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: 5e1a429eab
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '5e1a429eab'
${ noResults }
OpenBLAS/kernel/x86_64/zrot.S

368 lines
6.4 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 N	ARG1

  43. #define X	ARG2

  44. #define INCX	ARG3

  45. #define Y	ARG4

  46. #ifndef WINDOWS_ABI

  47. #define INCY	ARG5	/* r8  */

  48. #else

  49. #define INCY	%r10

  50. #endif

  51. 

  52. #define	I	%rax

  53. 

  54. #include "l1param.h"

  55. 

  56. 	PROLOGUE

  57. 	PROFCODE

  58. 

  59. #ifdef WINDOWS_ABI

  60. 	movq	48(%rsp), INCY

  61. 	FLD	72(%rsp)

  62. 	FLD	56(%rsp)

  63. #else

  64. 	FLD	24(%rsp)

  65. 	FLD	 8(%rsp)

  66. #endif

  67. 

  68. 	salq	$ZBASE_SHIFT, INCX

  69. 	salq	$ZBASE_SHIFT, INCY

  70. 

  71. 	testq	N, N

  72. 	jle	.L999

  73. 

  74. 	cmpq	$2 * SIZE, INCX

  75. 	jne	.L50

  76. 	cmpq	$2 * SIZE, INCY

  77. 	jne	.L50

  78. 

  79. 	movq	N,  I

  80. 	sarq	$1, I

  81. 	jle	.L15

  82. 	ALIGN_4

  83. 

  84. .L10:

  85. #if defined(PREFETCHW)

  86. 	PREFETCHW (PREFETCHSIZE +  0) - PREOFFSET(Y)

  87. #endif

  88. 

  89. 	FLD	0 * SIZE(X)

  90. 	FLD	0 * SIZE(Y)

  91. 

  92. 	fld	%st(1)

  93. 	fmul	%st(3), %st

  94. 

  95. 	fld	%st(1)

  96. 	fmul	%st(5), %st

  97. 

  98. 	faddp	%st, %st(1)

  99. 	FST	0 * SIZE(X)

  100. 

  101. 	fmul	%st(2), %st

  102. 	fxch	%st(1)

  103. 	fmul	%st(3), %st

  104. 

  105. 	fsubrp	%st, %st(1)

  106. 	FST	0 * SIZE(Y)

  107. 

  108. 	FLD	1 * SIZE(X)

  109. 	FLD	1 * SIZE(Y)

  110. 

  111. 	fld	%st(1)

  112. 	fmul	%st(3), %st

  113. 

  114. 	fld	%st(1)

  115. 	fmul	%st(5), %st

  116. 

  117. 	faddp	%st, %st(1)

  118. 	FST	1 * SIZE(X)

  119. 

  120. 	fmul	%st(2), %st

  121. 	fxch	%st(1)

  122. 	fmul	%st(3), %st

  123. 

  124. 	fsubrp	%st, %st(1)

  125. 	FST	1 * SIZE(Y)

  126. 

  127. #if defined(PREFETCHW)

  128. 	PREFETCHW (PREFETCHSIZE +  0) - PREOFFSET(Y)

  129. #endif

  130. 

  131. 	FLD	2 * SIZE(X)

  132. 	FLD	2 * SIZE(Y)

  133. 

  134. 	fld	%st(1)

  135. 	fmul	%st(3), %st

  136. 

  137. 	fld	%st(1)

  138. 	fmul	%st(5), %st

  139. 

  140. 	faddp	%st, %st(1)

  141. 	FST	2 * SIZE(X)

  142. 

  143. 	fmul	%st(2), %st

  144. 	fxch	%st(1)

  145. 	fmul	%st(3), %st

  146. 

  147. 	fsubrp	%st, %st(1)

  148. 	FST	2 * SIZE(Y)

  149. 

  150. 	FLD	3 * SIZE(X)

  151. 	FLD	3 * SIZE(Y)

  152. 

  153. 	fld	%st(1)

  154. 	fmul	%st(3), %st

  155. 

  156. 	fld	%st(1)

  157. 	fmul	%st(5), %st

  158. 

  159. 	faddp	%st, %st(1)

  160. 	FST	3 * SIZE(X)

  161. 

  162. 	fmul	%st(2), %st

  163. 	fxch	%st(1)

  164. 	fmul	%st(3), %st

  165. 

  166. 	fsubrp	%st, %st(1)

  167. 	FST	3 * SIZE(Y)

  168. 

  169. 	addq	$4 * SIZE, X

  170. 	addq	$4 * SIZE, Y

  171. 

  172. 	decq	I

  173. 	jg	.L10

  174. 	ALIGN_4

  175. 

  176. .L15:

  177. 	movq	N,  I

  178. 	andq	$1, I

  179. 	jle	.L999

  180. 	ALIGN_4

  181. 

  182. .L16:

  183. 	FLD	0 * SIZE(X)

  184. 	FLD	0 * SIZE(Y)

  185. 

  186. 	fld	%st(1)

  187. 	fmul	%st(3), %st

  188. 

  189. 	fld	%st(1)

  190. 	fmul	%st(5), %st

  191. 

  192. 	faddp	%st, %st(1)

  193. 	FST	0 * SIZE(X)

  194. 

  195. 	fmul	%st(2), %st

  196. 	fxch	%st(1)

  197. 	fmul	%st(3), %st

  198. 

  199. 	fsubrp	%st, %st(1)

  200. 	FST	0 * SIZE(Y)

  201. 

