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

388 lines
7.9 kB
Raw Permalink 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	16

  43. #define ARGS	 0

  44. 

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

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

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

  48. 

  49. #define RET	%eax

  50. #define	M	%ebx

  51. #define X	%ecx

  52. #define INCX	%edx

  53. #define I	%esi

  54. #define MM	%ebp

  55. #define XX	%edi

  56. #define TEMP	%ebx

  57. 

  58. #ifdef USE_MIN

  59. #define maxps	minps

  60. #define maxss	minss

  61. #endif

  62. 

  63. #ifndef HAVE_SSE2

  64. #define pxor	xorps

  65. #define movsd	movlps

  66. #endif

  67. 

  68. #include "l1param.h"

  69. 

  70. 	PROLOGUE

  71. 

  72. 	pushl	%ebp

  73. 	pushl	%edi

  74. 	pushl	%esi

  75. 	pushl	%ebx

  76. 

  77. 	PROFCODE

  78. 

  79. 	movl	STACK_M, M

  80. 	movl	STACK_X, X

  81. 	movl	STACK_INCX, INCX

  82. 

  83. 	pxor	%xmm0, %xmm0

  84. 	pxor	%xmm7, %xmm7

  85. 	xor	RET, RET

  86. 	testl	M, M

  87. 	jle	.L999

  88. 	testl	INCX, INCX

  89. 	jle	.L999

  90. 

  91. 	sall	$ZBASE_SHIFT, INCX

  92. 	movl	M, MM

  93. 	movl	X, XX

  94. 

  95. #ifdef USE_ABS

  96. #ifndef HAVE_SSE2

  97. 	subl	$8, %esp

  98. 	movl	$0x7fffffff, (%esp)

  99. 	movss	(%esp), %xmm7

  100. 	shufps	$0, %xmm7, %xmm7

  101. 	addl	$8, %esp

  102. #else

  103. 	cmpeqps	%xmm7, %xmm7

  104. 	psrld	$1, %xmm7		/* Generate USE_ABS */

  105. #endif

  106. #endif

  107. 

  108. 	movss	0 * SIZE(XX), %xmm0

  109. 	movss	1 * SIZE(XX), %xmm1

  110. 	addl	INCX, XX

  111. 	decl	MM

  112. 

  113. #ifdef USE_ABS

  114. 	andps	%xmm7, %xmm0

  115. 	andps	%xmm7, %xmm1

  116. #endif

  117. 	addps	%xmm1,  %xmm0

  118. 	shufps	$0, %xmm0, %xmm0

  119. 	cmpl	$2 * SIZE, INCX

  120. 	jne	.L70

  121. 

  122. .L30:

  123. 	movl	MM,  I

  124. 	sarl	$3, I

  125. 	jle	.L35

  126. 	ALIGN_4

  127. 

  128. .L31:

  129. #ifdef PREFETCH

  130. 	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(XX)

  131. #endif

  132. 

  133. 	movsd	 0 * SIZE(XX), %xmm1

  134. 	movhps	 2 * SIZE(XX), %xmm1

  135. 	movsd	 4 * SIZE(XX), %xmm2

  136. 	movhps	 6 * SIZE(XX), %xmm2

  137. 

  138. 	movaps	%xmm1, %xmm3

  139. 

  140. 	shufps	$0x88, %xmm2, %xmm1

  141. 	shufps	$0xdd, %xmm2, %xmm3

  142. 

  143. #ifdef USE_ABS

  144. 	andps	%xmm7, %xmm1

  145. 	andps	%xmm7, %xmm3

  146. 	addps	%xmm3,  %xmm1

  147. 	maxps	%xmm1,  %xmm0

  148. #endif

  149. 

  150. 	movsd	 8 * SIZE(XX), %xmm1

  151. 	movhps	10 * SIZE(XX), %xmm1

  152. 	movsd	12 * SIZE(XX), %xmm2

  153. 	movhps	14 * SIZE(XX), %xmm2

  154. 

  155. 	movaps	%xmm1, %xmm3

  156. 

  157. 	shufps	$0x88, %xmm2, %xmm1

  158. 	shufps	$0xdd, %xmm2, %xmm3

  159. 

  160. #ifdef USE_ABS

  161. 	andps	%xmm7, %xmm1

  162. 	andps	%xmm7, %xmm3

  163. #endif

  164. 

