@ -237,15 +270,14 @@ protected: BOOL mMouseVanish; - struct LLWindowWin32Thread; - LLWindowWin32Thread* mWindowThread = nullptr; - LLThreadSafeQueue> mFunctionQueue; - LLThreadSafeQueue> mMouseQueue; - void post(const std::function& func); - void postMouseButtonEvent(const std::function& func); - void recreateWindow(RECT window_rect, DWORD dw_ex_style, DWORD dw_style); + LLWindowWin32Thread* mWindowThread = nullptr; + LLThreadSafeQueue> mFunctionQueue; + LLThreadSafeQueue> mMouseQueue; + void post(const std::function& func); + void postMouseButtonEvent(const std::function& func); friend class LLWindowManager; + friend class LLWindowWin32Thread; }; class LLSplashScreenWin32 : public LLSplashScreen diff --git a/indra/newview/CMakeLists.txt b/indra/newview/CMakeLists.txt index 6d090be33a..631089f6ce 100644 --- a/indra/newview/CMakeLists.txt +++ b/indra/newview/CMakeLists.txt @@ -393,6 +393,7 @@ set(viewer_SOURCE_FILES llloginhandler.cpp lllogininstance.cpp llmachineid.cpp + llmainlooprepeater.cpp llmanip.cpp llmaniprotate.cpp llmanipscale.cpp @@ -1031,6 +1032,7 @@ set(viewer_HEADER_FILES llloginhandler.h lllogininstance.h llmachineid.h + llmainlooprepeater.h llmanip.h llmaniprotate.h llmanipscale.h @@ -1602,7 +1604,6 @@ if (WINDOWS) ${WINDOWS_LIBRARIES} comdlg32 dxguid - imm32 kernel32 odbc32 odbccp32 diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index 607e531e62..f15b5d0981 100644 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -3858,17 +3858,6 @@ Value 1 - MainWorkTime - - Comment - Max time per frame devoted to mainloop work queue (in milliseconds) - Persist - 1 - Type - F32 - Value - 0.1 - QueueInventoryFetchTimeout Comment @@ -12674,20 +12663,6 @@ Value 50 - ThreadPoolSizes - - Comment - Map of size overrides for specific thread pools. - Persist - 1 - Type - LLSD - Value - - General - 4 - - ThrottleBandwidthKBPS Comment diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 93e5c2e341..89756d0881 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -233,12 +233,11 @@ #include "llavatariconctrl.h" #include "llgroupiconctrl.h" #include "llviewerassetstats.h" -#include "workqueue.h" -using namespace LL; // Include for security api initialization #include "llsecapi.h" #include "llmachineid.h" +#include "llmainlooprepeater.h" #include "llcleanup.h" #include "llcoproceduremanager.h" @@ -367,10 +366,6 @@ BOOL gLogoutInProgress = FALSE; BOOL gSimulateMemLeak = FALSE; -// We don't want anyone, especially threads working on the graphics pipeline, -// to have to block due to this WorkQueue being full. -WorkQueue gMainloopWork("mainloop", 1024*1024); - //////////////////////////////////////////////////////////// // Internal globals... that should be removed. static std::string gArgs; @@ -386,6 +381,42 @@ static std::string gLaunchFileOnQuit; // Used on Win32 for other apps to identify our window (eg, win_setup) const char* const VIEWER_WINDOW_CLASSNAME = "Second Life"; +//-- LLDeferredTaskList ------------------------------------------------------ + +/** + * A list of deferred tasks. + * + * We sometimes need to defer execution of some code until the viewer gets idle, + * e.g. removing an inventory item from within notifyObservers() may not work out. + * + * Tasks added to this list will be executed in the next LLAppViewer::idle() iteration. + * All tasks are executed only once. + */ +class LLDeferredTaskList: public LLSingleton +{ + LLSINGLETON_EMPTY_CTOR(LLDeferredTaskList); + LOG_CLASS(LLDeferredTaskList); + + friend class LLAppViewer; + typedef boost::signals2::signal signal_t; + + void addTask(const signal_t::slot_type& cb) + { + mSignal.connect(cb); + } + + void run() + { + if (!mSignal.empty()) + { + mSignal(); + mSignal.disconnect_all_slots(); + } + } + + signal_t mSignal; +}; + //---------------------------------------------------------------------------- // List of entries from strings.xml to always replace @@ -943,6 +974,9 @@ bool LLAppViewer::init() } LL_INFOS("InitInfo") << "Cache initialization is done." << LL_ENDL ; + // Initialize the repeater service. + LLMainLoopRepeater::instance().start(); + // Initialize event recorder LLViewerEventRecorder::createInstance(); @@ -2158,6 +2192,8 @@ bool LLAppViewer::cleanup() SUBSYSTEM_CLEANUP(LLProxy); LLCore::LLHttp::cleanup(); + LLMainLoopRepeater::instance().stop(); + ll_close_fail_log(); LLError::LLCallStacks::cleanup(); @@ -4452,7 +4488,7 @@ bool LLAppViewer::initCache() void LLAppViewer::addOnIdleCallback(const boost::function& cb) { - gMainloopWork.post(cb); + LLDeferredTaskList::instance().addTask(cb); } void LLAppViewer::loadKeyBindings() @@ -4850,6 +4886,7 @@ void LLAppViewer::idle() LLNotificationsUI::LLToast::updateClass(); LLSmoothInterpolation::updateInterpolants(); LLMortician::updateClass(); + LLImageGL::updateClass(); LLFilePickerThread::clearDead(); //calls LLFilePickerThread::notify() LLDirPickerThread::clearDead(); F32 dt_raw = idle_timer.getElapsedTimeAndResetF32(); @@ -5226,19 +5263,8 @@ void LLAppViewer::idle() } } - // Service the WorkQueue we use for replies from worker threads. - // Use function statics for the timeslice setting so we only have to fetch - // and convert MainWorkTime once. - static F32 MainWorkTimeRaw = gSavedSettings.getF32("MainWorkTime"); - static F32Milliseconds MainWorkTimeMs(MainWorkTimeRaw); - // MainWorkTime is specified in fractional milliseconds, but std::chrono - // uses integer representations. What if we want less than a microsecond? - // Use nanoseconds. We're very sure we will never need to specify a - // MainWorkTime that would be larger than we could express in - // std::chrono::nanoseconds. - static std::chrono::nanoseconds MainWorkTimeNanoSec{ - std::chrono::nanoseconds::rep(MainWorkTimeMs.value() * 1000000)}; - gMainloopWork.runFor(MainWorkTimeNanoSec); + // Execute deferred tasks. + LLDeferredTaskList::instance().run(); // Handle shutdown process, for example, // wait for floaters to close, send quit message, diff --git a/indra/newview/llenvironment.cpp b/indra/newview/llenvironment.cpp index 1a66f10b8f..dba24b3d02 100644 --- a/indra/newview/llenvironment.cpp +++ b/indra/newview/llenvironment.cpp @@ -824,6 +824,7 @@ std::string env_selection_to_string(LLEnvironment::EnvSelection_t sel) #undef RTNENUM } +LLEnvironment* LLSimpleton::sInstance = nullptr; //------------------------------------------------------------------------- LLEnvironment::LLEnvironment(): mCloudScrollDelta(), diff --git a/indra/newview/llmainlooprepeater.cpp b/indra/newview/llmainlooprepeater.cpp new file mode 100644 index 0000000000..6736e9a950 --- /dev/null +++ b/indra/newview/llmainlooprepeater.cpp @@ -0,0 +1,88 @@ +/** + * @file llmachineid.cpp + * @brief retrieves unique machine ids + * + * $LicenseInfo:firstyear=2009&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2010, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#include "llviewerprecompiledheaders.h" +#include "llapr.h" +#include "llevents.h" +#include "llmainlooprepeater.h" + + + +// LLMainLoopRepeater +//----------------------------------------------------------------------------- + + +LLMainLoopRepeater::LLMainLoopRepeater(void): + mQueue(0) +{ + ; // No op. +} + + +void LLMainLoopRepeater::start(void) +{ + if(mQueue != 0) return; + + mQueue = new LLThreadSafeQueue(1024); + mMainLoopConnection = LLEventPumps::instance(). + obtain("mainloop").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMainLoop, this, _1)); + mRepeaterConnection = LLEventPumps::instance(). + obtain("mainlooprepeater").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMessage, this, _1)); +} + + +void LLMainLoopRepeater::stop(void) +{ + mMainLoopConnection.release(); + mRepeaterConnection.release(); + + delete mQueue; + mQueue = 0; +} + + +bool LLMainLoopRepeater::onMainLoop(LLSD const &) +{ + LLSD message; + while(mQueue->tryPopBack(message)) { + std::string pump = message["pump"].asString(); + if(pump.length() == 0 ) continue; // No pump. + LLEventPumps::instance().obtain(pump).post(message["payload"]); + } + return false; +} + + +bool LLMainLoopRepeater::onMessage(LLSD const & event) +{ + try { + mQueue->pushFront(event); + } catch(LLThreadSafeQueueError & e) { + LL_WARNS() << "could not repeat message (" << e.what() << ")" << + event.asString() << LL_ENDL; + } + return false; +} diff --git a/indra/newview/llmainlooprepeater.h b/indra/newview/llmainlooprepeater.h new file mode 100644 index 0000000000..2ec3a74e4a --- /dev/null +++ b/indra/newview/llmainlooprepeater.h @@ -0,0 +1,64 @@ +/** + * @file llmainlooprepeater.h + * @brief a service for repeating messages on the main loop. + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2010, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_LLMAINLOOPREPEATER_H +#define LL_LLMAINLOOPREPEATER_H + + +#include "llsd.h" +#include "llthreadsafequeue.h" + + +// +// A service which creates the pump 'mainlooprepeater' to which any thread can +// post a message that will be re-posted on the main loop. +// +// The posted message should contain two map elements: pump and payload. The +// pump value is a string naming the pump to which the message should be +// re-posted. The payload value is what will be posted to the designated pump. +// +class LLMainLoopRepeater: + public LLSingleton +{ + LLSINGLETON(LLMainLoopRepeater); +public: + // Start the repeater service. + void start(void); + + // Stop the repeater service. + void stop(void); + +private: + LLTempBoundListener mMainLoopConnection; + LLTempBoundListener mRepeaterConnection; + LLThreadSafeQueue * mQueue; + + bool onMainLoop(LLSD const &); + bool onMessage(LLSD const & event); +}; + + +#endif diff --git a/indra/newview/llselectmgr.cpp b/indra/newview/llselectmgr.cpp index bc00c518e9..53247031b4 100644 --- a/indra/newview/llselectmgr.cpp +++ b/indra/newview/llselectmgr.cpp @@ -97,6 +97,8 @@ #include "llglheaders.h" #include "llinventoryobserver.h" +LLSelectMgr* LLSimpleton::sInstance = nullptr; + LLViewerObject* getSelectedParentObject(LLViewerObject *object) ; // // Consts diff --git a/indra/newview/llstartup.cpp b/indra/newview/llstartup.cpp index 9a4149948c..57c5074804 100644 --- a/indra/newview/llstartup.cpp +++ b/indra/newview/llstartup.cpp @@ -205,9 +205,6 @@ #include "llstacktrace.h" -#include "threadpool.h" - - #if LL_WINDOWS #include "lldxhardware.h" #endif @@ -304,20 +301,6 @@ void callback_cache_name(const LLUUID& id, const std::string& full_name, bool is // local classes // -void launchThreadPool() -{ - LLSD poolSizes{ gSavedSettings.getLLSD("ThreadPoolSizes") }; - LLSD sizeSpec{ poolSizes["General"] }; - LLSD::Integer size{ sizeSpec.isInteger()? sizeSpec.asInteger() : 3 }; - LL_DEBUGS("ThreadPool") << "Instantiating General pool with " - << size << " threads" << LL_ENDL; - // Use a function-static ThreadPool: static duration, but instantiated - // only on demand. - // We don't want anyone, especially the main thread, to have to block - // due to this ThreadPool being full. - static LL::ThreadPool pool("General", size, 1024*1024); -} - void update_texture_fetch() { LLAppViewer::getTextureCache()->update(1); // unpauses the texture cache thread @@ -1506,9 +1489,6 @@ bool idle_startup() gAgentCamera.resetCamera(); display_startup(); - // start up the ThreadPool we'll use for textures et al. - launchThreadPool(); - // Initialize global class data needed for surfaces (i.e. textures) LL_DEBUGS("AppInit") << "Initializing sky..." << LL_ENDL; // Initialize all of the viewer object classes for the first time (doing things like texture fetches. diff --git a/indra/newview/llviewercamera.cpp b/indra/newview/llviewercamera.cpp index 5d8e80cc41..5ebce115f6 100644 --- a/indra/newview/llviewercamera.cpp +++ b/indra/newview/llviewercamera.cpp @@ -54,6 +54,8 @@ // System includes #include // for setprecision +LLViewerCamera* LLSimpleton::sInstance = nullptr; + LLTrace::CountStatHandle<> LLViewerCamera::sVelocityStat("camera_velocity"); LLTrace::CountStatHandle<> LLViewerCamera::sAngularVelocityStat("camera_angular_velocity"); diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index 498e4ef8bc..fbc5830a5c 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -679,9 +679,6 @@ void LLViewerTexture::init(bool firstinit) mVolumeList[LLRender::LIGHT_TEX].clear(); mVolumeList[LLRender::SCULPT_TEX].clear(); - - mMainQueue = LL::WorkQueue::getInstance("mainloop"); - mImageQueue = LL::WorkQueue::getInstance("LLImageGL"); } //virtual @@ -1625,26 +1622,17 @@ void LLViewerFetchedTexture::scheduleCreateTexture() { mNeedsCreateTexture = TRUE; #if LL_WINDOWS //flip to 0 to revert to single-threaded OpenGL texture uploads - auto mainq = mMainQueue.lock(); - if (mainq) - { - mainq->postTo( - mImageQueue, - // work to be done on LLImageGL worker thread - [this]() - { - //actually create the texture on a background thread - createTexture(); - }, - // callback to be run on main thread - [this]() - { - //finalize on main thread - postCreateTexture(); - unref(); - }); - } - else + if (!LLImageGLThread::sInstance->post([this]() + { + //actually create the texture on a background thread + createTexture(); + LLImageGLThread::sInstance->postCallback([this]() + { + //finalize on main thread + postCreateTexture(); + unref(); + }); + })) #endif { gTextureList.mCreateTextureList.insert(this); diff --git a/indra/newview/llviewertexture.h b/indra/newview/llviewertexture.h index 4cd4c7cd39..f9f1bfef44 100644 --- a/indra/newview/llviewertexture.h +++ b/indra/newview/llviewertexture.h @@ -35,7 +35,6 @@ #include "llrender.h" #include "llmetricperformancetester.h" #include "httpcommon.h" -#include "workqueue.h" #include #include @@ -214,9 +213,6 @@ protected: //do not use LLPointer here. LLViewerMediaTexture* mParcelMedia ; - LL::WorkQueue::weak_t mMainQueue; - LL::WorkQueue::weak_t mImageQueue; - static F32 sTexelPixelRatio; public: static const U32 sCurrentFileVersion; diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp index 5f62908009..0a8457eb2c 100644 --- a/indra/newview/llworld.cpp +++ b/indra/newview/llworld.cpp @@ -62,6 +62,8 @@ #include +LLWorld* LLSimpleton::sInstance = nullptr; + // // Globals // -- cgit v1.3 From 18de6c9b989cc7060f2a314f5b68cc102677823b Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 11 Nov 2021 10:23:16 -0500 Subject: SL-16094: Stylish braces! --- indra/llcommon/llsingleton.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index fdd5bdfea9..6042c0906c 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -858,7 +858,8 @@ public: static inline T& instance() { return *getInstance(); } static inline bool instanceExists() { return sInstance != nullptr; } - static void deleteSingleton() { + static void deleteSingleton() + { delete sInstance; sInstance = nullptr; } -- cgit v1.3 From 730b8427b5f40e5f62bca15c5109f50db5c10be5 Mon Sep 17 00:00:00 2001 From: "Brad Payne (Vir Linden)" Date: Tue, 9 Nov 2021 20:25:25 +0000 Subject: SL-16329 - track frame time and jitter (as average deviation frame to frame) in stats window --- indra/llcommon/lltracerecording.cpp | 27 +++++++++++++++++++- indra/llcommon/lltracerecording.h | 29 ++++++++++++++++++++++ indra/llui/llstatbar.cpp | 17 +++++++++++-- indra/llui/llstatbar.h | 6 +++-- indra/newview/llviewerstats.cpp | 11 +++++--- indra/newview/llviewerstats.h | 4 +-- .../newview/skins/default/xui/en/floater_stats.xml | 19 ++++++++++++++ 7 files changed, 103 insertions(+), 10 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/lltracerecording.cpp b/indra/llcommon/lltracerecording.cpp index c72a64d086..5ce1b337fe 100644 --- a/indra/llcommon/lltracerecording.cpp +++ b/indra/llcommon/lltracerecording.cpp @@ -858,7 +858,6 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S3 : NaN; } - F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { LL_PROFILE_ZONE_SCOPED; @@ -952,6 +951,32 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S : NaN; } +F64 PeriodicRecording::getPeriodMedian( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) +{ + LL_PROFILE_ZONE_SCOPED; + num_periods = llmin(num_periods, getNumRecordedPeriods()); + + std::vector buf; + for (S32 i = 1; i <= num_periods; i++) + { + Recording& recording = getPrevRecording(i); + if (recording.getDuration() > (F32Seconds)0.f) + { + if (recording.hasValue(stat)) + { + buf.push_back(recording.getMean(stat)); + } + } + } + if (buf.size()==0) + { + return 0.0f; + } + std::sort(buf.begin(), buf.end()); + + return F64((buf.size() % 2 == 0) ? (buf[buf.size() / 2 - 1] + buf[buf.size() / 2]) / 2 : buf[buf.size() / 2]); +} + F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { LL_PROFILE_ZONE_SCOPED; diff --git a/indra/llcommon/lltracerecording.h b/indra/llcommon/lltracerecording.h index 6715104613..1f3d37336a 100644 --- a/indra/llcommon/lltracerecording.h +++ b/indra/llcommon/lltracerecording.h @@ -599,6 +599,35 @@ namespace LLTrace return typename RelatedTypes::fractional_t(getPeriodMeanPerSec(static_cast&>(stat), num_periods)); } + F64 getPeriodMedian( const StatType& stat, S32 num_periods = S32_MAX); + + template + typename RelatedTypes::fractional_t getPeriodMedianPerSec(const StatType& stat, S32 num_periods = S32_MAX) + { + LL_PROFILE_ZONE_SCOPED; + num_periods = llmin(num_periods, getNumRecordedPeriods()); + + std::vector ::fractional_t> buf; + for (S32 i = 1; i <= num_periods; i++) + { + Recording& recording = getPrevRecording(i); + if (recording.getDuration() > (F32Seconds)0.f) + { + buf.push_back(recording.getPerSec(stat)); + } + } + std::sort(buf.begin(), buf.end()); + + return typename RelatedTypes::fractional_t((buf.size() % 2 == 0) ? (buf[buf.size() / 2 - 1] + buf[buf.size() / 2]) / 2 : buf[buf.size() / 2]); + } + + template + typename RelatedTypes::fractional_t getPeriodMedianPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) + { + LL_PROFILE_ZONE_SCOPED; + return typename RelatedTypes::fractional_t(getPeriodMedianPerSec(static_cast&>(stat), num_periods)); + } + // // PERIODIC STANDARD DEVIATION // diff --git a/indra/llui/llstatbar.cpp b/indra/llui/llstatbar.cpp index 8adcd664df..2449100952 100644 --- a/indra/llui/llstatbar.cpp +++ b/indra/llui/llstatbar.cpp @@ -160,6 +160,7 @@ LLStatBar::Params::Params() tick_spacing("tick_spacing", 0.f), decimal_digits("decimal_digits", 3), show_bar("show_bar", false), + show_median("show_median", false), show_history("show_history", false), scale_range("scale_range", true), num_frames("num_frames", 200), @@ -186,6 +187,7 @@ LLStatBar::LLStatBar(const Params& p) mNumShortHistoryFrames(p.num_frames_short), mMaxHeight(p.max_height), mDisplayBar(p.show_bar), + mShowMedian(p.show_median), mDisplayHistory(p.show_history), mOrientation(p.orientation), mAutoScaleMax(!p.bar_max.isProvided()), @@ -318,7 +320,14 @@ void LLStatBar::draw() min = frame_recording.getPeriodMinPerSec(count_stat, num_frames); max = frame_recording.getPeriodMaxPerSec(count_stat, num_frames); mean = frame_recording.getPeriodMeanPerSec(count_stat, num_frames); - display_value = mean; + if (mShowMedian) + { + display_value = frame_recording.getPeriodMedianPerSec(count_stat, num_frames); + } + else + { + display_value = mean; + } } break; case STAT_EVENT: @@ -344,7 +353,11 @@ void LLStatBar::draw() mean = frame_recording.getPeriodMean(sample_stat, num_frames); num_rapid_changes = calc_num_rapid_changes(frame_recording, sample_stat, RAPID_CHANGE_WINDOW); - if (num_rapid_changes / RAPID_CHANGE_WINDOW.value() > MAX_RAPID_CHANGES_PER_SEC) + if (mShowMedian) + { + display_value = frame_recording.getPeriodMedian(sample_stat, num_frames); + } + else if (num_rapid_changes / RAPID_CHANGE_WINDOW.value() > MAX_RAPID_CHANGES_PER_SEC) { display_value = mean; } diff --git a/indra/llui/llstatbar.h b/indra/llui/llstatbar.h index 1ff4c67fc5..6b481ca68f 100644 --- a/indra/llui/llstatbar.h +++ b/indra/llui/llstatbar.h @@ -44,9 +44,10 @@ public: bar_max, tick_spacing; - Optional show_bar, + Optional show_bar, show_history, - scale_range; + scale_range, + show_median; // default is mean Optional decimal_digits, num_frames, @@ -112,6 +113,7 @@ private: bool mDisplayBar, // Display the bar graph. mDisplayHistory, + mShowMedian, mAutoScaleMax, mAutoScaleMin; }; diff --git a/indra/newview/llviewerstats.cpp b/indra/newview/llviewerstats.cpp index 314c1a1f1e..ac8a657fb2 100644 --- a/indra/newview/llviewerstats.cpp +++ b/indra/newview/llviewerstats.cpp @@ -183,8 +183,9 @@ SimMeasurement SIM_UNACKED_BYTES("simtotalunackedbytes", "", LL_S SimMeasurement SIM_PHYSICS_MEM("physicsmemoryallocated", "", LL_SIM_STAT_SIMPHYSICSMEMORY); LLTrace::SampleStatHandle FRAMETIME_JITTER("frametimejitter", "Average delta between successive frame times"), - FRAMETIME_SLEW("frametimeslew", "Average delta between frame time and mean"), - SIM_PING("simpingstat"); + FRAMETIME_SLEW("frametimeslew", "Average delta between frame time and mean"), + FRAMETIME("frametime", "Measured frame time"), + SIM_PING("simpingstat"); LLTrace::EventStatHandle > AGENT_POSITION_SNAP("agentpositionsnap", "agent position corrections"); @@ -261,8 +262,12 @@ void LLViewerStats::updateFrameStats(const F64Seconds time_diff) // new "stutter" meter add(LLStatViewer::FRAMETIME_DOUBLED, time_diff >= 2.0 * mLastTimeDiff ? 