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-gen.h" 24 25 //------------------------------------------------------------------------------ 26 // CUDA preferred MemType 27 //------------------------------------------------------------------------------ 28 static int CeedGetPreferredMemType_Cuda_gen(CeedMemType *type) { 29 *type = CEED_MEM_DEVICE; 30 return 0; 31 } 32 33 //------------------------------------------------------------------------------ 34 // Backend init 35 //------------------------------------------------------------------------------ 36 static int CeedInit_Cuda_gen(const char *resource, Ceed ceed) { 37 int ierr; 38 const int nrc = 9; // number of characters in resource 39 if (strncmp(resource, "/gpu/cuda/gen", nrc)) 40 return CeedError(ceed, 1, "Cuda backend cannot use resource: %s", resource); 41 42 Ceed ceedshared; 43 CeedInit("/gpu/cuda/shared", &ceedshared); 44 ierr = CeedSetDelegate(ceed, ceedshared); CeedChk(ierr); 45 46 Ceed_Cuda_gen *data; 47 ierr = CeedCalloc(1,&data); CeedChk(ierr); 48 ierr = CeedSetData(ceed,(void *)&data); CeedChk(ierr); 49 ierr = CeedCudaInit(ceed, resource, nrc); CeedChk(ierr); 50 51 const char fallbackresource[] = "/cpu/self/ref/serial"; 52 ierr = CeedSetOperatorFallbackResource(ceed, fallbackresource); CeedChk(ierr); 53 54 ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "GetPreferredMemType", 55 CeedGetPreferredMemType_Cuda_gen); CeedChk(ierr); 56 ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "QFunctionCreate", 57 CeedQFunctionCreate_Cuda_gen); CeedChk(ierr); 58 ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "OperatorCreate", 59 CeedOperatorCreate_Cuda_gen); CeedChk(ierr); 60 return 0; 61 } 62 63 //------------------------------------------------------------------------------ 64 // Register backend 65 //------------------------------------------------------------------------------ 66 __attribute__((constructor)) 67 static void Register(void) { 68 CeedRegister("/gpu/cuda/gen", CeedInit_Cuda_gen, 40); 69 } 70 //------------------------------------------------------------------------------ 71