  165. 	addps	%xmm3,  %xmm1

  166. 	maxps	%xmm1,  %xmm0

  167. 

  168. 	addl	$16 * SIZE, XX

  169. 	decl	I

  170. 	jg	.L31

  171. 	ALIGN_4

  172. 

  173. .L35:

  174. 	andl	$7,  MM

  175. 	jle	.L40

  176. 

  177. 	testl	$4, MM

  178. 	je	.L36

  179. 

  180. 	movsd	 0 * SIZE(XX), %xmm1

  181. 	movhps	 2 * SIZE(XX), %xmm1

  182. 	movsd	 4 * SIZE(XX), %xmm2

  183. 	movhps	 6 * SIZE(XX), %xmm2

  184. 

  185. 	movaps	%xmm1, %xmm3

  186. 

  187. 	shufps	$0x88, %xmm2, %xmm1

  188. 	shufps	$0xdd, %xmm2, %xmm3

  189. 

  190. #ifdef USE_ABS

  191. 	andps	%xmm7, %xmm1

  192. 	andps	%xmm7, %xmm3

  193. #endif

  194. 	addps	%xmm3,  %xmm1

  195. 	maxps	%xmm1,  %xmm0

  196. 

  197. 	addl	$8 * SIZE, XX

  198. 	ALIGN_3

  199. 

  200. .L36:

  201. 	testl	$2, MM

  202. 	je	.L37

  203. 

  204. 	movss	0 * SIZE(XX), %xmm1

  205. 	movss	1 * SIZE(XX), %xmm2

  206. 	movss	2 * SIZE(XX), %xmm3

  207. 	movss	3 * SIZE(XX), %xmm4

  208. #ifdef USE_ABS

  209. 	andps	%xmm7, %xmm1

  210. 	andps	%xmm7, %xmm2

  211. 	andps	%xmm7, %xmm3

  212. 	andps	%xmm7, %xmm4

  213. #endif

  214. 	addps	%xmm2,  %xmm1

  215. 	addps	%xmm4,  %xmm3

  216. 	maxss	%xmm1,  %xmm0

  217. 	maxss	%xmm3,  %xmm0

  218. 	addl	$4 * SIZE, XX

  219. 	ALIGN_3

  220. 

  221. .L37:

  222. 	testl	$1, MM

  223. 	je	.L40

  224. 

  225. 	movss	0 * SIZE(XX), %xmm1

  226. 	movss	1 * SIZE(XX), %xmm2

  227. #ifdef USE_ABS

  228. 	andps	%xmm7, %xmm1

  229. 	andps	%xmm7, %xmm2

  230. #endif

  231. 	addps	%xmm2,  %xmm1

  232. 	maxss	%xmm1,  %xmm0

  233. 	ALIGN_4

  234. 

  235. .L40:

  236. 	movaps	%xmm0, %xmm1

  237. 	movhlps %xmm0, %xmm0

  238. 	maxps	%xmm1, %xmm0

  239. 	movaps	%xmm0, %xmm1

  240. 	shufps	$1, %xmm0, %xmm0

  241. 	maxss	%xmm1, %xmm0

  242. 	jmp	.L999

  243. 	ALIGN_4

  244. 

  245. .L70:

  246. 	movl	MM,  I

  247. 	sarl	$3, I

  248. 	jle	.L75

  249. 	ALIGN_4

  250. 

  251. .L71:

  252. #ifdef PREFETCH

  253. 	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(XX)

  254. #endif

  255. 

  256. 	movsd	0 * SIZE(XX), %xmm1

  257. 	addl	INCX, XX

  258. 	movhps	0 * SIZE(XX), %xmm1

  259. 	addl	INCX, XX

  260. 	movsd	0 * SIZE(XX), %xmm2

  261. 	addl	INCX, XX

  262. 	movhps	0 * SIZE(XX), %xmm2

  263. 	addl	INCX, XX

  264. 

  265. 	movaps	%xmm1, %xmm3

  266. 

  267. 	shufps	$0x88, %xmm2, %xmm1

  268. 	shufps	$0xdd, %xmm2, %xmm3

  269. 

  270. #ifdef USE_ABS

  271. 	andps	%xmm7, %xmm1

  272. 	andps	%xmm7, %xmm3

  273. #endif

  274. 	addps	%xmm3,  %xmm1

  275. 	maxps	%xmm1,  %xmm0

  276. 

