saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-06-29 03:30:52 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add file and line info in CudaCall and RocmCall macros (#15148)

Browse source 
This PR add file and line information so that it is easy to trouble shoot the issue of cuda error.
Update Rocm call as well for hipify.
This commit is contained in:
Tianlei Wu 2023-03-27 11:04:19 -07:00 committed by GitHub
parent ffcfb1ec98
commit 2e56620611
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
13 changed files with 108 additions and 111 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
onnxruntime/core/providers/cuda/cuda_call.cc
View file

@ -87,7 +87,7 @@ const char* CudaErrString<ncclResult_t>(ncclResult_t e) {
template <typename ERRTYPE, bool THRW>
std::conditional_t<THRW, void, Status> CudaCall(
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg) {
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg, const char* file, const int line) {
if (retCode != successCode) {
try {
#ifdef _WIN32
@ -105,10 +105,10 @@ std::conditional_t<THRW, void, Status> CudaCall(
cudaGetDevice(&currentCudaDevice);
cudaGetLastError(); // clear last CUDA error
static char str[1024];
snprintf(str, 1024, "%s failure %d: %s ; GPU=%d ; hostname=%s ; expr=%s; %s",
snprintf(str, 1024, "%s failure %d: %s ; GPU=%d ; hostname=%s ; file=%s ; line=%d ; expr=%s; %s",
libName, (int)retCode, CudaErrString(retCode), currentCudaDevice,
hostname,
exprString, msg);
file, line, exprString, msg);
if constexpr (THRW) {
// throw an exception with the error info
ORT_THROW(str);
@ -130,19 +130,19 @@ std::conditional_t<THRW, void, Status> CudaCall(
}
}
template Status CudaCall<cudaError, false>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg);
template void CudaCall<cudaError, true>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg);
template Status CudaCall<cublasStatus_t, false>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg);
template void CudaCall<cublasStatus_t, true>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg);
template Status CudaCall<cudnnStatus_t, false>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg);
template void CudaCall<cudnnStatus_t, true>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg);
template Status CudaCall<curandStatus_t, false>(curandStatus_t retCode, const char* exprString, const char* libName, curandStatus_t successCode, const char* msg);
template void CudaCall<curandStatus_t, true>(curandStatus_t retCode, const char* exprString, const char* libName, curandStatus_t successCode, const char* msg);
template Status CudaCall<cufftResult, false>(cufftResult retCode, const char* exprString, const char* libName, cufftResult successCode, const char* msg);
template void CudaCall<cufftResult, true>(cufftResult retCode, const char* exprString, const char* libName, cufftResult successCode, const char* msg);
template Status CudaCall<cudaError, false>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg, const char* file, const int line);
template void CudaCall<cudaError, true>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg, const char* file, const int line);
template Status CudaCall<cublasStatus_t, false>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg, const char* file, const int line);
template void CudaCall<cublasStatus_t, true>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg, const char* file, const int line);
template Status CudaCall<cudnnStatus_t, false>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg, const char* file, const int line);
template void CudaCall<cudnnStatus_t, true>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg, const char* file, const int line);
template Status CudaCall<curandStatus_t, false>(curandStatus_t retCode, const char* exprString, const char* libName, curandStatus_t successCode, const char* msg, const char* file, const int line);
template void CudaCall<curandStatus_t, true>(curandStatus_t retCode, const char* exprString, const char* libName, curandStatus_t successCode, const char* msg, const char* file, const int line);
template Status CudaCall<cufftResult, false>(cufftResult retCode, const char* exprString, const char* libName, cufftResult successCode, const char* msg, const char* file, const int line);
template void CudaCall<cufftResult, true>(cufftResult retCode, const char* exprString, const char* libName, cufftResult successCode, const char* msg, const char* file, const int line);
#ifdef ORT_USE_NCCL
template Status CudaCall<ncclResult_t, false>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg);
template void CudaCall<ncclResult_t, true>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg);
template Status CudaCall<ncclResult_t, false>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg, const char* file, const int line);
template void CudaCall<ncclResult_t, true>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg, const char* file, const int line);
#endif
} // namespace onnxruntime

