OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
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.
Tree: 64629cb5c7
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '64629cb5c7'
${ noResults }
OpenBLAS/kernel/x86_64/zasum_sse2.S

319 lines
6.7 kB
Raw Blame History 

  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. 

  39. #define ASSEMBLER

  40. #include "common.h"

  41. 

  42. #define M	ARG1	/* rdi */

  43. #define X	ARG2	/* rsi */

  44. #define INCX	ARG3	/* rdx */

  45. 

  46. #define I	%rax

  47. 

  48. #include "l1param.h"

  49. 

  50. 	PROLOGUE

  51. 	PROFCODE

  52. 

  53. 	SAVEREGISTERS

  54. 

  55. 	xorps	%xmm0, %xmm0

  56. 	testq	M, M

  57. 	jle	.L999

  58. 	testq	INCX, INCX

  59. 	jle	.L999

  60. 

  61. 	xorps	%xmm1, %xmm1

  62. 	xorps	%xmm2, %xmm2

  63. 	xorps	%xmm3, %xmm3

  64. 

  65. 	pcmpeqb	%xmm15, %xmm15

  66. 	psrlq	$1, %xmm15

  67. 

  68. 	salq	$ZBASE_SHIFT, INCX

  69. 

  70. 	cmpq	$2 * SIZE, INCX

  71. 	jne	.L40

  72. 

  73. 	subq	$-16 * SIZE, X

  74. 	addq	M, M

  75. 

  76. 	testq	$SIZE, X

  77. 	je	.L05

  78. 

  79. #ifdef movsd

  80. 	xorps	%xmm0, %xmm0

  81. #endif

  82. 	movsd	-16 * SIZE(X), %xmm0

  83. 	addq	$SIZE, X

  84. 

  85. 	andps	%xmm15, %xmm0

  86. 	subq	$1, M

  87. 	jle	.L999

  88. 	ALIGN_3

  89. 

  90. .L05:

  91. 	movq	M,  I

  92. 	sarq	$4, I

  93. 	jle	.L20

  94. 

  95. 	movaps	-16 * SIZE(X), %xmm4

  96. 	movaps	-14 * SIZE(X), %xmm5

  97. 	movaps	-12 * SIZE(X), %xmm6

  98. 	movaps	-10 * SIZE(X), %xmm7

  99. 

  100. 	movaps	 -8 * SIZE(X), %xmm8

  101. 	movaps	 -6 * SIZE(X), %xmm9

  102. 	movaps	 -4 * SIZE(X), %xmm10

  103. 	movaps	 -2 * SIZE(X), %xmm11

  104. 

  105. 	decq	I

  106. 	jle	.L11

  107. 	ALIGN_4

  108. 

  109. .L10:

  110. #ifdef PREFETCH

  111. 	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(X)

  112. #endif

  113. 

  114. 	andps	%xmm15, %xmm4

  115. 	addpd	%xmm4,  %xmm0

  116. 	movaps	  0 * SIZE(X), %xmm4

  117. 

  118. 	andps	%xmm15, %xmm5

  119. 	addpd	%xmm5,  %xmm1

  120. 	movaps	  2 * SIZE(X), %xmm5

  121. 

  122. 	andps	%xmm15, %xmm6

  123. 	addpd	%xmm6,  %xmm2

  124. 	movaps	  4 * SIZE(X), %xmm6

  125. 

  126. 	andps	%xmm15, %xmm7

  127. 	addpd	%xmm7,  %xmm3

  128. 	movaps	  6 * SIZE(X), %xmm7

  129. 

  130. #if defined(PREFETCH) && !defined(FETCH128)

  131. 	PREFETCH (PREFETCHSIZE +  64) - PREOFFSET(X)

  132. #endif

  133. 

  134. 	andps	%xmm15, %xmm8

  135. 	addpd	%xmm8,  %xmm0

  136. 	movaps	  8 * SIZE(X), %xmm8

  137. 

  138. 	andps	%xmm15, %xmm9

  139. 	addpd	%xmm9,  %xmm1

  140. 	movaps	 10 * SIZE(X), %xmm9

  141. 

  142. 	andps	%xmm15, %xmm10

  143. 	addpd	%xmm10, %xmm2

  144. 	movaps	 12 * SIZE(X), %xmm10

  145. 

  146. 	andps	%xmm15, %xmm11

  147. 	addpd	%xmm11, %xmm3

  148. 	movaps	 14 * SIZE(X), %xmm11

  149. 

  150. 	subq	$-16 * SIZE, X

  151. 	decq	I

  152. 	jg	.L10

  153. 	ALIGN_4

  154. 

  155. .L11:

  156. 	andps	%xmm15, %xmm4

  157. 	andps	%xmm15, %xmm5

  158. 	andps	%xmm15, %xmm6

  159. 	andps	%xmm15, %xmm7

  160. 

  161. 	addpd	%xmm4,  %xmm0

  162. 	addpd	%xmm5,  %xmm1

  163. 	addpd	%xmm6,  %xmm2

  164. 	addpd	%xmm7,  %xmm3

  165. 

  166. 	andps	%xmm15, %xmm8

  167. 	andps	%xmm15, %xmm9

  168. 	andps	%xmm15, %xmm10

  169. 	andps	%xmm15, %xmm11

  170. 

  171. 	addpd	%xmm8,  %xmm0

  172. 	addpd	%xmm9,  %xmm1

  173. 	addpd	%xmm10, %xmm2

  174. 	addpd	%xmm11, %xmm3

  175. 

