livie
/
phoenix-firestorm
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix Tracy memory profiling overloads for aligned allocations 

Fix disabling renderdoc support
Improve ll_aligned_alloc functions on darwin for 32 and 64byte aligned by utilizing posix_memalign
master
Rye 2025-02-02 02:43:46 -05:00
parent 51ed6b5424
commit 6fcd349f37
20 changed files with 168 additions and 126 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
autobuild.xml
View File

@ -2433,11 +2433,11 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors</string>
<key>archive</key>
<map>
<key>hash</key>
<string>226225ec049826c35adc5e897e0398ed64d4bedb</string>
<string>0c3d01b7e9e39c23f0f40c56a1a04d1fba08ead0</string>
<key>hash_algorithm</key>
<string>sha1</string>
<key>url</key>
<string>https://github.com/secondlife/3p-tracy/releases/download/v0.11.0%2Br1/tracy-v0.11.0.10376230034-darwin64-10376230034.tar.zst</string>
<string>https://github.com/secondlife/3p-tracy/releases/download/v0.11.1-r1/tracy-v0.11.1.11706699176-darwin64-11706699176.tar.zst</string>
</map>
<key>name</key>
<string>darwin64</string>
@ -2447,11 +2447,11 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors</string>
<key>archive</key>
<map>
<key>hash</key>
<string>8c5429d1a1486f40cf7e5e88a232222d1fa4f78e</string>
<string>b46cef5646a8d0471ab6256fe5119220fa238772</string>
<key>hash_algorithm</key>
<string>sha1</string>
<key>url</key>
<string>https://github.com/secondlife/3p-tracy/releases/download/v0.11.0%2Br1/tracy-v0.11.0.10376230034-windows64-10376230034.tar.zst</string>
<string>https://github.com/secondlife/3p-tracy/releases/download/v0.11.1-r1/tracy-v0.11.1.11706699176-windows64-11706699176.tar.zst</string>
</map>
<key>name</key>
<string>windows64</string>
@ -2461,11 +2461,11 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors</string>
<key>archive</key>
<map>
<key>hash</key>
<string>ed0664a009aba1dcf1246d845839f524e857162e</string>
<string>beab04c9ea6036b1851a485b65c66cf6a38f0be4</string>
<key>hash_algorithm</key>
<string>sha1</string>
<key>url</key>
<string>https://github.com/secondlife/3p-tracy/releases/download/v0.11.0%2Br1/tracy-v0.11.0.10376230034-linux64-10376230034.tar.zst</string>
<string>https://github.com/secondlife/3p-tracy/releases/download/v0.11.1-r1/tracy-v0.11.1.11706699176-linux64-11706699176.tar.zst</string>
</map>
<key>name</key>
<string>linux64</string>
@ -2478,7 +2478,7 @@ Copyright (c) 2012, 2014, 2015, 2016 nghttp2 contributors</string>
<key>copyright</key>
<string>Copyright (c) 2017-2024, Bartosz Taudul (wolf@nereid.pl)</string>
<key>version</key>
<string>v0.11.0.10376230034</string>
<string>v0.11.1.11706699176</string>
<key>name</key>
<string>tracy</string>
<key>canonical_repo</key>

1
indra/cmake/CMakeLists.txt
View File

@ -55,6 +55,7 @@ set(cmake_SOURCE_FILES
TemplateCheck.cmake
TinyEXR.cmake
TinyGLTF.cmake
Tracy.cmake
Tut.cmake
UI.cmake
UnixInstall.cmake

