1 // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2 // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3 // All Rights reserved. See files LICENSE and NOTICE for details. 4 // 5 // This file is part of CEED, a collection of benchmarks, miniapps, software 6 // libraries and APIs for efficient high-order finite element and spectral 7 // element discretizations for exascale applications. For more information and 8 // source code availability see http://github.com/ceed. 9 // 10 // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11 // a collaborative effort of two U.S. Department of Energy organizations (Office 12 // of Science and the National Nuclear Security Administration) responsible for 13 // the planning and preparation of a capable exascale ecosystem, including 14 // software, applications, hardware, advanced system engineering and early 15 // testbed platforms, in support of the nation's exascale computing imperative. 16 17 #include <ceed-backend.h> 18 #include <string.h> 19 #include <stdarg.h> 20 #include <nvrtc.h> 21 #include <cuda.h> 22 #include <cuda_runtime.h> 23 #include "ceed-cuda-shared.h" 24 25 static int CeedGetPreferredMemType_Cuda_shared(CeedMemType *type) { 26 *type = CEED_MEM_DEVICE; 27 return 0; 28 } 29 30 static int CeedInit_Cuda_shared(const char *resource, Ceed ceed) { 31 int ierr; 32 const int nrc = 9; // number of characters in resource 33 if (strncmp(resource, "/gpu/cuda/shared", nrc)) 34 return CeedError(ceed, 1, "Cuda backend cannot use resource: %s", resource); 35 36 Ceed ceedreg; 37 CeedInit("/gpu/cuda/reg", &ceedreg); 38 ierr = CeedSetDelegate(ceed, ceedreg); CeedChk(ierr); 39 40 const int rlen = strlen(resource); 41 const bool slash = (rlen>nrc) ? (resource[nrc] == '/') : false; 42 const int deviceID = (slash && rlen > nrc + 1) ? atoi(&resource[nrc + 1]) : 0; 43 44 ierr = cudaSetDevice(deviceID); CeedChk(ierr); 45 46 Ceed_Cuda_shared *data; 47 ierr = CeedCalloc(1,&data); CeedChk(ierr); 48 data->Q1d = -1; 49 data->P1d = -1; 50 data->grad = false; 51 52 ierr = CeedSetData(ceed,(void *)&data); CeedChk(ierr); 53 ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "GetPreferredMemType", 54 CeedGetPreferredMemType_Cuda_shared); CeedChk(ierr); 55 ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "BasisCreateTensorH1", 56 CeedBasisCreateTensorH1_Cuda_shared); CeedChk(ierr); 57 ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "BasisCreateH1", 58 CeedBasisCreateH1_Cuda_shared); CeedChk(ierr); 59 return 0; 60 } 61 62 __attribute__((constructor)) 63 static void Register(void) { 64 CeedRegister("/gpu/cuda/shared", CeedInit_Cuda_shared, 40); 65 } 66