Move TLS accesses close together.

TracyLua.hpp

@@ -180,13 +180,14 @@ static tracy_force_inline void SendLuaCallstack( lua_State* L, uint32_t depth )
     assert( dst - ptr == spaceNeeded + 4 );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
     MemWrite( &item->hdr.type, QueueType::CallstackAlloc );
     MemWrite( &item->callstackAlloc.ptr, (uint64_t)ptr );
     MemWrite( &item->callstackAlloc.nativePtr, (uint64_t)Callstack( depth ) );
-    MemWrite( &item->callstackAlloc.thread, GetThreadHandle() );
+    MemWrite( &item->callstackAlloc.thread, thread );
     tail.store( magic + 1, std::memory_order_release );
 }
 
@@ -227,6 +228,7 @@ static inline int LuaZoneBeginS( lua_State* L )
     memcpy( ptr + 12 + fsz + 1, dbg.source, ssz + 1 );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
@@ -238,7 +240,7 @@ static inline int LuaZoneBeginS( lua_State* L )
     MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
     MemWrite( &item->zoneBegin.cpu, cpu );
 #endif
-    MemWrite( &item->zoneBegin.thread, GetThreadHandle() );
+    MemWrite( &item->zoneBegin.thread, thread );
     MemWrite( &item->zoneBegin.srcloc, (uint64_t)ptr );
     tail.store( magic + 1, std::memory_order_release );
 
@@ -293,6 +295,7 @@ static inline int LuaZoneBeginNS( lua_State* L )
     memcpy( ptr + 12 + fsz + 1 + ssz + 1, name, nsz );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
@@ -304,7 +307,7 @@ static inline int LuaZoneBeginNS( lua_State* L )
     MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
     MemWrite( &item->zoneBegin.cpu, cpu );
 #endif
-    MemWrite( &item->zoneBegin.thread, GetThreadHandle() );
+    MemWrite( &item->zoneBegin.thread, thread );
     MemWrite( &item->zoneBegin.srcloc, (uint64_t)ptr );
     tail.store( magic + 1, std::memory_order_release );
 
@@ -359,6 +362,7 @@ static inline int LuaZoneBegin( lua_State* L )
     memcpy( ptr + 12 + fsz + 1, dbg.source, ssz + 1 );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
@@ -370,7 +374,7 @@ static inline int LuaZoneBegin( lua_State* L )
     MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
     MemWrite( &item->zoneBegin.cpu, cpu );
 #endif
-    MemWrite( &item->zoneBegin.thread, GetThreadHandle() );
+    MemWrite( &item->zoneBegin.thread, thread );
     MemWrite( &item->zoneBegin.srcloc, (uint64_t)ptr );
     tail.store( magic + 1, std::memory_order_release );
     return 0;
@@ -421,6 +425,7 @@ static inline int LuaZoneBeginN( lua_State* L )
     memcpy( ptr + 12 + fsz + 1 + ssz + 1, name, nsz );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
@@ -432,7 +437,7 @@ static inline int LuaZoneBeginN( lua_State* L )
     MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
     MemWrite( &item->zoneBegin.cpu, cpu );
 #endif
-    MemWrite( &item->zoneBegin.thread, GetThreadHandle() );
+    MemWrite( &item->zoneBegin.thread, thread );
     MemWrite( &item->zoneBegin.srcloc, (uint64_t)ptr );
     tail.store( magic + 1, std::memory_order_release );
     return 0;
@@ -453,6 +458,7 @@ static inline int LuaZoneEnd( lua_State* L )
 #endif
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
@@ -464,7 +470,7 @@ static inline int LuaZoneEnd( lua_State* L )
     MemWrite( &item->zoneEnd.time, Profiler::GetTime( cpu ) );
     MemWrite( &item->zoneEnd.cpu, cpu );
 #endif
-    MemWrite( &item->zoneEnd.thread, GetThreadHandle() );
+    MemWrite( &item->zoneEnd.thread, thread );
     tail.store( magic + 1, std::memory_order_release );
     return 0;
 }
@@ -484,6 +490,7 @@ static inline int LuaZoneText( lua_State* L )
     const auto size = strlen( txt );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto ptr = (char*)tracy_malloc( size+1 );
     memcpy( ptr, txt, size );
@@ -491,7 +498,7 @@ static inline int LuaZoneText( lua_State* L )
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
     MemWrite( &item->hdr.type, QueueType::ZoneText );
-    MemWrite( &item->zoneText.thread, GetThreadHandle() );
+    MemWrite( &item->zoneText.thread, thread );
     MemWrite( &item->zoneText.text, (uint64_t)ptr );
     tail.store( magic + 1, std::memory_order_release );
     return 0;
@@ -512,6 +519,7 @@ static inline int LuaZoneName( lua_State* L )
     const auto size = strlen( txt );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto ptr = (char*)tracy_malloc( size+1 );
     memcpy( ptr, txt, size );
@@ -519,7 +527,7 @@ static inline int LuaZoneName( lua_State* L )
     auto& tail = token->get_tail_index();
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
     MemWrite( &item->hdr.type, QueueType::ZoneName );
-    MemWrite( &item->zoneText.thread, GetThreadHandle() );
+    MemWrite( &item->zoneText.thread, thread );
     MemWrite( &item->zoneText.text, (uint64_t)ptr );
     tail.store( magic + 1, std::memory_order_release );
     return 0;
@@ -535,6 +543,7 @@ static inline int LuaMessage( lua_State* L )
     const auto size = strlen( txt );
 
