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
kusano 2b45e8
#ifndef USE_ALPHA
kusano 2b45e8
#define REAL_PART(a, b)  (a)
kusano 2b45e8
#define IMAGE_PART(a, b) (b)
kusano 2b45e8
#else
kusano 2b45e8
#define REAL_PART(a, b)  (alpha_r * (a) - alpha_i * (b))
kusano 2b45e8
#define IMAGE_PART(a, b) (alpha_i * (a) + alpha_r * (b))
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
#if defined(REAL_ONLY)
kusano 2b45e8
#define CMULT(a, b) (REAL_PART(a, b))
kusano 2b45e8
#elif defined(IMAGE_ONLY)
kusano 2b45e8
#define CMULT(a, b) (IMAGE_PART(a, b))
kusano 2b45e8
#else
kusano 2b45e8
#define CMULT(a, b) (REAL_PART(a, b) + IMAGE_PART(a, b))
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda,
kusano 2b45e8
#ifdef USE_ALPHA
kusano 2b45e8
	   FLOAT alpha_r, FLOAT alpha_i, 
kusano 2b45e8
#endif
kusano 2b45e8
	   FLOAT *b){
kusano 2b45e8
kusano 2b45e8
  BLASLONG i, j;
kusano 2b45e8
kusano 2b45e8
  FLOAT *a_offset, *a_offset1, *a_offset2, *a_offset3, *a_offset4;
kusano 2b45e8
  FLOAT *b_offset;
kusano 2b45e8
  FLOAT a1, a2, a3, a4, a5, a6, a7, a8;
kusano 2b45e8
kusano 2b45e8
  lda *= 2;
kusano 2b45e8
kusano 2b45e8
  a_offset = a;
kusano 2b45e8
  b_offset = b;
kusano 2b45e8
  
kusano 2b45e8
  j = (n >> 2);
kusano 2b45e8
  if (j > 0){
kusano 2b45e8
    do{
kusano 2b45e8
      a_offset1  = a_offset;
kusano 2b45e8
      a_offset2  = a_offset1 + lda;
kusano 2b45e8
      a_offset3  = a_offset2 + lda;
kusano 2b45e8
      a_offset4  = a_offset3 + lda;
kusano 2b45e8
      a_offset += 4 * lda;
kusano 2b45e8
      
kusano 2b45e8
      for (i = 0; i < m; i ++) {
kusano 2b45e8
kusano 2b45e8
	a1 = *(a_offset1 + 0);
kusano 2b45e8
	a2 = *(a_offset1 + 1);
kusano 2b45e8
	a3 = *(a_offset2 + 0);
kusano 2b45e8
	a4 = *(a_offset2 + 1);
kusano 2b45e8
	a5 = *(a_offset3 + 0);
kusano 2b45e8
	a6 = *(a_offset3 + 1);
kusano 2b45e8
	a7 = *(a_offset4 + 0);
kusano 2b45e8
	a8 = *(a_offset4 + 1);
kusano 2b45e8
kusano 2b45e8
	*(b_offset +  0) = CMULT(a1, a2);
kusano 2b45e8
	*(b_offset +  1) = CMULT(a3, a4);
kusano 2b45e8
	*(b_offset +  2) = CMULT(a5, a6);
kusano 2b45e8
	*(b_offset +  3) = CMULT(a7, a8);
kusano 2b45e8
kusano 2b45e8
	a_offset1 += 2;
kusano 2b45e8
	a_offset2 += 2;
kusano 2b45e8
	a_offset3 += 2;
kusano 2b45e8
	a_offset4 += 2;
kusano 2b45e8
	  
kusano 2b45e8
	b_offset +=  4;
kusano 2b45e8
kusano 2b45e8
      }
kusano 2b45e8
kusano 2b45e8
      j--;
kusano 2b45e8
    }while(j > 0);
kusano 2b45e8
  } /* end of if(j > 0) */
kusano 2b45e8
  
kusano 2b45e8
  if (n & 2) {
kusano 2b45e8
    a_offset1  = a_offset;
kusano 2b45e8
    a_offset2  = a_offset1 + lda;
kusano 2b45e8
    a_offset += 2 * lda;
kusano 2b45e8
      
kusano 2b45e8
      for (i = 0; i < m; i ++) {
kusano 2b45e8
kusano 2b45e8
	a1 = *(a_offset1 + 0);
kusano 2b45e8
	a2 = *(a_offset1 + 1);
kusano 2b45e8
	a3 = *(a_offset2 + 0);
kusano 2b45e8
	a4 = *(a_offset2 + 1);
kusano 2b45e8
kusano 2b45e8
	*(b_offset +  0) = CMULT(a1, a2);
kusano 2b45e8
	*(b_offset +  1) = CMULT(a3, a4);
kusano 2b45e8
kusano 2b45e8
	a_offset1 += 2;
kusano 2b45e8
	a_offset2 += 2;
kusano 2b45e8
	  
kusano 2b45e8
	b_offset +=  2;
kusano 2b45e8
kusano 2b45e8
      }
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  if (n & 1) {
kusano 2b45e8
    a_offset1  = a_offset;
kusano 2b45e8
      
kusano 2b45e8
    for (i = 0; i < m; i ++) {
kusano 2b45e8
      
kusano 2b45e8
      a1 = *(a_offset1 + 0);
kusano 2b45e8
      a2 = *(a_offset1 + 1);
kusano 2b45e8
      
kusano 2b45e8
      *(b_offset +  0) = CMULT(a1, a2);
kusano 2b45e8
      
kusano 2b45e8
      a_offset1 += 2;
kusano 2b45e8
      
kusano 2b45e8
      b_offset +=  1;
kusano 2b45e8
    }
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  return 0;
kusano 2b45e8
}