Mentions légales du service

Skip to content
Snippets Groups Projects
codelet_zsymm.c 4.21 KiB
Newer Older
 * @file starpu/codelet_zsymm.c
Mathieu Faverge's avatar
Mathieu Faverge committed
 * @copyright 2009-2014 The University of Tennessee and The University of
 *                      Tennessee Research Foundation. All rights reserved.
 * @copyright 2012-2020 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
 *                      Univ. Bordeaux. All rights reserved.
 * @brief Chameleon zsymm StarPU codelet
 * @author Hatem Ltaief
 * @author Jakub Kurzak
 * @author Mathieu Faverge
 * @author Emmanuel Agullo
 * @author Cedric Castagnede
BARROS DE ASSIS Lucas's avatar
BARROS DE ASSIS Lucas committed
 * @author Lucas Barros de Assis
#include "chameleon_starpu.h"
#include "runtime_codelet_z.h"
#if !defined(CHAMELEON_SIMULATION)
static void cl_zsymm_cpu_func(void *descr[], void *cl_arg)
{
    cham_side_t side;
    cham_uplo_t uplo;
    CHAMELEON_Complex64_t alpha;
    CHAMELEON_Complex64_t beta;
    CHAM_tile_t *tileC;

    tileA = cti_interface_get(descr[0]);
    tileB = cti_interface_get(descr[1]);
    tileC = cti_interface_get(descr[2]);
BARROS DE ASSIS Lucas's avatar
BARROS DE ASSIS Lucas committed
    starpu_codelet_unpack_args(cl_arg, &side, &uplo, &M, &N, &alpha, &beta);
#ifdef CHAMELEON_USE_CUDA
static void cl_zsymm_cuda_func(void *descr[], void *cl_arg)
{
    cham_side_t side;
    cham_uplo_t uplo;
    int M;
    int N;
    cuDoubleComplex alpha;
    cuDoubleComplex beta;
    CHAM_tile_t *tileC;

    tileA = cti_interface_get(descr[0]);
    tileB = cti_interface_get(descr[1]);
    tileC = cti_interface_get(descr[2]);
BARROS DE ASSIS Lucas's avatar
BARROS DE ASSIS Lucas committed
    starpu_codelet_unpack_args(cl_arg, &side, &uplo, &M, &N, &alpha, &beta);
Mathieu Faverge's avatar
Mathieu Faverge committed
    RUNTIME_getStream(stream);
        &alpha, tileA->mat, tileA->ld,
                tileB->mat, tileB->ld,
        &beta,  tileC->mat, tileC->ld,

#ifndef STARPU_CUDA_ASYNC
    cudaStreamSynchronize( stream );
#endif

    return;
}
#endif /* CHAMELEON_USE_CUDA */
#endif /* !defined(CHAMELEON_SIMULATION) */
CODELETS(zsymm, cl_zsymm_cpu_func, cl_zsymm_cuda_func, STARPU_CUDA_ASYNC)

/**
 *
 * @ingroup INSERT_TASK_Complex64_t
 *
 */
void INSERT_TASK_zsymm(const RUNTIME_option_t *options,
                      cham_side_t side, cham_uplo_t uplo,
                      int m, int n, int nb,
                      CHAMELEON_Complex64_t alpha, const CHAM_desc_t *A, int Am, int An,
                      const CHAM_desc_t *B, int Bm, int Bn,
                      CHAMELEON_Complex64_t beta, const CHAM_desc_t *C, int Cm, int Cn)
    if ( alpha == 0. ) {
        return INSERT_TASK_zlascal( options, ChamUpperLower, m, n, nb,
                                    beta, C, Cm, Cn );
    }

    (void)nb;
    struct starpu_codelet *codelet = &cl_zsymm;
    void (*callback)(void*) = options->profiling ? cl_zsymm_callback : NULL;
    starpu_option_request_t* schedopt = (starpu_option_request_t *)(options->request->schedopt);
    int workerid = (schedopt == NULL) ? -1 : schedopt->workerid;
    int accessC = ( beta == 0. ) ? STARPU_W : STARPU_RW;

    CHAMELEON_BEGIN_ACCESS_DECLARATION;
    CHAMELEON_ACCESS_R(A, Am, An);
    CHAMELEON_ACCESS_R(B, Bm, Bn);
    CHAMELEON_ACCESS_RW(C, Cm, Cn);
    CHAMELEON_END_ACCESS_DECLARATION;

    starpu_insert_task(
        starpu_mpi_codelet(codelet),
        STARPU_VALUE,    &side,                sizeof(int),
        STARPU_VALUE,    &uplo,                sizeof(int),
        STARPU_VALUE,       &m,                        sizeof(int),
        STARPU_VALUE,       &n,                        sizeof(int),
        STARPU_VALUE,   &alpha,         sizeof(CHAMELEON_Complex64_t),
        STARPU_R,               RTBLKADDR(A, CHAMELEON_Complex64_t, Am, An),
        STARPU_R,               RTBLKADDR(B, CHAMELEON_Complex64_t, Bm, Bn),
        STARPU_VALUE,    &beta,         sizeof(CHAMELEON_Complex64_t),
        accessC,                RTBLKADDR(C, CHAMELEON_Complex64_t, Cm, Cn),
        STARPU_PRIORITY,    options->priority,
        STARPU_CALLBACK,    callback,
#if defined(CHAMELEON_CODELETS_HAVE_NAME)
        STARPU_NAME, "zsymm",
#endif
        0);
}