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/ceed.h> 18 #include <ceed/backend.h> 19 #include "ceed-opt.h" 20 21 //------------------------------------------------------------------------------ 22 // Tensor Contract Core loop 23 //------------------------------------------------------------------------------ 24 static inline int CeedTensorContractApply_Core_Opt(CeedTensorContract contract, 25 CeedInt A, CeedInt B, CeedInt C, CeedInt J, const CeedScalar *restrict t, 26 CeedTransposeMode t_mode, const CeedInt add, const CeedScalar *restrict u, 27 CeedScalar *restrict v) { 28 CeedInt t_stride_0 = B, t_stride_1 = 1; 29 if (t_mode == CEED_TRANSPOSE) { 30 t_stride_0 = 1; t_stride_1 = J; 31 } 32 33 for (CeedInt a=0; a<A; a++) 34 for (CeedInt b=0; b<B; b++) 35 for (CeedInt j=0; j<J; j++) { 36 CeedScalar tq = t[j*t_stride_0 + b*t_stride_1]; 37 for (CeedInt c=0; c<C; c++) 38 v[(a*J+j)*C+c] += tq * u[(a*B+b)*C+c]; 39 } 40 41 return CEED_ERROR_SUCCESS; 42 } 43 44 //------------------------------------------------------------------------------ 45 // Tensor Contract Apply 46 //------------------------------------------------------------------------------ 47 static int CeedTensorContractApply_Opt(CeedTensorContract contract, CeedInt A, 48 CeedInt B, CeedInt C, CeedInt J, 49 const CeedScalar *restrict t, 50 CeedTransposeMode t_mode, const CeedInt add, 51 const CeedScalar *restrict u, 52 CeedScalar *restrict v) { 53 if (!add) 54 for (CeedInt q=0; q<A*J*C; q++) 55 v[q] = (CeedScalar) 0.0; 56 57 if (C == 1) 58 return CeedTensorContractApply_Core_Opt(contract, A, B, 1, J, t, t_mode, 59 add, u, v); 60 else 61 return CeedTensorContractApply_Core_Opt(contract, A, B, C, J, t, t_mode, 62 add, u, v); 63 64 return CEED_ERROR_SUCCESS; 65 } 66 67 //------------------------------------------------------------------------------ 68 // Tensor Contract Create 69 //------------------------------------------------------------------------------ 70 int CeedTensorContractCreate_Opt(CeedBasis basis, CeedTensorContract contract) { 71 int ierr; 72 Ceed ceed; 73 ierr = CeedTensorContractGetCeed(contract, &ceed); CeedChkBackend(ierr); 74 75 ierr = CeedSetBackendFunction(ceed, "TensorContract", contract, "Apply", 76 CeedTensorContractApply_Opt); CeedChkBackend(ierr); 77 78 return CEED_ERROR_SUCCESS; 79 } 80 //------------------------------------------------------------------------------ 81