strsmpl.c

Go to the documentation of this file.

#include "common.h"

/***************************************************************************/
int PLASMA_strsmpl(int N, int NRHS,
                   float *A, int LDA,
                   float *L, int *IPIV,
                   float *B, int LDB)
{
    int NB, IB, IBNB, NT;
    int status;
    plasma_context_t *plasma;
    PLASMA_sequence *sequence = NULL;
    PLASMA_request request = PLASMA_REQUEST_INITIALIZER;
    PLASMA_desc descA, descB, descL;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_strsmpl", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    /* Check input arguments */
    if (N < 0) {
        plasma_error("PLASMA_strsmpl", "illegal value of N");
        return -1;
    }
    if (NRHS < 0) {
        plasma_error("PLASMA_strsmpl", "illegal value of NRHS");
        return -2;
    }
    if (LDA < max(1, N)) {
        plasma_error("PLASMA_strsmpl", "illegal value of LDA");
        return -4;
    }
    if (LDB < max(1, N)) {
        plasma_error("PLASMA_strsmpl", "illegal value of LDB");
        return -8;
    }
    /* Quick return */
    if (min(N, NRHS) == 0)
        return PLASMA_SUCCESS;

    /* Tune NB & IB depending on N & NRHS; Set NBNB */
    status = plasma_tune(PLASMA_FUNC_SGESV, N, N, NRHS);
    if (status != PLASMA_SUCCESS) {
        plasma_error("PLASMA_strsmpl", "plasma_tune() failed");
        return status;
    }

    /* Set Mt, NT & NTRHS */
    NB    = PLASMA_NB;
    IB    = PLASMA_IB;
    IBNB  = IB*NB;
    NT    = (N%NB==0) ? (N/NB) : (N/NB+1);

    plasma_sequence_create(plasma, &sequence);

    descL = plasma_desc_init(
        PlasmaRealFloat,
        IB, NB, IBNB,
        NT*IB, NT*NB, 0, 0, NT*IB, NT*NB);
    descL.mat = L;

    if ( PLASMA_TRANSLATION == PLASMA_OUTOFPLACE ) {
        plasma_sooplap2tile( descA, A, NB, NB, LDA, N,    0, 0, N, N   , plasma_desc_mat_free(&(descA)) );
        plasma_sooplap2tile( descB, B, NB, NB, LDB, NRHS, 0, 0, N, NRHS, plasma_desc_mat_free(&(descA)); plasma_desc_mat_free(&(descB)));
    } else {
        plasma_siplap2tile( descA, A, NB, NB, LDA, N,    0, 0, N, N   );
        plasma_siplap2tile( descB, B, NB, NB, LDB, NRHS, 0, 0, N, NRHS);
    }

    /* Call the tile interface */
    PLASMA_strsmpl_Tile_Async(&descA, &descL, IPIV, &descB, sequence, &request);

    if ( PLASMA_TRANSLATION == PLASMA_OUTOFPLACE ) {
        plasma_sooptile2lap( descB, B, NB, NB, LDB, NRHS );
        plasma_dynamic_sync();
        plasma_desc_mat_free(&descA);
        plasma_desc_mat_free(&descB);
    } else {
        plasma_siptile2lap( descA, A, NB, NB, LDA, N    );
        plasma_siptile2lap( descB, B, NB, NB, LDB, NRHS );
        plasma_dynamic_sync();
    }

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

/***************************************************************************/
int PLASMA_strsmpl_Tile(PLASMA_desc *A, PLASMA_desc *L, int *IPIV, PLASMA_desc *B)
{
    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_strsmpl_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    plasma_sequence_create(plasma, &sequence);
    PLASMA_strsmpl_Tile_Async(A, L, IPIV, B, sequence, &request);
    plasma_dynamic_sync();
    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);
    return status;
}

/***************************************************************************/
int PLASMA_strsmpl_Tile_Async(PLASMA_desc *A, PLASMA_desc *L, int *IPIV, PLASMA_desc *B,
                              PLASMA_sequence *sequence, PLASMA_request *request)
{
    PLASMA_desc descA = *A;
    PLASMA_desc descL = *L;
    PLASMA_desc descB = *B;
    plasma_context_t *plasma;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_strsmpl_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    if (sequence == NULL) {
        plasma_fatal_error("PLASMA_strsmpl_Tile", "NULL sequence");
        return PLASMA_ERR_UNALLOCATED;
    }
    if (request == NULL) {
        plasma_fatal_error("PLASMA_strsmpl_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_strsmpl_Tile", "invalid first descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    if (plasma_desc_check(&descL) != PLASMA_SUCCESS) {
        plasma_error("PLASMA_strsmpl_Tile", "invalid second descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    if (plasma_desc_check(&descB) != PLASMA_SUCCESS) {
        plasma_error("PLASMA_strsmpl_Tile", "invalid third descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    /* Check input arguments */
    if (descA.nb != descA.mb || descB.nb != descB.mb) {
        plasma_error("PLASMA_strsmpl_Tile", "only square tiles supported");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    /* Quick return */
/*
    if (min(N, NRHS) == 0)
        return PLASMA_SUCCESS;
*/
    plasma_parallel_call_6(plasma_pstrsmpl,
        PLASMA_desc, descA,
        PLASMA_desc, descB,
        PLASMA_desc, descL,
        int*, IPIV,
        PLASMA_sequence*, sequence,
        PLASMA_request*, request);

    return PLASMA_SUCCESS;
}