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: a1ec94c258
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a1ec94c258'
${ noResults }
OpenBLAS/kernel/x86/xaxpy.S

357 lines
7.3 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_M		 4 + STACK + ARGS(%esp)

  46. #define STACK_ALPHA_R	16 + STACK + ARGS(%esp)

  47. #define STACK_ALPHA_I	32 + STACK + ARGS(%esp)

  48. #define STACK_X		48 + STACK + ARGS(%esp)

  49. #define STACK_INCX	52 + STACK + ARGS(%esp)

  50. #define STACK_Y		56 + STACK + ARGS(%esp)

  51. #define STACK_INCY	60 + STACK + ARGS(%esp)

  52. 

  53. #define M	%ebx

  54. #define X	%esi

  55. #define INCX	%ecx

  56. #define Y	%edi

  57. #define INCY	%edx

  58. 

  59. #ifndef CONJ

  60. #define ADD1	fsubrp

  61. #define ADD2	faddp

  62. #else

  63. #define ADD1	faddp

  64. #define ADD2	fsubrp

  65. #endif

  66. 

  67. 	PROLOGUE

  68. 

  69. 	pushl	%edi

  70. 	pushl	%esi

  71. 	pushl	%ebx

  72. 

  73. 	PROFCODE

  74. 

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

  76. 	EMMS

  77. #endif

  78. 

  79. 	FLD	STACK_ALPHA_I

  80. 	FLD	STACK_ALPHA_R

  81. 

  82. 	movl	STACK_M,    M

  83. 	movl	STACK_X,    X

  84. 	movl	STACK_INCX, INCX

  85. 	movl	STACK_Y,    Y

  86. 	movl	STACK_INCY, INCY

  87. 

  88. 	sall	$ZBASE_SHIFT, INCX

  89. 	sall	$ZBASE_SHIFT, INCY

  90. 

  91. 	testl	M, M

  92. 	jle	.L40

  93. 

  94. 	cmpl	$2 * SIZE, INCX

  95. 	jne	.L14

  96. 	cmpl	$2 * SIZE, INCY

  97. 	jne	.L14

  98. 

  99. 	movl	M,  %eax

  100. 	sarl	$2, %eax

  101. 	jle	.L15

  102. 	ALIGN_3

  103. 

  104. .L16:

  105. 	FLD	0 * SIZE(X)

  106. 	fmul	%st(1), %st

  107. 	FLD	1 * SIZE(X)

  108. 	fmul	%st(3), %st

  109. 	ADD1	%st, %st(1)

  110. 	FLD	0 * SIZE(Y)

  111. 	faddp	%st, %st(1)

  112. 	FST	0 * SIZE(Y)

  113. 

  114. 	FLD	0 * SIZE(X)

  115. 	fmul	%st(2), %st

  116. 	FLD	1 * SIZE(X)

  117. 	fmul	%st(2), %st

  118. 	ADD2	%st, %st(1)

  119. 	FLD	1 * SIZE(Y)

  120. 	faddp	%st, %st(1)

  121. 	FST	1 * SIZE(Y)

  122. 

  123. 	FLD	2 * SIZE(X)

  124. 	fmul	%st(1), %st

  125. 	FLD	3 * SIZE(X)

  126. 	fmul	%st(3), %st

  127. 	ADD1	%st, %st(1)

  128. 	FLD	2 * SIZE(Y)

  129. 	faddp	%st, %st(1)

  130. 	FST	2 * SIZE(Y)

  131. 

  132. 	FLD	2 * SIZE(X)

  133. 	fmul	%st(2), %st

  134. 	FLD	3 * SIZE(X)

  135. 	fmul	%st(2), %st

  136. 	ADD2	%st, %st(1)

  137. 	FLD	3 * SIZE(Y)

  138. 	faddp	%st, %st(1)

  139. 	FST	3 * SIZE(Y)

  140. 

  141. 	FLD	4 * SIZE(X)

  142. 	fmul	%st(1), %st

  143. 	FLD	5 * SIZE(X)

  144. 	fmul	%st(3), %st

  145. 	ADD1	%st, %st(1)

  146. 	FLD	4 * SIZE(Y)

  147. 	faddp	%st, %st(1)

  148. 	FST	4 * SIZE(Y)

  149. 

