1*7d8d0e25Snbeams // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2*7d8d0e25Snbeams // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3*7d8d0e25Snbeams // All Rights reserved. See files LICENSE and NOTICE for details. 4*7d8d0e25Snbeams // 5*7d8d0e25Snbeams // This file is part of CEED, a collection of benchmarks, miniapps, software 6*7d8d0e25Snbeams // libraries and APIs for efficient high-order finite element and spectral 7*7d8d0e25Snbeams // element discretizations for exascale applications. For more information and 8*7d8d0e25Snbeams // source code availability see http://github.com/ceed. 9*7d8d0e25Snbeams // 10*7d8d0e25Snbeams // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11*7d8d0e25Snbeams // a collaborative effort of two U.S. Department of Energy organizations (Office 12*7d8d0e25Snbeams // of Science and the National Nuclear Security Administration) responsible for 13*7d8d0e25Snbeams // the planning and preparation of a capable exascale ecosystem, including 14*7d8d0e25Snbeams // software, applications, hardware, advanced system engineering and early 15*7d8d0e25Snbeams // testbed platforms, in support of the nation's exascale computing imperative. 16*7d8d0e25Snbeams 17*7d8d0e25Snbeams #include <string.h> 18*7d8d0e25Snbeams #include <stdio.h> 19*7d8d0e25Snbeams #include "ceed-hip-gen.h" 20*7d8d0e25Snbeams 21*7d8d0e25Snbeams //------------------------------------------------------------------------------ 22*7d8d0e25Snbeams // Apply QFunction 23*7d8d0e25Snbeams //------------------------------------------------------------------------------ 24*7d8d0e25Snbeams static int CeedQFunctionApply_Hip_gen(CeedQFunction qf, CeedInt Q, 25*7d8d0e25Snbeams CeedVector *U, CeedVector *V) { 26*7d8d0e25Snbeams int ierr; 27*7d8d0e25Snbeams Ceed ceed; 28*7d8d0e25Snbeams ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChk(ierr); 29*7d8d0e25Snbeams return CeedError(ceed, 1, "Backend does not implement QFunctionApply"); 30*7d8d0e25Snbeams } 31*7d8d0e25Snbeams 32*7d8d0e25Snbeams //------------------------------------------------------------------------------ 33*7d8d0e25Snbeams // Destroy QFunction 34*7d8d0e25Snbeams //------------------------------------------------------------------------------ 35*7d8d0e25Snbeams static int CeedQFunctionDestroy_Hip_gen(CeedQFunction qf) { 36*7d8d0e25Snbeams int ierr; 37*7d8d0e25Snbeams CeedQFunction_Hip_gen *data; 38*7d8d0e25Snbeams ierr = CeedQFunctionGetData(qf, &data); CeedChk(ierr); 39*7d8d0e25Snbeams Ceed ceed; 40*7d8d0e25Snbeams ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChk(ierr); 41*7d8d0e25Snbeams ierr = hipFree(data->d_c); CeedChk_Hip(ceed, ierr); 42*7d8d0e25Snbeams ierr = CeedFree(&data->qFunctionSource); CeedChk(ierr); 43*7d8d0e25Snbeams ierr = CeedFree(&data); CeedChk(ierr); 44*7d8d0e25Snbeams return 0; 45*7d8d0e25Snbeams } 46*7d8d0e25Snbeams 47*7d8d0e25Snbeams //------------------------------------------------------------------------------ 48*7d8d0e25Snbeams // Load QFunction 49*7d8d0e25Snbeams //------------------------------------------------------------------------------ 50*7d8d0e25Snbeams static int loadHipFunction(CeedQFunction qf, char *c_src_file) { 51*7d8d0e25Snbeams int ierr; 52*7d8d0e25Snbeams Ceed ceed; 53*7d8d0e25Snbeams CeedQFunctionGetCeed(qf, &ceed); 54*7d8d0e25Snbeams CeedQFunction_Hip_gen *data; 55*7d8d0e25Snbeams ierr = CeedQFunctionGetData(qf, &data); CeedChk(ierr); 56*7d8d0e25Snbeams 57*7d8d0e25Snbeams // Find source file 58*7d8d0e25Snbeams char *hip_file; 59*7d8d0e25Snbeams ierr = CeedCalloc(HIP_MAX_PATH, &hip_file); CeedChk(ierr); 60*7d8d0e25Snbeams memcpy(hip_file, c_src_file, strlen(c_src_file)); 61*7d8d0e25Snbeams const char *last_dot = strrchr(hip_file, '.'); 62*7d8d0e25Snbeams if (!last_dot) 63*7d8d0e25Snbeams return CeedError(ceed, 1, "Cannot find file's extension!"); 