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/sparc/zsum.S

328 lines
6.5 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 N	%i0

  43. #define X	%i1

  44. #define INCX	%i2

  45. #define I	%i3

  46. 

  47. #ifdef DOUBLE

  48. #define c1	%f0

  49. #define c2	%f2

  50. #define t1	%f8

  51. #define t2	%f10

  52. #define t3	%f12

  53. #define t4	%f14

  54. 

  55. #define a1	%f16

  56. #define a2	%f18

  57. #define a3	%f20

  58. #define a4	%f22

  59. #define a5	%f24

  60. #define a6	%f26

  61. #define a7	%f28

  62. #define a8	%f30

  63. #else

  64. #define c1	%f0

  65. #define c2	%f1

  66. #define t1	%f4

  67. #define t2	%f5

  68. #define t3	%f6

  69. #define t4	%f7

  70. 

  71. #define a1	%f8

  72. #define a2	%f9

  73. #define a3	%f10

  74. #define a4	%f11

  75. #define a5	%f12

  76. #define a6	%f13

  77. #define a7	%f14

  78. #define a8	%f15

  79. #endif

  80. 

  81. 	PROLOGUE

  82. 	SAVESP

  83. 

  84. 	FCLR(0)

  85. 

  86. 	sll	INCX, ZBASE_SHIFT, INCX

  87. 

  88. 	FMOV	c1, c2

  89. 	FMOV	c1, t1

  90. 	FMOV	c1, t2

  91. 	FMOV	c1, t3

  92. 	FMOV	c1, t4

  93. 

  94. 	cmp	INCX, 0

  95. 	ble	.LL19

  96. 	nop

  97. 

  98. 	cmp	INCX, 2 * SIZE

  99. 	bne	.LL50

  100. 	nop

  101. 

  102. 	sra	N, 2, I

  103. 	cmp	I, 0

  104. 	ble,pn	%icc, .LL15

  105. 	nop

  106. 

  107. 	LDF	[X +  0 * SIZE], a1

  108. 	add	I, -1, I

  109. 	LDF	[X +  1 * SIZE], a2

  110. 	cmp	I, 0

  111. 	LDF	[X +  2 * SIZE], a3

  112. 	LDF	[X +  3 * SIZE], a4

  113. 	LDF	[X +  4 * SIZE], a5

  114. 	LDF	[X +  5 * SIZE], a6

  115. 	LDF	[X +  6 * SIZE], a7

  116. 	LDF	[X +  7 * SIZE], a8

  117. 

  118. 	ble,pt	%icc, .LL12

  119. 	add	X, 8 * SIZE, X

  120. 

  121. #define PREFETCHSIZE 32

  122. 

  123. .LL11:

  124. 	FADD	c1, t1, c1

  125. 	prefetch [X  + PREFETCHSIZE * SIZE], 0

  126. 	FMOV	a1, t1

  127. 	LDF	[X +  0 * SIZE], a1

  128. 

  129. 	FADD	c2, t2, c2

  130. 	add	I, -1, I

  131. 	FMOV	a2, t2

  132. 	LDF	[X +  1 * SIZE], a2

  133. 

  134. 	FADD	c1, t3, c1

  135. 	cmp	I, 0

  136. 	FMOV	a3, t3

  137. 	LDF	[X +  2 * SIZE], a3

  138. 

  139. 	FADD	c2, t4, c2

  140. 	nop

  141. 	FMOV	a4, t4

  142. 	LDF	[X +  3 * SIZE], a4

  143. 

  144. 	FADD	c1, t1, c1

  145. 	nop

  146. 	FMOV	a5, t1

  147. 	LDF	[X +  4 * SIZE], a5

  148. 

  149. 	FADD	c2, t2, c2

  150. 	nop

  151. 	FMOV	a6, t2

  152. 	LDF	[X +  5 * SIZE], a6

  153. 

  154. 	FADD	c1, t3, c1

  155. 	FMOV	a7, t3

  156. 	LDF	[X +  6 * SIZE], a7

  157. 	add	X, 8 * SIZE, X

  158. 

  159. 	FADD	c2, t4, c2

  160. 	FMOV	a8, t4

  161. 	bg,pt	%icc, .LL11

  162. 	LDF	[X -  1 * SIZE], a8

  163. 

  164. .LL12:

  165. 	FADD	c1, t1, c1

  166. 	FMOV	a1, t1

  167. 	FADD	c2, t2, c2

  168. 	FMOV	a2, t2

  169. 

  170. 	FADD	c1, t3, c1

  171. 	FMOV	a3, t3

  172. 	FADD	c2, t4, c2

  173. 	FMOV	a4, t4

  174. 

  175. 	FADD	c1, t1, c1

  176. 	FMOV	a5, t1

  177. 	FADD	c2, t2, c2

  178. 	FMOV	a6, t2

  179. 

