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/newview/llviewertexture.cpp | 311 ++++++++++++++++++++++++-------------- 1 file changed, 200 insertions(+), 111 deletions(-) (limited to 'indra/newview/llviewertexture.cpp') 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); -- cgit v1.3 From 52e1a45659341df68eca01822f189e0883afe66d Mon Sep 17 00:00:00 2001 From: Runitai Linden Date: Tue, 12 Oct 2021 14:46:19 -0500 Subject: SL-16166 Optimization pass on LLRenderPass::pushBatch --- indra/llrender/llgltexture.h | 2 +- indra/llrender/llrender.cpp | 18 +++- indra/llrender/llrender.h | 11 +++ indra/llrender/lltexture.h | 4 +- indra/llrender/llvertexbuffer.cpp | 155 ++++++++++++++++++++++++++++++++++ indra/llrender/llvertexbuffer.h | 11 ++- indra/newview/lldrawpool.cpp | 29 +++---- indra/newview/lldrawpoolmaterials.cpp | 1 + indra/newview/lldrawpoolsimple.cpp | 1 + indra/newview/lldrawpooltree.cpp | 1 + indra/newview/llspatialpartition.h | 7 +- indra/newview/llviewertexture.cpp | 6 +- indra/newview/pipeline.cpp | 52 +++++++----- indra/newview/pipeline.h | 3 +- 14 files changed, 247 insertions(+), 54 deletions(-) (limited to 'indra/newview/llviewertexture.cpp') diff --git a/indra/llrender/llgltexture.h b/indra/llrender/llgltexture.h index 071912c2c2..028457c510 100644 --- a/indra/llrender/llgltexture.h +++ b/indra/llrender/llgltexture.h @@ -176,7 +176,7 @@ private: protected: void setTexelsPerImage(); - //note: do not make this function public. +public: /*virtual*/ LLImageGL* getGLTexture() const ; protected: diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index b6711e44e3..34c5b77944 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -169,6 +169,7 @@ void LLTexUnit::refreshState(void) void LLTexUnit::activate(void) { + LL_PROFILE_ZONE_SCOPED; if (mIndex < 0) return; if ((S32)gGL.mCurrTextureUnitIndex != mIndex || gGL.mDirty) @@ -229,8 +230,20 @@ void LLTexUnit::disable(void) } } +void LLTexUnit::bindFast(LLTexture* texture) +{ + LLImageGL* gl_tex = texture->getGLTexture(); + + glActiveTextureARB(GL_TEXTURE0_ARB + mIndex); + gGL.mCurrTextureUnitIndex = mIndex; + mCurrTexture = gl_tex->getTexName(); + glBindTexture(sGLTextureType[gl_tex->getTarget()], mCurrTexture); + mHasMipMaps = gl_tex->mHasMipMaps; +} + bool LLTexUnit::bind(LLTexture* texture, bool for_rendering, bool forceBind) { + LL_PROFILE_ZONE_SCOPED; stop_glerror(); if (mIndex >= 0) { @@ -1243,8 +1256,6 @@ void LLRender::syncLightState() void LLRender::syncMatrices() { - stop_glerror(); - static const U32 name[] = { LLShaderMgr::MODELVIEW_MATRIX, @@ -1415,8 +1426,6 @@ void LLRender::syncMatrices() } } } - - stop_glerror(); } void LLRender::translatef(const GLfloat& x, const GLfloat& y, const GLfloat& z) @@ -1925,6 +1934,7 @@ void LLRender::end() } void LLRender::flush() { + LL_PROFILE_ZONE_SCOPED; if (mCount > 0) { if (!mUIOffset.empty()) diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h index c08c2d6881..658cdda504 100644 --- a/indra/llrender/llrender.h +++ b/indra/llrender/llrender.h @@ -161,6 +161,17 @@ public: bool bind(LLImageGL* texture, bool for_rendering = false, bool forceBind = false); bool bind(LLTexture* texture, bool for_rendering = false, bool forceBind = false); + // bind implementation for inner loops + // makes the following assumptions: + // - No need for gGL.flush() + // - texture is not null + // - gl_tex->getTexName() is not zero + // - This texture is not being bound redundantly + // - USE_SRGB_DECODE is disabled + // - mTexOptionsDirty is false + // - + void bindFast(LLTexture* texture); + // Binds a cubemap to this texture unit // (automatically enables the texture unit for cubemaps) bool bind(LLCubeMap* cubeMap); diff --git a/indra/llrender/lltexture.h b/indra/llrender/lltexture.h index 41481fb8a7..256d85ce5a 100644 --- a/indra/llrender/lltexture.h +++ b/indra/llrender/lltexture.h @@ -67,11 +67,9 @@ public: virtual S32 