1 : 0); + sample(LLStatViewer::FRAMETIME, time_diff); + // old stats that were never really used - sample(LLStatViewer::FRAMETIME_JITTER, F64Milliseconds (mLastTimeDiff - time_diff)); + F64Seconds jit = (F64Seconds) std::fabs((mLastTimeDiff - time_diff)); + LL_INFOS() << "times " << mLastTimeDiff << ", " << time_diff << " jit " << jit << LL_ENDL; + sample(LLStatViewer::FRAMETIME_JITTER, jit); F32Seconds average_frametime = gRenderStartTime.getElapsedTimeF32() / (F32)gFrameCount; sample(LLStatViewer::FRAMETIME_SLEW, F64Milliseconds (average_frametime - time_diff)); diff --git a/indra/newview/llviewerstats.h b/indra/newview/llviewerstats.h index 04870e0c26..ac8eccc0ca 100644 --- a/indra/newview/llviewerstats.h +++ b/indra/newview/llviewerstats.h @@ -218,8 +218,8 @@ extern SimMeasurement SIM_PHYSICS_MEM; extern LLTrace::SampleStatHandle FRAMETIME_JITTER, - FRAMETIME_SLEW, - SIM_PING; + FRAMETIME_SLEW, + SIM_PING; extern LLTrace::EventStatHandle > AGENT_POSITION_SNAP; diff --git a/indra/newview/skins/default/xui/en/floater_stats.xml b/indra/newview/skins/default/xui/en/floater_stats.xml index e4f735740b..6f84930c75 100644 --- a/indra/newview/skins/default/xui/en/floater_stats.xml +++ b/indra/newview/skins/default/xui/en/floater_stats.xml @@ -35,6 +35,25 @@ decimal_digits="1" show_bar="true" show_history="true"/> + + + Date: Thu, 4 Nov 2021 16:40:05 -0400 Subject: SL-16202: Instantiate LLSimpleton::sInstance generically instead of requiring a separate declaration for each subclass. The previous way produces errors in clang. (cherry picked from commit 8458ad8890cf0a11804996210d7bcfbdaa3eec2e) --- indra/llcommon/llsingleton.h | 3 +++ indra/llui/llviewereventrecorder.cpp | 2 -- indra/newview/llenvironment.cpp | 1 - indra/newview/llselectmgr.cpp | 2 -- indra/newview/llviewercamera.cpp | 2 -- indra/newview/llworld.cpp | 2 -- 6 files changed, 3 insertions(+), 9 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 10a8ecfedb..24d01812c9 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -865,4 +865,7 @@ public: } }; +template +T* LLSimpleton::sInstance{ nullptr }; + #endif diff --git a/indra/llui/llviewereventrecorder.cpp b/indra/llui/llviewereventrecorder.cpp index 5a44ec947a..cb000aef74 100644 --- a/indra/llui/llviewereventrecorder.cpp +++ b/indra/llui/llviewereventrecorder.cpp @@ -28,8 +28,6 @@ #include "llui.h" #include "llleap.h" -LLViewerEventRecorder* LLSimpleton::sInstance = nullptr; - LLViewerEventRecorder::LLViewerEventRecorder() { clear(UNDEFINED); diff --git a/indra/newview/llenvironment.cpp b/indra/newview/llenvironment.cpp index 0b914f07ab..4bec7fa111 100644 --- a/indra/newview/llenvironment.cpp +++ b/indra/newview/llenvironment.cpp @@ -824,7 +824,6 @@ std::string env_selection_to_string(LLEnvironment::EnvSelection_t sel) #undef RTNENUM } -LLEnvironment* LLSimpleton::sInstance = nullptr; //------------------------------------------------------------------------- LLEnvironment::LLEnvironment(): mCloudScrollDelta(), diff --git a/indra/newview/llselectmgr.cpp b/indra/newview/llselectmgr.cpp index e7670b5a73..6f136e50e0 100644 --- a/indra/newview/llselectmgr.cpp +++ b/indra/newview/llselectmgr.cpp @@ -97,8 +97,6 @@ #include "llglheaders.h" #include "llinventoryobserver.h" -LLSelectMgr* LLSimpleton::sInstance = nullptr; - LLViewerObject* getSelectedParentObject(LLViewerObject *object) ; // // Consts diff --git a/indra/newview/llviewercamera.cpp b/indra/newview/llviewercamera.cpp index 5ebce115f6..5d8e80cc41 100644 --- a/indra/newview/llviewercamera.cpp +++ b/indra/newview/llviewercamera.cpp @@ -54,8 +54,6 @@ // System includes #include // for setprecision -LLViewerCamera* LLSimpleton::sInstance = nullptr; - LLTrace::CountStatHandle<> LLViewerCamera::sVelocityStat("camera_velocity"); LLTrace::CountStatHandle<> LLViewerCamera::sAngularVelocityStat("camera_angular_velocity"); diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp index 4cb63d0ab8..d5cce6a52a 100644 --- a/indra/newview/llworld.cpp +++ b/indra/newview/llworld.cpp @@ -62,8 +62,6 @@ #include -LLWorld* LLSimpleton::sInstance = nullptr; - // // Globals // -- cgit v1.3 From f997bcd186d00e30132f32be007bb3978bf3a8f5 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Thu, 11 Nov 2021 10:23:16 -0500 Subject: SL-16094: Stylish braces! (cherry picked from commit 18de6c9b989cc7060f2a314f5b68cc102677823b) --- indra/llcommon/llsingleton.h | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 24d01812c9..da2d6fd984 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -859,9 +859,16 @@ public: static inline T& instance() { return *getInstance(); } static inline bool instanceExists() { return sInstance != nullptr; } +<<<<<<< HEAD static void deleteSingleton() { delete sInstance; sInstance = nullptr; +======= + static void deleteSingleton() + { + delete sInstance; + sInstance = nullptr; +>>>>>>> 18de6c9b98 (SL-16094: Stylish braces!) } }; -- cgit v1.3 From 3171aaad9b1f2757f8b0d8cbb784a45a7bbebafa Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Fri, 19 Nov 2021 14:57:36 -0500 Subject: SL-16094: fix merge glitch --- indra/llcommon/llsingleton.h | 6 ------ 1 file changed, 6 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index da2d6fd984..f85f961287 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -859,16 +859,10 @@ public: static inline T& instance() { return *getInstance(); } static inline bool instanceExists() { return sInstance != nullptr; } -<<<<<<< HEAD - static void deleteSingleton() { - delete sInstance; - sInstance = nullptr; -======= static void deleteSingleton() { delete sInstance; sInstance = nullptr; ->>>>>>> 18de6c9b98 (SL-16094: Stylish braces!) } }; -- cgit v1.3 From cc34e26ef7e74845e4af9e5c5d450c0b12a268e0 Mon Sep 17 00:00:00 2001 From: Runitai Linden Date: Mon, 22 Nov 2021 11:51:03 -0600 Subject: SL-16094 Add WorkQueue profile hooks --- indra/llcommon/workqueue.cpp | 2 ++ indra/llcommon/workqueue.h | 3 +++ 2 files changed, 5 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index 633594ceea..fbdbea2051 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -60,6 +60,7 @@ void LL::WorkQueue::runUntilClose() { for (;;) { + LL_PROFILE_ZONE_SCOPED; callWork(mQueue.pop()); } } @@ -90,6 +91,7 @@ bool LL::WorkQueue::runOne() bool LL::WorkQueue::runUntil(const TimePoint& until) { + LL_PROFILE_ZONE_SCOPED; // Should we subtract some slop to allow for typical Work execution time? // How much slop? Work work; diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index c25d787425..96574a18b9 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -260,6 +260,7 @@ namespace LL template bool runFor(const std::chrono::duration& timeslice) { + LL_PROFILE_ZONE_SCOPED; return runUntil(TimePoint::clock::now() + timeslice); } @@ -431,6 +432,7 @@ namespace LL bool WorkQueue::postTo(weak_t target, const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) { + LL_PROFILE_ZONE_SCOPED; // We're being asked to post to the WorkQueue at target. // target is a weak_ptr: have to lock it to check it. auto tptr = target.lock(); @@ -479,6 +481,7 @@ namespace LL template bool WorkQueue::postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable) { + LL_PROFILE_ZONE_SCOPED; // target is a weak_ptr: have to lock it to check it auto tptr = target.lock(); if (tptr) -- cgit v1.3 From 9b0d8c7e629597fd8e6dfb91a6b8f625b34ab274 Mon Sep 17 00:00:00 2001 From: Runitai Linden Date: Mon, 22 Nov 2021 18:42:56 -0600 Subject: SL-16094 More profile hooks for threading code, remove redundant wglCreateContextAttribs call --- indra/llcommon/llthreadsafequeue.h | 18 ++++++++++++++++++ indra/llcommon/threadpool.cpp | 6 +++++- indra/llcommon/threadsafeschedule.h | 34 ++++++++++++++++++++++++++++++---- indra/llrender/llimagegl.cpp | 2 ++ indra/llwindow/llwindowwin32.cpp | 2 +- 5 files changed, 56 insertions(+), 6 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 5c934791fe..2806506550 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -275,6 +275,7 @@ template template bool LLThreadSafeQueue::tryLock(CALLABLE&& callable) { + LL_PROFILE_ZONE_SCOPED; lock_t lock1(mLock, std::defer_lock); if (!lock1.try_lock()) return false; @@ -291,6 +292,7 @@ bool LLThreadSafeQueue::tryLockUntil( const std::chrono::time_point& until, CALLABLE&& callable) { + LL_PROFILE_ZONE_SCOPED; lock_t lock1(mLock, std::defer_lock); if (!lock1.try_lock_until(until)) return false; @@ -304,6 +306,7 @@ template template bool LLThreadSafeQueue::push_(lock_t& lock, T&& element) { + LL_PROFILE_ZONE_SCOPED; if (mStorage.size() >= mCapacity) return false; @@ -319,6 +322,7 @@ template template bool LLThreadSafeQueue::pushIfOpen(T&& element) { + LL_PROFILE_ZONE_SCOPED; lock_t lock1(mLock); while (true) { @@ -341,6 +345,7 @@ template template void LLThreadSafeQueue::push(T&& element) { + LL_PROFILE_ZONE_SCOPED; if (! pushIfOpen(std::forward(element))) { LLTHROW(LLThreadSafeQueueInterrupt()); @@ -352,6 +357,7 @@ template template bool LLThreadSafeQueue::tryPush(T&& element) { + LL_PROFILE_ZONE_SCOPED; return tryLock( [this, element=std::move(element)](lock_t& lock) { @@ -368,6 +374,7 @@ bool LLThreadSafeQueue::tryPushFor( const std::chrono::duration& timeout, T&& element) { + LL_PROFILE_ZONE_SCOPED; // Convert duration to time_point: passing the same timeout duration to // each of multiple calls is wrong. return tryPushUntil(std::chrono::steady_clock::now() + timeout, @@ -381,6 +388,7 @@ bool LLThreadSafeQueue::tryPushUntil( const std::chrono::time_point& until, T&& element) { + LL_PROFILE_ZONE_SCOPED; return tryLockUntil( until, [this, until, element=std::move(element)](lock_t& lock) @@ -413,6 +421,7 @@ template typename LLThreadSafeQueue::pop_result LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) { + LL_PROFILE_ZONE_SCOPED; // If mStorage is empty, there's no head element. if (mStorage.empty()) return mClosed? DONE : EMPTY; @@ -434,6 +443,7 @@ LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) template ElementT LLThreadSafeQueue::pop(void) { + LL_PROFILE_ZONE_SCOPED; lock_t lock1(mLock); ElementT value; while (true) @@ -462,6 +472,7 @@ ElementT LLThreadSafeQueue::pop(void) template bool LLThreadSafeQueue::tryPop(ElementT & element) { + LL_PROFILE_ZONE_SCOPED; return tryLock( [this, &element](lock_t& lock) { @@ -479,6 +490,7 @@ bool LLThreadSafeQueue::tryPopFor( const std::chrono::duration& timeout, ElementT& element) { + LL_PROFILE_ZONE_SCOPED; // Convert duration to time_point: passing the same timeout duration to // each of multiple calls is wrong. return tryPopUntil(std::chrono::steady_clock::now() + timeout, element); @@ -491,6 +503,7 @@ bool LLThreadSafeQueue::tryPopUntil( const std::chrono::time_point& until, ElementT& element) { + LL_PROFILE_ZONE_SCOPED; return tryLockUntil( until, [this, until, &element](lock_t& lock) @@ -510,6 +523,7 @@ LLThreadSafeQueue::tryPopUntil_( const std::chrono::time_point& until, ElementT& element) { + LL_PROFILE_ZONE_SCOPED; while (true) { pop_result popped = pop_(lock, element); @@ -536,6 +550,7 @@ LLThreadSafeQueue::tryPopUntil_( template size_t LLThreadSafeQueue::size(void) { + LL_PROFILE_ZONE_SCOPED; lock_t lock(mLock); return mStorage.size(); } @@ -544,6 +559,7 @@ size_t LLThreadSafeQueue::size(void) template void LLThreadSafeQueue::close() { + LL_PROFILE_ZONE_SCOPED; lock_t lock(mLock); mClosed = true; lock.unlock(); @@ -557,6 +573,7 @@ void LLThreadSafeQueue::close() template bool LLThreadSafeQueue::isClosed() { + LL_PROFILE_ZONE_SCOPED; lock_t lock(mLock); return mClosed; } @@ -565,6 +582,7 @@ bool LLThreadSafeQueue::isClosed() template bool LLThreadSafeQueue::done() { + LL_PROFILE_ZONE_SCOPED; lock_t lock(mLock); return mClosed && mStorage.empty(); } diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp index cf25cc838e..06e0dc5bfc 100644 --- a/indra/llcommon/threadpool.cpp +++ b/indra/llcommon/threadpool.cpp @@ -28,7 +28,11 @@ LL::ThreadPool::ThreadPool(const std::string& name, size_t threads, size_t capac for (size_t i = 0; i < threads; ++i) { std::string tname{ STRINGIZE(mName << ':' << (i+1) << '/' << threads) }; - mThreads.emplace_back(tname, [this, tname](){ run(tname); }); + mThreads.emplace_back(tname, [this, tname]() + { + LL_PROFILER_SET_THREAD_NAME(tname.c_str()); + run(tname); + }); } // Listen on "LLApp", and when the app is shutting down, close the queue // and join the workers. diff --git a/indra/llcommon/threadsafeschedule.h b/indra/llcommon/threadsafeschedule.h index c8ad23532b..601681d550 100644 --- a/indra/llcommon/threadsafeschedule.h +++ b/indra/llcommon/threadsafeschedule.h @@ -98,12 +98,14 @@ namespace LL // we could minimize redundancy by breaking out a common base class... void push(const DataTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; push(tuple_cons(Clock::now(), tuple)); } /// individually pass each component of the TimeTuple void push(const TimePoint& time, Args&&... args) { + LL_PROFILE_ZONE_SCOPED; push(TimeTuple(time, std::forward(args)...)); } @@ -114,6 +116,7 @@ namespace LL // and call that overload. void push(Args&&... args) { + LL_PROFILE_ZONE_SCOPED; push(Clock::now(), std::forward(args)...); } @@ -124,18 +127,21 @@ namespace LL /// DataTuple with implicit now bool tryPush(const DataTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; return tryPush(tuple_cons(Clock::now(), tuple)); } /// individually pass components bool tryPush(const TimePoint& time, Args&&... args) { + LL_PROFILE_ZONE_SCOPED; return tryPush(TimeTuple(time, std::forward(args)...)); } /// individually pass components with implicit now bool tryPush(Args&&... args) { + LL_PROFILE_ZONE_SCOPED; return tryPush(Clock::now(), std::forward(args)...); } @@ -148,6 +154,7 @@ namespace LL bool tryPushFor(const std::chrono::duration& timeout, const DataTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; return tryPushFor(timeout, tuple_cons(Clock::now(), tuple)); } @@ -156,6 +163,7 @@ namespace LL bool tryPushFor(const std::chrono::duration& timeout, const TimePoint& time, Args&&... args) { + LL_PROFILE_ZONE_SCOPED; return tryPushFor(TimeTuple(time, std::forward(args)...)); } @@ -164,6 +172,7 @@ namespace LL bool tryPushFor(const std::chrono::duration& timeout, Args&&... args) { + LL_PROFILE_ZONE_SCOPED; return tryPushFor(Clock::now(), std::forward(args)...); } @@ -176,6 +185,7 @@ namespace LL bool tryPushUntil(const std::chrono::time_point& until, const DataTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; return tryPushUntil(until, tuple_cons(Clock::now(), tuple)); } @@ -184,6 +194,7 @@ namespace LL bool tryPushUntil(const std::chrono::time_point& until, const TimePoint& time, Args&&... args) { + LL_PROFILE_ZONE_SCOPED; return tryPushUntil(until, TimeTuple(time, std::forward(args)...)); } @@ -192,6 +203,7 @@ namespace LL bool tryPushUntil(const std::chrono::time_point& until, Args&&... args) { + LL_PROFILE_ZONE_SCOPED; return tryPushUntil(until, Clock::now(), std::forward(args)...); } @@ -209,12 +221,14 @@ namespace LL // haven't yet jumped through those hoops. DataTuple pop() { + LL_PROFILE_ZONE_SCOPED; return tuple_cdr(popWithTime()); } /// pop TimeTuple by value TimeTuple popWithTime() { + LL_PROFILE_ZONE_SCOPED; lock_t lock(super::mLock); // We can't just sit around waiting forever, given that there may // be items in the queue that are not yet ready but will *become* @@ -254,6 +268,7 @@ namespace LL /// tryPop(DataTuple&) bool tryPop(DataTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; TimeTuple tt; if (! super::tryPop(tt)) return false; @@ -264,6 +279,7 @@ namespace LL /// for when Args has exactly one type bool tryPop(typename std::tuple_element<1, TimeTuple>::type& value) { + LL_PROFILE_ZONE_SCOPED; TimeTuple tt; if (! super::tryPop(tt)) return false; @@ -275,6 +291,7 @@ namespace LL template bool tryPopFor(const std::chrono::duration& timeout, Tuple& tuple) { + LL_PROFILE_ZONE_SCOPED; // It's important to use OUR tryPopUntil() implementation, rather // than delegating immediately to our base class. return tryPopUntil(Clock::now() + timeout, tuple); @@ -285,6 +302,7 @@ namespace LL bool tryPopUntil(const std::chrono::time_point& until, TimeTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; // super::tryPopUntil() wakes up when an item becomes available or // we hit 'until', whichever comes first. Thing is, the current // head of the queue could become ready sooner than either of @@ -304,20 +322,25 @@ namespace LL pop_result tryPopUntil_(lock_t& lock, const TimePoint& until, TimeTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; TimePoint adjusted = until; if (! super::mStorage.empty()) { + LL_PROFILE_ZONE_NAMED("tpu - adjust"); // use whichever is earlier: the head item's timestamp, or // the caller's limit adjusted = min(std::get<0>(super::mStorage.front()), adjusted); } // now delegate to base-class tryPopUntil_() pop_result popped; - while ((popped = pop_result(super::tryPopUntil_(lock, adjusted, tuple))) == WAITING) { - // If super::tryPopUntil_() returns WAITING, it means there's - // a head item, but it's not yet time. But it's worth looping - // back to recheck. + LL_PROFILE_ZONE_NAMED("tpu - super"); + while ((popped = pop_result(super::tryPopUntil_(lock, adjusted, tuple))) == WAITING) + { + // If super::tryPopUntil_() returns WAITING, it means there's + // a head item, but it's not yet time. But it's worth looping + // back to recheck. + } } return popped; } @@ -327,6 +350,7 @@ namespace LL bool tryPopUntil(const std::chrono::time_point& until, DataTuple& tuple) { + LL_PROFILE_ZONE_SCOPED; TimeTuple tt; if (! tryPopUntil(until, tt)) return false; @@ -339,6 +363,7 @@ namespace LL bool tryPopUntil(const std::chrono::time_point& until, typename std::tuple_element<1, TimeTuple>::type& value) { + LL_PROFILE_ZONE_SCOPED; TimeTuple tt; if (! tryPopUntil(until, tt)) return false; @@ -362,6 +387,7 @@ namespace LL // considering whether to deliver the current head element bool canPop(const TimeTuple& head) const override { + LL_PROFILE_ZONE_SCOPED; // an item with a future timestamp isn't yet ready to pop // (should we add some slop for overhead?) return std::get<0>(head) <= Clock::now(); diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 1b6920fe3b..e623baa653 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -2257,6 +2257,7 @@ LLImageGLThread::LLImageGLThread(LLWindow* window) : ThreadPool("LLImageGL", 1, 1024*1024) , mWindow(window) { + LL_PROFILE_ZONE_SCOPED; mFinished = false; mContext = mWindow->createSharedContext(); @@ -2264,6 +2265,7 @@ LLImageGLThread::LLImageGLThread(LLWindow* window) void LLImageGLThread::run() { + LL_PROFILE_ZONE_SCOPED; // We must perform setup on this thread before actually servicing our // WorkQueue, likewise cleanup afterwards. mWindow->makeContextCurrent(mContext); diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index b845f75ce4..777117b7c8 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -1697,7 +1697,7 @@ void* LLWindowWin32::createSharedContext() 0 }; - HGLRC rc = wglCreateContextAttribsARB(mhDC, mhRC, attribs); + HGLRC rc = 0; bool done = false; while (!done) -- cgit v1.3 From a32a45163d18f9b5998e469a356f870dbdb034ad Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 23 Nov 2021 09:58:54 -0500 Subject: SL-16094: Extend stringize() to support variadic arguments. It's useful to be able to say STRINGIZE(item0 << item1 << item2), and we use that a lot in our code base. But weird syntax aside, there are a couple advantages to being able to write stringize(item0, item1, item2). First, it allows stringize() to be used from within some other variadic function, without having to make that function a macro that accepts an arbitrary insertion-operator expression. There's no such thing as a member macro. Second, particularly for variadic functions, it allows us to optimize the single-argument case stringize(item0). A macro can't do that. When item0 is already a string of the desired char type, instead of streaming it into a std::ostringstream and retrieving it again, we can simply return the input string. When it's a pointer to the desired char type, we can directly construct the result string without the help of std::ostringstream. When it's a string of some other char type, we can engage ll_convert() to perform needed conversions. We generalize and optimize the generic gstringize() function, retaining the role of stringize() and wstringize() as thin wrappers that merely provide the desired char type. Optimizing the single-argument case requires separately defining gstringize() with two or more arguments: the general case. Then gstringize(arg) is delegated to a gstringize_impl class template so we can partially specialize to recognize a std::basic_string argument, as well as desired_char_type*. Both these specializations engage ll_convert(), which already handles the trivial case when no conversion is required. Use of ll_convert() in this role supercedes and generalizes the previous wstring_to_utf8str() and utf8str_to_wstring() overloads. Also introduce stream_to(std::ostream&, ...) to support variadic streaming to other destinations, e.g. a file, std::cout, ... --- indra/llcommon/stringize.h | 109 +++++++++++++++++++++++++++++++++------------ 1 file changed, 80 insertions(+), 29 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/stringize.h b/indra/llcommon/stringize.h index 38dd198ad3..bc91f188e4 100644 --- a/indra/llcommon/stringize.h +++ b/indra/llcommon/stringize.h @@ -31,58 +31,109 @@ #include #include +#include /** - * gstringize(item) encapsulates an idiom we use constantly, using - * operator<<(std::ostringstream&, TYPE) followed by std::ostringstream::str() - * or their wstring equivalents - * to render a string expressing some item. + * stream_to(std::ostream&, items, ...) streams each item in the parameter list + * to the passed std::ostream using the insertion operator <<. This can be + * used, for instance, to make a simple print() function, e.g.: + * + * @code + * template + * void print(Items&&... items) + * { + * stream_to(std::cout, std::forward(items)...); + * } + * @endcode */ -template -std::basic_string gstringize(const T& item) +// recursion tail +template +void stream_to(std::basic_ostream& out) {} +// stream one or more items +template +void stream_to(std::basic_ostream& out, T&& item, Items&&... items) { - std::basic_ostringstream out; - out << item; - return out.str(); + out << std::move(item); + stream_to(out, std::forward(items)...); } +// why we use function overloads, not function template specializations: +// http://www.gotw.ca/publications/mill17.htm + /** - *partial specialization of stringize for handling wstring - *TODO: we should have similar specializations for wchar_t[] but not until it is needed. + * gstringize(item, ...) encapsulates an idiom we use constantly, using + * operator<<(std::ostringstream&, TYPE) followed by std::ostringstream::str() + * or their wstring equivalents to render a string expressing one or more items. */ -inline std::string stringize(const std::wstring& item) +// two or more args - the case of a single argument is handled separately +template +auto gstringize(T0&& item0, T1&& item1, Items&&... items) { - return wstring_to_utf8str(item); + std::basic_ostringstream out; + stream_to(out, std::forward(item0), std::forward(item1), + std::forward(items)...); + return out.str(); } -/** - * Specialization of gstringize for std::string return types - */ -template -std::string stringize(const T& item) +// generic single argument: stream to out, as above +template +struct gstringize_impl { - return gstringize(item); + auto operator()(typename boost::call_traits::param_type arg) + { + std::basic_ostringstream out; + out << arg; + return out.str(); + } +}; + +// partially specialize for a single STRING argument - +// note that ll_convert(T) already handles the trivial case +template +struct gstringize_impl> +{ + auto operator()(const std::basic_string& arg) + { + return ll_convert>(arg); + } +}; + +// partially specialize for a single CHARTYPE* argument - +// since it's not a basic_string and we do want to optimize this common case +template +struct gstringize_impl +{ + auto operator()(const INCHAR* arg) + { + return ll_convert>(arg); + } +}; + +// gstringize(single argument) +template +auto gstringize(T&& item) +{ + // use decay so we don't require separate specializations for T, const + // T, T&, const T& ... + return gstringize_impl>()(std::forward(item)); } /** - * Specialization for generating wstring from string. - * Both a convenience function and saves a miniscule amount of overhead. + * Specialization of gstringize for std::string return types */ -inline std::wstring wstringize(const std::string& item) +template +auto stringize(Items&&... items) { - // utf8str_to_wstring() returns LLWString, which isn't necessarily the - // same as std::wstring - LLWString s(utf8str_to_wstring(item)); - return std::wstring(s.begin(), s.end()); + return gstringize(std::forward(items)...); } /** * Specialization of gstringize for std::wstring return types */ -template -std::wstring wstringize(const T& item) +template +auto wstringize(Items&&... items) { - return gstringize(item); + return gstringize(std::forward(items)...); } /** -- cgit v1.3 From adc2666dbb2444194a5df84711207def7eba074c Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 23 Nov 2021 10:11:16 -0500 Subject: SL-16094: Tweak llstring merge --- indra/llcommon/stringize.