SH-2031 High risk changeset, but potentially high reward. Addresses frame stalls in renderer by never using the fixed function pipeline if shaders are available.

8 files changed, 69 insertions, 27 deletions

diff --git a/indra/llrender/llcubemap.cpp b/indra/llrender/llcubemap.cpp
index fb22d7f1f5..1b10354c22 100644
--- a/indra/llrender/llcubemap.cpp
+++ b/indra/llrender/llcubemap.cpp
@@ -259,7 +259,7 @@ void LLCubeMap::setMatrix(S32 stage)
 	
 	if (mMatrixStage < 0) return;
 	
-	if (stage > 0)
+	//if (stage > 0)
 	{
 		gGL.getTexUnit(stage)->activate();
 	}
@@ -278,17 +278,17 @@ void LLCubeMap::setMatrix(S32 stage)
 	glLoadMatrixf((F32 *)trans.mMatrix);
 	glMatrixMode(GL_MODELVIEW);
 	
-	if (stage > 0)
+	/*if (stage > 0)
 	{
 		gGL.getTexUnit(0)->activate();
-	}
+	}*/
 }
 
 void LLCubeMap::restoreMatrix()
 {
 	if (mMatrixStage < 0) return;
 
-	if (mMatrixStage > 0)
+	//if (mMatrixStage > 0)
 	{
 		gGL.getTexUnit(mMatrixStage)->activate();
 	}
@@ -296,10 +296,10 @@ void LLCubeMap::restoreMatrix()
 	glPopMatrix();
 	glMatrixMode(GL_MODELVIEW);
 	
-	if (mMatrixStage > 0)
+	/*if (mMatrixStage > 0)
 	{
 		gGL.getTexUnit(0)->activate();
-	}
+	}*/
 }
 
 void LLCubeMap::setReflection (void)
diff --git a/indra/llrender/llgl.cpp b/indra/llrender/llgl.cpp
index e07ff0015c..8937726209 100644
--- a/indra/llrender/llgl.cpp
+++ b/indra/llrender/llgl.cpp
@@ -986,12 +986,12 @@ void LLGLManager::initExtensions()
 	}
 	if (mHasSync)
 	{
-		glFenceSync = (PFNGLFENCESYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glFenceSync");

-		glIsSync = (PFNGLISSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glIsSync");

-		glDeleteSync = (PFNGLDELETESYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glDeleteSync");

-		glClientWaitSync = (PFNGLCLIENTWAITSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glClientWaitSync");

-		glWaitSync = (PFNGLWAITSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glWaitSync");

-		glGetInteger64v = (PFNGLGETINTEGER64VPROC) GLH_EXT_GET_PROC_ADDRESS("glGetInteger64v");

+		glFenceSync = (PFNGLFENCESYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glFenceSync");
+		glIsSync = (PFNGLISSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glIsSync");
+		glDeleteSync = (PFNGLDELETESYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glDeleteSync");
+		glClientWaitSync = (PFNGLCLIENTWAITSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glClientWaitSync");
+		glWaitSync = (PFNGLWAITSYNCPROC) GLH_EXT_GET_PROC_ADDRESS("glWaitSync");
+		glGetInteger64v = (PFNGLGETINTEGER64VPROC) GLH_EXT_GET_PROC_ADDRESS("glGetInteger64v");
 		glGetSynciv = (PFNGLGETSYNCIVPROC) GLH_EXT_GET_PROC_ADDRESS("glGetSynciv");
 	}
 	if (mHasMapBufferRange)
@@ -1379,6 +1379,8 @@ void LLGLState::checkStates(const std::string& msg)
 	glGetIntegerv(GL_BLEND_SRC, &src);
 	glGetIntegerv(GL_BLEND_DST, &dst);
 	
+	stop_glerror();
+
 	BOOL error = FALSE;
 
 	if (src != GL_SRC_ALPHA || dst != GL_ONE_MINUS_SRC_ALPHA)
@@ -1399,7 +1401,9 @@ void LLGLState::checkStates(const std::string& msg)
 	{
 		LLGLenum state = iter->first;
 		LLGLboolean cur_state = iter->second;
+		stop_glerror();
 		LLGLboolean gl_state = glIsEnabled(state);
+		stop_glerror();
 		if(cur_state != gl_state)
 		{
 			dumpStates();
@@ -1424,11 +1428,11 @@ void LLGLState::checkStates(const std::string& msg)
 
 void LLGLState::checkTextureChannels(const std::string& msg)
 {
+#if 0
 	if (!gDebugGL)
 	{
 		return;
 	}
-
 	stop_glerror();
 
 	GLint activeTexture;
@@ -1594,6 +1598,7 @@ void LLGLState::checkTextureChannels(const std::string& msg)
 			LL_GL_ERRS << "GL texture state corruption detected.  " << msg << LL_ENDL;
 		}
 	}
+#endif
 }
 
 void LLGLState::checkClientArrays(const std::string& msg, U32 data_mask)
@@ -1710,7 +1715,7 @@ void LLGLState::checkClientArrays(const std::string& msg, U32 data_mask)
 		}
 	}
 
