Skip to content

PR #28782: [XLA:GPU] Annotate cuBLAS/cuDNN outputs to avoid initcheck failures #3361

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

PR #28782: [XLA:GPU] Annotate cuBLAS/cuDNN outputs to avoid initcheck failures #3361

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 18 additions & 0 deletions tsl/profiler/lib/nvtx_utils.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,7 @@ limitations under the License.
#include "nvtx3/nvToolsExt.h"
#include "nvtx3/nvToolsExtCuda.h"
#include "nvtx3/nvToolsExtCudaRt.h"
#include "nvtx3/nvToolsExtMemCudaRt.h"
#include "nvtx3/nvToolsExtPayload.h"
#include "third_party/gpus/cuda/include/cuda.h"

Expand Down Expand Up @@ -119,4 +120,21 @@ StringHandle RegisterString(ProfilerDomainHandle domain,
buffer.append(suffix);
return impl(buffer.c_str());
}

void MarkMemoryInitialized(void const* address, size_t size,
StreamHandle stream) {
auto domain = DefaultProfilerDomain();
nvtxMemVirtualRangeDesc_t range_desc{size, address};
nvtxMemMarkInitializedBatch_t regions_desc{
NVTX_EXT_COMPATID_MEM,
sizeof(nvtxMemMarkInitializedBatch_t),
NVTX_MEM_TYPE_VIRTUAL_ADDRESS,
/*regionDescCount=*/1,
sizeof(nvtxMemVirtualRangeDesc_t),
&range_desc};
nvtxMemCudaMarkInitialized(reinterpret_cast<nvtxDomainHandle_t>(domain),
reinterpret_cast<cudaStream_t>(stream),
/*isPerThreadStream=*/false, &regions_desc);
}

} // namespace tsl::profiler
5 changes: 5 additions & 0 deletions tsl/profiler/lib/nvtx_utils.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,5 +78,10 @@ void RangePush(ProfilerDomainHandle domain, StringHandle title,
// Register the schema of a custom payload type, for use with the more powerful
// version of RangePush
uint64_t RegisterSchema(ProfilerDomainHandle domain, const void* schemaAttr);

// Mark a memory region as initialized.
// This mitigates false positives from the compute sanitizer (initcheck).
void MarkMemoryInitialized(void const* address, size_t size,
StreamHandle stream);
} // namespace tsl::profiler
#endif // TENSORFLOW_TSL_PROFILER_LIB_NVTX_UTILS_H_
2 changes: 2 additions & 0 deletions tsl/profiler/lib/nvtx_utils_stub.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,4 +31,6 @@ uint64_t RegisterSchema(ProfilerDomainHandle, const void*) { return 0; }
StringHandle RegisterString(ProfilerDomainHandle, const std::string&) {
return {};
}
void MarkMemoryInitialized(void const* address, size_t size,
StreamHandle stream) {}
} // namespace tsl::profiler