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/stringize.h b/indra/llcommon/stringize.h index 8501beb16d..12df693910 100644 --- a/indra/llcommon/stringize.h +++ b/indra/llcommon/stringize.h @@ -53,7 +53,7 @@ void stream_to(std::basic_ostream& out) {} template void stream_to(std::basic_ostream& out, T&& item, Items&&... items) { - out << std::move(item); + out << std::forward(item); stream_to(out, std::forward(items)...); } @@ -113,8 +113,8 @@ struct gstringize_impl template auto gstringize(T&& item) { - // use decay so we don't require separate specializations for T, const - // T, T&, const T& ... + // use decay so we don't require separate specializations + // for T, const T, T&, const T& ... return gstringize_impl>()(std::forward(item)); } -- cgit v1.3 From 30cf50e6af3183680bd6413573eecd95b1f4fbb5 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 23 Nov 2021 14:25:16 -0500 Subject: SL-16094: Support ll_convert(const char*) and correspondingly, ll_convert(const wchar_t*). Now that we're using ll_convert() for single-argument stringize(arg), make sure it can efficiently handle the simple case of constructing a string from a const char pointer. --- indra/llcommon/llstring.h | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llstring.h b/indra/llcommon/llstring.h index 54e3f9ee63..d94f549480 100644 --- a/indra/llcommon/llstring.h +++ b/indra/llcommon/llstring.h @@ -529,6 +529,13 @@ struct ll_convert_impl T operator()(const T& in) const { return in; } }; +// simple construction from char* +template +struct ll_convert_impl +{ + T operator()(const typename T::value_type* in) const { return { in }; } +}; + // specialize ll_convert_impl to return EXPR #define ll_convert_alias(TO, FROM, EXPR) \ template<> \ -- cgit v1.3 From 2b96f89c2a374d72c0a8bc28a7b06ad4db7eae6e Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Tue, 23 Nov 2021 20:39:32 -0500 Subject: SL-16400: Add ThreadPool::start() method, and call it. It's sometimes important to finish other initialization before launching the threads in the ThreadPool, so make that an explicit step. In particular, we were launching the LLImageGL texture thread before initializing the GL context, resulting in all gray textures. --- indra/llcommon/threadpool.cpp | 10 +++++++--- indra/llcommon/threadpool.h | 9 +++++++++ indra/llrender/llimagegl.cpp | 5 +---- indra/llwindow/llwindowwin32.cpp | 1 + 4 files changed, 18 insertions(+), 7 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/threadpool.cpp b/indra/llcommon/threadpool.cpp index 06e0dc5bfc..ba914035e2 100644 --- a/indra/llcommon/threadpool.cpp +++ b/indra/llcommon/threadpool.cpp @@ -23,11 +23,15 @@ LL::ThreadPool::ThreadPool(const std::string& name, size_t threads, size_t capacity): mQueue(name, capacity), - mName("ThreadPool:" + name) + mName("ThreadPool:" + name), + mThreadCount(threads) +{} + +void LL::ThreadPool::start() { - for (size_t i = 0; i < threads; ++i) + for (size_t i = 0; i < mThreadCount; ++i) { - std::string tname{ STRINGIZE(mName << ':' << (i+1) << '/' << threads) }; + std::string tname{ stringize(mName, ':', (i+1), '/', mThreadCount) }; mThreads.emplace_back(tname, [this, tname]() { LL_PROFILER_SET_THREAD_NAME(tname.c_str()); diff --git a/indra/llcommon/threadpool.h b/indra/llcommon/threadpool.h index 1ca24aec58..b79c9b9090 100644 --- a/indra/llcommon/threadpool.h +++ b/indra/llcommon/threadpool.h @@ -32,6 +32,14 @@ namespace LL ThreadPool(const std::string& name, size_t threads=1, size_t capacity=1024); virtual ~ThreadPool(); + /** + * Launch the ThreadPool. Until this call, a constructed ThreadPool + * launches no threads. That permits coders to derive from ThreadPool, + * or store it as a member of some other class, but refrain from + * launching it until all other construction is complete. + */ + void start(); + /** * ThreadPool listens for application shutdown messages on the "LLApp" * LLEventPump. Call close() to shut down this ThreadPool early. @@ -54,6 +62,7 @@ namespace LL WorkQueue mQueue; std::string mName; + size_t mThreadCount; std::vector> mThreads; }; diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index eda61d3c74..894eb8c773 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -2261,6 +2261,7 @@ LLImageGLThread::LLImageGLThread(LLWindow* window) mFinished = false; mContext = mWindow->createSharedContext(); + ThreadPool::start(); } void LLImageGLThread::run() @@ -2268,10 +2269,6 @@ void LLImageGLThread::run() LL_PROFILE_ZONE_SCOPED; // We must perform setup on this thread before actually servicing our // WorkQueue, likewise cleanup afterwards. - while (mContext == nullptr) - { // HACK -- wait for mContext to be initialized since this thread will usually start before mContext is set - std::this_thread::sleep_for(std::chrono::milliseconds(1)); - } mWindow->makeContextCurrent(mContext); gGL.init(); ThreadPool::run(); diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index 51fd228619..af7b8d91f0 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -4495,6 +4495,7 @@ std::vector LLWindowWin32::getDynamicFallbackFontList() inline LLWindowWin32::LLWindowWin32Thread::LLWindowWin32Thread() : ThreadPool("Window Thread", 1, MAX_QUEUE_SIZE) { + ThreadPool::start(); } /** -- cgit v1.3 From 877a02dba1df8a5d7d9f40b04d6be834ed9864da Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 24 Nov 2021 09:38:56 -0500 Subject: SL-16094: Fix merge glitches from previous revert. --- indra/llcommon/llthreadsafequeue.h | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index a588175074..2806506550 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -85,8 +85,8 @@ public: LLThreadSafeQueue(U32 capacity = 1024); virtual ~LLThreadSafeQueue() {} - // Add an element to the queue (will block if the queue has - // reached capacity). + // Add an element to the queue (will block if the queue has reached + // capacity). // // This call will raise an interrupt error if the queue is closed while // the caller is blocked. @@ -95,6 +95,11 @@ public: // legacy name void pushFront(ElementT const & element) { return push(element); } + // Add an element to the queue (will block if the queue has reached + // capacity). Return false if the queue is closed before push is possible. + template + bool pushIfOpen(T&& element); + // Try to add an element to the queue without blocking. Returns // true only if the element was actually added. template @@ -314,8 +319,8 @@ bool LLThreadSafeQueue::push_(lock_t& lock, T&& element) template -template -void LLThreadSafeQueue::push(T&& element) +template +bool LLThreadSafeQueue::pushIfOpen(T&& element) { LL_PROFILE_ZONE_SCOPED; lock_t lock1(mLock); @@ -325,12 +330,10 @@ void LLThreadSafeQueue::push(T&& element) // drained or not: the moment either end calls close(), further push() // operations will fail. if (mClosed) - { - LLTHROW(LLThreadSafeQueueInterrupt()); - } + return false; if (push_(lock1, std::forward(element))) - return; + return true; // Storage Full. Wait for signal. mCapacityCond.wait(lock1); -- cgit v1.3 From 78d837789a3741c65c3334934d96a505a522ee43 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 24 Nov 2021 09:43:37 -0500 Subject: SL-16400: Make WorkQueue::runFor() and runUntil() stop when done. runFor(interval) and runUntil(timestamp) are intended, and documented, to run *no longer than* the specified time. Instead, the initial implementation always waited the full specified time, hoping for work to arrive. Fix that: once we clear work that's already pending, return right away. --- indra/llcommon/workqueue.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index 1e89d87cff..e7d40354aa 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -78,8 +78,8 @@ bool LL::WorkQueue::runUntil(const TimePoint& until) LL_PROFILE_ZONE_SCOPED; // Should we subtract some slop to allow for typical Work execution time? // How much slop? - Work work; - while (TimePoint::clock::now() < until && mQueue.tryPopUntil(until, work)) + // runUntil() is simply a time-bounded runPending(). + for (Work work; TimePoint::clock::now() < until && mQueue.tryPop(work); ) { callWork(work); } -- cgit v1.3 From 0b066539fe68dc5750900c3452189645c40adb45 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 24 Nov 2021 10:47:54 -0500 Subject: DRTVWR-546, SL-16220, SL-16094: Undo previous glthread branch revert. Reverting a merge is sticky: it tells git you never want to see that branch again. Merging the DRTVWR-546 branch, which contained the revert, into the glthread branch undid much of the development work on that branch. To restore it we must revert the revert. This reverts commit 029b41c0419e975bbb28454538b46dc69ce5d2ba. --- indra/llcommon/CMakeLists.txt | 3 +- indra/llcommon/llsingleton.h | 14 +- indra/llcommon/tests/threadsafeschedule_test.cpp | 4 +- indra/llcommon/tests/workqueue_test.cpp | 72 ++++- indra/llcommon/timing.cpp | 25 -- indra/llcommon/workqueue.cpp | 30 ++- indra/llcommon/workqueue.h | 197 +++++++++----- indra/llrender/llimagegl.cpp | 89 ++----- indra/llrender/llimagegl.h | 27 +- indra/llwindow/llwindowwin32.cpp | 321 +++-------------------- indra/llwindow/llwindowwin32.h | 36 +-- indra/newview/CMakeLists.txt | 3 +- indra/newview/app_settings/settings.xml | 25 ++ indra/newview/llappviewer.cpp | 51 +--- indra/newview/llmainlooprepeater.cpp | 88 ------- indra/newview/llmainlooprepeater.h | 64 ----- indra/newview/llstartup.cpp | 17 ++ indra/newview/llviewertexture.cpp | 34 ++- indra/newview/llviewertexture.h | 4 + 19 files changed, 371 insertions(+), 733 deletions(-) delete mode 100644 indra/llcommon/timing.cpp delete mode 100644 indra/newview/llmainlooprepeater.cpp delete mode 100644 indra/newview/llmainlooprepeater.h (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 9defa6b6c1..782f656406 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -119,8 +119,8 @@ set(llcommon_SOURCE_FILES lluriparser.cpp lluuid.cpp llworkerthread.cpp - timing.cpp u64.cpp + threadpool.cpp workqueue.cpp StackWalker.cpp ) @@ -256,6 +256,7 @@ set(llcommon_HEADER_FILES lockstatic.h stdtypes.h stringize.h + threadpool.h threadsafeschedule.h timer.h tuple.h diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index f85f961287..6042c0906c 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -847,14 +847,13 @@ template class LLSimpleton { public: - static T* sInstance; - - static void createInstance() - { + template + static void createInstance(ARGS&&... args) + { llassert(sInstance == nullptr); - sInstance = new T(); + sInstance = new T(std::forward(args)...); } - + static inline T* getInstance() { return sInstance; } static inline T& instance() { return *getInstance(); } static inline bool instanceExists() { return sInstance != nullptr; } @@ -864,6 +863,9 @@ public: delete sInstance; sInstance = nullptr; } + +private: + static T* sInstance; }; template diff --git a/indra/llcommon/tests/threadsafeschedule_test.cpp b/indra/llcommon/tests/threadsafeschedule_test.cpp index af67b9f492..c421cc7b1c 100644 --- a/indra/llcommon/tests/threadsafeschedule_test.cpp +++ b/indra/llcommon/tests/threadsafeschedule_test.cpp @@ -46,11 +46,11 @@ namespace tut // the real time required for each push() call. Explicitly increment // the timestamp for each one -- but since we're passing explicit // timestamps, make the queue reorder them. - queue.push(Queue::TimeTuple(Queue::Clock::now() + 20ms, "ghi")); + queue.push(Queue::TimeTuple(Queue::Clock::now() + 200ms, "ghi")); // Given the various push() overloads, you have to match the type // exactly: conversions are ambiguous. queue.push("abc"s); - queue.push(Queue::Clock::now() + 10ms, "def"); + queue.push(Queue::Clock::now() + 100ms, "def"); queue.close(); auto entry = queue.pop(); ensure_equals("failed to pop first", std::get<0>(entry), "abc"s); diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp index d5405400fd..bea3ad911b 100644 --- a/indra/llcommon/tests/workqueue_test.cpp +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -20,7 +20,10 @@ // external library headers // other Linden headers #include "../test/lltut.h" +#include "../test/catch_and_store_what_in.h" #include "llcond.h" +#include "llcoros.h" +#include "lleventcoro.h" #include "llstring.h" #include "stringize.h" @@ -138,7 +141,8 @@ namespace tut [](){ return 17; }, // Note that a postTo() *callback* can safely bind a reference to // a variable on the invoking thread, because the callback is run - // on the invoking thread. + // on the invoking thread. (Of course the bound variable must + // survive until the callback is called.) [&result](int i){ result = i; }); // this should post the callback to main qptr->runOne(); @@ -156,4 +160,70 @@ namespace tut main.runPending(); ensure_equals("failed to run string callback", alpha, "abc"); } + + template<> template<> + void object::test<5>() + { + set_test_name("postTo with void return"); + WorkQueue main("main"); + auto qptr = WorkQueue::getInstance("queue"); + std::string observe; + main.postTo( + qptr, + // The ONLY reason we can get away with binding a reference to + // 'observe' in our work callable is because we're directly + // calling qptr->runOne() on this same thread. It would be a + // mistake to do that if some other thread were servicing 'queue'. + [&observe](){ observe = "queue"; }, + [&observe](){ observe.append(";main"); }); + qptr->runOne(); + main.runOne(); + ensure_equals("failed to run both lambdas", observe, "queue;main"); + } + + template<> template<> + void object::test<6>() + { + set_test_name("waitForResult"); + std::string stored; + // Try to call waitForResult() on this thread's main coroutine. It + // should throw because the main coroutine must service the queue. + auto what{ catch_what( + [this, &stored](){ stored = queue.waitForResult( + [](){ return "should throw"; }); }) }; + ensure("lambda should not have run", stored.empty()); + ensure_not("waitForResult() should have thrown", what.empty()); + ensure(STRINGIZE("should mention waitForResult: " << what), + what.find("waitForResult") != std::string::npos); + + // Call waitForResult() on a coroutine, with a string result. + LLCoros::instance().launch( + "waitForResult string", + [this, &stored]() + { stored = queue.waitForResult( + [](){ return "string result"; }); }); + llcoro::suspend(); + // Nothing will have happened yet because, even if the coroutine did + // run immediately, all it did was to queue the inner lambda on + // 'queue'. Service it. + queue.runOne(); + llcoro::suspend(); + ensure_equals("bad waitForResult return", stored, "string result"); + + // Call waitForResult() on a coroutine, with a void callable. + stored.clear(); + bool done = false; + LLCoros::instance().launch( + "waitForResult void", + [this, &stored, &done]() + { + queue.waitForResult([&stored](){ stored = "ran"; }); + done = true; + }); + llcoro::suspend(); + queue.runOne(); + llcoro::suspend(); + ensure_equals("didn't run coroutine", stored, "ran"); + ensure("void waitForResult() didn't return", done); + } } // namespace tut diff --git a/indra/llcommon/timing.cpp b/indra/llcommon/timing.cpp deleted file mode 100644 index c2dc695ef3..0000000000 --- a/indra/llcommon/timing.cpp +++ /dev/null @@ -1,25 +0,0 @@ -/** - * @file timing.cpp - * @brief This file will be deprecated in the future. - * - * $LicenseInfo:firstyear=2000&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index e7d40354aa..c74dada2e4 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -26,8 +26,9 @@ using Mutex = LLCoros::Mutex; using Lock = LLCoros::LockType; -LL::WorkQueue::WorkQueue(const std::string& name): - super(makeName(name)) +LL::WorkQueue::WorkQueue(const std::string& name, size_t capacity): + super(makeName(name)), + mQueue(capacity) { // TODO: register for "LLApp" events so we can implicitly close() on // viewer shutdown. @@ -38,6 +39,21 @@ void LL::WorkQueue::close() mQueue.close(); } +size_t LL::WorkQueue::size() +{ + return mQueue.size(); +} + +bool LL::WorkQueue::isClosed() +{ + return mQueue.isClosed(); +} + +bool LL::WorkQueue::done() +{ + return mQueue.done(); +} + void LL::WorkQueue::runUntilClose() { try @@ -130,3 +146,13 @@ void LL::WorkQueue::error(const std::string& msg) { LL_ERRS("WorkQueue") << msg << LL_ENDL; } + +void LL::WorkQueue::checkCoroutine(const std::string& method) +{ + // By convention, the default coroutine on each thread has an empty name + // string. See also LLCoros::logname(). + if (LLCoros::getName().empty()) + { + LLTHROW(Error("Do not call " + method + " from a thread's default coroutine")); + } +} diff --git a/indra/llcommon/workqueue.h b/indra/llcommon/workqueue.h index 8e4b38c2f3..96574a18b9 100644 --- a/indra/llcommon/workqueue.h +++ b/indra/llcommon/workqueue.h @@ -12,14 +12,14 @@ #if ! defined(LL_WORKQUEUE_H) #define LL_WORKQUEUE_H +#include "llcoros.h" +#include "llexception.h" #include "llinstancetracker.h" #include "threadsafeschedule.h" #include +#include // std::current_exception #include // std::function -#include #include -#include // std::pair -#include namespace LL { @@ -45,11 +45,16 @@ namespace LL using TimedWork = Queue::TimeTuple; using Closed = Queue::Closed; + struct Error: public LLException + { + Error(const std::string& what): LLException(what) {} + }; + /** * You may omit the WorkQueue name, in which case a unique name is * synthesized; for practical purposes that makes it anonymous. */ - WorkQueue(const std::string& name = std::string()); + WorkQueue(const std::string& name = std::string(), size_t capacity=1024); /** * Since the point of WorkQueue is to pass work to some other worker @@ -59,15 +64,36 @@ namespace LL */ void close(); + /** + * WorkQueue supports multiple producers and multiple consumers. In + * the general case it's misleading to test size(), since any other + * thread might change it the nanosecond the lock is released. On that + * basis, some might argue against publishing a size() method at all. + * + * But there are two specific cases in which a test based on size() + * might be reasonable: + * + * * If you're the only producer, noticing that size() == 0 is + * meaningful. + * * If you're the only consumer, noticing that size() > 0 is + * meaningful. + */ + size_t size(); + /// producer end: are we prevented from pushing any additional items? + bool isClosed(); + /// consumer end: are we done, is the queue entirely drained? + bool done(); + /*---------------------- fire and forget API -----------------------*/ /// fire-and-forget, but at a particular (future?) time template void post(const TimePoint& time, CALLABLE&& callable) { - // Defer reifying an arbitrary CALLABLE until we hit this method. - // All other methods should accept CALLABLEs of arbitrary type to - // avoid multiple levels of std::function indirection. + // Defer reifying an arbitrary CALLABLE until we hit this or + // postIfOpen(). All other methods should accept CALLABLEs of + // arbitrary type to avoid multiple levels of std::function + // indirection. mQueue.push(TimedWork(time, std::move(callable))); } @@ -82,6 +108,47 @@ namespace LL post(TimePoint::clock::now(), std::move(callable)); } + /** + * post work for a particular time, unless the queue is closed before + * we can post + */ + template + bool postIfOpen(const TimePoint& time, CALLABLE&& callable) + { + // Defer reifying an arbitrary CALLABLE until we hit this or + // post(). All other methods should accept CALLABLEs of arbitrary + // type to avoid multiple levels of std::function indirection. + return mQueue.pushIfOpen(TimedWork(time, std::move(callable))); + } + + /** + * post work, unless the queue is closed before we can post + */ + template + bool postIfOpen(CALLABLE&& callable) + { + return postIfOpen(TimePoint::clock::now(), std::move(callable)); + } + + /** + * Post work to be run at a specified time to another WorkQueue, which + * may or may not still exist and be open. Return true if we were able + * to post. + */ + template + static bool postMaybe(weak_t target, const TimePoint& time, CALLABLE&& callable); + + /** + * Post work to another WorkQueue, which may or may not still exist + * and be open. Return true if we were able to post. + */ + template + static bool postMaybe(weak_t target, CALLABLE&& callable) + { + return postMaybe(target, TimePoint::clock::now(), + std::forward(callable)); + } + /** * Launch a callable returning bool that will trigger repeatedly at * specified interval, until the callable returns false. @@ -115,63 +182,8 @@ namespace LL // Studio compile errors that seem utterly unrelated to this source // code. template - bool postTo(WorkQueue::weak_t target, - const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback) - { - // We're being asked to post to the WorkQueue at target. - // target is a weak_ptr: have to lock it to check it. - auto tptr = target.lock(); - if (! tptr) - // can't post() if the target WorkQueue has been destroyed - return false; - - // Here we believe target WorkQueue still exists. Post to it a - // lambda that packages our callable, our callback and a weak_ptr - // to this originating WorkQueue. - tptr->post( - time, - [reply = super::getWeak(), - callable = std::move(callable), - callback = std::move(callback)] - () - { - // Call the callable in any case -- but to minimize - // copying the result, immediately bind it into a reply - // lambda. The reply lambda also binds the original - // callback, so that when we, the originating WorkQueue, - // finally receive and process the reply lambda, we'll - // call the bound callback with the bound result -- on the - // same thread that originally called postTo(). - auto rlambda = - [result = callable(), - callback = std::move(callback)] - () - { callback(std::move(result)); }; - // Check if this originating WorkQueue still exists. - // Remember, the outer lambda is now running on a thread - // servicing the target WorkQueue, and real time has - // elapsed since postTo()'s tptr->post() call. - // reply is a weak_ptr: have to lock it to check it. - auto rptr = reply.lock(); - if (rptr) - { - // Only post reply lambda if the originating WorkQueue - // still exists. If not -- who would we tell? Log it? - try - { - rptr->post(std::move(rlambda)); - } - catch (const Closed&) - { - // Originating WorkQueue might still exist, but - // might be Closed. Same thing: just discard the - // callback. - } - } - }); - // looks like we were able to post() - return true; - } + bool postTo(weak_t target, + const TimePoint& time, CALLABLE&& callable, FOLLOWUP&& callback); /** * Post work to another WorkQueue, requesting a specific callback to @@ -181,10 +193,36 @@ namespace LL * inaccessible. */ template - bool