Refactor GPU context id handling.

The gpu backends each used a bare 255 as the sentinel for a not-yet-initialized context id. Define InvalidGpuContextId (-1) in TracyQueue.hpp and replace the scattered 255s and their asserts with it. Id exhaustion is handled by an error message and an assert in the new NextGpuContextId() getter. Non-assert builds continue execution, at which point they are no longer valid. Handling this code path is out of scope here. Various attempts at handling the exhaustion problem have been otherwise purged from the API implementations.
2026-09-02 16:38:30 +00:00 · 2026-08-15 19:55:03 +02:00
parent 54ecee1f74
commit 8fdfd08af0
14 changed files with 109 additions and 112 deletions
--- a/manual/tracy.md
+++ b/manual/tracy.md
@@ -1566,6 +1566,8 @@ Tracy provides bindings for profiling OpenGL, Vulkan, Direct3D 11, Direct3D 12,

 Note that the CPU and GPU timers may be unsynchronized unless you create a calibrated context, but the availability of calibrated contexts is limited. You can try to correct the desynchronization of uncalibrated contexts in the profiler's options (section [5.4](#options)).

+GPU contexts are identified by a process-wide unique id, assigned when the context is created (the context creation macros do this automatically). Ids are 8 bits wide (0--255) and are allocated sequentially, so a process can create at most 256 GPU contexts.
+
 > [!TIP]
 > **Check the scope**
 >
@@ -2203,7 +2205,7 @@ Moreover, there are two sets of functions described below. The standard set send

 A GPU context can be created with the `___tracy_emit_gpu_new_context` function (or the serialized variant). You'll need to specify:

- `context` -- a unique context id.
+- `context` -- a unique context id (8-bit, 0--255; see section [3.10](#gpuprofiling)).

 - `gpuTime` -- an initial GPU timestamp.

--- a/manual/tracy.tex
+++ b/manual/tracy.tex
@@ -1776,6 +1776,8 @@ Tracy provides bindings for profiling OpenGL, Vulkan, Direct3D 11, Direct3D 12,

 Note that the CPU and GPU timers may be unsynchronized unless you create a calibrated context, but the availability of calibrated contexts is limited. You can try to correct the desynchronization of uncalibrated contexts in the profiler's options (section~\ref{options}).

+GPU contexts are identified by a process-wide unique id, assigned when the context is created (the context creation macros do this automatically). Ids are 8 bits wide (0--255) and are allocated sequentially, so a process can create at most 256 GPU contexts.
+
 \begin{bclogo}[
 noborder=true,
 couleur=black!5,
@@ -2480,7 +2482,7 @@ Moreover, there are two sets of functions described below. The standard set send
 A GPU context can be created with the \texttt{\_\_\_tracy\_emit\_gpu\_new\_context} function (or the serialized variant). You'll need to specify:

 \begin{itemize}
-\item \texttt{context} -- a unique context id.
+\item \texttt{context} -- a unique context id (8-bit, 0--255; see section~\ref{gpuprofiling}).
 \item \texttt{gpuTime} -- an initial GPU timestamp.
 \item \texttt{period} -- the timestamp period of the GPU.
 \item \texttt{flags} -- the flags to use.
--- a/public/client/TracyProfiler.cpp
+++ b/public/client/TracyProfiler.cpp
@@ -1240,7 +1240,7 @@ struct ProfilerData
    moodycamel::ConcurrentQueue<QueueItem> queue;
    Profiler profiler;
    std::atomic<uint32_t> lockCounter { 0 };
-    std::atomic<uint8_t> gpuCtxCounter { 0 };
+    std::atomic<uint32_t> gpuCtxCounter { 0 };
    std::atomic<ThreadNameData*> threadNameData { nullptr };
 };

@@ -1385,7 +1385,7 @@ TRACY_API Profiler& MANGLED_NAME_BASED_ON_CONFIG(GetProfiler)() { return GetProf
 TRACY_API moodycamel::ConcurrentQueue<QueueItem>& GetQueue() { return GetProfilerData().queue; }
 TRACY_API int64_t GetInitTime() { return GetProfilerData().initTime; }
 TRACY_API std::atomic<uint32_t>& GetLockCounter() { return GetProfilerData().lockCounter; }
-TRACY_API std::atomic<uint8_t>& GetGpuCtxCounter() { return GetProfilerData().gpuCtxCounter; }
+TRACY_API std::atomic<uint32_t>& GetGpuCtxCounter() { return GetProfilerData().gpuCtxCounter; }
 TRACY_API GpuCtxWrapper& GetGpuCtx() { return GetProfilerThreadData().gpuCtx; }
 TRACY_API uint32_t GetThreadHandle() { return detail::GetThreadHandleImpl(); }
 std::atomic<ThreadNameData*>& GetThreadNameData() { return GetProfilerData().threadNameData; }
@@ -1471,7 +1471,7 @@ static EarlyMainThreadTokenRepair init_order(104) s_earlyMainThreadTokenRepair;
 #  endif

 std::atomic<uint32_t> init_order(104) s_lockCounter( 0 );
-std::atomic<uint8_t> init_order(104) s_gpuCtxCounter( 0 );
+std::atomic<uint32_t> init_order(104) s_gpuCtxCounter( 0 );

 thread_local GpuCtxWrapper init_order(104) s_gpuCtx { nullptr };

