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 <string.h> 18 #include <libxsmm.h> 19 #include "ceed-xsmm.h" 20 21 // Blocked Tensor Contract 22 static int CeedTensorContract_Xsmm_Blocked(CeedTensorContract contract, 23 CeedInt A, CeedInt B, CeedInt C, CeedInt J, const CeedScalar *restrict t, 24 CeedTransposeMode tmode, const CeedInt Add, const CeedScalar *restrict u, 25 CeedScalar *restrict v) { 26 CeedScalar alpha = 1.0, beta = 1.0; 27 char transu = 'N', transt = 'N'; 28 if (tmode == CEED_TRANSPOSE) 29 transt = 'T'; 30 31 if (!Add) 32 beta = 0.0; 33 34 for (CeedInt a=0; a<A; a++) 35 // libXSMM GEMM 36 libxsmm_dgemm(&transu, &transt, &C, &J, &B, 37 &alpha, &u[a*B*C], NULL, &t[0], NULL, 38 &beta, &v[a*J*C], NULL); 39 return 0; 40 } 41 42 // Serial Tensor Contact 43 static int CeedTensorContract_Xsmm_Serial(CeedTensorContract contract, 44 CeedInt A, CeedInt B, CeedInt C, CeedInt J, const CeedScalar *restrict t, 45 CeedTransposeMode tmode, const CeedInt Add, const CeedScalar *restrict u, 46 CeedScalar *restrict v) { 47 CeedScalar alpha = 1.0, beta = 1.0; 48 char transu = 'N', transt = 'N'; 49 if ((tmode == CEED_TRANSPOSE && C != 1) 50 || (tmode == CEED_NOTRANSPOSE && C == 1)) 51 transt = 'T'; 52 53 if (!Add) 54 beta = 0.0; 55 56 if (C != 1) 57 for (CeedInt a=0; a<A; a++) 58 // libXSMM GEMM 59 libxsmm_dgemm(&transu, &transt, &C, &J, &B, 60 &alpha, &u[a*B*C], NULL, &t[0], NULL, 61 &beta, &v[a*J*C], NULL); 62 else 63 // libXSMM GEMM 64 libxsmm_dgemm(&transt, &transu, &J, &A, &B, 65 &alpha, &t[0], NULL, &u[0], NULL, 66 &beta, &v[0], NULL); 67 68 return 0; 69 } 70 71 // Switch for Tensor Contract 72 static int CeedTensorContractApply_Xsmm(CeedTensorContract contract, CeedInt A, 73 CeedInt B, CeedInt C, CeedInt J, 74 const CeedScalar *restrict t, 75 CeedTransposeMode tmode, 76 const CeedInt Add, 77 const CeedScalar *restrict u, 78 CeedScalar *restrict v) { 79 CeedInt blksize = 8; 80 81 if (C % blksize) 82 CeedTensorContract_Xsmm_Serial(contract, A, B, C, J, t, tmode, Add, u, v); 83 else 84 CeedTensorContract_Xsmm_Blocked(contract, A, B, C, J, t, tmode, Add, u, v); 85 86 return 0; 87 } 88 89 static int CeedTensorContractDestroy_Xsmm(CeedTensorContract contract) { 90 return 0; 91 } 92 93 int CeedTensorContractCreate_Xsmm(CeedBasis basis, CeedTensorContract contract) { 94 int ierr; 95 Ceed ceed; 96 ierr = CeedTensorContractGetCeed(contract, &ceed); CeedChk(ierr); 97 98 ierr = CeedSetBackendFunction(ceed, "TensorContract", contract, "Apply", 99 CeedTensorContractApply_Xsmm); CeedChk(ierr); 100 ierr = CeedSetBackendFunction(ceed, "TensorContract", contract, "Destroy", 101 CeedTensorContractDestroy_Xsmm); CeedChk(ierr); 102 103 return 0; 104 } 105