/** * Copyright 2019 Huawei Technologies Co., Ltd * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #ifndef MINDSPORE_CCSRC_RUNTIME_DEVICE_GPU_GPU_COMMON_H_ #define MINDSPORE_CCSRC_RUNTIME_DEVICE_GPU_GPU_COMMON_H_ #include #include #include #include #include "utils/log_adapter.h" #include "utils/trace_base.h" #include "include/curand.h" namespace mindspore { namespace device { namespace gpu { #define CHECK_OP_RET_WITH_EXCEPT(expression, message) \ { \ bool success = (expression); \ if (!success) { \ MS_LOG(EXCEPTION) << "Op Error: " << message << " | Error Number: " << success; \ } \ } #define CHECK_OP_RET_WITH_ERROR(expression, message) \ { \ bool success = (expression); \ if (!success) { \ MS_LOG(ERROR) << "Op Error: " << message << " | Error Number: " << success; \ } \ } #define CHECK_RET_WITH_RETURN_ERROR(expression, message) \ { \ bool success = (expression); \ if (!success) { \ MS_LOG(ERROR) << message; \ return false; \ } \ } #define CHECK_CUDA_RET_WITH_ERROR(node, expression, message) \ { \ cudaError_t status = (expression); \ if (status != cudaSuccess) { \ MS_LOG(ERROR) << "CUDA Error: " << message << " | Error Number: " << status << " " << cudaGetErrorString(status) \ << trace::DumpSourceLines(node.lock()); \ } \ } #define CHECK_CUDA_RET_WITH_ERROR_NOTRACE(expression, message) \ { \ cudaError_t status = (expression); \ if (status != cudaSuccess) { \ MS_LOG(ERROR) << "CUDA Error: " << message << " | Error Number: " << status << " " \ << cudaGetErrorString(status); \ } \ } #define CHECK_CUDA_RET_WITH_RETURN_ERROR_NOTRACE(expression, message) \ { \ cudaError_t status = (expression); \ if (status != cudaSuccess) { \ MS_LOG(ERROR) << "CUDA Error: " << message << " | Error Number: " << status << " " \ << cudaGetErrorString(status); \ return false; \ } \ } #define CHECK_CUDA_RET_WITH_EXCEPT(node, expression, message) \ { \ cudaError_t status = (expression); \ if (status != cudaSuccess) { \ MS_LOG(EXCEPTION) << "CUDA Error: " << message << " | Error Number: " << status << " " \ << cudaGetErrorString(status) << trace::DumpSourceLines(node.lock()); \ } \ } #define CHECK_CUDA_RET_WITH_EXCEPT_NOTRACE(expression, message) \ { \ cudaError_t status = (expression); \ if (status != cudaSuccess) { \ MS_LOG(EXCEPTION) << "CUDA Error: " << message << " | Error Number: " << status << " " \ << cudaGetErrorString(status); \ } \ } #define CHECK_CUDNN_RET_WITH_EXCEPT(node, expression, message) \ { \ cudnnStatus_t status = (expression); \ if (status != CUDNN_STATUS_SUCCESS) { \ MS_LOG(EXCEPTION) << "cuDNN Error: " << message << " | Error Number: " << status << " " \ << cudnnGetErrorString(status) << trace::DumpSourceLines(node.lock()); \ } \ } #define CHECK_CUDNN_RET_WITH_EXCEPT_NOTRACE(expression, message) \ { \ cudnnStatus_t status = (expression); \ if (status != CUDNN_STATUS_SUCCESS) { \ MS_LOG(EXCEPTION) << "cuDNN Error: " << message << " | Error Number: " << status << " " \ << cudnnGetErrorString(status); \ } \ } #define CHECK_CUDNN_RET_WITH_ERROR_NOTRACE(expression, message) \ { \ cudnnStatus_t status = (expression); \ if (status != CUDNN_STATUS_SUCCESS) { \ MS_LOG(ERROR) << "cuDNN Error: " << message << " | Error Number: " << status << " " \ << cudnnGetErrorString(status); \ } \ } #define CHECK_CUDNN_RET_WITH_ERROR(node, expression, message) \ { \ cudnnStatus_t status = (expression); \ if (status != CUDNN_STATUS_SUCCESS) { \ MS_LOG(ERROR) << "cuDNN Error: " << message << " | Error Number: " << status << " " \ << cudnnGetErrorString(status) << trace::DumpSourceLines(node.lock()); \ } \ } #define CHECK_CUBLAS_RET_WITH_EXCEPT_NOTRACE(expression, message) \ { \ cublasStatus_t status = (expression); \ if (status != CUBLAS_STATUS_SUCCESS) { \ MS_LOG(EXCEPTION) << "cuBLAS