Separate message for zone begin with alloc src loc and callstack.

TracyLua.hpp

@@ -227,7 +227,7 @@ static inline int LuaZoneBeginS( lua_State* L )
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::ZoneBeginAllocSrcLoc );
+    MemWrite( &item->hdr.type, QueueType::ZoneBeginAllocSrcLocCallstack );
 #ifdef TRACY_RDTSCP_OPT
     MemWrite( &item->zoneBegin.time, Profiler::GetTime( item->zoneBegin.cpu ) );
 #else
@@ -293,7 +293,7 @@ static inline int LuaZoneBeginNS( lua_State* L )
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::ZoneBeginAllocSrcLoc );
+    MemWrite( &item->hdr.type, QueueType::ZoneBeginAllocSrcLocCallstack );
 #ifdef TRACY_RDTSCP_OPT
     MemWrite( &item->zoneBegin.time, Profiler::GetTime( item->zoneBegin.cpu ) );
 #else

client/TracyProfiler.cpp

@@ -1316,6 +1316,7 @@ static void FreeAssociatedMemory( const QueueItem& item )
         tracy_free( (void*)ptr );
         break;
     case QueueType::ZoneBeginAllocSrcLoc:
+    case QueueType::ZoneBeginAllocSrcLocCallstack:
         ptr = MemRead<uint64_t>( &item.zoneBegin.srcloc );
         tracy_free( (void*)ptr );
         break;
@@ -1392,6 +1393,7 @@ Profiler::DequeueStatus Profiler::Dequeue( moodycamel::ConsumerToken& token )
                     tracy_free( (void*)ptr );
                     break;
                 case QueueType::ZoneBeginAllocSrcLoc:
+                case QueueType::ZoneBeginAllocSrcLocCallstack:
                     ptr = MemRead<uint64_t>( &item->zoneBegin.srcloc );
                     SendSourceLocationPayload( ptr );
                     tracy_free( (void*)ptr );

common/TracyQueue.hpp

@@ -12,6 +12,7 @@ enum class QueueType : uint8_t
     ZoneName,
     Message,
     ZoneBeginAllocSrcLoc,
+    ZoneBeginAllocSrcLocCallstack,
     CallstackMemory,
     Callstack,
     CallstackAlloc,
@@ -327,6 +328,7 @@ static const size_t QueueDataSize[] = {
     sizeof( QueueHeader ) + sizeof( QueueZoneText ),        // zone name
     sizeof( QueueHeader ) + sizeof( QueueMessage ),
     sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // allocated source location
+    sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // allocated source location, callstack
     sizeof( QueueHeader ) + sizeof( QueueCallstackMemory ),
     sizeof( QueueHeader ) + sizeof( QueueCallstack ),
     sizeof( QueueHeader ) + sizeof( QueueCallstack ),       // callstack alloc

server/TracyWorker.cpp

@@ -2316,6 +2316,9 @@ bool Worker::Process( const QueueItem& ev )
     case QueueType::ZoneBeginAllocSrcLoc:
         ProcessZoneBeginAllocSrcLoc( ev.zoneBegin );
         break;
+    case QueueType::ZoneBeginAllocSrcLocCallstack:
+        ProcessZoneBeginAllocSrcLocCallstack( ev.zoneBegin );
+        break;
     case QueueType::ZoneEnd:
         ProcessZoneEnd( ev.zoneEnd );
         break;
@@ -2407,6 +2410,7 @@ bool Worker::Process( const QueueItem& ev )
         ProcessCallstackMemory( ev.callstackMemory );
         break;
     case QueueType::Callstack:
+    case QueueType::CallstackAlloc:
         ProcessCallstack( ev.callstack );
         break;
     case QueueType::CallstackFrameSize:
@@ -2470,13 +2474,11 @@ void Worker::ProcessZoneBeginCallstack( const QueueZoneBegin& ev )
     next.zone = zone;
 }
 
-void Worker::ProcessZoneBeginAllocSrcLoc( const QueueZoneBegin& ev )
+void Worker::ProcessZoneBeginAllocSrcLocImpl( ZoneEvent* zone, const QueueZoneBegin& ev )
 {
     auto it = m_pendingSourceLocationPayload.find( ev.srcloc );
     assert( it != m_pendingSourceLocationPayload.end() );
 
-    auto zone = m_slab.AllocInit<ZoneEvent>();
-
     zone->start = TscTime( ev.time );
     zone->end = -1;
     zone->srcloc = it->second;
@@ -2492,6 +2494,22 @@ void Worker::ProcessZoneBeginAllocSrcLoc( const QueueZoneBegin& ev )
     m_pendingSourceLocationPayload.erase( it );
 }
 
+void Worker::ProcessZoneBeginAllocSrcLoc( const QueueZoneBegin& ev )
+{
+    auto zone = m_slab.AllocInit<ZoneEvent>();
+    ProcessZoneBeginAllocSrcLocImpl( zone, ev );
+}
+
+void Worker::ProcessZoneBeginAllocSrcLocCallstack( const QueueZoneBegin& ev )
+{
+    auto zone = m_slab.AllocInit<ZoneEvent>();
+    ProcessZoneBeginAllocSrcLocImpl( zone, ev );
+
+    auto& next = m_nextCallstack[ev.thread];
+    next.type = NextCallstackType::Zone;
+    next.zone = zone;
+}
+
 void Worker::ProcessZoneEnd( const QueueZoneEnd& ev )
 {
     auto tit = m_threadMap.find( ev.thread );

server/TracyWorker.hpp

@@ -351,6 +351,7 @@ private:
     tracy_force_inline void ProcessZoneBegin( const QueueZoneBegin& ev );
     tracy_force_inline void ProcessZoneBeginCallstack( const QueueZoneBegin& ev );
     tracy_force_inline void ProcessZoneBeginAllocSrcLoc( const QueueZoneBegin& ev );
+    tracy_force_inline void ProcessZoneBeginAllocSrcLocCallstack( const QueueZoneBegin& ev );
     tracy_force_inline void ProcessZoneEnd( const QueueZoneEnd& ev );
     tracy_force_inline void ProcessZoneValidation( const QueueZoneValidation& ev );
     tracy_force_inline void ProcessFrameMark( const QueueFrameMark& ev );
@@ -387,6 +388,7 @@ private:
     tracy_force_inline void ProcessSysTime( const QueueSysTime& ev );
 
     tracy_force_inline void ProcessZoneBeginImpl( ZoneEvent* zone, const QueueZoneBegin& ev );
+    tracy_force_inline void ProcessZoneBeginAllocSrcLocImpl( ZoneEvent* zone, const QueueZoneBegin& ev );
     tracy_force_inline void ProcessGpuZoneBeginImpl( GpuEvent* zone, const QueueGpuZoneBegin& ev );
 
     void ZoneStackFailure( uint64_t thread, const ZoneEvent* ev );