Automated merge with https://bitbucket.org/lindenlab/viewer-release

3 files changed, 114 insertions, 18 deletions

diff --git a/indra/newview/llfeaturemanager.cpp b/indra/newview/llfeaturemanager.cpp
index d0555477ea..92e6f70566 100755
--- a/indra/newview/llfeaturemanager.cpp
+++ b/indra/newview/llfeaturemanager.cpp
@@ -417,13 +417,67 @@ bool LLFeatureManager::parseFeatureTable(std::string filename)
 	return parse_ok;
 }
 
+F32 gpu_benchmark();
+
 bool LLFeatureManager::loadGPUClass()
 {
+	//get memory bandwidth from benchmark
+	F32 gbps = gpu_benchmark();
+
+	if (gbps < 0.f)
+	{ //couldn't bench, use GLVersion
+
+		if (gGLManager.mGLVersion < 2.f)
+		{
+			mGPUClass = GPU_CLASS_0;
+		}
+		else if (gGLManager.mGLVersion < 3.f)
+		{
+			mGPUClass = GPU_CLASS_1;
+		}
+		else if (gGLManager.mGLVersion < 3.3f)
+		{
+			mGPUClass = GPU_CLASS_2;
+		}
+		else if (gGLManager.mGLVersion < 4.f)
+		{
+			mGPUClass = GPU_CLASS_3;
+		}
+		else 
+		{
+			mGPUClass = GPU_CLASS_4;
+		}
+	}
+	else if (gbps < 5.f)
+	{
+		mGPUClass = GPU_CLASS_0;
+	}
+	else if (gbps < 10.f)
+	{
+		mGPUClass = GPU_CLASS_1;
+	}
+	else if (gbps < 20.f)
+	{
+		mGPUClass = GPU_CLASS_2;
+	}
+	else if (gbps < 40.f)
+	{
+		mGPUClass = GPU_CLASS_3;
+	}
+	else if (gbps < 80.f)
+	{
+		mGPUClass = GPU_CLASS_4;
+	}
+	else 
+	{
+		mGPUClass = GPU_CLASS_5;
+	}
+	
 	// defaults
-	mGPUClass = GPU_CLASS_UNKNOWN;
 	mGPUString = gGLManager.getRawGLString();
-	mGPUSupported = FALSE;
+	mGPUSupported = TRUE;
 
+#if 0
 	// first table is in the app dir
 	std::string app_path = gDirUtilp->getAppRODataDir();
 	app_path += gDirUtilp->getDirDelimiter();
@@ -451,8 +505,8 @@ bool LLFeatureManager::loadGPUClass()
 	{
 		parse_ok = parseGPUTable(app_path);
 	}
-
-	return parse_ok; // indicates that the file parsed correctly, not that the gpu was recognized
+#endif
+	return true; // indicates that the file parsed correctly, not that the gpu was recognized
 }
 
 	
@@ -730,6 +784,7 @@ void LLFeatureManager::init()
 
 void LLFeatureManager::applyRecommendedSettings()
 {
+	loadGPUClass();
 	// apply saved settings
 	// cap the level at 2 (high)
 	U32 level = llmax(GPU_CLASS_0, llmin(mGPUClass, GPU_CLASS_5));
diff --git a/indra/newview/llglsandbox.cpp b/indra/newview/llglsandbox.cpp
index c386030329..0548de3895 100755
--- a/indra/newview/llglsandbox.cpp
+++ b/indra/newview/llglsandbox.cpp
@@ -879,13 +879,32 @@ void LLViewerObjectList::renderObjectBeacons()
 }
 
 
-void gpu_benchmark()
+F32 gpu_benchmark()
 {
-	if (!LLGLSLShader::sNoFixedFunction)
+	if (!gGLManager.mHasShaderObjects)
 	{ //don't bother benchmarking the fixed function
-		return;
+		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
@@ -951,6 +970,8 @@ void gpu_benchmark()
 	//wait for any previoius GL commands to finish
 	glFinish();
 	
+	bool busted_finish = false;
+
 	for (S32 c = -1; c < samples; ++c)
 	{
 		LLTimer timer;
@@ -965,7 +986,18 @@ void gpu_benchmark()
 		}
 		
 		//wait for current batch of copies to finish
-		glFinish();
+		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();
 
@@ -976,13 +1008,20 @@ void gpu_benchmark()
 
 			F32 gbps = gb/time;
 
-			results.push_back(gbps);
+			if (!gGLManager.mHasTimerQuery && !busted_finish && gbps > 128.f)
+			{ //unrealistically high bandwidth for a card without timer queries, glFinish is probably ignored
+				busted_finish = true;
+			}
+			else
+			{
+				results.push_back(gbps);
+			}		
 		}
 	}
 
 	gBenchmarkProgram.unbind();
 
-	LLGLSLShader::finishProfile();
+	LLGLSLShader::finishProfile(false);
 	
 	LLImageGL::deleteTextures(count, source);
 
@@ -993,20 +1032,22 @@ void gpu_benchmark()
 
 	LL_INFOS() << "Memory bandwidth is " << llformat("%.3f", gbps) << "GB/sec according to CPU timers" << LL_ENDL;
 	
-	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;
-
 	if (gGLManager.mHasTimerQuery)
 	{
+		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;
 	}
 	else
 	{
 		LL_INFOS() << "ARB_timer_query unavailable." << LL_ENDL;
 	}
+
+	return gbps;
 }
 
diff --git a/indra/newview/llviewermenu.cpp b/indra/newview/llviewermenu.cpp
index 8c9429c05d..57e295b8d4 100755
--- a/indra/newview/llviewermenu.cpp
+++ b/indra/newview/llviewermenu.cpp
@@ -7187,7 +7187,7 @@ class LLAdvancedClickRenderProfile: public view_listener_t
 	}
 };
 
-void gpu_benchmark();
+F32 gpu_benchmark();
 
 class LLAdvancedClickRenderBenchmark: public view_listener_t
 {