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 17ec3da8bcSJed Brown #include <ceed/ceed.h> 18ec3da8bcSJed Brown #include <ceed/backend.h> 193d576824SJeremy L Thompson #include <hip/hip_runtime.h> 207d8d0e25Snbeams #include <stdio.h> 213d576824SJeremy L Thompson #include <string.h> 227d8d0e25Snbeams #include "ceed-hip-gen.h" 233d576824SJeremy 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; 32e15f9bd0SJeremy L Thompson ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChkBackend(ierr); 33e15f9bd0SJeremy L Thompson return CeedError(ceed, CEED_ERROR_BACKEND, 34e15f9bd0SJeremy L Thompson "Backend does not implement QFunctionApply"); 357d8d0e25Snbeams } 367d8d0e25Snbeams 377d8d0e25Snbeams //------------------------------------------------------------------------------ 387d8d0e25Snbeams // Destroy QFunction 397d8d0e25Snbeams //------------------------------------------------------------------------------ 407d8d0e25Snbeams static int CeedQFunctionDestroy_Hip_gen(CeedQFunction qf) { 417d8d0e25Snbeams int ierr; 427d8d0e25Snbeams CeedQFunction_Hip_gen *data; 43e15f9bd0SJeremy L Thompson ierr = CeedQFunctionGetData(qf, &data); CeedChkBackend(ierr); 447d8d0e25Snbeams Ceed ceed; 45e15f9bd0SJeremy L Thompson ierr = CeedQFunctionGetCeed(qf, &ceed); CeedChkBackend(ierr); 467d8d0e25Snbeams ierr = hipFree(data->d_c); CeedChk_Hip(ceed, ierr); 47e15f9bd0SJeremy L Thompson ierr = CeedFree(&data->qFunctionSource); CeedChkBackend(ierr); 48e15f9bd0SJeremy L Thompson ierr = CeedFree(&data); CeedChkBackend(ierr); 49e15f9bd0SJeremy L Thompson return CEED_ERROR_SUCCESS; 507d8d0e25Snbeams } 517d8d0e25Snbeams 527d8d0e25Snbeams //------------------------------------------------------------------------------ 537d8d0e25Snbeams // Load QFunction 547d8d0e25Snbeams //------------------------------------------------------------------------------ 557d8d0e25Snbeams static int loadHipFunction(CeedQFunction qf, char *c_src_file) { 567d8d0e25Snbeams int ierr; 577d8d0e25Snbeams Ceed ceed; 587d8d0e25Snbeams CeedQFunctionGetCeed(qf, &ceed); 597d8d0e25Snbeams CeedQFunction_Hip_gen *data; 60e15f9bd0SJeremy L Thompson ierr = CeedQFunctionGetData(qf, &data); CeedChkBackend(ierr); 617d8d0e25Snbeams 627d8d0e25Snbeams // Find source file 637d8d0e25Snbeams char *hip_file; 64e15f9bd0SJeremy L Thompson ierr = CeedCalloc(HIP_MAX_PATH, &hip_file); CeedChkBackend(ierr); 657d8d0e25Snbeams memcpy(hip_file, c_src_file, strlen(c_src_file)); 667d8d0e25Snbeams const char *last_dot = strrchr(hip_file, '.'); 677d8d0e25Snbeams if (!last_dot) 687d8d0e25Snbeams return CeedError(ceed, 1, "Cannot find file's extension!"); 697d8d0e25Snbeams const size_t hip_path_len = last_dot - hip_file; 707d8d0e25Snbeams strncpy(&hip_file[hip_path_len], ".h", 3); 717d8d0e25Snbeams 727d8d0e25Snbeams // Open source file 737d8d0e25Snbeams FILE *fp; 747d8d0e25Snbeams long lSize; 757d8d0e25Snbeams char *buffer; 767d8d0e25Snbeams fp = fopen (hip_file, "rb"); 777d8d0e25Snbeams if (!fp) 787d8d0e25Snbeams // LCOV_EXCL_START 7942e449dbSjeremylt return CeedError(ceed, 1, "Couldn't open the Hip file for the QFunction."); 807d8d0e25Snbeams // LCOV_EXCL_STOP 817d8d0e25Snbeams 827d8d0e25Snbeams // Compute size of source file 837d8d0e25Snbeams fseek(fp, 0L, SEEK_END); 847d8d0e25Snbeams lSize = ftell(fp); 857d8d0e25Snbeams rewind(fp); 867d8d0e25Snbeams 877d8d0e25Snbeams // Allocate memory for entire content 88e15f9bd0SJeremy L Thompson ierr = CeedCalloc(lSize+1, &buffer); CeedChkBackend(ierr); 897d8d0e25Snbeams 907d8d0e25Snbeams // Copy the file into the buffer 917d8d0e25Snbeams if (1 != fread(buffer, lSize, 1, fp)) { 927d8d0e25Snbeams // LCOV_EXCL_START 937d8d0e25Snbeams fclose(fp); 94e15f9bd0SJeremy L Thompson ierr = CeedFree(&buffer); CeedChkBackend(ierr); 9542e449dbSjeremylt return CeedError(ceed, 1, "Couldn't read the Hip file for the QFunction."); 967d8d0e25Snbeams // LCOV_EXCL_STOP 977d8d0e25Snbeams } 987d8d0e25Snbeams 997d8d0e25Snbeams // Append typedef and save source string 1007d8d0e25Snbeams // FIXME: the magic number 16 should be defined somewhere... 1017d8d0e25Snbeams char *fields_string = 1027d8d0e25Snbeams "typedef struct { const CeedScalar* inputs[16]; CeedScalar* outputs[16]; } Fields_Hip_gen;"; 1037d8d0e25Snbeams ierr = CeedMalloc(1 + strlen(fields_string) + strlen(buffer), 104e15f9bd0SJeremy L Thompson &data->qFunctionSource); CeedChkBackend(ierr); 1057d8d0e25Snbeams memcpy(data->qFunctionSource, fields_string, strlen(fields_string)); 1067d8d0e25Snbeams memcpy(data->qFunctionSource + strlen(fields_string), buffer, 1077d8d0e25Snbeams strlen(buffer) + 1); 1087d8d0e25Snbeams 1097d8d0e25Snbeams // Cleanup 110e15f9bd0SJeremy L Thompson ierr = CeedFree(&buffer); CeedChkBackend(ierr); 1117d8d0e25Snbeams fclose(fp); 112e15f9bd0SJeremy L Thompson ierr = CeedFree(&hip_file); CeedChkBackend(ierr); 113e15f9bd0SJeremy L Thompson return CEED_ERROR_SUCCESS; 1147d8d0e25Snbeams } 1157d8d0e25Snbeams 1167d8d0e25Snbeams //------------------------------------------------------------------------------ 1177d8d0e25Snbeams // Create QFunction 1187d8d0e25Snbeams //------------------------------------------------------------------------------ 1197d8d0e25Snbeams int CeedQFunctionCreate_Hip_gen(CeedQFunction qf) { 1207d8d0e25Snbeams int ierr; 1217d8d0e25Snbeams Ceed ceed; 1227d8d0e25Snbeams CeedQFunctionGetCeed(qf, &ceed); 1237d8d0e25Snbeams CeedQFunction_Hip_gen *data; 124e15f9bd0SJeremy L Thompson ierr = CeedCalloc(1, &data); CeedChkBackend(ierr); 125e15f9bd0SJeremy L Thompson ierr = CeedQFunctionSetData(qf, data); CeedChkBackend(ierr); 1267d8d0e25Snbeams 127*43e1b16fSJeremy L Thompson // Read source 1287d8d0e25Snbeams char *source; 129e15f9bd0SJeremy L Thompson ierr = CeedQFunctionGetSourcePath(qf, &source); CeedChkBackend(ierr); 130*43e1b16fSJeremy L Thompson // Empty source path indicates user must supply Q-Function 131*43e1b16fSJeremy L Thompson if (source[0] != '\0') { 132*43e1b16fSJeremy L Thompson ierr = CeedQFunctionGetKernelName(qf, &data->qFunctionName); 133*43e1b16fSJeremy L Thompson CeedChkBackend(ierr); 134*43e1b16fSJeremy L Thompson ierr = loadHipFunction(qf, source); CeedChkBackend(ierr); 135*43e1b16fSJeremy L Thompson } 1367d8d0e25Snbeams 1377d8d0e25Snbeams ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Apply", 138e15f9bd0SJeremy L Thompson CeedQFunctionApply_Hip_gen); CeedChkBackend(ierr); 1397d8d0e25Snbeams ierr = CeedSetBackendFunction(ceed, "QFunction", qf, "Destroy", 140e15f9bd0SJeremy L Thompson CeedQFunctionDestroy_Hip_gen); CeedChkBackend(ierr); 141e15f9bd0SJeremy L Thompson return CEED_ERROR_SUCCESS; 1427d8d0e25Snbeams } 1437d8d0e25Snbeams //------------------------------------------------------------------------------ 144