4
onnxruntime/core/providers/cuda/cuda_provider_factory.cc
View file

@ -109,8 +109,8 @@ struct ProviderInfo_CUDA_Impl : ProviderInfo_CUDA {
return cuda::Impl_Cast(static_cast<cudaStream_t>(stream), input_data, output_data, count);
}
Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { return CudaCall<cudaError, false>(cudaError(retCode), exprString, libName, cudaError(successCode), msg); }
void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { CudaCall<cudaError, true>(cudaError(retCode), exprString, libName, cudaError(successCode), msg); }
Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { return CudaCall<cudaError, false>(cudaError(retCode), exprString, libName, cudaError(successCode), msg, file, line); }
void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { CudaCall<cudaError, true>(cudaError(retCode), exprString, libName, cudaError(successCode), msg, file, line); }
void CopyGpuToCpu(void* dst_ptr, const void* src_ptr, const size_t size, const OrtMemoryInfo& dst_location, const OrtMemoryInfo& src_location) override {
ORT_ENFORCE(dst_location.device.Type() == OrtDevice::CPU);

7
onnxruntime/core/providers/cuda/cuda_provider_factory.h
View file

@ -20,7 +20,6 @@ class NvtxRangeCreator;
}
struct ProviderInfo_CUDA {
virtual OrtStatus* SetCurrentGpuDeviceId(_In_ int device_id) = 0;
virtual OrtStatus* GetCurrentGpuDeviceId(_In_ int* device_id) = 0;
@ -33,8 +32,8 @@ struct ProviderInfo_CUDA {
virtual void cuda__Impl_Cast(void* stream, const double* input_data, float* output_data, size_t count) = 0;
virtual void cuda__Impl_Cast(void* stream, const float* input_data, double* output_data, size_t count) = 0;
virtual Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
virtual void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
virtual void CopyGpuToCpu(void* dst_ptr, const void* src_ptr, const size_t size, const OrtMemoryInfo& dst_location, const OrtMemoryInfo& src_location) = 0;
virtual void cudaMemcpy_HostToDevice(void* dst, const void* src, size_t count) = 0;
@ -60,7 +59,7 @@ struct ProviderInfo_CUDA {
virtual bool TestAll() = 0;
#endif
protected:
protected:
~ProviderInfo_CUDA() = default; // Can only be destroyed through a subclass instance
};

