zlaswp.c

Go to the documentation of this file.

#include "common.h"

/***************************************************************************/
int PLASMA_zlaswp(int N, PLASMA_Complex64_t *A, int LDA,
                  int K1, int K2, int *IPIV, int INCX)
{
    int NB;
    int status;
    plasma_context_t *plasma;
    PLASMA_sequence *sequence = NULL;
    PLASMA_request request = PLASMA_REQUEST_INITIALIZER;
    PLASMA_desc descA;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_zlaswp", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    /* Check input arguments */
    if (N < 0) {
        plasma_error("PLASMA_zlaswp", "illegal value of N");
        return -1;
    }
    if (LDA < max(1, N)) {
        plasma_error("PLASMA_zlaswp", "illegal value of LDA");
        return -3;
    }

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

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

    /* Set NT & NTRHS */
    NB = PLASMA_NB;

    plasma_sequence_create(plasma, &sequence);

    if ( PLASMA_TRANSLATION == PLASMA_OUTOFPLACE ) {
        plasma_zooplap2tile( descA, A, NB, NB, LDA, N, 0, 0, K2, N, plasma_desc_mat_free(&(descA)) );
    } else {
        plasma_ziplap2tile( descA, A, NB, NB, LDA, N, 0, 0, K2, N);
    }

    /* Call the tile interface */
    PLASMA_zlaswp_Tile_Async(&descA, K1, K2, IPIV, INCX, sequence, &request);

    if ( PLASMA_TRANSLATION == PLASMA_OUTOFPLACE ) {
        plasma_zooptile2lap( descA, A, NB, NB, LDA, N );
        plasma_dynamic_sync();
        plasma_desc_mat_free(&descA);
    } else {
        plasma_ziptile2lap( descA, A, NB, NB, LDA, N );
        plasma_dynamic_sync();
    }

    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);
    return status;
}

/***************************************************************************/
int PLASMA_zlaswp_Tile(PLASMA_desc *A, int K1, int K2, int *IPIV, int INCX)
{
    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_zlaswp_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    plasma_sequence_create(plasma, &sequence);
    PLASMA_zlaswp_Tile_Async(A, K1, K2, IPIV, INCX, sequence, &request);
    plasma_dynamic_sync();
    status = sequence->status;
    plasma_sequence_destroy(plasma, sequence);
    return status;
}

/***************************************************************************/
int PLASMA_zlaswp_Tile_Async(PLASMA_desc *A, int K1, int K2, int *IPIV, int INCX,
                             PLASMA_sequence *sequence, PLASMA_request *request)
{
    PLASMA_desc descA = *A;
    plasma_context_t *plasma;

    plasma = plasma_context_self();
    if (plasma == NULL) {
        plasma_fatal_error("PLASMA_zlaswp_Tile", "PLASMA not initialized");
        return PLASMA_ERR_NOT_INITIALIZED;
    }
    if (sequence == NULL) {
        plasma_fatal_error("PLASMA_zlaswp_Tile", "NULL sequence");
        return PLASMA_ERR_UNALLOCATED;
    }
    if (request == NULL) {
        plasma_fatal_error("PLASMA_zlaswp_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_zlaswp_Tile", "invalid first descriptor");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }

    if ( (K1 != 1) || (K2 != descA.m) ) {
        plasma_error("PLASMA_zlaswp_Tile", "invalid K1 or K2 (1..M is the only interval supported right now)");
        return plasma_request_fail(sequence, request, PLASMA_ERR_ILLEGAL_VALUE);
    }

    plasma_dynamic_call_3(
        plasma_pzbarrier_tl2pnl,
        PLASMA_desc, descA,
        PLASMA_sequence*, sequence,
        PLASMA_request*,  request);

    /* swap */
    plasma_dynamic_call_5(
        plasma_pzlaswp,
        PLASMA_desc, descA,
        int *,       IPIV,
        int,         INCX,
        PLASMA_sequence*, sequence,
        PLASMA_request*,  request);
    
    plasma_dynamic_call_3(
        plasma_pzbarrier_pnl2tl,
        PLASMA_desc, descA,
        PLASMA_sequence*, sequence,
        PLASMA_request*,  request);

    return PLASMA_SUCCESS;
}