added a toggle "LLGLManager::mDebugGPU" for debugging certain CPUs.

added more debug code for EXT-6791: [crashhunters] Intel 965 Crash in glCopyTexSubImage2D

3 files changed, 64 insertions, 135 deletions

diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index 236594d602..823208b095 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -185,9 +185,6 @@ PFNGLRENDERBUFFERSTORAGEMULTISAMPLEEXTPROC glRenderbufferStorageMultisampleEXT =
 // GL_EXT_framebuffer_blit
 PFNGLBLITFRAMEBUFFEREXTPROC glBlitFramebufferEXT = NULL;
 
-// GL_EXT_blend_func_separate
-PFNGLBLENDFUNCSEPARATEEXTPROC glBlendFuncSeparateEXT = NULL;
-
 // GL_ARB_draw_buffers
 PFNGLDRAWBUFFERSARBPROC glDrawBuffersARB = NULL;
 
@@ -327,7 +324,6 @@ LLGLManager::LLGLManager() :
 	mHasCompressedTextures(FALSE),
 	mHasFramebufferObject(FALSE),
 	mHasFramebufferMultisample(FALSE),
-	mHasBlendFuncSeparate(FALSE),
 
 	mHasVertexBufferObject(FALSE),
 	mHasPBuffer(FALSE),
@@ -356,6 +352,8 @@ LLGLManager::LLGLManager() :
 
 	mHasSeparateSpecularColor(FALSE),
 
+	mDebugGPU(FALSE),
+
 	mDriverVersionMajor(1),
 	mDriverVersionMinor(0),
 	mDriverVersionRelease(0),
@@ -523,11 +521,23 @@ bool LLGLManager::initGL()
 		return false;
 	}
 	
+	setToDebugGPU();
 
 	initGLStates();
 	return true;
 }
 
+void LLGLManager::setToDebugGPU()
+{
+	//"MOBILE INTEL(R) 965 EXPRESS CHIP", 
+	if (mGLRenderer.find("INTEL") != std::string::npos && mGLRenderer.find("965") != std::string::npos)
+	{
+		mDebugGPU = TRUE ;
+	}
+
+	return ;
+}
+
 void LLGLManager::getGLInfo(LLSD& info)
 {
 	info["GLInfo"]["GLVendor"] = std::string((const char *)glGetString(GL_VENDOR));
@@ -637,11 +647,6 @@ void LLGLManager::initExtensions()
 #else
 	mHasDrawBuffers = FALSE;
 # endif
-# if GL_EXT_blend_func_separate
-	mHasBlendFuncSeparate = TRUE;
-#else
-	mHasBlendFuncSeparate = FALSE;
-# endif
 	mHasMipMapGeneration = FALSE;
 	mHasSeparateSpecularColor = FALSE;
 	mHasAnisotropic = FALSE;
@@ -668,7 +673,6 @@ void LLGLManager::initExtensions()
 		&& ExtensionExists("GL_EXT_packed_depth_stencil", gGLHExts.mSysExts);
 	mHasFramebufferMultisample = mHasFramebufferObject && ExtensionExists("GL_EXT_framebuffer_multisample", gGLHExts.mSysExts);
 	mHasDrawBuffers = ExtensionExists("GL_ARB_draw_buffers", gGLHExts.mSysExts);
-	mHasBlendFuncSeparate = ExtensionExists("GL_EXT_blend_func_separate", gGLHExts.mSysExts);
 	mHasTextureRectangle = ExtensionExists("GL_ARB_texture_rectangle", gGLHExts.mSysExts);
 #if !LL_DARWIN
 	mHasPointParameters = !mIsATI && ExtensionExists("GL_ARB_point_parameters", gGLHExts.mSysExts);
@@ -692,7 +696,6 @@ void LLGLManager::initExtensions()
 		mHasFramebufferObject = FALSE;
 		mHasFramebufferMultisample = FALSE;
 		mHasDrawBuffers = FALSE;
-		mHasBlendFuncSeparate = FALSE;
 		mHasMipMapGeneration = FALSE;
 		mHasSeparateSpecularColor = FALSE;
 		mHasAnisotropic = FALSE;
@@ -717,7 +720,6 @@ void LLGLManager::initExtensions()
 		mHasShaderObjects = FALSE;
 		mHasVertexShader = FALSE;
 		mHasFragmentShader = FALSE;
-		mHasBlendFuncSeparate = FALSE;
 		LL_WARNS("RenderInit") << "GL extension support forced to SIMPLE level via LL_GL_BASICEXT" << LL_ENDL;
 	}
 	if (getenv("LL_GL_BLACKLIST"))	/* Flawfinder: ignore */
@@ -746,8 +748,7 @@ void LLGLManager::initExtensions()
 		if (strchr(blacklist,'r')) mHasDrawBuffers = FALSE;//S
 		if (strchr(blacklist,'s')) mHasFramebufferMultisample = FALSE;
 		if (strchr(blacklist,'t')) mHasTextureRectangle = FALSE;
-		if (strchr(blacklist,'u')) mHasBlendFuncSeparate = FALSE;//S
-		
+
 	}
 #endif // LL_LINUX || LL_SOLARIS
 	