32
onnxruntime/core/providers/cuda/shared_inc/cuda_call.h
View file

@ -13,32 +13,32 @@ namespace onnxruntime {
template <typename ERRTYPE, bool THRW>
std::conditional_t<THRW, void, Status> CudaCall(
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg = "");
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg, const char* file, const int line);
#define CUDA_CALL(expr) (CudaCall<cudaError, false>((expr), #expr, "CUDA", cudaSuccess))
#define CUBLAS_CALL(expr) (CudaCall<cublasStatus_t, false>((expr), #expr, "CUBLAS", CUBLAS_STATUS_SUCCESS))
#define CUDA_CALL(expr) (CudaCall<cudaError, false>((expr), #expr, "CUDA", cudaSuccess, "", __FILE__, __LINE__))
#define CUBLAS_CALL(expr) (CudaCall<cublasStatus_t, false>((expr), #expr, "CUBLAS", CUBLAS_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define CUSPARSE_CALL(expr) (CudaCall<cusparseStatus_t, false>((expr), #expr, "CUSPARSE", CUSPARSE_STATUS_SUCCESS))
#define CURAND_CALL(expr) (CudaCall<curandStatus_t, false>((expr), #expr, "CURAND", CURAND_STATUS_SUCCESS))
#define CUDNN_CALL(expr) (CudaCall<cudnnStatus_t, false>((expr), #expr, "CUDNN", CUDNN_STATUS_SUCCESS))
#define CUDNN_CALL2(expr, m) (CudaCall<cudnnStatus_t, false>((expr), #expr, "CUDNN", CUDNN_STATUS_SUCCESS, m))
#define CUSPARSE_CALL(expr) (CudaCall<cusparseStatus_t, false>((expr), #expr, "CUSPARSE", CUSPARSE_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define CURAND_CALL(expr) (CudaCall<curandStatus_t, false>((expr), #expr, "CURAND", CURAND_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define CUDNN_CALL(expr) (CudaCall<cudnnStatus_t, false>((expr), #expr, "CUDNN", CUDNN_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define CUDNN_CALL2(expr, m) (CudaCall<cudnnStatus_t, false>((expr), #expr, "CUDNN", CUDNN_STATUS_SUCCESS, m, __FILE__, __LINE__))
#define CUFFT_CALL(expr) (CudaCall<cufftResult, false>((expr), #expr, "CUFFT", CUFFT_SUCCESS))
#define CUFFT_CALL(expr) (CudaCall<cufftResult, false>((expr), #expr, "CUFFT", CUFFT_SUCCESS, "", __FILE__, __LINE__))
#define CUDA_CALL_THROW(expr) (CudaCall<cudaError, true>((expr), #expr, "CUDA", cudaSuccess))
#define CUBLAS_CALL_THROW(expr) (CudaCall<cublasStatus_t, true>((expr), #expr, "CUBLAS", CUBLAS_STATUS_SUCCESS))
#define CUDA_CALL_THROW(expr) (CudaCall<cudaError, true>((expr), #expr, "CUDA", cudaSuccess, "", __FILE__, __LINE__))
#define CUBLAS_CALL_THROW(expr) (CudaCall<cublasStatus_t, true>((expr), #expr, "CUBLAS", CUBLAS_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define CUSPARSE_CALL_THROW(expr) (CudaCall<cusparseStatus_t, true>((expr), #expr, "CUSPARSE", CUSPARSE_STATUS_SUCCESS))
#define CURAND_CALL_THROW(expr) (CudaCall<curandStatus_t, true>((expr), #expr, "CURAND", CURAND_STATUS_SUCCESS))
#define CUSPARSE_CALL_THROW(expr) (CudaCall<cusparseStatus_t, true>((expr), #expr, "CUSPARSE", CUSPARSE_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define CURAND_CALL_THROW(expr) (CudaCall<curandStatus_t, true>((expr), #expr, "CURAND", CURAND_STATUS_SUCCESS, "", __FILE__, __LINE__))
// the cudnn configuration call that doesn't need set stream
#define CUDNN_CALL_THROW(expr) (CudaCall<cudnnStatus_t, true>((expr), #expr, "CUDNN", CUDNN_STATUS_SUCCESS))
#define CUDNN_CALL_THROW(expr) (CudaCall<cudnnStatus_t, true>((expr), #expr, "CUDNN", CUDNN_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define CUFFT_CALL_THROW(expr) (CudaCall<cufftResult, true>((expr), #expr, "CUFFT", CUFFT_SUCCESS))
#define CUFFT_CALL_THROW(expr) (CudaCall<cufftResult, true>((expr), #expr, "CUFFT", CUFFT_SUCCESS, "", __FILE__, __LINE__))
#ifdef ORT_USE_NCCL
#define NCCL_CALL(expr) (CudaCall<ncclResult_t, false>((expr), #expr, "NCCL", ncclSuccess))
#define NCCL_CALL_THROW(expr) (CudaCall<ncclResult_t, true>((expr), #expr, "NCCL", ncclSuccess))
#define NCCL_CALL(expr) (CudaCall<ncclResult_t, false>((expr), #expr, "NCCL", ncclSuccess, "", __FILE__, __LINE__))
#define NCCL_CALL_THROW(expr) (CudaCall<ncclResult_t, true>((expr), #expr, "NCCL", ncclSuccess, "", __FILE__, __LINE__))
#endif
} // namespace onnxruntime

18
onnxruntime/core/providers/migraphx/migraphx_call.cc
View file

@ -31,7 +31,7 @@ const char* RocmErrString<hipError_t>(hipError_t x) {
template <typename ERRTYPE, bool THRW>
std::conditional_t<THRW, void, Status> RocmCall(
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg) {
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg, const char* file, const int line) {
if (retCode != successCode) {
try {
char hostname[HOST_NAME_MAX];
@ -41,10 +41,10 @@ std::conditional_t<THRW, void, Status> RocmCall(
(void)hipGetDevice(&currentHipDevice);
(void)hipGetLastError(); // clear last HIP error
static char str[1024];
snprintf(str, 1024, "%s failure %d: %s ; GPU=%d ; hostname=%s ; expr=%s; %s",
snprintf(str, 1024, "%s failure %d: %s ; GPU=%d ; hostname=%s ; file=%s ; line=%d ; expr=%s; %s",
libName, (int)retCode, RocmErrString(retCode), currentHipDevice,
hostname,
exprString, msg);
file, line, exprString, msg);
if constexpr (THRW) {
// throw an exception with the error info
ORT_THROW(str);
@ -66,11 +66,11 @@ std::conditional_t<THRW, void, Status> RocmCall(
}
}
template Status RocmCall<hipError_t, false>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg);
template void RocmCall<hipError_t, true>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg);
template Status RocmCall<rocblas_status, false>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg);
template void RocmCall<rocblas_status, true>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg);
template Status RocmCall<miopenStatus_t, false>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg);
template void RocmCall<miopenStatus_t, true>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg);
template Status RocmCall<hipError_t, false>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg, const char* file, const int line);
template void RocmCall<hipError_t, true>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg, const char* file, const int line);
template Status RocmCall<rocblas_status, false>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg, const char* file, const int line);
template void RocmCall<rocblas_status, true>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg, const char* file, const int line);
template Status RocmCall<miopenStatus_t, false>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg, const char* file, const int line);
template void RocmCall<miopenStatus_t, true>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg, const char* file, const int line);
} // namespace onnxruntime

