core_zunmlq.c

Go to the documentation of this file.

#include <lapacke.h>
#include "common.h"

/***************************************************************************/
#if defined(PLASMA_HAVE_WEAK)
#pragma weak CORE_zunmlq = PCORE_zunmlq
#define CORE_zunmlq PCORE_zunmlq
#endif
int CORE_zunmlq(int side, int trans,
                int M, int N, int K, int IB,
                PLASMA_Complex64_t *A, int LDA,
                PLASMA_Complex64_t *T, int LDT,
                PLASMA_Complex64_t *C, int LDC,
                PLASMA_Complex64_t *WORK, int LDWORK)
{
    int i, kb;
    int i1, i3;
    int nq, nw;
    int ic = 0;
    int jc = 0;
    int ni = N;
    int mi = M;

    /* Check input arguments */
    if ((side != PlasmaLeft) && (side != PlasmaRight)) {
        coreblas_error(1, "Illegal value of side");
        return -1;
    }
    /*
     * NQ is the order of Q and NW is the minimum dimension of WORK
     */
    if (side == PlasmaLeft) {
        nq = M;
        nw = N;
    }
    else {
        nq = N;
        nw = M;
    }

    if ((trans != PlasmaNoTrans) && (trans != PlasmaConjTrans)) {
        coreblas_error(2, "Illegal value of trans");
        return -2;
    }
    if (M < 0) {
        coreblas_error(3, "Illegal value of M");
        return -3;
    }
    if (N < 0) {
        coreblas_error(4, "Illegal value of N");
        return -4;
    }
    if ((K < 0) || (K > nq)) {
        coreblas_error(5, "Illegal value of K");
        return -5;
    }
    if ((IB < 0) || ( (IB == 0) && ((M > 0) && (N > 0)) )) {
        coreblas_error(6, "Illegal value of IB");
        return -6;
    }
    if ((LDA < max(1,K)) && (K > 0)) {
        coreblas_error(8, "Illegal value of LDA");
        return -8;
    }
    if ((LDC < max(1,M)) && (M > 0)) {
        coreblas_error(12, "Illegal value of LDC");
        return -12;
    }
    if ((LDWORK < max(1,nw)) && (nw > 0)) {
        coreblas_error(14, "Illegal value of LDWORK");
        return -14;
    }

    /* Quick return */
    if ((M == 0) || (N == 0) || (K == 0))
        return PLASMA_SUCCESS;

    if (((side == PlasmaLeft) && (trans == PlasmaNoTrans))
        || ((side == PlasmaRight) && (trans != PlasmaNoTrans))) {
        i1 = 0;
        i3 = IB;
    }
    else {
        i1 = ( ( K-1 ) / IB )*IB;
        i3 = -IB;
    }

    if( trans == PlasmaNoTrans) {
        trans = PlasmaConjTrans;
    }
    else {
        trans = PlasmaNoTrans;
    }

    for(i = i1; (i >- 1) && (i < K); i+=i3 ) {
        kb = min(IB, K-i);

        if (side == PlasmaLeft) {
            /*
             * H or H' is applied to C(i:m,1:n)
             */
            mi = M - i;
            ic = i;
        }
        else {
            /*
             * H or H' is applied to C(1:m,i:n)
             */
            ni = N - i;
            jc = i;
        }
        /*
         * Apply H or H'
         */
        LAPACKE_zlarfb_work(LAPACK_COL_MAJOR,
            lapack_const(side),
            lapack_const(trans),
            lapack_const(PlasmaForward),
            lapack_const(PlasmaRowwise),
            mi, ni, kb,
            &A[LDA*i+i], LDA,
            &T[LDT*i], LDT,
            &C[LDC*jc+ic], LDC,
            WORK, LDWORK);
    }
    return PLASMA_SUCCESS;
}

/***************************************************************************/
void QUARK_CORE_zunmlq(Quark *quark, Quark_Task_Flags *task_flags,
                       int side, int trans,
                       int m, int n, int k, int ib, int nb,
                       PLASMA_Complex64_t *A, int lda,
                       PLASMA_Complex64_t *T, int ldt,
                       PLASMA_Complex64_t *C, int ldc)
{
    DAG_CORE_UNMLQ;
    QUARK_Insert_Task(quark, CORE_zunmlq_quark, task_flags,
        sizeof(PLASMA_enum),                &side,  VALUE,
        sizeof(PLASMA_enum),                &trans, VALUE,
        sizeof(int),                        &m,     VALUE,
        sizeof(int),                        &n,     VALUE,
        sizeof(int),                        &k,     VALUE,
        sizeof(int),                        &ib,    VALUE,
        sizeof(PLASMA_Complex64_t)*nb*nb,    A,             INPUT | QUARK_REGION_U,
        sizeof(int),                        &lda,   VALUE,
        sizeof(PLASMA_Complex64_t)*ib*nb,    T,             INPUT,
        sizeof(int),                        &ldt,   VALUE,
        sizeof(PLASMA_Complex64_t)*nb*nb,    C,             INOUT,
        sizeof(int),                        &ldc,   VALUE,
        sizeof(PLASMA_Complex64_t)*ib*nb,    NULL,          SCRATCH,
        sizeof(int),                        &nb,    VALUE,
        0);
}

/***************************************************************************/
#if defined(PLASMA_HAVE_WEAK)
#pragma weak CORE_zunmlq_quark = PCORE_zunmlq_quark
#define CORE_zunmlq_quark PCORE_zunmlq_quark
#endif
void CORE_zunmlq_quark(Quark *quark)
{
    int side;
    int trans;
    int m;
    int n;
    int k;
    int ib;
    PLASMA_Complex64_t *A;
    int lda;
    PLASMA_Complex64_t *T;
    int ldt;
    PLASMA_Complex64_t *C;
    int ldc;
    PLASMA_Complex64_t *WORK;
    int ldwork;

    quark_unpack_args_14(quark, side, trans, m, n, k, ib,
                         A, lda, T, ldt, C, ldc, WORK, ldwork);
    CORE_zunmlq(side, trans, m, n, k, ib, 
                A, lda, T, ldt, C, ldc, WORK, ldwork);
}