diff options
Diffstat (limited to 'indra/newview/llglsandbox.cpp')
| -rwxr-xr-x[-rw-r--r--] | indra/newview/llglsandbox.cpp | 196 |
1 files changed, 190 insertions, 6 deletions
diff --git a/indra/newview/llglsandbox.cpp b/indra/newview/llglsandbox.cpp index 1208c9378e..c80dec0e75 100644..100755 --- a/indra/newview/llglsandbox.cpp +++ b/indra/newview/llglsandbox.cpp @@ -62,6 +62,7 @@ #include "llresmgr.h" #include "pipeline.h" #include "llspatialpartition.h" +#include "llviewershadermgr.h" // Height of the yellow selection highlight posts for land const F32 PARCEL_POST_HEIGHT = 0.666f; @@ -79,10 +80,10 @@ void LLToolSelectRect::handleRectangleSelection(S32 x, S32 y, MASK mask) S32 top = llmax(y, mDragStartY); S32 bottom =llmin(y, mDragStartY); - left = llround((F32) left * LLUI::sGLScaleFactor.mV[VX]); - right = llround((F32) right * LLUI::sGLScaleFactor.mV[VX]); - top = llround((F32) top * LLUI::sGLScaleFactor.mV[VY]); - bottom = llround((F32) bottom * LLUI::sGLScaleFactor.mV[VY]); + left = ll_round((F32) left * LLUI::getScaleFactor().mV[VX]); + right = ll_round((F32) right * LLUI::getScaleFactor().mV[VX]); + top = ll_round((F32) top * LLUI::getScaleFactor().mV[VY]); + bottom = ll_round((F32) bottom * LLUI::getScaleFactor().mV[VY]); F32 old_far_plane = LLViewerCamera::getInstance()->getFar(); F32 old_near_plane = LLViewerCamera::getInstance()->getNear(); @@ -623,7 +624,8 @@ void LLViewerParcelMgr::renderCollisionSegments(U8* segments, BOOL use_pass, LLV LLGLDepthTest gls_depth(GL_TRUE, GL_FALSE); LLGLDisable cull(GL_CULL_FACE); - if (mCollisionBanned == BA_BANNED) + if (mCollisionBanned == BA_BANNED || + regionp->getRegionFlag(REGION_FLAGS_BLOCK_FLYOVER)) { collision_height = BAN_HEIGHT; } @@ -767,7 +769,6 @@ void draw_line_cube(F32 width, const LLVector3& center) gGL.vertex3f(center.mV[VX] + width ,center.mV[VY] - width,center.mV[VZ] - width); } - void LLViewerObjectList::renderObjectBeacons() { if (mDebugBeacons.empty()) @@ -878,3 +879,186 @@ void LLViewerObjectList::renderObjectBeacons() } +F32 gpu_benchmark() +{ + if (!gGLManager.mHasShaderObjects || !gGLManager.mHasTimerQuery) + { // don't bother benchmarking the fixed function + // or venerable drivers which don't support accurate timing anyway + // and are likely to be correctly identified by the GPU table already. + return -1.f; + } + + if (gBenchmarkProgram.mProgramObject == 0) + { + LLViewerShaderMgr::instance()->initAttribsAndUniforms(); + + gBenchmarkProgram.mName = "Benchmark Shader"; + gBenchmarkProgram.mFeatures.attachNothing = true; + gBenchmarkProgram.mShaderFiles.clear(); + gBenchmarkProgram.mShaderFiles.push_back(std::make_pair("interface/benchmarkV.glsl", GL_VERTEX_SHADER_ARB)); + gBenchmarkProgram.mShaderFiles.push_back(std::make_pair("interface/benchmarkF.glsl", GL_FRAGMENT_SHADER_ARB)); + gBenchmarkProgram.mShaderLevel = 1; + if (!gBenchmarkProgram.createShader(NULL, NULL)) + { + return -1.f; + } + } + + LLGLDisable blend(GL_BLEND); + + //measure memory bandwidth by: + // - allocating a batch of textures and render targets + // - rendering those textures to those render targets + // - recording time taken + // - taking the median time for a given number of samples + + //resolution of textures/render targets + const U32 res = 1024; + + //number of textures + const U32 count = 32; + + //number of samples to take + const S32 samples = 64; + + if (gGLManager.mHasTimerQuery) + { + LLGLSLShader::initProfile(); + } + + LLRenderTarget dest[count]; + U32 source[count]; + LLImageGL::generateTextures(count, source); + std::vector<F32> results; + + //build a random texture + U8* pixels = new U8[res*res*4]; + + for (U32 i = 0; i < res*res*4; ++i) + { + pixels[i] = (U8) ll_rand(255); + } + + + gGL.setColorMask(true, true); + LLGLDepthTest depth(GL_FALSE); + + for (U32 i = 0; i < count; ++i) + { //allocate render targets and textures + dest[i].allocate(res,res,GL_RGBA,false, false, LLTexUnit::TT_TEXTURE, true); + dest[i].bindTarget(); + dest[i].clear(); + dest[i].flush(); + + gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, source[i]); + LLImageGL::setManualImage(GL_TEXTURE_2D, 0, GL_RGBA, res,res,GL_RGBA, GL_UNSIGNED_BYTE, pixels); + } + + delete [] pixels; + + //make a dummy triangle to draw with + LLPointer<LLVertexBuffer> buff = new LLVertexBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0, GL_STATIC_DRAW_ARB); + buff->allocateBuffer(3, 0, true); + + LLStrider<LLVector3> v; + LLStrider<LLVector2> tc; + + buff->getVertexStrider(v); + + v[0].set(-1,1,0); + v[1].set(-1,-3,0); + v[2].set(3,1,0); + + buff->flush(); + + gBenchmarkProgram.bind(); + + bool busted_finish = false; + + buff->setBuffer(LLVertexBuffer::MAP_VERTEX); + glFinish(); + + for (S32 c = -1; c < samples; ++c) + { + LLTimer timer; + timer.start(); + + for (U32 i = 0; i < count; ++i) + { + dest[i].bindTarget(); + gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, source[i]); + buff->drawArrays(LLRender::TRIANGLES, 0, 3); + dest[i].flush(); + } + + //wait for current batch of copies to finish + if (busted_finish) + { + //read a pixel off the last target since some drivers seem to ignore glFinish + dest[count-1].bindTarget(); + U32 pixel = 0; + glReadPixels(0,0,1,1,GL_RGBA, GL_UNSIGNED_BYTE, &pixel); + dest[count-1].flush(); + } + else + { + glFinish(); + } + + F32 time = timer.getElapsedTimeF32(); + + if (c >= 0) // <-- ignore the first sample as it tends to be artificially slow + { + //store result in gigabytes per second + F32 gb = (F32) ((F64) (res*res*8*count))/(1000000000); + + F32 gbps = gb/time; + + if (!gGLManager.mHasTimerQuery && !busted_finish && gbps > 128.f) + { //unrealistically high bandwidth for a card without timer queries, glFinish is probably ignored + busted_finish = true; + LL_WARNS() << "GPU Benchmark detected GL driver with broken glFinish implementation." << LL_ENDL; + } + else + { + results.push_back(gbps); + } + } + } + + gBenchmarkProgram.unbind(); + + if (gGLManager.mHasTimerQuery) + { + LLGLSLShader::finishProfile(false); + } + + LLImageGL::deleteTextures(count, source); + + std::sort(results.begin(), results.end()); + + F32 gbps = results[results.size()/2]; + + LL_INFOS() << "Memory bandwidth is " << llformat("%.3f", gbps) << "GB/sec according to CPU timers" << LL_ENDL; + +#if LL_DARWIN + if (gbps > 512.f) + { + LL_WARNS() << "Memory bandwidth is improbably high and likely incorrect; discarding result." << LL_ENDL; + //OSX is probably lying, discard result + gbps = -1.f; + } +#endif + + F32 ms = gBenchmarkProgram.mTimeElapsed/1000000.f; + F32 seconds = ms/1000.f; + + F64 samples_drawn = res*res*count*samples; + F32 samples_sec = (samples_drawn/1000000000.0)/seconds; + gbps = samples_sec*8; + + LL_INFOS() << "Memory bandwidth is " << llformat("%.3f", gbps) << "GB/sec according to ARB_timer_query" << LL_ENDL; + + return gbps; +} + |