@@ -1490,7 +1490,7 @@ TRACY_API Profiler& MANGLED_NAME_BASED_ON_CONFIG(GetProfiler)() { return s_profi
 TRACY_API moodycamel::ConcurrentQueue<QueueItem>& GetQueue() { return s_queue; }
 TRACY_API int64_t GetInitTime() { return s_initTime.val; }
 TRACY_API std::atomic<uint32_t>& GetLockCounter() { return s_lockCounter; }
-TRACY_API std::atomic<uint8_t>& GetGpuCtxCounter() { return s_gpuCtxCounter; }
+TRACY_API std::atomic<uint32_t>& GetGpuCtxCounter() { return s_gpuCtxCounter; }
 TRACY_API GpuCtxWrapper& GetGpuCtx() { return s_gpuCtx; }
 TRACY_API uint32_t GetThreadHandle() { return s_threadHandle.val; }

@@ -1501,6 +1501,18 @@ TRACY_API LuaZoneState& GetLuaZoneState() { return s_luaZoneState; }
 #  endif
 #endif

+TRACY_API int32_t NextGpuContextId()
+{
+    const auto id = GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed );
+    if( id > UINT8_MAX )
+    {
+        Profiler::LogString( MessageSourceType::Tracy, MessageSeverity::Error, 0, 0, "Tracy: more than 256 GPU contexts in this process; gpu context ids are 8-bit" );
+        assert( false );
+        return InvalidGpuContextId;
+    }
+    return int32_t( id );
+}
+
 TRACY_API bool ProfilerAvailable() { return s_instance != nullptr; }
 TRACY_API bool ProfilerAllocatorAvailable() { return !RpThreadShutdown; }

--- a/public/client/TracyProfiler.hpp
+++ b/public/client/TracyProfiler.hpp
@@ -95,7 +95,8 @@ TRACY_API moodycamel::ConcurrentQueue<QueueItem>::ExplicitProducer* GetToken();
 TRACY_API Profiler& MANGLED_NAME_BASED_ON_CONFIG(GetProfiler)();
 tracy_force_inline Profiler& GetProfiler() { return MANGLED_NAME_BASED_ON_CONFIG(GetProfiler)(); }
 TRACY_API std::atomic<uint32_t>& GetLockCounter();
-TRACY_API std::atomic<uint8_t>& GetGpuCtxCounter();
+TRACY_API std::atomic<uint32_t>& GetGpuCtxCounter();
+TRACY_API int32_t NextGpuContextId();
 TRACY_API GpuCtxWrapper& GetGpuCtx();
 TRACY_API uint32_t GetThreadHandle();
 TRACY_API bool ProfilerAvailable();
--- a/public/client/TracyRocprof.cpp
+++ b/public/client/TracyRocprof.cpp
@@ -79,14 +79,9 @@ uint8_t gpu_context_allocate( ToolData* data )
    float timestamp_period = 1.0f;
    data->previous_cpu_time = cpu_timestamp;

-    // Allocate the process-unique GPU context ID. There's a max of 255 available;
-    // if we are recreating devices a lot we may exceed that. Don't do that, or
-    // wrap around and get weird (but probably still usable) numbers.
-    uint8_t context_id = tracy::GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed );
-    if( context_id >= 255 )
-    {
-        context_id %= 255;
-    }
+    // Allocate the process-unique GPU context ID. There's a max of 256 available
+    // (ids are 8-bit); if we are recreating devices a lot we may exceed that.
+    uint8_t context_id = uint8_t( tracy::NextGpuContextId() );

    uint8_t context_flags = 0;
 #ifdef TRACY_ROCPROF_CALIBRATION
--- a/public/common/TracyQueue.hpp
+++ b/public/common/TracyQueue.hpp
@@ -536,6 +536,8 @@ enum GpuContextFlags : uint8_t
    GpuContextCalibration   = 1 << 0
 };

+constexpr int32_t InvalidGpuContextId = -1;
+
 struct QueueGpuNewContext
 {
    int64_t cpuTime;
--- a/public/tracy/TracyCUDA.hpp
+++ b/public/tracy/TracyCUDA.hpp
@@ -576,7 +576,7 @@ namespace tracy

            auto item = Profiler::QueueSerial();
            tracyMemWrite(item->hdr.type, QueueType::GpuContextName);
-            tracyMemWrite(item->gpuContextNameFat.context, m_tracyGpuContext);
+            tracyMemWrite(item->gpuContextNameFat.context, (uint8_t)m_tracyGpuContext);
            tracyMemWrite(item->gpuContextNameFat.ptr, (uint64_t)ptr);
            tracyMemWrite(item->gpuContextNameFat.size, len);
            SubmitQueueItem(item);
@@ -623,7 +623,7 @@ namespace tracy
                tracyMemWrite(item->gpuCalibration.gpuTime, (int64_t)tCUpti);
                tracyMemWrite(item->gpuCalibration.cpuTime, tTracy);
                tracyMemWrite(item->gpuCalibration.cpuDelta, deltaTicksCUpti);
-                tracyMemWrite(item->gpuCalibration.context, m_tracyGpuContext);
+                tracyMemWrite(item->gpuCalibration.context, (uint8_t)m_tracyGpuContext);
                Profiler::QueueSerialFinish();
            }
            #endif
@@ -644,8 +644,8 @@ namespace tracy
            //uint32_t timelineId = tracy::GetThreadHandle();
            uint32_t timelineId = tracyTimelineId(cudaContextId, cudaStreamId);
            uint16_t queryId = m_queryIdGen.fetch_add(2);
-            tracyAnnounceGpuTimestamp(apiStart, apiEnd, queryId, m_tracyGpuContext, pSrcLoc, timelineId);
-            tracySubmitGpuTimestamp(gpuStart, gpuEnd, queryId, m_tracyGpuContext);
+            tracyAnnounceGpuTimestamp(apiStart, apiEnd, queryId, (uint8_t)m_tracyGpuContext, pSrcLoc, timelineId);
+            tracySubmitGpuTimestamp(gpuStart, gpuEnd, queryId, (uint8_t)m_tracyGpuContext);
        }

        void OnEventsProcessed() {
@@ -1410,17 +1410,17 @@ namespace tracy

        };

