17f5b9731SStan Tomov // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 27f5b9731SStan Tomov // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 37f5b9731SStan Tomov // All Rights reserved. See files LICENSE and NOTICE for details. 482b77998SStan Tomov // 582b77998SStan Tomov // This file is part of CEED, a collection of benchmarks, miniapps, software 682b77998SStan Tomov // libraries and APIs for efficient high-order finite element and spectral 782b77998SStan Tomov // element discretizations for exascale applications. For more information and 882b77998SStan Tomov // source code availability see http://github.com/ceed. 982b77998SStan Tomov // 1082b77998SStan Tomov // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 1182b77998SStan Tomov // a collaborative effort of two U.S. Department of Energy organizations (Office 1282b77998SStan Tomov // of Science and the National Nuclear Security Administration) responsible for 1382b77998SStan Tomov // the planning and preparation of a capable exascale ecosystem, including 1482b77998SStan Tomov // software, applications, hardware, advanced system engineering and early 1582b77998SStan Tomov // testbed platforms, in support of the nation's exascale computing imperative. 1682b77998SStan Tomov 1738612b08SStan Tomov #include "ceed-magma.h" 1852d6035fSJeremy L Thompson 19*e0582403Sabdelfattah83 static int CeedDestroy_Magma(Ceed ceed) { 20*e0582403Sabdelfattah83 int ierr; 21*e0582403Sabdelfattah83 Ceed_Magma *data; 22*e0582403Sabdelfattah83 ierr = CeedGetData(ceed, (void *)&data); CeedChk(ierr); 23*e0582403Sabdelfattah83 magma_queue_destroy( data->queue ); 24*e0582403Sabdelfattah83 ierr = CeedFree(&data); CeedChk(ierr); 25*e0582403Sabdelfattah83 return 0; 26*e0582403Sabdelfattah83 } 27*e0582403Sabdelfattah83 2882b77998SStan Tomov static int CeedInit_Magma(const char *resource, Ceed ceed) { 291dc2661bSVeselin Dobrev int ierr; 302a847359SStan Tomov if (strcmp(resource, "/gpu/magma")) 317f5b9731SStan Tomov // LCOV_EXCL_START 327f5b9731SStan Tomov return CeedError(ceed, 1, "Magma backend cannot use resource: %s", resource); 337f5b9731SStan Tomov // LCOV_EXCL_STOP 347f5b9731SStan Tomov 357f5b9731SStan Tomov 367f5b9731SStan Tomov // Create refrence CEED that implementation will be dispatched 377f5b9731SStan Tomov // through unless overridden 387f5b9731SStan Tomov Ceed ceedref; 397f5b9731SStan Tomov CeedInit("/gpu/cuda/ref", &ceedref); 407f5b9731SStan Tomov ierr = CeedSetDelegate(ceed, ceedref); CeedChk(ierr); 4193fbe329SStan Tomov 421dc2661bSVeselin Dobrev ierr = magma_init(); 437f5b9731SStan Tomov if (ierr) 447f5b9731SStan Tomov // LCOV_EXCL_START 457f5b9731SStan Tomov return CeedError(ceed, 1, "error in magma_init(): %d\n", ierr); 467f5b9731SStan Tomov // LCOV_EXCL_STOP 4793fbe329SStan Tomov 48*e0582403Sabdelfattah83 Ceed_Magma *data; 49*e0582403Sabdelfattah83 ierr = CeedCalloc(sizeof(Ceed_Magma), &data); CeedChk(ierr); 50*e0582403Sabdelfattah83 ierr = CeedSetData(ceed, (void *)&data); CeedChk(ierr); 51*e0582403Sabdelfattah83 52*e0582403Sabdelfattah83 // kernel selection 53*e0582403Sabdelfattah83 data->basis_kernel_mode = MAGMA_KERNEL_DIM_SPECIFIC; 54*e0582403Sabdelfattah83 55*e0582403Sabdelfattah83 // kernel max threads per thread-block 56*e0582403Sabdelfattah83 data->maxthreads[0] = 128; // for 1D kernels 57*e0582403Sabdelfattah83 data->maxthreads[1] = 128; // for 2D kernels 58*e0582403Sabdelfattah83 data->maxthreads[2] = 64; // for 3D kernels 59*e0582403Sabdelfattah83 60*e0582403Sabdelfattah83 // create a queue that uses the null stream 61*e0582403Sabdelfattah83 magma_getdevice( &(data->device) ); 62*e0582403Sabdelfattah83 magma_queue_create_from_cuda(data->device, NULL, NULL, NULL, &(data->queue)); 63*e0582403Sabdelfattah83 64868539c2SNatalie Beams ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "ElemRestrictionCreate", 65868539c2SNatalie Beams CeedElemRestrictionCreate_Magma); CeedChk(ierr); 66868539c2SNatalie Beams ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, 67868539c2SNatalie Beams "ElemRestrictionCreateBlocked", 68868539c2SNatalie Beams CeedElemRestrictionCreateBlocked_Magma); CeedChk(ierr); 697f5b9731SStan Tomov ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "BasisCreateTensorH1", 707f5b9731SStan Tomov CeedBasisCreateTensorH1_Magma); CeedChk(ierr); 717f5b9731SStan Tomov ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "BasisCreateH1", 727f5b9731SStan Tomov CeedBasisCreateH1_Magma); CeedChk(ierr); 73a8c028e3SNatalie Beams ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "OperatorCreate", 74a8c028e3SNatalie Beams CeedOperatorCreate_Magma); CeedChk(ierr); 75*e0582403Sabdelfattah83 ierr = CeedSetBackendFunction(ceed, "Ceed", ceed, "Destroy", 76*e0582403Sabdelfattah83 CeedDestroy_Magma); CeedChk(ierr); 7782b77998SStan Tomov return 0; 7882b77998SStan Tomov } 7982b77998SStan Tomov 8082b77998SStan Tomov __attribute__((constructor)) 8182b77998SStan Tomov static void Register(void) { 8244951fc6Sjeremylt CeedRegister("/gpu/magma", CeedInit_Magma,20); 8382b77998SStan Tomov } 847f5b9731SStan Tomov 85