8
onnxruntime/core/providers/migraphx/migraphx_call.h
View file

@ -14,9 +14,9 @@ namespace onnxruntime {
template <typename ERRTYPE, bool THRW>
std::conditional_t<THRW, void, Status> RocmCall(
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg = "");
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg, const char* file, const int line);
#define HIP_CALL(expr) (RocmCall<hipError_t, false>((expr), #expr, "HIP", hipSuccess))
#define HIP_CALL_THROW(expr) (RocmCall<hipError_t, true>((expr), #expr, "HIP", hipSuccess))
#define HIP_CALL(expr) (RocmCall<hipError_t, false>((expr), #expr, "HIP", hipSuccess, "", __FILE__, __LINE__))
#define HIP_CALL_THROW(expr) (RocmCall<hipError_t, true>((expr), #expr, "HIP", hipSuccess, "", __FILE__, __LINE__))
}
} // namespace onnxruntime

30
onnxruntime/core/providers/rocm/rocm_call.cc
View file

@ -90,7 +90,7 @@ const char* RocmErrString<ncclResult_t>(ncclResult_t e) {
template <typename ERRTYPE, bool THRW>
std::conditional_t<THRW, void, Status> RocmCall(
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg) {
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg, const char* file, const int line) {
if (retCode != successCode) {
try {
#ifdef _WIN32
@ -108,10 +108,10 @@ std::conditional_t<THRW, void, Status> RocmCall(
ORT_IGNORE_RETURN_VALUE(hipGetDevice(&currentHipDevice)); // void to silence nodiscard
ORT_IGNORE_RETURN_VALUE(hipGetLastError()); // clear last ROCM error; void to silence nodiscard
static char str[1024];
snprintf(str, 1024, "%s failure %d: %s ; GPU=%d ; hostname=%s ; expr=%s; %s",
snprintf(str, 1024, "%s failure %d: %s ; GPU=%d ; hostname=%s ; file=%s ; line=%d ; expr=%s; %s",
libName, (int)retCode, RocmErrString(retCode), currentHipDevice,
hostname,
exprString, msg);
file, line, exprString, msg);
if constexpr (THRW) {
// throw an exception with the error info
ORT_THROW(str);
@ -133,19 +133,19 @@ std::conditional_t<THRW, void, Status> RocmCall(
}
}
template Status RocmCall<hipError_t, false>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg);
template void RocmCall<hipError_t, true>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg);
template Status RocmCall<rocblas_status, false>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg);
template void RocmCall<rocblas_status, true>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg);
template Status RocmCall<miopenStatus_t, false>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg);
template void RocmCall<miopenStatus_t, true>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg);
template Status RocmCall<hiprandStatus_t, false>(hiprandStatus_t retCode, const char* exprString, const char* libName, hiprandStatus_t successCode, const char* msg);
template void RocmCall<hiprandStatus_t, true>(hiprandStatus_t retCode, const char* exprString, const char* libName, hiprandStatus_t successCode, const char* msg);
template Status RocmCall<hipfftResult, false>(hipfftResult retCode, const char* exprString, const char* libName, hipfftResult successCode, const char* msg);
template void RocmCall<hipfftResult, true>(hipfftResult retCode, const char* exprString, const char* libName, hipfftResult successCode, const char* msg);
template Status RocmCall<hipError_t, false>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg, const char* file, const int line);
template void RocmCall<hipError_t, true>(hipError_t retCode, const char* exprString, const char* libName, hipError_t successCode, const char* msg, const char* file, const int line);
template Status RocmCall<rocblas_status, false>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg, const char* file, const int line);
template void RocmCall<rocblas_status, true>(rocblas_status retCode, const char* exprString, const char* libName, rocblas_status successCode, const char* msg, const char* file, const int line);
template Status RocmCall<miopenStatus_t, false>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg, const char* file, const int line);
template void RocmCall<miopenStatus_t, true>(miopenStatus_t retCode, const char* exprString, const char* libName, miopenStatus_t successCode, const char* msg, const char* file, const int line);
template Status RocmCall<hiprandStatus_t, false>(hiprandStatus_t retCode, const char* exprString, const char* libName, hiprandStatus_t successCode, const char* msg, const char* file, const int line);
template void RocmCall<hiprandStatus_t, true>(hiprandStatus_t retCode, const char* exprString, const char* libName, hiprandStatus_t successCode, const char* msg, const char* file, const int line);
template Status RocmCall<hipfftResult, false>(hipfftResult retCode, const char* exprString, const char* libName, hipfftResult successCode, const char* msg, const char* file, const int line);
template void RocmCall<hipfftResult, true>(hipfftResult retCode, const char* exprString, const char* libName, hipfftResult successCode, const char* msg, const char* file, const int line);
#ifdef ORT_USE_NCCL
template Status RocmCall<ncclResult_t, false>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg);
template void RocmCall<ncclResult_t, true>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg);
template Status RocmCall<ncclResult_t, false>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg, const char* file, const int line);
template void RocmCall<ncclResult_t, true>(ncclResult_t retCode, const char* exprString, const char* libName, ncclResult_t successCode, const char* msg, const char* file, const int line);
#endif
} // namespace onnxruntime

