You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

symm_ucopy_16.c 10 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274
  1. /*********************************************************************/
  2. /* Copyright 2009, 2010 The University of Texas at Austin. */
  3. /* All rights reserved. */
  4. /* */
  5. /* Redistribution and use in source and binary forms, with or */
  6. /* without modification, are permitted provided that the following */
  7. /* conditions are met: */
  8. /* */
  9. /* 1. Redistributions of source code must retain the above */
  10. /* copyright notice, this list of conditions and the following */
  11. /* disclaimer. */
  12. /* */
  13. /* 2. Redistributions in binary form must reproduce the above */
  14. /* copyright notice, this list of conditions and the following */
  15. /* disclaimer in the documentation and/or other materials */
  16. /* provided with the distribution. */
  17. /* */
  18. /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
  19. /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
  20. /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
  21. /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
  22. /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
  23. /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
  24. /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
  25. /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
  26. /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
  27. /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
  28. /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
  29. /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
  30. /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
  31. /* POSSIBILITY OF SUCH DAMAGE. */
  32. /* */
  33. /* The views and conclusions contained in the software and */
  34. /* documentation are those of the authors and should not be */
  35. /* interpreted as representing official policies, either expressed */
  36. /* or implied, of The University of Texas at Austin. */
  37. /*********************************************************************/
  38. #include <stdio.h>
  39. #include "common.h"
  40. int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY, FLOAT *b){
  41. BLASLONG i, js, offset;
  42. FLOAT data01, data02, data03, data04, data05, data06, data07, data08;
  43. FLOAT data09, data10, data11, data12, data13, data14, data15, data16;
  44. FLOAT *ao1, *ao2, *ao3, *ao4, *ao5, *ao6, *ao7, *ao8;
  45. FLOAT *ao9, *ao10, *ao11, *ao12, *ao13, *ao14, *ao15, *ao16;
  46. js = (n >> 4);
  47. while (js > 0){
  48. offset = posX - posY;
  49. if (offset > 0) ao1 = a + posY + (posX + 0) * lda; else ao1 = a + posX + 0 + posY * lda;
  50. if (offset > -1) ao2 = a + posY + (posX + 1) * lda; else ao2 = a + posX + 1 + posY * lda;
  51. if (offset > -2) ao3 = a + posY + (posX + 2) * lda; else ao3 = a + posX + 2 + posY * lda;
  52. if (offset > -3) ao4 = a + posY + (posX + 3) * lda; else ao4 = a + posX + 3 + posY * lda;
  53. if (offset > -4) ao5 = a + posY + (posX + 4) * lda; else ao5 = a + posX + 4 + posY * lda;
  54. if (offset > -5) ao6 = a + posY + (posX + 5) * lda; else ao6 = a + posX + 5 + posY * lda;
