Callstack capture for ZoneBegin.

Tracy.hpp

@@ -66,9 +66,19 @@
 #define TracyFree( ptr ) tracy::Profiler::MemFree( ptr );
 
 #ifdef TRACY_HAS_CALLSTACK
+#  define ZoneScopedS( depth ) static const tracy::SourceLocation __tracy_source_location { nullptr, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::ScopedZone ___tracy_scoped_zone( &__tracy_source_location, depth );
+#  define ZoneScopedNS( name, depth ) static const tracy::SourceLocation __tracy_source_location { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::ScopedZone ___tracy_scoped_zone( &__tracy_source_location, depth );
+#  define ZoneScopedCS( color, depth ) __tracy_source_location { nullptr, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::ScopedZone ___tracy_scoped_zone( &__tracy_source_location, depth );
+#  define ZoneScopedNCS( name, color, depth ) static const tracy::SourceLocation __tracy_source_location { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::ScopedZone ___tracy_scoped_zone( &__tracy_source_location, depth );
+
 #  define TracyAllocS( ptr, size, depth ) tracy::Profiler::MemAllocCallstack( ptr, size, depth );
 #  define TracyFreeS( ptr, depth ) tracy::Profiler::MemFreeCallstack( ptr, depth );
 #else
+#  define ZoneScopedS( depth ) ZoneScoped
+#  define ZoneScopedNS( name, depth ) ZoneScopedN( name )
+#  define ZoneScopedCS( color, depth ) ZoneScopedC( color )
+#  define ZoneScopedNCS( name, color, depth ) ZoneScopedNC( name, color )
+
 #  define TracyAllocS( ptr, size, depth ) TracyAlloc( ptr, size )
 #  define TracyFreeS( ptr, depth ) TracyFree( ptr )
 #endif

client/TracyProfiler.cpp

@@ -389,6 +389,11 @@ Profiler::DequeueStatus Profiler::Dequeue( moodycamel::ConsumerToken& token )
                     SendSourceLocationPayload( ptr );
                     tracy_free( (void*)ptr );
                     break;
+                case QueueType::Callstack:
+                    ptr = MemRead<uint64_t>( &item->callstack.ptr );
+                    SendCallstackPayload( ptr );
+                    tracy_free( (void*)ptr );
+                    break;
                 default:
                     assert( false );
                     break;

client/TracyProfiler.hpp

@@ -238,6 +238,21 @@ public:
         s_profiler.m_serialLock.unlock();
     }
 
+    static tracy_force_inline void SendCallstack( int depth, uint64_t thread )
+    {
+#ifdef TRACY_HAS_CALLSTACK
+        auto ptr = Callstack( depth );
+        Magic magic;
+        auto& token = s_token.ptr;
+        auto& tail = token->get_tail_index();
+        auto item = token->enqueue_begin<moodycamel::CanAlloc>( magic );
+        MemWrite( &item->hdr.type, QueueType::Callstack );
+        MemWrite( &item->callstack.ptr, ptr );
+        MemWrite( &item->callstack.thread, thread );
+        tail.store( magic + 1, std::memory_order_release );
+#endif
+    }
+
     static bool ShouldExit();
 
 private:

client/TracyScoped.hpp

@@ -36,6 +36,29 @@ public:
         tail.store( magic + 1, std::memory_order_release );
     }
 
+    tracy_force_inline ScopedZone( const SourceLocation* srcloc, int depth )
+    {
+        const auto thread = GetThreadHandle();
+        m_thread = thread;
+        Magic magic;
+        auto& token = s_token.ptr;
+        auto& tail = token->get_tail_index();
+        auto item = token->enqueue_begin<moodycamel::CanAlloc>( magic );
+        MemWrite( &item->hdr.type, QueueType::ZoneBeginCallstack );
+#ifdef TRACY_RDTSCP_OPT
+        MemWrite( &item->zoneBegin.time, Profiler::GetTime( item->zoneBegin.cpu ) );
+#else
+        uint32_t cpu;
+        MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
+        MemWrite( &item->zoneBegin.cpu, cpu );
+#endif
+        MemWrite( &item->zoneBegin.thread, thread );
+        MemWrite( &item->zoneBegin.srcloc, (uint64_t)srcloc );
+        tail.store( magic + 1, std::memory_order_release );
+
+        s_profiler.SendCallstack( depth, thread );
+    }
+
     tracy_force_inline ~ScopedZone()
     {
         Magic magic;

common/TracyQueue.hpp

@@ -12,8 +12,10 @@ enum class QueueType : uint8_t
     Message,
     ZoneBeginAllocSrcLoc,
     CallstackMemory,
+    Callstack,
     Terminate,
     ZoneBegin,
+    ZoneBeginCallstack,
     ZoneEnd,
     FrameMarkMsg,
     SourceLocation,
@@ -216,6 +218,12 @@ struct QueueCallstackMemory
     uint64_t ptr;
 };
 
+struct QueueCallstack
+{
+    uint64_t ptr;
+    uint64_t thread;
+};
+
 struct QueueCallstackFrame
 {
     uint64_t ptr;
@@ -259,6 +267,7 @@ struct QueueItem
         QueueMemAlloc memAlloc;
         QueueMemFree memFree;
         QueueCallstackMemory callstackMemory;
+        QueueCallstack callstack;
         QueueCallstackFrame callstackFrame;
     };
 };
@@ -272,9 +281,11 @@ static const size_t QueueDataSize[] = {
     sizeof( QueueHeader ) + sizeof( QueueMessage ),
     sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // allocated source location
     sizeof( QueueHeader ) + sizeof( QueueCallstackMemory ),
+    sizeof( QueueHeader ) + sizeof( QueueCallstack ),
     // above items must be first
     sizeof( QueueHeader ),                                  // terminate
     sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),
+    sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // callstack
     sizeof( QueueHeader ) + sizeof( QueueZoneEnd ),
     sizeof( QueueHeader ) + sizeof( QueueFrameMark ),
     sizeof( QueueHeader ) + sizeof( QueueSourceLocation ),