     Magic magic;
+    const auto thread = GetThreadHandle();
     auto token = GetToken();
     auto ptr = (char*)tracy_malloc( size+1 );
     memcpy( ptr, txt, size );
@@ -543,7 +552,7 @@ static inline int LuaMessage( lua_State* L )
     auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
     MemWrite( &item->hdr.type, QueueType::Message );
     MemWrite( &item->message.time, Profiler::GetTime() );
-    MemWrite( &item->message.thread, GetThreadHandle() );
+    MemWrite( &item->message.thread, thread );
     MemWrite( &item->message.text, (uint64_t)ptr );
     tail.store( magic + 1, std::memory_order_release );
     return 0;

TracyOpenGL.hpp

@@ -102,13 +102,14 @@ public:
 
         const float period = 1.f;
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::GpuNewContext );
         MemWrite( &item->gpuNewContext.cpuTime, tcpu );
         MemWrite( &item->gpuNewContext.gpuTime, tgpu );
-        MemWrite( &item->gpuNewContext.thread, GetThreadHandle() );
+        MemWrite( &item->gpuNewContext.thread, thread );
         MemWrite( &item->gpuNewContext.period, period );
         MemWrite( &item->gpuNewContext.context, m_context );
         MemWrite( &item->gpuNewContext.accuracyBits, (uint8_t)bits );
@@ -238,9 +239,8 @@ public:
         const auto queryId = GetGpuCtx().ptr->NextQueryId();
         glQueryCounter( GetGpuCtx().ptr->TranslateOpenGlQueryId( queryId ), GL_TIMESTAMP );
 
-        const auto thread = GetThreadHandle();
-
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );

TracyVulkan.hpp

@@ -211,13 +211,14 @@ public:
         vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, ctx->m_query, queryId );
 
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::GpuZoneBegin );
         MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
         MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
-        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
+        MemWrite( &item->gpuZoneBegin.thread, thread );
         MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
         MemWrite( &item->gpuZoneBegin.context, ctx->GetId() );
         tail.store( magic + 1, std::memory_order_release );
@@ -233,12 +234,12 @@ public:
 #ifdef TRACY_ON_DEMAND
         if( !m_active ) return;
 #endif
-        const auto thread = GetThreadHandle();
 
         const auto queryId = ctx->NextQueryId();
         vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, ctx->m_query, queryId );
 
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );

client/TracyLock.hpp

@@ -124,12 +124,13 @@ public:
 #endif
 
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::LockRelease );
         MemWrite( &item->lockRelease.id, m_id );
-        MemWrite( &item->lockRelease.thread, GetThreadHandle() );
+        MemWrite( &item->lockRelease.thread, thread );
         MemWrite( &item->lockRelease.time, Profiler::GetTime() );
         tail.store( magic + 1, std::memory_order_release );
     }
@@ -156,12 +157,13 @@ public:
         if( ret )
         {
             Magic magic;
+            const auto thread = GetThreadHandle();
             auto token = GetToken();
             auto& tail = token->get_tail_index();
             auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
             MemWrite( &item->hdr.type, QueueType::LockObtain );
             MemWrite( &item->lockObtain.id, m_id );
-            MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+            MemWrite( &item->lockObtain.thread, thread );
             MemWrite( &item->lockObtain.time, Profiler::GetTime() );
             tail.store( magic + 1, std::memory_order_release );
         }
@@ -183,12 +185,13 @@ public:
 #endif
 