4
onnxruntime/core/providers/rocm/rocm_provider_factory.cc
View file

@ -104,8 +104,8 @@ struct ProviderInfo_ROCM_Impl : ProviderInfo_ROCM {
return rocm::Impl_Cast(static_cast<hipStream_t>(stream), input_data, output_data, count);
}
Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { return RocmCall<hipError_t, false>(hipError_t(retCode), exprString, libName, hipError_t(successCode), msg); }
void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { RocmCall<hipError_t, true>(hipError_t(retCode), exprString, libName, hipError_t(successCode), msg); }
Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { return RocmCall<hipError_t, false>(hipError_t(retCode), exprString, libName, hipError_t(successCode), msg, file, line); }
void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { RocmCall<hipError_t, true>(hipError_t(retCode), exprString, libName, hipError_t(successCode), msg, file, line); }
void CopyGpuToCpu(void* dst_ptr, const void* src_ptr, const size_t size, const OrtMemoryInfo& dst_location, const OrtMemoryInfo& src_location) override {
ORT_ENFORCE(dst_location.device.Type() == OrtDevice::CPU);

4
onnxruntime/core/providers/rocm/rocm_provider_factory.h
View file

@ -29,8 +29,8 @@ struct ProviderInfo_ROCM {
virtual void rocm__Impl_Cast(void* stream, const double* input_data, float* output_data, size_t count) = 0;
virtual void rocm__Impl_Cast(void* stream, const float* input_data, double* output_data, size_t count) = 0;
virtual Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
virtual void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
virtual void CopyGpuToCpu(void* dst_ptr, const void* src_ptr, const size_t size, const OrtMemoryInfo& dst_location, const OrtMemoryInfo& src_location) = 0;
virtual void rocmMemcpy_HostToDevice(void* dst, const void* src, size_t count) = 0;