6
indra/cmake/Tracy.cmake
View File

@ -15,6 +15,7 @@ endif()
if (USE_TRACY)
option(USE_TRACY_ON_DEMAND "Use on-demand Tracy profiling." ON)
option(USE_TRACY_LOCAL_ONLY "Disallow remote Tracy profiling." OFF)
option(USE_TRACY_GPU "Use Tracy GPU profiling" OFF)
use_system_binary(tracy)
use_prebuilt_binary(tracy)
@ -31,9 +32,8 @@ if (USE_TRACY)
target_compile_definitions(ll::tracy INTERFACE -DTRACY_NO_BROADCAST=1 -DTRACY_ONLY_LOCALHOST=1)
endif ()
# GHA runners don't always provide invariant TSC support, but always build with LL_TESTS enabled
if (DARWIN AND LL_TESTS)
target_compile_definitions(ll::tracy INTERFACE -DTRACY_TIMER_FALLBACK=1)
if (USE_TRACY_GPU AND NOT DARWIN) # Tracy OpenGL mode is incompatible with macOS/iOS
target_compile_definitions(ll::tracy INTERFACE -DLL_PROFILER_ENABLE_TRACY_OPENGL=1)
endif ()
# See: indra/llcommon/llprofiler.h

6
indra/llcommon/linden_common.h
View File

@ -28,12 +28,6 @@
#define LL_LINDEN_COMMON_H
#include "llprofiler.h"
#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))
#define _aligned_free(X) tracy_aligned_free((X))
#endif
// *NOTE: Please keep includes here to a minimum!
//

76
indra/llcommon/llcommon.cpp
View File

@ -33,23 +33,12 @@
#include "lltracethreadrecorder.h"
#include "llcleanup.h"
thread_local bool gProfilerEnabled = false;
#if (TRACY_ENABLE)
#if LL_PROFILER_CONFIGURATION >= LL_PROFILER_CONFIG_TRACY && TRACY_ENABLE
// Override new/delete for tracy memory profiling
void* ll_tracy_new(size_t size)
{
void* ptr;
if (gProfilerEnabled)
{
//LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY;
ptr = (malloc)(size);
}
else
{
ptr = (malloc)(size);
}
void* ptr = (malloc)(size);
if (!ptr)
{
throw std::bad_alloc();
@ -58,6 +47,29 @@ void* ll_tracy_new(size_t size)
return ptr;
}
void* ll_tracy_aligned_new(size_t size, size_t alignment)
{
void* ptr = ll_aligned_malloc_fallback(size, alignment);
if (!ptr)
{
throw std::bad_alloc();
}
LL_PROFILE_ALLOC(ptr, size);
return ptr;
}
void ll_tracy_delete(void* ptr)
{
LL_PROFILE_FREE(ptr);
(free)(ptr);
}
void ll_tracy_aligned_delete(void* ptr)
{
LL_PROFILE_FREE(ptr);
ll_aligned_free_fallback(ptr);
}
void* operator new(size_t size)
{
return ll_tracy_new(size);
@ -68,18 +80,14 @@ void* operator new[](std::size_t count)
return ll_tracy_new(count);
}
void ll_tracy_delete(void* ptr)
void* operator new(size_t size, std::align_val_t align)
{
LL_PROFILE_FREE(ptr);
if (gProfilerEnabled)
{
//LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY;
(free)(ptr);
}
else
{
(free)(ptr);
}
return ll_tracy_aligned_new(size, (size_t)align);
}
void* operator new[](std::size_t count, std::align_val_t align)
{
return ll_tracy_aligned_new(count, (size_t)align);
}
void operator delete(void *ptr) noexcept
@ -92,27 +100,17 @@ void operator delete[](void* ptr) noexcept
ll_tracy_delete(ptr);
}
// C-style malloc/free can't be so easily overridden, so we define tracy versions and use
// a pre-processor #define in linden_common.h to redirect to them. The parens around the native
// functions below prevents recursive substitution by the preprocessor.
//
// Unaligned mallocs are rare in LL code but hooking them causes problems in 3p lib code (looking at
// you, Havok), so we'll only capture the aligned version.
void *tracy_aligned_malloc(size_t size, size_t alignment)
void operator delete(void *ptr, std::align_val_t align) noexcept
{
auto ptr = ll_aligned_malloc_fallback(size, alignment);
if (ptr) LL_PROFILE_ALLOC(ptr, size);
return ptr;
ll_tracy_aligned_delete(ptr);
}
void tracy_aligned_free(void *memblock)
void operator delete[](void* ptr, std::align_val_t align) noexcept
{
LL_PROFILE_FREE(memblock);
ll_aligned_free_fallback(memblock);
ll_tracy_aligned_delete(ptr);
}
#endif
#endif // TRACY_ENABLE && !LL_PROFILER_ENABLE_TRACY_OPENGL
//static
bool LLCommon::sAprInitialized = false;