-        CUDACtx(uint8_t gpuContextID = 255)
+        CUDACtx(int32_t gpuContextID = InvalidGpuContextId)
        {
            ZoneScoped;

-            if (gpuContextID != 255) {
+            if (gpuContextID != InvalidGpuContextId) {
                m_tracyGpuContext = gpuContextID;
                return;
            }

-            m_tracyGpuContext = GetGpuCtxCounter().fetch_add(1, std::memory_order_relaxed);
-            assert(m_tracyGpuContext != 255);
+            m_tracyGpuContext = NextGpuContextId();
+            assert(m_tracyGpuContext != InvalidGpuContextId);

            TracyTimestamp tTracy;
            CUptiTimestamp tCUpti;
@@ -1434,7 +1434,7 @@ namespace tracy
            tracyMemWrite(item->gpuNewContext.thread, (uint32_t)0);
            tracyMemWrite(item->gpuNewContext.period, 1.0f);
            tracyMemWrite(item->gpuNewContext.type, GpuContextType::CUDA);
-            tracyMemWrite(item->gpuNewContext.context, m_tracyGpuContext);
+            tracyMemWrite(item->gpuNewContext.context, (uint8_t)m_tracyGpuContext);
            #if TRACY_CUDA_CALIBRATED_CONTEXT
            tracyMemWrite(item->gpuNewContext.flags, GpuContextCalibration);
            #else
@@ -1470,7 +1470,7 @@ namespace tracy
            CUDACtx* ctx = nullptr;
            std::mutex m;
            int ref_count = 0;
-            uint8_t ctx_id = 255;
+            int32_t ctx_id = InvalidGpuContextId;
            static Singleton& Get() {
                static Singleton singleton;
                return singleton;
@@ -1481,7 +1481,7 @@ namespace tracy
        ProfilerStats stats = {};
        #endif

-        uint8_t m_tracyGpuContext = 255;
+        int32_t m_tracyGpuContext = InvalidGpuContextId;
        static constexpr size_t cacheline = 64;
        alignas(cacheline) std::atomic<uint16_t> m_queryIdGen = 0;
    };
--- a/public/tracy/TracyD3D11.hpp
+++ b/public/tracy/TracyD3D11.hpp
@@ -121,7 +121,7 @@ public:
        }

        // ready to roll
-        m_contextId = GetGpuCtxCounter().fetch_add(1);
+        m_contextId = NextGpuContextId();
        m_immediateDevCtx->Begin(m_disjointQuery);
        m_previousCheckpoint = m_nextCheckpoint = 0;

@@ -131,7 +131,7 @@ public:
        MemWrite( &item->gpuNewContext.gpuTime, tgpu );
        MemWrite( &item->gpuNewContext.thread, uint32_t(0) );   // #TODO: why not GetThreadHandle()?
        MemWrite( &item->gpuNewContext.period, 1.0f );
-        MemWrite( &item->gpuNewContext.context, m_contextId);
+        MemWrite( &item->gpuNewContext.context, uint8_t(m_contextId));
        MemWrite( &item->gpuNewContext.flags, GpuContextFlags(0) );
        MemWrite( &item->gpuNewContext.type, GpuContextType::Direct3D11 );

@@ -167,7 +167,7 @@ public:

        auto item = Profiler::QueueSerial();
        MemWrite( &item->hdr.type, QueueType::GpuContextName );
-        MemWrite( &item->gpuContextNameFat.context, m_contextId );
+        MemWrite( &item->gpuContextNameFat.context, uint8_t(m_contextId) );
        MemWrite( &item->gpuContextNameFat.ptr, (uint64_t)ptr );
        MemWrite( &item->gpuContextNameFat.size, len );
 #ifdef TRACY_ON_DEMAND
@@ -240,7 +240,7 @@ public:
            MemWrite(&item->hdr.type, QueueType::GpuTime);
            MemWrite(&item->gpuTime.gpuTime, static_cast<int64_t>(timestamp));
            MemWrite(&item->gpuTime.queryId, static_cast<uint16_t>(k));
-            MemWrite(&item->gpuTime.context, m_contextId);
+            MemWrite(&item->gpuTime.context, uint8_t(m_contextId));
            Profiler::QueueSerialFinish();
        }

@@ -287,7 +287,7 @@ private:
            YieldThread();  // busy-wait :-( attempt to reduce power usage with _mm_pause() & friends...
    }

-    tracy_force_inline uint8_t GetContextId() const
+    tracy_force_inline int32_t GetContextId() const
    {
        return m_contextId;
    }
@@ -298,7 +298,7 @@ private:
    ID3D11Query* m_queries[MaxQueries];
    ID3D11Query* m_disjointQuery = nullptr;

-    uint8_t m_contextId = 255;  // NOTE: apparently, 255 means invalid id; is this documented anywhere?
+    int32_t m_contextId = InvalidGpuContextId;

    uintptr_t m_queryCounter = 0;

@@ -380,7 +380,7 @@ public:
        MemWrite( &item->gpuZoneEnd.cpuTime, Profiler::GetTime() );
        MemWrite( &item->gpuZoneEnd.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneEnd.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneEnd.context, m_ctx->GetContextId() );
+        MemWrite( &item->gpuZoneEnd.context, uint8_t(m_ctx->GetContextId()) );
        Profiler::QueueSerialFinish();
    }