getWidth(S32 discard_level = -1) const; virtual S32 getHeight(S32 discard_level = -1) const; virtual bool isActiveFetching(); + virtual LLImageGL* getGLTexture() const; private: - //note: do not make this function public. - virtual LLImageGL* getGLTexture() const; - virtual void updateBindStatsForTester(); }; #endif diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 0449ac392c..7b9388123f 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -787,6 +787,18 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi placeFence(); } +void LLVertexBuffer::drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const +{ + mMappable = false; + gGL.syncMatrices(); + + U16* idx = ((U16*)(U8*)mAlignedIndexOffset) + indices_offset; + + LL_PROFILER_GPU_ZONEC("gl.DrawRangeElements", 0xFFFF00) + glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, + idx); +} + void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const { llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL); @@ -2272,6 +2284,21 @@ bool LLVertexBuffer::bindGLBuffer(bool force_bind) return ret; } +bool LLVertexBuffer::bindGLBufferFast() +{ + if (mGLBuffer != sGLRenderBuffer || !sVBOActive) + { + glBindBufferARB(GL_ARRAY_BUFFER_ARB, mGLBuffer); + sGLRenderBuffer = mGLBuffer; + sBindCount++; + sVBOActive = true; + + return true; + } + + return false; +} + static LLTrace::BlockTimerStatHandle FTM_BIND_GL_INDICES("Bind Indices"); bool LLVertexBuffer::bindGLIndices(bool force_bind) @@ -2297,6 +2324,21 @@ bool LLVertexBuffer::bindGLIndices(bool force_bind) return ret; } +bool LLVertexBuffer::bindGLIndicesFast() +{ + if (mGLIndices != sGLRenderIndices || !sIBOActive) + { + glBindBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB, mGLIndices); + sGLRenderIndices = mGLIndices; + sBindCount++; + sIBOActive = true; + + return true; + } + + return false; +} + void LLVertexBuffer::flush() { if (useVBOs()) @@ -2487,6 +2529,26 @@ void LLVertexBuffer::setBuffer(U32 data_mask) } } +void LLVertexBuffer::setBufferFast(U32 data_mask) +{ + //set up pointers if the data mask is different ... + bool setup = (sLastMask != data_mask); + + + const bool bindBuffer = bindGLBufferFast(); + const bool bindIndices = bindGLIndicesFast(); + + setup = setup || bindBuffer || bindIndices; + + setupClientArrays(data_mask); + + if (data_mask && setup) + { + setupVertexBuffer(data_mask); // subclass specific setup (virtual function) + sSetCount++; + } +} + // virtual (default) void LLVertexBuffer::setupVertexBuffer(U32 data_mask) { @@ -2644,6 +2706,99 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask) llglassertok(); } +void LLVertexBuffer::setupVertexBufferFast(U32 data_mask) +{ + U8* base = (U8*)mAlignedOffset; + + if (data_mask & MAP_NORMAL) + { + S32 loc = TYPE_NORMAL; + void* ptr = (void*)(base + mOffsets[TYPE_NORMAL]); + glVertexAttribPointerARB(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_NORMAL], ptr); + } + if (data_mask & MAP_TEXCOORD3) + { + S32 loc = TYPE_TEXCOORD3; + void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD3]); + glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD3], ptr); + } + if (data_mask & MAP_TEXCOORD2) + { + S32 loc = TYPE_TEXCOORD2; + void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD2]); + glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD2], ptr); + } + if (data_mask & MAP_TEXCOORD1) + { + S32 loc = TYPE_TEXCOORD1; + void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD1]); + glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], ptr); + } + if (data_mask & MAP_TANGENT) + { + S32 loc = TYPE_TANGENT; + void* ptr = (void*)(base + mOffsets[TYPE_TANGENT]); + glVertexAttribPointerARB(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TANGENT], ptr); + } + if (data_mask & MAP_TEXCOORD0) + { + S32 loc = TYPE_TEXCOORD0; + void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD0]); + glVertexAttribPointerARB(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD0], ptr); + } + if (data_mask & MAP_COLOR) + { + S32 loc = TYPE_COLOR; + //bind emissive instead of color pointer if emissive is present + void* ptr = (data_mask & MAP_EMISSIVE) ? (void*)(base + mOffsets[TYPE_EMISSIVE]) : (void*)(base + mOffsets[TYPE_COLOR]); + glVertexAttribPointerARB(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_COLOR], ptr); + } + if (data_mask & MAP_EMISSIVE) + { + S32 loc = TYPE_EMISSIVE; + void* ptr = (void*)(base + mOffsets[TYPE_EMISSIVE]); + glVertexAttribPointerARB(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr); + + if (!(data_mask & MAP_COLOR)) + { //map emissive to color channel when color is not also being bound to avoid unnecessary shader swaps + loc = TYPE_COLOR; + glVertexAttribPointerARB(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr); + } + } + if (data_mask & MAP_WEIGHT) + { + S32 loc = TYPE_WEIGHT; + void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT]); + glVertexAttribPointerARB(loc, 1, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT], ptr); + } + if (data_mask & MAP_WEIGHT4) + { + S32 loc = TYPE_WEIGHT4; + void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT4]); + glVertexAttribPointerARB(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT4], ptr); + } + if (data_mask & MAP_CLOTHWEIGHT) + { + S32 loc = TYPE_CLOTHWEIGHT; + void* ptr = (void*)(base + mOffsets[TYPE_CLOTHWEIGHT]); + glVertexAttribPointerARB(loc, 4, GL_FLOAT, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_CLOTHWEIGHT], ptr); + } + if (data_mask & MAP_TEXTURE_INDEX) + { +#if !LL_DARWIN + S32 loc = TYPE_TEXTURE_INDEX; + void* ptr = (void*)(base + mOffsets[TYPE_VERTEX] + 12); + glVertexAttribIPointer(loc, 1, GL_UNSIGNED_INT, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); +#endif + } + if (data_mask & MAP_VERTEX) + { + S32 loc = TYPE_VERTEX; + void* ptr = (void*)(base + mOffsets[TYPE_VERTEX]); + glVertexAttribPointerARB(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); + } +} + LLVertexBuffer::MappedRegion::MappedRegion(S32 type, S32 index, S32 count) : mType(type), mIndex(index), mCount(count) { diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h index 1d60970df4..51ed85510e 100644 --- a/indra/llrender/llvertexbuffer.h +++ b/indra/llrender/llvertexbuffer.h @@ -210,13 +210,17 @@ protected: virtual ~LLVertexBuffer(); // use unref() - virtual void setupVertexBuffer(U32 data_mask); // pure virtual, called from mapBuffer() + virtual void setupVertexBuffer(U32 data_mask); + void setupVertexBufferFast(U32 data_mask); + void setupVertexArray(); void genBuffer(U32 size); void genIndices(U32 size); bool bindGLBuffer(bool force_bind = false); + bool bindGLBufferFast(); bool bindGLIndices(bool force_bind = false); + bool bindGLIndicesFast(); bool bindGLArray(); void releaseBuffer(); void releaseIndices(); @@ -239,6 +243,8 @@ public: // set for rendering virtual void setBuffer(U32 data_mask); // calls setupVertexBuffer() if data_mask is not 0 + void setBufferFast(U32 data_mask); // calls setupVertexBufferFast(), assumes data_mask is not 0 among other assumptions + void flush(); //flush pending data to GL memory // allocate buffer bool allocateBuffer(S32 nverts, S32 nindices, bool create); @@ -290,6 +296,9 @@ public: void drawArrays(U32 mode, U32 offset, U32 count) const; void drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const; + //implementation for inner loops that does no safety checking + void drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const; + //for debugging, validate data in given range is valid void validateRange(U32 start, U32 end, U32 count, U32 offset) const; diff --git a/indra/newview/lldrawpool.cpp b/indra/newview/lldrawpool.cpp index d583a692f9..01dd307187 100644 --- a/indra/newview/lldrawpool.cpp +++ b/indra/newview/lldrawpool.cpp @@ -404,6 +404,7 @@ void LLRenderPass::renderTexture(U32 