49
indra/llcommon/llmemory.h
View File

@ -231,8 +231,6 @@ inline void* ll_aligned_malloc_32(size_t size) // returned hunk MUST be freed wi
LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY;
#if defined(LL_WINDOWS)
void* ret = _aligned_malloc(size, 32);
#elif defined(LL_DARWIN)
void* ret = ll_aligned_malloc_fallback( size, 32 );
#else
void *ret;
if (0 != posix_memalign(&ret, 32, size))
@ -248,8 +246,31 @@ inline void ll_aligned_free_32(void *p)
LL_PROFILE_FREE(p);
#if defined(LL_WINDOWS)
_aligned_free(p);
#elif defined(LL_DARWIN)
ll_aligned_free_fallback( p );
#else
free(p); // posix_memalign() is compatible with heap deallocator
#endif
}
inline void* ll_aligned_malloc_64(size_t size) // returned hunk MUST be freed with ll_aligned_free_32().
{
LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY;
#if defined(LL_WINDOWS)
void* ret = _aligned_malloc(size, 64);
#else
void *ret;
if (0 != posix_memalign(&ret, 64, size))
return nullptr;
#endif
LL_PROFILE_ALLOC(ret, size);
return ret;
}
inline void ll_aligned_free_64(void *p)
{
LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY;
LL_PROFILE_FREE(p);
#if defined(LL_WINDOWS)
_aligned_free(p);
#else
free(p); // posix_memalign() is compatible with heap deallocator
#endif
@ -261,19 +282,23 @@ LL_FORCE_INLINE void* ll_aligned_malloc(size_t size)
{
LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY;
void* ret;
if (LL_DEFAULT_HEAP_ALIGN % ALIGNMENT == 0)
if constexpr (LL_DEFAULT_HEAP_ALIGN % ALIGNMENT == 0)
{
ret = malloc(size);
LL_PROFILE_ALLOC(ret, size);
}
else if (ALIGNMENT == 16)
else if constexpr (ALIGNMENT == 16)
{
ret = ll_aligned_malloc_16(size);
}
else if (ALIGNMENT == 32)
else if constexpr (ALIGNMENT == 32)
{
ret = ll_aligned_malloc_32(size);
}
else if constexpr (ALIGNMENT == 64)
{
ret = ll_aligned_malloc_64(size);
}
else
{
ret = ll_aligned_malloc_fallback(size, ALIGNMENT);
@ -285,16 +310,20 @@ template<size_t ALIGNMENT>
LL_FORCE_INLINE void ll_aligned_free(void* ptr)
{
LL_PROFILE_ZONE_SCOPED_CATEGORY_MEMORY;
if (ALIGNMENT == LL_DEFAULT_HEAP_ALIGN)
if constexpr (ALIGNMENT == LL_DEFAULT_HEAP_ALIGN)
{
LL_PROFILE_FREE(ptr);
free(ptr);
}
else if (ALIGNMENT == 16)
else if constexpr (ALIGNMENT == 16)
{
ll_aligned_free_16(ptr);
}
else if (ALIGNMENT == 32)
else if constexpr (ALIGNMENT == 32)
{
return ll_aligned_free_32(ptr);
}
else if constexpr (ALIGNMENT == 64)
{
return ll_aligned_free_32(ptr);
}

