1c532df63SYohann // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2c532df63SYohann // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3c532df63SYohann // All Rights reserved. See files LICENSE and NOTICE for details. 4c532df63SYohann // 5c532df63SYohann // This file is part of CEED, a collection of benchmarks, miniapps, software 6c532df63SYohann // libraries and APIs for efficient high-order finite element and spectral 7c532df63SYohann // element discretizations for exascale applications. For more information and 8c532df63SYohann // source code availability see http://github.com/ceed. 9c532df63SYohann // 10c532df63SYohann // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11c532df63SYohann // a collaborative effort of two U.S. Department of Energy organizations (Office 12c532df63SYohann // of Science and the National Nuclear Security Administration) responsible for 13c532df63SYohann // the planning and preparation of a capable exascale ecosystem, including 14c532df63SYohann // software, applications, hardware, advanced system engineering and early 15c532df63SYohann // testbed platforms, in support of the nation's exascale computing imperative. 16c532df63SYohann 17c532df63SYohann #include <ceed-backend.h> 18c532df63SYohann #include <string.h> 19c532df63SYohann #include <stdarg.h> 20c532df63SYohann #include <nvrtc.h> 21c532df63SYohann #include <cuda.h> 22c532df63SYohann #include <cuda_runtime.h> 23c532df63SYohann #include "ceed-cuda-shared.h" 24c532df63SYohann 25*ab213215SJeremy L Thompson //------------------------------------------------------------------------------ 26*ab213215SJeremy L Thompson // Backend init 27*ab213215SJeremy L Thompson //------------------------------------------------------------------------------ 28c532df63SYohann static int CeedInit_Cuda_shared(const char *resource, Ceed ceed) { 29c532df63SYohann int ierr; 30c532df63SYohann const int nrc = 9; // number of characters in resource 31c532df63SYohann if (strncmp(resource, "/gpu/cuda/shared", nrc)) 32c532df63SYohann return CeedError(ceed, 1, "Cuda backend cannot use resource: %s", resource); 33c532df63SYohann 34c532df63SYohann Ceed ceedreg; 35c532df63SYohann CeedInit("/gpu/cuda/reg", &ceedreg); 36a4999eddSjeremylt ierr = CeedSetDelegate(ceed, ceedreg); CeedChk(ierr); 37c532df63SYohann 38c532df63SYohann Ceed_Cuda_shared *data; 39c532df63SYohann ierr = CeedCalloc(1,&data); CeedChk(ierr); 40c532df63SYohann ierr = CeedSetData(ceed,(void *)&data); CeedChk(ierr); 41abfaacbbSSander Arens ierr = CeedCudaInit(ceed, resource, nrc); CeedChk(ierr); 42abfaacbbSSander Arens 43c532df63SYohann ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "BasisCreateTensorH1", 445afe0718Sjeremylt CeedBasisCreateTensorH1_Cuda_shared); 455afe0718Sjeremylt CeedChk(ierr); 46c532df63SYohann return 0; 47c532df63SYohann } 48c532df63SYohann 49*ab213215SJeremy L Thompson //------------------------------------------------------------------------------ 50*ab213215SJeremy L Thompson // Register backend 51*ab213215SJeremy L Thompson //------------------------------------------------------------------------------ 52c532df63SYohann __attribute__((constructor)) 53c532df63SYohann static void Register(void) { 54c532df63SYohann CeedRegister("/gpu/cuda/shared", CeedInit_Cuda_shared, 40); 55c532df63SYohann } 56*ab213215SJeremy L Thompson //------------------------------------------------------------------------------ 57