@@ -407,7 +407,7 @@ private:
        MemWrite( &item->gpuZoneBegin.srcloc, sourceLocation );
        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, m_ctx->GetContextId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t(m_ctx->GetContextId()) );
        Profiler::QueueSerialFinish();
    }

@@ -433,7 +433,6 @@ static inline void DestroyD3D11Context( D3D11Ctx* ctx )
    tracy_free( ctx );
 }
 }
-
 #undef TracyD3D11Panic

 using TracyD3D11Ctx = tracy::D3D11Ctx*;
--- a/public/tracy/TracyD3D12.hpp
+++ b/public/tracy/TracyD3D12.hpp
@@ -76,7 +76,7 @@ namespace tracy
    {
        friend class D3D12ZoneScope;

-        uint8_t m_contextId = 255;  // 255 represents "invalid id"
+        int32_t m_contextId = InvalidGpuContextId;

        std::mutex m_collectionMutex;

@@ -128,7 +128,7 @@ namespace tracy
                MemWrite(&item->gpuCalibration.gpuTime, int64_t(gpuTimestamp));
                MemWrite(&item->gpuCalibration.cpuTime, int64_t(cpuTimestamp));
                MemWrite(&item->gpuCalibration.cpuDelta, cpuDeltaNS);
-                MemWrite(&item->gpuCalibration.context, GetId());
+                MemWrite(&item->gpuCalibration.context, static_cast<uint8_t>(GetId()));
                SubmitQueueItem(item);
            }
        }
@@ -239,7 +239,7 @@ namespace tracy
            cpuTimestamp = Profiler::GetTime();

            // All setup/init checks completed: ready to create the context.
-            m_contextId = GetGpuCtxCounter().fetch_add(1);
+            m_contextId = NextGpuContextId();
            ZoneValue(m_contextId);

            auto* item = Profiler::QueueSerial();
@@ -288,7 +288,7 @@ namespace tracy
            m_queue->Release();
        }

-        tracy_force_inline uint8_t GetId() const
+        tracy_force_inline int32_t GetId() const
        {
            return m_contextId;
        }
@@ -300,7 +300,7 @@ namespace tracy

            auto item = Profiler::QueueSerial();
            MemWrite( &item->hdr.type, QueueType::GpuContextName );
-            MemWrite( &item->gpuContextNameFat.context, GetId());
+            MemWrite( &item->gpuContextNameFat.context, static_cast<uint8_t>(GetId()));
            MemWrite( &item->gpuContextNameFat.ptr, (uint64_t)ptr );
            MemWrite( &item->gpuContextNameFat.size, len );
            SubmitQueueItem(item);
@@ -458,7 +458,7 @@ namespace tracy
            MemWrite(&item->hdr.type, QueueType::GpuTime);
            MemWrite(&item->gpuTime.gpuTime, static_cast<int64_t>(gpuTimestamp));
            MemWrite(&item->gpuTime.queryId, static_cast<uint16_t>(queryId));
-            MemWrite(&item->gpuTime.context, GetId());
+            MemWrite(&item->gpuTime.context, static_cast<uint8_t>(GetId()));
            Profiler::QueueSerialFinish();
            m_shadowBuffer[queryId] = gpuTimestamp;
            TracyD3D12Debug(
@@ -608,7 +608,7 @@ namespace tracy
            MemWrite( &item->gpuZoneBegin.srcloc, srcLocationAddr );
            MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
            MemWrite( &item->gpuZoneBegin.queryId, static_cast<uint16_t>( m_queryId ) );
-            MemWrite( &item->gpuZoneBegin.context, m_ctx->GetId() );
+            MemWrite( &item->gpuZoneBegin.context, static_cast<uint8_t>(m_ctx->GetId()) );
            Profiler::QueueSerialFinish();
        }

@@ -669,7 +669,7 @@ namespace tracy
                MemWrite(&item->gpuZoneEnd.cpuTime, Profiler::GetTime());
                MemWrite(&item->gpuZoneEnd.thread, GetThreadHandle());
                MemWrite(&item->gpuZoneEnd.queryId, static_cast<uint16_t>(queryId));
-                MemWrite(&item->gpuZoneEnd.context, m_ctx->GetId());
+                MemWrite( &item->gpuZoneEnd.context, static_cast<uint8_t>(m_ctx->GetId()) );
                Profiler::QueueSerialFinish();
 #ifdef TRACY_ON_DEMAND
            }
@@ -695,12 +695,6 @@ namespace tracy
    {
        auto* ctx = static_cast<D3D12QueueCtx*>(tracy_malloc(sizeof(D3D12QueueCtx)));
        new (ctx) D3D12QueueCtx{ device, queue };
-        // constructor may have failed:
-        if (ctx->GetId() == 255)
-        {
-            DestroyD3D12Context(ctx);
-            return nullptr;
-        }
        return ctx;
    }

--- a/public/tracy/TracyMetal.hmm
+++ b/public/tracy/TracyMetal.hmm
@@ -125,11 +125,6 @@ public:
        ZoneScopedNC("tracy::MetalCtx::Create", Color::Red4);
        auto ctx = static_cast<MetalCtx*>(tracy_malloc(sizeof(MetalCtx)));
        new (ctx) MetalCtx(device);
-        if (ctx->m_contextId == 255)
-        {
-            TracyMetalPanic({assert(false);} return nullptr, "ERROR: unable to create context.");
-            Destroy(ctx);
-        }
        return ctx;
    }