@@ -795,14 +796,6 @@ void LLGLManager::initExtensions()
 	{
 		LL_INFOS("RenderInit") << "Couldn't initialize GL_ARB_fragment_shader" << LL_ENDL;
 	}
-	if (!mHasBlendFuncSeparate)
-	{
-		LL_INFOS("RenderInit") << "Couldn't initialize GL_EXT_blend_func_separate" << LL_ENDL;
-	}
-	if (!mHasDrawBuffers)
-	{
-		LL_INFOS("RenderInit") << "Couldn't initialize GL_ARB_draw_buffers" << LL_ENDL;
-	}
 
 	// Disable certain things due to known bugs
 	if (mIsIntel && mHasMipMapGeneration)
@@ -873,10 +866,6 @@ void LLGLManager::initExtensions()
 	{
 		glDrawBuffersARB = (PFNGLDRAWBUFFERSARBPROC) GLH_EXT_GET_PROC_ADDRESS("glDrawBuffersARB");
 	}
-	if (mHasBlendFuncSeparate)
-	{
-		glBlendFuncSeparateEXT = (PFNGLBLENDFUNCSEPARATEEXTPROC) GLH_EXT_GET_PROC_ADDRESS("glBlendFuncSeparateEXT");
-	}
 #if (!LL_LINUX && !LL_SOLARIS) || LL_LINUX_NV_GL_HEADERS
 	// This is expected to be a static symbol on Linux GL implementations, except if we use the nvidia headers - bah
 	glDrawRangeElements = (PFNGLDRAWRANGEELEMENTSPROC)GLH_EXT_GET_PROC_ADDRESS("glDrawRangeElements");
@@ -1039,9 +1028,24 @@ void flush_glerror()
 	glGetError();
 }
 
-void do_assert_glerror()
+void assert_glerror()
 {
-	if (LL_UNLIKELY(!gGLManager.mInited))
+	if (!gGLActive)
+	{
+		//llwarns << "GL used while not active!" << llendl;
+
+		if (gDebugSession)
+		{
+			//ll_fail("GL used while not active");
+		}
+	}
+
+	if (gNoRender || !gDebugGL) 
+	{
+		return;
+	}
+	
+	if (!gGLManager.mInited)
 	{
 		LL_ERRS("RenderInit") << "GL not initialized" << LL_ENDL;
 	}
@@ -1049,9 +1053,10 @@ void do_assert_glerror()
 	GLenum error;
 	error = glGetError();
 	BOOL quit = FALSE;
-	while (LL_UNLIKELY(error))
+	while (error)
 	{
 		quit = TRUE;
+#ifndef LL_LINUX // *FIX: !  This should be an error for linux as well.
 		GLubyte const * gl_error_msg = gluErrorString(error);
 		if (NULL != gl_error_msg)
 		{
@@ -1075,6 +1080,7 @@ void do_assert_glerror()
 			}
 		}
 		error = glGetError();
+#endif
 	}
 
 	if (quit)
@@ -1090,25 +1096,6 @@ void do_assert_glerror()
 	}
 }
 
-void assert_glerror()
-{
-	if (!gGLActive)
-	{
-		//llwarns << "GL used while not active!" << llendl;
-
-		if (gDebugSession)
-		{
-			//ll_fail("GL used while not active");
-		}
-	}
-
-	if (!gNoRender && gDebugGL) 
-	{
-		do_assert_glerror();
-	}
-}
-	
-
 void clear_glerror()
 {
 	//  Create or update texture to be used with this data 
@@ -1122,7 +1109,7 @@ void clear_glerror()
 //
 
 // Static members
-boost::unordered_map<LLGLenum, LLGLboolean> LLGLState::sStateMap;
+std::map<LLGLenum, LLGLboolean> LLGLState::sStateMap;
 
