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: e1afb23811
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e1afb23811'
${ noResults }
OpenBLAS/kernel/alpha/axpy.S

429 lines
9.0 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. 

  43. #define PREFETCHSIZE 40

  44. 

  45. 	PROLOGUE

  46. 	PROFCODE

  47. 	.frame	$sp, 16, $26, 0

  48. 

  49. 	ldq	$24,   0($sp)

  50. 	fmov	$f19,  $f30

  51. 	ldl	$23,   8($sp)

  52. 	lda	$sp, -16($sp)

  53. #ifndef PROFILE

  54. 	.prologue 0

  55. #else

  56. 	.prologue 1

  57. #endif

  58. 

  59. 	nop

  60. 	sra	$16,  3,  $1

  61. 	stt	$f2,   0($sp)

  62. 	cmpeq	$21,  1,  $3

  63. 

  64. 	stt	$f3,   8($sp)

  65. 	cmpeq	$23,  1, $4

  66. 	and	$16,  7,  $2

  67. 	ble	$16, $End

  68. 

  69. 	and	$3,  $4,  $3

  70. 	fbeq	$f30, $End

  71. 

  72. 	beq	$3,  $Sub

  73. 	ble	$1,  $Remain

  74. 	.align 4

  75. 

  76. 	LD	$f10,  0*SIZE($20)

  77. 	LD	$f11,  1*SIZE($20)

  78. 	LD	$f12,  2*SIZE($20)

  79. 	LD	$f13,  3*SIZE($20)

  80. 

  81. 	LD	$f18,  0*SIZE($24)

  82. 	LD	$f19,  1*SIZE($24)

  83. 	LD	$f20,  2*SIZE($24)

  84. 	LD	$f21,  3*SIZE($24)

  85. 

  86. 	LD	$f14,  4*SIZE($20)

  87. 	LD	$f15,  5*SIZE($20)

  88. 	LD	$f16,  6*SIZE($20)

  89. 	LD	$f17,  7*SIZE($20)

  90. 

  91. 	LD	$f22,  4*SIZE($24)

  92. 	LD	$f23,  5*SIZE($24)

  93. 	LD	$f24,  6*SIZE($24)

  94. 	LD	$f25,  7*SIZE($24)

  95. 

  96. 	subq	$1,   1,  $1

  97. 	addq	$20, 8*SIZE, $20

  98. 	unop

  99. 	ble	$1,  $LoopEnd

  100. 	.align 4

  101. 

  102. $Loop:

  103. 	ldt	$f31, PREFETCHSIZE * SIZE($24)

  104. 	ldl	$31,  PREFETCHSIZE * SIZE($20)

  105. 

  106. 	MUL	$f30, $f10, $f26		# ctemp1 = da * atemp1

  107. 	LD	$f10,  0*SIZE($20)

  108. 	MUL	$f30, $f11, $f27

  109. 	LD	$f11,  1*SIZE($20)

  110. 

  111. 	MUL	$f30, $f12, $f28

  112. 	LD	$f12,  2*SIZE($20)

  113. 	MUL	$f30, $f13, $f29

  114. 	LD	$f13,  3*SIZE($20)

  115. 

  116. 	ADD	$f18, $f26, $f0

  117. 	LD	$f18,  8*SIZE($24)

  118. 	MUL	$f30, $f14, $f26		# ctemp1 = da * atemp1

  119. 	LD	$f14,  4*SIZE($20)

  120. 

  121. 	ADD	$f19, $f27, $f1

  122. 	LD	$f19,  9*SIZE($24)

  123. 	MUL	$f30, $f15, $f27

  124. 	LD	$f15,  5*SIZE($20)

  125. 

  126. 	ADD	$f20, $f28, $f2

  127. 	LD	$f20, 10*SIZE($24)

  128. 	MUL	$f30, $f16, $f28

  129. 	LD	$f16,  6*SIZE($20)

  130. 

  131. 	ADD	$f21, $f29, $f3

  132. 	LD	$f21, 11*SIZE($24)

  133. 	MUL	$f30, $f17, $f29

  134. 	LD	$f17, 7*SIZE($20)

  135. 

