sgelqs.c

Go to the documentation of this file.

#include "common.h"

/***************************************************************************/
int PLASMA_sgelqs(int M, int N, int NRHS,
                  float *A, int LDA,
                  float *T,
                  float *B, int LDB)
{
    int NB, IB, IBNB, MT, NT;
    int status;
    plasma_context_t *plasma;
    PLASMA_sequence *sequence = NULL;
    PLASMA_request request = PLASMA_REQUEST_INITIALIZER;
    PLASMA_desc descA, descB, descT;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_sgelqs", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    /* Check input arguments */
    if (M < 0) {
        plasma_error("PLASMA_sgelqs", "illegal value of M");
        return -1;
    }
    if (N < 0 || M > N) {
        plasma_error("PLASMA_sgelqs", "illegal value of N");
        return -2;
    }
    if (NRHS < 0) {
        plasma_error("PLASMA_sgelqs", "illegal value of N");
        return -3;
    }
    if (LDA < max(1, M)) {
        plasma_error("PLASMA_sgelqs", "illegal value of LDA");
        return -5;
    }
    if (LDB < max(1, max(1, N))) {
        plasma_error("PLASMA_sgelqs", "illegal value of LDB");
        return -8;
    }
    /* Quick return */
    if (min(M, min(N, NRHS)) == 0) {
        return PLASMA_SUCCESS;
    }

    /* Tune NB & IB depending on M, N & NRHS; Set NBNBSIZE */
    status = plasma_tune(PLASMA_FUNC_SGELS, M, N, NRHS);
    if (status != PLASMA_SUCCESS) {
        plasma_error("PLASMA_sgelqs", "plasma_tune() failed");
        return status;
    }

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

    plasma_sequence_create(plasma, &sequence);

    if (plasma->householder == PLASMA_FLAT_HOUSEHOLDER) {
        descT = plasma_desc_init(
            PlasmaRealFloat,
            IB, NB, IBNB,
            MT*IB, NT*NB, 0, 0, MT*IB, NT*NB);
    }
    else {
        /* Double the size of T to accomodate the tree reduction phase */
        descT = plasma_desc_init(
            PlasmaRealFloat,
            IB, NB, IBNB,
            MT*IB, 2*NT*NB, 0, 0, MT*IB, 2*NT*NB);
    }
    descT.mat = T;

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

    /* Call the tile interface */
    PLASMA_sgelqs_Tile_Async(&descA, &descT, &descB, sequence, &request);

    if ( PLASMA_TRANSLATION == PLASMA_OUTOFPLACE ) {
        plasma_sooptile2lap( descA, A, NB, NB, LDA, N    );
        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_sgelqs_Tile(PLASMA_desc *A, PLASMA_desc *T, 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_sgelqs_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    plasma_sequence_create(plasma, &sequence);
    PLASMA_sgelqs_Tile_Async(A, T, B, sequence, &request);
    plasma_dynamic_sync();
    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);
    return status;
}

/***************************************************************************/
int PLASMA_sgelqs_Tile_Async(PLASMA_desc *A, PLASMA_desc *T, PLASMA_desc *B,
                             PLASMA_sequence *sequence, PLASMA_request *request)
{
    PLASMA_desc descA = *A;
    PLASMA_desc descT = *T;
    PLASMA_desc descB = *B;
    plasma_context_t *plasma;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_sgelqs_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    if (sequence == NULL) {
        plasma_fatal_error("PLASMA_sgelqs_Tile", "NULL sequence");
        return PLASMA_ERR_UNALLOCATED;
    }
    if (request == NULL) {
        plasma_fatal_error("PLASMA_sgelqs_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_sgelqs_Tile", "invalid first descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    if (plasma_desc_check(&descT) != PLASMA_SUCCESS) {
        plasma_error("PLASMA_sgelqs_Tile", "invalid second descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    if (plasma_desc_check(&descB) != PLASMA_SUCCESS) {
        plasma_error("PLASMA_sgelqs_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_sgelqs_Tile", "only square tiles supported");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    /* Quick return */
/*
    if (min(M, min(N, NRHS)) == 0) {
        return PLASMA_SUCCESS;
    }
*/
    plasma_parallel_call_3(plasma_pstile_zero,
        PLASMA_desc, plasma_desc_submatrix(descB, descA.m, 0, descA.n-descA.m, descB.n),
        PLASMA_sequence*, sequence,
        PLASMA_request*, request);

    plasma_parallel_call_9(plasma_pstrsm,
        PLASMA_enum, PlasmaLeft,
        PLASMA_enum, PlasmaLower,
        PLASMA_enum, PlasmaNoTrans,
        PLASMA_enum, PlasmaNonUnit,
        float, 1.0,
        PLASMA_desc, plasma_desc_submatrix(descA, 0, 0, descA.m, descA.m),
        PLASMA_desc, plasma_desc_submatrix(descB, 0, 0, descA.m, descB.n),
        PLASMA_sequence*, sequence,
        PLASMA_request*, request);

    if (plasma->householder == PLASMA_FLAT_HOUSEHOLDER) {
        plasma_parallel_call_7(plasma_psormlq,
            PLASMA_enum, PlasmaLeft,
            PLASMA_enum, PlasmaTrans,
            PLASMA_desc, descA,
            PLASMA_desc, descB,
            PLASMA_desc, descT,
            PLASMA_sequence*, sequence,
            PLASMA_request*, request);
    }
    else {
        plasma_dynamic_call_8(plasma_psormlqrh,
            PLASMA_enum, PlasmaLeft,
            PLASMA_enum, PlasmaTrans,
            PLASMA_desc, descA,
            PLASMA_desc, descB,
            PLASMA_desc, descT,
            PLASMA_enum, PLASMA_RHBLK,
            PLASMA_sequence*, sequence,
            PLASMA_request*, request);
    }

    return PLASMA_SUCCESS;
}