Scarfski/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2026-06-08 08:33:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update stale MEMORY2 comment to reflect cbid tracker removal

Browse Source 
The comment still described consuming/not-consuming cudaCallSiteInfo
entries, but memory CBIDs are no longer tracked so no entry exists.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Basil Milanich
2026-04-13 10:28:13 -05:00
parent 58809f95ff
commit 0a0566abc5
1 changed files with 4 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
public/tracy/TracyCUDA.hpp
View File

@@ -1181,13 +1181,10 @@ namespace tracy
{
ZoneNamedN(kernel, "tracy::CUDACtx::DoProcessDeviceEvent[malloc/free]", instrument);
CUpti_ActivityMemory3* memory3 = (CUpti_ActivityMemory3*)record;
// Do NOT call matchActivityToAPICall here. The handler only needs
// address, size, and timestamp — apiCall is never used. More importantly,
// consuming the cudaCallSiteInfo entry would break kernel/memcpy zone
// correlation for graphs that mix alloc nodes with other node types: all
// activities in one graph launch share the same correlationId, so consuming
// it here would prevent matchGraphActivityToAPICall from finding it for the
// subsequent kernel or memcpy records from the same launch.
// No API call correlation needed — this handler only uses address,
// size, and timestamp directly from the activity record. Memory
// API CBIDs are intentionally excluded from cbidRuntimeTrackers /
// cbidDriverTrackers so no cudaCallSiteInfo entry is created.
static constexpr const char* graph_name = "CUDA Memory Allocation";
if (memory3->memoryOperationType == CUPTI_ACTIVITY_MEMORY_OPERATION_TYPE_ALLOCATION){
auto& memAllocAddress = PersistentState::Get().memAllocAddress;