Scarfski/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2026-10-18 06:45:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

Review fixes: correct stale comments, fix fetch() missing lock, tidy helper

Browse Source 
- Fix wrong comments on graph launch tracker entries: they claimed
  correlation works "via CUPTI_ACTIVITY_KIND_GRAPH_TRACE", but that
  approach was rejected (GRAPH_TRACE suppresses per-kernel records).
  The actual mechanism is the shared correlationId across all nodes
  in one graph launch.
- Fix ConcurrentHashMap::fetch() missing its read lock — a pre-existing
  data race now exercised by the new graph correlation hot path.
- Cache PersistentState::Get().cudaGraphCurrentLaunch in a local ref
  inside matchGraphActivityToAPICall instead of calling Get() twice.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Basil Milanich
2026-04-06 14:00:01 -05:00
parent 0b6e27934d
commit c8ebc6f21e
1 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
public/tracy/TracyCUDA.hpp
View File

@@ -337,6 +337,7 @@ struct ConcurrentHashMap {
}
auto fetch(TKey key, TValue& value) {
ZoneNamed(fetch, instrument);
auto lock = acquire_read_lock();
auto it = mapping.find(key);
if (it != mapping.end()) {
value = it->second;
@@ -782,8 +783,8 @@ namespace tracy
{ CUPTI_RUNTIME_TRACE_CBID_cudaEventQuery_v3020, NON_STREAM_FUNC() },
{ CUPTI_RUNTIME_TRACE_CBID_cudaStreamWaitEvent_v3020, NON_STREAM_FUNC() },
{ CUPTI_RUNTIME_TRACE_CBID_cudaDeviceSynchronize_v3020, NON_STREAM_FUNC() },
// Graph launch: tracked so we can correlate GPU activities back to
// the cudaGraphLaunch call site via CUPTI_ACTIVITY_KIND_GRAPH_TRACE
// Graph launch: tracked so all CONCURRENT_KERNEL/MEMCPY/MEMSET activities
// sharing the launch's correlationId can be correlated back to this call site.
{ CUPTI_RUNTIME_TRACE_CBID_cudaGraphLaunch_v10000, GET_STREAM_FUNC(cudaGraphLaunch_v10000_params, stream) },
{ CUPTI_RUNTIME_TRACE_CBID_cudaGraphLaunch_ptsz_v10000, GET_STREAM_FUNC(cudaGraphLaunch_v10000_params, stream) },
};
@@ -842,8 +843,8 @@ namespace tracy
{ CUPTI_DRIVER_TRACE_CBID_cuEventSynchronize, NON_STREAM_FUNC() },
{ CUPTI_DRIVER_TRACE_CBID_cuCtxSynchronize, NON_STREAM_FUNC() },
{ CUPTI_DRIVER_TRACE_CBID_cuStreamWaitEvent, GET_STREAM_FUNC(cuStreamWaitEvent_params, hStream) },
// Graph launch: tracked so we can correlate GPU activities back to
// the cuGraphLaunch call site via CUPTI_ACTIVITY_KIND_GRAPH_TRACE
// Graph launch: tracked so all CONCURRENT_KERNEL/MEMCPY/MEMSET activities
// sharing the launch's correlationId can be correlated back to this call site.
{ CUPTI_DRIVER_TRACE_CBID_cuGraphLaunch, GET_STREAM_FUNC(cuGraphLaunch_params, hStream) },
{ CUPTI_DRIVER_TRACE_CBID_cuGraphLaunch_ptsz, GET_STREAM_FUNC(cuGraphLaunch_params, hStream) },
};
@@ -916,13 +917,14 @@ namespace tracy
// via either path.
static bool matchGraphActivityToAPICall(uint32_t correlationId, uint32_t graphId,
APICallInfo& apiCallInfo, const char* kind) {
auto& graphLaunchCache = PersistentState::Get().cudaGraphCurrentLaunch;
if (!matchActivityToAPICall(correlationId, apiCallInfo)) {
if (graphId == 0 || !PersistentState::Get().cudaGraphCurrentLaunch.fetch(graphId, apiCallInfo)) {
if (graphId == 0 || !graphLaunchCache.fetch(graphId, apiCallInfo)) {
matchError(correlationId, kind);
return false;
}
} else if (graphId != 0) {
PersistentState::Get().cudaGraphCurrentLaunch.insert_or_assign(graphId, apiCallInfo);
graphLaunchCache.insert_or_assign(graphId, apiCallInfo);
}
return true;
}