1*7fcac036SJeremy L Thompson // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2*7fcac036SJeremy L Thompson // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3*7fcac036SJeremy L Thompson // All Rights reserved. See files LICENSE and NOTICE for details. 4*7fcac036SJeremy L Thompson // 5*7fcac036SJeremy L Thompson // This file is part of CEED, a collection of benchmarks, miniapps, software 6*7fcac036SJeremy L Thompson // libraries and APIs for efficient high-order finite element and spectral 7*7fcac036SJeremy L Thompson // element discretizations for exascale applications. For more information and 8*7fcac036SJeremy L Thompson // source code availability see http://github.com/ceed. 9*7fcac036SJeremy L Thompson // 10*7fcac036SJeremy L Thompson // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11*7fcac036SJeremy L Thompson // a collaborative effort of two U.S. Department of Energy organizations (Office 12*7fcac036SJeremy L Thompson // of Science and the National Nuclear Security Administration) responsible for 13*7fcac036SJeremy L Thompson // the planning and preparation of a capable exascale ecosystem, including 14*7fcac036SJeremy L Thompson // software, applications, hardware, advanced system engineering and early 15*7fcac036SJeremy L Thompson // testbed platforms, in support of the nation's exascale computing imperative. 16*7fcac036SJeremy L Thompson 17*7fcac036SJeremy L Thompson #ifndef _ceed_cuda_common_h 18*7fcac036SJeremy L Thompson #define _ceed_cuda_common_h 19*7fcac036SJeremy L Thompson 20*7fcac036SJeremy L Thompson #include <ceed/ceed.h> 21*7fcac036SJeremy L Thompson #include <ceed/backend.h> 22*7fcac036SJeremy L Thompson #include <cublas_v2.h> 23*7fcac036SJeremy L Thompson #include <cuda.h> 24*7fcac036SJeremy L Thompson #include <cuda_runtime.h> 25*7fcac036SJeremy L Thompson 26*7fcac036SJeremy L Thompson #define QUOTE(...) #__VA_ARGS__ 27*7fcac036SJeremy L Thompson 28*7fcac036SJeremy L Thompson #define CUDA_MAX_PATH 256 29*7fcac036SJeremy L Thompson 30*7fcac036SJeremy L Thompson #define CeedChk_Cu(ceed, x) \ 31*7fcac036SJeremy L Thompson do { \ 32*7fcac036SJeremy L Thompson CUresult cuda_result = x; \ 33*7fcac036SJeremy L Thompson if (cuda_result != CUDA_SUCCESS) { \ 34*7fcac036SJeremy L Thompson const char *msg; \ 35*7fcac036SJeremy L Thompson cuGetErrorName(cuda_result, &msg); \ 36*7fcac036SJeremy L Thompson return CeedError((ceed), CEED_ERROR_BACKEND, msg); \ 37*7fcac036SJeremy L Thompson } \ 38*7fcac036SJeremy L Thompson } while (0) 39*7fcac036SJeremy L Thompson 40*7fcac036SJeremy L Thompson #define CeedChk_Cublas(ceed, x) \ 41*7fcac036SJeremy L Thompson do { \ 42*7fcac036SJeremy L Thompson cublasStatus_t cublas_result = x; \ 43*7fcac036SJeremy L Thompson if (cublas_result != CUBLAS_STATUS_SUCCESS) { \ 44*7fcac036SJeremy L Thompson const char *msg = cublasGetErrorName(cublas_result); \ 45*7fcac036SJeremy L Thompson return CeedError((ceed), CEED_ERROR_BACKEND, msg); \ 46*7fcac036SJeremy L Thompson } \ 47*7fcac036SJeremy L Thompson } while (0) 48*7fcac036SJeremy L Thompson 49*7fcac036SJeremy L Thompson #define CASE(name) case name: return #name 50*7fcac036SJeremy L Thompson // LCOV_EXCL_START 51*7fcac036SJeremy L Thompson static const char *cublasGetErrorName(cublasStatus_t error) { 52*7fcac036SJeremy L Thompson switch (error) { 53*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_SUCCESS); 54*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_NOT_INITIALIZED); 55*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_ALLOC_FAILED); 56*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_INVALID_VALUE); 57*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_ARCH_MISMATCH); 58*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_MAPPING_ERROR); 59*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_EXECUTION_FAILED); 60*7fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_INTERNAL_ERROR); 61*7fcac036SJeremy L Thompson default: return "CUBLAS_STATUS_UNKNOWN_ERROR"; 62*7fcac036SJeremy L Thompson } 63*7fcac036SJeremy L Thompson } 64*7fcac036SJeremy L Thompson // LCOV_EXCL_STOP 65*7fcac036SJeremy L Thompson 66*7fcac036SJeremy L Thompson typedef struct { 67*7fcac036SJeremy L Thompson int deviceId; 68*7fcac036SJeremy L Thompson cublasHandle_t cublasHandle; 69*7fcac036SJeremy L Thompson struct cudaDeviceProp deviceProp; 70*7fcac036SJeremy L Thompson } Ceed_Cuda; 71*7fcac036SJeremy L Thompson 72*7fcac036SJeremy L Thompson CEED_INTERN int CeedCudaInit(Ceed ceed, const char *resource, int nrc); 73*7fcac036SJeremy L Thompson 74*7fcac036SJeremy L Thompson CEED_INTERN int CeedDestroy_Cuda(Ceed ceed); 75*7fcac036SJeremy L Thompson 76*7fcac036SJeremy L Thompson #endif // _ceed_cuda_common_h