@@ -147,7 +142,7 @@ public:

        auto* item = Profiler::QueueSerial();
        MemWrite( &item->hdr.type, QueueType::GpuContextName );
-        MemWrite( &item->gpuContextNameFat.context, m_contextId );
+        MemWrite( &item->gpuContextNameFat.context, static_cast<uint8_t>(m_contextId) );
        MemWrite( &item->gpuContextNameFat.ptr, (uint64_t)ptr );
        MemWrite( &item->gpuContextNameFat.size, len );
        SubmitQueueItem(item);
@@ -259,7 +254,7 @@ public:
            MemWrite(&item->hdr.type, QueueType::GpuTime);
            MemWrite(&item->gpuTime.gpuTime, static_cast<int64_t>(t_start));
            MemWrite(&item->gpuTime.queryId, static_cast<uint16_t>(k));
-            MemWrite(&item->gpuTime.context, m_contextId);
+            MemWrite(&item->gpuTime.context, static_cast<uint8_t>(m_contextId));
            Profiler::QueueSerialFinish();
            }
            {
@@ -267,7 +262,7 @@ public:
            MemWrite(&item->hdr.type, QueueType::GpuTime);
            MemWrite(&item->gpuTime.gpuTime, static_cast<int64_t>(t_end));
            MemWrite(&item->gpuTime.queryId, static_cast<uint16_t>(k+1));
-            MemWrite(&item->gpuTime.context, m_contextId);
+            MemWrite(&item->gpuTime.context, static_cast<uint8_t>(m_contextId));
            Profiler::QueueSerialFinish();
            }
            m_mostRecentTimestamp = (t_end > m_mostRecentTimestamp) ? t_end : m_mostRecentTimestamp;
@@ -342,7 +337,7 @@ private:

        float period = 1.0f;
        
-        m_contextId = GetGpuCtxCounter().fetch_add(1);
+        m_contextId = NextGpuContextId();
        
        auto* item = Profiler::QueueSerial();
        MemWrite(&item->hdr.type, QueueType::GpuNewContext);
@@ -350,7 +345,7 @@ private:
        MemWrite(&item->gpuNewContext.gpuTime, int64_t(gpuTimestamp));
        MemWrite(&item->gpuNewContext.thread, uint32_t(0)); // TODO: why not GetThreadHandle()?
        MemWrite(&item->gpuNewContext.period, period);
-        MemWrite(&item->gpuNewContext.context, m_contextId);
+        MemWrite(&item->gpuNewContext.context, static_cast<uint8_t>(m_contextId));
        //MemWrite(&item->gpuNewContext.flags, GpuContextCalibration);
        MemWrite(&item->gpuNewContext.flags, GpuContextFlags(0));
        MemWrite(&item->gpuNewContext.type, GpuContextType::Metal);
@@ -421,7 +416,7 @@ private:
        return Query{ buffer, idx };
    }

-    tracy_force_inline uint8_t GetContextId() const
+    tracy_force_inline int32_t GetContextId() const
    {
        return m_contextId;
    }
@@ -463,7 +458,7 @@ private:
        return counterSampleBuffer;
    }

-    uint8_t m_contextId = 255;
+    int32_t m_contextId = InvalidGpuContextId;

    id<MTLDevice> m_device = nil;
    id<MTLCounterSampleBuffer> m_counterSampleBuffers [2] = {};
@@ -602,7 +597,7 @@ private:
        MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, ctx->GetContextId() );
+        MemWrite( &item->gpuZoneBegin.context, static_cast<uint8_t>(ctx->GetContextId()) );
        Profiler::QueueSerialFinish();
        
        TracyMetalDebugMasked(1<<2, TracyAllocN((void*)(uintptr_t)queryId, 1, "TracyMetalGpuZone"));
@@ -615,7 +610,7 @@ private:
        MemWrite( &item->gpuZoneEnd.cpuTime, Profiler::GetTime() );
        MemWrite( &item->gpuZoneEnd.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneEnd.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneEnd.context, ctx->GetContextId() );
+        MemWrite( &item->gpuZoneEnd.context, static_cast<uint8_t>(ctx->GetContextId()) );
        Profiler::QueueSerialFinish();
        
        TracyMetalDebugMasked(1<<2, TracyAllocN((void*)(uintptr_t)queryId, 1, "TracyMetalGpuZone"));
--- a/public/tracy/TracyOpenCL.hpp
+++ b/public/tracy/TracyOpenCL.hpp
@@ -79,12 +79,12 @@ namespace tracy {
        static constexpr size_t QueryCount = 64 * 1024;

        OpenCLCtx(cl_context context, cl_device_id device)
-            : m_contextId(GetGpuCtxCounter().fetch_add(1, std::memory_order_relaxed))
+            : m_contextId(NextGpuContextId())
            , m_head(0)
            , m_tail(0)
        {
            int64_t tcpu, tgpu;
-            TRACY_CL_ASSERT(m_contextId != 255);
+            TRACY_CL_ASSERT(m_contextId != InvalidGpuContextId);

            cl_int err = CL_SUCCESS;
            cl_command_queue queue = clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &err);
@@ -200,7 +200,7 @@ namespace tracy {
            }
        }

-        tracy_force_inline uint8_t GetId() const
+        tracy_force_inline int32_t GetId() const
        {
            return m_contextId;
        }
@@ -222,7 +222,7 @@ namespace tracy {

    private:

-        unsigned int m_contextId;
+        int32_t m_contextId;

