/** * * @copyright (c) 2009-2014 The University of Tennessee and The University * of Tennessee Research Foundation. * All rights reserved. * @copyright (c) 2012-2014 Inria. All rights reserved. * @copyright (c) 2012-2014 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria, Univ. Bordeaux. All rights reserved. * **/ /** * * @file codelet_zhemm.c * * MORSE codelets kernel * MORSE is a software package provided by Univ. of Tennessee, * Univ. of California Berkeley and Univ. of Colorado Denver * * @version 2.5.0 * @comment This file has been automatically generated * from Plasma 2.5.0 for MORSE 1.0.0 * @author Hatem Ltaief * @author Jakub Kurzak * @author Mathieu Faverge * @author Emmanuel Agullo * @author Cedric Castagnede * @date 2010-11-15 * @precisions normal z -> c * **/ #include "morse_starpu.h" #include "codelet_z.h" /** * * @ingroup CORE_MORSE_Complex64_t * **/ void MORSE_TASK_zhemm(MORSE_option_t *options, MORSE_enum side, MORSE_enum uplo, int m, int n, int nb, MORSE_Complex64_t alpha, MORSE_desc_t *A, int Am, int An, int lda, MORSE_desc_t *B, int Bm, int Bn, int ldb, MORSE_Complex64_t beta, MORSE_desc_t *C, int Cm, int Cn, int ldc) { (void)nb; struct starpu_codelet *codelet = &cl_zhemm; void (*callback)(void*) = options->profiling ? cl_zhemm_callback : NULL; if ( morse_desc_islocal( A, Am, An ) || morse_desc_islocal( B, Bm, Bn ) || morse_desc_islocal( C, Cm, Cn ) ) { starpu_insert_task( codelet, STARPU_VALUE, &side, sizeof(MORSE_enum), STARPU_VALUE, &uplo, sizeof(MORSE_enum), STARPU_VALUE, &m, sizeof(int), STARPU_VALUE, &n, sizeof(int), STARPU_VALUE, &alpha, sizeof(MORSE_Complex64_t), STARPU_R, RTBLKADDR(A, MORSE_Complex64_t, Am, An), STARPU_VALUE, &lda, sizeof(int), STARPU_R, RTBLKADDR(B, MORSE_Complex64_t, Bm, Bn), STARPU_VALUE, &ldb, sizeof(int), STARPU_VALUE, &beta, sizeof(MORSE_Complex64_t), STARPU_RW, RTBLKADDR(C, MORSE_Complex64_t, Cm, Cn), STARPU_VALUE, &ldc, sizeof(int), STARPU_PRIORITY, options->priority, STARPU_CALLBACK, callback, 0); } } static void cl_zhemm_cpu_func(void *descr[], void *cl_arg) { MORSE_enum side; MORSE_enum uplo; int M; int N; MORSE_Complex64_t alpha; MORSE_Complex64_t *A; int LDA; MORSE_Complex64_t *B; int LDB; MORSE_Complex64_t beta; MORSE_Complex64_t *C; int LDC; A = (MORSE_Complex64_t *)STARPU_MATRIX_GET_PTR(descr[0]); B = (MORSE_Complex64_t *)STARPU_MATRIX_GET_PTR(descr[1]); C = (MORSE_Complex64_t *)STARPU_MATRIX_GET_PTR(descr[2]); starpu_codelet_unpack_args(cl_arg, &side, &uplo, &M, &N, &alpha, &LDA, &LDB, &beta, &LDC); cblas_zhemm( CblasColMajor, (CBLAS_SIDE)side, (CBLAS_UPLO)uplo, M, N, CBLAS_SADDR(alpha), A, LDA, B, LDB, CBLAS_SADDR(beta), C, LDC); } #ifdef CHAMELEON_USE_CUDA static void cl_zhemm_cuda_func(void *descr[], void *cl_arg) { MORSE_enum side; MORSE_enum uplo; int M; int N; cuDoubleComplex alpha; cuDoubleComplex *A; int LDA; cuDoubleComplex *B; int LDB; cuDoubleComplex beta; cuDoubleComplex *C; int LDC; A = (cuDoubleComplex *)STARPU_MATRIX_GET_PTR(descr[0]); B = (cuDoubleComplex *)STARPU_MATRIX_GET_PTR(descr[1]); C = (cuDoubleComplex *)STARPU_MATRIX_GET_PTR(descr[2]); starpu_codelet_unpack_args(cl_arg, &side, &uplo, &M, &N, &alpha, &LDA, &LDB, &beta, &LDC); CUstream stream = starpu_cuda_get_local_stream(); cublasSetKernelStream( stream ); cublasZhemm( morse_lapack_const(side), morse_lapack_const(uplo), M, N, alpha, A, LDA, B, LDB, beta, C, LDC); #ifndef STARPU_CUDA_ASYNC cudaStreamSynchronize( stream ); #endif return; } #endif /* * Codelet definition */ CODELETS(zhemm, 3, cl_zhemm_cpu_func, cl_zhemm_cuda_func, STARPU_CUDA_ASYNC)