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 "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 "XGEMV "
kusano 2b45e8
#elif defined(DOUBLE)
kusano 2b45e8
#define ERROR_NAME "ZGEMV "
kusano 2b45e8
#else
kusano 2b45e8
#define ERROR_NAME "CGEMV "
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
static int (*gemv_thread[])(BLASLONG, BLASLONG, FLOAT *, FLOAT *, BLASLONG,  FLOAT * , BLASLONG, FLOAT *, BLASLONG, FLOAT *, int) = {
kusano 2b45e8
#ifdef XDOUBLE
kusano 2b45e8
  xgemv_thread_n, xgemv_thread_t, xgemv_thread_r, xgemv_thread_c, xgemv_thread_o, xgemv_thread_u, xgemv_thread_s, xgemv_thread_d, 
kusano 2b45e8
#elif defined DOUBLE
kusano 2b45e8
  zgemv_thread_n, zgemv_thread_t, zgemv_thread_r, zgemv_thread_c, zgemv_thread_o, zgemv_thread_u, zgemv_thread_s, zgemv_thread_d, 
kusano 2b45e8
#else
kusano 2b45e8
  cgemv_thread_n, cgemv_thread_t, cgemv_thread_r, cgemv_thread_c, cgemv_thread_o, cgemv_thread_u, cgemv_thread_s, cgemv_thread_d, 
kusano 2b45e8
#endif
kusano 2b45e8
};
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
#ifndef CBLAS
kusano 2b45e8
kusano 2b45e8
void NAME(char *TRANS, blasint *M, blasint *N,
kusano 2b45e8
	 FLOAT *ALPHA, FLOAT *a, blasint *LDA,
kusano 2b45e8
	 FLOAT *x, blasint *INCX,
kusano 2b45e8
	 FLOAT *BETA,  FLOAT *y, blasint *INCY){
kusano 2b45e8
  
kusano 2b45e8
  char trans = *TRANS;
kusano 2b45e8
  blasint m = *M;
kusano 2b45e8
  blasint n = *N;
kusano 2b45e8
  blasint lda = *LDA;
kusano 2b45e8
  blasint incx = *INCX;
kusano 2b45e8
  blasint incy = *INCY;
kusano 2b45e8
kusano 2b45e8
  FLOAT *buffer;
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
  int nthreads;
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  int (*gemv[])(BLASLONG, BLASLONG, BLASLONG, FLOAT, FLOAT, FLOAT *, BLASLONG,
kusano 2b45e8
		FLOAT * , BLASLONG, FLOAT *, BLASLONG, FLOAT *) = {
kusano 2b45e8
		  GEMV_N, GEMV_T, GEMV_R, GEMV_C,
kusano 2b45e8
		  GEMV_O, GEMV_U, GEMV_S, GEMV_D,
kusano 2b45e8
		};
kusano 2b45e8
  
kusano 2b45e8
  blasint    info;
kusano 2b45e8
  blasint    lenx, leny;
kusano 2b45e8
  blasint    i;
kusano 2b45e8
kusano 2b45e8
  PRINT_DEBUG_NAME;
kusano 2b45e8
kusano 2b45e8
  FLOAT alpha_r = *(ALPHA + 0);
kusano 2b45e8
  FLOAT alpha_i = *(ALPHA + 1);
kusano 2b45e8
kusano 2b45e8
  FLOAT beta_r  = *(BETA + 0);
kusano 2b45e8
  FLOAT beta_i  = *(BETA + 1);
kusano 2b45e8
kusano 2b45e8
  TOUPPER(trans);
kusano 2b45e8
kusano 2b45e8
  info = 0;
kusano 2b45e8
kusano 2b45e8
  i    = -1;
kusano 2b45e8
kusano 2b45e8
  if (trans == 'N')  i = 0;
kusano 2b45e8
  if (trans == 'T')  i = 1;
kusano 2b45e8
  if (trans == 'R')  i = 2;
kusano 2b45e8
  if (trans == 'C')  i = 3;
kusano 2b45e8
  if (trans == 'O')  i = 4;
kusano 2b45e8
  if (trans == 'U')  i = 5;
kusano 2b45e8
  if (trans == 'S')  i = 6;
kusano 2b45e8
  if (trans == 'D')  i = 7;
kusano 2b45e8
kusano 2b45e8
  if (incy == 0)      info = 11;
kusano 2b45e8
  if (incx == 0)      info = 8;
kusano 2b45e8
  if (lda < MAX(1,m)) info = 6;
kusano 2b45e8
  if (n < 0) 	      info = 3;
kusano 2b45e8
  if (m < 0) 	      info = 2;
kusano 2b45e8
  if (i < 0)          info = 1;
kusano 2b45e8
kusano 2b45e8
  trans = i;
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,
kusano 2b45e8
	   enum CBLAS_TRANSPOSE TransA,
kusano 2b45e8
	   blasint m, blasint n,
kusano 2b45e8
	   FLOAT *ALPHA,
kusano 2b45e8
	   FLOAT  *a, blasint lda,
kusano 2b45e8
	   FLOAT  *x, blasint incx,
kusano 2b45e8
	   FLOAT *BETA,
kusano 2b45e8
	   FLOAT  *y, blasint incy){
kusano 2b45e8
kusano 2b45e8
  FLOAT *buffer;
kusano 2b45e8
  blasint    lenx, leny;
kusano 2b45e8
  int trans;
kusano 2b45e8
  blasint info, t;
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
  int nthreads;
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  int (*gemv[])(BLASLONG, BLASLONG, BLASLONG, FLOAT, FLOAT, FLOAT *, BLASLONG,
kusano 2b45e8
	    FLOAT * , BLASLONG, FLOAT *, BLASLONG, FLOAT *) = {
kusano 2b45e8
	      GEMV_N, GEMV_T, GEMV_R, GEMV_C,
kusano 2b45e8
	      GEMV_O, GEMV_U, GEMV_S, GEMV_D,
kusano 2b45e8
	    };
kusano 2b45e8
kusano 2b45e8
  PRINT_DEBUG_CNAME;
kusano 2b45e8
kusano 2b45e8
  FLOAT alpha_r = *(ALPHA + 0);
kusano 2b45e8
  FLOAT alpha_i = *(ALPHA + 1);
kusano 2b45e8
kusano 2b45e8
  FLOAT beta_r  = *(BETA + 0);
kusano 2b45e8
  FLOAT beta_i  = *(BETA + 1);
kusano 2b45e8
kusano 2b45e8
  trans = -1;
kusano 2b45e8
  info  =  0;
kusano 2b45e8
kusano 2b45e8
  if (order == CblasColMajor) {
kusano 2b45e8
    if (TransA == CblasNoTrans)     trans = 0;
kusano 2b45e8
    if (TransA == CblasTrans)       trans = 1;
kusano 2b45e8
    if (TransA == CblasConjNoTrans) trans = 2;
kusano 2b45e8
    if (TransA == CblasConjTrans)   trans = 3;
kusano 2b45e8
    
