kusano 2b45e8
/*********************************************************************/
kusano 2b45e8
/* Copyright 2009, 2010 The University of Texas at Austin.           */
kusano 2b45e8
/* All rights reserved.                                              */
kusano 2b45e8
/*                                                                   */
kusano 2b45e8
/* Redistribution and use in source and binary forms, with or        */
kusano 2b45e8
/* without modification, are permitted provided that the following   */
kusano 2b45e8
/* conditions are met:                                               */
kusano 2b45e8
/*                                                                   */
kusano 2b45e8
/*   1. Redistributions of source code must retain the above         */
kusano 2b45e8
/*      copyright notice, this list of conditions and the following  */
kusano 2b45e8
/*      disclaimer.                                                  */
kusano 2b45e8
/*                                                                   */
kusano 2b45e8
/*   2. Redistributions in binary form must reproduce the above      */
kusano 2b45e8
/*      copyright notice, this list of conditions and the following  */
kusano 2b45e8
/*      disclaimer in the documentation and/or other materials       */
kusano 2b45e8
/*      provided with the distribution.                              */
kusano 2b45e8
/*                                                                   */
kusano 2b45e8
/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
kusano 2b45e8
/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
kusano 2b45e8
/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
kusano 2b45e8
/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
kusano 2b45e8
/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
kusano 2b45e8
/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
kusano 2b45e8
/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
kusano 2b45e8
/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
kusano 2b45e8
/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
kusano 2b45e8
/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
kusano 2b45e8
/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
kusano 2b45e8
/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
kusano 2b45e8
/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
kusano 2b45e8
/*    POSSIBILITY OF SUCH DAMAGE.                                    */
kusano 2b45e8
/*                                                                   */
kusano 2b45e8
/* The views and conclusions contained in the software and           */
kusano 2b45e8
/* documentation are those of the authors and should not be          */
kusano 2b45e8
/* interpreted as representing official policies, either expressed   */
kusano 2b45e8
/* or implied, of The University of Texas at Austin.                 */
kusano 2b45e8
/*********************************************************************/
kusano 2b45e8
kusano 2b45e8
#include <stdio.h></stdio.h>
kusano 2b45e8
#include <ctype.h></ctype.h>
kusano 2b45e8
#include "common.h"
kusano 2b45e8
#ifdef FUNCTION_PROFILE
kusano 2b45e8
#include "functable.h"
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
#ifdef XDOUBLE
kusano 2b45e8
#define ERROR_NAME "QTBMV "
kusano 2b45e8
#elif defined(DOUBLE)
kusano 2b45e8
#define ERROR_NAME "DTBMV "
kusano 2b45e8
#else
kusano 2b45e8
#define ERROR_NAME "STBMV "
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
static int (*tbmv[])(BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG, void *) = {
kusano 2b45e8
#ifdef XDOUBLE
kusano 2b45e8
  qtbmv_NUU, qtbmv_NUN, qtbmv_NLU, qtbmv_NLN,
kusano 2b45e8
  qtbmv_TUU, qtbmv_TUN, qtbmv_TLU, qtbmv_TLN,
kusano 2b45e8
#elif defined(DOUBLE)
kusano 2b45e8
  dtbmv_NUU, dtbmv_NUN, dtbmv_NLU, dtbmv_NLN,
kusano 2b45e8
  dtbmv_TUU, dtbmv_TUN, dtbmv_TLU, dtbmv_TLN,
kusano 2b45e8
#else
kusano 2b45e8
  stbmv_NUU, stbmv_NUN, stbmv_NLU, stbmv_NLN,
kusano 2b45e8
  stbmv_TUU, stbmv_TUN, stbmv_TLU, stbmv_TLN,
kusano 2b45e8
#endif
kusano 2b45e8
};
kusano 2b45e8
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
static int (*tbmv_thread[])(BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, int) = {
kusano 2b45e8
#ifdef XDOUBLE
kusano 2b45e8
  qtbmv_thread_NUU, qtbmv_thread_NUN, qtbmv_thread_NLU, qtbmv_thread_NLN,
kusano 2b45e8
  qtbmv_thread_TUU, qtbmv_thread_TUN, qtbmv_thread_TLU, qtbmv_thread_TLN,
kusano 2b45e8
#elif defined(DOUBLE)
kusano 2b45e8
  dtbmv_thread_NUU, dtbmv_thread_NUN, dtbmv_thread_NLU, dtbmv_thread_NLN,
kusano 2b45e8
  dtbmv_thread_TUU, dtbmv_thread_TUN, dtbmv_thread_TLU, dtbmv_thread_TLN,
kusano 2b45e8
#else
kusano 2b45e8
  stbmv_thread_NUU, stbmv_thread_NUN, stbmv_thread_NLU, stbmv_thread_NLN,
kusano 2b45e8
  stbmv_thread_TUU, stbmv_thread_TUN, stbmv_thread_TLU, stbmv_thread_TLN,
kusano 2b45e8
#endif
kusano 2b45e8
};
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
#ifndef CBLAS
kusano 2b45e8
kusano 2b45e8
void NAME(char *UPLO, char *TRANS, char *DIAG,
kusano 2b45e8
	 blasint *N, blasint *K, 
kusano 2b45e8
	 FLOAT *a, blasint *LDA, FLOAT *x, blasint *INCX){
kusano 2b45e8
  
