core_zgeqrt.c

Go to the documentation of this file.

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

/***************************************************************************/
#if defined(PLASMA_HAVE_WEAK)
#pragma weak CORE_zgeqrt = PCORE_zgeqrt
#define CORE_zgeqrt PCORE_zgeqrt
#endif
int CORE_zgeqrt(int M, int N, int IB,
                PLASMA_Complex64_t *A, int LDA,
                PLASMA_Complex64_t *T, int LDT,
                PLASMA_Complex64_t *TAU,
                PLASMA_Complex64_t *WORK)
{
    int i, k, sb;

    /* Check input arguments */
    if (M < 0) {
        coreblas_error(1, "Illegal value of M");
        return -1;
    }
    if (N < 0) {
        coreblas_error(2, "Illegal value of N");
        return -2;
    }
    if ((IB < 0) || ( (IB == 0) && ((M > 0) && (N > 0)) )) {
        coreblas_error(3, "Illegal value of IB");
        return -3;
    }
    if ((LDA < max(1,M)) && (M > 0)) {
        coreblas_error(5, "Illegal value of LDA");
        return -5;
    }
    if ((LDT < max(1,IB)) && (IB > 0)) {
        coreblas_error(7, "Illegal value of LDT");
        return -7;
    }

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

    k = min(M, N);

    for(i = 0; i < k; i += IB) {
        sb = min(IB, k-i);

        LAPACKE_zgeqr2_work(LAPACK_COL_MAJOR, M-i, sb, 
                            &A[LDA*i+i], LDA, &TAU[i], WORK);

        LAPACKE_zlarft_work(LAPACK_COL_MAJOR,
            lapack_const(PlasmaForward),
            lapack_const(PlasmaColumnwise),
            M-i, sb,
            &A[LDA*i+i], LDA, &TAU[i],
            &T[LDT*i], LDT);

        if (N > i+sb) {
            LAPACKE_zlarfb_work(
                LAPACK_COL_MAJOR,
                lapack_const(PlasmaLeft),
                lapack_const(PlasmaConjTrans),
                lapack_const(PlasmaForward),
                lapack_const(PlasmaColumnwise),
                M-i, N-i-sb, sb,
                &A[LDA*i+i],      LDA,
                &T[LDT*i],        LDT,
                &A[LDA*(i+sb)+i], LDA,
                WORK, N-i-sb);
        }
    }
    return PLASMA_SUCCESS;
}

/***************************************************************************/
void QUARK_CORE_zgeqrt(Quark *quark, Quark_Task_Flags *task_flags,
                       int m, int n, int ib, int nb,
                       PLASMA_Complex64_t *A, int lda,
                       PLASMA_Complex64_t *T, int ldt)
{
    DAG_CORE_GEQRT;
    QUARK_Insert_Task(quark, CORE_zgeqrt_quark, task_flags,
        sizeof(int),                        &m,     VALUE,
        sizeof(int),                        &n,     VALUE,
        sizeof(int),                        &ib,    VALUE,
        sizeof(PLASMA_Complex64_t)*nb*nb,    A,             INOUT,
        sizeof(int),                        &lda,   VALUE,
        sizeof(PLASMA_Complex64_t)*ib*nb,    T,             OUTPUT,
        sizeof(int),                        &ldt,   VALUE,
        sizeof(PLASMA_Complex64_t)*nb,       NULL,          SCRATCH,
        sizeof(PLASMA_Complex64_t)*ib*nb,    NULL,          SCRATCH,
        0);
}

/***************************************************************************/
#if defined(PLASMA_HAVE_WEAK)
#pragma weak CORE_zgeqrt_quark = PCORE_zgeqrt_quark
#define CORE_zgeqrt_quark PCORE_zgeqrt_quark
#endif
void CORE_zgeqrt_quark(Quark *quark)
{
    int m;
    int n;
    int ib;
    PLASMA_Complex64_t *A;
    int lda;
    PLASMA_Complex64_t *T;
    int ldt;
    PLASMA_Complex64_t *TAU;
    PLASMA_Complex64_t *WORK;

    quark_unpack_args_9(quark, m, n, ib, A, lda, T, ldt, TAU, WORK);
    CORE_zgeqrt(m, n, ib, A, lda, T, ldt, TAU, WORK);
}