Exemplo n.º 1
0
void CORE_zgetrf_incpiv_quark(Quark *quark)
{
    int m;
    int n;
    int ib;
    PLASMA_Complex64_t *A;
    int lda;
    int *IPIV;
    PLASMA_sequence *sequence;
    PLASMA_request *request;
    PLASMA_bool check_info;
    int iinfo;

    int info;

    quark_unpack_args_10(quark, m, n, ib, A, lda, IPIV, sequence, request, check_info, iinfo);
    CORE_zgetrf_incpiv(m, n, ib, A, lda, IPIV, &info);
    if (info != PLASMA_SUCCESS && check_info)
        plasma_sequence_flush(quark, sequence, request, iinfo+info);
}
Exemplo n.º 2
0
/***************************************************************************//**
 *  Parallel tile LU factorization - static scheduling
 **/
void plasma_pzgetrf_incpiv(plasma_context_t *plasma)
{
    PLASMA_desc A;
    PLASMA_desc L;
    int *IPIV;
    PLASMA_sequence *sequence;
    PLASMA_request *request;

    int k, m, n;
    int next_k;
    int next_m;
    int next_n;
    int ldak, ldam;
    int info;
    int tempkn, tempkm, tempmm, tempnn;
    int ib = PLASMA_IB;
    PLASMA_Complex64_t *work;

    plasma_unpack_args_5(A, L, IPIV, sequence, request);
    if (sequence->status != PLASMA_SUCCESS)
        return;
    work = (PLASMA_Complex64_t*)plasma_private_alloc(plasma, ib*L.nb, L.dtyp);
    ss_init(A.mt, A.nt, -1);

    k = 0;
    n = PLASMA_RANK;
    while (n >= A.nt) {
        k++;
        n = n-A.nt+k;
    }
    m = k;

    while (k < min(A.mt, A.nt) && n < A.nt && !ss_aborted()) {
        next_n = n;
        next_m = m;
        next_k = k;

        next_m++;
        if (next_m == A.mt) {
            next_n += PLASMA_SIZE;
            while (next_n >= A.nt && next_k < min(A.mt, A.nt)) {
                next_k++;
                next_n = next_n-A.nt+next_k;
            }
            next_m = next_k;
        }

        tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
        tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
        tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
        tempnn = n == A.nt-1 ? A.n-n*A.nb : A.nb;

        ldak = BLKLDD(A, k);
        ldam = BLKLDD(A, m);

        if (n == k) {
            if (m == k) {
                ss_cond_wait(k, k, k-1);
                CORE_zgetrf_incpiv(
                    tempkm, tempkn, ib,
                    A(k, k), ldak,
                    IPIV(k, k), &info);
                if (info != 0 && m == A.mt-1) {
                    plasma_request_fail(sequence, request, info + A.nb*k);
                    ss_abort();
                }
                ss_cond_set(k, k, k);
            }
            else {
                ss_cond_wait(m, k, k-1);
                CORE_ztstrf(
                    tempmm, tempkn, ib, A.nb,
                    A(k, k), ldak,
                    A(m, k), ldam,
                    L(m, k), L.mb,
                    IPIV(m, k),
                    work, L.nb, &info);
                if (info != 0 && m == A.mt-1) {
                    plasma_request_fail(sequence, request, info + A.nb*k);
                    ss_abort();
                }
                ss_cond_set(m, k, k);
            }
        }
        else {
            if (m == k) {
                ss_cond_wait(k, k, k);
                ss_cond_wait(k, n, k-1);
                CORE_zgessm(
                    tempkm, tempnn, tempkm, ib,
                    IPIV(k, k),
                    A(k, k), ldak,
                    A(k, n), ldak);
            }
            else {
                ss_cond_wait(m, k, k);
                ss_cond_wait(m, n, k-1);
                CORE_zssssm(
                    A.nb, tempnn, tempmm, tempnn, A.nb, ib,
                    A(k, n), ldak,
                    A(m, n), ldam,
                    L(m, k), L.mb,
                    A(m, k), ldam,
                    IPIV(m, k));
                ss_cond_set(m, n, k);
            }
        }
        n = next_n;
        m = next_m;
        k = next_k;
    }
    plasma_private_free(plasma, work);
    ss_finalize();
}