zplghe.c

Go to the documentation of this file.

#include "common.h"

/***************************************************************************/
int PLASMA_zplghe( double bump, int N,
                   PLASMA_Complex64_t *A, int LDA,
                   unsigned long long int seed )
{
    int NB;
    int status;
    plasma_context_t *plasma;
    PLASMA_sequence *sequence = NULL;
    PLASMA_request request = PLASMA_REQUEST_INITIALIZER;
    PLASMA_desc descA;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_zplghe", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    /* Check input arguments */
    if (N < 0) {
        plasma_error("PLASMA_zplghe", "illegal value of N");
        return -2;
    }
    if (LDA < max(1, N)) {
        plasma_error("PLASMA_zplghe", "illegal value of LDA");
        return -4;
    }
    /* Quick return */
    if (max(0, N) == 0)
        return PLASMA_SUCCESS;

    /* Tune NB depending on M, N & NRHS; Set NBNB */
    status = plasma_tune(PLASMA_FUNC_ZGEMM, N, N, 0);
    if (status != PLASMA_SUCCESS) {
        plasma_error("PLASMA_zplghe", "plasma_tune() failed");
        return status;
    }
    
    /* Set NT */
    NB = PLASMA_NB;
    plasma_sequence_create(plasma, &sequence);
    
    descA = plasma_desc_init(
        PlasmaComplexDouble, NB, NB, NB*NB,
        LDA, N, 0, 0, N, N);
    descA.mat = A;

    /* Call the tile interface */
    PLASMA_zplghe_Tile_Async( bump, &descA, seed, sequence, &request );

    plasma_ziptile2lap( descA, A, NB, NB, LDA, N );
    plasma_dynamic_sync();

    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);

    return status;
}

/***************************************************************************/
int PLASMA_zplghe_Tile( double bump, PLASMA_desc *A,
                        unsigned long long int seed )
{
    plasma_context_t *plasma;
    PLASMA_sequence *sequence = NULL;
    PLASMA_request request = PLASMA_REQUEST_INITIALIZER;
    int status;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_zplghe_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    plasma_sequence_create(plasma, &sequence);
    PLASMA_zplghe_Tile_Async( bump, A, seed, sequence, &request );
    plasma_dynamic_sync();
    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);
    return status;
}

/***************************************************************************/
int PLASMA_zplghe_Tile_Async( double          bump,
                              PLASMA_desc     *A,
                              unsigned long long int seed,
                              PLASMA_sequence *sequence, 
                              PLASMA_request  *request)
{
    PLASMA_desc descA = *A;
    plasma_context_t *plasma;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_zplghe_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    if (sequence == NULL) {
        plasma_fatal_error("PLASMA_zplghe_Tile", "NULL sequence");
        return PLASMA_ERR_UNALLOCATED;
    }
    if (request == NULL) {
        plasma_fatal_error("PLASMA_zplghe_Tile", "NULL request");
        return PLASMA_ERR_UNALLOCATED;
    }
    /* Check sequence status */
    if (sequence->status == PLASMA_SUCCESS)
        request->status = PLASMA_SUCCESS;
    else
        return plasma_request_fail(sequence, request, PLASMA_ERR_SEQUENCE_FLUSHED);

    /* Check descriptors for correctness */
    if (plasma_desc_check(&descA) != PLASMA_SUCCESS) {
        plasma_error("PLASMA_zplghe_Tile", "invalid descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    /* Check input arguments */
    if (descA.nb != descA.mb) {
        plasma_error("PLASMA_zplghe_Tile", "only square tiles supported");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }

    /* Quick return */
    if (min( descA.m, descA.n ) == 0)
        return PLASMA_SUCCESS;

    plasma_parallel_call_5(plasma_pzplghe,
        double,           bump,
        PLASMA_desc,      descA,
        unsigned long long int, seed,
        PLASMA_sequence*, sequence,
        PLASMA_request*,  request);

    return PLASMA_SUCCESS;
}