  55. if (offset > -6) ao7 = a + posY + (posX + 6) * lda; else ao7 = a + posX + 6 + posY * lda;
  56. if (offset > -7) ao8 = a + posY + (posX + 7) * lda; else ao8 = a + posX + 7 + posY * lda;
  57. if (offset > -8) ao9 = a + posY + (posX + 8) * lda; else ao9 = a + posX + 8 + posY * lda;
  58. if (offset > -9) ao10 = a + posY + (posX + 9) * lda; else ao10 = a + posX + 9 + posY * lda;
  59. if (offset > -10) ao11 = a + posY + (posX + 10) * lda; else ao11 = a + posX + 10 + posY * lda;
  60. if (offset > -11) ao12 = a + posY + (posX + 11) * lda; else ao12 = a + posX + 11 + posY * lda;
  61. if (offset > -12) ao13 = a + posY + (posX + 12) * lda; else ao13 = a + posX + 12 + posY * lda;
  62. if (offset > -13) ao14 = a + posY + (posX + 13) * lda; else ao14 = a + posX + 13 + posY * lda;
  63. if (offset > -14) ao15 = a + posY + (posX + 14) * lda; else ao15 = a + posX + 14 + posY * lda;
  64. if (offset > -15) ao16 = a + posY + (posX + 15) * lda; else ao16 = a + posX + 15 + posY * lda;
  65. i = m;
  66. while (i > 0) {
  67. data01 = *(ao1 + 0);
  68. data02 = *(ao2 + 0);
  69. data03 = *(ao3 + 0);
  70. data04 = *(ao4 + 0);
  71. data05 = *(ao5 + 0);
  72. data06 = *(ao6 + 0);
  73. data07 = *(ao7 + 0);
  74. data08 = *(ao8 + 0);
  75. data09 = *(ao9 + 0);
  76. data10 = *(ao10 + 0);
  77. data11 = *(ao11 + 0);
  78. data12 = *(ao12 + 0);
  79. data13 = *(ao13 + 0);
  80. data14 = *(ao14 + 0);
  81. data15 = *(ao15 + 0);
  82. data16 = *(ao16 + 0);
  83. if (offset > 0) ao1 ++; else ao1 += lda;
  84. if (offset > -1) ao2 ++; else ao2 += lda;
  85. if (offset > -2) ao3 ++; else ao3 += lda;
  86. if (offset > -3) ao4 ++; else ao4 += lda;
  87. if (offset > -4) ao5 ++; else ao5 += lda;
  88. if (offset > -5) ao6 ++; else ao6 += lda;
  89. if (offset > -6) ao7 ++; else ao7 += lda;
  90. if (offset > -7) ao8 ++; else ao8 += lda;
  91. if (offset > -8) ao9 ++; else ao9 += lda;
  92. if (offset > -9) ao10 ++; else ao10 += lda;
  93. if (offset > -10) ao11 ++; else ao11 += lda;
  94. if (offset > -11) ao12 ++; else ao12 += lda;
  95. if (offset > -12) ao13 ++; else ao13 += lda;
  96. if (offset > -13) ao14 ++; else ao14 += lda;
  97. if (offset > -14) ao15 ++; else ao15 += lda;
  98. if (offset > -15) ao16 ++; else ao16 += lda;
  99. b[ 0] = data01;
  100. b[ 1] = data02;
  101. b[ 2] = data03;
  102. b[ 3] = data04;
  103. b[ 4] = data05;
  104. b[ 5] = data06;
  105. b[ 6] = data07;
  106. b[ 7] = data08;
  107. b[ 8] = data09;
  108. b[ 9] = data10;
  109. b[10] = data11;
  110. b[11] = data12;
  111. b[12] = data13;
  112. b[13] = data14;
  113. b[14] = data15;
  114. b[15] = data16;
  115. b += 16;
  116. offset --;
  117. i --;
  118. }
  119. posX += 16;
  120. js --;
  121. }
  122. if (n & 8) {
  123. offset = posX - posY;
  124. if (offset > 0) ao1 = a + posY + (posX + 0) * lda; else ao1 = a + posX + 0 + posY * lda;
  125. if (offset > -1) ao2 = a + posY + (posX + 1) * lda; else ao2 = a + posX + 1 + posY * lda;
  126. if (offset > -2) ao3 = a + posY + (posX + 2) * lda; else ao3 = a + posX + 2 + posY * lda;
  127. if (offset > -3) ao4 = a + posY + (posX + 3) * lda; else ao4 = a + posX + 3 + posY * lda;
