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: f66ca05b31
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'f66ca05b31'
${ noResults }
OpenBLAS/kernel/x86/zrot.S

408 lines
7.2 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 STACK	12

  43. #define ARGS     0

  44. 

  45. #define STACK_N		 4 + STACK + ARGS(%esp)

  46. #define STACK_X		 8 + STACK + ARGS(%esp)

  47. #define STACK_INCX	12 + STACK + ARGS(%esp)

  48. #define STACK_Y		16 + STACK + ARGS(%esp)

  49. #define STACK_INCY	20 + STACK + ARGS(%esp)

  50. #define STACK_C		24 + STACK + ARGS(%esp)

  51. #ifdef XDOUBLE

  52. #define STACK_S		40 + STACK + ARGS(%esp)

  53. #elif defined DOUBLE

  54. #define STACK_S		32 + STACK + ARGS(%esp)

  55. #else

  56. #define STACK_S		28 + STACK + ARGS(%esp)

  57. #endif

  58. 

  59. #define N	%ebx

  60. #define X	%esi

  61. #define INCX	%ecx

  62. #define Y	%edi

  63. #define INCY	%edx

  64. 

  65. #define	I	%eax

  66. 

  67. #ifdef PENTIUM4

  68. #define PREFETCH	prefetcht0

  69. #define PREFETCH_SIZE	144

  70. #endif

  71. 

  72. #ifdef OPTERON

  73. #define PREFETCH	prefetchw

  74. #define PREFETCH_SIZE	144

  75. #endif

  76. 

  77. 	PROLOGUE

  78. 

  79. 	pushl	%edi

  80. 	pushl	%esi

  81. 	pushl	%ebx

  82. 

  83. 	PROFCODE

  84. 

  85. #if defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)

  86. 	EMMS

  87. #endif

  88. 

  89. 	movl	STACK_N,     N

  90. 	movl	STACK_X,     X

  91. 	movl	STACK_INCX,  INCX

  92. 	movl	STACK_Y,     Y

  93. 	movl	STACK_INCY,  INCY

  94. 

  95. 	FLD	STACK_S

  96. 	FLD	STACK_C

  97. 

  98. 	sall	$ZBASE_SHIFT, INCX

  99. 	sall	$ZBASE_SHIFT, INCY

  100. 

  101. 	testl	N, N

  102. 	jle	.L999

  103. 

  104. 	cmpl	$2 * SIZE, INCX

  105. 	jne	.L50

  106. 	cmpl	$2 * SIZE, INCY

  107. 	jne	.L50

  108. 

  109. 	movl	N,  I

  110. 	sarl	$1, I

  111. 	jle	.L15

  112. 	ALIGN_4

  113. 

  114. .L10:

  115. #ifdef PENTIUM4

  116. 	PREFETCH	(PREFETCH_SIZE + 0) * SIZE(X)

  117. #endif

  118. #ifdef OPTERON

  119. 	PREFETCH	(PREFETCH_SIZE + 0) * SIZE(X)

  120. #endif

  121. 

  122. 	FLD	0 * SIZE(X)

  123. 	FLD	0 * SIZE(Y)

  124. 

  125. 	fld	%st(1)

  126. 	fmul	%st(3), %st

  127. 

  128. 	fld	%st(1)

  129. 	fmul	%st(5), %st

  130. 

  131. 	faddp	%st, %st(1)

  132. 	FST	0 * SIZE(X)

  133. 

  134. 	fmul	%st(2), %st

  135. 	fxch	%st(1)

  136. 	fmul	%st(3), %st

  137. 

  138. 	fsubrp	%st, %st(1)

  139. 	FST	0 * SIZE(Y)

  140. 

  141. 	FLD	1 * SIZE(X)

  142. 	FLD	1 * SIZE(Y)

  143. 

  144. 	fld	%st(1)

  145. 	fmul	%st(3), %st

  146. 

  147. 	fld	%st(1)

  148. 	fmul	%st(5), %st

  149. 

  150. 	faddp	%st, %st(1)

  151. 	FST	1 * SIZE(X)

  152. 

  153. 	fmul	%st(2), %st

  154. 	fxch	%st(1)

  155. 	fmul	%st(3), %st

  156. 