        EventInfo m_query[QueryCount];
        unsigned int m_head; // index at which a new event should be inserted
@@ -251,7 +251,7 @@ namespace tracy {
            MemWrite(&item->gpuZoneBegin.srcloc, (uint64_t)srcLoc);
            MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
            MemWrite(&item->gpuZoneBegin.queryId, (uint16_t)m_beginQueryId);
-            MemWrite(&item->gpuZoneBegin.context, ctx->GetId());
+            MemWrite(&item->gpuZoneBegin.context, uint8_t(ctx->GetId()));
            Profiler::QueueSerialFinish();
        }

@@ -276,7 +276,7 @@ namespace tracy {
            MemWrite(&item->gpuZoneBegin.srcloc, (uint64_t)srcLoc);
            MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
            MemWrite(&item->gpuZoneBegin.queryId, (uint16_t)m_beginQueryId);
-            MemWrite(&item->gpuZoneBegin.context, ctx->GetId());
+            MemWrite(&item->gpuZoneBegin.context, uint8_t(ctx->GetId()));
            Profiler::QueueSerialFinish();
        }

@@ -300,7 +300,7 @@ namespace tracy {
            MemWrite(&item->gpuZoneBegin.srcloc, srcloc);
            MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
            MemWrite(&item->gpuZoneBegin.queryId, (uint16_t)m_beginQueryId);
-            MemWrite(&item->gpuZoneBegin.context, ctx->GetId());
+            MemWrite(&item->gpuZoneBegin.context, uint8_t(ctx->GetId()));
            Profiler::QueueSerialFinish();
        }

@@ -324,7 +324,7 @@ namespace tracy {
            MemWrite(&item->gpuZoneBegin.srcloc, srcloc);
            MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
            MemWrite(&item->gpuZoneBegin.queryId, (uint16_t)m_beginQueryId);
-            MemWrite(&item->gpuZoneBegin.context, ctx->GetId());
+            MemWrite(&item->gpuZoneBegin.context, uint8_t(ctx->GetId()));
            Profiler::QueueSerialFinish();
        }

@@ -346,7 +346,7 @@ namespace tracy {
            MemWrite(&item->gpuZoneEnd.cpuTime, Profiler::GetTime());
            MemWrite(&item->gpuZoneEnd.thread, GetThreadHandle());
            MemWrite(&item->gpuZoneEnd.queryId, (uint16_t)queryId);
-            MemWrite(&item->gpuZoneEnd.context, m_ctx->GetId());
+            MemWrite(&item->gpuZoneEnd.context, uint8_t(m_ctx->GetId()));
            Profiler::QueueSerialFinish();
        }

--- a/public/tracy/TracyOpenGL.hpp
+++ b/public/tracy/TracyOpenGL.hpp
@@ -114,14 +114,14 @@ class GpuCtx

 public:
    GpuCtx()
-        : m_context( GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed ) )
+        : m_context( NextGpuContextId() )
        , m_head( 0 )
        , m_tail( 0 )
        , m_supportsQueryBufferObject( false )
    {
        ZoneScopedC( Color::Red4 );

-        assert( m_context != 255 );
+        assert( m_context != InvalidGpuContextId );

        if( !CheckFeature( "GL_ARB_timer_query" ) && !CheckFeature( "GL_EXT_disjoint_timer_query" ) )
        {
@@ -169,7 +169,7 @@ public:
        MemWrite( &item->gpuNewContext.gpuTime, tgpu );
        MemWrite( &item->gpuNewContext.thread, thread );
        MemWrite( &item->gpuNewContext.period, period );
-        MemWrite( &item->gpuNewContext.context, m_context );
+        MemWrite( &item->gpuNewContext.context, uint8_t( m_context ) );
 #ifdef TRACY_OPENGL_AUTO_CALIBRATION
        MemWrite( &item->gpuNewContext.flags, GpuContextFlags( GpuContextCalibration ) );
 #else
@@ -190,7 +190,7 @@ public:
        memcpy( ptr, name, len );

        TracyLfqPrepare( QueueType::GpuContextName );
-        MemWrite( &item->gpuContextNameFat.context, m_context );
+        MemWrite( &item->gpuContextNameFat.context, uint8_t( m_context ) );
        MemWrite( &item->gpuContextNameFat.ptr, (uint64_t)ptr );
        MemWrite( &item->gpuContextNameFat.size, len );
 #ifdef TRACY_ON_DEMAND
@@ -227,7 +227,7 @@ public:
            TracyLfqPrepare( QueueType::GpuTime );
            MemWrite( &item->gpuTime.gpuTime, (int64_t)time );
            MemWrite( &item->gpuTime.queryId, (uint16_t)m_tail );
-            MemWrite( &item->gpuTime.context, m_context );
+            MemWrite( &item->gpuTime.context, uint8_t( m_context ) );
            TracyLfqCommit;

            m_tail = ( m_tail + 1 ) % QueryCount;
@@ -311,7 +311,7 @@ private:
        MemWrite( &item->gpuCalibration.gpuTime, tgpu );
        MemWrite( &item->gpuCalibration.cpuTime, refCpu );
        MemWrite( &item->gpuCalibration.cpuDelta, delta );
-        MemWrite( &item->gpuCalibration.context, m_context );
+        MemWrite( &item->gpuCalibration.context, uint8_t( m_context ) );
        TracyLfqCommit;
    }
 #endif
@@ -329,13 +329,13 @@ private:
        return m_query[id];
    }

