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: 1a7b8c650d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '1a7b8c650d'
${ noResults }
OpenBLAS/kernel/x86/scal.S

359 lines
7.0 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. 	PROLOGUE

  43. 

  44. 	pushl	%edi

  45. 	pushl	%esi

  46. 	pushl	%ebx

  47. 

  48. 	PROFCODE

  49. 

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

  51. 	EMMS

  52. #endif

  53. 

  54. 	movl	16(%esp),%edx

  55. 	FLD	28(%esp)

  56. 

  57. #ifdef XDOUBLE

  58. 	movl	44(%esp),%edi

  59. 	movl	48(%esp),%esi

  60. 	movl    64(%esp),%ecx

  61. #elif defined(DOUBLE)

  62. 	movl	36(%esp),%edi

  63. 	movl	40(%esp),%esi

  64. 	movl	56(%esp),%ecx

  65. #else

  66. 	movl	32(%esp),%edi

  67. 	movl	36(%esp),%esi

  68. 	movl	52(%esp),%ecx

  69. #endif

  70. 

  71. 	ftst

  72. 	fnstsw	%ax

  73. 	andb	$68, %ah

  74. 	je	.L300		# Alpha != ZERO

  75. 

  76. 	cmpl	$1,%ecx		# dummy2 flag

  77. 	je	.L300

  78. 

  79. /* Alpha == ZERO */

  80. 	cmpl	$1,%esi

  81. 	jne	.L104

  82. 

  83. 	movl	%edx, %ecx		# ecx = n

  84. 	sarl	$3,   %ecx		# (n >> 3)

  85. 	jle	.L102

  86. 	ALIGN_4

  87. 

  88. .L101:

  89. #ifndef XDOUBLE

  90. 	FSTU	0 * SIZE(%edi)

  91. 	FSTU	1 * SIZE(%edi)

  92. 	FSTU	2 * SIZE(%edi)

  93. 	FSTU	3 * SIZE(%edi)

  94. 	FSTU	4 * SIZE(%edi)

  95. 	FSTU	5 * SIZE(%edi)

  96. 	FSTU	6 * SIZE(%edi)

  97. 	FSTU	7 * SIZE(%edi)

  98. #else

  99. 	fld	%st

  100. 	FST	0 * SIZE(%edi)

  101. 	fld	%st

  102. 	FST	1 * SIZE(%edi)

  103. 	fld	%st

  104. 	FST	2 * SIZE(%edi)

  105. 	fld	%st

  106. 	FST	3 * SIZE(%edi)

  107. 	fld	%st

  108. 	FST	4 * SIZE(%edi)

  109. 	fld	%st

  110. 	FST	5 * SIZE(%edi)

  111. 	fld	%st

  112. 	FST	6 * SIZE(%edi)

  113. 	fld	%st

  114. 	FST	7 * SIZE(%edi)

  115. #endif

  116. 

  117. 	addl	$8 * SIZE, %edi

  118. 	decl	%ecx

  119. 	jg	.L101

  120. 	ALIGN_4

  121. 

  122. .L102:

  123. 	movl	%edx, %ecx

  124. 	andl	$7,   %ecx

  125. 	jle	.L999

  126. 	ALIGN_4

  127. 

  128. .L103:

  129. #ifndef XDOUBLE

  130. 	FSTU	0 * SIZE(%edi)

  131. #else

  132. 	fld	%st

  133. 	FST	0 * SIZE(%edi)

  134. #endif

  135. 	addl	$SIZE, %edi

  136. 	decl	%ecx

  137. 	jg	.L103

  138. 	jmp	.L999

  139. 	ALIGN_4

  140. 

  141. .L104:

  142. 	sall	$BASE_SHIFT, %esi

  143. 

  144. 	movl	%edx, %ecx		# ecx = n

  145. 	sarl	$3,   %ecx		# (n >> 3)

  146. 	jle	.L106

  147. 	ALIGN_4

  148. 