 GLboolean LLGLDepthTest::sDepthEnabled = GL_FALSE; // OpenGL default
 GLenum LLGLDepthTest::sDepthFunc = GL_LESS; // OpenGL default
@@ -1170,7 +1157,7 @@ void LLGLState::resetTextureStates()
 void LLGLState::dumpStates() 
 {
 	LL_INFOS("RenderState") << "GL States:" << LL_ENDL;
-	for (boost::unordered_map<LLGLenum, LLGLboolean>::iterator iter = sStateMap.begin();
+	for (std::map<LLGLenum, LLGLboolean>::iterator iter = sStateMap.begin();
 		 iter != sStateMap.end(); ++iter)
 	{
 		LL_INFOS("RenderState") << llformat(" 0x%04x : %s",(S32)iter->first,iter->second?"TRUE":"FALSE") << LL_ENDL;
@@ -1206,7 +1193,7 @@ void LLGLState::checkStates(const std::string& msg)
 		}
 	}
 	
-	for (boost::unordered_map<LLGLenum, LLGLboolean>::iterator iter = sStateMap.begin();
+	for (std::map<LLGLenum, LLGLboolean>::iterator iter = sStateMap.begin();
 		 iter != sStateMap.end(); ++iter)
 	{
 		LLGLenum state = iter->first;
diff --git a/indra/llrender/llgl.h b/indra/llrender/llgl.h
index c4f5d91e1a..620aa5bee0 100644
--- a/indra/llrender/llgl.h
+++ b/indra/llrender/llgl.h
@@ -36,7 +36,7 @@
 // This file contains various stuff for handling gl extensions and other gl related stuff.
 
 #include <string>
-#include <boost/unordered_map.hpp>
+#include <map>
 #include <list>
 
 #include "llerror.h"
@@ -87,7 +87,6 @@ public:
 	BOOL mHasCompressedTextures;
 	BOOL mHasFramebufferObject;
 	BOOL mHasFramebufferMultisample;
-	BOOL mHasBlendFuncSeparate;
 	
 	// ARB Extensions
 	BOOL mHasVertexBufferObject;
@@ -120,6 +119,9 @@ public:
 
 	// Misc extensions
 	BOOL mHasSeparateSpecularColor;
+
+	//whether this GPU is in the debug list.
+	BOOL mDebugGPU;
 	
 	S32 mDriverVersionMajor;
 	S32 mDriverVersionMinor;
@@ -148,6 +150,7 @@ private:
 	void initExtensions();
 	void initGLStates();
 	void initGLImages();
+	void setToDebugGPU();
 };
 
 extern LLGLManager gGLManager;
@@ -241,7 +244,7 @@ public:
 	static void checkClientArrays(const std::string& msg = "", U32 data_mask = 0x0001);
 	
 protected:
-	static boost::unordered_map<LLGLenum, LLGLboolean> sStateMap;
+	static std::map<LLGLenum, LLGLboolean> sStateMap;
 	
 public:
 	enum { CURRENT_STATE = -2 };
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 1a48c8a06c..ee26f9819c 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -150,12 +150,12 @@ void LLImageGL::checkTexSize(bool forced) const
 			if (gDebugSession)
 			{
 				gFailLog << "wrong texture size and discard level!" << 
-					mWidth << " Height: " << mHeight << " Current Level: " << mCurrentDiscardLevel << std::endl;
+					mWidth << " Height: " << mHeight << " Current Level: " << (S32)mCurrentDiscardLevel << std::endl;
 			}
 			else
 			{
 				llerrs << "wrong texture size and discard level: width: " << 
-					mWidth << " Height: " << mHeight << " Current Level: " << mCurrentDiscardLevel << llendl ;
+					mWidth << " Height: " << mHeight << " Current Level: " << (S32)mCurrentDiscardLevel << llendl ;
 			}
 		}
 
@@ -1057,8 +1057,12 @@ BOOL LLImageGL::setSubImageFromFrameBuffer(S32 fb_x, S32 fb_y, S32 x_pos, S32 y_
 {
 	if (gGL.getTexUnit(0)->bind(this, false, true))
 	{
-		checkTexSize(true) ;
-		llcallstacks << fb_x << " : " << fb_y << " : " << x_pos << " : " << y_pos << " : " << width << " : " << height << llcallstacksendl ;
+		if(gGLManager.mDebugGPU)
+		{
+			llinfos << "Calling glCopyTexSubImage2D(...)" << llendl ;
+			checkTexSize(true) ;
+			llcallstacks << fb_x << " : " << fb_y << " : " << x_pos << " : " << y_pos << " : " << width << " : " << height << llcallstacksendl ;
+		}
 
 		glCopyTexSubImage2D(GL_TEXTURE_2D, 0, fb_x, fb_y, x_pos, y_pos, width, height);
 		mGLTextureCreated = true;
@@ -1654,7 +1658,7 @@ void LLImageGL::calcAlphaChannelOffsetAndStride()
 	}
 }
 