  157. 	fsubrp	%st, %st(1)

  158. 	FST	1 * SIZE(Y)

  159. 

  160. #ifdef PENTIUM4

  161. 	PREFETCH	(PREFETCH_SIZE + 0) * SIZE(Y)

  162. #endif

  163. #ifdef OPTERON

  164. 	PREFETCH	(PREFETCH_SIZE + 0) * SIZE(Y)

  165. #endif

  166. 

  167. 	FLD	2 * SIZE(X)

  168. 	FLD	2 * SIZE(Y)

  169. 

  170. 	fld	%st(1)

  171. 	fmul	%st(3), %st

  172. 

  173. 	fld	%st(1)

  174. 	fmul	%st(5), %st

  175. 

  176. 	faddp	%st, %st(1)

  177. 	FST	2 * SIZE(X)

  178. 

  179. 	fmul	%st(2), %st

  180. 	fxch	%st(1)

  181. 	fmul	%st(3), %st

  182. 

  183. 	fsubrp	%st, %st(1)

  184. 	FST	2 * SIZE(Y)

  185. 

  186. 	FLD	3 * SIZE(X)

  187. 	FLD	3 * SIZE(Y)

  188. 

  189. 	fld	%st(1)

  190. 	fmul	%st(3), %st

  191. 

  192. 	fld	%st(1)

  193. 	fmul	%st(5), %st

  194. 

  195. 	faddp	%st, %st(1)

  196. 	FST	3 * SIZE(X)

  197. 

  198. 	fmul	%st(2), %st

  199. 	fxch	%st(1)

  200. 	fmul	%st(3), %st

  201. 

  202. 	fsubrp	%st, %st(1)

  203. 	FST	3 * SIZE(Y)

  204. 

  205. 	addl	$4 * SIZE, X

  206. 	addl	$4 * SIZE, Y

  207. 

  208. 	decl	I

  209. 	jg	.L10

  210. 	ALIGN_4

  211. 

  212. .L15:

  213. 	movl	N,  I

  214. 	andl	$1, I

  215. 	jle	.L999

  216. 	ALIGN_4

  217. 

  218. .L16:

  219. 	FLD	0 * SIZE(X)

  220. 	FLD	0 * SIZE(Y)

  221. 

  222. 	fld	%st(1)

  223. 	fmul	%st(3), %st

  224. 

  225. 	fld	%st(1)

  226. 	fmul	%st(5), %st

  227. 

  228. 	faddp	%st, %st(1)

  229. 	FST	0 * SIZE(X)

  230. 

  231. 	fmul	%st(2), %st

  232. 	fxch	%st(1)

  233. 	fmul	%st(3), %st

  234. 

  235. 	fsubrp	%st, %st(1)

  236. 	FST	0 * SIZE(Y)

  237. 

  238. 	FLD	1 * SIZE(X)

  239. 	FLD	1 * SIZE(Y)

  240. 

  241. 	fld	%st(1)

  242. 	fmul	%st(3), %st

  243. 

  244. 	fld	%st(1)

  245. 	fmul	%st(5), %st

  246. 

  247. 	faddp	%st, %st(1)

  248. 	FST	1 * SIZE(X)

  249. 

  250. 	fmul	%st(2), %st

  251. 	fxch	%st(1)

  252. 	fmul	%st(3), %st

  253. 

  254. 	fsubrp	%st, %st(1)

  255. 	FST	1 * SIZE(Y)

  256. 	jmp	.L999

  257. 	ALIGN_4

  258. 

  259. .L50:

  260. 	movl	N,  I

  261. 	sarl	$1, I

  262. 	jle	.L55

  263. 	ALIGN_4

  264. 

  265. .L51:

  266. 	FLD	0 * SIZE(X)

  267. 	FLD	0 * SIZE(Y)

  268. 

  269. 	fld	%st(1)

  270. 	fmul	%st(3), %st

  271. 

  272. 	fld	%st(1)

  273. 	fmul	%st(5), %st

  274. 

  275. 	faddp	%st, %st(1)

  276. 	FST	0 * SIZE(X)

  277. 

  278. 	fmul	%st(2), %st

  279. 	fxch	%st(1)

  280. 	fmul	%st(3), %st

  281. 

