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: 549a9f1dbb
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '549a9f1dbb'
${ noResults }
OpenBLAS/kernel/x86_64/amax_sse.S

476 lines
8.5 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. #ifdef USE_MIN

  49. #define maxps	minps

  50. #define maxss	minss

  51. #endif

  52. 

  53. #include "l1param.h"

  54. 

  55. 	PROLOGUE

  56. 	PROFCODE

  57. 

  58. 	SAVEREGISTERS

  59. 

  60. 	xorps	%xmm0, %xmm0

  61. 	leaq	(, INCX, SIZE), INCX

  62. 

  63. 	testq	M, M

  64. 	jle	.L999

  65. 

  66. #ifdef USE_ABS

  67. 	pcmpeqb	%xmm15, %xmm15

  68. 	psrld	$1,    %xmm15

  69. #endif

  70. 

  71. 	movss	(X), %xmm0

  72. 	shufps	$0,  %xmm0, %xmm0

  73. #ifdef USE_ABS

  74. 	andps	%xmm15, %xmm0

  75. #endif

  76. 	movaps	%xmm0, %xmm1

  77. 	movaps	%xmm0, %xmm2

  78. 	movaps	%xmm0, %xmm3

  79. 	addq	INCX, X

  80. 	decq	M

  81. 	jle	.L999

  82. 

  83. 	cmpq	$SIZE, INCX

  84. 	jne	.L40

  85. 

  86. 	subq	$-32 * SIZE, X

  87. 

  88. 	cmpq	$3, M

  89. 	jle	.L17

  90. 

  91. 	testq	$SIZE, X

  92. 	je	.L05

  93. 

  94. 	movss	-32 * SIZE(X), %xmm1

  95. 	shufps	$0, %xmm1, %xmm1

  96. #ifdef USE_ABS

  97. 	andps	%xmm15, %xmm1

  98. #endif

  99. 	decq	M

  100. 	addq	$SIZE, X

  101. 	ALIGN_3

  102. 

  103. .L05:

  104. 	testq	$2 * SIZE, X

  105. 	je	.L06

  106. 

  107. 	movsd	-32 * SIZE(X), %xmm2

  108. 	unpcklps  %xmm2, %xmm2

  109. #ifdef USE_ABS

  110. 	andps	%xmm15, %xmm2

  111. #endif

  112. 	subq	$2, M

  113. 	addq	$2 * SIZE, X

  114. 	ALIGN_3

  115. 

  116. .L06:

  117. 	movq	M,  I

  118. 	sarq	$5, I

  119. 	jle	.L15

  120. 

  121. 	movaps	-32 * SIZE(X), %xmm4

  122. 	movaps	-28 * SIZE(X), %xmm5

  123. 	movaps	-24 * SIZE(X), %xmm6

  124. 	movaps	-20 * SIZE(X), %xmm7

  125. 

  126. 	decq	I

  127. 	jle	.L12

  128. 	ALIGN_4

  129. 

  130. .L11:

  131. #ifdef PREFETCH

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

  133. #endif

  134. 

  135. #ifdef USE_ABS

  136. 	andps	%xmm15, %xmm4

  137. #endif

  138. 	maxps	%xmm4, %xmm0

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

  140. 

  141. #ifdef USE_ABS

  142. 	andps	%xmm15, %xmm5

  143. #endif

  144. 	maxps	%xmm5, %xmm1

  145. 	movaps	-12 * SIZE(X), %xmm5

  146. 

  147. #ifdef USE_ABS

  148. 	andps	%xmm15, %xmm6

  149. #endif

  150. 	maxps	%xmm6, %xmm2

  151. 	movaps	 -8 * SIZE(X), %xmm6

  152. 

  153. #ifdef USE_ABS

  154. 	andps	%xmm15, %xmm7

  155. #endif

  156. 	maxps	%xmm7, %xmm3

  157. 	movaps	 -4 * SIZE(X), %xmm7

  158. 

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

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

  161. #endif

  162. 

  163. #ifdef USE_ABS

  164. 	andps	%xmm15, %xmm4

  165. #endif

  166. 	maxps	%xmm4, %xmm0

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

  168. 

  169. #ifdef USE_ABS

  170. 	andps	%xmm15, %xmm5

  171. #endif

  172. 	maxps	%xmm5, %xmm1

  173. 	movaps	  4 * SIZE(X), %xmm5

  174. 

  175. #ifdef USE_ABS

  176. 	andps	%xmm15, %xmm6

  177. #endif

  178. 	maxps	%xmm6, %xmm2

  179. 	movaps	  8 * SIZE(X), %xmm6

  180. 

  181. #ifdef USE_ABS

  182. 	andps	%xmm15, %xmm7

  183. #endif

  184. 	maxps	%xmm7, %xmm3

  185. 	movaps	 12 * SIZE(X), %xmm7

  186. 

  187. 	subq	$-32 * SIZE, X

  188. 	decq	I

  189. 	jg	.L11

  190. 	ALIGN_4

  191. 

