dorglq.c

Go to the documentation of this file.

#include "common.h"

/***************************************************************************/
int PLASMA_dorglq(int M, int N, int K,
                  double *A, int LDA,
                  double *T,
                  double *Q, int LDQ)
{
    int NB, IB, IBNB, KT, NT;
    int status;
    plasma_context_t *plasma;
    PLASMA_sequence *sequence = NULL;
    PLASMA_request request = PLASMA_REQUEST_INITIALIZER;
    PLASMA_desc descA, descQ, descT;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_dorglq", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    if (M < 0) {
        plasma_error("PLASMA_dorglq", "illegal value of M");
        return -1;
    }
    if (N < M) {
        plasma_error("PLASMA_dorglq", "illegal value of N");
        return -2;
    }
    if (K < 0 || K > M) {
        plasma_error("PLASMA_dorglq", "illegal value of K");
        return -3;
    }
    if (LDA < max(1, M)) {
        plasma_error("PLASMA_dorglq", "illegal value of LDA");
        return -5;
    }
    if (LDQ < max(1, M)) {
        plasma_error("PLASMA_dorglq", "illegal value of LDQ");
        return -8;
    }
    /* Quick return - currently NOT equivalent to LAPACK's:
     * CALL DLASET( 'Full', MAX( M, N ), NRHS, ZERO, ZERO, B, LDQ ) */
    if (min(M, min(N, K)) == 0)
        return PLASMA_SUCCESS;

    /* Tune NB & IB depending on M, N & NRHS; Set NBNB */
    status = plasma_tune(PLASMA_FUNC_DGELS, M, N, 0);
    if (status != PLASMA_SUCCESS) {
        plasma_error("PLASMA_dorglq", "plasma_tune() failed");
        return status;
    }

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

    plasma_sequence_create(plasma, &sequence);

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

    if ( PLASMA_TRANSLATION == PLASMA_OUTOFPLACE ) {
        plasma_dooplap2tile( descA, A, NB, NB, LDA, N, 0, 0, K, N, plasma_desc_mat_free(&(descA)) );
        plasma_dooplap2tile( descQ, Q, NB, NB, LDQ, N, 0, 0, M, N, plasma_desc_mat_free(&(descA)); plasma_desc_mat_free(&(descQ)));
    } else {
        plasma_diplap2tile( descA, A, NB, NB, LDA, N, 0, 0, K, N);
        plasma_diplap2tile( descQ, Q, NB, NB, LDQ, N, 0, 0, M, N);
    }

    /* Call the tile interface */
    PLASMA_dorglq_Tile_Async(&descA, &descT, &descQ, sequence, &request);

    if ( PLASMA_TRANSLATION == PLASMA_OUTOFPLACE ) {
        plasma_dooptile2lap( descQ, Q, NB, NB, LDQ, N );
        plasma_dynamic_sync();
        plasma_desc_mat_free(&descA);
        plasma_desc_mat_free(&descQ);
    } else {
        plasma_diptile2lap( descA, A, NB, NB, LDA, N );
        plasma_diptile2lap( descQ, Q, NB, NB, LDQ, N );
        plasma_dynamic_sync();
    }
        
    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);
    return status;
}

/***************************************************************************/
int PLASMA_dorglq_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_dorglq_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    plasma_sequence_create(plasma, &sequence);
    PLASMA_dorglq_Tile_Async(A, T, B, sequence, &request);
    plasma_dynamic_sync();
    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);
    return status;
}

/***************************************************************************/
int PLASMA_dorglq_Tile_Async(PLASMA_desc *A, PLASMA_desc *T, PLASMA_desc *Q,
                             PLASMA_sequence *sequence, PLASMA_request *request)
{
    PLASMA_desc descA = *A;
    PLASMA_desc descT = *T;
    PLASMA_desc descQ = *Q;
    plasma_context_t *plasma;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_dorglq_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    if (sequence == NULL) {
        plasma_fatal_error("PLASMA_dorglq_Tile", "NULL sequence");
        return PLASMA_ERR_UNALLOCATED;
    }
    if (request == NULL) {
        plasma_fatal_error("PLASMA_dorglq_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_dorglq_Tile", "invalid first descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    if (plasma_desc_check(&descT) != PLASMA_SUCCESS) {
        plasma_error("PLASMA_dorglq_Tile", "invalid second descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    if (plasma_desc_check(&descQ) != PLASMA_SUCCESS) {
        plasma_error("PLASMA_dorglq_Tile", "invalid third descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    /* Check input arguments */
    if (descA.nb != descA.mb || descQ.nb != descQ.mb) {
        plasma_error("PLASMA_dorglq_Tile", "only square tiles supported");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }
    /* Quick return - currently NOT equivalent to LAPACK's:
     * CALL DLASET( 'Full', MAX( M, N ), NRHS, ZERO, ZERO, Q, LDQ ) */
/*
    if (min(M, N) == 0)
        return PLASMA_SUCCESS;
*/
    if (plasma->householder == PLASMA_FLAT_HOUSEHOLDER) {
        plasma_dynamic_call_5(plasma_pdorglq,
            PLASMA_desc, descA,
            PLASMA_desc, descQ,
            PLASMA_desc, descT,
            PLASMA_sequence*, sequence,
            PLASMA_request*, request);
    }
    else {
        plasma_dynamic_call_6(plasma_pdorglqrh,
            PLASMA_desc, descA,
            PLASMA_desc, descQ,
            PLASMA_desc, descT,
            PLASMA_enum, PLASMA_RHBLK,
            PLASMA_sequence*, sequence,
            PLASMA_request*, request);
    }

    return PLASMA_SUCCESS;
}