kusano 2b45e8
  char uplo_arg  = *UPLO;
kusano 2b45e8
  char trans_arg = *TRANS;
kusano 2b45e8
  char diag_arg  = *DIAG;
kusano 2b45e8
  
kusano 2b45e8
  blasint n    = *N;
kusano 2b45e8
  blasint k    = *K;
kusano 2b45e8
  blasint lda  = *LDA;
kusano 2b45e8
  blasint incx = *INCX;
kusano 2b45e8
kusano 2b45e8
  blasint info;
kusano 2b45e8
  int uplo;
kusano 2b45e8
  int unit;
kusano 2b45e8
  int trans;
kusano 2b45e8
  FLOAT *buffer;
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
  int nthreads;
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  PRINT_DEBUG_NAME;
kusano 2b45e8
kusano 2b45e8
  TOUPPER(uplo_arg);
kusano 2b45e8
  TOUPPER(trans_arg);
kusano 2b45e8
  TOUPPER(diag_arg);
kusano 2b45e8
kusano 2b45e8
  trans = -1;
kusano 2b45e8
  unit  = -1;
kusano 2b45e8
  uplo  = -1;
kusano 2b45e8
kusano 2b45e8
  if (trans_arg == 'N') trans = 0;
kusano 2b45e8
  if (trans_arg == 'T') trans = 1;
kusano 2b45e8
  if (trans_arg == 'R') trans = 0;
kusano 2b45e8
  if (trans_arg == 'C') trans = 1;
kusano 2b45e8
kusano 2b45e8
  if (diag_arg  == 'U') unit  = 0;
kusano 2b45e8
  if (diag_arg  == 'N') unit  = 1;
kusano 2b45e8
kusano 2b45e8
  if (uplo_arg  == 'U') uplo  = 0;
kusano 2b45e8
  if (uplo_arg  == 'L') uplo  = 1;
kusano 2b45e8
kusano 2b45e8
  info = 0;
kusano 2b45e8
kusano 2b45e8
  if (incx == 0)          info =  9;
kusano 2b45e8
  if (lda < k + 1)        info =  7;
kusano 2b45e8
  if (k < 0)              info =  5;
kusano 2b45e8
  if (n < 0)              info =  4;
kusano 2b45e8
  if (unit  < 0)          info =  3;
kusano 2b45e8
  if (trans < 0)          info =  2;
kusano 2b45e8
  if (uplo  < 0)          info =  1;
kusano 2b45e8
kusano 2b45e8
  if (info != 0) {
kusano 2b45e8
    BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
kusano 2b45e8
    return;
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
#else
kusano 2b45e8
kusano 2b45e8
void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
kusano 2b45e8
	   enum CBLAS_TRANSPOSE TransA, enum CBLAS_DIAG Diag,
kusano 2b45e8
	   blasint n, blasint k, FLOAT  *a, blasint lda, FLOAT  *x, blasint incx) {
kusano 2b45e8
kusano 2b45e8
  int trans, uplo, unit;
kusano 2b45e8
  blasint info;
kusano 2b45e8
  FLOAT *buffer;
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
  int nthreads;
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  PRINT_DEBUG_CNAME;
kusano 2b45e8
kusano 2b45e8
  unit  = -1;
kusano 2b45e8
  uplo  = -1;
kusano 2b45e8
  trans = -1;
kusano 2b45e8
  info  =  0;
kusano 2b45e8
kusano 2b45e8
  if (order == CblasColMajor) {
kusano 2b45e8
    if (Uplo == CblasUpper)         uplo  = 0;
kusano 2b45e8
    if (Uplo == CblasLower)         uplo  = 1;
kusano 2b45e8
kusano 2b45e8
    if (TransA == CblasNoTrans)     trans = 0;
kusano 2b45e8
    if (TransA == CblasTrans)       trans = 1;
kusano 2b45e8
    if (TransA == CblasConjNoTrans) trans = 0;
kusano 2b45e8
    if (TransA == CblasConjTrans)   trans = 1;
kusano 2b45e8
    