  149. .L105:

  150. #ifndef XDOUBLE

  151. 	FSTU	0 * SIZE(%edi)

  152. 	addl	%esi, %edi

  153. 	FSTU	0 * SIZE(%edi)

  154. 	addl	%esi, %edi

  155. 	FSTU	0 * SIZE(%edi)

  156. 	addl	%esi, %edi

  157. 	FSTU	0 * SIZE(%edi)

  158. 	addl	%esi, %edi

  159. 	FSTU	0 * SIZE(%edi)

  160. 	addl	%esi, %edi

  161. 	FSTU	0 * SIZE(%edi)

  162. 	addl	%esi, %edi

  163. 	FSTU	0 * SIZE(%edi)

  164. 	addl	%esi, %edi

  165. 	FSTU	0 * SIZE(%edi)

  166. 	addl	%esi, %edi

  167. #else

  168. 	fld	%st

  169. 	FST	0 * SIZE(%edi)

  170. 	addl	%esi, %edi

  171. 	fld	%st

  172. 	FST	0 * SIZE(%edi)

  173. 	addl	%esi, %edi

  174. 	fld	%st

  175. 	FST	0 * SIZE(%edi)

  176. 	addl	%esi, %edi

  177. 	fld	%st

  178. 	FST	0 * SIZE(%edi)

  179. 	addl	%esi, %edi

  180. 	fld	%st

  181. 	FST	0 * SIZE(%edi)

  182. 	addl	%esi, %edi

  183. 	fld	%st

  184. 	FST	0 * SIZE(%edi)

  185. 	addl	%esi, %edi

  186. 	fld	%st

  187. 	FST	0 * SIZE(%edi)

  188. 	addl	%esi, %edi

  189. 	fld	%st

  190. 	FST	0 * SIZE(%edi)

  191. 	addl	%esi, %edi

  192. #endif

  193. 

  194. 	decl	%ecx

  195. 	jg	.L105

  196. 	ALIGN_4

  197. 

  198. .L106:

  199. 	movl	%edx, %ecx

  200. 	andl	$7,   %ecx

  201. 	jle	.L999

  202. 	ALIGN_4

  203. 

  204. .L107:

  205. #ifndef XDOUBLE

  206. 	FSTU	0 * SIZE(%edi)

  207. #else

  208. 	fld	%st

  209. 	FST	0 * SIZE(%edi)

  210. #endif

  211. 	addl	%esi, %edi

  212. 	decl	%ecx

  213. 	jg	.L107

  214. 	jmp	.L999

  215. 	ALIGN_4

  216. 

  217. /* Alpha != ZERO */

  218. 

  219. .L300:

  220. 	cmpl	$1,%esi

  221. 	jne	.L304

  222. 

  223. 	movl	%edx, %ecx		# ecx = n

  224. 	sarl	$3,   %ecx		# (n >> 3)

  225. 	jle	.L302

  226. 	ALIGN_4

  227. 

  228. .L301:

  229. 	FLD	0 * SIZE(%edi)

  230. 	fmul	%st(1), %st

  231. 	FST	0 * SIZE(%edi)

  232. 

  233. 	FLD	1 * SIZE(%edi)

  234. 	fmul	%st(1), %st

  235. 	FST	1 * SIZE(%edi)

  236. 

  237. 	FLD	2 * SIZE(%edi)

  238. 	fmul	%st(1), %st

  239. 	FST	2 * SIZE(%edi)

  240. 

  241. 	FLD	3 * SIZE(%edi)

  242. 	fmul	%st(1), %st

  243. 	FST	3 * SIZE(%edi)

  244. 

  245. 	FLD	4 * SIZE(%edi)

  246. 	fmul	%st(1), %st

  247. 	FST	4 * SIZE(%edi)

  248. 

  249. 	FLD	5 * SIZE(%edi)

  250. 	fmul	%st(1), %st

  251. 	FST	5 * SIZE(%edi)

  252. 