  136. 	ST	$f0,   0*SIZE($24)

  137. 	ADD	$f22, $f26, $f0

  138. 	ST	$f1,   1*SIZE($24)

  139. 	ADD	$f23, $f27, $f1

  140. 

  141. 	ST	$f2,   2*SIZE($24)

  142. 	ADD	$f24, $f28, $f2

  143. 	ST	$f3,   3*SIZE($24)

  144. 	ADD	$f25, $f29, $f3

  145. 

  146. 	LD	$f22, 12*SIZE($24)

  147. 	LD	$f23, 13*SIZE($24)

  148. 	LD	$f24, 14*SIZE($24)

  149. 	LD	$f25, 15*SIZE($24)

  150. 

  151. 	ST	$f0,  4*SIZE($24)

  152. 	ST	$f1,  5*SIZE($24)

  153. 	ST	$f2,  6*SIZE($24)

  154. 	ST	$f3,  7*SIZE($24)

  155. 

  156. 	subq	$1,  1, $1

  157. 	addq	$24, 8*SIZE, $24

  158. 	addq	$20, 8*SIZE, $20

  159. 	bgt	$1, $Loop

  160. 	.align 4

  161. 

  162. $LoopEnd:

  163. 	MUL	$f30, $f10, $f26		# ctemp1 = da * atemp1

  164. 	MUL	$f30, $f11, $f27

  165. 	MUL	$f30, $f12, $f28

  166. 	MUL	$f30, $f13, $f29

  167. 

  168. 	ADD	$f18, $f26, $f0

  169. 	MUL	$f30, $f14, $f26		# ctemp1 = da * atemp1

  170. 	ADD	$f19, $f27, $f1

  171. 	MUL	$f30, $f15, $f27

  172. 

  173. 	ADD	$f20, $f28, $f2

  174. 	MUL	$f30, $f16, $f28

  175. 	ADD	$f21, $f29, $f3

  176. 	MUL	$f30, $f17, $f29

  177. 

  178. 	ST	$f0,   0*SIZE($24)

  179. 	ADD	$f22, $f26, $f0

  180. 	ST	$f1,   1*SIZE($24)

  181. 	ADD	$f23, $f27, $f1

  182. 

  183. 	ST	$f2,   2*SIZE($24)

  184. 	ADD	$f24, $f28, $f2

  185. 	ST	$f3,   3*SIZE($24)

  186. 	ADD	$f25, $f29, $f3

  187. 

  188. 	ST	$f0,   4*SIZE($24)

  189. 	ST	$f1,   5*SIZE($24)

  190. 	ST	$f2,   6*SIZE($24)

  191. 	ST	$f3,   7*SIZE($24)

  192. 	addq	$24, 8*SIZE, $24

  193. 	.align 4

  194. 

  195. $Remain:

  196. 	ble	$2, $End

  197. 	.align 4

  198. 

  199. $RemainLoop:

  200. 	LD	$f10,  0*SIZE($20)

  201. 	LD	$f11,  0*SIZE($24)

  202. 	addq	$20, SIZE, $20

  203. 	addq	$24, SIZE, $24

  204. 

  205. 	MUL	$f30, $f10, $f12

  206. 	subq	$2,  1,  $2

  207. 	ADD	$f11, $f12, $f13

  208. 	ST	$f13,  -1*SIZE($24)

  209. 	bgt	$2,  $RemainLoop

  210. 	.align 4

  211. 

  212. $End:

  213. 	ldt	$f2,   0($sp)

  214. 	ldt	$f3,   8($sp)

  215. 	lda	$sp,  16($sp)

  216. 	ret

  217. 	.align 4

  218. 

  219. $Sub:

  220. 	SXSUBL	$16,  SIZE, $22

  221. 	subq	$1,  1, $4

  222. 	ble	$1, $SubRemain

  223. 	.align 4

  224. 

  225. 	LD	$f10,  0($20)

  226. 	SXADDQ	$21, $20, $20

  227. 