-	if (glIsEnabled(GL_TEXTURE_2D))
+	/*if (glIsEnabled(GL_TEXTURE_2D))
 	{
 		if (!(data_mask & 0x0008))
 		{
@@ -1733,7 +1738,7 @@ void LLGLState::checkClientArrays(const std::string& msg, U32 data_mask)
 				gFailLog << "GL does not have GL_TEXTURE_2D enabled on channel 1." << std::endl;
 			}
 		}
-	}
+	}*/
 
 	glClientActiveTextureARB(GL_TEXTURE0_ARB);
 	gGL.getTexUnit(0)->activate();
diff --git a/indra/llrender/llglslshader.cpp b/indra/llrender/llglslshader.cpp
index ad2c662dfc..c582858413 100644
--- a/indra/llrender/llglslshader.cpp
+++ b/indra/llrender/llglslshader.cpp
@@ -49,6 +49,7 @@ using std::make_pair;
 using std::string;
 
 GLhandleARB LLGLSLShader::sCurBoundShader = 0;
+bool LLGLSLShader::sNoFixedFunction = false;
 
 BOOL shouldChange(const LLVector4& v1, const LLVector4& v2)
 {
@@ -376,6 +377,7 @@ BOOL LLGLSLShader::link(BOOL suppress_errors)
 
 void LLGLSLShader::bind()
 {
+	gGL.flush();
 	if (gGLManager.mHasShaderObjects)
 	{
 		glUseProgramObjectARB(mProgramObject);
@@ -390,6 +392,7 @@ void LLGLSLShader::bind()
 
 void LLGLSLShader::unbind()
 {
+	gGL.flush();
 	if (gGLManager.mHasShaderObjects)
 	{
 		stop_glerror();
diff --git a/indra/llrender/llglslshader.h b/indra/llrender/llglslshader.h
index 4922eb6d67..24562c3c42 100644
--- a/indra/llrender/llglslshader.h
+++ b/indra/llrender/llglslshader.h
@@ -67,6 +67,7 @@ public:
 	LLGLSLShader();
 
 	static GLhandleARB sCurBoundShader;
+	static bool sNoFixedFunction;
 
 	void unload();
 	BOOL createShader(std::vector<std::string> * attributes,
diff --git a/indra/llrender/llimagegl.cpp b/indra/llrender/llimagegl.cpp
index 60a5962234..9ca3a23d52 100644
--- a/indra/llrender/llimagegl.cpp
+++ b/indra/llrender/llimagegl.cpp
@@ -1414,6 +1414,8 @@ BOOL LLImageGL::readBackRaw(S32 discard_level, LLImageRaw* imageraw, bool compre
 
 void LLImageGL::deleteDeadTextures()
 {
+	bool reset = false;
+
 	while (!sDeadTextureList.empty())
 	{
 		GLuint tex = sDeadTextureList.front();
@@ -1426,12 +1428,22 @@ void LLImageGL::deleteDeadTextures()
 			{
 				tex_unit->unbind(tex_unit->getCurrType());
 				stop_glerror();
+
+				if (i > 0)
+				{
+					reset = true;
+				}
 			}
 		}
 		
 		glDeleteTextures(1, &tex);
 		stop_glerror();
 	}
+
+	if (reset)
+	{
+		gGL.getTexUnit(0)->activate();
+	}
 }
 		
 void LLImageGL::destroyGLTexture()
diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp
index 1d82dda30f..70df1dd1d1 100644
--- a/indra/llrender/llrender.cpp
+++ b/indra/llrender/llrender.cpp
@@ -46,6 +46,7 @@ S32	gGLViewport[4];
 
 U32 LLRender::sUICalls = 0;
 U32 LLRender::sUIVerts = 0;
+U32 LLTexUnit::sWhiteTexture = 0;
 
 static const U32 LL_NUM_TEXTURE_LAYERS = 32; 
 static const U32 LL_NUM_LIGHT_UNITS = 8;
@@ -126,7 +127,8 @@ void LLTexUnit::refreshState(void)
 	// Per apple spec, don't call glEnable/glDisable when index exceeds max texture units
 	// http://www.mailinglistarchive.com/html/mac-opengl@lists.apple.com/2008-07/msg00653.html
 	//
-	bool enableDisable = (mIndex < gGLManager.mNumTextureUnits) && mCurrTexType != LLTexUnit::TT_MULTISAMPLE_TEXTURE;
+	bool enableDisable = !LLGLSLShader::sNoFixedFunction && 
+		(mIndex < gGLManager.mNumTextureUnits) && mCurrTexType != LLTexUnit::TT_MULTISAMPLE_TEXTURE;
 		