type, U32 mask, BOOL batch_textures) void LLRenderPass::pushBatches(U32 type, U32 mask, BOOL texture, BOOL batch_textures) { + LL_PROFILE_ZONE_SCOPED; for (LLCullResult::drawinfo_iterator i = gPipeline.beginRenderMap(type); i != gPipeline.endRenderMap(type); ++i) { LLDrawInfo* pparams = *i; @@ -452,6 +453,7 @@ void LLRenderPass::applyModelMatrix(const LLDrawInfo& params) void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL batch_textures) { + LL_PROFILE_ZONE_SCOPED; if (!params.mCount) { return; @@ -469,7 +471,7 @@ void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL ba { if (params.mTextureList[i].notNull()) { - gGL.getTexUnit(i)->bind(params.mTextureList[i], TRUE); + gGL.getTexUnit(i)->bindFast(params.mTextureList[i]); } } } @@ -477,8 +479,7 @@ void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL ba { //not batching textures or batch has only 1 texture -- might need a texture matrix if (params.mTexture.notNull()) { - params.mTexture->addTextureStats(params.mVSize); - gGL.getTexUnit(0)->bind(params.mTexture, TRUE) ; + gGL.getTexUnit(0)->bindFast(params.mTexture); if (params.mTextureMatrix) { tex_setup = true; @@ -495,19 +496,17 @@ void LLRenderPass::pushBatch(LLDrawInfo& params, U32 mask, BOOL texture, BOOL ba } } - if (params.mVertexBuffer.notNull()) - { - if (params.mGroup) - { - params.mGroup->rebuildMesh(); - } + { + if (params.mGroup) + { + params.mGroup->rebuildMesh(); + } - LLGLEnableFunc stencil_test(GL_STENCIL_TEST, params.mSelected, &LLGLCommonFunc::selected_stencil_test); - - params.mVertexBuffer->setBuffer(mask); - params.mVertexBuffer->drawRange(params.mDrawMode, params.mStart, params.mEnd, params.mCount, params.mOffset); - gPipeline.addTrianglesDrawn(params.mCount, params.mDrawMode); - } + LLGLEnableFunc stencil_test(GL_STENCIL_TEST, params.mSelected, &LLGLCommonFunc::selected_stencil_test); + + params.mVertexBuffer->setBufferFast(mask); + params.mVertexBuffer->drawRangeFast(params.mDrawMode, params.mStart, params.mEnd, params.mCount, params.mOffset); + } if (tex_setup) { diff --git a/indra/newview/lldrawpoolmaterials.cpp b/indra/newview/lldrawpoolmaterials.cpp index 05b0c1f1a9..d45e387e10 100644 --- a/indra/newview/lldrawpoolmaterials.cpp +++ b/indra/newview/lldrawpoolmaterials.cpp @@ -106,6 +106,7 @@ void LLDrawPoolMaterials::endDeferredPass(S32 pass) void LLDrawPoolMaterials::renderDeferred(S32 pass) { + LL_PROFILE_ZONE_SCOPED; static const U32 type_list[] = { LLRenderPass::PASS_MATERIAL, diff --git a/indra/newview/lldrawpoolsimple.cpp b/indra/newview/lldrawpoolsimple.cpp index 74e6665a96..ba8549904b 100644 --- a/indra/newview/lldrawpoolsimple.cpp +++ b/indra/newview/lldrawpoolsimple.cpp @@ -471,6 +471,7 @@ void LLDrawPoolSimple::endDeferredPass(S32 pass) void LLDrawPoolSimple::renderDeferred(S32 pass) { + LL_PROFILE_ZONE_SCOPED; LLGLDisable blend(GL_BLEND); LLGLDisable alpha_test(GL_ALPHA_TEST); diff --git a/indra/newview/lldrawpooltree.cpp b/indra/newview/lldrawpooltree.cpp index 0d5195bdbf..a1ff020068 100644 --- a/indra/newview/lldrawpooltree.cpp +++ b/indra/newview/lldrawpooltree.cpp @@ -153,6 +153,7 @@ void LLDrawPoolTree::beginDeferredPass(S32 pass) void LLDrawPoolTree::renderDeferred(S32 pass) { + LL_PROFILE_ZONE_SCOPED; render(pass); } diff --git a/indra/newview/llspatialpartition.h b/indra/newview/llspatialpartition.h index 919f386d29..6ef82fac9c 100644 --- a/indra/newview/llspatialpartition.h +++ b/indra/newview/llspatialpartition.h @@ -41,6 +41,7 @@ #include "llviewercamera.h" #include "llvector4a.h" #include +#include #define SG_STATE_INHERIT_MASK (OCCLUDED) #define SG_INITIAL_STATE_MASK (DIRTY | GEOM_DIRTY) @@ -216,10 +217,10 @@ public: typedef std::vector > sg_vector_t; typedef std::vector > bridge_list_t; typedef std::vector > drawmap_elem_t; - typedef std::map draw_map_t; + typedef std::unordered_map draw_map_t; typedef std::vector > buffer_list_t; - typedef std::map