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::LockMark );
         MemWrite( &item->lockMark.id, m_id );
-        MemWrite( &item->lockMark.thread, GetThreadHandle() );
+        MemWrite( &item->lockMark.thread, thread );
         MemWrite( &item->lockMark.srcloc, (uint64_t)srcloc );
         tail.store( magic + 1, std::memory_order_release );
     }
@@ -317,12 +320,13 @@ public:
 #endif
 
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::LockRelease );
         MemWrite( &item->lockRelease.id, m_id );
-        MemWrite( &item->lockRelease.thread, GetThreadHandle() );
+        MemWrite( &item->lockRelease.thread, thread );
         MemWrite( &item->lockRelease.time, Profiler::GetTime() );
         tail.store( magic + 1, std::memory_order_release );
     }
@@ -349,12 +353,13 @@ public:
         if( ret )
         {
             Magic magic;
+            const auto thread = GetThreadHandle();
             auto token = GetToken();
             auto& tail = token->get_tail_index();
             auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
             MemWrite( &item->hdr.type, QueueType::LockObtain );
             MemWrite( &item->lockObtain.id, m_id );
-            MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+            MemWrite( &item->lockObtain.thread, thread );
             MemWrite( &item->lockObtain.time, Profiler::GetTime() );
             tail.store( magic + 1, std::memory_order_release );
         }
@@ -424,12 +429,13 @@ public:
 #endif
 
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::LockSharedRelease );
         MemWrite( &item->lockRelease.id, m_id );
-        MemWrite( &item->lockRelease.thread, GetThreadHandle() );
+        MemWrite( &item->lockRelease.thread, thread );
         MemWrite( &item->lockRelease.time, Profiler::GetTime() );
         tail.store( magic + 1, std::memory_order_release );
     }
@@ -456,12 +462,13 @@ public:
         if( ret )
         {
             Magic magic;
+            const auto thread = GetThreadHandle();
             auto token = GetToken();
             auto& tail = token->get_tail_index();
             auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
             MemWrite( &item->hdr.type, QueueType::LockSharedObtain );
             MemWrite( &item->lockObtain.id, m_id );
-            MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+            MemWrite( &item->lockObtain.thread, thread );
             MemWrite( &item->lockObtain.time, Profiler::GetTime() );
             tail.store( magic + 1, std::memory_order_release );
         }
@@ -483,12 +490,13 @@ public:
 #endif
 
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::LockMark );
         MemWrite( &item->lockMark.id, m_id );
-        MemWrite( &item->lockMark.thread, GetThreadHandle() );
+        MemWrite( &item->lockMark.thread, thread );
         MemWrite( &item->lockMark.srcloc, (uint64_t)srcloc );
         tail.store( magic + 1, std::memory_order_release );
     }

client/TracyProfiler.cpp

@@ -492,8 +492,8 @@ LONG WINAPI CrashFilter( PEXCEPTION_POINTERS pExp )
     }
 
     {
-        const auto thread = GetThreadHandle();
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
@@ -730,8 +730,8 @@ static void CrashHandler( int signal, siginfo_t* info, void* /*ucontext*/ )
     }
 
     {
-        const auto thread = GetThreadHandle();
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
@@ -2065,11 +2065,11 @@ TracyCZoneCtx ___tracy_emit_zone_begin( const struct ___tracy_source_location_da
     ctx.active = active;
 #endif
     if( !ctx.active ) return ctx;
-    const auto thread = tracy::GetThreadHandle();
     const auto id = tracy::GetProfiler().GetNextZoneId();
     ctx.id = id;
 
     tracy::Magic magic;
+    const auto thread = tracy::GetThreadHandle();
     auto token = tracy::GetToken();
     auto& tail = token->get_tail_index();
 #ifndef TRACY_NO_VERIFY
@@ -2107,11 +2107,11 @@ TracyCZoneCtx ___tracy_emit_zone_begin_callstack( const struct ___tracy_source_l
     ctx.active = active;
 #endif
     if( !ctx.active ) return ctx;
-    const auto thread = tracy::GetThreadHandle();
     const auto id = tracy::GetProfiler().GetNextZoneId();
     ctx.id = id;
 