  228. 	LD	$f11,  0($20)

  229. 	SXADDQ	$21, $20, $20

  230. 	LD	$f12,  0($20)

  231. 	SXADDQ	$21, $20, $20

  232. 

  233. 	LD	$f13,  0($20)

  234. 	SXADDQ	$21, $20, $20

  235. 	LD	$f18,  0($24)

  236. 	SXADDQ	$23, $24, $22

  237. 

  238. 	LD	$f19,  0($22)

  239. 	SXADDQ	$23, $22, $22

  240. 	LD	$f20,  0($22)

  241. 	SXADDQ	$23, $22, $22

  242. 

  243. 	LD	$f21,  0($22)

  244. 	SXADDQ	$23, $22, $22

  245. 	LD	$f14,  0($20)

  246. 	SXADDQ	$21, $20, $20

  247. 

  248. 	LD	$f15,  0($20)

  249. 	SXADDQ	$21, $20, $20

  250. 	LD	$f16,  0($20)

  251. 	SXADDQ	$21, $20, $20

  252. 

  253. 	LD	$f17,  0($20)

  254. 	SXADDQ	$21, $20, $20

  255. 	LD	$f22,  0($22)

  256. 	SXADDQ	$23, $22, $22

  257. 

  258. 	LD	$f23,  0($22)

  259. 	SXADDQ	$23, $22, $22

  260. 	LD	$f24,  0($22)

  261. 	SXADDQ	$23, $22, $22

  262. 

  263. 	LD	$f25,  0($22)

  264. 	SXADDQ	$23, $22, $22

  265. 	unop

  266. 	ble	$4,  $SubLoopEnd

  267. 	.align 4

  268. 

  269. $SubLoop:

  270. 	MUL	$f30, $f10, $f26		# ctemp1 = da * atemp1

  271. 	LD	$f10,  0($20)

  272. 	unop

  273. 	SXADDQ	$21, $20, $20

  274. 

  275. 	MUL	$f30, $f11, $f27

  276. 	LD	$f11,  0($20)

  277. 	unop

  278. 	SXADDQ	$21, $20, $20

  279. 

  280. 	MUL	$f30, $f12, $f28

  281. 	LD	$f12,  0($20)

  282. 	unop

  283. 	SXADDQ	$21, $20, $20

  284. 

  285. 	MUL	$f30, $f13, $f29

  286. 	LD	$f13,  0($20)

  287. 	unop

  288. 	SXADDQ	$21, $20, $20

  289. 

  290. 	ADD	$f18, $f26, $f0

  291. 	MUL	$f30, $f14, $f26		# ctemp1 = da * atemp1

  292. 	LD	$f14,  0($20)

  293. 	SXADDQ	$21, $20, $20

  294. 

  295. 	ADD	$f19, $f27, $f1

  296. 	MUL	$f30, $f15, $f27

  297. 	LD	$f15,  0($20)

  298. 	SXADDQ	$21, $20, $20

  299. 

  300. 	ADD	$f20, $f28, $f2

  301. 	MUL	$f30, $f16, $f28

  302. 	LD	$f16,  0($20)

  303. 	SXADDQ	$21, $20, $20

  304. 

  305. 	ADD	$f21, $f29, $f3

  306. 	MUL	$f30, $f17, $f29

  307. 	LD	$f17,  0($20)

  308. 	SXADDQ	$21, $20, $20

  309. 

  310. 	ST	$f0,   0($24)

  311. 	SXADDQ	$23, $24, $24

  312. 	ADD	$f22, $f26, $f0

  313. 	unop

  314. 

  315. 	ST	$f1,   0($24)

  316. 	SXADDQ	$23, $24, $24

  317. 	ADD	$f23, $f27, $f1

  318. 	unop

  319. 

  320. 	ST	$f2,   0($24)

  321. 	SXADDQ	$23, $24, $24

  322. 	ADD	$f24, $f28, $f2

  323. 	unop

  324. 

  325. 	ST	$f3,   0($24)

  326. 	SXADDQ	$23, $24, $24

  327. 	ADD	$f25, $f29, $f3

  328. 	unop

  329. 