32
indra/llcommon/llprofiler.h
View File

@ -74,23 +74,18 @@
#define LL_PROFILER_CONFIGURATION LL_PROFILER_CONFIG_FAST_TIMER
#endif
extern thread_local bool gProfilerEnabled;
#if defined(LL_PROFILER_CONFIGURATION) && (LL_PROFILER_CONFIGURATION > LL_PROFILER_CONFIG_NONE)
#if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY || LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER
#include "tracy/Tracy.hpp"
// Enable OpenGL profiling
#define LL_PROFILER_ENABLE_TRACY_OPENGL 0
// Enable RenderDoc labeling
#define LL_PROFILER_ENABLE_RENDER_DOC 0
//#define LL_PROFILER_ENABLE_RENDER_DOC 0
#endif
#if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY
#define LL_PROFILER_FRAME_END FrameMark
#define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ); gProfilerEnabled = true;
#define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name );
#define LL_RECORD_BLOCK_TIME(name) ZoneScoped // Want descriptive names; was: ZoneNamedN( ___tracy_scoped_zone, #name, true );
#define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, name, true );
#define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB
@ -133,7 +128,7 @@ extern thread_local bool gProfilerEnabled;
#endif
#if LL_PROFILER_CONFIGURATION == LL_PROFILER_CONFIG_TRACY_FAST_TIMER
#define LL_PROFILER_FRAME_END FrameMark
#define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name ); gProfilerEnabled = true;
#define LL_PROFILER_SET_THREAD_NAME( name ) tracy::SetThreadName( name );
#define LL_RECORD_BLOCK_TIME(name) ZoneNamedN(___tracy_scoped_zone, #name, true); const LLTrace::BlockTimer& LL_GLUE_TOKENS(block_time_recorder, __LINE__)(LLTrace::timeThisBlock(name)); (void)LL_GLUE_TOKENS(block_time_recorder, __LINE__);
#define LL_PROFILE_ZONE_NAMED(name) ZoneNamedN( ___tracy_scoped_zone, #name, true );
#define LL_PROFILE_ZONE_NAMED_COLOR(name,color) ZoneNamedNC( ___tracy_scopped_zone, name, color, true ) // RGB
@ -158,23 +153,20 @@ extern thread_local bool gProfilerEnabled;
#endif // LL_PROFILER
#if LL_PROFILER_ENABLE_TRACY_OPENGL
#define LL_PROFILE_GPU_ZONE(name) TracyGpuZone(name)
#define LL_PROFILE_GPU_ZONEC(name,color) TracyGpuZoneC(name,color)
#define LL_PROFILE_GPU_ZONE(name) TracyGpuZone(name)
#define LL_PROFILE_GPU_ZONEC(name,color) TracyGpuZoneC(name,color)
#define LL_PROFILER_GPU_COLLECT TracyGpuCollect
#define LL_PROFILER_GPU_CONTEXT TracyGpuContext
// disable memory tracking (incompatible with GPU tracing
#define LL_PROFILE_ALLOC(ptr, size) (void)(ptr); (void)(size);
#define LL_PROFILE_FREE(ptr) (void)(ptr);
#define LL_PROFILER_GPU_CONTEXT_NAMED TracyGpuContextName
#else
#define LL_PROFILE_GPU_ZONE(name) (void)name;
#define LL_PROFILE_GPU_ZONEC(name,color) (void)name;(void)color;
#define LL_PROFILE_GPU_ZONE(name) (void)name;
#define LL_PROFILE_GPU_ZONEC(name,color) (void)name;(void)color;
#define LL_PROFILER_GPU_COLLECT
#define LL_PROFILER_GPU_CONTEXT
#define LL_PROFILER_GPU_CONTEXT_NAMED(name) (void)name;
#endif // LL_PROFILER_ENABLE_TRACY_OPENGL
#define LL_LABEL_OBJECT_GL(type, name, length, label)
#if !LL_DARWIN && LL_PROFILER_CONFIGURATION > 1
#if LL_PROFILER_CONFIGURATION >= LL_PROFILER_CONFIG_TRACY
#define LL_PROFILE_ALLOC(ptr, size) TracyAlloc(ptr, size)
#define LL_PROFILE_FREE(ptr) TracyFree(ptr)
#else
@ -182,8 +174,6 @@ extern thread_local bool gProfilerEnabled;
#define LL_PROFILE_FREE(ptr) (void)(ptr);
#endif
#endif
#if LL_PROFILER_ENABLE_RENDER_DOC
#define LL_LABEL_OBJECT_GL(type, name, length, label) glObjectLabel(type, name, length, label)
#else