     tracy::Magic magic;
+    const auto thread = tracy::GetThreadHandle();
     auto token = tracy::GetToken();
     auto& tail = token->get_tail_index();
 #ifndef TRACY_NO_VERIFY
@@ -2145,8 +2145,8 @@ TracyCZoneCtx ___tracy_emit_zone_begin_callstack( const struct ___tracy_source_l
 void ___tracy_emit_zone_end( TracyCZoneCtx ctx )
 {
     if( !ctx.active ) return;
-    const auto thread = tracy::GetThreadHandle();
     tracy::Magic magic;
+    const auto thread = tracy::GetThreadHandle();
     auto token = tracy::GetToken();
     auto& tail = token->get_tail_index();
 #ifndef TRACY_NO_VERIFY
@@ -2176,8 +2176,8 @@ void ___tracy_emit_zone_end( TracyCZoneCtx ctx )
 void ___tracy_emit_zone_text( TracyCZoneCtx ctx, const char* txt, size_t size )
 {
     if( !ctx.active ) return;
-    const auto thread = tracy::GetThreadHandle();
     tracy::Magic magic;
+    const auto thread = tracy::GetThreadHandle();
     auto token = tracy::GetToken();
     auto ptr = (char*)tracy::tracy_malloc( size+1 );
     memcpy( ptr, txt, size );
@@ -2204,8 +2204,8 @@ void ___tracy_emit_zone_text( TracyCZoneCtx ctx, const char* txt, size_t size )
 void ___tracy_emit_zone_name( TracyCZoneCtx ctx, const char* txt, size_t size )
 {
     if( !ctx.active ) return;
-    const auto thread = tracy::GetThreadHandle();
     tracy::Magic magic;
+    const auto thread = tracy::GetThreadHandle();
     auto token = tracy::GetToken();
     auto ptr = (char*)tracy::tracy_malloc( size+1 );
     memcpy( ptr, txt, size );

client/TracyProfiler.hpp

@@ -251,6 +251,7 @@ public:
         if( !GetProfiler().IsConnected() ) return;
 #endif
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto ptr = (char*)tracy_malloc( size+1 );
         memcpy( ptr, txt, size );
@@ -259,7 +260,7 @@ public:
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::Message );
         MemWrite( &item->message.time, GetTime() );
-        MemWrite( &item->message.thread, GetThreadHandle() );
+        MemWrite( &item->message.thread, thread );
         MemWrite( &item->message.text, (uint64_t)ptr );
         tail.store( magic + 1, std::memory_order_release );
     }
@@ -270,12 +271,13 @@ public:
         if( !GetProfiler().IsConnected() ) return;
 #endif
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::MessageLiteral );
         MemWrite( &item->message.time, GetTime() );
-        MemWrite( &item->message.thread, GetThreadHandle() );
+        MemWrite( &item->message.thread, thread );
         MemWrite( &item->message.text, (uint64_t)txt );
         tail.store( magic + 1, std::memory_order_release );
     }
@@ -286,6 +288,7 @@ public:
         if( !GetProfiler().IsConnected() ) return;
 #endif
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto ptr = (char*)tracy_malloc( size+1 );
         memcpy( ptr, txt, size );
@@ -294,7 +297,7 @@ public:
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::MessageColor );
         MemWrite( &item->messageColor.time, GetTime() );
-        MemWrite( &item->messageColor.thread, GetThreadHandle() );
+        MemWrite( &item->messageColor.thread, thread );
         MemWrite( &item->messageColor.text, (uint64_t)ptr );
         MemWrite( &item->messageColor.r, uint8_t( ( color       ) & 0xFF ) );
         MemWrite( &item->messageColor.g, uint8_t( ( color >> 8  ) & 0xFF ) );
@@ -308,12 +311,13 @@ public:
         if( !GetProfiler().IsConnected() ) return;
 #endif
         Magic magic;
+        const auto thread = GetThreadHandle();
         auto token = GetToken();
         auto& tail = token->get_tail_index();
         auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
         MemWrite( &item->hdr.type, QueueType::MessageLiteralColor );
         MemWrite( &item->messageColor.time, GetTime() );
-        MemWrite( &item->messageColor.thread, GetThreadHandle() );
+        MemWrite( &item->messageColor.thread, thread );
         MemWrite( &item->messageColor.text, (uint64_t)txt );
         MemWrite( &item->messageColor.r, uint8_t( ( color       ) & 0xFF ) );
         MemWrite( &item->messageColor.g, uint8_t( ( color >> 8  ) & 0xFF ) );