postTo(WorkQueue::weak_t target, - CALLABLE&& callable, FOLLOWUP&& callback) + bool postTo(weak_t target, CALLABLE&& callable, FOLLOWUP&& callback) + { + return postTo(target, TimePoint::clock::now(), + std::move(callable), std::move(callback)); + } + + /** + * Post work to another WorkQueue to be run at a specified time, + * blocking the calling coroutine until then, returning the result to + * caller on completion. + * + * In general, we assume that each thread's default coroutine is busy + * servicing its WorkQueue or whatever. To try to prevent mistakes, we + * forbid calling waitForResult() from a thread's default coroutine. + */ + template + auto waitForResult(const TimePoint& time, CALLABLE&& callable); + + /** + * Post work to another WorkQueue, blocking the calling coroutine + * until then, returning the result to caller on completion. + * + * In general, we assume that each thread's default coroutine is busy + * servicing its WorkQueue or whatever. To try to prevent mistakes, we + * forbid calling waitForResult() from a thread's default coroutine. + */ + template + auto waitForResult(CALLABLE&& callable) { - return postTo(target, TimePoint::clock::now(), std::move(callable), std::move(callback)); + return waitForResult(TimePoint::clock::now(), std::move(callable)); } /*--------------------------- worker API ---------------------------*/ @@ -233,6 +271,23 @@ namespace LL bool runUntil(const TimePoint& until); private: + template + static auto makeReplyLambda(CALLABLE&& callable, FOLLOWUP&& callback); + /// general case: arbitrary C++ return type + template + struct MakeReplyLambda; + /// specialize for CALLABLE returning void + template + struct MakeReplyLambda; + + /// general case: arbitrary C++ return type + template + struct WaitForResult; + /// specialize for CALLABLE returning void + template + struct WaitForResult; + + static void checkCoroutine(const std::string& method); static void error(const std::string& msg); static std::string makeName(const std::string& name); void callWork(const Queue::DataTuple& work); @@ -254,8 +309,8 @@ namespace LL { public: // bind the desired data - BackJack(WorkQueue::weak_t target, - const WorkQueue::TimePoint& start, + BackJack(weak_t target, + const TimePoint& start, const std::chrono::duration& interval, CALLABLE&& callable): mTarget(target), @@ -302,8 +357,8 @@ namespace LL } private: - WorkQueue::weak_t mTarget; - WorkQueue::TimePoint mStart; + weak_t mTarget; + TimePoint mStart; std::chrono::duration mInterval; CALLABLE mCallable; }; diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index ed0e3fb345..894eb8c773 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -172,31 +172,19 @@ BOOL is_little_endian() return (*c == 0x78) ; } -LLImageGLThread* LLImageGLThread::sInstance = nullptr; - //static void LLImageGL::initClass(LLWindow* window, S32 num_catagories, BOOL skip_analyze_alpha /* = false */) { LL_PROFILE_ZONE_SCOPED; sSkipAnalyzeAlpha = skip_analyze_alpha; - LLImageGLThread::sInstance = new LLImageGLThread(window); - LLImageGLThread::sInstance->start(); -} - -//static -void LLImageGL::updateClass() -{ - LL_PROFILE_ZONE_SCOPED; - LLImageGLThread::sInstance->executeCallbacks(); + LLImageGLThread::createInstance(window); } //static void LLImageGL::cleanupClass() { LL_PROFILE_ZONE_SCOPED; - LLImageGLThread::sInstance->mFunctionQueue.close(); - delete LLImageGLThread::sInstance; - LLImageGLThread::sInstance = nullptr; + LLImageGLThread::deleteSingleton(); } //static @@ -504,6 +492,9 @@ void LLImageGL::init(BOOL usemipmaps) #endif mCategory = -1; + + // Sometimes we have to post work for the main thread. + mMainQueue = LL::WorkQueue::getInstance("mainloop"); } void LLImageGL::cleanup() @@ -1536,8 +1527,7 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_ } //if we're on the image loading thread, be sure to delete old_texname and update mTexName on the main thread - if (LLImageGLThread::sInstance != nullptr && - LLThread::currentID() == LLImageGLThread::sInstance->getID()) + if (! on_main_thread()) { { LL_PROFILE_ZONE_NAMED("cglt - sync"); @@ -1554,7 +1544,9 @@ BOOL LLImageGL::createGLTexture(S32 discard_level, const U8* data_in, BOOL data_ } ref(); - LLImageGLThread::sInstance->postCallback([=]() + LL::WorkQueue::postMaybe( + mMainQueue, + [=]() { LL_PROFILE_ZONE_NAMED("cglt - delete callback"); if (old_texname != 0) @@ -2259,7 +2251,11 @@ void LLImageGL::resetCurTexSizebar() */ LLImageGLThread::LLImageGLThread(LLWindow* window) - : LLThread("LLImageGL"), mWindow(window) + // We want exactly one thread, but a very large capacity: we never want + // anyone, especially inner-loop render code, to have to block on post() + // because we're full. + : ThreadPool("LLImageGL", 1, 1024*1024) + , mWindow(window) { LL_PROFILE_ZONE_SCOPED; mFinished = false; @@ -2268,61 +2264,6 @@ LLImageGLThread::LLImageGLThread(LLWindow* window) ThreadPool::start(); } -// post a function to be executed on the LLImageGL background thread - -bool LLImageGLThread::post(const std::function& func) -{ - try - { - mFunctionQueue.post(func); - } - catch (LLThreadSafeQueueInterrupt e) - { - return false; - } - - return true; -} - -//post a callback to be executed on the main thread - -bool LLImageGLThread::postCallback(const std::function& callback) -{ - try - { - if (!mCallbackQueue.tryPost(callback)) - { - mPendingCallbackQ.push(callback); - } - } - catch (LLThreadSafeQueueInterrupt e) - { - //thread is closing, drop request - return false; - } - - return true; -} - -void LLImageGLThread::executeCallbacks() -{ - LL_PROFILE_ZONE_SCOPED; - //executed from main thread - mCallbackQueue.runPending(); - - while (!mPendingCallbackQ.empty()) - { - if (mCallbackQueue.tryPost(mPendingCallbackQ.front())) - { - mPendingCallbackQ.pop(); - } - else - { - break; - } - } -} - void LLImageGLThread::run() { LL_PROFILE_ZONE_SCOPED; @@ -2330,7 +2271,7 @@ void LLImageGLThread::run() // WorkQueue, likewise cleanup afterwards. mWindow->makeContextCurrent(mContext); gGL.init(); - mFunctionQueue.runUntilClose(); + ThreadPool::run(); gGL.shutdown(); mWindow->destroySharedContext(mContext); } diff --git a/indra/llrender/llimagegl.h b/indra/llrender/llimagegl.h index bb46dbc639..ae773bb362 100644 --- a/indra/llrender/llimagegl.h +++ b/indra/llrender/llimagegl.h @@ -37,6 +37,7 @@ #include "llunits.h" #include "llthreadsafequeue.h" #include "llrender.h" +#include "threadpool.h" #include "workqueue.h" class LLTextureAtlas ; @@ -198,6 +199,7 @@ private: void freePickMask(); LLPointer mSaveData; // used for destroyGL/restoreGL + LL::WorkQueue::weak_t mMainQueue; U8* mPickMask; //downsampled bitmap approximation of alpha channel. NULL if no alpha channel U16 mPickMaskWidth; U16 mPickMaskHeight; @@ -271,7 +273,6 @@ public: public: static void initClass(LLWindow* window, S32 num_catagories, BOOL skip_analyze_alpha = false); - static void updateClass(); static void cleanupClass() ; private: @@ -307,34 +308,24 @@ public: }; -class LLImageGLThread : public LLThread +class LLImageGLThread : public LLSimpleton, LL::ThreadPool { public: LLImageGLThread(LLWindow* window); // post a function to be executed on the LLImageGL background thread - bool post(const std::function& func); - - //post a callback to be executed on the main thread - bool postCallback(const std::function& callback); - - void executeCallbacks(); + template + bool post(CALLABLE&& func) + { + return getQueue().postIfOpen(std::forward(func)); + } void run() override; - // Work Queue for background thread - LL::WorkQueue mFunctionQueue; - - // Work Queue for main thread (run from updateClass) - LL::WorkQueue mCallbackQueue; - +private: LLWindow* mWindow; void* mContext = nullptr; LLAtomicBool mFinished; - - std::queue> mPendingCallbackQ; - - static LLImageGLThread* sInstance; }; diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp index 149a92ffff..062dd02903 100644 --- a/indra/llwindow/llwindowwin32.cpp +++ b/indra/llwindow/llwindowwin32.cpp @@ -183,23 +183,19 @@ DWORD LLWindowWin32::sWinIMESentenceMode = IME_SMODE_AUTOMATIC; LLCoordWindow LLWindowWin32::sWinIMEWindowPosition(-1,-1); // The following class LLWinImm delegates Windows IMM APIs. -// We need this because some language versions of Windows, -// e.g., US version of Windows XP, doesn't install IMM32.DLL -// as a default, and we can't link against imm32.lib statically. -// I believe DLL loading of this type is best suited to do -// in a static initialization of a class. What I'm not sure is -// whether it follows the Linden Conding Standard... -// See http://wiki.secondlife.com/wiki/Coding_standards#Static_Members +// It was originally introduced to support US Windows XP, on which we needed +// to dynamically load IMM32.DLL and use GetProcAddress to resolve its entry +// points. Now that that's moot, we retain this wrapper only for hooks for +// metrics. class LLWinImm { public: - static bool isAvailable() { return sTheInstance.mHImmDll != NULL; } + static bool isAvailable() { return true; } public: // Wrappers for IMM API. static BOOL isIME(HKL hkl); - static HWND getDefaultIMEWnd(HWND hwnd); static HIMC getContext(HWND hwnd); static BOOL releaseContext(HWND hwnd, HIMC himc); static BOOL getOpenStatus(HIMC himc); @@ -213,236 +209,96 @@ public: static BOOL setCompositionFont(HIMC himc, LPLOGFONTW logfont); static BOOL setCandidateWindow(HIMC himc, LPCANDIDATEFORM candidate_form); static BOOL notifyIME(HIMC himc, DWORD action, DWORD index, DWORD value); - -private: - LLWinImm(); - ~LLWinImm(); - -private: - // Pointers to IMM API. - BOOL (WINAPI *mImmIsIME)(HKL); - HWND (WINAPI *mImmGetDefaultIMEWnd)(HWND); - HIMC (WINAPI *mImmGetContext)(HWND); - BOOL (WINAPI *mImmReleaseContext)(HWND, HIMC); - BOOL (WINAPI *mImmGetOpenStatus)(HIMC); - BOOL (WINAPI *mImmSetOpenStatus)(HIMC, BOOL); - BOOL (WINAPI *mImmGetConversionStatus)(HIMC, LPDWORD, LPDWORD); - BOOL (WINAPI *mImmSetConversionStatus)(HIMC, DWORD, DWORD); - BOOL (WINAPI *mImmGetCompostitionWindow)(HIMC, LPCOMPOSITIONFORM); - BOOL (WINAPI *mImmSetCompostitionWindow)(HIMC, LPCOMPOSITIONFORM); - LONG (WINAPI *mImmGetCompositionString)(HIMC, DWORD, LPVOID, DWORD); - BOOL (WINAPI *mImmSetCompositionString)(HIMC, DWORD, LPVOID, DWORD, LPVOID, DWORD); - BOOL (WINAPI *mImmSetCompositionFont)(HIMC, LPLOGFONTW); - BOOL (WINAPI *mImmSetCandidateWindow)(HIMC, LPCANDIDATEFORM); - BOOL (WINAPI *mImmNotifyIME)(HIMC, DWORD, DWORD, DWORD); - -private: - HMODULE mHImmDll; - static LLWinImm sTheInstance; }; -LLWinImm LLWinImm::sTheInstance; - -LLWinImm::LLWinImm() : mHImmDll(NULL) -{ - // Check system metrics - if ( !GetSystemMetrics( SM_IMMENABLED ) ) - return; - - mHImmDll = LoadLibraryA("Imm32"); - if (mHImmDll != NULL) - { - mImmIsIME = (BOOL (WINAPI *)(HKL)) GetProcAddress(mHImmDll, "ImmIsIME"); - mImmGetDefaultIMEWnd = (HWND (WINAPI *)(HWND)) GetProcAddress(mHImmDll, "ImmGetDefaultIMEWnd"); - mImmGetContext = (HIMC (WINAPI *)(HWND)) GetProcAddress(mHImmDll, "ImmGetContext"); - mImmReleaseContext = (BOOL (WINAPI *)(HWND, HIMC)) GetProcAddress(mHImmDll, "ImmReleaseContext"); - mImmGetOpenStatus = (BOOL (WINAPI *)(HIMC)) GetProcAddress(mHImmDll, "ImmGetOpenStatus"); - mImmSetOpenStatus = (BOOL (WINAPI *)(HIMC, BOOL)) GetProcAddress(mHImmDll, "ImmSetOpenStatus"); - mImmGetConversionStatus = (BOOL (WINAPI *)(HIMC, LPDWORD, LPDWORD)) GetProcAddress(mHImmDll, "ImmGetConversionStatus"); - mImmSetConversionStatus = (BOOL (WINAPI *)(HIMC, DWORD, DWORD)) GetProcAddress(mHImmDll, "ImmSetConversionStatus"); - mImmGetCompostitionWindow = (BOOL (WINAPI *)(HIMC, LPCOMPOSITIONFORM)) GetProcAddress(mHImmDll, "ImmGetCompositionWindow"); - mImmSetCompostitionWindow = (BOOL (WINAPI *)(HIMC, LPCOMPOSITIONFORM)) GetProcAddress(mHImmDll, "ImmSetCompositionWindow"); - mImmGetCompositionString= (LONG (WINAPI *)(HIMC, DWORD, LPVOID, DWORD)) GetProcAddress(mHImmDll, "ImmGetCompositionStringW"); - mImmSetCompositionString= (BOOL (WINAPI *)(HIMC, DWORD, LPVOID, DWORD, LPVOID, DWORD)) GetProcAddress(mHImmDll, "ImmSetCompositionStringW"); - mImmSetCompositionFont = (BOOL (WINAPI *)(HIMC, LPLOGFONTW)) GetProcAddress(mHImmDll, "ImmSetCompositionFontW"); - mImmSetCandidateWindow = (BOOL (WINAPI *)(HIMC, LPCANDIDATEFORM)) GetProcAddress(mHImmDll, "ImmSetCandidateWindow"); - mImmNotifyIME = (BOOL (WINAPI *)(HIMC, DWORD, DWORD, DWORD)) GetProcAddress(mHImmDll, "ImmNotifyIME"); - - if (mImmIsIME == NULL || - mImmGetDefaultIMEWnd == NULL || - mImmGetContext == NULL || - mImmReleaseContext == NULL || - mImmGetOpenStatus == NULL || - mImmSetOpenStatus == NULL || - mImmGetConversionStatus == NULL || - mImmSetConversionStatus == NULL || - mImmGetCompostitionWindow == NULL || - mImmSetCompostitionWindow == NULL || - mImmGetCompositionString == NULL || - mImmSetCompositionString == NULL || - mImmSetCompositionFont == NULL || - mImmSetCandidateWindow == NULL || - mImmNotifyIME == NULL) - { - // If any of the above API entires are not found, we can't use IMM API. - // So, turn off the IMM support. We should log some warning message in - // the case, since it is very unusual; these APIs are available from - // the beginning, and all versions of IMM32.DLL should have them all. - // Unfortunately, this code may be executed before initialization of - // the logging channel (LL_WARNS()), and we can't do it here... Yes, this - // is one of disadvantages to use static constraction to DLL loading. - FreeLibrary(mHImmDll); - mHImmDll = NULL; - - // If we unload the library, make sure all the function pointers are cleared - mImmIsIME = NULL; - mImmGetDefaultIMEWnd = NULL; - mImmGetContext = NULL; - mImmReleaseContext = NULL; - mImmGetOpenStatus = NULL; - mImmSetOpenStatus = NULL; - mImmGetConversionStatus = NULL; - mImmSetConversionStatus = NULL; - mImmGetCompostitionWindow = NULL; - mImmSetCompostitionWindow = NULL; - mImmGetCompositionString = NULL; - mImmSetCompositionString = NULL; - mImmSetCompositionFont = NULL; - mImmSetCandidateWindow = NULL; - mImmNotifyIME = NULL; - } - } -} - - // static BOOL LLWinImm::isIME(HKL hkl) { - if ( sTheInstance.mImmIsIME ) - return sTheInstance.mImmIsIME(hkl); - return FALSE; + return ImmIsIME(hkl); } // static HIMC LLWinImm::getContext(HWND hwnd) { - if ( sTheInstance.mImmGetContext ) - return sTheInstance.mImmGetContext(hwnd); - return 0; + return ImmGetContext(hwnd); } //static BOOL LLWinImm::releaseContext(HWND hwnd, HIMC himc) { - if ( sTheInstance.mImmIsIME ) - return sTheInstance.mImmReleaseContext(hwnd, himc); - return FALSE; + return ImmReleaseContext(hwnd, himc); } // static BOOL LLWinImm::getOpenStatus(HIMC himc) { - if ( sTheInstance.mImmGetOpenStatus ) - return sTheInstance.mImmGetOpenStatus(himc); - return FALSE; + return ImmGetOpenStatus(himc); } // static BOOL LLWinImm::setOpenStatus(HIMC himc, BOOL status) { - if ( sTheInstance.mImmSetOpenStatus ) - return sTheInstance.mImmSetOpenStatus(himc, status); - return FALSE; + return ImmSetOpenStatus(himc, status); } // static BOOL LLWinImm::getConversionStatus(HIMC himc, LPDWORD conversion, LPDWORD sentence) { - if ( sTheInstance.mImmGetConversionStatus ) - return sTheInstance.mImmGetConversionStatus(himc, conversion, sentence); - return FALSE; + return ImmGetConversionStatus(himc, conversion, sentence); } // static BOOL LLWinImm::setConversionStatus(HIMC himc, DWORD conversion, DWORD sentence) { - if ( sTheInstance.mImmSetConversionStatus ) - return sTheInstance.mImmSetConversionStatus(himc, conversion, sentence); - return FALSE; + return ImmSetConversionStatus(himc, conversion, sentence); } // static BOOL LLWinImm::getCompositionWindow(HIMC himc, LPCOMPOSITIONFORM form) { - if ( sTheInstance.mImmGetCompostitionWindow ) - return sTheInstance.mImmGetCompostitionWindow(himc, form); - return FALSE; + return ImmGetCompositionWindow(himc, form); } // static BOOL LLWinImm::setCompositionWindow(HIMC himc, LPCOMPOSITIONFORM form) { - if ( sTheInstance.mImmSetCompostitionWindow ) - return sTheInstance.mImmSetCompostitionWindow(himc, form); - return FALSE; + return ImmSetCompositionWindow(himc, form); } // static LONG LLWinImm::getCompositionString(HIMC himc, DWORD index, LPVOID data, DWORD length) { - if ( sTheInstance.mImmGetCompositionString ) - return sTheInstance.mImmGetCompositionString(himc, index, data, length); - return FALSE; + return ImmGetCompositionString(himc, index, data, length); } // static BOOL LLWinImm::setCompositionString(HIMC himc, DWORD index, LPVOID pComp, DWORD compLength, LPVOID pRead, DWORD readLength) { - if ( sTheInstance.mImmSetCompositionString ) - return sTheInstance.mImmSetCompositionString(himc, index, pComp, compLength, pRead, readLength); - return FALSE; + return ImmSetCompositionString(himc, index, pComp, compLength, pRead, readLength); } // static BOOL LLWinImm::setCompositionFont(HIMC himc, LPLOGFONTW pFont) { - if ( sTheInstance.mImmSetCompositionFont ) - return sTheInstance.mImmSetCompositionFont(himc, pFont); - return FALSE; + return ImmSetCompositionFont(himc, pFont); } // static BOOL LLWinImm::setCandidateWindow(HIMC himc, LPCANDIDATEFORM form) { - if ( sTheInstance.mImmSetCandidateWindow ) - return sTheInstance.mImmSetCandidateWindow(himc, form); - return FALSE; + return ImmSetCandidateWindow(himc, form); } // static BOOL LLWinImm::notifyIME(HIMC himc, DWORD action, DWORD index, DWORD value) { - if ( sTheInstance.mImmNotifyIME ) - return sTheInstance.mImmNotifyIME(himc, action, index, value); - return FALSE; + return ImmNotifyIME(himc, action, index, value); } - -// ---------------------------------------------------------------------------------------- -LLWinImm::~LLWinImm() -{ - if (mHImmDll != NULL) - { - FreeLibrary(mHImmDll); - mHImmDll = NULL; - } -} - - class LLMonitorInfo { public: @@ -552,8 +408,7 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, : LLWindow(callbacks, fullscreen, flags) { sMainThreadId = LLThread::currentID(); - mWindowThread = new LLWindowWin32Thread(this); - mWindowThread->start(); + mWindowThread = new LLWindowWin32Thread(); //MAINT-516 -- force a load of opengl32.dll just in case windows went sideways LoadLibrary(L"opengl32.dll"); @@ -624,7 +479,6 @@ LLWindowWin32::LLWindowWin32(LLWindowCallbacks* callbacks, // Make an instance of our window then define the window class mhInstance = GetModuleHandle(NULL); - mWndProc = NULL; // Init Direct Input - needed for joystick / Spacemouse @@ -1048,17 +902,13 @@ void LLWindowWin32::close() // Something killed the window while we were busy destroying gl or handle somehow got broken LL_WARNS("Window") << "Failed to destroy Window, invalid handle!" << LL_ENDL; } - mWindowHandle = NULL; - mWindowThread->mFinished = true; }); - - while (!mWindowThread->isStopped()) - { - //nudge window thread - PostMessage(mWindowHandle, WM_USER + 0x0017, 0xB0B0, 0x1337); - std::this_thread::sleep_for(std::chrono::milliseconds(1)); - } + // Even though the above lambda might not yet have run, we've already + // bound mWindowHandle into it by value, which should suffice for the + // operations we're asking. That's the last time WE should touch it. + mWindowHandle = NULL; + mWindowThread->close(); } BOOL LLWindowWin32::isValid() @@ -1351,51 +1201,7 @@ BOOL LLWindowWin32::switchContext(BOOL fullscreen, const LLCoordScreen& size, BO << " Fullscreen: " << mFullscreen << LL_ENDL; - auto oldHandle = mWindowHandle; - - //zero out mWindowHandle and mhDC before destroying window so window thread falls back to peekmessage - mWindowHandle = 0; - mhDC = 0; - - if (oldHandle && !destroy_window_handler(oldHandle)) - { - LL_WARNS("Window") << "Failed to properly close window before recreating it!" << LL_ENDL; - } - - mWindowHandle = NULL; - mhDC = 0; - - mWindowThread->post( - [this, window_rect, dw_ex_style, dw_style]() - { - mWindowHandle = CreateWindowEx(dw_ex_style, - mWindowClassName, - mWindowTitle, - WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, - window_rect.left, // x pos - window_rect.top, // y pos - window_rect.right - window_rect.left, // width - window_rect.bottom - window_rect.top, // height - NULL, - NULL, - mhInstance, - NULL); - - if (mWindowHandle) - { - mhDC = GetDC(mWindowHandle); - } - } - ); - - // HACK wait for above handle to become populated - // TODO: use a future - int count = 1024; - while (!mhDC && count > 0) - { - Sleep(10); - --count; - } + recreateWindow(window_rect, dw_ex_style, dw_style); if (mWindowHandle) { @@ -1723,48 +1529,7 @@ const S32 max_format = (S32)num_formats - 1; mhDC = 0; // Zero The Device Context } - auto oldHandle = mWindowHandle; - mWindowHandle = NULL; - mhDC = 0; - - // Destroy The Window - if (oldHandle && !destroy_window_handler(oldHandle)) - { - LL_WARNS("Window") << "Failed to properly close window!" << LL_ENDL; - } - - mWindowThread->post( - [this, window_rect, dw_ex_style, dw_style]() - { - mWindowHandle = CreateWindowEx(dw_ex_style, - mWindowClassName, - mWindowTitle, - WS_CLIPSIBLINGS | WS_CLIPCHILDREN | dw_style, - window_rect.left, // x pos - window_rect.top, // y pos - window_rect.right - window_rect.left, // width - window_rect.bottom - window_rect.top, // height - NULL, - NULL, - mhInstance, - NULL); - - if (mWindowHandle) - { - mhDC = GetDC(mWindowHandle); - } - } - ); - - // HACK wait for above handle to become populated - // TODO: use a future - int count = 1024; - while (!mhDC && count > 0) - { - PostMessage(oldHandle, WM_USER + 8, 0x1717, 0x3b3b); - Sleep(10); - --count; - } + recreateWindow(window_rect, dw_ex_style, dw_style); if (mWindowHandle) { @@ -2347,7 +2112,7 @@ void LLWindowWin32::gatherInput() } - if (mWindowThread->mFunctionQueue.size() > 0) + if (mWindowThread->getQueue().size()) { LL_PROFILE_ZONE_NAMED("gi - PostMessage"); kickWindowThread(); @@ -2455,17 +2220,6 @@ LRESULT CALLBACK LLWindowWin32::mainWindowProc(HWND h_wnd, UINT u_msg, WPARAM w_ if (NULL != window_imp) { - // Has user provided their own window callback? - if (NULL != window_imp->mWndProc) - { - LL_PROFILE_ZONE_NAMED("mwp - WndProc"); - if (!window_imp->mWndProc(h_wnd, u_msg, w_param, l_param)) - { - // user has handled window message - return 0; - } - } - // Juggle to make sure we can get negative positions for when // mouse is outside window. LLCoordWindow window_coord((S32)(S16)LOWORD(l_param), (S32)(S16)HIWORD(l_param)); @@ -4746,10 +4500,8 @@ std::vector LLWindowWin32::getDynamicFallbackFontList() #endif // LL_WINDOWS -inline LLWindowWin32Thread::LLWindowWin32Thread(LLWindowWin32* window) - : LLThread("Window Thread"), - mWindow(window), - mFunctionQueue(MAX_QUEUE_SIZE) +inline LLWindowWin32::LLWindowWin32Thread::LLWindowWin32Thread() + : ThreadPool("Window Thread", 1, MAX_QUEUE_SIZE) { ThreadPool::start(); } @@ -4813,7 +4565,7 @@ void LLWindowWin32::LLWindowWin32Thread::run() { MSG msg; BOOL status; - if (mWindow->mhDC == 0) + if (mhDC == 0) { LL_PROFILE_ZONE_NAMED("w32t - PeekMessage"); logger.onChange("PeekMessage(", std::hex, mWindowHandle, ")"); @@ -4840,11 +4592,7 @@ void LLWindowWin32::LLWindowWin32Thread::run() LL_PROFILE_ZONE_NAMED("w32t - Function Queue"); logger.onChange("runPending()"); //process any pending