4
indra/llrender/llglslshader.cpp
View File

@ -543,7 +543,7 @@ bool LLGLSLShader::createShader()
}
}
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
setLabel(mName.c_str());
#endif
@ -2061,7 +2061,7 @@ LLUUID LLGLSLShader::hash()
return hash_obj.digest();
}
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
void LLGLSLShader::setLabel(const char* label) {
LL_LABEL_OBJECT_GL(GL_PROGRAM, mProgramObject, strlen(label), label);
}

4
indra/llrender/llglslshader.h
View File

@ -360,7 +360,7 @@ public:
// hacky flag used for optimization in LLDrawPoolAlpha
bool mCanBindFast = false;
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
void setLabel(const char* label);
#endif
@ -380,7 +380,7 @@ extern LLGLSLShader gSolidColorProgram;
//Alpha mask shader (declared here so llappearance can access properly)
extern LLGLSLShader gAlphaMaskProgram;
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
#define LL_SET_SHADER_LABEL(shader) shader.setLabel(#shader)
#else
#define LL_SET_SHADER_LABEL(shader, label)

2
indra/llrender/llvertexbuffer.cpp
View File

@ -885,7 +885,7 @@ bool LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_of
return true;
}
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
void LLVertexBuffer::setLabel(const char* label) {
LL_LABEL_OBJECT_GL(GL_BUFFER, mGLBuffer, strlen(label), label);
}

4
indra/llrender/llvertexbuffer.h
View File

@ -279,7 +279,7 @@ public:
//for debugging, validate data in given range is valid
bool validateRange(U32 start, U32 end, U32 count, U32 offset) const;
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
void setLabel(const char* label);
#endif
@ -340,7 +340,7 @@ public:
static U32 sVertexCount;
};
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
#define LL_LABEL_VERTEX_BUFFER(buf, name) buf->setLabel(name)
#else
#define LL_LABEL_VERTEX_BUFFER(buf, name)

7
indra/llwindow/llwindowwin32.cpp
View File

@ -1666,6 +1666,11 @@ const S32 max_format = (S32)num_formats - 1;
return false;
}
// Setup Tracy gpu context
{
LL_PROFILER_GPU_CONTEXT;
}
// Disable vertical sync for swap
toggleVSync(enable_vsync);
@ -1697,8 +1702,6 @@ const S32 max_format = (S32)num_formats - 1;
swapBuffers();
}
LL_PROFILER_GPU_CONTEXT;
return true;
}

37
indra/newview/llappviewer.cpp
View File

@ -1324,6 +1324,7 @@ bool LLAppViewer::frame()
bool LLAppViewer::doFrame()
{
LL_RECORD_BLOCK_TIME(FTM_FRAME);
LL_PROFILE_GPU_ZONE("Frame");
{
// and now adjust the visuals from previous frame.
if(LLPerfStats::tunables.userAutoTuneEnabled && LLPerfStats::tunables.tuningFlag != LLPerfStats::Tunables::Nothing)
@ -1413,24 +1414,26 @@ bool LLAppViewer::doFrame()
if (!LLApp::isExiting())
{
LL_PROFILE_ZONE_NAMED_CATEGORY_APP("df JoystickKeyboard");
pingMainloopTimeout("Main:JoystickKeyboard");
// Scan keyboard for movement keys. Command keys and typing
// are handled by windows callbacks. Don't do this until we're
// done initializing. JC
if (gViewerWindow
&& (gHeadlessClient || gViewerWindow->getWindow()->getVisible())
&& gViewerWindow->getActive()
&& !gViewerWindow->getWindow()->getMinimized()
&& LLStartUp::getStartupState() == STATE_STARTED
&& (gHeadlessClient || !gViewerWindow->getShowProgress())
&& !gFocusMgr.focusLocked())
{
LLPerfStats::RecordSceneTime T (LLPerfStats::StatType_t::RENDER_IDLE);
joystick->scanJoystick();
gKeyboard->scanKeyboard();
gViewerInput.scanMouse();
LL_PROFILE_ZONE_NAMED_CATEGORY_APP("df JoystickKeyboard");
pingMainloopTimeout("Main:JoystickKeyboard");
// Scan keyboard for movement keys. Command keys and typing
// are handled by windows callbacks. Don't do this until we're
// done initializing. JC
if (gViewerWindow
&& (gHeadlessClient || gViewerWindow->getWindow()->getVisible())
&& gViewerWindow->getActive()
&& !gViewerWindow->getWindow()->getMinimized()
&& LLStartUp::getStartupState() == STATE_STARTED
&& (gHeadlessClient || !gViewerWindow->getShowProgress())
&& !gFocusMgr.focusLocked())
{
LLPerfStats::RecordSceneTime T(LLPerfStats::StatType_t::RENDER_IDLE);
joystick->scanJoystick();
gKeyboard->scanKeyboard();
gViewerInput.scanMouse();
}
}
// Update state based on messages, user input, object idle.