64*7d8d0e25Snbeams const size_t hip_path_len = last_dot - hip_file; 65*7d8d0e25Snbeams strncpy(&hip_file[hip_path_len], ".h", 3); 66*7d8d0e25Snbeams 67*7d8d0e25Snbeams // Open source file 68*7d8d0e25Snbeams FILE *fp; 69*7d8d0e25Snbeams long lSize; 70*7d8d0e25Snbeams char *buffer; 71*7d8d0e25Snbeams fp = fopen (hip_file, "rb"); 72*7d8d0e25Snbeams if (!fp) 73*7d8d0e25Snbeams // LCOV_EXCL_START 74*7d8d0e25Snbeams CeedError(ceed, 1, "Couldn't open the Hip file for the QFunction."); 75*7d8d0e25Snbeams // LCOV_EXCL_STOP 76*7d8d0e25Snbeams 77*7d8d0e25Snbeams // Compute size of source file 78*7d8d0e25Snbeams fseek(fp, 0L, SEEK_END); 79*7d8d0e25Snbeams lSize = ftell(fp); 80*7d8d0e25Snbeams rewind(fp); 81*7d8d0e25Snbeams 82*7d8d0e25Snbeams // Allocate memory for entire content 83*7d8d0e25Snbeams ierr = CeedCalloc(lSize+1, &buffer); CeedChk(ierr); 84*7d8d0e25Snbeams 85*7d8d0e25Snbeams // Copy the file into the buffer 86*7d8d0e25Snbeams if (1 != fread(buffer, lSize, 1, fp)) { 87*7d8d0e25Snbeams // LCOV_EXCL_START 88*7d8d0e25Snbeams fclose(fp); 89*7d8d0e25Snbeams ierr = CeedFree(&buffer); CeedChk(ierr); 90*7d8d0e25Snbeams CeedError(ceed, 1, "Couldn't read the Hip file for the QFunction."); 91*7d8d0e25Snbeams // LCOV_EXCL_STOP 92*7d8d0e25Snbeams } 93*7d8d0e25Snbeams 94*7d8d0e25Snbeams // Append typedef and save source string 95*7d8d0e25Snbeams // FIXME: the magic number 16 should be defined somewhere... 96*7d8d0e25Snbeams char *fields_string = 97*7d8d0e25Snbeams "typedef struct { const CeedScalar* inputs[16]; CeedScalar* outputs[16]; } Fields_Hip_gen;"; 98*7d8d0e25Snbeams ierr = CeedMalloc(1 + strlen(fields_string) + strlen(buffer), 99*7d8d0e25Snbeams &data->qFunctionSource); CeedChk(ierr); 100*7d8d0e25Snbeams memcpy(data->qFunctionSource, fields_string, strlen(fields_string)); 101*7d8d0e25Snbeams memcpy(data->qFunctionSource + strlen(fields_string), buffer, 102*7d8d0e25Snbeams strlen(buffer) + 1); 103*7d8d0e25Snbeams 104*7d8d0e25Snbeams // Cleanup 105*7d8d0e25Snbeams ierr = CeedFree(&buffer); CeedChk(ierr); 106*7d8d0e25Snbeams fclose(fp); 107*7d8d0e25Snbeams ierr = CeedFree(&hip_file); CeedChk(ierr); 108*7d8d0e25Snbeams return 0; 109*7d8d0e25Snbeams } 110*7d8d0e25Snbeams 111*7d8d0e25Snbeams //------------------------------------------------------------------------------ 112*7d8d0e25Snbeams // Create QFunction 113*7d8d0e25Snbeams //------------------------------------------------------------------------------ 114*7d8d0e25Snbeams int CeedQFunctionCreate_Hip_gen(CeedQFunction qf) { 115*7d8d0e25Snbeams int ierr; 116*7d8d0e25Snbeams Ceed ceed; 117*7d8d0e25Snbeams CeedQFunctionGetCeed(qf, &ceed); 118*7d8d0e25Snbeams CeedQFunction_Hip_gen *data; 119*7d8d0e25Snbeams ierr = CeedCalloc(1, &data); CeedChk(ierr); 120*7d8d0e25Snbeams ierr = CeedQFunctionSetData(qf, data); CeedChk(ierr); 121*7d8d0e25Snbeams 122*7d8d0e25Snbeams char *source; 123*7d8d0e25Snbeams ierr = CeedQFunctionGetSourcePath(qf, &source); CeedChk(ierr); 124*7d8d0e25Snbeams const char *funname = strrchr(source, ':') + 1; 125*7d8d0e25Snbeams data->qFunctionName = (char *)funname; 126*7d8d0e25Snbeams const int filenamelen = funname - source; 127*7d8d0e25Snbeams char filename[filenamelen]; 128*7d8d0e25Snbeams memcpy(filename, source, filenamelen - 1); 129*7d8d0e25Snbeams filename[filenamelen - 1] = '\0'; 130*7d8d0e25Snbeams ierr = loadHipFunction(qf, filename); CeedChk(ierr); 131*7d8d0e25Snbeams 132*7d8d0e25Snbeams ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Apply", 133*7d8d0e25Snbeams CeedQFunctionApply_Hip_gen); CeedChk(ierr); 134*7d8d0e25Snbeams ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Destroy", 135*7d8d0e25Snbeams CeedQFunctionDestroy_Hip_gen); CeedChk(ierr); 136*7d8d0e25Snbeams return 0; 137*7d8d0e25Snbeams } 138*7d8d0e25Snbeams //------------------------------------------------------------------------------ 139