  330. 	LD	$f18,  0($22)

  331. 	SXADDQ	$23, $22, $22

  332. 	LD	$f19,  0($22)

  333. 	SXADDQ	$23, $22, $22

  334. 

  335. 	LD	$f20,  0($22)

  336. 	SXADDQ	$23, $22, $22

  337. 	LD	$f21,  0($22)

  338. 	SXADDQ	$23, $22, $22

  339. 

  340. 	LD	$f22,  0($22)

  341. 	SXADDQ	$23, $22, $22

  342. 	LD	$f23,  0($22)

  343. 	SXADDQ	$23, $22, $22

  344. 

  345. 	LD	$f24,  0($22)

  346. 	SXADDQ	$23, $22, $22

  347. 	LD	$f25,  0($22)

  348. 	SXADDQ	$23, $22, $22

  349. 

  350. 	ST	$f0,  0($24)

  351. 	SXADDQ	$23, $24, $24

  352. 	ST	$f1,  0($24)

  353. 	SXADDQ	$23, $24, $24

  354. 	ST	$f2,  0($24)

  355. 	SXADDQ	$23, $24, $24

  356. 	ST	$f3,  0($24)

  357. 	SXADDQ	$23, $24, $24

  358. 

  359. 	subq	$4,   1,  $4

  360. 	bgt	$4, $SubLoop

  361. 	.align 4

  362. 

  363. $SubLoopEnd:

  364. 	MUL	$f30, $f10, $f26		# ctemp1 = da * atemp1

  365. 	MUL	$f30, $f11, $f27

  366. 	MUL	$f30, $f12, $f28

  367. 	MUL	$f30, $f13, $f29

  368. 

  369. 	ADD	$f18, $f26, $f0

  370. 	MUL	$f30, $f14, $f26		# ctemp1 = da * atemp1

  371. 	ADD	$f19, $f27, $f1

  372. 	MUL	$f30, $f15, $f27

  373. 

  374. 	ADD	$f20, $f28, $f2

  375. 	MUL	$f30, $f16, $f28

  376. 	ADD	$f21, $f29, $f3

  377. 	MUL	$f30, $f17, $f29

  378. 

  379. 	ST	$f0,   0($24)

  380. 	SXADDQ	$23, $24, $24

  381. 	ST	$f1,   0($24)

  382. 	SXADDQ	$23, $24, $24

  383. 

  384. 	ST	$f2,   0($24)

  385. 	SXADDQ	$23, $24, $24

  386. 	ST	$f3,   0($24)

  387. 	SXADDQ	$23, $24, $24

  388. 

  389. 	ADD	$f22, $f26, $f0

  390. 	ADD	$f23, $f27, $f1

  391. 	ADD	$f24, $f28, $f2

  392. 	ADD	$f25, $f29, $f3

  393. 

  394. 	ST	$f0,   0($24)

  395. 	SXADDQ	$23, $24, $24

  396. 	ST	$f1,   0($24)

  397. 	SXADDQ	$23, $24, $24

  398. 

  399. 	ST	$f2,   0($24)

  400. 	SXADDQ	$23, $24, $24

  401. 	ST	$f3,   0($24)

  402. 	SXADDQ	$23, $24, $24

  403. 	.align 4

  404. 

  405. $SubRemain:

  406. 	ble	$2, $SubEnd

  407. 	.align 4

  408. 

  409. $SubRemainLoop:

  410. 	LD	$f10,  0($20)

  411. 	LD	$f11,  0($24)

  412. 	SXADDQ	$21, $20, $20

  413. 

  414. 	MUL	$f30, $f10, $f12

  415. 	subq	$2,  1,  $2

  416. 	ADD	$f11, $f12, $f13

  417. 	ST	$f13,  0($24)

  418. 	SXADDQ	$23, $24, $24

  419. 

  420. 	bgt	$2,  $SubRemainLoop

  421. 	.align 4

  422. 

  423. $SubEnd:

  424. 	ldt	$f2,   0($sp)

  425. 	ldt	$f3,   8($sp)

  426. 	lda	$sp,  16($sp)

  427. 	ret

  428. 	EPILOGUE