Merge remote-tracking branch 'remotes/origin/DRTVWR-541' into DRTVWR-546

# Conflicts:
#	indra/llcommon/linden_common.h

10 files changed, 49 insertions, 5 deletions

diff --git a/indra/llcommon/linden_common.h b/indra/llcommon/linden_common.h
index b2c5be6b76..a228fd22be 100644
--- a/indra/llcommon/linden_common.h
+++ b/indra/llcommon/linden_common.h
@@ -28,7 +28,7 @@
 #define LL_LINDEN_COMMON_H
 
 #include "llprofiler.h"
-#if (TRACY_ENABLE)  // hooks for memory profiling
+#if TRACY_ENABLE && !defined(LL_PROFILER_ENABLE_TRACY_OPENGL)  // hooks for memory profiling
 void *tracy_aligned_malloc(size_t size, size_t alignment);
 void  tracy_aligned_free(void *memblock);
 #define _aligned_malloc(X, Y) tracy_aligned_malloc((X), (Y))
@@ -68,6 +68,4 @@ void  tracy_aligned_free(void *memblock);
 #include "llerror.h"
 #include "llfile.h"
 
-#include "llprofiler.h" // must be before fast timer; needed due to LLThreads potentially needing access to tracy
-
 #endif
diff --git a/indra/llcommon/llprofiler.h b/indra/llcommon/llprofiler.h
index 62e649913b..62ec20fa44 100644
--- a/indra/llcommon/llprofiler.h
+++ b/indra/llcommon/llprofiler.h
@@ -45,6 +45,9 @@
 //      #define TRACY_ONLY_LOCALHOST 1
         #define TRACY_ONLY_IPV4      1
         #include "Tracy.hpp"
+
+        // Mutually exclusive with detailed memory tracing
+        #define LL_PROFILER_ENABLE_TRACY_OPENGL 0
     #endif
 
     #if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY
diff --git a/indra/llrender/llglheaders.h b/indra/llrender/llglheaders.h
index 6bca3623e0..3d93cc0762 100644
--- a/indra/llrender/llglheaders.h
+++ b/indra/llrender/llglheaders.h
@@ -812,4 +812,23 @@ extern void glGetBufferPointervARB (GLenum, GLenum, GLvoid* *);
 #define GL_RENDERBUFFER_FREE_MEMORY_ATI            0x87FD
 #endif
 
+#if defined(TRACY_ENABLE) && LL_PROFILER_ENABLE_TRACY_OPENGL
+    // Tracy uses the following:
+    //    glGenQueries
+    //    glGetQueryiv
+    //    glGetQueryObjectiv
+    #define glGenQueries        glGenQueriesARB
+    #define glGetQueryiv        glGetQueryivARB
+    #define glGetQueryObjectiv  glGetQueryObjectivARB
+    #include <tracy/TracyOpenGL.hpp>
+
+    #define LL_PROFILER_GPU_ZONEC(name,color) TracyGpuZoneC(name,color);
+    #define LL_PROFILER_GPU_COLLECT           TracyGpuCollect
+    #define LL_PROFILER_GPU_CONTEXT           TracyGpuContext
+#else
+    #define LL_PROFILER_GPU_ZONEC(name,color) (void)name;(void)color;
+    #define LL_PROFILER_GPU_COLLECT
+    #define LL_PROFILER_GPU_CONTEXT
+#endif
+
 #endif // LL_LLGLHEADERS_H
diff --git a/indra/llrender/llrendertarget.cpp b/indra/llrender/llrendertarget.cpp
index e3c0255290..401085a00b 100644
--- a/indra/llrender/llrendertarget.cpp
+++ b/indra/llrender/llrendertarget.cpp
@@ -437,11 +437,13 @@ void LLRenderTarget::bindTarget()
 									GL_COLOR_ATTACHMENT1,
 									GL_COLOR_ATTACHMENT2,
 									GL_COLOR_ATTACHMENT3};
+			LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x4000FF )
 			glDrawBuffersARB(mTex.size(), drawbuffers);
 		}
 			
 		if (mTex.empty())
 		{ //no color buffer to draw to
+			LL_PROFILER_GPU_ZONEC( "gl.DrawBuffer", 0x0000FF )
 			glDrawBuffer(GL_NONE);
 			glReadBuffer(GL_NONE);
 		}
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 7d2b09ca4a..b69b644ebc 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -615,6 +615,7 @@ void LLVertexBuffer::drawArrays(U32 mode, const std::vector<LLVector3>& pos, con
 		glNormalPointer(GL_FLOAT, 0, norm[0].mV);
 	}
 	LLGLSLShader::startProfile();
+	LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF0000 )
 	glDrawArrays(sGLMode[mode], 0, count);
 	LLGLSLShader::stopProfile(count, mode);
 }
@@ -654,6 +655,7 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto
 	}
 
