diff options
author | 2016-07-01 11:52:36 -0800 | |
---|---|---|
committer | 2016-07-01 13:11:19 -0700 | |
commit | f2a69f30fb159f4dd7d6550562e6e0c2e1f05c3a (patch) | |
tree | 45541cd6481cc925f1f7b90944201c32b1a66970 | |
parent | fa70c501de3117059e98f1718f2ad641fd2185f5 (diff) |
Handle cudaMemcpyAsync_v3020 API callback in GpuTracer.
Reduce verbosity of logging.
Change: 126437651
-rw-r--r-- | tensorflow/core/common_runtime/gpu/gpu_tracer.cc | 23 |
1 files changed, 14 insertions, 9 deletions
diff --git a/tensorflow/core/common_runtime/gpu/gpu_tracer.cc b/tensorflow/core/common_runtime/gpu/gpu_tracer.cc index 585285258f..82d8b7129a 100644 --- a/tensorflow/core/common_runtime/gpu/gpu_tracer.cc +++ b/tensorflow/core/common_runtime/gpu/gpu_tracer.cc @@ -479,20 +479,25 @@ void GPUTracerImpl::AddCorrelationId(uint32 correlation_id, if (cbInfo->callbackSite == CUPTI_API_ENTER) { auto *params = reinterpret_cast<const cuLaunchKernel_params *>( cbInfo->functionParams); - VLOG(2) << "LAUNCH stream " << params->hStream << " correllation " - << cbInfo->correlationId << " kernel " << cbInfo->symbolName; + if (VLOG_IS_ON(2)) { + VLOG(2) << "LAUNCH stream " << params->hStream << " correllation " + << cbInfo->correlationId << " kernel " << cbInfo->symbolName; + } const string annotation = tls_annotation ? tls_annotation : cbInfo->symbolName; tracer->AddCorrelationId(cbInfo->correlationId, annotation); } } else if ((domain == CUPTI_CB_DOMAIN_RUNTIME_API) && - (cbid == CUPTI_RUNTIME_TRACE_CBID_cudaMemcpy_v3020)) { + (cbid == CUPTI_RUNTIME_TRACE_CBID_cudaMemcpy_v3020 || + cbid == CUPTI_RUNTIME_TRACE_CBID_cudaMemcpyAsync_v3020)) { if (cbInfo->callbackSite == CUPTI_API_ENTER) { - auto *funcParams = reinterpret_cast<const cudaMemcpy_v3020_params *>( - cbInfo->functionParams); - size_t count = funcParams->count; - enum cudaMemcpyKind kind = funcParams->kind; - VLOG(2) << "MEMCPY count " << count << " kind " << kind; + if (VLOG_IS_ON(2)) { + auto *funcParams = reinterpret_cast<const cudaMemcpy_v3020_params *>( + cbInfo->functionParams); + size_t count = funcParams->count; + enum cudaMemcpyKind kind = funcParams->kind; + VLOG(2) << "MEMCPY count " << count << " kind " << kind; + } if (tls_annotation) { const string annotation = tls_annotation; tracer->AddCorrelationId(cbInfo->correlationId, annotation); @@ -510,7 +515,7 @@ void GPUTracerImpl::AddCorrelationId(uint32 correlation_id, tracer->AddCorrelationId(cbInfo->correlationId, annotation); } } else { - LOG(WARNING) << "Unhandled API Callback for " << domain << " " << cbid; + VLOG(1) << "Unhandled API Callback for " << domain << " " << cbid; } } |