5
indra/newview/llappviewerlinux.cpp
View File

@ -73,6 +73,11 @@ static void exceptionTerminateHandler()
int main( int argc, char **argv )
{
// Call Tracy first thing to have it allocate memory
// https://github.com/wolfpld/tracy/issues/196
LL_PROFILER_FRAME_END;
LL_PROFILER_SET_THREAD_NAME("App");
gArgC = argc;
gArgV = argv;

5
indra/newview/llappviewermacosx.cpp
View File

@ -231,6 +231,11 @@ void infos(const std::string& message)
int main( int argc, char **argv )
{
// Call Tracy first thing to have it allocate memory
// https://github.com/wolfpld/tracy/issues/196
LL_PROFILER_FRAME_END;
LL_PROFILER_SET_THREAD_NAME("App");
// Store off the command line args for use later.
gArgC = argc;
gArgV = argv;

4
indra/newview/lldrawpool.h
View File

@ -204,7 +204,7 @@ public:
NUM_RENDER_TYPES,
};
#ifdef LL_PROFILER_ENABLE_RENDER_DOC
#if LL_PROFILER_ENABLE_RENDER_DOC
static inline const char* lookupPassName(U32 pass)
{
switch (pass)
@ -340,7 +340,7 @@ public:
}
}
#else
static inline const char* lookupPass(U32 pass) { return ""; }
static inline const char* lookupPassName(U32 pass) { return ""; }
#endif
LLRenderPass(const U32 type);

11
indra/newview/llheroprobemanager.cpp
View File

@ -81,6 +81,7 @@ void LLHeroProbeManager::update()
}
LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY;
LL_PROFILE_GPU_ZONE("hero manager update");
llassert(!gCubeSnapshot); // assert a snapshot is not in progress
if (LLAppViewer::instance()->logoutRequestSent())
{
@ -282,6 +283,9 @@ void LLHeroProbeManager::renderProbes()
// In effect this simulates single-bounce lighting.
void LLHeroProbeManager::updateProbeFace(LLReflectionMap* probe, U32 face, bool is_dynamic, F32 near_clip)
{
LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY;
LL_PROFILE_GPU_ZONE("hero probe update");
// hacky hot-swap of camera specific render targets
gPipeline.mRT = &gPipeline.mHeroProbeRT;
@ -352,7 +356,7 @@ void LLHeroProbeManager::updateProbeFace(LLReflectionMap* probe, U32 face, bool
for (int i = 0; i < mMipChain.size(); ++i)
{
LL_PROFILE_GPU_ZONE("probe mip");
LL_PROFILE_GPU_ZONE("hero probe mip");
mMipChain[i].bindTarget();
if (i == 0)
{
@ -379,7 +383,7 @@ void LLHeroProbeManager::updateProbeFace(LLReflectionMap* probe, U32 face, bool
if (mip >= 0)
{
LL_PROFILE_GPU_ZONE("probe mip copy");
LL_PROFILE_GPU_ZONE("hero probe mip copy");
mTexture->bind(0);
glCopyTexSubImage3D(GL_TEXTURE_CUBE_MAP_ARRAY, mip, 0, 0, sourceIdx * 6 + face, 0, 0, res, res);
@ -427,7 +431,7 @@ void LLHeroProbeManager::generateRadiance(LLReflectionMap* probe)
for (int i = 0; i < mMipChain.size() / 4; ++i)
{
LL_PROFILE_GPU_ZONE("probe radiance gen");
LL_PROFILE_GPU_ZONE("hero probe radiance gen");
static LLStaticHashedString sMipLevel("mipLevel");
static LLStaticHashedString sRoughness("roughness");
static LLStaticHashedString sWidth("u_width");
@ -474,6 +478,7 @@ void LLHeroProbeManager::updateUniforms()
}
LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY;
LL_PROFILE_GPU_ZONE("hpmu - uniforms")
LLMatrix4a modelview;
modelview.loadu(gGLModelView);