 	LLGLSLShader::startProfile();
+    LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x80FF80 )
 	glDrawElements(sGLMode[mode], num_indices, GL_UNSIGNED_SHORT, indicesp);
 	LLGLSLShader::stopProfile(num_indices, mode);
 }
@@ -763,6 +765,7 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi
 
 	stop_glerror();
 	LLGLSLShader::startProfile();
+    LL_PROFILER_GPU_ZONEC( "gl.DrawRangeElements", 0xFFFF00 )
 	glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, 
 		idx);
 	LLGLSLShader::stopProfile(count, mode);
@@ -814,6 +817,7 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const
 
 	stop_glerror();
 	LLGLSLShader::startProfile();
+    LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0xA0FFA0 )
 	glDrawElements(sGLMode[mode], count, GL_UNSIGNED_SHORT,
 		((U16*) getIndicesPointer()) + indices_offset);
 	LLGLSLShader::stopProfile(count, mode);
@@ -861,6 +865,7 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const
 		stop_glerror();
 		LLGLSLShader::startProfile();
 		stop_glerror();
+        LL_PROFILER_GPU_ZONEC( "gl.DrawArrays", 0xFF4040 )
 		glDrawArrays(sGLMode[mode], first, count);
 		stop_glerror();
 		LLGLSLShader::stopProfile(count, mode);
diff --git a/indra/llwindow/llwindowwin32.cpp b/indra/llwindow/llwindowwin32.cpp
index 585905893b..714e07da31 100644
--- a/indra/llwindow/llwindowwin32.cpp
+++ b/indra/llwindow/llwindowwin32.cpp
@@ -1748,6 +1748,8 @@ const	S32   max_format  = (S32)num_formats - 1;
 		return FALSE;
 	}
 
+	LL_PROFILER_GPU_CONTEXT
+
 	if (!gGLManager.initGL())
 	{
 		close();
@@ -3475,6 +3477,8 @@ BOOL LLWindowWin32::resetDisplayResolution()
 void LLWindowWin32::swapBuffers()
 {
 	SwapBuffers(mhDC);
+
+    LL_PROFILER_GPU_COLLECT
 }
 
 
diff --git a/indra/newview/llappviewerwin32.cpp b/indra/newview/llappviewerwin32.cpp
index 9daea515e5..fb53a7648d 100644
--- a/indra/newview/llappviewerwin32.cpp
+++ b/indra/newview/llappviewerwin32.cpp
@@ -323,6 +323,10 @@ int APIENTRY WINMAIN(HINSTANCE hInstance,
                      PWSTR     pCmdLine,
                      int       nCmdShow)
 {
+    // Call Tracy first thing to have it allocate memory
+    // https://github.com/wolfpld/tracy/issues/196
+    LL_PROFILER_FRAME_END
+
 	const S32 MAX_HEAPS = 255;
 	DWORD heap_enable_lfh_error[MAX_HEAPS];
 	S32 num_heaps = 0;
diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp
index 4a802ad9aa..6e55d8f66a 100644
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -585,6 +585,7 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color)
 						glTexCoordPointer(2, GL_FLOAT, 8, vol_face.mTexCoords);
 					}
 					gGL.syncMatrices();
+					LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x00FF00 );
 					glDrawElements(GL_TRIANGLES, vol_face.mNumIndices, GL_UNSIGNED_SHORT, vol_face.mIndices);
 					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
 				}
diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp
index efa4a7fd66..0adf58a0bf 100644
--- a/indra/newview/llspatialpartition.cpp
+++ b/indra/newview/llspatialpartition.cpp
@@ -2700,11 +2700,17 @@ void renderPhysicsShape(LLDrawable* drawable, LLVOVolume* volume)
 			glVertexPointer(3, GL_FLOAT, 16, phys_volume->mHullPoints);
 			gGL.diffuseColor4fv(line_color.mV);
 			gGL.syncMatrices();
-			glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);
+			{
+				LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x20FF20 )
+				glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);
+			}
 			
 			gGL.diffuseColor4fv(color.mV);
 			glPolygonMode(GL_FRONT_AND_BACK, GL_FILL);
-			glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);			
+			{
+				LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x40FF40 )
+				glDrawElements(GL_TRIANGLES, phys_volume->mNumHullIndices, GL_UNSIGNED_SHORT, phys_volume->mHullIndices);
+			}
 		}
 		else
 		{
@@ -3222,6 +3228,7 @@ void renderRaycast(LLDrawable* drawablep)
 						gGL.diffuseColor4f(0,1,1,0.5f);
 						glVertexPointer(3, GL_FLOAT, sizeof(LLVector4a), face.mPositions);
 						gGL.syncMatrices();
+						LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0x60FF60 );
 						glDrawElements(GL_TRIANGLES, face.mNumIndices, GL_UNSIGNED_SHORT, face.mIndices);
 					}
 					
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index e7c2d4db39..4e28c8c493 100644
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -11146,6 +11146,7 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar)
 		if (LLPipeline::sRenderDeferred)
 		{
 			GLuint buff = GL_COLOR_ATTACHMENT0;
+			LL_PROFILER_GPU_ZONEC( "gl.DrawBuffersARB", 0x8000FF );
 			glDrawBuffersARB(1, &buff);
 		}