34
onnxruntime/core/providers/rocm/shared_inc/rocm_call.h
View file

@ -13,27 +13,27 @@ namespace onnxruntime {
template <typename ERRTYPE, bool THRW>
std::conditional_t<THRW, void, Status> RocmCall(
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg = "");
ERRTYPE retCode, const char* exprString, const char* libName, ERRTYPE successCode, const char* msg, const char* file, const int line);
#define HIP_CALL(expr) (RocmCall<hipError_t, false>((expr), #expr, "HIP", hipSuccess))
#define ROCBLAS_CALL(expr) (RocmCall<rocblas_status, false>((expr), #expr, "ROCBLAS", rocblas_status_success))
#define HIPSPARSE_CALL(expr) (RocmCall<hipsparseStatus_t, false>((expr), #expr, "HIPSPARSE", HIPSPARSE_STATUS_SUCCESS))
#define HIPRAND_CALL(expr) (RocmCall<hiprandStatus_t, false>((expr), #expr, "HIPRAND", HIPRAND_STATUS_SUCCESS))
#define MIOPEN_CALL(expr) (RocmCall<miopenStatus_t, false>((expr), #expr, "MIOPEN", miopenStatusSuccess))
#define MIOPEN_CALL2(expr, m) (RocmCall<miopenStatus_t, false>((expr), #expr, "MIOPEN", miopenStatusSuccess, m))
#define HIPFFT_CALL(expr) (RocmCall<hipfftResult, false>((expr), #expr, "HIPFFT", HIPFFT_SUCCESS))
#define HIP_CALL(expr) (RocmCall<hipError_t, false>((expr), #expr, "HIP", hipSuccess, "", __FILE__, __LINE__))
#define ROCBLAS_CALL(expr) (RocmCall<rocblas_status, false>((expr), #expr, "ROCBLAS", rocblas_status_success, "", __FILE__, __LINE__))
#define HIPSPARSE_CALL(expr) (RocmCall<hipsparseStatus_t, false>((expr), #expr, "HIPSPARSE", HIPSPARSE_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define HIPRAND_CALL(expr) (RocmCall<hiprandStatus_t, false>((expr), #expr, "HIPRAND", HIPRAND_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define MIOPEN_CALL(expr) (RocmCall<miopenStatus_t, false>((expr), #expr, "MIOPEN", miopenStatusSuccess, "", __FILE__, __LINE__))
#define MIOPEN_CALL2(expr, m) (RocmCall<miopenStatus_t, false>((expr), #expr, "MIOPEN", miopenStatusSuccess, m, __FILE__, __LINE__))
#define HIPFFT_CALL(expr) (RocmCall<hipfftResult, false>((expr), #expr, "HIPFFT", HIPFFT_SUCCESS, "", __FILE__, __LINE__))
#define HIP_CALL_THROW(expr) (RocmCall<hipError_t, true>((expr), #expr, "HIP", hipSuccess))
#define ROCBLAS_CALL_THROW(expr) (RocmCall<rocblas_status, true>((expr), #expr, "ROCBLAS", rocblas_status_success))
#define HIPSPARSE_CALL_THROW(expr) (RocmCall<hipsparseStatus_t, true>((expr), #expr, "HIPSPARSE", HIPSPARSE_STATUS_SUCCESS))
#define HIPRAND_CALL_THROW(expr) (RocmCall<hiprandStatus_t, true>((expr), #expr, "HIPRAND", HIPRAND_STATUS_SUCCESS))
#define MIOPEN_CALL_THROW(expr) (RocmCall<miopenStatus_t, true>((expr), #expr, "MIOPEN", miopenStatusSuccess))
#define MIOPEN_CALL_THROW2(expr, m) (RocmCall<miopenStatus_t, true>((expr), #expr, "MIOPEN", miopenStatusSuccess, m))
#define HIPFFT_CALL_THROW(expr) (RocmCall<hipfftResult, true>((expr), #expr, "HIPFFT", HIPFFT_SUCCESS))
#define HIP_CALL_THROW(expr) (RocmCall<hipError_t, true>((expr), #expr, "HIP", hipSuccess, "", __FILE__, __LINE__))
#define ROCBLAS_CALL_THROW(expr) (RocmCall<rocblas_status, true>((expr), #expr, "ROCBLAS", rocblas_status_success, "", __FILE__, __LINE__))
#define HIPSPARSE_CALL_THROW(expr) (RocmCall<hipsparseStatus_t, true>((expr), #expr, "HIPSPARSE", HIPSPARSE_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define HIPRAND_CALL_THROW(expr) (RocmCall<hiprandStatus_t, true>((expr), #expr, "HIPRAND", HIPRAND_STATUS_SUCCESS, "", __FILE__, __LINE__))
#define MIOPEN_CALL_THROW(expr) (RocmCall<miopenStatus_t, true>((expr), #expr, "MIOPEN", miopenStatusSuccess, "", __FILE__, __LINE__))
#define MIOPEN_CALL_THROW2(expr, m) (RocmCall<miopenStatus_t, true>((expr), #expr, "MIOPEN", miopenStatusSuccess, m, __FILE__, __LINE__))
#define HIPFFT_CALL_THROW(expr) (RocmCall<hipfftResult, true>((expr), #expr, "HIPFFT", HIPFFT_SUCCESS, "", __FILE__, __LINE__))
#ifdef ORT_USE_NCCL
#define NCCL_CALL(expr) (RocmCall<ncclResult_t, false>((expr), #expr, "NCCL", ncclSuccess))
#define NCCL_CALL_THROW(expr) (RocmCall<ncclResult_t, true>((expr), #expr, "NCCL", ncclSuccess))
#define NCCL_CALL(expr) (RocmCall<ncclResult_t, false>((expr), #expr, "NCCL", ncclSuccess, "", __FILE__, __LINE__))
#define NCCL_CALL_THROW(expr) (RocmCall<ncclResult_t, true>((expr), #expr, "NCCL", ncclSuccess, "", __FILE__, __LINE__))
#endif
} // namespace onnxruntime