Error: " << message << " | Error Number: " << status; \ } \ } #define CHECK_CUBLAS_RET_WITH_EXCEPT(node, expression, message) \ { \ cublasStatus_t status = (expression); \ if (status != CUBLAS_STATUS_SUCCESS) { \ MS_LOG(EXCEPTION) << "cuBLAS Error: " << message << " | Error Number: " << status \ << trace::DumpSourceLines(node.lock()); \ } \ } #define CHECK_CUBLAS_RET_WITH_ERROR(expression, message) \ { \ cublasStatus_t status = (expression); \ if (status != CUBLAS_STATUS_SUCCESS) { \ MS_LOG(ERROR) << "cuBLAS Error: " << message << " | Error Number: " << status; \ } \ } #define CHECK_CUSOLVER_RET_WITH_EXCEPT_NOTRACE(expression, message) \ { \ cusolverStatus_t status = (expression); \ if (status != CUSOLVER_STATUS_SUCCESS) { \ MS_LOG(EXCEPTION) << "cusolver Error: " << message << " | Error Number: " << status; \ } \ } #define CHECK_CUSOLVER_RET_WITH_EXCEPT(node, expression, message) \ { \ cusolverStatus_t status = (expression); \ if (status != CUSOLVER_STATUS_SUCCESS) { \ MS_LOG(EXCEPTION) << "cusolver Error: " << message << " | Error Number: " << status \ << trace::DumpSourceLines(node.lock()); \ ; \ } \ } #define CHECK_CUSOLVER_RET_WITH_ERROR(expression, message) \ { \ cusolverStatus_t status = (expression); \ if (status != CUSOLVER_STATUS_SUCCESS) { \ MS_LOG(ERROR) << "cusolver Error: " << message << " | Error Number: " << status; \ } \ } #define CHECK_NCCL_RET_WITH_EXCEPT(node, expression, message) \ { \ int result = (expression); \ if (result != ncclSuccess) { \ MS_LOG(EXCEPTION) << "NCCL Error: " << message << " | Error Number: " << result \ << trace::DumpSourceLines(node.lock()); \ } \ } #define VARIABLE_NOT_USED(var) \ { (void)(var); } inline bool CheckNullInput(const std::vector &input_shape) { // If input_shape.size() == 0, it means a scalar input; If input_shape.size() != 0 and input_shape contains 0, // it means a null input. Just return a null output. if (input_shape.size() != 0) { if (std::any_of(input_shape.begin(), input_shape.end(), [](size_t i) { return i == 0; })) { return true; } } return false; } #define CHECK_NULL_INPUT(input_shape) mindspore::device::gpu::CheckNullInput(input_shape) inline const char *CurandGetErrorString(curandStatus_t status) { switch (status) { case CURAND_STATUS_VERSION_MISMATCH: return "Header file and linked library version do not match."; case CURAND_STATUS_NOT_INITIALIZED: return "Generator not initialized."; case CURAND_STATUS_ALLOCATION_FAILED: return "Memory allocation failed."; case CURAND_STATUS_TYPE_ERROR: return "Generator is wrong type."; case CURAND_STATUS_OUT_OF_RANGE: return "Argument out of range."; case CURAND_STATUS_LENGTH_NOT_MULTIPLE: return "Length requested is not a multiple of dimension."; case CURAND_STATUS_DOUBLE_PRECISION_REQUIRED: return "GPU does not have double precision required by MRG32k3a."; case CURAND_STATUS_LAUNCH_FAILURE: return "Kernel launch failure."; case CURAND_STATUS_PREEXISTING_FAILURE: return "Preexisting failure on library entry."; case CURAND_STATUS_INITIALIZATION_FAILED: return "Initialization of CUDA failed."; case CURAND_STATUS_ARCH_MISMATCH: return "Architecture mismatch, GPU does not support requested feature."; case CURAND_STATUS_INTERNAL_ERROR: return "Internal library error."; default: return "Unknown the curandStatus."; } } #define CHECK_CURAND_RET_WITH_EXCEPT(expression, message) \ { \ curandStatus_t status = (expression); \ if (status != CURAND_STATUS_SUCCESS) { \ MS_LOG(EXCEPTION) << "CUDA curand Error: " << message << " | curandStatus: " << status << " " \ << mindspore::device::gpu::CurandGetErrorString(status); \ } \ } } // namespace gpu } // namespace device } // namespace mindspore #endif // MINDSPORE_CCSRC_RUNTIME_DEVICE_GPU_GPU_COMMON_H_