functions - std::function curFunc; - while (mFunctionQueue.tryPopBack(curFunc)) - { - curFunc(); - } + getQueue().runPending(); } #if 0 @@ -4857,11 +4605,6 @@ void LLWindowWin32::LLWindowWin32Thread::run() } } -void LLWindowWin32Thread::post(const std::function& func) -{ - mFunctionQueue.pushFront(func); -} - void LLWindowWin32::post(const std::function& func) { mFunctionQueue.pushFront(func); diff --git a/indra/llwindow/llwindowwin32.h b/indra/llwindow/llwindowwin32.h index 8d0193abc8..b02815e990 100644 --- a/indra/llwindow/llwindowwin32.h +++ b/indra/llwindow/llwindowwin32.h @@ -36,44 +36,12 @@ #include "llthread.h" #include "llthreadsafequeue.h" #include "llmutex.h" +#include "workqueue.h" // Hack for async host by name #define LL_WM_HOST_RESOLVED (WM_APP + 1) typedef void (*LLW32MsgCallback)(const MSG &msg); -class LLWindowWin32; - -// Thread that owns the Window Handle -class LLWindowWin32Thread : public LLThread -{ -public: - class Message - { - public: - LRESULT mMsg; - }; - - static const int MAX_QUEUE_SIZE = 2048; - - LLThreadSafeQueue mMessageQueue; - LLThreadSafeQueue> mFunctionQueue; - - bool mFinished = false; - - LLWindowWin32Thread(LLWindowWin32* window); - - void run() override; - - void post(const std::function& func); - -private: - - // call PeekMessage and pull enqueue messages for later processing - void gatherInput(); - LLWindowWin32* mWindow = nullptr; - -}; - class LLWindowWin32 : public LLWindow { public: @@ -218,7 +186,6 @@ protected: HGLRC mhRC = 0; // OpenGL rendering context HDC mhDC = 0; // Windows Device context handle HINSTANCE mhInstance; // handle to application instance - WNDPROC mWndProc; // user-installable window proc RECT mOldMouseClip; // Screen rect to which the mouse cursor was globally constrained before we changed it in clipMouse() WPARAM mLastSizeWParam; F32 mOverrideAspectRatio; @@ -280,7 +247,6 @@ protected: void kickWindowThread(HWND windowHandle=0); friend class LLWindowManager; - friend class LLWindowWin32Thread; }; class LLSplashScreenWin32 : public LLSplashScreen diff --git a/indra/newview/CMakeLists.txt b/indra/newview/CMakeLists.txt index 9b636e5e5d..5f085bb9ad 100644 --- a/indra/newview/CMakeLists.txt +++ b/indra/newview/CMakeLists.txt @@ -393,7 +393,6 @@ set(viewer_SOURCE_FILES llloginhandler.cpp lllogininstance.cpp llmachineid.cpp - llmainlooprepeater.cpp llmanip.cpp llmaniprotate.cpp llmanipscale.cpp @@ -1032,7 +1031,6 @@ set(viewer_HEADER_FILES llloginhandler.h lllogininstance.h llmachineid.h - llmainlooprepeater.h llmanip.h llmaniprotate.h llmanipscale.h @@ -1604,6 +1602,7 @@ if (WINDOWS) ${WINDOWS_LIBRARIES} comdlg32 dxguid + imm32 kernel32 odbc32 odbccp32 diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index 2d821b7451..058da4b66d 100644 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -3871,6 +3871,17 @@ Value 1 + MainWorkTime + + Comment + Max time per frame devoted to mainloop work queue (in milliseconds) + Persist + 1 + Type + F32 + Value + 0.1 + QueueInventoryFetchTimeout Comment @@ -12667,6 +12678,20 @@ Value + ThreadPoolSizes + + Comment + Map of size overrides for specific thread pools. + Persist + 1 + Type + LLSD + Value + + General + 4 + + ThrottleBandwidthKBPS Comment diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 5b3cbff9b8..e13c0a2472 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -233,11 +233,12 @@ #include "llavatariconctrl.h" #include "llgroupiconctrl.h" #include "llviewerassetstats.h" +#include "workqueue.h" +using namespace LL; // Include for security api initialization #include "llsecapi.h" #include "llmachineid.h" -#include "llmainlooprepeater.h" #include "llcleanup.h" #include "llcoproceduremanager.h" @@ -366,6 +367,10 @@ BOOL gLogoutInProgress = FALSE; BOOL gSimulateMemLeak = FALSE; +// We don't want anyone, especially threads working on the graphics pipeline, +// to have to block due to this WorkQueue being full. +WorkQueue gMainloopWork("mainloop", 1024*1024); + //////////////////////////////////////////////////////////// // Internal globals... that should be removed. static std::string gArgs; @@ -381,42 +386,6 @@ static std::string gLaunchFileOnQuit; // Used on Win32 for other apps to identify our window (eg, win_setup) const char* const VIEWER_WINDOW_CLASSNAME = "Second Life"; -//-- LLDeferredTaskList ------------------------------------------------------ - -/** - * A list of deferred tasks. - * - * We sometimes need to defer execution of some code until the viewer gets idle, - * e.g. removing an inventory item from within notifyObservers() may not work out. - * - * Tasks added to this list will be executed in the next LLAppViewer::idle() iteration. - * All tasks are executed only once. - */ -class LLDeferredTaskList: public LLSingleton -{ - LLSINGLETON_EMPTY_CTOR(LLDeferredTaskList); - LOG_CLASS(LLDeferredTaskList); - - friend class LLAppViewer; - typedef boost::signals2::signal signal_t; - - void addTask(const signal_t::slot_type& cb) - { - mSignal.connect(cb); - } - - void run() - { - if (!mSignal.empty()) - { - mSignal(); - mSignal.disconnect_all_slots(); - } - } - - signal_t mSignal; -}; - //---------------------------------------------------------------------------- // List of entries from strings.xml to always replace @@ -973,9 +942,6 @@ bool LLAppViewer::init() } LL_INFOS("InitInfo") << "Cache initialization is done." << LL_ENDL ; - // Initialize the repeater service. - LLMainLoopRepeater::instance().start(); - // Initialize event recorder LLViewerEventRecorder::createInstance(); @@ -2217,8 +2183,6 @@ bool LLAppViewer::cleanup() SUBSYSTEM_CLEANUP(LLProxy); LLCore::LLHttp::cleanup(); - LLMainLoopRepeater::instance().stop(); - ll_close_fail_log(); LLError::LLCallStacks::cleanup(); @@ -4550,7 +4514,7 @@ bool LLAppViewer::initCache() void LLAppViewer::addOnIdleCallback(const boost::function& cb) { - LLDeferredTaskList::instance().addTask(cb); + gMainloopWork.post(cb); } void LLAppViewer::loadKeyBindings() @@ -4948,7 +4912,6 @@ void LLAppViewer::idle() LLNotificationsUI::LLToast::updateClass(); LLSmoothInterpolation::updateInterpolants(); LLMortician::updateClass(); - LLImageGL::updateClass(); LLFilePickerThread::clearDead(); //calls LLFilePickerThread::notify() LLDirPickerThread::clearDead(); F32 dt_raw = idle_timer.getElapsedTimeAndResetF32(); diff --git a/indra/newview/llmainlooprepeater.cpp b/indra/newview/llmainlooprepeater.cpp deleted file mode 100644 index 6736e9a950..0000000000 --- a/indra/newview/llmainlooprepeater.cpp +++ /dev/null @@ -1,88 +0,0 @@ -/** - * @file llmachineid.cpp - * @brief retrieves unique machine ids - * - * $LicenseInfo:firstyear=2009&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#include "llviewerprecompiledheaders.h" -#include "llapr.h" -#include "llevents.h" -#include "llmainlooprepeater.h" - - - -// LLMainLoopRepeater -//----------------------------------------------------------------------------- - - -LLMainLoopRepeater::LLMainLoopRepeater(void): - mQueue(0) -{ - ; // No op. -} - - -void LLMainLoopRepeater::start(void) -{ - if(mQueue != 0) return; - - mQueue = new LLThreadSafeQueue(1024); - mMainLoopConnection = LLEventPumps::instance(). - obtain("mainloop").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMainLoop, this, _1)); - mRepeaterConnection = LLEventPumps::instance(). - obtain("mainlooprepeater").listen(LLEventPump::inventName(), boost::bind(&LLMainLoopRepeater::onMessage, this, _1)); -} - - -void LLMainLoopRepeater::stop(void) -{ - mMainLoopConnection.release(); - mRepeaterConnection.release(); - - delete mQueue; - mQueue = 0; -} - - -bool LLMainLoopRepeater::onMainLoop(LLSD const &) -{ - LLSD message; - while(mQueue->tryPopBack(message)) { - std::string pump = message["pump"].asString(); - if(pump.length() == 0 ) continue; // No pump. - LLEventPumps::instance().obtain(pump).post(message["payload"]); - } - return false; -} - - -bool LLMainLoopRepeater::onMessage(LLSD const & event) -{ - try { - mQueue->pushFront(event); - } catch(LLThreadSafeQueueError & e) { - LL_WARNS() << "could not repeat message (" << e.what() << ")" << - event.asString() << LL_ENDL; - } - return false; -} diff --git a/indra/newview/llmainlooprepeater.h b/indra/newview/llmainlooprepeater.h deleted file mode 100644 index 2ec3a74e4a..0000000000 --- a/indra/newview/llmainlooprepeater.h +++ /dev/null @@ -1,64 +0,0 @@ -/** - * @file llmainlooprepeater.h - * @brief a service for repeating messages on the main loop. - * - * $LicenseInfo:firstyear=2010&license=viewerlgpl$ - * Second Life Viewer Source Code - * Copyright (C) 2010, Linden Research, Inc. - * - * This library is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; - * version 2.1 of the License only. - * - * This library is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#ifndef LL_LLMAINLOOPREPEATER_H -#define LL_LLMAINLOOPREPEATER_H - - -#include "llsd.h" -#include "llthreadsafequeue.h" - - -// -// A service which creates the pump 'mainlooprepeater' to which any thread can -// post a message that will be re-posted on the main loop. -// -// The posted message should contain two map elements: pump and payload. The -// pump value is a string naming the pump to which the message should be -// re-posted. The payload value is what will be posted to the designated pump. -// -class LLMainLoopRepeater: - public LLSingleton -{ - LLSINGLETON(LLMainLoopRepeater); -public: - // Start the repeater service. - void start(void); - - // Stop the repeater service. - void stop(void); - -private: - LLTempBoundListener mMainLoopConnection; - LLTempBoundListener mRepeaterConnection; - LLThreadSafeQueue * mQueue; - - bool onMainLoop(LLSD const &); - bool onMessage(LLSD const & event); -}; - - -#endif diff --git a/indra/newview/llstartup.cpp b/indra/newview/llstartup.cpp index 8d21b04511..df066fb7ed 100644 --- a/indra/newview/llstartup.cpp +++ b/indra/newview/llstartup.cpp @@ -206,6 +206,9 @@ #include "llstacktrace.h" +#include "threadpool.h" + + #if LL_WINDOWS #include "lldxhardware.h" #endif @@ -303,6 +306,20 @@ void callback_cache_name(const LLUUID& id, const std::string& full_name, bool is // local classes // +void launchThreadPool() +{ + LLSD poolSizes{ gSavedSettings.getLLSD("ThreadPoolSizes") }; + LLSD sizeSpec{ poolSizes["General"] }; + LLSD::Integer size{ sizeSpec.isInteger()? sizeSpec.asInteger() : 3 }; + LL_DEBUGS("ThreadPool") << "Instantiating General pool with " + << size << " threads" << LL_ENDL; + // Use a function-static ThreadPool: static duration, but instantiated + // only on demand. + // We don't want anyone, especially the main thread, to have to block + // due to this ThreadPool being full. + static LL::ThreadPool pool("General", size, 1024*1024); +} + void update_texture_fetch() { LLAppViewer::getTextureCache()->update(1); // unpauses the texture cache thread diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index e6ac701644..f932acd48c 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -679,6 +679,9 @@ void LLViewerTexture::init(bool firstinit) mVolumeList[LLRender::LIGHT_TEX].clear(); mVolumeList[LLRender::SCULPT_TEX].clear(); + + mMainQueue = LL::WorkQueue::getInstance("mainloop"); + mImageQueue = LL::WorkQueue::getInstance("LLImageGL"); } //virtual @@ -1622,17 +1625,26 @@ void LLViewerFetchedTexture::scheduleCreateTexture() { mNeedsCreateTexture = TRUE; #if LL_WINDOWS //flip to 0 to revert to single-threaded OpenGL texture uploads - if (!LLImageGLThread::sInstance->post([this]() - { - //actually create the texture on a background thread - createTexture(); - LLImageGLThread::sInstance->postCallback([this]() - { - //finalize on main thread - postCreateTexture(); - unref(); - }); - })) + auto mainq = mMainQueue.lock(); + if (mainq) + { + mainq->postTo( + mImageQueue, + // work to be done on LLImageGL worker thread + [this]() + { + //actually create the texture on a background thread + createTexture(); + }, + // callback to be run on main thread + [this]() + { + //finalize on main thread + postCreateTexture(); + unref(); + }); + } + else #endif { gTextureList.mCreateTextureList.insert(this); diff --git a/indra/newview/llviewertexture.h b/indra/newview/llviewertexture.h index f9f1bfef44..4cd4c7cd39 100644 --- a/indra/newview/llviewertexture.h +++ b/indra/newview/llviewertexture.h @@ -35,6 +35,7 @@ #include "llrender.h" #include "llmetricperformancetester.h" #include "httpcommon.h" +#include "workqueue.h" #include #include @@ -213,6 +214,9 @@ protected: //do not use LLPointer here. LLViewerMediaTexture* mParcelMedia ; + LL::WorkQueue::weak_t mMainQueue; + LL::WorkQueue::weak_t mImageQueue; + static F32 sTexelPixelRatio; public: static const U32 sCurrentFileVersion; -- cgit v1.3 From 04ebc11a2d8a2e59abda5061e35e504fc30504d2 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 24 Nov 2021 12:56:48 -0500 Subject: SL-16094: Fix WorkQueue test for correct behavior of runFor(). Turns out that one of our WorkQueue integration tests was relying on the incorrect runFor() behavior that we just fixed, so the test broke. Now that runFor() doesn't wait around for work to be posted, use an explicit wait loop instead. To support this, add LLCond::get(functor), where functor must accept a const reference to the stored data. This new get() returns whatever the functor returns, allowing a caller to peek at the stored data. Also use universal references for all remaining LLCond functor arguments. --- indra/llcommon/llcond.h | 52 +++++++++++++++++++++++---------- indra/llcommon/tests/workqueue_test.cpp | 12 ++++++-- 2 files changed, 45 insertions(+), 19 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcond.h b/indra/llcommon/llcond.h index c08acb66a1..da6e6affe1 100644 --- a/indra/llcommon/llcond.h +++ b/indra/llcommon/llcond.h @@ -67,15 +67,30 @@ public: LLCond(const LLCond&) = delete; LLCond& operator=(const LLCond&) = delete; - /// get() returns the stored DATA by value -- so to use get(), DATA must - /// be copyable. The only way to get a non-const reference -- to modify - /// the stored DATA -- is via update_one() or update_all(). + /** + * get() returns the stored DATA by value -- so to use get(), DATA must + * be copyable. The only way to get a non-const reference -- to modify + * the stored DATA -- is via update_one() or update_all(). + */ value_type get() { LockType lk(mMutex); return mData; } + /** + * get(functor) returns whatever the functor returns. It allows us to peek + * at the stored DATA without copying the whole thing. The functor must + * accept a const reference to DATA. If you want to modify DATA, call + * update_one() or update_all() instead. + */ + template + auto get(FUNC&& func) + { + LockType lk(mMutex); + return std::forward(func)(const_data()); + } + /** * Pass update_one() an invocable accepting non-const (DATA&). The * invocable will presumably modify the referenced DATA. update_one() @@ -86,11 +101,11 @@ public: * update_one() when DATA is a struct or class. */ template - void update_one(MODIFY modify) + void update_one(MODIFY&& modify) { { // scope of lock can/should end before notify_one() LockType lk(mMutex); - modify(mData); + std::forward(modify)(mData); } mCond.notify_one(); } @@ -105,11 +120,11 @@ public: * update_all() when DATA is a struct or class. */ template - void update_all(MODIFY modify) + void update_all(MODIFY&& modify) { { // scope of lock can/should end before notify_all() LockType lk(mMutex); - modify(mData); + std::forward(modify)(mData); } mCond.notify_all(); } @@ -122,7 +137,7 @@ public: * wait() on the condition_variable. */ template - void wait(Pred pred) + void wait(Pred&& pred) { LockType lk(mMutex); // We must iterate explicitly since the predicate accepted by @@ -133,7 +148,7 @@ public: // But what if they instead pass a predicate accepting non-const // (DATA&)? Such a predicate could modify mData, which would be Bad. // Forbid that. - while (! pred(const_cast(mData))) + while (! std::forward(pred)(const_data())) { mCond.wait(lk); } @@ -150,7 +165,7 @@ public: * returning true. */ template - bool wait_for(const std::chrono::duration& timeout_duration, Pred pred) + bool wait_for(const std::chrono::duration& timeout_duration, Pred&& pred) { // Instead of replicating wait_until() logic, convert duration to // time_point and just call wait_until(). @@ -159,7 +174,8 @@ public: // wrong! We'd keep pushing the timeout time farther and farther into // the future. This way, we establish a definite timeout time and // stick to it. - return wait_until(std::chrono::steady_clock::now() + timeout_duration, pred); + return wait_until(std::chrono::steady_clock::now() + timeout_duration, + std::forward(pred)); } /** @@ -169,9 +185,9 @@ public: * generic wait_for() method. */ template - bool wait_for(F32Milliseconds timeout_duration, Pred pred) + bool wait_for(F32Milliseconds timeout_duration, Pred&& pred) { - return wait_for(convert(timeout_duration), pred); + return wait_for(convert(timeout_duration), std::forward(pred)); } protected: @@ -189,6 +205,10 @@ protected: } private: + // It's important to pass a const ref to certain user-specified functors + // that aren't supposed to be able to modify mData. + const value_type& const_data() const { return mData; } + /** * Pass wait_until() a chrono::time_point, indicating the time at which we * should stop waiting, and a predicate accepting (const DATA&), returning @@ -209,21 +229,21 @@ private: * honoring a fixed timeout. */ template - bool wait_until(const std::chrono::time_point& timeout_time, Pred pred) + bool wait_until(const std::chrono::time_point& timeout_time, Pred&& pred) { LockType lk(mMutex); // We advise the caller to pass a predicate accepting (const DATA&). // But what if they instead pass a predicate accepting non-const // (DATA&)? Such a predicate could modify mData, which would be Bad. // Forbid that. - while (! pred(const_cast(mData))) + while (! std::forward(pred)(const_data())) { if (cv_status::timeout == mCond.wait_until(lk, timeout_time)) { // It's possible that wait_until() timed out AND the predicate // became true more or less simultaneously. Even though // wait_until() timed out, check the predicate one more time. - return pred(const_cast(mData)); + return std::forward(pred)(const_data()); } } return true; diff --git a/indra/llcommon/tests/workqueue_test.cpp b/indra/llcommon/tests/workqueue_test.cpp index bea3ad911b..1d73f7aa0d 100644 --- a/indra/llcommon/tests/workqueue_test.cpp +++ b/indra/llcommon/tests/workqueue_test.cpp @@ -99,9 +99,15 @@ namespace tut return (++count < 3); }); // no convenient way to close() our queue while we've got a - // postEvery() running, so run until we think we should have exhausted - // the iterations - queue.runFor(10*interval); + // postEvery() running, so run until we have exhausted the iterations + // or we time out waiting + for (auto finish = start + 10*interval; + WorkQueue::TimePoint::clock::now() < finish && + data.get([](const Shared& data){ return data.size(); }) < 3; ) + { + queue.runPending(); + std::this_thread::sleep_for(interval/10); + } // Take a copy of the captured deque. Shared result = data.get(); ensure_equals("called wrong number of times", result.size(), 3); -- cgit v1.3 From c37c727a9b196d6c4053cadcd4c27684a4d93b8e Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 10:02:27 -0800 Subject: SL-16606: Add categories --- indra/llcommon/CMakeLists.txt | 1 + indra/llcommon/llprofilercategories.h | 280 ++++++++++++++++++++++++++++++++++ 2 files changed, 281 insertions(+) create mode 100644 indra/llcommon/llprofilercategories.h (limited to 'indra/llcommon') diff --git a/indra/llcommon/CMakeLists.txt b/indra/llcommon/CMakeLists.txt index 782f656406..ca8b5e946f 100644 --- a/indra/llcommon/CMakeLists.txt +++ b/indra/llcommon/CMakeLists.txt @@ -202,6 +202,7 @@ set(llcommon_HEADER_FILES llnametable.h llpointer.h llprofiler.h + llprofilercategories.h llpounceable.h llpredicate.h llpreprocessor.h diff --git a/indra/llcommon/llprofilercategories.h b/indra/llcommon/llprofilercategories.h new file mode 100644 index 0000000000..8db29468cc --- /dev/null +++ b/indra/llcommon/llprofilercategories.h @@ -0,0 +1,280 @@ +/** + * @file llprofiler_ategories.h + * @brief Profiling categories to minimize Tracy memory usage when viewing captures. + * + * $LicenseInfo:firstyear=2022&license=viewerlgpl$ + * Second Life Viewer Source Code + * Copyright (C) 2022, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_PROFILER_CATEGORIES_H +#define LL_PROFILER_CATEGORIES_H + +// A Tracy capture can quickly consume memory. Use these defines to selectively turn on/off Tracy profiling for these categories. +// The biggest memory usage ones are: +// +// LL_PROFILER_CATEGORY_ENABLE_DRAWPOOL +// LL_PROFILER_CATEGORY_ENABLE_LLSD +// LL_PROFILER_CATEGORY_ENABLE_MEMORY +// LL_PROFILER_CATEGORY_ENABLE_SHADERS +// +// NOTE: You can still manually use: +// LL_PROFILE_ZONE_SCOPED(); +// LL_PROFILE_ZONE_NAMED("name"); +// but just be aware that those will ALWAYS show up in a Tracy capture +// a) using more memory, and +// b) adding visual clutter. +#define LL_PROFILER_CATEGORY_ENABLE_APP 1 +#define LL_PROFILER_CATEGORY_ENABLE_AVATAR 1 +#define LL_PROFILER_CATEGORY_ENABLE_DISPLAY 1 +#define LL_PROFILER_CATEGORY_ENABLE_DRAWABLE 1 +#define LL_PROFILER_CATEGORY_ENABLE_DRAWPOOL 1 +#define LL_PROFILER_CATEGORY_ENABLE_ENVIRONMENT 1 +#define LL_PROFILER_CATEGORY_ENABLE_FACE 1 +#define LL_PROFILER_CATEGORY_ENABLE_LLSD 1 +#define LL_PROFILER_CATEGORY_ENABLE_LOGGING 1 +#define LL_PROFILER_CATEGORY_ENABLE_MATERIAL 1 +#define LL_PROFILER_CATEGORY_ENABLE_MEDIA 1 +#define LL_PROFILER_CATEGORY_ENABLE_MEMORY 1 +#define LL_PROFILER_CATEGORY_ENABLE_NETWORK 1 +#define LL_PROFILER_CATEGORY_ENABLE_OCTREE 1 +#define LL_PROFILER_CATEGORY_ENABLE_PIPELINE 1 +#define LL_PROFILER_CATEGORY_ENABLE_SHADER 1 +#define LL_PROFILER_CATEGORY_ENABLE_SPATIAL 1 +#define LL_PROFILER_CATEGORY_ENABLE_STATS 1 +#define LL_PROFILER_CATEGORY_ENABLE_STRING 1 +#define LL_PROFILER_CATEGORY_ENABLE_TEXTURE 1 +#define LL_PROFILER_CATEGORY_ENABLE_THREAD 1 +#define LL_PROFILER_CATEGORY_ENABLE_UI 1 +#define LL_PROFILER_CATEGORY_ENABLE_VIEWER 1 +#define LL_PROFILER_CATEGORY_ENABLE_VERTEX 1 +#define LL_PROFILER_CATEGORY_ENABLE_VOLUME 1 +#define LL_PROFILER_CATEGORY_ENABLE_WIN32 1 + +#if LL_PROFILER_CATEGORY_ENABLE_APP + #define LL_PROFILE_ZONE_NAMED_CATEGORY_APP LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_APP LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_APP(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_APP +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_AVATAR + #define LL_PROFILE_ZONE_NAMED_CATEGORY_AVATAR LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_AVATAR LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_AVATAR(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_AVATAR +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_DISPLAY + #define LL_PROFILE_ZONE_NAMED_CATEGORY_DISPLAY LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_DISPLAY(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_DRAWABLE + #define