4
indra/newview/llreflectionmapmanager.cpp
View File

@ -210,6 +210,7 @@ void LLReflectionMapManager::update()
}
LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY;
LL_PROFILE_GPU_ZONE("reflection manager update");
llassert(!gCubeSnapshot); // assert a snapshot is not in progress
if (LLAppViewer::instance()->logoutRequestSent())
{
@ -696,6 +697,8 @@ void LLReflectionMapManager::doProbeUpdate()
// In effect this simulates single-bounce lighting.
void LLReflectionMapManager::updateProbeFace(LLReflectionMap* probe, U32 face)
{
LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY;
LL_PROFILE_GPU_ZONE("probe update");
// hacky hot-swap of camera specific render targets
gPipeline.mRT = &gPipeline.mAuxillaryRT;
@ -1011,6 +1014,7 @@ void LLReflectionMapManager::updateUniforms()
}
LL_PROFILE_ZONE_SCOPED_CATEGORY_DISPLAY;
LL_PROFILE_GPU_ZONE("rmmu - uniforms")
// structure for packing uniform buffer object
// see class3/deferred/reflectionProbeF.glsl

2
indra/newview/llviewerdisplay.cpp
View File

@ -407,6 +407,7 @@ static void update_tp_display(bool minimized)
void display(bool rebuild, F32 zoom_factor, int subfield, bool for_snapshot)
{
LL_PROFILE_ZONE_NAMED_CATEGORY_DISPLAY("Render");
LL_PROFILE_GPU_ZONE("Render");
LLPerfStats::RecordSceneTime T (LLPerfStats::StatType_t::RENDER_DISPLAY); // render time capture - This is the main stat for overall rendering.
@ -709,6 +710,7 @@ void display(bool rebuild, F32 zoom_factor, int subfield, bool for_snapshot)
if (gPipeline.RenderMirrors && !gSnapshot)
{
LL_PROFILE_ZONE_NAMED_CATEGORY_DISPLAY("Update hero probes");
LL_PROFILE_GPU_ZONE("hero manager")
gPipeline.mHeroProbeManager.update();
gPipeline.mHeroProbeManager.renderProbes();
}