kusano 2b45e8
    if (Diag == CblasUnit)          unit  = 0;
kusano 2b45e8
    if (Diag == CblasNonUnit)       unit  = 1;
kusano 2b45e8
kusano 2b45e8
    info = -1;
kusano 2b45e8
kusano 2b45e8
    if (incx == 0)          info =  9;
kusano 2b45e8
    if (lda < k + 1)        info =  7;
kusano 2b45e8
    if (k < 0)              info =  5;
kusano 2b45e8
    if (n < 0)              info =  4;
kusano 2b45e8
    if (unit  < 0)          info =  3;
kusano 2b45e8
    if (trans < 0)          info =  2;
kusano 2b45e8
    if (uplo  < 0)          info =  1;
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  if (order == CblasRowMajor) {
kusano 2b45e8
    if (Uplo == CblasUpper)         uplo  = 1;
kusano 2b45e8
    if (Uplo == CblasLower)         uplo  = 0;
kusano 2b45e8
kusano 2b45e8
    if (TransA == CblasNoTrans)     trans = 1;
kusano 2b45e8
    if (TransA == CblasTrans)       trans = 0;
kusano 2b45e8
    if (TransA == CblasConjNoTrans) trans = 1;
kusano 2b45e8
    if (TransA == CblasConjTrans)   trans = 0;
kusano 2b45e8
kusano 2b45e8
    if (Diag == CblasUnit)          unit  = 0;
kusano 2b45e8
    if (Diag == CblasNonUnit)       unit  = 1;
kusano 2b45e8
kusano 2b45e8
    info = -1;
kusano 2b45e8
kusano 2b45e8
    if (incx == 0)          info =  9;
kusano 2b45e8
    if (lda < k + 1)        info =  7;
kusano 2b45e8
    if (k < 0)              info =  5;
kusano 2b45e8
    if (n < 0)              info =  4;
kusano 2b45e8
    if (unit  < 0)          info =  3;
kusano 2b45e8
    if (trans < 0)          info =  2;
kusano 2b45e8
    if (uplo  < 0)          info =  1;
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  if (info >= 0) {
kusano 2b45e8
    BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
kusano 2b45e8
    return;
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
#endif
kusano 2b45e8
  
kusano 2b45e8
  if (n == 0) return;
kusano 2b45e8
  
kusano 2b45e8
  IDEBUG_START;
kusano 2b45e8
kusano 2b45e8
  FUNCTION_PROFILE_START();
kusano 2b45e8
kusano 2b45e8
  if (incx < 0 ) x -= (n - 1) * incx;
kusano 2b45e8
kusano 2b45e8
  buffer = (FLOAT *)blas_memory_alloc(1);
kusano 2b45e8
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
  nthreads = num_cpu_avail(2);
kusano 2b45e8
kusano 2b45e8
  if (nthreads == 1) {
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  (tbmv[(trans<<2) | (uplo<<1) | unit])(n, k, a, lda, x, incx, buffer);
kusano 2b45e8
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
  } else {
kusano 2b45e8
kusano 2b45e8
    (tbmv_thread[(trans<<2) | (uplo<<1) | unit])(n, k, a, lda, x, incx, buffer, nthreads);
kusano 2b45e8
kusano 2b45e8
  }
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  blas_memory_free(buffer);
kusano 2b45e8
kusano 2b45e8
  FUNCTION_PROFILE_END(1, n * k / 2 + n, n * k);
kusano 2b45e8
kusano 2b45e8
  IDEBUG_END;
kusano 2b45e8
kusano 2b45e8
  return;
kusano 2b45e8
}