17d8d0e25Snbeams // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 27d8d0e25Snbeams // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 37d8d0e25Snbeams // All Rights reserved. See files LICENSE and NOTICE for details. 47d8d0e25Snbeams // 57d8d0e25Snbeams // This file is part of CEED, a collection of benchmarks, miniapps, software 67d8d0e25Snbeams // libraries and APIs for efficient high-order finite element and spectral 77d8d0e25Snbeams // element discretizations for exascale applications. For more information and 87d8d0e25Snbeams // source code availability see http://github.com/ceed. 97d8d0e25Snbeams // 107d8d0e25Snbeams // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 117d8d0e25Snbeams // a collaborative effort of two U.S. Department of Energy organizations (Office 127d8d0e25Snbeams // of Science and the National Nuclear Security Administration) responsible for 137d8d0e25Snbeams // the planning and preparation of a capable exascale ecosystem, including 147d8d0e25Snbeams // software, applications, hardware, advanced system engineering and early 157d8d0e25Snbeams // testbed platforms, in support of the nation's exascale computing imperative. 167d8d0e25Snbeams 17*3d576824SJeremy L Thompson #include <ceed.h> 18*3d576824SJeremy L Thompson #include <ceed-backend.h> 19*3d576824SJeremy L Thompson #include <hip/hip_runtime.h> 207d8d0e25Snbeams #include <stdio.h> 21*3d576824SJeremy L Thompson #include <string.h> 227d8d0e25Snbeams #include "ceed-hip-gen.h" 23*3d576824SJeremy L Thompson #include "../hip/ceed-hip.h" 247d8d0e25Snbeams 257d8d0e25Snbeams //------------------------------------------------------------------------------ 267d8d0e25Snbeams // Apply QFunction 277d8d0e25Snbeams //------------------------------------------------------------------------------ 287d8d0e25Snbeams static int CeedQFunctionApply_Hip_gen(CeedQFunction qf, CeedInt Q, 297d8d0e25Snbeams CeedVector *U, CeedVector *V) { 307d8d0e25Snbeams int ierr; 317d8d0e25Snbeams Ceed ceed; 327d8d0e25Snbeams ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChk(ierr); 337d8d0e25Snbeams return CeedError(ceed, 1, "Backend does not implement QFunctionApply"); 347d8d0e25Snbeams } 357d8d0e25Snbeams 367d8d0e25Snbeams //------------------------------------------------------------------------------ 377d8d0e25Snbeams // Destroy QFunction 387d8d0e25Snbeams //------------------------------------------------------------------------------ 397d8d0e25Snbeams static int CeedQFunctionDestroy_Hip_gen(CeedQFunction qf) { 407d8d0e25Snbeams int ierr; 417d8d0e25Snbeams CeedQFunction_Hip_gen *data; 427d8d0e25Snbeams ierr = CeedQFunctionGetData(qf, &data); CeedChk(ierr); 437d8d0e25Snbeams Ceed ceed; 447d8d0e25Snbeams ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChk(ierr); 457d8d0e25Snbeams ierr = hipFree(data->d_c); CeedChk_Hip(ceed, ierr); 467d8d0e25Snbeams ierr = CeedFree(&data->qFunctionSource); CeedChk(ierr); 477d8d0e25Snbeams ierr = CeedFree(&data); CeedChk(ierr); 487d8d0e25Snbeams return 0; 497d8d0e25Snbeams } 507d8d0e25Snbeams 517d8d0e25Snbeams //------------------------------------------------------------------------------ 527d8d0e25Snbeams // Load QFunction 537d8d0e25Snbeams //------------------------------------------------------------------------------ 547d8d0e25Snbeams static int loadHipFunction(CeedQFunction qf, char *c_src_file) { 557d8d0e25Snbeams int ierr; 567d8d0e25Snbeams Ceed ceed; 577d8d0e25Snbeams CeedQFunctionGetCeed(qf, &ceed); 587d8d0e25Snbeams CeedQFunction_Hip_gen *data; 597d8d0e25Snbeams ierr = CeedQFunctionGetData(qf, &data); CeedChk(ierr); 607d8d0e25Snbeams 617d8d0e25Snbeams // Find source file 627d8d0e25Snbeams char *hip_file; 637d8d0e25Snbeams ierr = CeedCalloc(HIP_MAX_PATH, &hip_file); CeedChk(ierr); 647d8d0e25Snbeams memcpy(hip_file, c_src_file, strlen(c_src_file)); 657d8d0e25Snbeams const char *last_dot = strrchr(hip_file, '.'); 