  277. 	movsd	0 * SIZE(XX), %xmm1

  278. 	addl	INCX, XX

  279. 	movhps	0 * SIZE(XX), %xmm1

  280. 	addl	INCX, XX

  281. 	movsd	0 * SIZE(XX), %xmm2

  282. 	addl	INCX, XX

  283. 	movhps	0 * SIZE(XX), %xmm2

  284. 	addl	INCX, XX

  285. 

  286. 	movaps	%xmm1, %xmm3

  287. 

  288. 	shufps	$0x88, %xmm2, %xmm1

  289. 	shufps	$0xdd, %xmm2, %xmm3

  290. 

  291. #ifdef USE_ABS

  292. 	andps	%xmm7, %xmm1

  293. 	andps	%xmm7, %xmm3

  294. #endif

  295. 	addps	%xmm3,  %xmm1

  296. 	maxps	%xmm1,  %xmm0

  297. 	decl	I

  298. 	jg	.L71

  299. 	ALIGN_4

  300. 

  301. .L75:

  302. 	andl	$7,  MM

  303. 	jle	.L80

  304. 

  305. 	testl	$4, MM

  306. 	je	.L76

  307. 

  308. 	movsd	0 * SIZE(XX), %xmm1

  309. 	addl	INCX, XX

  310. 	movhps	0 * SIZE(XX), %xmm1

  311. 	addl	INCX, XX

  312. 	movsd	0 * SIZE(XX), %xmm2

  313. 	addl	INCX, XX

  314. 	movhps	0 * SIZE(XX), %xmm2

  315. 	addl	INCX, XX

  316. 

  317. 	movaps	%xmm1, %xmm3

  318. 

  319. 	shufps	$0x88, %xmm2, %xmm1

  320. 	shufps	$0xdd, %xmm2, %xmm3

  321. 

  322. #ifdef USE_ABS

  323. 	andps	%xmm7, %xmm1

  324. 	andps	%xmm7, %xmm3

  325. #endif

  326. 	addps	%xmm3,  %xmm1

  327. 	maxps	%xmm1,  %xmm0

  328. 	ALIGN_3

  329. 

  330. .L76:

  331. 	testl	$2, MM

  332. 	je	.L77

  333. 

  334. 	movss	0 * SIZE(XX), %xmm1

  335. 	movss	1 * SIZE(XX), %xmm2

  336. 	addl	INCX, XX

  337. 	movss	0 * SIZE(XX), %xmm3

  338. 	movss	1 * SIZE(XX), %xmm4

  339. 	addl	INCX, XX

  340. 

  341. #ifdef USE_ABS

  342. 	andps	%xmm7, %xmm1

  343. 	andps	%xmm7, %xmm2

  344. 	andps	%xmm7, %xmm3

  345. 	andps	%xmm7, %xmm4

  346. #endif

  347. 	addps	%xmm2,  %xmm1

  348. 	addps	%xmm4,  %xmm3

  349. 	maxss	%xmm1,  %xmm0

  350. 	maxss	%xmm3,  %xmm0

  351. 	ALIGN_3

  352. 

  353. .L77:

  354. 	testl	$1, MM

  355. 	je	.L80

  356. 

  357. 	movss	0 * SIZE(XX), %xmm1

  358. 	movss	1 * SIZE(XX), %xmm2

  359. #ifdef USE_ABS

  360. 	andps	%xmm7, %xmm1

  361. 	andps	%xmm7, %xmm2

  362. #endif

  363. 	addps	%xmm2,  %xmm1

  364. 	maxss	%xmm1,  %xmm0

  365. 	ALIGN_4

  366. 

  367. .L80:

  368. 	movaps	%xmm0, %xmm1

  369. 	movhlps %xmm0, %xmm0

  370. 	maxps	%xmm1, %xmm0

  371. 	movaps	%xmm0, %xmm1

  372. 	shufps	$1, %xmm0, %xmm0

  373. 	maxss	%xmm1, %xmm0

  374. 	ALIGN_4

  375. 

  376. .L999:

  377. 	subl	$8, %esp

  378. 	movss	%xmm0, (%esp)

  379. 	flds	(%esp)

  380. 	addl	$8, %esp

  381. 	popl	%ebx

  382. 	popl	%esi

  383. 	popl	%edi

  384. 	popl	%ebp

  385. 	ret

  386. 

  387. 	EPILOGUE