Blame thirdparty/openblas/xianyi-OpenBLAS-e6e87a2/kernel/generic/trmm_utcopy_2.c

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
int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){
kusano 2b45e8
kusano 2b45e8
  BLASLONG i, js;
kusano 2b45e8
  BLASLONG X;
kusano 2b45e8
kusano 2b45e8
  FLOAT data01, data02, data03, data04;
kusano 2b45e8
  FLOAT *ao1, *ao2;
kusano 2b45e8
kusano 2b45e8
  js = (n >> 1);
kusano 2b45e8
kusano 2b45e8
  if (js > 0){
kusano 2b45e8
    do {
kusano 2b45e8
      X = posX;
kusano 2b45e8
kusano 2b45e8
      if (posX <= posY) {
kusano 2b45e8
	ao1 = a + posX + (posY + 0) * lda;
kusano 2b45e8
	ao2 = a + posX + (posY + 1) * lda;
kusano 2b45e8
      } else {
kusano 2b45e8
	ao1 = a + posY + (posX + 0) * lda;
kusano 2b45e8
	ao2 = a + posY + (posX + 1) * lda;
kusano 2b45e8
      }
kusano 2b45e8
kusano 2b45e8
      i = (m >> 1);
kusano 2b45e8
      if (i > 0) {
kusano 2b45e8
	do {
kusano 2b45e8
	  if (X < posY) {
kusano 2b45e8
	    ao1 += 2;
kusano 2b45e8
	    ao2 += 2;
kusano 2b45e8
	    b   += 4;
kusano 2b45e8
kusano 2b45e8
	  } else 
kusano 2b45e8
	    if (X > posY) {
kusano 2b45e8
	      data01 = *(ao1 + 0);
kusano 2b45e8
	      data02 = *(ao1 + 1);
kusano 2b45e8
	      data03 = *(ao2 + 0);
kusano 2b45e8
	      data04 = *(ao2 + 1);
kusano 2b45e8
	  
kusano 2b45e8
	      b[ 0] = data01;
kusano 2b45e8
	      b[ 1] = data02;
kusano 2b45e8
	      b[ 2] = data03;
kusano 2b45e8
	      b[ 3] = data04;
kusano 2b45e8
kusano 2b45e8
	      ao1 += 2 * lda;
kusano 2b45e8
	      ao2 += 2 * lda;
kusano 2b45e8
	      b += 4;
kusano 2b45e8
kusano 2b45e8
	    } else {
kusano 2b45e8
#ifdef UNIT	      
kusano 2b45e8
	      data03 = *(ao2 + 0);
kusano 2b45e8
	  
kusano 2b45e8
	      b[ 0] = ONE;
kusano 2b45e8
	      b[ 1] = ZERO;
kusano 2b45e8
	      b[ 2] = data03;
kusano 2b45e8
	      b[ 3] = ONE;
kusano 2b45e8
#else
kusano 2b45e8
	      data01 = *(ao1 + 0);
kusano 2b45e8
	      data03 = *(ao2 + 0);
kusano 2b45e8
	      data04 = *(ao2 + 1);
kusano 2b45e8
	  
kusano 2b45e8
	      b[ 0] = data01;
kusano 2b45e8
	      b[ 1] = ZERO;
kusano 2b45e8
	      b[ 2] = data03;
kusano 2b45e8
	      b[ 3] = data04;
kusano 2b45e8
#endif
kusano 2b45e8
kusano 2b45e8
	      ao1 += 2 * lda;
kusano 2b45e8
	      ao2 += 2 * lda;
kusano 2b45e8
	      b += 4;
kusano 2b45e8
	    }
kusano 2b45e8
kusano 2b45e8
	  X += 2;
kusano 2b45e8
	  i --;
kusano 2b45e8
	} while (i > 0);
kusano 2b45e8
      }
kusano 2b45e8
kusano 2b45e8
      if (m & 1) {
kusano 2b45e8
      
kusano 2b45e8
	if (X < posY) {
kusano 2b45e8
	  ao1 += 1;
kusano 2b45e8
	  ao2 += 1;
kusano 2b45e8
	  b += 2;
kusano 2b45e8
	} else 
kusano 2b45e8
	  if (X > posY) {
kusano 2b45e8
	    data01 = *(ao1 + 0);
kusano 2b45e8
	    data02 = *(ao1 + 1);
kusano 2b45e8
	  
kusano 2b45e8
	    b[ 0] = data01;
kusano 2b45e8
	    b[ 1] = data02;
kusano 2b45e8
	    ao1 += lda;
kusano 2b45e8
	    b += 2;
kusano 2b45e8
	  } else {
kusano 2b45e8
#ifdef UNIT	      
kusano 2b45e8
	    b[ 0] = ONE;
kusano 2b45e8
	    b[ 1] = ZERO;
kusano 2b45e8
#else
kusano 2b45e8
	    data01 = *(ao1 + 0);
kusano 2b45e8
	    
kusano 2b45e8
	    b[ 0] = data01;
kusano 2b45e8
	    b[ 1] = ZERO;
kusano 2b45e8
#endif
kusano 2b45e8
	    ao1 += lda;
kusano 2b45e8
	    b += 2;
kusano 2b45e8
	  }
kusano 2b45e8
      }
kusano 2b45e8
kusano 2b45e8
      posY += 2;
kusano 2b45e8
      js --;
kusano 2b45e8
    } while (js > 0);
kusano 2b45e8
  } /* End of main loop */
kusano 2b45e8
kusano 2b45e8
  if (n & 1){
kusano 2b45e8
    X = posX;
kusano 2b45e8
    
kusano 2b45e8
    if (posX <= posY) {
kusano 2b45e8
      ao1 = a + posX + (posY + 0) * lda;
kusano 2b45e8
    } else {
kusano 2b45e8
      ao1 = a + posY + (posX + 0) * lda;
kusano 2b45e8
    }
kusano 2b45e8
    
kusano 2b45e8
    i = m;
kusano 2b45e8
    if (m > 0) {
kusano 2b45e8
      do {
kusano 2b45e8
	if (X < posY) {
kusano 2b45e8
	  b += 1;
kusano 2b45e8
  	  ao1 += 1;
kusano 2b45e8
	} else
kusano 2b45e8
	  if (X > posY) {
kusano 2b45e8
	    data01 = *(ao1 + 0);
kusano 2b45e8
	    b[ 0] = data01;
kusano 2b45e8
	    b += 1;
kusano 2b45e8
	    ao1 += lda;
kusano 2b45e8
	  } else {
kusano 2b45e8
#ifdef UNIT
kusano 2b45e8
	    b[ 0] = ONE;
kusano 2b45e8
#else
kusano 2b45e8
	    data01 = *(ao1 + 0);
kusano 2b45e8
	    b[ 0] = data01;
kusano 2b45e8
#endif
kusano 2b45e8
	    b += 1;
kusano 2b45e8
	    ao1 += lda;
kusano 2b45e8
	  }
kusano 2b45e8
	
kusano 2b45e8
	X += 1;
kusano 2b45e8
	i --;
kusano 2b45e8
      } while (i > 0);
kusano 2b45e8
    }
kusano 2b45e8
  }
kusano 2b45e8
kusano 2b45e8
  return 0;
kusano 2b45e8
}