LL_PROFILE_ZONE_NAMED_CATEGORY_DRAWABLE LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_DRAWABLE LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_DRAWABLE(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_DRAWABLE +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_DRAWPOOL + #define LL_PROFILE_ZONE_NAMED_CATEGORY_DRAWPOOL LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_DRAWPOOL LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_DRAWPOOL(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_DRAWPOOL +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_ENVIRONMENT + #define LL_PROFILE_ZONE_NAMED_CATEGORY_ENVIRONMENT LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_ENVIRONMENT LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_ENVIRONMENT(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_ENVIRONMENT +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_FACE + #define LL_PROFILE_ZONE_NAMED_CATEGORY_FACE LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_FACE LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_FACE(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_FACE +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_LLSD + #define LL_PROFILE_ZONE_NAMED_CATEGORY_LLSD LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_LLSD(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_LOGGING + #define LL_PROFILE_ZONE_NAMED_CATEGORY_LOGGING LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_LOGGING(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_MATERIAL + #define LL_PROFILE_ZONE_NAMED_CATEGORY_MATERIAL LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_MATERIAL LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_MATERIAL(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_MATERIAL +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_MEDIA + #define LL_PROFILE_ZONE_NAMED_CATEGORY_MEDIA LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_MEDIA LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_MEDIA(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_MEDIA +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_MEMORY + #define LL_PROFILE_ZONE_NAMED_CATEGORY_MEMORY LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_MEMORY(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_NETWORK + #define LL_PROFILE_ZONE_NAMED_CATEGORY_NETWORK LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_NETWORK LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_NETWORK(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_NETWORK +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_OCTREE + #define LL_PROFILE_ZONE_NAMED_CATEGORY_OCTREE LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_OCTREE LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_OCTREE(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_OCTREE +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_PIPELINE + #define LL_PROFILE_ZONE_NAMED_CATEGORY_PIPELINE LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_PIPELINE(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_PIPELINE +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_SHADER + #define LL_PROFILE_ZONE_NAMED_CATEGORY_SHADER LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_SHADER(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_SHADER +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_SPATIAL + #define LL_PROFILE_ZONE_NAMED_CATEGORY_SPATIAL LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_SPATIAL LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_SPATIAL(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_SPATIAL +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_STATS + #define LL_PROFILE_ZONE_NAMED_CATEGORY_STATS LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_STATS(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_STRING + #define LL_PROFILE_ZONE_NAMED_CATEGORY_STRING LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_STRING LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_STRING(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_STRING +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_TEXTURE + #define LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_TEXTURE(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_TEXTURE +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_THREAD + #define LL_PROFILE_ZONE_NAMED_CATEGORY_THREAD LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_THREAD(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_UI + #define LL_PROFILE_ZONE_NAMED_CATEGORY_UI LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_UI LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_UI(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_UI +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_VERTEX + #define LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_VIEWER + #define LL_PROFILE_ZONE_NAMED_CATEGORY_VIEWER LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_VIEWER LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_VIEWER(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_VIEWER +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_VOLUME + #define LL_PROFILE_ZONE_NAMED_CATEGORY_VOLUME LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_VOLUME(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME +#endif + +#if LL_PROFILER_CATEGORY_ENABLE_WIN32 + #define LL_PROFILE_ZONE_NAMED_CATEGORY_WIN32 LL_PROFILE_ZONE_NAMED + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_WIN32 LL_PROFILE_ZONE_SCOPED +#else + #define LL_PROFILE_ZONE_NAMED_CATEGORY_WIN32(name) + #define LL_PROFILE_ZONE_SCOPED_CATEGORY_WIN32 +#endif + +#endif // LL_PROFILER_CATEGORIES_H + -- cgit v1.3 From f9031ee02d19afe01023936eacc867dcdef01861 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 11:22:56 -0800 Subject: SL-16606: Include profiler categories automatically --- indra/llcommon/llprofiler.h | 40 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 40 insertions(+) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h index ca60d23248..f9d7ae7ce4 100644 --- a/indra/llcommon/llprofiler.h +++ b/indra/llcommon/llprofiler.h @@ -27,6 +27,44 @@ #ifndef LL_PROFILER_H #define LL_PROFILER_H +// If you use the default macros LL_PROFILE_ZONE_SCOPED and LL_PROFILE_ZONE_NAMED to profile code ... +// +// void foo() +// { +// LL_PROFILE_ZONE_SCOPED; +// : +// +// { +// LL_PROFILE_ZONE_NAMED("widget bar"); +// : +// } +// { +// LL_PROFILE_ZONE_NAMED("widget qux"); +// : +// } +// } +// +// ... please be aware that ALL these will show up in a Tracy capture which can quickly exhaust memory. +// Instead, use LL_PROFILE_ZONE_SCOPED_CATEGORY_* and LL_PROFILE_ZONE_NAMED_CATEGORY_* to profile code ... +// +// void foo() +// { +// LL_PROFILE_ZONE_SCOPED_CATEGORY_UI; +// : +// +// { +// LL_PROFILE_ZONE_NAMED_CATEGORY_UI("widget bar"); +// : +// } +// { +// LL_PROFILE_ZONE_NAMED_CATEGORY_UI("widget qux"); +// : +// } +// } +// +// ... as these can be selectively turned on/off. This will minimize memory usage and visual clutter in a Tracy capture. +// See llprofiler_categories.h for more details on profiling categories. + #define LL_PROFILER_CONFIG_NONE 0 // No profiling #define LL_PROFILER_CONFIG_FAST_TIMER 1 // Profiling on: Only Fast Timers #define LL_PROFILER_CONFIG_TRACY 2 // Profiling on: Only Tracy @@ -108,4 +146,6 @@ extern thread_local bool gProfilerEnabled; #define LL_PROFILER_SET_THREAD_NAME( name ) (void)(name) #endif // LL_PROFILER +#include "llprofilercategories.h" + #endif // LL_PROFILER_H -- cgit v1.3 From 3a9ce9f22667cf3e0a1d6def9c478a66310ef0ad Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 12:19:23 -0800 Subject: SL-16606: Add profiler category LLSD --- indra/llcommon/llsd.cpp | 22 +++++++++++----------- indra/llcommon/llsd.h | 14 +++++++------- indra/llxml/llcontrol.h | 2 +- 3 files changed, 19 insertions(+), 19 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llsd.cpp b/indra/llcommon/llsd.cpp index 605f6bf0e3..807b3d13f8 100644 --- a/indra/llcommon/llsd.cpp +++ b/indra/llcommon/llsd.cpp @@ -400,7 +400,7 @@ namespace ImplMap& ImplMap::makeMap(LLSD::Impl*& var) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; if (shared()) { ImplMap* i = new ImplMap(mData); @@ -415,21 +415,21 @@ namespace bool ImplMap::has(const LLSD::String& k) const { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; DataMap::const_iterator i = mData.find(k); return i != mData.end(); } LLSD ImplMap::get(const LLSD::String& k) const { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; DataMap::const_iterator i = mData.find(k); return (i != mData.end()) ? i->second : LLSD(); } LLSD ImplMap::getKeys() const { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; LLSD keys = LLSD::emptyArray(); DataMap::const_iterator iter = mData.begin(); while (iter != mData.end()) @@ -442,13 +442,13 @@ namespace void ImplMap::insert(const LLSD::String& k, const LLSD& v) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; mData.insert(DataMap::value_type(k, v)); } void ImplMap::erase(const LLSD::String& k) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; mData.erase(k); } @@ -690,7 +690,7 @@ const LLSD::Impl& LLSD::Impl::safe(const Impl* impl) ImplMap& LLSD::Impl::makeMap(Impl*& var) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; ImplMap* im = new ImplMap; reset(var, im); return *im; @@ -896,12 +896,12 @@ void LLSD::erase(const String& k) { makeMap(impl).erase(k); } LLSD& LLSD::operator[](const String& k) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; return makeMap(impl).ref(k); } const LLSD& LLSD::operator[](const String& k) const { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; return safe(impl).ref(k); } @@ -928,12 +928,12 @@ void LLSD::erase(Integer i) { makeArray(impl).erase(i); } LLSD& LLSD::operator[](Integer i) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; return makeArray(impl).ref(i); } const LLSD& LLSD::operator[](Integer i) const { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; return safe(impl).ref(i); } diff --git a/indra/llcommon/llsd.h b/indra/llcommon/llsd.h index b8ddf21596..24cb9bbce1 100644 --- a/indra/llcommon/llsd.h +++ b/indra/llcommon/llsd.h @@ -290,16 +290,16 @@ public: LLSD& with(const String&, const LLSD&); LLSD& operator[](const String&); - LLSD& operator[](const char* c) - { - LL_PROFILE_ZONE_SCOPED; - return (*this)[String(c)]; + LLSD& operator[](const char* c) + { + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; + return (*this)[String(c)]; } const LLSD& operator[](const String&) const; const LLSD& operator[](const char* c) const - { - LL_PROFILE_ZONE_SCOPED; - return (*this)[String(c)]; + { + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; + return (*this)[String(c)]; } //@} diff --git a/indra/llxml/llcontrol.h b/indra/llxml/llcontrol.h index 088502c017..0839c02c50 100644 --- a/indra/llxml/llcontrol.h +++ b/indra/llxml/llcontrol.h @@ -247,7 +247,7 @@ public: // generic getter template T get(const std::string& name) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_LLSD; LLControlVariable* control = getControl(name); LLSD value; eControlType type = TYPE_COUNT; -- cgit v1.3 From 6e306cd7ce77e75c98205360b7d8b38531319900 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 12:20:12 -0800 Subject: SL-16606: Add profiler category LOGGING --- indra/llcommon/llerror.cpp | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llerror.cpp b/indra/llcommon/llerror.cpp index 17a5ec5776..2fae9fdfaa 100644 --- a/indra/llcommon/llerror.cpp +++ b/indra/llcommon/llerror.cpp @@ -109,7 +109,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING int syslogPriority = LOG_CRIT; switch (level) { case LLError::LEVEL_DEBUG: syslogPriority = LOG_DEBUG; break; @@ -167,7 +167,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING if (LLError::getAlwaysFlush()) { mFile << message << std::endl; @@ -234,7 +234,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING // The default colors for error, warn and debug are now a bit more pastel // and easier to read on the default (black) terminal background but you // now have the option to set the color of each via an environment variables: @@ -274,7 +274,7 @@ namespace { LL_FORCE_INLINE void writeANSI(const std::string& ansi_code, const std::string& message) { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING static std::string s_ansi_bold = createBoldANSI(); // bold text static std::string s_ansi_reset = createResetANSI(); // reset // ANSI color code escape sequence, message, and reset in one fprintf call @@ -311,7 +311,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING mBuffer->addLine(message); } @@ -338,7 +338,7 @@ namespace { virtual void recordMessage(LLError::ELevel level, const std::string& message) override { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING debugger_print(message); } }; @@ -1220,7 +1220,7 @@ namespace void writeToRecorders(const LLError::CallSite& site, const std::string& message) { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING LLError::ELevel level = site.mLevel; SettingsConfigPtr s = Globals::getInstance()->getSettingsConfig(); @@ -1355,7 +1355,7 @@ namespace LLError bool Log::shouldLog(CallSite& site) { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING LLMutexTrylock lock(getMutex(), 5); if (!lock.isLocked()) { @@ -1400,7 +1400,7 @@ namespace LLError void Log::flush(const std::ostringstream& out, const CallSite& site) { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_LOGGING LLMutexTrylock lock(getMutex(),5); if (!lock.isLocked()) { -- cgit v1.3 From 9f2be2a0547f5e827a91cbcd9162cbe8f9cdfb53 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 12:25:21 -0800 Subject: SL-16606: Add profiler category MEMORY --- indra/llcommon/llcommon.cpp | 4 ++-- indra/llcommon/llmemory.h | 20 ++++++++++---------- 2 files changed, 12 insertions(+), 12 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llcommon.cpp b/indra/llcommon/llcommon.cpp index 25a809dad2..d2c4e66160 100644 --- a/indra/llcommon/llcommon.cpp +++ b/indra/llcommon/llcommon.cpp @@ -42,7 +42,7 @@ void *operator new(size_t size) void* ptr; if (gProfilerEnabled) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; ptr = (malloc)(size); } else @@ -62,7 +62,7 @@ void operator delete(void *ptr) noexcept TracyFree(ptr); if (gProfilerEnabled) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; (free)(ptr); } else diff --git a/indra/llcommon/llmemory.h b/indra/llcommon/llmemory.h index 41023b4ba4..ac6c969d70 100644 --- a/indra/llcommon/llmemory.h +++ b/indra/llcommon/llmemory.h @@ -136,7 +136,7 @@ public: \ #else inline void* ll_aligned_malloc_fallback( size_t size, int align ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; #if defined(LL_WINDOWS) void* ret = _aligned_malloc(size, align); #else @@ -157,7 +157,7 @@ public: \ inline void ll_aligned_free_fallback( void* ptr ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; LL_PROFILE_FREE(ptr); #if defined(LL_WINDOWS) _aligned_free(ptr); @@ -174,7 +174,7 @@ public: \ inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed with ll_aligned_free_16(). { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; #if defined(LL_WINDOWS) void* ret = _aligned_malloc(size, 16); #elif defined(LL_DARWIN) @@ -190,7 +190,7 @@ inline void* ll_aligned_malloc_16(size_t size) // returned hunk MUST be freed wi inline void ll_aligned_free_16(void *p) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; LL_PROFILE_FREE(p); #if defined(LL_WINDOWS) _aligned_free(p); @@ -203,7 +203,7 @@ inline void ll_aligned_free_16(void *p) inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // returned hunk MUST be freed with ll_aligned_free_16(). { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; LL_PROFILE_FREE(ptr); #if defined(LL_WINDOWS) void* ret = _aligned_realloc(ptr, size, 16); @@ -228,7 +228,7 @@ inline void* ll_aligned_realloc_16(void* ptr, size_t size, size_t old_size) // r inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed with ll_aligned_free_32(). { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; #if defined(LL_WINDOWS) void* ret = _aligned_malloc(size, 32); #elif defined(LL_DARWIN) @@ -244,7 +244,7 @@ inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed wi inline void ll_aligned_free_32(void *p) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; LL_PROFILE_FREE(p); #if defined(LL_WINDOWS) _aligned_free(p); @@ -259,7 +259,7 @@ inline void ll_aligned_free_32(void *p) template LL_FORCE_INLINE void* ll_aligned_malloc(size_t size) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; void* ret; if (LL_DEFAULT_HEAP_ALIGN % ALIGNMENT == 0) { @@ -284,7 +284,7 @@ LL_FORCE_INLINE void* ll_aligned_malloc(size_t size) template LL_FORCE_INLINE void ll_aligned_free(void* ptr) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; if (ALIGNMENT == LL_DEFAULT_HEAP_ALIGN) { LL_PROFILE_FREE(ptr); @@ -309,7 +309,7 @@ LL_FORCE_INLINE void ll_aligned_free(void* ptr) // inline void ll_memcpy_nonaliased_aligned_16(char* __restrict dst, const char* __restrict src, size_t bytes) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY; assert(src != NULL); assert(dst != NULL); assert(bytes > 0); -- cgit v1.3 From 31b0e8cef83780de19fc713791a30f56108b75f6 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 12:47:54 -0800 Subject: SL-16606: Add profiler category STATS --- indra/llcommon/llfasttimer.cpp | 4 +- indra/llcommon/lltrace.h | 18 +++--- indra/llcommon/lltraceaccumulators.cpp | 20 +++---- indra/llcommon/lltraceaccumulators.h | 26 ++++----- indra/llcommon/lltracerecording.cpp | 100 ++++++++++++++++----------------- indra/llcommon/lltracerecording.h | 46 +++++++-------- 6 files changed, 107 insertions(+), 107 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llfasttimer.cpp b/indra/llcommon/llfasttimer.cpp index d38946004f..2612d0f07c 100644 --- a/indra/llcommon/llfasttimer.cpp +++ b/indra/llcommon/llfasttimer.cpp @@ -222,7 +222,7 @@ void BlockTimer::bootstrapTimerTree() // this preserves partial order derived from current frame's observations void BlockTimer::incrementalUpdateTimerTree() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; for(block_timer_tree_df_post_iterator_t it = begin_block_timer_tree_df_post(BlockTimer::getRootTimeBlock()); it != end_block_timer_tree_df_post(); ++it) @@ -263,7 +263,7 @@ void BlockTimer::incrementalUpdateTimerTree() void BlockTimer::updateTimes() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; // walk up stack of active timers and accumulate current time while leaving timing structures active BlockTimerStackRecord* stack_record = LLThreadLocalSingletonPointer::getInstance(); if (!stack_record) return; diff --git a/indra/llcommon/lltrace.h b/indra/llcommon/lltrace.h index 4051c558a4..fcd8753f75 100644 --- a/indra/llcommon/lltrace.h +++ b/indra/llcommon/lltrace.h @@ -227,7 +227,7 @@ public: void setName(const char* name) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mName = name; setKey(name); } @@ -236,13 +236,13 @@ public: StatType& allocations() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return static_cast&>(*(StatType*)this); } StatType& deallocations() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return static_cast&>(*(StatType*)this); } }; @@ -264,7 +264,7 @@ struct MeasureMem { static size_t measureFootprint(const T& value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return sizeof(T) + value.getMemFootprint(); } }; @@ -274,7 +274,7 @@ struct MeasureMem { static size_t measureFootprint(const T& value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return U32Bytes(value).value(); } }; @@ -284,7 +284,7 @@ struct MeasureMem { static size_t measureFootprint(const T* value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; if (!value) { return 0; @@ -329,7 +329,7 @@ struct MeasureMem, IS_MEM_TRACKABLE, IS_BYTES> { static size_t measureFootprint(const std::basic_string& value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return value.capacity() * sizeof(T); } }; @@ -338,7 +338,7 @@ struct MeasureMem, IS_MEM_TRACKABLE, IS_BYTES> template inline void claim_alloc(MemStatHandle& measurement, const T& value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; #if LL_TRACE_ENABLED S32 size = MeasureMem::measureFootprint(value); if(size == 0) return; @@ -351,7 +351,7 @@ inline void claim_alloc(MemStatHandle& measurement, const T& value) template inline void disclaim_alloc(MemStatHandle& measurement, const T& value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; #if LL_TRACE_ENABLED S32 size = MeasureMem::measureFootprint(value); if(size == 0) return; diff --git a/indra/llcommon/lltraceaccumulators.cpp b/indra/llcommon/lltraceaccumulators.cpp index 8e9aaee0e6..34299f5a29 100644 --- a/indra/llcommon/lltraceaccumulators.cpp +++ b/indra/llcommon/lltraceaccumulators.cpp @@ -41,7 +41,7 @@ extern MemStatHandle gTraceMemStat; AccumulatorBufferGroup::AccumulatorBufferGroup() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; claim_alloc(gTraceMemStat, mCounts.capacity() * sizeof(CountAccumulator)); claim_alloc(gTraceMemStat, mSamples.capacity() * sizeof(SampleAccumulator)); claim_alloc(gTraceMemStat, mEvents.capacity() * sizeof(EventAccumulator)); @@ -56,7 +56,7 @@ AccumulatorBufferGroup::AccumulatorBufferGroup(const AccumulatorBufferGroup& oth mStackTimers(other.mStackTimers), mMemStats(other.mMemStats) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; claim_alloc(gTraceMemStat, mCounts.capacity() * sizeof(CountAccumulator)); claim_alloc(gTraceMemStat, mSamples.capacity() * sizeof(SampleAccumulator)); claim_alloc(gTraceMemStat, mEvents.capacity() * sizeof(EventAccumulator)); @@ -66,7 +66,7 @@ AccumulatorBufferGroup::AccumulatorBufferGroup(const AccumulatorBufferGroup& oth AccumulatorBufferGroup::~AccumulatorBufferGroup() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; disclaim_alloc(gTraceMemStat, mCounts.capacity() * sizeof(CountAccumulator)); disclaim_alloc(gTraceMemStat, mSamples.capacity() * sizeof(SampleAccumulator)); disclaim_alloc(gTraceMemStat, mEvents.capacity() * sizeof(EventAccumulator)); @@ -76,7 +76,7 @@ AccumulatorBufferGroup::~AccumulatorBufferGroup() void AccumulatorBufferGroup::handOffTo(AccumulatorBufferGroup& other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; other.mCounts.reset(&mCounts); other.mSamples.reset(&mSamples); other.mEvents.reset(&mEvents); @@ -86,7 +86,7 @@ void AccumulatorBufferGroup::handOffTo(AccumulatorBufferGroup& other) void AccumulatorBufferGroup::makeCurrent() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mCounts.makeCurrent(); mSamples.makeCurrent(); mEvents.makeCurrent(); @@ -109,7 +109,7 @@ void AccumulatorBufferGroup::makeCurrent() //static void AccumulatorBufferGroup::clearCurrent() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; AccumulatorBuffer::clearCurrent(); AccumulatorBuffer::clearCurrent(); AccumulatorBuffer::clearCurrent(); @@ -124,7 +124,7 @@ bool AccumulatorBufferGroup::isCurrent() const void AccumulatorBufferGroup::append( const AccumulatorBufferGroup& other ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mCounts.addSamples(other.mCounts, SEQUENTIAL); mSamples.addSamples(other.mSamples, SEQUENTIAL); mEvents.addSamples(other.mEvents, SEQUENTIAL); @@ -134,7 +134,7 @@ void AccumulatorBufferGroup::append( const AccumulatorBufferGroup& other ) void AccumulatorBufferGroup::merge( const AccumulatorBufferGroup& other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mCounts.addSamples(other.mCounts, NON_SEQUENTIAL); mSamples.addSamples(other.mSamples, NON_SEQUENTIAL); mEvents.addSamples(other.mEvents, NON_SEQUENTIAL); @@ -145,7 +145,7 @@ void AccumulatorBufferGroup::merge( const AccumulatorBufferGroup& other) void AccumulatorBufferGroup::reset(AccumulatorBufferGroup* other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mCounts.reset(other ? &other->mCounts : NULL); mSamples.reset(other ? &other->mSamples : NULL); mEvents.reset(other ? &other->mEvents : NULL); @@ -155,7 +155,7 @@ void AccumulatorBufferGroup::reset(AccumulatorBufferGroup* other) void AccumulatorBufferGroup::sync() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; if (isCurrent()) { F64SecondsImplicit time_stamp = LLTimer::getTotalSeconds(); diff --git a/indra/llcommon/lltraceaccumulators.h b/indra/llcommon/lltraceaccumulators.h index b183fcd14a..7267a44300 100644 --- a/indra/llcommon/lltraceaccumulators.h +++ b/indra/llcommon/lltraceaccumulators.h @@ -66,7 +66,7 @@ namespace LLTrace : mStorageSize(0), mStorage(NULL) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; const AccumulatorBuffer& other = *getDefaultBuffer(); resize(sNextStorageSlot); for (S32 i = 0; i < sNextStorageSlot; i++) @@ -77,7 +77,7 @@ namespace LLTrace ~AccumulatorBuffer() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; if (isCurrent()) { LLThreadLocalSingletonPointer::setInstance(NULL); @@ -100,7 +100,7 @@ namespace LLTrace : mStorageSize(0), mStorage(NULL) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; resize(sNextStorageSlot); for (S32 i = 0; i < sNextStorageSlot; i++) { @@ -110,7 +110,7 @@ namespace LLTrace void addSamples(const AccumulatorBuffer& other, EBufferAppendType append_type) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; llassert(mStorageSize >= sNextStorageSlot && other.mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -120,7 +120,7 @@ namespace LLTrace void copyFrom(const AccumulatorBuffer& other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; llassert(mStorageSize >= sNextStorageSlot && other.mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -130,7 +130,7 @@ namespace LLTrace void reset(const AccumulatorBuffer* other = NULL) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; llassert(mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -140,7 +140,7 @@ namespace LLTrace void sync(F64SecondsImplicit time_stamp) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; llassert(mStorageSize >= sNextStorageSlot); for (size_t i = 0; i < sNextStorageSlot; i++) { @@ -166,7 +166,7 @@ namespace LLTrace // NOTE: this is not thread-safe. We assume that slots are reserved in the main thread before any child threads are spawned size_t reserveSlot() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; size_t next_slot = sNextStorageSlot++; if (next_slot >= mStorageSize) { @@ -180,7 +180,7 @@ namespace LLTrace void resize(size_t new_size) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; if (new_size <= mStorageSize) return; ACCUMULATOR* old_storage = mStorage; @@ -221,7 +221,7 @@ namespace LLTrace static self_t* getDefaultBuffer() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; static bool sInitialized = false; if (!sInitialized) { @@ -336,7 +336,7 @@ namespace LLTrace void sample(F64 value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; F64SecondsImplicit time_stamp = LLTimer::getTotalSeconds(); // store effect of last value @@ -550,7 +550,7 @@ namespace LLTrace void addSamples(const MemAccumulator& other, EBufferAppendType append_type) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mAllocations.addSamples(other.mAllocations, append_type); mDeallocations.addSamples(other.mDeallocations, append_type); @@ -569,7 +569,7 @@ namespace LLTrace void reset(const MemAccumulator* other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mSize.reset(other ? &other->mSize : NULL); mAllocations.reset(other ? &other->mAllocations : NULL); mDeallocations.reset(other ? &other->mDeallocations : NULL); diff --git a/indra/llcommon/lltracerecording.cpp b/indra/llcommon/lltracerecording.cpp index 5ce1b337fe..1613af1dcf 100644 --- a/indra/llcommon/lltracerecording.cpp +++ b/indra/llcommon/lltracerecording.cpp @@ -50,7 +50,7 @@ Recording::Recording(EPlayState state) : mElapsedSeconds(0), mActiveBuffers(NULL) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; claim_alloc(gTraceMemStat, this); mBuffers = new AccumulatorBufferGroup(); claim_alloc(gTraceMemStat, mBuffers); @@ -60,14 +60,14 @@ Recording::Recording(EPlayState state) Recording::Recording( const Recording& other ) : mActiveBuffers(NULL) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; claim_alloc(gTraceMemStat, this); *this = other; } Recording& Recording::operator = (const Recording& other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; // this will allow us to seamlessly start without affecting any data we've acquired from other setPlayState(PAUSED); @@ -88,7 +88,7 @@ Recording& Recording::operator = (const Recording& other) Recording::~Recording() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; disclaim_alloc(gTraceMemStat, this); disclaim_alloc(gTraceMemStat, mBuffers); @@ -107,7 +107,7 @@ void Recording::update() #if LL_TRACE_ENABLED if (isStarted()) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mElapsedSeconds += mSamplingTimer.getElapsedTimeF64(); // must have @@ -128,7 +128,7 @@ void Recording::update() void Recording::handleReset() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; #if LL_TRACE_ENABLED mBuffers.write()->reset(); @@ -139,7 +139,7 @@ void Recording::handleReset() void Recording::handleStart() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; #if LL_TRACE_ENABLED mSamplingTimer.reset(); mBuffers.setStayUnique(true); @@ -151,7 +151,7 @@ void Recording::handleStart() void Recording::handleStop() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; #if LL_TRACE_ENABLED mElapsedSeconds += mSamplingTimer.getElapsedTimeF64(); // must have thread recorder running on this thread @@ -583,20 +583,20 @@ PeriodicRecording::PeriodicRecording( S32 num_periods, EPlayState state) mNumRecordedPeriods(0), mRecordingPeriods(num_periods ? num_periods : 1) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; setPlayState(state); claim_alloc(gTraceMemStat, this); } PeriodicRecording::~PeriodicRecording() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; disclaim_alloc(gTraceMemStat, this); } void PeriodicRecording::nextPeriod() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; if (mAutoResize) { mRecordingPeriods.push_back(Recording()); @@ -611,7 +611,7 @@ void PeriodicRecording::nextPeriod() void PeriodicRecording::appendRecording(Recording& recording) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; getCurRecording().appendRecording(recording); nextPeriod(); } @@ -619,7 +619,7 @@ void PeriodicRecording::appendRecording(Recording& recording) void PeriodicRecording::appendPeriodicRecording( PeriodicRecording& other ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; if (other.mRecordingPeriods.empty()) return; getCurRecording().update(); @@ -693,7 +693,7 @@ void PeriodicRecording::appendPeriodicRecording( PeriodicRecording& other ) F64Seconds PeriodicRecording::getDuration() const { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; F64Seconds duration; S32 num_periods = mRecordingPeriods.size(); for (S32 i = 1; i <= num_periods; i++) @@ -707,7 +707,7 @@ F64Seconds PeriodicRecording::getDuration() const LLTrace::Recording PeriodicRecording::snapshotCurRecording() const { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; Recording recording_copy(getCurRecording()); recording_copy.stop(); return recording_copy; @@ -750,19 +750,19 @@ const Recording& PeriodicRecording::getPrevRecording( S32 offset ) const void PeriodicRecording::handleStart() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; getCurRecording().start(); } void PeriodicRecording::handleStop() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; getCurRecording().pause(); } void PeriodicRecording::handleReset() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; getCurRecording().stop(); if (mAutoResize) @@ -786,13 +786,13 @@ void PeriodicRecording::handleReset() void PeriodicRecording::handleSplitTo(PeriodicRecording& other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; getCurRecording().splitTo(other.getCurRecording()); } F64 PeriodicRecording::getPeriodMin( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -814,7 +814,7 @@ F64 PeriodicRecording::getPeriodMin( const StatType& stat, S32 F64 PeriodicRecording::getPeriodMax( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -837,7 +837,7 @@ F64 PeriodicRecording::getPeriodMax( const StatType& stat, S32 // calculates means using aggregates per period F64 PeriodicRecording::getPeriodMean( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 mean = 0; @@ -860,7 +860,7 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S3 F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 period_mean = getPeriodMean(stat, num_periods); @@ -885,7 +885,7 @@ F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -907,7 +907,7 @@ F64 PeriodicRecording::getPeriodMin( const StatType& stat, S3 F64 PeriodicRecording::getPeriodMax(const StatType& stat, S32 num_periods /*= S32_MAX*/) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -930,7 +930,7 @@ F64 PeriodicRecording::getPeriodMax(const StatType& stat, S32 F64 PeriodicRecording::getPeriodMean( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); S32 valid_period_count = 0; @@ -953,7 +953,7 @@ F64 PeriodicRecording::getPeriodMean( const StatType& stat, S F64 PeriodicRecording::getPeriodMedian( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); std::vector buf; @@ -979,7 +979,7 @@ F64 PeriodicRecording::getPeriodMedian( const StatType& stat, F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 period_mean = getPeriodMean(stat, num_periods); @@ -1005,7 +1005,7 @@ F64 PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes min_val(std::numeric_limits::max()); @@ -1025,7 +1025,7 @@ F64Kilobytes PeriodicRecording::getPeriodMin(const MemStatHandle& stat, S32 num_ F64Kilobytes PeriodicRecording::getPeriodMax(const StatType& stat, S32 num_periods /*= S32_MAX*/) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes max_val(0.0); @@ -1045,7 +1045,7 @@ F64Kilobytes PeriodicRecording::getPeriodMax(const MemStatHandle& stat, S32 num_ F64Kilobytes PeriodicRecording::getPeriodMean( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes mean(0); @@ -1066,7 +1066,7 @@ F64Kilobytes PeriodicRecording::getPeriodMean(const MemStatHandle& stat, S32 num F64Kilobytes PeriodicRecording::getPeriodStandardDeviation( const StatType& stat, S32 num_periods /*= S32_MAX*/ ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64Kilobytes period_mean = getPeriodMean(stat, num_periods); @@ -1100,7 +1100,7 @@ F64Kilobytes PeriodicRecording::getPeriodStandardDeviation(const MemStatHandle& void ExtendableRecording::extend() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; // push the data back to accepted recording mAcceptedRecording.appendRecording(mPotentialRecording); // flush data, so we can start from scratch @@ -1109,26 +1109,26 @@ void ExtendableRecording::extend() void ExtendableRecording::handleStart() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mPotentialRecording.start(); } void ExtendableRecording::handleStop() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mPotentialRecording.pause(); } void ExtendableRecording::handleReset() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mAcceptedRecording.reset(); mPotentialRecording.reset(); } void ExtendableRecording::handleSplitTo(ExtendableRecording& other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mPotentialRecording.splitTo(other.mPotentialRecording); } @@ -1145,7 +1145,7 @@ ExtendablePeriodicRecording::ExtendablePeriodicRecording() void ExtendablePeriodicRecording::extend() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; // push the data back to accepted recording mAcceptedRecording.appendPeriodicRecording(mPotentialRecording); // flush data, so we can start from scratch @@ -1155,26 +1155,26 @@ void ExtendablePeriodicRecording::extend() void ExtendablePeriodicRecording::handleStart() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mPotentialRecording.start(); } void ExtendablePeriodicRecording::handleStop() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mPotentialRecording.pause(); } void ExtendablePeriodicRecording::handleReset() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mAcceptedRecording.reset(); mPotentialRecording.reset(); } void ExtendablePeriodicRecording::handleSplitTo(ExtendablePeriodicRecording& other) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; mPotentialRecording.splitTo(other.mPotentialRecording); } @@ -1189,7 +1189,7 @@ PeriodicRecording& get_frame_recording() void LLStopWatchControlsMixinCommon::start() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; switch (mPlayState) { case STOPPED: @@ -1211,7 +1211,7 @@ void LLStopWatchControlsMixinCommon::start() void LLStopWatchControlsMixinCommon::stop() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; switch (mPlayState) { case STOPPED: @@ -1231,7 +1231,7 @@ void LLStopWatchControlsMixinCommon::stop() void LLStopWatchControlsMixinCommon::pause() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; switch (mPlayState) { case STOPPED: @@ -1251,7 +1251,7 @@ void LLStopWatchControlsMixinCommon::pause() void LLStopWatchControlsMixinCommon::unpause() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; switch (mPlayState) { case STOPPED: @@ -1271,7 +1271,7 @@ void LLStopWatchControlsMixinCommon::unpause() void LLStopWatchControlsMixinCommon::resume() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; switch (mPlayState) { case STOPPED: @@ -1292,7 +1292,7 @@ void LLStopWatchControlsMixinCommon::resume() void LLStopWatchControlsMixinCommon::restart() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; switch (mPlayState) { case STOPPED: @@ -1316,13 +1316,13 @@ void LLStopWatchControlsMixinCommon::restart() void LLStopWatchControlsMixinCommon::reset() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; handleReset(); } void LLStopWatchControlsMixinCommon::setPlayState( EPlayState state ) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; switch(state) { case STOPPED: diff --git a/indra/llcommon/lltracerecording.h b/indra/llcommon/lltracerecording.h index 1f3d37336a..556b7470cf 100644 --- a/indra/llcommon/lltracerecording.h +++ b/indra/llcommon/lltracerecording.h @@ -355,7 +355,7 @@ namespace LLTrace template S32 getSampleCount(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); S32 num_samples = 0; @@ -375,7 +375,7 @@ namespace LLTrace template typename T::value_t getPeriodMin(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -398,7 +398,7 @@ namespace LLTrace template T getPeriodMin(const CountStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return T(getPeriodMin(static_cast&>(stat), num_periods)); } @@ -406,7 +406,7 @@ namespace LLTrace template T getPeriodMin(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return T(getPeriodMin(static_cast&>(stat), num_periods)); } @@ -414,7 +414,7 @@ namespace LLTrace template T getPeriodMin(const EventStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return T(getPeriodMin(static_cast&>(stat), num_periods)); } @@ -424,7 +424,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMinPerSec(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); typename RelatedTypes::fractional_t min_val(std::numeric_limits::max()); @@ -439,7 +439,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMinPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodMinPerSec(static_cast&>(stat), num_periods)); } @@ -451,7 +451,7 @@ namespace LLTrace template typename T::value_t getPeriodMax(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); bool has_value = false; @@ -474,7 +474,7 @@ namespace LLTrace template T getPeriodMax(const CountStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return T(getPeriodMax(static_cast&>(stat), num_periods)); } @@ -482,7 +482,7 @@ namespace LLTrace template T getPeriodMax(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return T(getPeriodMax(static_cast&>(stat), num_periods)); } @@ -490,7 +490,7 @@ namespace LLTrace template T getPeriodMax(const EventStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return T(getPeriodMax(static_cast&>(stat), num_periods)); } @@ -500,7 +500,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMaxPerSec(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); F64 max_val = std::numeric_limits::min(); @@ -515,7 +515,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMaxPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodMaxPerSec(static_cast&>(stat), num_periods)); } @@ -527,7 +527,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMean(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); typename RelatedTypes::fractional_t mean(0); @@ -548,14 +548,14 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMean(const CountStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodMean(static_cast&>(stat), num_periods)); } F64 getPeriodMean(const StatType& stat, S32 num_periods = S32_MAX); template typename RelatedTypes::fractional_t getPeriodMean(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodMean(static_cast&>(stat), num_periods)); } @@ -563,7 +563,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMean(const EventStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodMean(static_cast&>(stat), num_periods)); } @@ -573,7 +573,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMeanPerSec(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); typename RelatedTypes::fractional_t mean = 0; @@ -595,7 +595,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMeanPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodMeanPerSec(static_cast&>(stat), num_periods)); } @@ -604,7 +604,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMedianPerSec(const StatType& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; num_periods = llmin(num_periods, getNumRecordedPeriods()); std::vector ::fractional_t> buf; @@ -624,7 +624,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodMedianPerSec(const CountStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodMedianPerSec(static_cast&>(stat), num_periods)); } @@ -637,7 +637,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodStandardDeviation(const SampleStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodStandardDeviation(static_cast&>(stat), num_periods)); } @@ -645,7 +645,7 @@ namespace LLTrace template typename RelatedTypes::fractional_t getPeriodStandardDeviation(const EventStatHandle& stat, S32 num_periods = S32_MAX) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; return typename RelatedTypes::fractional_t(getPeriodStandardDeviation(static_cast&>(stat), num_periods)); } -- cgit v1.3 From 12fd860636e8d45087f94c8252212c103f49e1ad Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 12:49:11 -0800 Subject: SL-16606: Add profiler category STRING --- indra/llcommon/llstring.