8
onnxruntime/core/providers/shared_library/provider_interfaces.h
View file

@ -165,8 +165,8 @@ struct ProviderHost {
virtual void cuda__Impl_Cast(void* stream, const double* input_data, float* output_data, size_t count) = 0;
virtual void cuda__Impl_Cast(void* stream, const float* input_data, double* output_data, size_t count) = 0;
virtual Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
virtual void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
#endif
#ifdef USE_ROCM
@ -179,8 +179,8 @@ struct ProviderHost {
virtual void rocm__Impl_Cast(void* stream, const double* input_data, float* output_data, size_t count) = 0;
virtual void rocm__Impl_Cast(void* stream, const float* input_data, double* output_data, size_t count) = 0;
virtual Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) = 0;
virtual Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
virtual void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) = 0;
#endif
virtual std::unordered_set<NodeIndex> GetCpuPreferredNodes(const onnxruntime::GraphViewer& graph,

30
onnxruntime/core/providers/tensorrt/tensorrt_execution_provider.cc
View file

@ -194,33 +194,33 @@ void Impl_Cast(
} // namespace cuda
template <>
Status CudaCall<cudaError, false>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg) {
return g_host->CudaCall_false(retCode, exprString, libName, successCode, msg);
Status CudaCall<cudaError, false>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg, const char* file, const int line) {
return g_host->CudaCall_false(retCode, exprString, libName, successCode, msg, file, line);
}
template <>
void CudaCall<cudaError, true>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg) {
return g_host->CudaCall_true(retCode, exprString, libName, successCode, msg);
void CudaCall<cudaError, true>(cudaError retCode, const char* exprString, const char* libName, cudaError successCode, const char* msg, const char* file, const int line) {
return g_host->CudaCall_true(retCode, exprString, libName, successCode, msg, file, line);
}
template <>
Status CudaCall<cublasStatus_t, false>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg) {
return g_host->CudaCall_false(retCode, exprString, libName, successCode, msg);
Status CudaCall<cublasStatus_t, false>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg, const char* file, const int line) {
return g_host->CudaCall_false(retCode, exprString, libName, successCode, msg, file, line);
}
template <>
void CudaCall<cublasStatus_t, true>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg) {
return g_host->CudaCall_true(retCode, exprString, libName, successCode, msg);
void CudaCall<cublasStatus_t, true>(cublasStatus_t retCode, const char* exprString, const char* libName, cublasStatus_t successCode, const char* msg, const char* file, const int line) {
return g_host->CudaCall_true(retCode, exprString, libName, successCode, msg, file, line);
}
template <>
Status CudaCall<cudnnStatus_t, false>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg) {
return g_host->CudaCall_false(retCode, exprString, libName, successCode, msg);
Status CudaCall<cudnnStatus_t, false>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg, const char* file, const int line) {
return g_host->CudaCall_false(retCode, exprString, libName, successCode, msg, file, line);
}
template <>
void CudaCall<cudnnStatus_t, true>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg) {
return g_host->CudaCall_true(retCode, exprString, libName, successCode, msg);
void CudaCall<cudnnStatus_t, true>(cudnnStatus_t retCode, const char* exprString, const char* libName, cudnnStatus_t successCode, const char* msg, const char* file, const int line) {
return g_host->CudaCall_true(retCode, exprString, libName, successCode, msg, file, line);
}
class Memcpy final : public OpKernel {
@ -1496,8 +1496,7 @@ common::Status TensorrtExecutionProvider::Compile(const std::vector<FusedNodeAnd
}
if (detailed_build_log_) {
auto engine_build_stop = std::chrono::steady_clock::now();
LOGS_DEFAULT(INFO) << "TensorRT engine build for " << trt_node_name_with_precision << " took: " <<
std::chrono::duration_cast<std::chrono::milliseconds>(engine_build_stop - engine_build_start).count() << "ms" << std::endl;
LOGS_DEFAULT(INFO) << "TensorRT engine build for " << trt_node_name_with_precision << " took: " << std::chrono::duration_cast<std::chrono::milliseconds>(engine_build_stop - engine_build_start).count() << "ms" << std::endl;
}
if (engine_cache_enable_) {
std::unique_ptr<nvinfer1::IHostMemory> serializedModel(trt_engine->serialize());
@ -1899,8 +1898,7 @@ common::Status TensorrtExecutionProvider::Compile(const std::vector<FusedNodeAnd
trt_builder->buildEngineWithConfig(*trt_state->network->get(), *trt_config));
if (detailed_build_log_) {
auto engine_build_stop = std::chrono::steady_clock::now();
LOGS_DEFAULT(INFO) << "TensorRT engine build for " << trt_state->trt_node_name_with_precision << " took: " <<
std::chrono::duration_cast<std::chrono::milliseconds>(engine_build_stop - engine_build_start).count() << "ms" << std::endl;
LOGS_DEFAULT(INFO) << "TensorRT engine build for " << trt_state->trt_node_name_with_precision << " took: " << std::chrono::duration_cast<std::chrono::milliseconds>(engine_build_stop - engine_build_start).count() << "ms" << std::endl;
}
}
if (trt_state->engine == nullptr) {