buffer_texture_map_t; - typedef std::map buffer_map_t; + typedef std::unordered_map buffer_texture_map_t; + typedef std::unordered_map buffer_map_t; struct CompareDistanceGreater { diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index 274f53a160..949e71a4c9 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -845,14 +845,14 @@ void LLViewerTexture::addTextureStats(F32 virtual_size, BOOL needs_gltexture) co { //flag to reset the values because the old values are used. resetMaxVirtualSizeResetCounter(); - mMaxVirtualSize = virtual_size; - mAdditionalDecodePriority = 0.f; + mMaxVirtualSize = virtual_size; + mAdditionalDecodePriority = 0.f; mNeedsGLTexture = needs_gltexture; } else if (virtual_size > mMaxVirtualSize) { mMaxVirtualSize = virtual_size; - } + } } void LLViewerTexture::resetTextureStats() diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index c0b469af81..5ef3819de4 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -358,7 +358,6 @@ bool LLPipeline::sRenderAttachedLights = true; bool LLPipeline::sRenderAttachedParticles = true; bool LLPipeline::sRenderDeferred = false; S32 LLPipeline::sVisibleLightCount = 0; -F32 LLPipeline::sMinRenderSize = 0.f; bool LLPipeline::sRenderingHUDs; F32 LLPipeline::sDistortionWaterClipPlaneMargin = 1.0125f; @@ -2550,13 +2549,6 @@ void LLPipeline::markNotCulled(LLSpatialGroup* group, LLCamera& camera) return; } - const LLVector4a* bounds = group->getBounds(); - if (sMinRenderSize > 0.f && - llmax(llmax(bounds[1][0], bounds[1][1]), bounds[1][2]) < sMinRenderSize) - { - return; - } - assertInitialized(); if (!group->getSpatialPartition()->mRenderByGroup) @@ -3480,7 +3472,6 @@ void LLPipeline::stateSort(LLSpatialGroup* group, LLCamera& camera) group->mLastUpdateDistance = group->mDistance; } } - } void LLPipeline::stateSort(LLSpatialBridge* bridge, LLCamera& camera, BOOL fov_changed) @@ -3787,6 +3778,27 @@ void renderSoundHighlights(LLDrawable* drawablep) } } +void LLPipeline::touchTextures(LLDrawInfo* info) +{ + LL_PROFILE_ZONE_SCOPED; + for (auto& tex : info->mTextureList) + { + if (tex.notNull()) + { + LLImageGL* gl_tex = tex->getGLTexture(); + if (gl_tex && gl_tex->updateBindStats(gl_tex->mTextureMemory)) + { + tex->setActive(); + } + } + } + + if (info->mTexture.notNull()) + { + info->mTexture->addTextureStats(info->mVSize); + } +} + void LLPipeline::postSort(LLCamera& camera) { LL_RECORD_BLOCK_TIME(FTM_STATESORT_POSTSORT); @@ -3839,20 +3851,14 @@ void LLPipeline::postSort(LLCamera& camera) for (LLSpatialGroup::drawmap_elem_t::iterator k = src_vec.begin(); k != src_vec.end(); ++k) { - if (sMinRenderSize > 0.f) - { - LLVector4a bounds; - bounds.setSub((*k)->mExtents[1],(*k)->mExtents[0]); - - if (llmax(llmax(bounds[0], bounds[1]), bounds[2]) > sMinRenderSize) - { - sCull->pushDrawInfo(j->first, *k); - } - } - else - { - sCull->pushDrawInfo(j->first, *k); - } + LLDrawInfo* info = *k; + + sCull->pushDrawInfo(j->first, info); + if (!sShadowRender && !sReflectionRender) + { + touchTextures(info); + addTrianglesDrawn(info->mCount, info->mDrawMode); + } } } diff --git a/indra/newview/pipeline.h b/indra/newview/pipeline.h index 5605d26410..8ffbddca21 100644 --- a/indra/newview/pipeline.h +++ b/indra/newview/pipeline.h @@ -265,6 +265,8 @@ public: void stateSort(LLSpatialBridge* bridge, LLCamera& camera, BOOL fov_changed = FALSE); void stateSort(LLDrawable* drawablep, LLCamera& camera); void postSort(LLCamera& camera); + //update stats for textures in given DrawInfo + void touchTextures(LLDrawInfo* info); void forAllVisibleDrawables(void (*func)(LLDrawable*)); void renderObjects(U32 type, U32 mask, bool texture = true, bool batch_texture = false); @@ -596,7 +598,6 @@ public: static bool sRenderAttachedParticles; static bool sRenderDeferred; static S32 sVisibleLightCount; - static F32 sMinRenderSize; static bool sRenderingHUDs; static F32 sDistortionWaterClipPlaneMargin; -- cgit v1.3