21
indra/newview/pipeline.cpp
View File

@ -7182,11 +7182,11 @@ extern LLPointer<LLImageGL> gEXRImage;
void LLPipeline::tonemap(LLRenderTarget* src, LLRenderTarget* dst)
{
LL_PROFILE_GPU_ZONE("tonemap");
dst->bindTarget();
// gamma correct lighting
{
LL_PROFILE_GPU_ZONE("tonemap");
static LLCachedControl<bool> buildNoPost(gSavedSettings, "RenderDisablePostProcessing", false);
LLGLDepthTest depth(GL_FALSE, GL_FALSE);
@ -7235,11 +7235,11 @@ void LLPipeline::tonemap(LLRenderTarget* src, LLRenderTarget* dst)
void LLPipeline::gammaCorrect(LLRenderTarget* src, LLRenderTarget* dst)
{
LL_PROFILE_GPU_ZONE("gamma correct");
dst->bindTarget();
// gamma correct lighting
{
LL_PROFILE_GPU_ZONE("gamma correct");
LLGLDepthTest depth(GL_FALSE, GL_FALSE);
static LLCachedControl<bool> buildNoPost(gSavedSettings, "RenderDisablePostProcessing", false);
@ -7290,9 +7290,9 @@ void LLPipeline::copyScreenSpaceReflections(LLRenderTarget* src, LLRenderTarget*
void LLPipeline::generateGlow(LLRenderTarget* src)
{
LL_PROFILE_GPU_ZONE("glow generate");
if (sRenderGlow)
{
LL_PROFILE_GPU_ZONE("glow");
mGlow[2].bindTarget();
mGlow[2].clear();
@ -7401,6 +7401,7 @@ void LLPipeline::generateGlow(LLRenderTarget* src)
void LLPipeline::applyCAS(LLRenderTarget* src, LLRenderTarget* dst)
{
static LLCachedControl<F32> cas_sharpness(gSavedSettings, "RenderCASSharpness", 0.4f);
LL_PROFILE_GPU_ZONE("cas");
if (cas_sharpness == 0.0f || !gCASProgram.isComplete())
{
gPipeline.copyRenderTarget(src, dst);
@ -7445,6 +7446,7 @@ void LLPipeline::applyCAS(LLRenderTarget* src, LLRenderTarget* dst)
void LLPipeline::applyFXAA(LLRenderTarget* src, LLRenderTarget* dst)
{
LL_PROFILE_GPU_ZONE("FXAA");
{
llassert(!gCubeSnapshot);
bool multisample = RenderFSAAType == 1 && gFXAAProgram[0].isComplete() && mFXAAMap.isComplete();
@ -7536,7 +7538,7 @@ void LLPipeline::generateSMAABuffers(LLRenderTarget* src)
// Present everything.
if (multisample)
{
LL_PROFILE_GPU_ZONE("aa");
LL_PROFILE_GPU_ZONE("SMAA Edge");
static LLCachedControl<U32> aa_quality(gSavedSettings, "RenderFSAASamples", 0U);
U32 fsaa_quality = std::clamp(aa_quality(), 0U, 3U);
@ -7648,13 +7650,13 @@ void LLPipeline::generateSMAABuffers(LLRenderTarget* src)
void LLPipeline::applySMAA(LLRenderTarget* src, LLRenderTarget* dst)
{
LL_PROFILE_GPU_ZONE("SMAA");
llassert(!gCubeSnapshot);
bool multisample = RenderFSAAType == 2 && gSMAAEdgeDetectProgram[0].isComplete() && mFXAAMap.isComplete() && mSMAABlendBuffer.isComplete();
// Present everything.
if (multisample)
{
LL_PROFILE_GPU_ZONE("aa");
static LLCachedControl<U32> aa_quality(gSavedSettings, "RenderFSAASamples", 0U);
U32 fsaa_quality = std::clamp(aa_quality(), 0U, 3U);
@ -7732,8 +7734,9 @@ void LLPipeline::copyRenderTarget(LLRenderTarget* src, LLRenderTarget* dst)
void LLPipeline::combineGlow(LLRenderTarget* src, LLRenderTarget* dst)
{
// Go ahead and do our glow combine here in our destination. We blit this later into the front buffer.
LL_PROFILE_GPU_ZONE("glow combine");
// Go ahead and do our glow combine here in our destination. We blit this later into the front buffer.
dst->bindTarget();
{
@ -7752,6 +7755,7 @@ void LLPipeline::combineGlow(LLRenderTarget* src, LLRenderTarget* dst)
void LLPipeline::renderDoF(LLRenderTarget* src, LLRenderTarget* dst)
{
LL_PROFILE_GPU_ZONE("dof");
{
bool dof_enabled =
(RenderDepthOfFieldInEditMode || !LLToolMgr::getInstance()->inBuildMode()) &&
@ -7762,7 +7766,6 @@ void LLPipeline::renderDoF(LLRenderTarget* src, LLRenderTarget* dst)
if (dof_enabled)
{
LL_PROFILE_GPU_ZONE("dof");
LLGLDisable blend(GL_BLEND);
// depth of field focal plane calculations