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llstring.cpp b/indra/llcommon/llstring.cpp index bdea1e76ea..7f501f2e77 100644 --- a/indra/llcommon/llstring.cpp +++ b/indra/llcommon/llstring.cpp @@ -1317,7 +1317,7 @@ bool LLStringUtil::formatDatetime(std::string& replacement, std::string token, template<> S32 LLStringUtil::format(std::string& s, const format_map_t& substitutions) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STRING; S32 res = 0; std::string output; @@ -1390,7 +1390,7 @@ S32 LLStringUtil::format(std::string& s, const format_map_t& substitutions) template<> S32 LLStringUtil::format(std::string& s, const LLSD& substitutions) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STRING; S32 res = 0; if (!substitutions.isMap()) -- cgit v1.3 From b504c692554d492113a10ef45427fe0ab0d8a85d Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 12:55:53 -0800 Subject: SL-16606: Add profiler category THREAD --- indra/llcommon/llmutex.cpp | 24 ++++++++++---------- indra/llcommon/llsingleton.h | 2 +- indra/llcommon/llthread.cpp | 14 ++++++------ indra/llcommon/llthreadsafequeue.h | 36 +++++++++++++++--------------- indra/llcommon/threadsafeschedule.h | 44 ++++++++++++++++++------------------- indra/llcommon/workqueue.cpp | 8 +++---- 6 files changed, 64 insertions(+), 64 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llmutex.cpp b/indra/llcommon/llmutex.cpp index a49002b5dc..0273dd5970 100644 --- a/indra/llcommon/llmutex.cpp +++ b/indra/llcommon/llmutex.cpp @@ -44,7 +44,7 @@ LLMutex::~LLMutex() void LLMutex::lock() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if(isSelfLocked()) { //redundant lock mCount++; @@ -66,7 +66,7 @@ void LLMutex::lock() void LLMutex::unlock() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if (mCount > 0) { //not the root unlock mCount--; @@ -87,7 +87,7 @@ void LLMutex::unlock() bool LLMutex::isLocked() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if (!mMutex.try_lock()) { return true; @@ -111,7 +111,7 @@ LLThread::id_t LLMutex::lockingThread() const bool LLMutex::trylock() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if(isSelfLocked()) { //redundant lock mCount++; @@ -150,20 +150,20 @@ LLCondition::~LLCondition() void LLCondition::wait() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD std::unique_lock< std::mutex > lock(mMutex); mCond.wait(lock); } void LLCondition::signal() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD mCond.notify_one(); } void LLCondition::broadcast() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD mCond.notify_all(); } @@ -173,7 +173,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex) : mMutex(mutex), mLocked(false) { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if (mMutex) mLocked = mMutex->trylock(); } @@ -182,7 +182,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms) : mMutex(mutex), mLocked(false) { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if (!mMutex) return; @@ -197,7 +197,7 @@ LLMutexTrylock::LLMutexTrylock(LLMutex* mutex, U32 aTries, U32 delay_ms) LLMutexTrylock::~LLMutexTrylock() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if (mMutex && mLocked) mMutex->unlock(); } @@ -209,7 +209,7 @@ LLMutexTrylock::~LLMutexTrylock() // LLScopedLock::LLScopedLock(std::mutex* mutex) : mMutex(mutex) { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if(mutex) { mutex->lock(); @@ -228,7 +228,7 @@ LLScopedLock::~LLScopedLock() void LLScopedLock::unlock() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if(mLocked) { mMutex->unlock(); diff --git a/indra/llcommon/llsingleton.h b/indra/llcommon/llsingleton.h index 6042c0906c..51ef514cf7 100644 --- a/indra/llcommon/llsingleton.h +++ b/indra/llcommon/llsingleton.h @@ -455,7 +455,7 @@ public: static DERIVED_TYPE* getInstance() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // We know the viewer has LLSingleton dependency circularities. If you // feel strongly motivated to eliminate them, cheers and good luck. // (At that point we could consider a much simpler locking mechanism.) diff --git a/indra/llcommon/llthread.cpp b/indra/llcommon/llthread.cpp index 11f5a015f1..a807acc56e 100644 --- a/indra/llcommon/llthread.cpp +++ b/indra/llcommon/llthread.cpp @@ -333,7 +333,7 @@ bool LLThread::runCondition(void) // Stop thread execution if requested until unpaused. void LLThread::checkPause() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD mDataLock->lock(); // This is in a while loop because the pthread API allows for spurious wakeups. @@ -365,20 +365,20 @@ void LLThread::setQuitting() // static LLThread::id_t LLThread::currentID() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD return std::this_thread::get_id(); } // static void LLThread::yield() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD std::this_thread::yield(); } void LLThread::wake() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD mDataLock->lock(); if(!shouldSleep()) { @@ -389,7 +389,7 @@ void LLThread::wake() void LLThread::wakeLocked() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD if(!shouldSleep()) { mRunCondition->signal(); @@ -398,13 +398,13 @@ void LLThread::wakeLocked() void LLThread::lockData() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD mDataLock->lock(); } void LLThread::unlockData() { - LL_PROFILE_ZONE_SCOPED + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD mDataLock->unlock(); } diff --git a/indra/llcommon/llthreadsafequeue.h b/indra/llcommon/llthreadsafequeue.h index 2806506550..68d79cdd12 100644 --- a/indra/llcommon/llthreadsafequeue.h +++ b/indra/llcommon/llthreadsafequeue.h @@ -275,7 +275,7 @@ template template bool LLThreadSafeQueue::tryLock(CALLABLE&& callable) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock1(mLock, std::defer_lock); if (!lock1.try_lock()) return false; @@ -292,7 +292,7 @@ bool LLThreadSafeQueue::tryLockUntil( const std::chrono::time_point& until, CALLABLE&& callable) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock1(mLock, std::defer_lock); if (!lock1.try_lock_until(until)) return false; @@ -306,7 +306,7 @@ template template bool LLThreadSafeQueue::push_(lock_t& lock, T&& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; if (mStorage.size() >= mCapacity) return false; @@ -322,7 +322,7 @@ template template bool LLThreadSafeQueue::pushIfOpen(T&& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock1(mLock); while (true) { @@ -345,7 +345,7 @@ template template void LLThreadSafeQueue::push(T&& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; if (! pushIfOpen(std::forward(element))) { LLTHROW(LLThreadSafeQueueInterrupt()); @@ -357,7 +357,7 @@ template template bool LLThreadSafeQueue::tryPush(T&& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryLock( [this, element=std::move(element)](lock_t& lock) { @@ -374,7 +374,7 @@ bool LLThreadSafeQueue::tryPushFor( const std::chrono::duration& timeout, T&& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // Convert duration to time_point: passing the same timeout duration to // each of multiple calls is wrong. return tryPushUntil(std::chrono::steady_clock::now() + timeout, @@ -388,7 +388,7 @@ bool LLThreadSafeQueue::tryPushUntil( const std::chrono::time_point& until, T&& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryLockUntil( until, [this, until, element=std::move(element)](lock_t& lock) @@ -421,7 +421,7 @@ template typename LLThreadSafeQueue::pop_result LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // If mStorage is empty, there's no head element. if (mStorage.empty()) return mClosed? DONE : EMPTY; @@ -443,7 +443,7 @@ LLThreadSafeQueue::pop_(lock_t& lock, ElementT& element) template ElementT LLThreadSafeQueue::pop(void) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock1(mLock); ElementT value; while (true) @@ -472,7 +472,7 @@ ElementT LLThreadSafeQueue::pop(void) template bool LLThreadSafeQueue::tryPop(ElementT & element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryLock( [this, &element](lock_t& lock) { @@ -490,7 +490,7 @@ bool LLThreadSafeQueue::tryPopFor( const std::chrono::duration& timeout, ElementT& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // Convert duration to time_point: passing the same timeout duration to // each of multiple calls is wrong. return tryPopUntil(std::chrono::steady_clock::now() + timeout, element); @@ -503,7 +503,7 @@ bool LLThreadSafeQueue::tryPopUntil( const std::chrono::time_point& until, ElementT& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryLockUntil( until, [this, until, &element](lock_t& lock) @@ -523,7 +523,7 @@ LLThreadSafeQueue::tryPopUntil_( const std::chrono::time_point& until, ElementT& element) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; while (true) { pop_result popped = pop_(lock, element); @@ -550,7 +550,7 @@ LLThreadSafeQueue::tryPopUntil_( template size_t LLThreadSafeQueue::size(void) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock(mLock); return mStorage.size(); } @@ -559,7 +559,7 @@ size_t LLThreadSafeQueue::size(void) template void LLThreadSafeQueue::close() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock(mLock); mClosed = true; lock.unlock(); @@ -573,7 +573,7 @@ void LLThreadSafeQueue::close() template bool LLThreadSafeQueue::isClosed() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock(mLock); return mClosed; } @@ -582,7 +582,7 @@ bool LLThreadSafeQueue::isClosed() template bool LLThreadSafeQueue::done() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock(mLock); return mClosed && mStorage.empty(); } diff --git a/indra/llcommon/threadsafeschedule.h b/indra/llcommon/threadsafeschedule.h index 601681d550..3e0da94c02 100644 --- a/indra/llcommon/threadsafeschedule.h +++ b/indra/llcommon/threadsafeschedule.h @@ -98,14 +98,14 @@ namespace LL // we could minimize redundancy by breaking out a common base class... void push(const DataTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; push(tuple_cons(Clock::now(), tuple)); } /// individually pass each component of the TimeTuple void push(const TimePoint& time, Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; push(TimeTuple(time, std::forward(args)...)); } @@ -116,7 +116,7 @@ namespace LL // and call that overload. void push(Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; push(Clock::now(), std::forward(args)...); } @@ -127,21 +127,21 @@ namespace LL /// DataTuple with implicit now bool tryPush(const DataTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPush(tuple_cons(Clock::now(), tuple)); } /// individually pass components bool tryPush(const TimePoint& time, Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPush(TimeTuple(time, std::forward(args)...)); } /// individually pass components with implicit now bool tryPush(Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPush(Clock::now(), std::forward(args)...); } @@ -154,7 +154,7 @@ namespace LL bool tryPushFor(const std::chrono::duration& timeout, const DataTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPushFor(timeout, tuple_cons(Clock::now(), tuple)); } @@ -163,7 +163,7 @@ namespace LL bool tryPushFor(const std::chrono::duration& timeout, const TimePoint& time, Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPushFor(TimeTuple(time, std::forward(args)...)); } @@ -172,7 +172,7 @@ namespace LL bool tryPushFor(const std::chrono::duration& timeout, Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPushFor(Clock::now(), std::forward(args)...); } @@ -185,7 +185,7 @@ namespace LL bool tryPushUntil(const std::chrono::time_point& until, const DataTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPushUntil(until, tuple_cons(Clock::now(), tuple)); } @@ -194,7 +194,7 @@ namespace LL bool tryPushUntil(const std::chrono::time_point& until, const TimePoint& time, Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPushUntil(until, TimeTuple(time, std::forward(args)...)); } @@ -203,7 +203,7 @@ namespace LL bool tryPushUntil(const std::chrono::time_point& until, Args&&... args) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tryPushUntil(until, Clock::now(), std::forward(args)...); } @@ -221,14 +221,14 @@ namespace LL // haven't yet jumped through those hoops. DataTuple pop() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; return tuple_cdr(popWithTime()); } /// pop TimeTuple by value TimeTuple popWithTime() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; lock_t lock(super::mLock); // We can't just sit around waiting forever, given that there may // be items in the queue that are not yet ready but will *become* @@ -268,7 +268,7 @@ namespace LL /// tryPop(DataTuple&) bool tryPop(DataTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; TimeTuple tt; if (! super::tryPop(tt)) return false; @@ -279,7 +279,7 @@ namespace LL /// for when Args has exactly one type bool tryPop(typename std::tuple_element<1, TimeTuple>::type& value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; TimeTuple tt; if (! super::tryPop(tt)) return false; @@ -291,7 +291,7 @@ namespace LL template bool tryPopFor(const std::chrono::duration& timeout, Tuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // It's important to use OUR tryPopUntil() implementation, rather // than delegating immediately to our base class. return tryPopUntil(Clock::now() + timeout, tuple); @@ -302,7 +302,7 @@ namespace LL bool tryPopUntil(const std::chrono::time_point& until, TimeTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // super::tryPopUntil() wakes up when an item becomes available or // we hit 'until', whichever comes first. Thing is, the current // head of the queue could become ready sooner than either of @@ -322,7 +322,7 @@ namespace LL pop_result tryPopUntil_(lock_t& lock, const TimePoint& until, TimeTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; TimePoint adjusted = until; if (! super::mStorage.empty()) { @@ -350,7 +350,7 @@ namespace LL bool tryPopUntil(const std::chrono::time_point& until, DataTuple& tuple) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; TimeTuple tt; if (! tryPopUntil(until, tt)) return false; @@ -363,7 +363,7 @@ namespace LL bool tryPopUntil(const std::chrono::time_point& until, typename std::tuple_element<1, TimeTuple>::type& value) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; TimeTuple tt; if (! tryPopUntil(until, tt)) return false; @@ -387,7 +387,7 @@ namespace LL // considering whether to deliver the current head element bool canPop(const TimeTuple& head) const override { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // an item with a future timestamp isn't yet ready to pop // (should we add some slop for overhead?) return std::get<0>(head) <= Clock::now(); diff --git a/indra/llcommon/workqueue.cpp b/indra/llcommon/workqueue.cpp index c74dada2e4..eb06890468 100644 --- a/indra/llcommon/workqueue.cpp +++ b/indra/llcommon/workqueue.cpp @@ -60,7 +60,7 @@ void LL::WorkQueue::runUntilClose() { for (;;) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; callWork(mQueue.pop()); } } @@ -71,7 +71,7 @@ void LL::WorkQueue::runUntilClose() bool LL::WorkQueue::runPending() { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; for (Work work; mQueue.tryPop(work); ) { callWork(work); @@ -91,7 +91,7 @@ bool LL::WorkQueue::runOne() bool LL::WorkQueue::runUntil(const TimePoint& until) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; // Should we subtract some slop to allow for typical Work execution time? // How much slop? // runUntil() is simply a time-bounded runPending(). @@ -129,7 +129,7 @@ void LL::WorkQueue::callWork(const Queue::DataTuple& work) void LL::WorkQueue::callWork(const Work& work) { - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; try { work(); -- cgit v1.3 From 8aa22b41650f13f3752643276ab1449de4fa3ab8 Mon Sep 17 00:00:00 2001 From: Ptolemy Date: Thu, 13 Jan 2022 13:05:43 -0800 Subject: SL-16606: Add profiler category STATS --- indra/llcommon/lltracethreadrecorder.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/lltracethreadrecorder.cpp b/indra/llcommon/lltracethreadrecorder.cpp index 7ae1e72784..090d3297a0 100644 --- a/indra/llcommon/lltracethreadrecorder.cpp +++ b/indra/llcommon/lltracethreadrecorder.cpp @@ -277,7 +277,7 @@ void ThreadRecorder::pushToParent() void ThreadRecorder::pullFromChildren() { #if LL_TRACE_ENABLED - LL_PROFILE_ZONE_SCOPED; + LL_PROFILE_ZONE_SCOPED_CATEGORY_STATS; if (mActiveRecordings.empty()) return; { LLMutexLock lock(&mChildListMutex); -- cgit v1.3 From d9a68339d5aa18af349e347d6ed74bc01824cec7 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 23 Feb 2022 16:51:33 -0600 Subject: SL-16815 and SL-16906 Avoid redundant bumpmap generation, add some assertions around ref counting and (hack) fix crash on shutdown from dangling texture reference (reduced to 1 dangling texture from several hundred, can't find the remaining reference). --- indra/llcommon/llrefcount.cpp | 5 ++++- indra/llcommon/llrefcount.h | 16 ++++++++++++++-- indra/llrender/llimagegl.cpp | 8 +++++--- indra/newview/lldrawpoolbump.cpp | 4 ++-- indra/newview/llviewertexture.cpp | 3 +-- 5 files changed, 26 insertions(+), 10 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llrefcount.cpp b/indra/llcommon/llrefcount.cpp index 29a5ca6f24..5cbd346411 100644 --- a/indra/llcommon/llrefcount.cpp +++ b/indra/llcommon/llrefcount.cpp @@ -29,6 +29,9 @@ #include "llerror.h" +// maximum reference count before sounding memory leak alarm +const S32 gMaxRefCount = 65536; + LLRefCount::LLRefCount(const LLRefCount& other) : mRef(0) { @@ -47,7 +50,7 @@ LLRefCount::LLRefCount() : LLRefCount::~LLRefCount() { - if (mRef != 0) + if (mRef != LL_REFCOUNT_FREE && mRef != 0) { LL_ERRS() << "deleting non-zero reference" << LL_ENDL; } diff --git a/indra/llcommon/llrefcount.h b/indra/llcommon/llrefcount.h index 7e4af6ea66..2080da1565 100644 --- a/indra/llcommon/llrefcount.h +++ b/indra/llcommon/llrefcount.h @@ -37,6 +37,10 @@ class LLMutex; // see llthread.h for LLThreadSafeRefCount //---------------------------------------------------------------------------- +//nonsense but recognizable value for freed LLRefCount (aids in debugging) +#define LL_REFCOUNT_FREE 1234567890 +extern const S32 gMaxRefCount; + class LL_COMMON_API LLRefCount { protected: @@ -47,17 +51,25 @@ protected: public: LLRefCount(); + inline void validateRefCount() const + { + llassert(mRef > 0); // ref count below 0, likely corrupted + llassert(mRef < gMaxRefCount); // ref count excessive, likely memory leak + } + inline void ref() const { mRef++; + validateRefCount(); } inline S32 unref() const { - llassert(mRef >= 1); + validateRefCount(); if (0 == --mRef) { - delete this; + mRef = LL_REFCOUNT_FREE; // set to nonsense yet recognizable value to aid in debugging + delete this; return 0; } return mRef; diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp index 9e26247e1e..5cf3a2a15f 100644 --- a/indra/llrender/llimagegl.cpp +++ b/indra/llrender/llimagegl.cpp @@ -437,7 +437,7 @@ LLImageGL::LLImageGL( LLImageGL::~LLImageGL() { - if (!mExternalTexture) + if (!mExternalTexture && gGLManager.mInited) { LLImageGL::cleanup(); sImageList.erase(this); @@ -1629,14 +1629,16 @@ void LLImageGLThread::updateClass() } }; + // post update to background thread if available, otherwise execute immediately - auto queue = LLImageGLThread::sEnabled ? LL::WorkQueue::getInstance("LLImageGL") : nullptr; - if (queue) + auto queue = LL::WorkQueue::getInstance("LLImageGL"); + if (sEnabled) { queue->post(func); } else { + llassert(queue == nullptr); func(); } } diff --git a/indra/newview/lldrawpoolbump.cpp b/indra/newview/lldrawpoolbump.cpp index 2b151487fd..1d5419b515 100644 --- a/indra/newview/lldrawpoolbump.cpp +++ b/indra/newview/lldrawpoolbump.cpp @@ -1051,8 +1051,8 @@ void LLBumpImageList::onSourceLoaded( BOOL success, LLViewerTexture *src_vi, LLI } } - //if (iter->second->getWidth() != src->getWidth() || - // iter->second->getHeight() != src->getHeight()) // bump not cached yet or has changed resolution + if (iter->second->getWidth() != src->getWidth() || + iter->second->getHeight() != src->getHeight()) // bump not cached yet or has changed resolution { LLPointer dst_image = new LLImageRaw(src->getWidth(), src->getHeight(), 1); U8* dst_data = dst_image->getData(); diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index fa2c070ea0..e3ac56d0d3 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -1619,7 +1619,6 @@ void LLViewerFetchedTexture::scheduleCreateTexture() mNeedsCreateTexture = TRUE; if (preCreateTexture()) { - ref(); #if LL_IMAGEGL_THREAD_CHECK //grab a copy of the raw image data to make sure it isn't modified pending texture creation U8* data = mRawImage->getData(); @@ -1635,6 +1634,7 @@ void LLViewerFetchedTexture::scheduleCreateTexture() auto mainq = LLImageGLThread::sEnabled ? mMainQueue.lock() : nullptr; if (mainq) { + ref(); mainq->postTo( mImageQueue, // work to be done on LLImageGL worker thread @@ -1681,7 +1681,6 @@ void LLViewerFetchedTexture::scheduleCreateTexture() else { gTextureList.mCreateTextureList.insert(this); - unref(); } } } -- cgit v1.3 From 0e954a9afd7cc300bdd4cadfc25baa7f2607e5a4 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 9 Mar 2022 12:48:52 -0600 Subject: SL-16972 Per feedback from Ansariel, only bump up max heap size on 64-bit builds. --- indra/llcommon/llprocessor.cpp | 14 -------------- indra/llcommon/llprocessor.h | 14 ++++++++++++++ indra/newview/app_settings/settings.xml | 15 +++++++++++++-- indra/newview/llappviewer.cpp | 9 ++++++--- 4 files changed, 33 insertions(+), 19 deletions(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llprocessor.cpp b/indra/llcommon/llprocessor.cpp index 5d16a4b74d..818df07bb2 100644 --- a/indra/llcommon/llprocessor.cpp +++ b/indra/llcommon/llprocessor.cpp @@ -44,20 +44,6 @@ #include "llsd.h" -#if LL_MSVC && _M_X64 -# define LL_X86_64 1 -# define LL_X86 1 -#elif LL_MSVC && _M_IX86 -# define LL_X86 1 -#elif LL_GNUC && ( defined(__amd64__) || defined(__x86_64__) ) -# define LL_X86_64 1 -# define LL_X86 1 -#elif LL_GNUC && ( defined(__i386__) ) -# define LL_X86 1 -#elif LL_GNUC && ( defined(__powerpc__) || defined(__ppc__) ) -# define LL_PPC 1 -#endif - class LLProcessorInfoImpl; // foward declaration for the mImpl; namespace diff --git a/indra/llcommon/llprocessor.h b/indra/llcommon/llprocessor.h index 90e5bc59ee..b77eb22c3a 100644 --- a/indra/llcommon/llprocessor.h +++ b/indra/llcommon/llprocessor.h @@ -29,6 +29,20 @@ #define LLPROCESSOR_H #include "llunits.h" +#if LL_MSVC && _M_X64 +# define LL_X86_64 1 +# define LL_X86 1 +#elif LL_MSVC && _M_IX86 +# define LL_X86 1 +#elif LL_GNUC && ( defined(__amd64__) || defined(__x86_64__) ) +# define LL_X86_64 1 +# define LL_X86 1 +#elif LL_GNUC && ( defined(__i386__) ) +# define LL_X86 1 +#elif LL_GNUC && ( defined(__powerpc__) || defined(__ppc__) ) +# define LL_PPC 1 +#endif + class LLProcessorInfoImpl; class LL_COMMON_API LLProcessorInfo diff --git a/indra/newview/app_settings/settings.xml b/indra/newview/app_settings/settings.xml index ebf9bfcfe8..17071d4a65 100644 --- a/indra/newview/app_settings/settings.xml +++ b/indra/newview/app_settings/settings.xml @@ -6481,13 +6481,24 @@ MaxHeapSize Comment - Maximum heap size (GB) + Maximum heap size on 32-bit builds (GB) Persist 1 Type F32 Value - 4.0 + 1.6 + + MaxHeapSize64 + + Comment + Maximum heap size on 64-bit builds (GB) + Persist + 1 + Type + F32 + Value + 16.0 MaxPersistentNotifications diff --git a/indra/newview/llappviewer.cpp b/indra/newview/llappviewer.cpp index 66c44ef6a6..61fc693b99 100644 --- a/indra/newview/llappviewer.cpp +++ b/indra/newview/llappviewer.cpp @@ -1309,10 +1309,13 @@ void LLAppViewer::initMaxHeapSize() //------------------------------------------------------------------------------------------ //currently SL is built under 32-bit setting, we set its max heap size no more than 1.6 GB. - //F32 max_heap_size_gb = llmin(1.6f, (F32)gSavedSettings.getF32("MaxHeapSize")) ; - F32Gigabytes max_heap_size_gb = (F32Gigabytes)gSavedSettings.getF32("MaxHeapSize") ; + #ifndef LL_X86_64 + F32Gigabytes max_heap_size_gb = (F32Gigabytes)gSavedSettings.getF32("MaxHeapSize") ; +#else + F32Gigabytes max_heap_size_gb = (F32Gigabytes)gSavedSettings.getF32("MaxHeapSize64"); +#endif - LLMemory::initMaxHeapSizeGB(max_heap_size_gb); + LLMemory::initMaxHeapSizeGB(max_heap_size_gb); } static LLTrace::BlockTimerStatHandle FTM_MESSAGES("System Messages"); -- cgit v1.3 From 9b2df75c87d8ef06177f1591716cbe913b66de1e Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 25 Mar 2022 13:05:50 -0500 Subject: SL-17077 Ensure profiling is disabled if gpu_benchmark fails. Log the source of the real crash for when the viewer inevitably crashes later. --- indra/llcommon/llexception.cpp | 5 +++++ indra/newview/llfeaturemanager.cpp | 8 +++++++- 2 files changed, 12 insertions(+), 1 deletion(-) (limited to 'indra/llcommon') diff --git a/indra/llcommon/llexception.cpp b/indra/llcommon/llexception.cpp index b584b0ff8b..46560b5e4c 100644 --- a/indra/llcommon/llexception.cpp +++ b/indra/llcommon/llexception.cpp @@ -97,6 +97,11 @@ static const U32 STATUS_MSC_EXCEPTION = 0xE06D7363; // compiler specific U32 msc_exception_filter(U32 code, struct _EXCEPTION_POINTERS *exception_infop) { + const auto stack = to_string(boost::stacktrace::stacktrace()); + LL_WARNS() << "SEH Exception handled (that probably shouldn't be): Code " << code + << "\n Stack trace: \n" + << stack << LL_ENDL; + if (code == STATUS_MSC_EXCEPTION) { // C++ exception, go on diff --git a/indra/newview/llfeaturemanager.cpp b/indra/newview/llfeaturemanager.cpp index 3240f169b3..e934041e2e 100644 --- a/indra/newview/llfeaturemanager.cpp +++ b/indra/newview/llfeaturemanager.cpp @@ -381,7 +381,10 @@ F32 gpu_benchmark(); F32 logExceptionBenchmark() { - // Todo: make a wrapper/class for SEH exceptions + // FIXME: gpu_benchmark uses many C++ classes on the stack to control state. + // SEH exceptions with our current exception handling options do not call + // destructors for these classes, resulting in an undefined state should + // this handler be invoked. F32 gbps = -1; __try { @@ -389,6 +392,9 @@ F32 logExceptionBenchmark() } __except (msc_exception_filter(GetExceptionCode(), GetExceptionInformation())) { + // HACK - ensure that profiling is disabled + LLGLSLShader::finishProfile(false); + // convert to C++ styled exception char integer_string[32]; sprintf(integer_string, "SEH, code: %lu\n", GetExceptionCode()); -- cgit v1.3