  282. 	fsubrp	%st, %st(1)

  283. 	FST	0 * SIZE(Y)

  284. 

  285. 	FLD	1 * SIZE(X)

  286. 	FLD	1 * SIZE(Y)

  287. 

  288. 	fld	%st(1)

  289. 	fmul	%st(3), %st

  290. 

  291. 	fld	%st(1)

  292. 	fmul	%st(5), %st

  293. 

  294. 	faddp	%st, %st(1)

  295. 	FST	1 * SIZE(X)

  296. 

  297. 	fmul	%st(2), %st

  298. 	fxch	%st(1)

  299. 	fmul	%st(3), %st

  300. 

  301. 	fsubrp	%st, %st(1)

  302. 	FST	1 * SIZE(Y)

  303. 

  304. 	addl	INCX, X

  305. 	addl	INCY, Y

  306. 

  307. 	FLD	0 * SIZE(X)

  308. 	FLD	0 * SIZE(Y)

  309. 

  310. 	fld	%st(1)

  311. 	fmul	%st(3), %st

  312. 

  313. 	fld	%st(1)

  314. 	fmul	%st(5), %st

  315. 

  316. 	faddp	%st, %st(1)

  317. 	FST	0 * SIZE(X)

  318. 

  319. 	fmul	%st(2), %st

  320. 	fxch	%st(1)

  321. 	fmul	%st(3), %st

  322. 

  323. 	fsubrp	%st, %st(1)

  324. 	FST	0 * SIZE(Y)

  325. 

  326. 	FLD	1 * SIZE(X)

  327. 	FLD	1 * SIZE(Y)

  328. 

  329. 	fld	%st(1)

  330. 	fmul	%st(3), %st

  331. 

  332. 	fld	%st(1)

  333. 	fmul	%st(5), %st

  334. 

  335. 	faddp	%st, %st(1)

  336. 	FST	1 * SIZE(X)

  337. 

  338. 	fmul	%st(2), %st

  339. 	fxch	%st(1)

  340. 	fmul	%st(3), %st

  341. 

  342. 	fsubrp	%st, %st(1)

  343. 	FST	1 * SIZE(Y)

  344. 

  345. 	addl	INCX, X

  346. 	addl	INCY, Y

  347. 

  348. 	decl	I

  349. 	jg	.L51

  350. 	ALIGN_4

  351. 

  352. .L55:

  353. 	movl	N,  I

  354. 	andl	$1, I

  355. 	jle	.L999

  356. 	ALIGN_4

  357. 

  358. .L56:

  359. 	FLD	0 * SIZE(X)

  360. 	FLD	0 * SIZE(Y)

  361. 

  362. 	fld	%st(1)

  363. 	fmul	%st(3), %st

  364. 

  365. 	fld	%st(1)

  366. 	fmul	%st(5), %st

  367. 

  368. 	faddp	%st, %st(1)

  369. 	FST	0 * SIZE(X)

  370. 

  371. 	fmul	%st(2), %st

  372. 	fxch	%st(1)

  373. 	fmul	%st(3), %st

  374. 

  375. 	fsubrp	%st, %st(1)

  376. 	FST	0 * SIZE(Y)

  377. 

  378. 	FLD	1 * SIZE(X)

  379. 	FLD	1 * SIZE(Y)

  380. 

  381. 	fld	%st(1)

  382. 	fmul	%st(3), %st

  383. 

  384. 	fld	%st(1)

  385. 	fmul	%st(5), %st

  386. 

  387. 	faddp	%st, %st(1)

  388. 	FST	1 * SIZE(X)

  389. 

  390. 	fmul	%st(2), %st

  391. 	fxch	%st(1)

  392. 	fmul	%st(3), %st

  393. 

  394. 	fsubrp	%st, %st(1)

  395. 	FST	1 * SIZE(Y)

  396. 	ALIGN_4

  397. 

  398. .L999:

  399. 	ffreep	%st(0)

  400. 	ffreep	%st(0)

  401. 

  402. 	popl	%ebx

  403. 	popl	%esi

  404. 	popl	%edi

  405. 	ret

  406. 

  407. 	EPILOGUE