feat add tracy profiler

third_party/tracy/public/common/TracyProtocol.hpp

@@ -9,7 +9,7 @@ namespace tracy
 
 constexpr unsigned Lz4CompressBound( unsigned isize ) { return isize + ( isize / 255 ) + 16; }
 
-enum : uint32_t { ProtocolVersion = 74 };
+enum : uint32_t { ProtocolVersion = 76 };
 enum : uint16_t { BroadcastVersion = 3 };
 
 using lz4sz_t = uint32_t;
@@ -95,7 +95,6 @@ struct WelcomeMessage
     double timerMul;
     int64_t initBegin;
     int64_t initEnd;
-    uint64_t delay;
     uint64_t resolution;
     uint64_t epoch;
     uint64_t exectime;

third_party/tracy/public/common/TracyQueue.hpp

@@ -61,6 +61,7 @@ enum class QueueType : uint8_t
     ThreadWakeup,
     GpuTime,
     GpuContextName,
+    GpuAnnotationName,
     CallstackFrameSize,
     SymbolInformation,
     ExternalNameMetadata,
@@ -111,6 +112,7 @@ enum class QueueType : uint8_t
     SecondStringData,
     MemNamePayload,
     ThreadGroupHint,
+    GpuZoneAnnotation,
     StringData,
     ThreadName,
     PlotName,
@@ -331,7 +333,7 @@ struct QueuePlotDataInt : public QueuePlotDataBase
     int64_t val;
 };
 
-struct QueuePlotDataFloat : public QueuePlotDataBase 
+struct QueuePlotDataFloat : public QueuePlotDataBase
 {
     float val;
 };
@@ -406,7 +408,8 @@ enum class GpuContextType : uint8_t
     Direct3D11,
     Metal,
     Custom,
-    CUDA
+    CUDA,
+    Rocprof
 };
 
 enum GpuContextFlags : uint8_t
@@ -446,6 +449,15 @@ struct QueueGpuZoneEnd
     uint8_t context;
 };
 
+struct QueueGpuZoneAnnotation
+{
+    int64_t noteId;
+    double value;
+    uint32_t thread;
+    uint16_t queryId;
+    uint8_t context;
+};
+
 struct QueueGpuTime
 {
     int64_t gpuTime;
@@ -467,7 +479,7 @@ struct QueueGpuTimeSync
     int64_t cpuTime;
     uint8_t context;
 };
-    
+
 struct QueueGpuContextName
 {
     uint8_t context;
@@ -479,6 +491,18 @@ struct QueueGpuContextNameFat : public QueueGpuContextName
     uint16_t size;
 };
 
+struct QueueGpuAnnotationName
+{
+    int64_t noteId;
+    uint8_t context;
+};
+
+struct QueueGpuAnnotationNameFat : public QueueGpuAnnotationName
+{
+    uint64_t ptr;
+    uint16_t size;
+};
+
 struct QueueMemNamePayload
 {
     uint64_t name;
@@ -756,6 +780,8 @@ struct QueueItem
         QueueGpuTimeSync gpuTimeSync;
         QueueGpuContextName gpuContextName;
         QueueGpuContextNameFat gpuContextNameFat;
+        QueueGpuAnnotationName gpuAnnotationName;
+        QueueGpuAnnotationNameFat gpuAnnotationNameFat;
         QueueMemAlloc memAlloc;
         QueueMemFree memFree;
         QueueMemDiscard memDiscard;
@@ -789,6 +815,7 @@ struct QueueItem
         QueueSourceCodeNotAvailable sourceCodeNotAvailable;
         QueueFiberEnter fiberEnter;
         QueueFiberLeave fiberLeave;
+        QueueGpuZoneAnnotation zoneAnnotation;
     };
 };
 #pragma pack( pop )
@@ -849,6 +876,7 @@ static constexpr size_t QueueDataSize[] = {
     sizeof( QueueHeader ) + sizeof( QueueThreadWakeup ),
     sizeof( QueueHeader ) + sizeof( QueueGpuTime ),
     sizeof( QueueHeader ) + sizeof( QueueGpuContextName ),
+    sizeof( QueueHeader ) + sizeof( QueueGpuAnnotationName ),
     sizeof( QueueHeader ) + sizeof( QueueCallstackFrameSize ),
     sizeof( QueueHeader ) + sizeof( QueueSymbolInformation ),
     sizeof( QueueHeader ),                                  // ExternalNameMetadata - not for wire transfer
@@ -900,6 +928,7 @@ static constexpr size_t QueueDataSize[] = {
     sizeof( QueueHeader ),                                  // second string data
     sizeof( QueueHeader ) + sizeof( QueueMemNamePayload ),
     sizeof( QueueHeader ) + sizeof( QueueThreadGroupHint ),
+    sizeof( QueueHeader ) + sizeof( QueueGpuZoneAnnotation ), // GPU zone annotation
     // keep all QueueStringTransfer below
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // string data
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // thread name

third_party/tracy/public/common/TracySystem.cpp

@@ -10,7 +10,7 @@
 #  endif
 #  include <windows.h>
 #  include <malloc.h>
-#  include "TracyUwp.hpp"
+#  include "TracyWinFamily.hpp"
 #else
 #  include <pthread.h>
 #  include <string.h>
@@ -137,7 +137,7 @@ TRACY_API void SetThreadName( const char* name )
 TRACY_API void SetThreadNameWithHint( const char* name, int32_t groupHint )
 {
 #if defined _WIN32
-#  ifdef TRACY_UWP
+#  if defined TRACY_WIN32_NO_DESKTOP
     static auto _SetThreadDescription = &::SetThreadDescription;
 #  else
     static auto _SetThreadDescription = (t_SetThreadDescription)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "SetThreadDescription" );
@@ -246,7 +246,7 @@ TRACY_API const char* GetThreadName( uint32_t id )
 #endif
 
 #if defined _WIN32
-# ifdef TRACY_UWP
+# if defined TRACY_WIN32_NO_DESKTOP
    static auto _GetThreadDescription = &::GetThreadDescription;
 # else
    static auto _GetThreadDescription = (t_GetThreadDescription)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "GetThreadDescription" );

third_party/tracy/public/common/TracyUwp.hpp

@@ -1,11 +0,0 @@
-#ifndef __TRACYUWP_HPP__
-#define __TRACYUWP_HPP__
-
-#ifdef _WIN32
-#  include <winapifamily.h>
-#  if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP) && !WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
-#    define TRACY_UWP
-#  endif
-#endif
-
-#endif

third_party/tracy/public/common/TracyVersion.hpp

@@ -7,7 +7,7 @@ namespace Version
 {
 enum { Major = 0 };
 enum { Minor = 12 };
-enum { Patch = 2 };
+enum { Patch = 4 };
 }
 }
 

third_party/tracy/public/common/TracyWinFamily.hpp

@@ -0,0 +1,16 @@
+#ifndef __TRACYWINFAMILY_HPP__
+#define __TRACYWINFAMILY_HPP__
+
+#ifdef _WIN32
+#  include <winapifamily.h>
+#  if !WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
+#    define TRACY_WIN32_NO_DESKTOP
+#    if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_GAMES)
+#      define TRACY_GDK
+#    elif WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP)
+#      define TRACY_UWP
+#    endif
+#  endif
+#endif
+
+#endif