  192. .L12:

  193. #ifdef USE_ABS

  194. 	andps	%xmm15, %xmm4

  195. #endif

  196. 	maxps	%xmm4, %xmm0

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

  198. 

  199. #ifdef USE_ABS

  200. 	andps	%xmm15, %xmm5

  201. #endif

  202. 	maxps	%xmm5, %xmm1

  203. 	movaps	-12 * SIZE(X), %xmm5

  204. 

  205. #ifdef USE_ABS

  206. 	andps	%xmm15, %xmm6

  207. #endif

  208. 	maxps	%xmm6, %xmm2

  209. 	movaps	 -8 * SIZE(X), %xmm6

  210. 

  211. #ifdef USE_ABS

  212. 	andps	%xmm15, %xmm7

  213. #endif

  214. 	maxps	%xmm7, %xmm3

  215. 	movaps	 -4 * SIZE(X), %xmm7

  216. 

  217. #ifdef USE_ABS

  218. 	andps	%xmm15, %xmm4

  219. #endif

  220. 	maxps	%xmm4, %xmm0

  221. 

  222. #ifdef USE_ABS

  223. 	andps	%xmm15, %xmm5

  224. #endif

  225. 	maxps	%xmm5, %xmm1

  226. 

  227. #ifdef USE_ABS

  228. 	andps	%xmm15, %xmm6

  229. #endif

  230. 	maxps	%xmm6, %xmm2

  231. 

  232. #ifdef USE_ABS

  233. 	andps	%xmm15, %xmm7

  234. #endif

  235. 	maxps	%xmm7, %xmm3

  236. 

  237. 	subq	$-32 * SIZE, X

  238. 	ALIGN_3

  239. 

  240. 

  241. .L15:

  242. 	testq	$16, M

  243. 	je	.L16

  244. 

  245. 	movaps	-32 * SIZE(X), %xmm4

  246. #ifdef USE_ABS

  247. 	andps	%xmm15, %xmm4

  248. #endif

  249. 	maxps	%xmm4, %xmm0

  250. 

  251. 	movaps	-28 * SIZE(X), %xmm5

  252. #ifdef USE_ABS

  253. 	andps	%xmm15, %xmm5

  254. #endif

  255. 	maxps	%xmm5, %xmm1

  256. 

  257. 	movaps	-24 * SIZE(X), %xmm6

  258. #ifdef USE_ABS

  259. 	andps	%xmm15, %xmm6

  260. #endif

  261. 	maxps	%xmm6, %xmm2

  262. 

  263. 	movaps	-20 * SIZE(X), %xmm7

  264. #ifdef USE_ABS

  265. 	andps	%xmm15, %xmm7

  266. #endif

  267. 	maxps	%xmm7, %xmm3

  268. 

  269. 	addq	$16 * SIZE, X

  270. 	ALIGN_3

  271. 

  272. .L16:

  273. 	testq	$8, M

  274. 	je	.L17

  275. 

  276. 	movaps	-32 * SIZE(X), %xmm4

  277. #ifdef USE_ABS

  278. 	andps	%xmm15, %xmm4

  279. #endif

  280. 	maxps	%xmm4, %xmm0

  281. 

  282. 	movaps	-28 * SIZE(X), %xmm5

  283. #ifdef USE_ABS

  284. 	andps	%xmm15, %xmm5

  285. #endif

  286. 	maxps	%xmm5, %xmm1

  287. 	addq	$8 * SIZE, X

  288. 	ALIGN_3

  289. 

  290. .L17:

  291. 	testq	$4, M

  292. 	je	.L18

  293. 

  294. 	movaps	-32 * SIZE(X), %xmm4

  295. #ifdef USE_ABS

  296. 	andps	%xmm15, %xmm4

  297. #endif

  298. 	maxps	%xmm4, %xmm2

  299. 	addq	$4 * SIZE, X

  300. 	ALIGN_3

  301. 

  302. .L18:

  303. 	testq	$2, M

  304. 	je	.L19

  305. 

  306. 	movsd	-32 * SIZE(X), %xmm4

  307. 	unpcklps %xmm4, %xmm4

  308. #ifdef USE_ABS

  309. 	andps	%xmm15, %xmm4

  310. #endif

  311. 	maxps	%xmm4, %xmm3

  312. 	addq	$2 * SIZE, X

  313. 	ALIGN_3

  314. 

  315. .L19:

  316. 	testq	$1, M

  317. 	je	.L998

  318. 

  319. 	movss	-32 * SIZE(X), %xmm4

  320. #ifdef USE_ABS

  321. 	andps	%xmm15, %xmm4

  322. #endif

  323. 	maxss	%xmm4, %xmm0

  324. 	jmp	.L998

  325. 	ALIGN_3

  326. 

  327. .L40:

  328. 	movq	M,  I

  329. 	sarq	$3, I

  330. 	jle	.L45

  331. 	ALIGN_4

  332. 