-void LLImageGL::analyzeAlpha(const void* data_in, U32 w, U32 h)
+void LLImageGL::analyzeAlpha(const void* data_in, S32 w, S32 h)
 {
 	if(!mNeedsAlphaAndPickMask)
 	{
@@ -1662,91 +1666,26 @@ void LLImageGL::analyzeAlpha(const void* data_in, U32 w, U32 h)
 	}
 
 	U32 length = w * h;
-	U32 alphatotal = 0;
+	const GLubyte* current = ((const GLubyte*) data_in) + mAlphaOffset ;
 	
-	U32 sample[16];
-	memset(sample, 0, sizeof(U32)*16);
-
-	// generate histogram of quantized alpha.
-	// also add-in the histogram of a 2x2 box-sampled version.  The idea is
-	// this will mid-skew the data (and thus increase the chances of not
-	// being used as a mask) from high-frequency alpha maps which
-	// suffer the worst from aliasing when used as alpha masks.
-	if (w >= 2 && h >= 2)
-	{
-		llassert(w%2 == 0);
-		llassert(h%2 == 0);
-		const GLubyte* rowstart = ((const GLubyte*) data_in) + mAlphaOffset;
-		for (U32 y = 0; y < h; y+=2)
-		{
-			const GLubyte* current = rowstart;
-			for (U32 x = 0; x < w; x+=2)
-			{
-				const U32 s1 = current[0];
-				alphatotal += s1;
-				const U32 s2 = current[w * mAlphaStride];
-				alphatotal += s2;
-				current += mAlphaStride;
-				const U32 s3 = current[0];
-				alphatotal += s3;
-				const U32 s4 = current[w * mAlphaStride];
-				alphatotal += s4;
-				current += mAlphaStride;
-
-				++sample[s1/16];
-				++sample[s2/16];
-				++sample[s3/16];
-				++sample[s4/16];
-
-				const U32 asum = (s1+s2+s3+s4);
-				alphatotal += asum;
-				sample[asum/(16*4)] += 4;
-			}
-			
-			rowstart += 2 * w * mAlphaStride;
-		}
-		length *= 2; // we sampled everything twice, essentially
-	}
-	else
+	S32 sample[16];
+	memset(sample, 0, sizeof(S32)*16);
+
+	for (U32 i = 0; i < length; i++)
 	{
-		const GLubyte* current = ((const GLubyte*) data_in) + mAlphaOffset;
-		for (U32 i = 0; i < length; i++)
-		{
-			const U32 s1 = *current;
-			alphatotal += s1;
-			++sample[s1/16];
-			current += mAlphaStride;
-		}
+		++sample[*current/16];
+		current += mAlphaStride ;
 	}
-	
-	// if more than 1/16th of alpha samples are mid-range, this
-	// shouldn't be treated as a 1-bit mask
 
-	// also, if all of the alpha samples are clumped on one half
-	// of the range (but not at an absolute extreme), then consider
-	// this to be an intentional effect and don't treat as a mask.
-
-	U32 midrangetotal = 0;
+	U32 total = 0;
 	for (U32 i = 4; i < 11; i++)
 	{
-		midrangetotal += sample[i];
-	}
-	U32 lowerhalftotal = 0;
-	for (U32 i = 0; i < 8; i++)
-	{
-		lowerhalftotal += sample[i];
-	}
-	U32 upperhalftotal = 0;
-	for (U32 i = 8; i < 16; i++)
-	{
-		upperhalftotal += sample[i];
+		total += sample[i];
 	}
 
-	if (midrangetotal > length/16 || // lots of midrange, or
-	    (lowerhalftotal == length && alphatotal != 0) || // all close to transparent but not all totally transparent, or
-	    (upperhalftotal == length && alphatotal != 255*length)) // all close to opaque but not all totally opaque
+	if (total > length/16)
 	{
-		mIsMask = FALSE; // not suitable for masking
+		mIsMask = FALSE;
 	}
 	else
 	{