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-impl.h> 18 #include <string.h> 19 #include "ceed-ref.h" 20 21 static int CeedElemRestrictionApply_Ref(CeedElemRestriction r, 22 CeedTransposeMode tmode, CeedInt ncomp, 23 CeedTransposeMode lmode, CeedVector u, 24 CeedVector v, CeedRequest *request) { 25 CeedElemRestriction_Ref *impl = r->data; 26 int ierr; 27 const CeedScalar *uu; 28 CeedScalar *vv; 29 CeedInt esize = r->nelem*r->elemsize; 30 31 ierr = CeedVectorGetArrayRead(u, CEED_MEM_HOST, &uu); CeedChk(ierr); 32 ierr = CeedVectorGetArray(v, CEED_MEM_HOST, &vv); CeedChk(ierr); 33 if (tmode == CEED_NOTRANSPOSE) { 34 // Perform: v = r * u 35 if (!impl->indices) { 36 for (CeedInt i=0; i<esize*ncomp; i++) vv[i] = uu[i]; 37 } else if (ncomp == 1) { 38 for (CeedInt i=0; i<esize; i++) vv[i] = uu[impl->indices[i]]; 39 } else { 40 // vv is (elemsize x ncomp x nelem), column-major 41 if (lmode == CEED_NOTRANSPOSE) { // u is (ndof x ncomp), column-major 42 for (CeedInt e = 0; e < r->nelem; e++) 43 for (CeedInt d = 0; d < ncomp; d++) 44 for (CeedInt i=0; i<r->elemsize; i++) { 45 vv[i+r->elemsize*(d+ncomp*e)] = 46 uu[impl->indices[i+r->elemsize*e]+r->ndof*d]; 47 } 48 } else { // u is (ncomp x ndof), column-major 49 for (CeedInt e = 0; e < r->nelem; e++) 50 for (CeedInt d = 0; d < ncomp; d++) 51 for (CeedInt i=0; i<r->elemsize; i++) { 52 vv[i+r->elemsize*(d+ncomp*e)] = 53 uu[d+ncomp*impl->indices[i+r->elemsize*e]]; 54 } 55 } 56 } 57 } else { 58 // Note: in transpose mode, we perform: v += r^t * u 59 if (!impl->indices) { 60 for (CeedInt i=0; i<esize; i++) vv[i] += uu[i]; 61 } else if (ncomp == 1) { 62 for (CeedInt i=0; i<esize; i++) vv[impl->indices[i]] += uu[i]; 63 } else { 64 // u is (elemsize x ncomp x nelem) 65 if (lmode == CEED_NOTRANSPOSE) { // vv is (ndof x ncomp), column-major 66 for (CeedInt e = 0; e < r->nelem; e++) 67 for (CeedInt d = 0; d < ncomp; d++) 68 for (CeedInt i=0; i<r->elemsize; i++) { 69 vv[impl->indices[i+r->elemsize*e]+r->ndof*d] += 70 uu[i+r->elemsize*(d+e*ncomp)]; 71 } 72 } else { // vv is (ncomp x ndof), column-major 73 for (CeedInt e = 0; e < r->nelem; e++) 74 for (CeedInt d = 0; d < ncomp; d++) 75 for (CeedInt i=0; i<r->elemsize; i++) { 76 vv[d+ncomp*impl->indices[i+r->elemsize*e]] += 77 uu[i+r->elemsize*(d+e*ncomp)]; 78 } 79 } 80 } 81 } 82 ierr = CeedVectorRestoreArrayRead(u, &uu); CeedChk(ierr); 83 ierr = CeedVectorRestoreArray(v, &vv); CeedChk(ierr); 84 if (request != CEED_REQUEST_IMMEDIATE && request != CEED_REQUEST_ORDERED) 85 *request = NULL; 86 return 0; 87 } 88 89 static int CeedElemRestrictionDestroy_Ref(CeedElemRestriction r) { 90 CeedElemRestriction_Ref *impl = r->data; 91 int ierr; 92 93 ierr = CeedFree(&impl->indices_allocated); CeedChk(ierr); 94 ierr = CeedFree(&r->data); CeedChk(ierr); 95 return 0; 96 } 97 98 int CeedElemRestrictionCreate_Ref(CeedElemRestriction r, 99 CeedMemType mtype, 100 CeedCopyMode cmode, const CeedInt *indices) { 101 int ierr; 102 CeedElemRestriction_Ref *impl; 103 104 if (mtype != CEED_MEM_HOST) 105 return CeedError(r->ceed, 1, "Only MemType = HOST supported"); 106 ierr = CeedCalloc(1,&impl); CeedChk(ierr); 107 switch (cmode) { 108 case CEED_COPY_VALUES: 109 ierr = CeedMalloc(r->nelem*r->elemsize, &impl->indices_allocated); 110 CeedChk(ierr); 111 memcpy(impl->indices_allocated, indices, 112 r->nelem * r->elemsize * sizeof(indices[0])); 113 impl->indices = impl->indices_allocated; 114 break; 115 case CEED_OWN_POINTER: 116 impl->indices_allocated = (CeedInt *)indices; 117 impl->indices = impl->indices_allocated; 118 break; 119 case CEED_USE_POINTER: 120 impl->indices = indices; 121 } 122 r->data = impl; 123 r->Apply = CeedElemRestrictionApply_Ref; 124 r->Destroy = CeedElemRestrictionDestroy_Ref; 125 return 0; 126 } 127