  150. 	FLD	4 * SIZE(X)

  151. 	fmul	%st(2), %st

  152. 	FLD	5 * SIZE(X)

  153. 	fmul	%st(2), %st

  154. 	ADD2	%st, %st(1)

  155. 	FLD	5 * SIZE(Y)

  156. 	faddp	%st, %st(1)

  157. 	FST	5 * SIZE(Y)

  158. 

  159. 	FLD	6 * SIZE(X)

  160. 	fmul	%st(1), %st

  161. 	FLD	7 * SIZE(X)

  162. 	fmul	%st(3), %st

  163. 	ADD1	%st, %st(1)

  164. 	FLD	6 * SIZE(Y)

  165. 	faddp	%st, %st(1)

  166. 	FST	6 * SIZE(Y)

  167. 

  168. 	FLD	6 * SIZE(X)

  169. 	fmul	%st(2), %st

  170. 	FLD	7 * SIZE(X)

  171. 	fmul	%st(2), %st

  172. 	ADD2	%st, %st(1)

  173. 	FLD	7 * SIZE(Y)

  174. 	faddp	%st, %st(1)

  175. 	FST	7 * SIZE(Y)

  176. 

  177. #ifdef HAVE_3DNOW

  178. 	prefetch	20 * SIZE(X)

  179. 	prefetchw	20 * SIZE(Y)

  180. #endif

  181. 

  182. 	addl	$8 * SIZE, X

  183. 	addl	$8 * SIZE, Y

  184. 	decl	%eax

  185. 	jg	.L16

  186. 	ALIGN_3

  187. 

  188. .L15:

  189. 	movl	M,  %eax

  190. 	andl	$3, %eax

  191. 	jle	.L40

  192. 	ALIGN_3

  193. 

  194. .L22:

  195. 	FLD	0 * SIZE(X)

  196. 	fmul	%st(1), %st

  197. 	FLD	1 * SIZE(X)

  198. 	fmul	%st(3), %st

  199. 	ADD1	%st, %st(1)

  200. 	FLD	0 * SIZE(Y)

  201. 	faddp	%st, %st(1)

  202. 	FST	0 * SIZE(Y)

  203. 

  204. 	FLD	0 * SIZE(X)

  205. 	fmul	%st(2), %st

  206. 	FLD	1 * SIZE(X)

  207. 	fmul	%st(2), %st

  208. 	ADD2	%st, %st(1)

  209. 	FLD	1 * SIZE(Y)

  210. 	faddp	%st, %st(1)

  211. 	FST	1 * SIZE(Y)

  212. 

  213. 	addl	$2 * SIZE, X

  214. 	addl	$2 * SIZE, Y

  215. 	decl	%eax

  216. 	jg	.L22

  217. 	jmp	.L40

  218. 	ALIGN_3

  219. 

  220. .L14:

  221. 	movl	M, %eax

  222. 	sarl	$2,   %eax

  223. 	jle	.L28

  224. 	ALIGN_3

  225. 

  226. .L29:

  227. 	FLD	0 * SIZE(X)

  228. 	fmul	%st(1), %st

  229. 	FLD	1 * SIZE(X)

  230. 	fmul	%st(3), %st

  231. 	ADD1	%st, %st(1)

  232. 	FLD	0 * SIZE(Y)

  233. 	faddp	%st, %st(1)

  234. 	FST	0 * SIZE(Y)

  235. 

  236. 	FLD	0 * SIZE(X)

  237. 	fmul	%st(2), %st

  238. 	FLD	1 * SIZE(X)

  239. 	fmul	%st(2), %st

  240. 	ADD2	%st, %st(1)

  241. 	FLD	1 * SIZE(Y)

  242. 	faddp	%st, %st(1)

  243. 	FST	1 * SIZE(Y)

  244. 

  245. 	addl	INCX, X

  246. 	addl	INCY, Y

  247. 