 	if (mCurrTexType != TT_NONE)
 	{
@@ -184,7 +186,8 @@ void LLTexUnit::enable(eTextureType type)
 		mCurrTexType = type;
 
 		gGL.flush();
-		if (type != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
+		if (!LLGLSLShader::sNoFixedFunction && 
+			type != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
 			mIndex < gGLManager.mNumTextureUnits)
 		{
 			glEnable(sGLTextureType[type]);
@@ -201,7 +204,8 @@ void LLTexUnit::disable(void)
 		activate();
 		unbind(mCurrTexType);
 		gGL.flush();
-		if (mCurrTexType != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
+		if (!LLGLSLShader::sNoFixedFunction &&
+			mCurrTexType != LLTexUnit::TT_MULTISAMPLE_TEXTURE &&
 			mIndex < gGLManager.mNumTextureUnits)
 		{
 			glDisable(sGLTextureType[mCurrTexType]);
@@ -403,7 +407,14 @@ void LLTexUnit::unbind(eTextureType type)
 
 		activate();
 		mCurrTexture = 0;
-		glBindTexture(sGLTextureType[type], 0);
+		if (LLGLSLShader::sNoFixedFunction && type == LLTexUnit::TT_TEXTURE)
+		{
+			glBindTexture(sGLTextureType[type], sWhiteTexture);
+		}
+		else
+		{
+			glBindTexture(sGLTextureType[type], 0);
+		}
 		stop_glerror();
 	}
 }
@@ -474,6 +485,11 @@ void LLTexUnit::setTextureFilteringOption(LLTexUnit::eTextureFilterOptions optio
 
 void LLTexUnit::setTextureBlendType(eTextureBlendType type)
 {
+	if (LLGLSLShader::sNoFixedFunction)
+	{ //texture blend type means nothing when using shaders
+		return;
+	}
+
 	if (mIndex < 0) return;
 
 	// Do nothing if it's already correctly set.
@@ -594,6 +610,11 @@ GLint LLTexUnit::getTextureSourceType(eTextureBlendSrc src, bool isAlpha)
 
 void LLTexUnit::setTextureCombiner(eTextureBlendOp op, eTextureBlendSrc src1, eTextureBlendSrc src2, bool isAlpha)
 {
+	if (LLGLSLShader::sNoFixedFunction)
+	{ //register combiners do nothing when not using fixed function
+		return;
+	}	
+
 	if (mIndex < 0) return;
 
 	activate();
diff --git a/indra/llrender/llrender.h b/indra/llrender/llrender.h
index 41e7b35341..9eedebe2ce 100644
--- a/indra/llrender/llrender.h
+++ b/indra/llrender/llrender.h
@@ -52,6 +52,8 @@ class LLTexUnit
 {
 	friend class LLRender;
 public:
+	static U32 sWhiteTexture;
+
 	typedef enum
 	{
 		TT_TEXTURE = 0,			// Standard 2D Texture
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 82c5efe0ac..1180afa631 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -35,6 +35,8 @@
 #include "llmemtype.h"
 #include "llrender.h"
 #include "llvector4a.h"
+#include "llglslshader.h"
+
 
 //============================================================================
 
@@ -1113,8 +1115,7 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, offset, length, 
 							GL_MAP_WRITE_BIT | 
 							GL_MAP_FLUSH_EXPLICIT_BIT | 
-							GL_MAP_INVALIDATE_RANGE_BIT | 
-							GL_MAP_UNSYNCHRONIZED_BIT);
+							GL_MAP_INVALIDATE_RANGE_BIT);
 #endif
 					}
 					else
@@ -1122,8 +1123,7 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 #ifdef GL_ARB_map_buffer_range
 						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, 0, mSize, 
 							GL_MAP_WRITE_BIT | 
-							GL_MAP_FLUSH_EXPLICIT_BIT |
-							GL_MAP_UNSYNCHRONIZED_BIT);
+							GL_MAP_FLUSH_EXPLICIT_BIT);
 #endif
 					}
 				}
@@ -1280,8 +1280,7 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, 
 							GL_MAP_WRITE_BIT | 
 							GL_MAP_FLUSH_EXPLICIT_BIT | 
-							GL_MAP_INVALIDATE_RANGE_BIT |
-							GL_MAP_UNSYNCHRONIZED_BIT);
+							GL_MAP_INVALIDATE_RANGE_BIT);
 #endif
 					}
 					else
@@ -1289,8 +1288,7 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 #ifdef GL_ARB_map_buffer_range
 						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, sizeof(U16)*mNumIndices, 
 							GL_MAP_WRITE_BIT | 
-							GL_MAP_FLUSH_EXPLICIT_BIT |
-							GL_MAP_UNSYNCHRONIZED_BIT);
+							GL_MAP_FLUSH_EXPLICIT_BIT);
 #endif
 					}
 				}