  176. 	subq	$-16 * SIZE, X

  177. 	ALIGN_3

  178. 

  179. .L20:

  180. 	andq	$15,  M

  181. 	jle	.L998

  182. 

  183. 	testq	$8, M

  184. 	je	.L21

  185. 

  186. 	movaps	-16 * SIZE(X), %xmm4

  187. 	movaps	-14 * SIZE(X), %xmm5

  188. 	movaps	-12 * SIZE(X), %xmm6

  189. 	movaps	-10 * SIZE(X), %xmm7

  190. 

  191. 	andps	%xmm15, %xmm4

  192. 	andps	%xmm15, %xmm5

  193. 	andps	%xmm15, %xmm6

  194. 	andps	%xmm15, %xmm7

  195. 

  196. 	addpd	%xmm4, %xmm0

  197. 	addpd	%xmm5, %xmm1

  198. 	addpd	%xmm6, %xmm2

  199. 	addpd	%xmm7, %xmm3

  200. 	addq	$8 * SIZE, X

  201. 	ALIGN_3

  202. 

  203. .L21:

  204. 	testq	$4, M

  205. 	je	.L22

  206. 

  207. 	movaps	-16 * SIZE(X), %xmm4

  208. 	movaps	-14 * SIZE(X), %xmm5

  209. 

  210. 	andps	%xmm15, %xmm4

  211. 	andps	%xmm15, %xmm5

  212. 	addpd	%xmm4, %xmm0

  213. 	addpd	%xmm5, %xmm1

  214. 

  215. 	addq	$4 * SIZE, X

  216. 	ALIGN_3

  217. 

  218. .L22:

  219. 	testq	$2, M

  220. 	je	.L23

  221. 

  222. 	movaps	-16 * SIZE(X), %xmm6

  223. 	andps	%xmm15, %xmm6

  224. 	addpd	%xmm6, %xmm3

  225. 	addq	$2 * SIZE, X

  226. 

  227. .L23:

  228. 	testq	$1, M

  229. 	je	.L998

  230. 

  231. #ifdef movsd

  232. 	xorps	%xmm4, %xmm4

  233. #endif

  234. 	movsd	-16 * SIZE(X), %xmm4

  235. 	andps	%xmm15, %xmm4

  236. 	addsd	%xmm4, %xmm0

  237. 	jmp	.L998

  238. 	ALIGN_3

  239. 

  240. 

  241. .L40:

  242. 	movq	M,  I

  243. 	sarq	$2, I

  244. 	jle	.L60

  245. 	ALIGN_4

  246. 

  247. .L50:

  248. #if defined(OPTERON) || defined(BARCELONA) || defined(SHANGHAI)

  249. 	prefetcht0	PREFETCHSIZE * SIZE(X)

  250. #endif

  251. 

  252. #ifdef PENTIUM4

  253. 	prefetchnta	PREFETCHSIZE * SIZE(X)

  254. #endif

  255. 

  256. 	movsd	0 * SIZE(X), %xmm4

  257. 	movhpd	1 * SIZE(X), %xmm4

  258. 	addq	INCX, X

  259. 	andpd	%xmm15, %xmm4

  260. 	addpd	%xmm4, %xmm0

  261. 

  262. 	movsd	0 * SIZE(X), %xmm5

  263. 	movhpd	1 * SIZE(X), %xmm5

  264. 	addq	INCX, X

  265. 	andpd	%xmm15, %xmm5

  266. 	addpd	%xmm5, %xmm1

  267. 

  268. 	movsd	0 * SIZE(X), %xmm6

  269. 	movhpd	1 * SIZE(X), %xmm6

  270. 	addq	INCX, X

  271. 	andpd	%xmm15, %xmm6

  272. 	addpd	%xmm6, %xmm2

  273. 

  274. 	movsd	0 * SIZE(X), %xmm7

  275. 	movhpd	1 * SIZE(X), %xmm7

  276. 	addq	INCX, X

  277. 	andpd	%xmm15, %xmm7

  278. 	addpd	%xmm7, %xmm3

  279. 

  280. 	decq	I

  281. 	jg	.L50

  282. 	ALIGN_4

  283. 

  284. .L60:

  285. 	andq	$3,  M

  286. 	jle	.L998

  287. 	ALIGN_4

  288. 

  289. 

  290. .L61:

  291. 	movsd	0 * SIZE(X), %xmm4

  292. 	movhpd	1 * SIZE(X), %xmm4

  293. 	andpd	%xmm15, %xmm4

  294. 	addpd	%xmm4, %xmm0

  295. 	addq	INCX, X

  296. 	decq	M

  297. 	jg	.L61

  298. 	ALIGN_4

  299. 

  300. .L998:

  301. 	addpd	%xmm1, %xmm0

  302. 	addpd	%xmm3, %xmm2

  303. 	addpd	%xmm2, %xmm0

  304. 

  305. #ifndef HAVE_SSE3

  306. 	movhlps	%xmm0, %xmm1

  307. 	addsd	%xmm1, %xmm0

  308. #else

  309. 	haddpd	%xmm0, %xmm0

  310. #endif

  311. 	ALIGN_4

  312. 

  313. .L999:

  314. 	RESTOREREGISTERS

  315. 

  316. 	ret

  317. 

  318. 	EPILOGUE