667d8d0e25Snbeams if (!last_dot) 677d8d0e25Snbeams return CeedError(ceed, 1, "Cannot find file's extension!"); 687d8d0e25Snbeams const size_t hip_path_len = last_dot - hip_file; 697d8d0e25Snbeams strncpy(&hip_file[hip_path_len], ".h", 3); 707d8d0e25Snbeams 717d8d0e25Snbeams // Open source file 727d8d0e25Snbeams FILE *fp; 737d8d0e25Snbeams long lSize; 747d8d0e25Snbeams char *buffer; 757d8d0e25Snbeams fp = fopen (hip_file, "rb"); 767d8d0e25Snbeams if (!fp) 777d8d0e25Snbeams // LCOV_EXCL_START 787d8d0e25Snbeams CeedError(ceed, 1, "Couldn't open the Hip file for the QFunction."); 797d8d0e25Snbeams // LCOV_EXCL_STOP 807d8d0e25Snbeams 817d8d0e25Snbeams // Compute size of source file 827d8d0e25Snbeams fseek(fp, 0L, SEEK_END); 837d8d0e25Snbeams lSize = ftell(fp); 847d8d0e25Snbeams rewind(fp); 857d8d0e25Snbeams 867d8d0e25Snbeams // Allocate memory for entire content 877d8d0e25Snbeams ierr = CeedCalloc(lSize+1, &buffer); CeedChk(ierr); 887d8d0e25Snbeams 897d8d0e25Snbeams // Copy the file into the buffer 907d8d0e25Snbeams if (1 != fread(buffer, lSize, 1, fp)) { 917d8d0e25Snbeams // LCOV_EXCL_START 927d8d0e25Snbeams fclose(fp); 937d8d0e25Snbeams ierr = CeedFree(&buffer); CeedChk(ierr); 947d8d0e25Snbeams CeedError(ceed, 1, "Couldn't read the Hip file for the QFunction."); 957d8d0e25Snbeams // LCOV_EXCL_STOP 967d8d0e25Snbeams } 977d8d0e25Snbeams 987d8d0e25Snbeams // Append typedef and save source string 997d8d0e25Snbeams // FIXME: the magic number 16 should be defined somewhere... 1007d8d0e25Snbeams char *fields_string = 1017d8d0e25Snbeams "typedef struct { const CeedScalar* inputs[16]; CeedScalar* outputs[16]; } Fields_Hip_gen;"; 1027d8d0e25Snbeams ierr = CeedMalloc(1 + strlen(fields_string) + strlen(buffer), 1037d8d0e25Snbeams &data->qFunctionSource); CeedChk(ierr); 1047d8d0e25Snbeams memcpy(data->qFunctionSource, fields_string, strlen(fields_string)); 1057d8d0e25Snbeams memcpy(data->qFunctionSource + strlen(fields_string), buffer, 1067d8d0e25Snbeams strlen(buffer) + 1); 1077d8d0e25Snbeams 1087d8d0e25Snbeams // Cleanup 1097d8d0e25Snbeams ierr = CeedFree(&buffer); CeedChk(ierr); 1107d8d0e25Snbeams fclose(fp); 1117d8d0e25Snbeams ierr = CeedFree(&hip_file); CeedChk(ierr); 1127d8d0e25Snbeams return 0; 1137d8d0e25Snbeams } 1147d8d0e25Snbeams 1157d8d0e25Snbeams //------------------------------------------------------------------------------ 1167d8d0e25Snbeams // Create QFunction 1177d8d0e25Snbeams //------------------------------------------------------------------------------ 1187d8d0e25Snbeams int CeedQFunctionCreate_Hip_gen(CeedQFunction qf) { 1197d8d0e25Snbeams int ierr; 1207d8d0e25Snbeams Ceed ceed; 1217d8d0e25Snbeams CeedQFunctionGetCeed(qf, &ceed); 1227d8d0e25Snbeams CeedQFunction_Hip_gen *data; 1237d8d0e25Snbeams ierr = CeedCalloc(1, &data); CeedChk(ierr); 1247d8d0e25Snbeams ierr = CeedQFunctionSetData(qf, data); CeedChk(ierr); 1257d8d0e25Snbeams 1267d8d0e25Snbeams char *source; 1277d8d0e25Snbeams ierr = CeedQFunctionGetSourcePath(qf, &source); CeedChk(ierr); 1287d8d0e25Snbeams const char *funname = strrchr(source, ':') + 1; 1297d8d0e25Snbeams data->qFunctionName = (char *)funname; 1307d8d0e25Snbeams const int filenamelen = funname - source; 1317d8d0e25Snbeams char filename[filenamelen]; 1327d8d0e25Snbeams memcpy(filename, source, filenamelen - 1); 1337d8d0e25Snbeams filename[filenamelen - 1] = '\0'; 1347d8d0e25Snbeams ierr = loadHipFunction(qf, filename); CeedChk(ierr); 1357d8d0e25Snbeams 1367d8d0e25Snbeams ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Apply", 1377d8d0e25Snbeams CeedQFunctionApply_Hip_gen); CeedChk(ierr); 1387d8d0e25Snbeams ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Destroy", 1397d8d0e25Snbeams CeedQFunctionDestroy_Hip_gen); CeedChk(ierr); 1407d8d0e25Snbeams return 0; 1417d8d0e25Snbeams } 1427d8d0e25Snbeams //------------------------------------------------------------------------------ 143