  333. .L41:

  334. 	movss	(X), %xmm4

  335. 	addq	INCX, X

  336. #ifdef USE_ABS

  337. 	andps	%xmm15, %xmm4

  338. #endif

  339. 	maxss	%xmm4, %xmm0

  340. 

  341. 	movss	(X), %xmm5

  342. 	addq	INCX, X

  343. #ifdef USE_ABS

  344. 	andps	%xmm15, %xmm5

  345. #endif

  346. 	maxss	%xmm5, %xmm1

  347. 

  348. 	movss	(X), %xmm6

  349. 	addq	INCX, X

  350. #ifdef USE_ABS

  351. 	andps	%xmm15, %xmm6

  352. #endif

  353. 	maxss	%xmm6, %xmm2

  354. 

  355. 	movss	(X), %xmm7

  356. 	addq	INCX, X

  357. #ifdef USE_ABS

  358. 	andps	%xmm15, %xmm7

  359. #endif

  360. 	maxss	%xmm7, %xmm3

  361. 

  362. 	movss	(X), %xmm4

  363. 	addq	INCX, X

  364. #ifdef USE_ABS

  365. 	andps	%xmm15, %xmm4

  366. #endif

  367. 	maxss	%xmm4, %xmm0

  368. 

  369. 	movss	(X), %xmm5

  370. 	addq	INCX, X

  371. #ifdef USE_ABS

  372. 	andps	%xmm15, %xmm5

  373. #endif

  374. 	maxss	%xmm5, %xmm1

  375. 

  376. 	movss	(X), %xmm6

  377. 	addq	INCX, X

  378. #ifdef USE_ABS

  379. 	andps	%xmm15, %xmm6

  380. #endif

  381. 	maxss	%xmm6, %xmm2

  382. 

  383. 	movss	(X), %xmm7

  384. 	addq	INCX, X

  385. #ifdef USE_ABS

  386. 	andps	%xmm15, %xmm7

  387. #endif

  388. 	maxss	%xmm7, %xmm3

  389. 

  390. 	decq	I

  391. 	jg	.L41

  392. 	ALIGN_4

  393. 

  394. .L45:

  395. 	testq	$4, M

  396. 	je	.L46

  397. 

  398. 	movss	(X), %xmm4

  399. 	addq	INCX, X

  400. #ifdef USE_ABS

  401. 	andps	%xmm15, %xmm4

  402. #endif

  403. 	maxss	%xmm4, %xmm0

  404. 

  405. 	movss	(X), %xmm5

  406. 	addq	INCX, X

  407. #ifdef USE_ABS

  408. 	andps	%xmm15, %xmm5

  409. #endif

  410. 	maxss	%xmm5, %xmm1

  411. 

  412. 	movss	(X), %xmm6

  413. 	addq	INCX, X

  414. #ifdef USE_ABS

  415. 	andps	%xmm15, %xmm6

  416. #endif

  417. 	maxss	%xmm6, %xmm2

  418. 

  419. 	movss	(X), %xmm7

  420. 	addq	INCX, X

  421. #ifdef USE_ABS

  422. 	andps	%xmm15, %xmm7

  423. #endif

  424. 	maxss	%xmm7, %xmm3

  425. 	ALIGN_3

  426. 

  427. .L46:

  428. 	testq	$2, M

  429. 	je	.L47

  430. 

  431. 	movss	(X), %xmm4

  432. 	addq	INCX, X

  433. #ifdef USE_ABS

  434. 	andps	%xmm15, %xmm4

  435. #endif

  436. 	maxss	%xmm4, %xmm0

  437. 

  438. 	movss	(X), %xmm5

  439. 	addq	INCX, X

  440. #ifdef USE_ABS

  441. 	andps	%xmm15, %xmm5

  442. #endif

  443. 	maxss	%xmm5, %xmm1

  444. 	ALIGN_3

  445. 

  446. .L47:

  447. 	testq	$1, M

  448. 	je	.L998

  449. 

  450. 	movss	(X), %xmm4

  451. 	addq	INCX, X

  452. #ifdef USE_ABS

  453. 	andps	%xmm15, %xmm4

  454. #endif

  455. 	maxss	%xmm4, %xmm2

  456. 	ALIGN_4

  457. 

  458. .L998:

  459. 	maxps	%xmm1, %xmm0

  460. 	maxps	%xmm3, %xmm2

  461. 	maxps	%xmm2, %xmm0

  462. 	movaps	%xmm0, %xmm1

  463. 	movhlps %xmm0, %xmm0

  464. 	maxps	%xmm1, %xmm0

  465. 	movaps	%xmm0, %xmm1

  466. 	shufps	$1, %xmm0, %xmm0

  467. 	maxss	%xmm1, %xmm0

  468. 	ALIGN_4

  469. 

  470. .L999:

  471. 	RESTOREREGISTERS

  472. 

  473. 	ret

  474. 

  475. 	EPILOGUE