-    tracy_force_inline uint8_t GetId() const
+    tracy_force_inline int32_t GetId() const
    {
        return m_context;
    }

    unsigned int m_query[QueryCount];
-    uint8_t m_context;
+    int32_t m_context;

    unsigned int m_head;
    unsigned int m_tail;
@@ -367,7 +367,7 @@ public:
        MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
        memset( &item->gpuZoneBegin.thread, 0, sizeof( item->gpuZoneBegin.thread ) );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, GetGpuCtx().ptr->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( GetGpuCtx().ptr->GetId() ) );
        MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
        TracyLfqCommit;
    }
@@ -395,7 +395,7 @@ public:
 #endif
        MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, GetGpuCtx().ptr->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( GetGpuCtx().ptr->GetId() ) );
        MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
        TracyLfqCommit;
    }
@@ -418,7 +418,7 @@ public:
        MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
        memset( &item->gpuZoneBegin.thread, 0, sizeof( item->gpuZoneBegin.thread ) );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, GetGpuCtx().ptr->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( GetGpuCtx().ptr->GetId() ) );
        MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
        TracyLfqCommit;
    }
@@ -447,7 +447,7 @@ public:
        const auto srcloc = Profiler::AllocSourceLocation( line, source, sourceSz, function, functionSz, name, nameSz );
        MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, GetGpuCtx().ptr->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( GetGpuCtx().ptr->GetId() ) );
        MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
        TracyLfqCommit;
    }
@@ -466,7 +466,7 @@ public:
        MemWrite( &item->gpuZoneEnd.cpuTime, Profiler::GetTime() );
        memset( &item->gpuZoneEnd.thread, 0, sizeof( item->gpuZoneEnd.thread ) );
        MemWrite( &item->gpuZoneEnd.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneEnd.context, GetGpuCtx().ptr->GetId() );
+        MemWrite( &item->gpuZoneEnd.context, uint8_t( GetGpuCtx().ptr->GetId() ) );
        TracyLfqCommit;
    }

--- a/public/tracy/TracyVulkan.hpp
+++ b/public/tracy/TracyVulkan.hpp
@@ -105,7 +105,7 @@ public:
 #endif
        : m_device( device )
        , m_timeDomain( VK_TIME_DOMAIN_DEVICE_EXT )
-        , m_context( GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed ) )
+        , m_context( NextGpuContextId() )
        , m_head( 0 )
        , m_tail( 0 )
        , m_oldCnt( 0 )
@@ -114,7 +114,7 @@ public:
        , m_vkGetCalibratedTimestampsEXT( vkGetCalibratedTimestampsEXT )
 #endif
    {
-        assert( m_context != 255 );
+        assert( m_context != InvalidGpuContextId );

 #if defined TRACY_VK_USE_SYMBOL_TABLE
        PopulateSymbolTable(instance, instanceProcAddr, deviceProcAddr);
@@ -190,7 +190,7 @@ public:
 #endif
        : m_device( device )
        , m_timeDomain( VK_TIME_DOMAIN_DEVICE_EXT )
-        , m_context( GetGpuCtxCounter().fetch_add(1, std::memory_order_relaxed) )
+        , m_context( NextGpuContextId() )
        , m_head( 0 )
        , m_tail( 0 )
        , m_oldCnt( 0 )
@@ -199,7 +199,7 @@ public:
        , m_vkGetCalibratedTimestampsEXT( vkGetCalibratedTimestampsEXT )
 #endif
    {
-        assert( m_context != 255);
+        assert( m_context != InvalidGpuContextId);

 #if defined TRACY_VK_USE_SYMBOL_TABLE
        PopulateSymbolTable(instance, instanceProcAddr, deviceProcAddr);
@@ -242,7 +242,7 @@ public:

        auto item = Profiler::QueueSerial();
        MemWrite( &item->hdr.type, QueueType::GpuContextName );
-        MemWrite( &item->gpuContextNameFat.context, m_context );
+        MemWrite( &item->gpuContextNameFat.context, uint8_t( m_context ) );
        MemWrite( &item->gpuContextNameFat.ptr, (uint64_t)ptr );
        MemWrite( &item->gpuContextNameFat.size, len );
 #ifdef TRACY_ON_DEMAND
@@ -309,7 +309,7 @@ public:
            MemWrite( &item->hdr.type, QueueType::GpuTime );
            MemWrite( &item->gpuTime.gpuTime, m_res[idx * 2] );
            MemWrite( &item->gpuTime.queryId, uint16_t( wrappedTail + idx ) );
-            MemWrite( &item->gpuTime.context, m_context );
+            MemWrite( &item->gpuTime.context, uint8_t( m_context ) );
            Profiler::QueueSerialFinish();
        }

@@ -327,7 +327,7 @@ public:
                MemWrite( &item->gpuCalibration.gpuTime, tgpu );
                MemWrite( &item->gpuCalibration.cpuTime, refCpu );
                MemWrite( &item->gpuCalibration.cpuDelta, delta );
-                MemWrite( &item->gpuCalibration.context, m_context );
+                MemWrite( &item->gpuCalibration.context, uint8_t( m_context ) );
                Profiler::QueueSerialFinish();
            }
        }
@@ -345,7 +345,7 @@ public:
        return id % m_queryCount;
    }

-    tracy_force_inline uint8_t GetId() const
+    tracy_force_inline int32_t GetId() const
    {
        return m_context;
    }
@@ -473,7 +473,7 @@ private:
        MemWrite( &item->gpuNewContext.gpuTime, tgpu );
        memset( &item->gpuNewContext.thread, 0, sizeof( item->gpuNewContext.thread ) );
        MemWrite( &item->gpuNewContext.period, period );
-        MemWrite( &item->gpuNewContext.context, m_context );
+        MemWrite( &item->gpuNewContext.context, uint8_t( m_context ) );
        MemWrite( &item->gpuNewContext.flags, GpuContextFlags( flags ) );
        MemWrite( &item->gpuNewContext.type, GpuContextType::Vulkan );