  202. 	FLD	1 * SIZE(X)

  203. 	FLD	1 * SIZE(Y)

  204. 

  205. 	fld	%st(1)

  206. 	fmul	%st(3), %st

  207. 

  208. 	fld	%st(1)

  209. 	fmul	%st(5), %st

  210. 

  211. 	faddp	%st, %st(1)

  212. 	FST	1 * SIZE(X)

  213. 

  214. 	fmul	%st(2), %st

  215. 	fxch	%st(1)

  216. 	fmul	%st(3), %st

  217. 

  218. 	fsubrp	%st, %st(1)

  219. 	FST	1 * SIZE(Y)

  220. 	jmp	.L999

  221. 	ALIGN_4

  222. 

  223. .L50:

  224. 	movq	N,  I

  225. 	sarq	$1, I

  226. 	jle	.L55

  227. 	ALIGN_4

  228. 

  229. .L51:

  230. 	FLD	0 * SIZE(X)

  231. 	FLD	0 * SIZE(Y)

  232. 

  233. 	fld	%st(1)

  234. 	fmul	%st(3), %st

  235. 

  236. 	fld	%st(1)

  237. 	fmul	%st(5), %st

  238. 

  239. 	faddp	%st, %st(1)

  240. 	FST	0 * SIZE(X)

  241. 

  242. 	fmul	%st(2), %st

  243. 	fxch	%st(1)

  244. 	fmul	%st(3), %st

  245. 

  246. 	fsubrp	%st, %st(1)

  247. 	FST	0 * SIZE(Y)

  248. 

  249. 	FLD	1 * SIZE(X)

  250. 	FLD	1 * SIZE(Y)

  251. 

  252. 	fld	%st(1)

  253. 	fmul	%st(3), %st

  254. 

  255. 	fld	%st(1)

  256. 	fmul	%st(5), %st

  257. 

  258. 	faddp	%st, %st(1)

  259. 	FST	1 * SIZE(X)

  260. 

  261. 	fmul	%st(2), %st

  262. 	fxch	%st(1)

  263. 	fmul	%st(3), %st

  264. 

  265. 	fsubrp	%st, %st(1)

  266. 	FST	1 * SIZE(Y)

  267. 

  268. 	addq	INCX, X

  269. 	addq	INCY, Y

  270. 

  271. 	FLD	0 * SIZE(X)

  272. 	FLD	0 * SIZE(Y)

  273. 

  274. 	fld	%st(1)

  275. 	fmul	%st(3), %st

  276. 

  277. 	fld	%st(1)

  278. 	fmul	%st(5), %st

  279. 

  280. 	faddp	%st, %st(1)

  281. 	FST	0 * SIZE(X)

  282. 

  283. 	fmul	%st(2), %st

  284. 	fxch	%st(1)

  285. 	fmul	%st(3), %st

  286. 

  287. 	fsubrp	%st, %st(1)

  288. 	FST	0 * SIZE(Y)

  289. 

  290. 	FLD	1 * SIZE(X)

  291. 	FLD	1 * SIZE(Y)

  292. 

  293. 	fld	%st(1)

  294. 	fmul	%st(3), %st

  295. 

  296. 	fld	%st(1)

  297. 	fmul	%st(5), %st

  298. 

  299. 	faddp	%st, %st(1)

  300. 	FST	1 * SIZE(X)

  301. 

  302. 	fmul	%st(2), %st

  303. 	fxch	%st(1)

  304. 	fmul	%st(3), %st

  305. 

  306. 	fsubrp	%st, %st(1)

  307. 	FST	1 * SIZE(Y)

  308. 

  309. 	addq	INCX, X

  310. 	addq	INCY, Y

  311. 

  312. 	decq	I

  313. 	jg	.L51

  314. 	ALIGN_4

  315. 

  316. .L55:

  317. 	movq	N,  I

  318. 	andq	$1, I

  319. 	jle	.L999

  320. 	ALIGN_4

  321. 

  322. .L56:

  323. 	FLD	0 * SIZE(X)

  324. 	FLD	0 * SIZE(Y)

  325. 

  326. 	fld	%st(1)

  327. 	fmul	%st(3), %st

  328. 

  329. 	fld	%st(1)

  330. 	fmul	%st(5), %st

  331. 

  332. 	faddp	%st, %st(1)

  333. 	FST	0 * SIZE(X)

  334. 

  335. 	fmul	%st(2), %st

  336. 	fxch	%st(1)

  337. 	fmul	%st(3), %st

  338. 

  339. 	fsubrp	%st, %st(1)

  340. 	FST	0 * SIZE(Y)

  341. 

  342. 	FLD	1 * SIZE(X)

  343. 	FLD	1 * SIZE(Y)

  344. 

  345. 	fld	%st(1)

  346. 	fmul	%st(3), %st

  347. 

  348. 	fld	%st(1)

  349. 	fmul	%st(5), %st

  350. 

  351. 	faddp	%st, %st(1)

  352. 	FST	1 * SIZE(X)

  353. 

  354. 	fmul	%st(2), %st

  355. 	fxch	%st(1)

  356. 	fmul	%st(3), %st

  357. 

  358. 	fsubrp	%st, %st(1)

  359. 	FST	1 * SIZE(Y)

  360. 	ALIGN_4

  361. 

  362. .L999:

  363. 	ffreep	%st

  364. 	ffreep	%st

  365. 	ret

  366. 

  367. 	EPILOGUE