  128. if (offset > -4) ao5 = a + posY + (posX + 4) * lda; else ao5 = a + posX + 4 + posY * lda;
  129. if (offset > -5) ao6 = a + posY + (posX + 5) * lda; else ao6 = a + posX + 5 + posY * lda;
  130. if (offset > -6) ao7 = a + posY + (posX + 6) * lda; else ao7 = a + posX + 6 + posY * lda;
  131. if (offset > -7) ao8 = a + posY + (posX + 7) * lda; else ao8 = a + posX + 7 + posY * lda;
  132. i = m;
  133. while (i > 0) {
  134. data01 = *(ao1 + 0);
  135. data02 = *(ao2 + 0);
  136. data03 = *(ao3 + 0);
  137. data04 = *(ao4 + 0);
  138. data05 = *(ao5 + 0);
  139. data06 = *(ao6 + 0);
  140. data07 = *(ao7 + 0);
  141. data08 = *(ao8 + 0);
  142. if (offset > 0) ao1 ++; else ao1 += lda;
  143. if (offset > -1) ao2 ++; else ao2 += lda;
  144. if (offset > -2) ao3 ++; else ao3 += lda;
  145. if (offset > -3) ao4 ++; else ao4 += lda;
  146. if (offset > -4) ao5 ++; else ao5 += lda;
  147. if (offset > -5) ao6 ++; else ao6 += lda;
  148. if (offset > -6) ao7 ++; else ao7 += lda;
  149. if (offset > -7) ao8 ++; else ao8 += lda;
  150. b[ 0] = data01;
  151. b[ 1] = data02;
  152. b[ 2] = data03;
  153. b[ 3] = data04;
  154. b[ 4] = data05;
  155. b[ 5] = data06;
  156. b[ 6] = data07;
  157. b[ 7] = data08;
  158. b += 8;
  159. offset --;
  160. i --;
  161. }
  162. posX += 8;
  163. }
  164. if (n & 4) {
  165. offset = posX - posY;
  166. if (offset > 0) ao1 = a + posY + (posX + 0) * lda; else ao1 = a + posX + 0 + posY * lda;
  167. if (offset > -1) ao2 = a + posY + (posX + 1) * lda; else ao2 = a + posX + 1 + posY * lda;
  168. if (offset > -2) ao3 = a + posY + (posX + 2) * lda; else ao3 = a + posX + 2 + posY * lda;
  169. if (offset > -3) ao4 = a + posY + (posX + 3) * lda; else ao4 = a + posX + 3 + posY * lda;
  170. i = m;
  171. while (i > 0) {
  172. data01 = *(ao1 + 0);
  173. data02 = *(ao2 + 0);
  174. data03 = *(ao3 + 0);
  175. data04 = *(ao4 + 0);
  176. if (offset > 0) ao1 ++; else ao1 += lda;
  177. if (offset > -1) ao2 ++; else ao2 += lda;
  178. if (offset > -2) ao3 ++; else ao3 += lda;
  179. if (offset > -3) ao4 ++; else ao4 += lda;
  180. b[ 0] = data01;
  181. b[ 1] = data02;
  182. b[ 2] = data03;
  183. b[ 3] = data04;
  184. b += 4;
  185. offset --;
  186. i --;
  187. }
  188. posX += 4;
  189. }
  190. if (n & 2) {
  191. offset = posX - posY;
  192. if (offset > 0) ao1 = a + posY + (posX + 0) * lda; else ao1 = a + posX + 0 + posY * lda;
  193. if (offset > -1) ao2 = a + posY + (posX + 1) * lda; else ao2 = a + posX + 1 + posY * lda;
  194. i = m;
  195. while (i > 0) {
  196. data01 = *(ao1 + 0);
  197. data02 = *(ao2 + 0);
  198. if (offset > 0) ao1 ++; else ao1 += lda;
  199. if (offset > -1) ao2 ++; else ao2 += lda;
  200. b[ 0] = data01;
  201. b[ 1] = data02;
  202. b += 2;
  203. offset --;
  204. i --;
  205. }
  206. posX += 2;
  207. }
  208. if (n & 1) {
  209. offset = posX - posY;
  210. if (offset > 0) ao1 = a + posY + (posX + 0) * lda; else ao1 = a + posX + 0 + posY * lda;
  211. i = m;
  212. while (i > 0) {
  213. data01 = *(ao1 + 0);
  214. if (offset > 0) ao1 ++; else ao1 += lda;
  215. b[ 0] = data01;
  216. b ++;
  217. offset --;
  218. i --;
  219. }
  220. }
  221. return 0;
  222. }