10
onnxruntime/core/session/provider_bridge_ort.cc
View file

@ -205,8 +205,8 @@ struct ProviderHostImpl : ProviderHost {
void cuda__Impl_Cast(void* stream, const double* input_data, float* output_data, size_t count) override { return GetProviderInfo_CUDA().cuda__Impl_Cast(stream, input_data, output_data, count); }
void cuda__Impl_Cast(void* stream, const float* input_data, double* output_data, size_t count) override { return GetProviderInfo_CUDA().cuda__Impl_Cast(stream, input_data, output_data, count); }
Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { return GetProviderInfo_CUDA().CudaCall_false(retCode, exprString, libName, successCode, msg); }
void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { GetProviderInfo_CUDA().CudaCall_true(retCode, exprString, libName, successCode, msg); }
Status CudaCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { return GetProviderInfo_CUDA().CudaCall_false(retCode, exprString, libName, successCode, msg, file, line); }
void CudaCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { GetProviderInfo_CUDA().CudaCall_true(retCode, exprString, libName, successCode, msg, file, line); }
#endif
#ifdef USE_ROCM
@ -220,8 +220,8 @@ struct ProviderHostImpl : ProviderHost {
void rocm__Impl_Cast(void* stream, const double* input_data, float* output_data, size_t count) override { return GetProviderInfo_ROCM().rocm__Impl_Cast(stream, input_data, output_data, count); }
void rocm__Impl_Cast(void* stream, const float* input_data, double* output_data, size_t count) override { return GetProviderInfo_ROCM().rocm__Impl_Cast(stream, input_data, output_data, count); }
Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { return GetProviderInfo_ROCM().RocmCall_false(retCode, exprString, libName, successCode, msg); }
void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg) override { GetProviderInfo_ROCM().RocmCall_true(retCode, exprString, libName, successCode, msg); }
Status RocmCall_false(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { return GetProviderInfo_ROCM().RocmCall_false(retCode, exprString, libName, successCode, msg, file, line); }
void RocmCall_true(int retCode, const char* exprString, const char* libName, int successCode, const char* msg, const char* file, const int line) override { GetProviderInfo_ROCM().RocmCall_true(retCode, exprString, libName, successCode, msg, file, line); }
#endif
std::string GetEnvironmentVar(const std::string& var_name) override { return Env::Default().GetEnvironmentVar(var_name); }
@ -1936,7 +1936,7 @@ ORT_API_STATUS_IMPL(OrtApis::SessionOptionsAppendExecutionProvider_Dnnl,
auto factory = onnxruntime::DnnlProviderFactoryCreator::Create(dnnl_options);
if (!factory) {
return OrtApis::CreateStatus(ORT_FAIL,
"SessionOptionsAppendExecutionProvider_Dnnl: Failed to load shared library");
"SessionOptionsAppendExecutionProvider_Dnnl: Failed to load shared library");
}
options->provider_factories.push_back(factory);