@@ -517,7 +517,7 @@ private:
    int64_t m_qpcToNs;
 #endif
    int64_t m_prevCalibration;
-    uint8_t m_context;
+    int32_t m_context;

    std::atomic<uint64_t> m_head;
    uint64_t m_tail;
@@ -553,7 +553,7 @@ public:
        MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, ctx->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( ctx->GetId() ) );
        Profiler::QueueSerialFinish();
    }

@@ -587,7 +587,7 @@ public:
        MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, ctx->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( ctx->GetId() ) );
        Profiler::QueueSerialFinish();
    }

@@ -613,7 +613,7 @@ public:
        MemWrite( &item->gpuZoneBegin.srcloc, srcloc );
        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, ctx->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( ctx->GetId() ) );
        Profiler::QueueSerialFinish();
    }

@@ -648,7 +648,7 @@ public:
        MemWrite( &item->gpuZoneBegin.srcloc, srcloc );
        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, ctx->GetId() );
+        MemWrite( &item->gpuZoneBegin.context, uint8_t( ctx->GetId() ) );
        Profiler::QueueSerialFinish();
    }

@@ -667,7 +667,7 @@ public:
        MemWrite( &item->gpuZoneEnd.cpuTime, Profiler::GetTime() );
        MemWrite( &item->gpuZoneEnd.thread, GetThreadHandle() );
        MemWrite( &item->gpuZoneEnd.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneEnd.context, m_ctx->GetId() );
+        MemWrite( &item->gpuZoneEnd.context, uint8_t( m_ctx->GetId() ) );
        Profiler::QueueSerialFinish();
    }

--- a/public/tracy/TracyWebGPU.hpp
+++ b/public/tracy/TracyWebGPU.hpp
@@ -96,7 +96,7 @@ namespace tracy
    {
        friend class WebGPUZoneScope;

-        uint8_t m_contextId = 255;  // 255 represents "invalid id"
+        int32_t m_contextId = InvalidGpuContextId;

        std::mutex m_collectionMutex;

@@ -528,7 +528,7 @@ namespace tracy
            m_shadowBuffer.resize(m_queryLimit, gpuTimestamp);

            // All setup completed: register the context.
-            m_contextId = GetGpuCtxCounter().fetch_add(1);
+            m_contextId = NextGpuContextId();
            ZoneValue(m_contextId);

            auto* item = Profiler::QueueSerial();
@@ -560,7 +560,7 @@ namespace tracy
            if (m_instance)       { wgpuInstanceRelease(m_instance);     m_instance       = nullptr; }
        }

-        tracy_force_inline uint8_t GetId() const
+        tracy_force_inline int32_t GetId() const
        {
            return m_contextId;
        }
@@ -572,7 +572,7 @@ namespace tracy

            auto item = Profiler::QueueSerial();
            MemWrite(&item->hdr.type, QueueType::GpuContextName);
-            MemWrite(&item->gpuContextNameFat.context, GetId());
+            MemWrite(&item->gpuContextNameFat.context, static_cast<uint8_t>(GetId()));
            MemWrite(&item->gpuContextNameFat.ptr, (uint64_t)ptr);
            MemWrite(&item->gpuContextNameFat.size, len);
            SubmitQueueItem(item);
@@ -682,7 +682,7 @@ namespace tracy
            MemWrite(&item->hdr.type, QueueType::GpuTime);
            MemWrite(&item->gpuTime.gpuTime, static_cast<int64_t>(gpuTimestamp));
            MemWrite(&item->gpuTime.queryId, static_cast<uint16_t>(queryId));
-            MemWrite(&item->gpuTime.context, GetId());
+            MemWrite(&item->gpuTime.context, static_cast<uint8_t>(GetId()));
            Profiler::QueueSerialFinish();
            m_shadowBuffer[queryId] = gpuTimestamp;
        }
@@ -808,7 +808,7 @@ namespace tracy
            MemWrite(&item->gpuZoneBegin.srcloc, srcLocationAddr);
            MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
            MemWrite(&item->gpuZoneBegin.queryId, static_cast<uint16_t>(m_queryId));
-            MemWrite(&item->gpuZoneBegin.context, m_ctx->GetId());
+            MemWrite(&item->gpuZoneBegin.context, static_cast<uint8_t>(m_ctx->GetId()));
            Profiler::QueueSerialFinish();
        }

@@ -901,7 +901,7 @@ namespace tracy
                MemWrite(&item->gpuZoneEnd.cpuTime, Profiler::GetTime());
                MemWrite(&item->gpuZoneEnd.thread, GetThreadHandle());
                MemWrite(&item->gpuZoneEnd.queryId, static_cast<uint16_t>(queryId));
-                MemWrite(&item->gpuZoneEnd.context, m_ctx->GetId());
+                MemWrite(&item->gpuZoneEnd.context, static_cast<uint8_t>(m_ctx->GetId()));
                Profiler::QueueSerialFinish();
 #ifdef TRACY_ON_DEMAND
            }
@@ -923,11 +923,6 @@ namespace tracy
    {
        auto* ctx = static_cast<WebGPUQueueCtx*>(tracy_malloc(sizeof(WebGPUQueueCtx)));
        new (ctx) WebGPUQueueCtx{ instance, device, queue };
-        if (ctx->GetId() == 255)
-        {
-            DestroyWebGPUContext(ctx);
-            return nullptr;
-        }
        return ctx;
    }