1*241a4b83SYohann // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2*241a4b83SYohann // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3*241a4b83SYohann // All Rights reserved. See files LICENSE and NOTICE for details. 4*241a4b83SYohann // 5*241a4b83SYohann // This file is part of CEED, a collection of benchmarks, miniapps, software 6*241a4b83SYohann // libraries and APIs for efficient high-order finite element and spectral 7*241a4b83SYohann // element discretizations for exascale applications. For more information and 8*241a4b83SYohann // source code availability see http://github.com/ceed. 9*241a4b83SYohann // 10*241a4b83SYohann // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11*241a4b83SYohann // a collaborative effort of two U.S. Department of Energy organizations (Office 12*241a4b83SYohann // of Science and the National Nuclear Security Administration) responsible for 13*241a4b83SYohann // the planning and preparation of a capable exascale ecosystem, including 14*241a4b83SYohann // software, applications, hardware, advanced system engineering and early 15*241a4b83SYohann // testbed platforms, in support of the nation's exascale computing imperative. 16*241a4b83SYohann 17*241a4b83SYohann #include <ceed-backend.h> 18*241a4b83SYohann #include <string.h> 19*241a4b83SYohann #include <stdarg.h> 20*241a4b83SYohann #include <nvrtc.h> 21*241a4b83SYohann #include <cuda.h> 22*241a4b83SYohann #include <cuda_runtime.h> 23*241a4b83SYohann #include "ceed-cuda-gen.h" 24*241a4b83SYohann 25*241a4b83SYohann static int CeedGetPreferredMemType_Cuda_gen(CeedMemType *type) { 26*241a4b83SYohann *type = CEED_MEM_DEVICE; 27*241a4b83SYohann return 0; 28*241a4b83SYohann } 29*241a4b83SYohann 30*241a4b83SYohann static int CeedInit_Cuda_gen(const char *resource, Ceed ceed) { 31*241a4b83SYohann int ierr; 32*241a4b83SYohann const int nrc = 9; // number of characters in resource 33*241a4b83SYohann if (strncmp(resource, "/gpu/cuda/gen", nrc)) 34*241a4b83SYohann return CeedError(ceed, 1, "Cuda backend cannot use resource: %s", resource); 35*241a4b83SYohann 36*241a4b83SYohann Ceed ceedshared; 37*241a4b83SYohann CeedInit("/gpu/cuda/shared", &ceedshared); 38*241a4b83SYohann ierr = CeedSetDelegate(ceed, ceedshared); CeedChk(ierr); 39*241a4b83SYohann 40*241a4b83SYohann const int rlen = strlen(resource); 41*241a4b83SYohann const bool slash = (rlen>nrc) ? (resource[nrc] == '/') : false; 42*241a4b83SYohann const int deviceID = (slash && rlen > nrc + 1) ? atoi(&resource[nrc + 1]) : 0; 43*241a4b83SYohann 44*241a4b83SYohann ierr = cudaSetDevice(deviceID); CeedChk(ierr); 45*241a4b83SYohann 46*241a4b83SYohann Ceed_Cuda_gen *data; 47*241a4b83SYohann ierr = CeedCalloc(1,&data); CeedChk(ierr); 48*241a4b83SYohann 49*241a4b83SYohann ierr = CeedSetData(ceed,(void *)&data); CeedChk(ierr); 50*241a4b83SYohann ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "GetPreferredMemType", 51*241a4b83SYohann CeedGetPreferredMemType_Cuda_gen); CeedChk(ierr); 52*241a4b83SYohann ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "QFunctionCreate", 53*241a4b83SYohann CeedQFunctionCreate_Cuda_gen); CeedChk(ierr); 54*241a4b83SYohann ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "OperatorCreate", 55*241a4b83SYohann CeedOperatorCreate_Cuda_gen); CeedChk(ierr); 56*241a4b83SYohann ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "CompositeOperatorCreate", 57*241a4b83SYohann CeedCompositeOperatorCreate_Cuda_gen); CeedChk(ierr); 58*241a4b83SYohann return 0; 59*241a4b83SYohann } 60*241a4b83SYohann 61*241a4b83SYohann __attribute__((constructor)) 62*241a4b83SYohann static void Register(void) { 63*241a4b83SYohann CeedRegister("/gpu/cuda/gen", CeedInit_Cuda_gen, 40); 64*241a4b83SYohann } 65