17fcac036SJeremy L Thompson // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 27fcac036SJeremy L Thompson // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 37fcac036SJeremy L Thompson // All Rights reserved. See files LICENSE and NOTICE for details. 47fcac036SJeremy L Thompson // 57fcac036SJeremy L Thompson // This file is part of CEED, a collection of benchmarks, miniapps, software 67fcac036SJeremy L Thompson // libraries and APIs for efficient high-order finite element and spectral 77fcac036SJeremy L Thompson // element discretizations for exascale applications. For more information and 87fcac036SJeremy L Thompson // source code availability see http://github.com/ceed. 97fcac036SJeremy L Thompson // 107fcac036SJeremy L Thompson // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 117fcac036SJeremy L Thompson // a collaborative effort of two U.S. Department of Energy organizations (Office 127fcac036SJeremy L Thompson // of Science and the National Nuclear Security Administration) responsible for 137fcac036SJeremy L Thompson // the planning and preparation of a capable exascale ecosystem, including 147fcac036SJeremy L Thompson // software, applications, hardware, advanced system engineering and early 157fcac036SJeremy L Thompson // testbed platforms, in support of the nation's exascale computing imperative. 167fcac036SJeremy L Thompson 177fcac036SJeremy L Thompson #ifndef _ceed_cuda_common_h 187fcac036SJeremy L Thompson #define _ceed_cuda_common_h 197fcac036SJeremy L Thompson 207fcac036SJeremy L Thompson #include <ceed/ceed.h> 217fcac036SJeremy L Thompson #include <ceed/backend.h> 227fcac036SJeremy L Thompson #include <cublas_v2.h> 237fcac036SJeremy L Thompson #include <cuda.h> 247fcac036SJeremy L Thompson #include <cuda_runtime.h> 257fcac036SJeremy L Thompson 267fcac036SJeremy L Thompson #define QUOTE(...) #__VA_ARGS__ 277fcac036SJeremy L Thompson 287fcac036SJeremy L Thompson #define CeedChk_Cu(ceed, x) \ 297fcac036SJeremy L Thompson do { \ 307fcac036SJeremy L Thompson CUresult cuda_result = x; \ 317fcac036SJeremy L Thompson if (cuda_result != CUDA_SUCCESS) { \ 327fcac036SJeremy L Thompson const char *msg; \ 337fcac036SJeremy L Thompson cuGetErrorName(cuda_result, &msg); \ 347fcac036SJeremy L Thompson return CeedError((ceed), CEED_ERROR_BACKEND, msg); \ 357fcac036SJeremy L Thompson } \ 367fcac036SJeremy L Thompson } while (0) 377fcac036SJeremy L Thompson 387fcac036SJeremy L Thompson #define CeedChk_Cublas(ceed, x) \ 397fcac036SJeremy L Thompson do { \ 407fcac036SJeremy L Thompson cublasStatus_t cublas_result = x; \ 417fcac036SJeremy L Thompson if (cublas_result != CUBLAS_STATUS_SUCCESS) { \ 427fcac036SJeremy L Thompson const char *msg = cublasGetErrorName(cublas_result); \ 437fcac036SJeremy L Thompson return CeedError((ceed), CEED_ERROR_BACKEND, msg); \ 447fcac036SJeremy L Thompson } \ 457fcac036SJeremy L Thompson } while (0) 467fcac036SJeremy L Thompson 477fcac036SJeremy L Thompson #define CASE(name) case name: return #name 487fcac036SJeremy L Thompson // LCOV_EXCL_START 497fcac036SJeremy L Thompson static const char *cublasGetErrorName(cublasStatus_t error) { 507fcac036SJeremy L Thompson switch (error) { 517fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_SUCCESS); 527fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_NOT_INITIALIZED); 537fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_ALLOC_FAILED); 547fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_INVALID_VALUE); 557fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_ARCH_MISMATCH); 567fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_MAPPING_ERROR); 577fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_EXECUTION_FAILED); 587fcac036SJeremy L Thompson CASE(CUBLAS_STATUS_INTERNAL_ERROR); 597fcac036SJeremy L Thompson default: return "CUBLAS_STATUS_UNKNOWN_ERROR"; 607fcac036SJeremy L Thompson } 617fcac036SJeremy L Thompson } 627fcac036SJeremy L Thompson // LCOV_EXCL_STOP 637fcac036SJeremy L Thompson 647fcac036SJeremy L Thompson typedef struct { 657fcac036SJeremy L Thompson int deviceId; 667fcac036SJeremy L Thompson cublasHandle_t cublasHandle; 677fcac036SJeremy L Thompson struct cudaDeviceProp deviceProp; 687fcac036SJeremy L Thompson } Ceed_Cuda; 697fcac036SJeremy L Thompson 70*f87d896cSJeremy L Thompson CEED_INTERN int CeedCudaInit(Ceed ceed, const char *resource); 717fcac036SJeremy L Thompson 727fcac036SJeremy L Thompson CEED_INTERN int CeedDestroy_Cuda(Ceed ceed); 737fcac036SJeremy L Thompson 747fcac036SJeremy L Thompson #endif // _ceed_cuda_common_h