  248. 	FLD	0 * SIZE(X)

  249. 	fmul	%st(1), %st

  250. 	FLD	1 * SIZE(X)

  251. 	fmul	%st(3), %st

  252. 	ADD1	%st, %st(1)

  253. 	FLD	0 * SIZE(Y)

  254. 	faddp	%st, %st(1)

  255. 	FST	0 * SIZE(Y)

  256. 

  257. 	FLD	0 * SIZE(X)

  258. 	fmul	%st(2), %st

  259. 	FLD	1 * SIZE(X)

  260. 	fmul	%st(2), %st

  261. 	ADD2	%st, %st(1)

  262. 	FLD	1 * SIZE(Y)

  263. 	faddp	%st, %st(1)

  264. 	FST	1 * SIZE(Y)

  265. 

  266. 	addl	INCX, X

  267. 	addl	INCY, Y

  268. 

  269. 	FLD	0 * SIZE(X)

  270. 	fmul	%st(1), %st

  271. 	FLD	1 * SIZE(X)

  272. 	fmul	%st(3), %st

  273. 	ADD1	%st, %st(1)

  274. 	FLD	0 * SIZE(Y)

  275. 	faddp	%st, %st(1)

  276. 	FST	0 * SIZE(Y)

  277. 

  278. 	FLD	0 * SIZE(X)

  279. 	fmul	%st(2), %st

  280. 	FLD	1 * SIZE(X)

  281. 	fmul	%st(2), %st

  282. 	ADD2	%st, %st(1)

  283. 	FLD	1 * SIZE(Y)

  284. 	faddp	%st, %st(1)

  285. 	FST	1 * SIZE(Y)

  286. 

  287. 	addl	INCX, X

  288. 	addl	INCY, Y

  289. 

  290. 	FLD	0 * SIZE(X)

  291. 	fmul	%st(1), %st

  292. 	FLD	1 * SIZE(X)

  293. 	fmul	%st(3), %st

  294. 	ADD1	%st, %st(1)

  295. 	FLD	0 * SIZE(Y)

  296. 	faddp	%st, %st(1)

  297. 	FST	0 * SIZE(Y)

  298. 

  299. 	FLD	0 * SIZE(X)

  300. 	fmul	%st(2), %st

  301. 	FLD	1 * SIZE(X)

  302. 	fmul	%st(2), %st

  303. 	ADD2	%st, %st(1)

  304. 	FLD	1 * SIZE(Y)

  305. 	faddp	%st, %st(1)

  306. 	FST	1 * SIZE(Y)

  307. 

  308. 	addl	INCX, X

  309. 	addl	INCY, Y

  310. 

  311. 	decl	%eax

  312. 	jg	.L29

  313. 	ALIGN_3

  314. 

  315. .L28:

  316. 	movl	M,  %eax

  317. 	andl	$3, %eax

  318. 	jle	.L40

  319. 	ALIGN_3

  320. 

  321. .L35:

  322. 	FLD	0 * SIZE(X)

  323. 	fmul	%st(1), %st

  324. 	FLD	1 * SIZE(X)

  325. 	fmul	%st(3), %st

  326. 	ADD1	%st, %st(1)

  327. 	FLD	0 * SIZE(Y)

  328. 	faddp	%st, %st(1)

  329. 	FST	0 * SIZE(Y)

  330. 

  331. 	FLD	0 * SIZE(X)

  332. 	fmul	%st(2), %st

  333. 	FLD	1 * SIZE(X)

  334. 	fmul	%st(2), %st

  335. 	ADD2	%st, %st(1)

  336. 	FLD	1 * SIZE(Y)

  337. 	faddp	%st, %st(1)

  338. 	FST	1 * SIZE(Y)

  339. 

  340. 	addl	INCX, X

  341. 	addl	INCY, Y

  342. 

  343. 	decl	%eax

  344. 	jg	.L35

  345. 	ALIGN_3

  346. 

  347. .L40:

  348. 	ffreep	%st(0)

  349. 	ffreep	%st(0)

  350. 	xorl	%eax,%eax

  351. 	popl	%ebx

  352. 	popl	%esi

  353. 	popl	%edi

  354. 	ret

  355. 

  356. 	EPILOGUE