  253. 	FLD	6 * SIZE(%edi)

  254. 	fmul	%st(1), %st

  255. 	FST	6 * SIZE(%edi)

  256. 

  257. 	FLD	7 * SIZE(%edi)

  258. 	fmul	%st(1), %st

  259. 	FST	7 * SIZE(%edi)

  260. 

  261. 	addl	$8 * SIZE, %edi

  262. 	decl	%ecx

  263. 	jg	.L301

  264. 	ALIGN_4

  265. 

  266. .L302:

  267. 	movl	%edx, %ecx

  268. 	andl	$7,   %ecx

  269. 	jle	.L999

  270. 	ALIGN_4

  271. 

  272. .L303:

  273. 	FLD	0 * SIZE(%edi)

  274. 	fmul	%st(1), %st

  275. 	FST	0 * SIZE(%edi)

  276. 	addl	$SIZE, %edi

  277. 	decl	%ecx

  278. 	jg	.L303

  279. 	jmp	.L999

  280. 	ALIGN_4

  281. 

  282. .L304:

  283. 	sall	$BASE_SHIFT, %esi

  284. 

  285. 	movl	%edx, %ecx		# ecx = n

  286. 	sarl	$3,   %ecx		# (n >> 3)

  287. 	jle	.L306

  288. 	ALIGN_4

  289. 

  290. .L305:

  291. 	FLD	0 * SIZE(%edi)

  292. 	fmul	%st(1), %st

  293. 	FST	0 * SIZE(%edi)

  294. 	addl	%esi, %edi

  295. 

  296. 	FLD	0 * SIZE(%edi)

  297. 	fmul	%st(1), %st

  298. 	FST	0 * SIZE(%edi)

  299. 	addl	%esi, %edi

  300. 

  301. 	FLD	0 * SIZE(%edi)

  302. 	fmul	%st(1), %st

  303. 	FST	0 * SIZE(%edi)

  304. 	addl	%esi, %edi

  305. 

  306. 	FLD	0 * SIZE(%edi)

  307. 	fmul	%st(1), %st

  308. 	FST	0 * SIZE(%edi)

  309. 	addl	%esi, %edi

  310. 

  311. 	FLD	0 * SIZE(%edi)

  312. 	fmul	%st(1), %st

  313. 	FST	0 * SIZE(%edi)

  314. 	addl	%esi, %edi

  315. 

  316. 	FLD	0 * SIZE(%edi)

  317. 	fmul	%st(1), %st

  318. 	FST	0 * SIZE(%edi)

  319. 	addl	%esi, %edi

  320. 

  321. 	FLD	0 * SIZE(%edi)

  322. 	fmul	%st(1), %st

  323. 	FST	0 * SIZE(%edi)

  324. 	addl	%esi, %edi

  325. 

  326. 	FLD	0 * SIZE(%edi)

  327. 	fmul	%st(1), %st

  328. 	FST	0 * SIZE(%edi)

  329. 	addl	%esi, %edi

  330. 

  331. 	decl	%ecx

  332. 	jg	.L305

  333. 	ALIGN_4

  334. 

  335. .L306:

  336. 	movl	%edx, %ecx

  337. 	andl	$7,   %ecx

  338. 	jle	.L999

  339. 	ALIGN_4

  340. 

  341. .L307:

  342. 	FLD	0 * SIZE(%edi)

  343. 	fmul	%st(1), %st

  344. 	FST	0 * SIZE(%edi)

  345. 	addl	%esi, %edi

  346. 	decl	%ecx

  347. 	jg	.L307

  348. 	ALIGN_4

  349. 

  350. .L999:

  351. 	ffreep	%st(0)

  352. 	xorl	%eax,%eax

  353. 	popl	%ebx

  354. 	popl	%esi

  355. 	popl	%edi

  356. 	ret

  357. 

  358. 	EPILOGUE