184a01de5SJeremy L Thompson // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 284a01de5SJeremy L Thompson // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 384a01de5SJeremy L Thompson // All Rights reserved. See files LICENSE and NOTICE for details. 484a01de5SJeremy L Thompson // 584a01de5SJeremy L Thompson // This file is part of CEED, a collection of benchmarks, miniapps, software 684a01de5SJeremy L Thompson // libraries and APIs for efficient high-order finite element and spectral 784a01de5SJeremy L Thompson // element discretizations for exascale applications. For more information and 884a01de5SJeremy L Thompson // source code availability see http://github.com/ceed. 984a01de5SJeremy L Thompson // 1084a01de5SJeremy L Thompson // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 1184a01de5SJeremy L Thompson // a collaborative effort of two U.S. Department of Energy organizations (Office 1284a01de5SJeremy L Thompson // of Science and the National Nuclear Security Administration) responsible for 1384a01de5SJeremy L Thompson // the planning and preparation of a capable exascale ecosystem, including 1484a01de5SJeremy L Thompson // software, applications, hardware, advanced system engineering and early 1584a01de5SJeremy L Thompson // testbed platforms, in support of the nation's exascale computing imperative. 1684a01de5SJeremy L Thompson 17ec3da8bcSJed Brown #include <ceed/ceed.h> 18ec3da8bcSJed Brown #include <ceed/backend.h> 193d576824SJeremy L Thompson #include <stdbool.h> 203d576824SJeremy L Thompson #include <string.h> 2184a01de5SJeremy L Thompson #include "ceed-avx.h" 2284a01de5SJeremy L Thompson 23f10650afSjeremylt //------------------------------------------------------------------------------ 24f10650afSjeremylt // Backend Init 25f10650afSjeremylt //------------------------------------------------------------------------------ 2684a01de5SJeremy L Thompson static int CeedInit_Avx(const char *resource, Ceed ceed) { 2784a01de5SJeremy L Thompson int ierr; 28d1d35e2fSjeremylt if (strcmp(resource, "/cpu/self") && strcmp(resource, "/cpu/self/avx") && 29d1d35e2fSjeremylt strcmp(resource, "/cpu/self/avx/blocked")) 30c042f62fSJeremy L Thompson // LCOV_EXCL_START 31e15f9bd0SJeremy L Thompson return CeedError(ceed, CEED_ERROR_BACKEND, 32e15f9bd0SJeremy L Thompson "AVX backend cannot use resource: %s", resource); 33c042f62fSJeremy L Thompson // LCOV_EXCL_STOP 34e15f9bd0SJeremy L Thompson ierr = CeedSetDeterministic(ceed, true); CeedChkBackend(ierr); 3584a01de5SJeremy L Thompson 365f67fadeSJeremy L Thompson // Create reference CEED that implementation will be dispatched 3784a01de5SJeremy L Thompson // through unless overridden 38d1d35e2fSjeremylt Ceed ceed_ref; 39d1d35e2fSjeremylt CeedInit("/cpu/self/opt/blocked", &ceed_ref); 40d1d35e2fSjeremylt ierr = CeedSetDelegate(ceed, ceed_ref); CeedChkBackend(ierr); 4184a01de5SJeremy L Thompson 42*80a9ef05SNatalie Beams if (CEED_SCALAR_TYPE == CEED_SCALAR_FP64) { 432f86a920SJeremy L Thompson ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "TensorContractCreate", 44*80a9ef05SNatalie Beams CeedTensorContractCreate_f64_Avx); 45*80a9ef05SNatalie Beams CeedChkBackend(ierr); 46*80a9ef05SNatalie Beams } else { 47*80a9ef05SNatalie Beams ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "TensorContractCreate", 48*80a9ef05SNatalie Beams CeedTensorContractCreate_f32_Avx); 49*80a9ef05SNatalie Beams CeedChkBackend(ierr); 50*80a9ef05SNatalie Beams } 51*80a9ef05SNatalie Beams 52e15f9bd0SJeremy L Thompson return CEED_ERROR_SUCCESS; 5384a01de5SJeremy L Thompson } 5484a01de5SJeremy L Thompson 55f10650afSjeremylt //------------------------------------------------------------------------------ 56f10650afSjeremylt // Backend Register 57f10650afSjeremylt //------------------------------------------------------------------------------ 581d013790SJed Brown CEED_INTERN int CeedRegister_Avx_Blocked(void) { 591d013790SJed Brown return CeedRegister("/cpu/self/avx/blocked", CeedInit_Avx, 30); 6084a01de5SJeremy L Thompson } 61f10650afSjeremylt //------------------------------------------------------------------------------ 62