kusano 2b45e8
    info = -1;
kusano 2b45e8
kusano 2b45e8
    if (incy == 0)	  info = 11;
kusano 2b45e8
    if (incx == 0)	  info = 8;
kusano 2b45e8
    if (lda < MAX(1, m))  info = 6;
kusano 2b45e8
    if (n < 0)		  info = 3;
kusano 2b45e8
    if (m < 0)		  info = 2;
kusano 2b45e8
    if (trans < 0)        info = 1;
kusano 2b45e8
    
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  if (order == CblasRowMajor) {
kusano 2b45e8
    if (TransA == CblasNoTrans)     trans = 1;
kusano 2b45e8
    if (TransA == CblasTrans)       trans = 0;
kusano 2b45e8
    if (TransA == CblasConjNoTrans) trans = 3;
kusano 2b45e8
    if (TransA == CblasConjTrans)   trans = 2;
kusano 2b45e8
kusano 2b45e8
    info = -1;
kusano 2b45e8
kusano 2b45e8
    t = n;
kusano 2b45e8
    n = m;
kusano 2b45e8
    m = t;
kusano 2b45e8
kusano 2b45e8
    if (incy == 0)	  info = 11;
kusano 2b45e8
    if (incx == 0)	  info = 8;
kusano 2b45e8
    if (lda < MAX(1, m))  info = 6;
kusano 2b45e8
    if (n < 0)		  info = 3;
kusano 2b45e8
    if (m < 0)		  info = 2;
kusano 2b45e8
    if (trans < 0)        info = 1;
kusano 2b45e8
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
  /*  Quick return if possible. */
kusano 2b45e8
kusano 2b45e8
  if (m == 0 || n == 0) return;
kusano 2b45e8
kusano 2b45e8
  lenx = n;
kusano 2b45e8
  leny = m;
kusano 2b45e8
kusano 2b45e8
  if (trans & 1) lenx = m;
kusano 2b45e8
  if (trans & 1) leny = n;
kusano 2b45e8
kusano 2b45e8
  if (beta_r != ONE || beta_i != ZERO) SCAL_K(leny, 0, 0, beta_r, beta_i, y, abs(incy), NULL, 0, NULL, 0);
kusano 2b45e8
kusano 2b45e8
  if (alpha_r == ZERO && alpha_i == ZERO) return;
kusano 2b45e8
kusano 2b45e8
  IDEBUG_START;
kusano 2b45e8
kusano 2b45e8
  FUNCTION_PROFILE_START();
kusano 2b45e8
kusano 2b45e8
  if (incx < 0) x -= (lenx - 1) * incx * 2;
kusano 2b45e8
  if (incy < 0) y -= (leny - 1) * incy * 2;
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
    (gemv[(int)trans])(m, n, 0, alpha_r, alpha_i, a, lda, x, incx, y, incy, buffer);
kusano 2b45e8
kusano 2b45e8
#ifdef SMP
kusano 2b45e8
    
kusano 2b45e8
  } else {
kusano 2b45e8
    
kusano 2b45e8
    (gemv_thread[(int)trans])(m, n, ALPHA, a, lda, x, incx, y, incy, buffer, nthreads);
kusano 2b45e8
kusano 2b45e8
  }
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
  blas_memory_free(buffer);
kusano 2b45e8
kusano 2b45e8
  FUNCTION_PROFILE_END(4, m * n + m + n,  2 * m * n);
kusano 2b45e8
kusano 2b45e8
  IDEBUG_END;
kusano 2b45e8
kusano 2b45e8
  return;
kusano 2b45e8
}