  180. 	FADD	c1, t3, c1

  181. 	FMOV	a7, t3

  182. 	FADD	c2, t4, c2

  183. 	FMOV	a8, t4

  184. 

  185. .LL15:

  186. 	and	N, 3, I

  187. 	cmp	I,  0

  188. 	ble,a,pn %icc, .LL19

  189. 	nop

  190. 

  191. .LL16:

  192. 	LDF	[X +  0 * SIZE], a1

  193. 	LDF	[X +  1 * SIZE], a2

  194. 	add	I, -1, I

  195. 	cmp	I, 0

  196. 	FADD	c1, t1, c1

  197. 	FADD	c2, t2, c2

  198. 	FMOV	a1, t1

  199. 	FMOV	a2, t2

  200. 	bg,pt	%icc, .LL16

  201. 	add	X, 2 * SIZE, X

  202. 

  203. .LL19:

  204. 	FADD	c1, t1, c1

  205. 	FADD	c2, t2, c2

  206. 	FADD	c1, t3, c1

  207. 	FADD	c2, t4, c2

  208. 

  209. 	FADD	c1, c2, c1

  210. 	return	%i7 + 8

  211. 	clr	%g0

  212. 

  213. .LL50:

  214. 	sra	N, 2, I

  215. 	cmp	I, 0

  216. 	ble,pn	%icc, .LL55

  217. 	nop

  218. 

  219. 	LDF	[X +  0 * SIZE], a1

  220. 	LDF	[X +  1 * SIZE], a2

  221. 	add	X, INCX, X

  222. 	LDF	[X +  0 * SIZE], a3

  223. 	LDF	[X +  1 * SIZE], a4

  224. 	add	X, INCX, X

  225. 	LDF	[X +  0 * SIZE], a5

  226. 	LDF	[X +  1 * SIZE], a6

  227. 	add	X, INCX, X

  228. 	add	I, -1, I

  229. 	LDF	[X +  0 * SIZE], a7

  230. 	cmp	I, 0

  231. 	LDF	[X +  1 * SIZE], a8

  232. 

  233. 	ble,pt	%icc, .LL52

  234. 	add	X, INCX, X

  235. 

  236. .LL51:

  237. 	FADD	c1, t1, c1

  238. 	add	I, -1, I

  239. 	FMOV	a1, t1

  240. 	LDF	[X +  0 * SIZE], a1

  241. 

  242. 	FADD	c2, t2, c2

  243. 	cmp	I, 0

  244. 	FMOV	a2, t2

  245. 	LDF	[X +  1 * SIZE], a2

  246. 	add	X, INCX, X

  247. 

  248. 	FADD	c1, t3, c1

  249. 	FMOV	a3, t3

  250. 	LDF	[X +  0 * SIZE], a3

  251. 

  252. 	FADD	c2, t4, c2

  253. 	FMOV	a4, t4

  254. 	LDF	[X +  1 * SIZE], a4

  255. 	add	X, INCX, X

  256. 

  257. 	FADD	c1, t1, c1

  258. 	FMOV	a5, t1

  259. 	LDF	[X +  0 * SIZE], a5

  260. 

  261. 	FADD	c2, t2, c2

  262. 	FMOV	a6, t2

  263. 	LDF	[X +  1 * SIZE], a6

  264. 	add	X, INCX, X

  265. 

  266. 	FADD	c1, t3, c1

  267. 	FMOV	a7, t3

  268. 	LDF	[X +  0 * SIZE], a7

  269. 

  270. 	FADD	c2, t4, c2

  271. 	FMOV	a8, t4

  272. 	LDF	[X +  1 * SIZE], a8

  273. 

  274. 	bg,pt	%icc, .LL51

  275. 	add	X, INCX, X

  276. 

  277. .LL52:

  278. 	FADD	c1, t1, c1

  279. 	FMOV	a1, t1

  280. 	FADD	c2, t2, c2

  281. 	FMOV	a2, t2

  282. 

  283. 	FADD	c1, t3, c1

  284. 	FMOV	a3, t3

  285. 	FADD	c2, t4, c2

  286. 	FMOV	a4, t4

  287. 

  288. 	FADD	c1, t1, c1

  289. 	FMOV	a5, t1

  290. 	FADD	c2, t2, c2

  291. 	FMOV	a6, t2

  292. 

  293. 	FADD	c1, t3, c1

  294. 	FMOV	a7, t3

  295. 	FADD	c2, t4, c2

  296. 	FMOV	a8, t4

  297. 

  298. .LL55:

  299. 	and	N, 3, I

  300. 	cmp	I,  0

  301. 	ble,a,pn %icc, .LL59

  302. 	nop

  303. 

  304. .LL56:

  305. 	LDF	[X +  0 * SIZE], a1

  306. 	LDF	[X +  1 * SIZE], a2

  307. 	FADD	c1, t1, c1

  308. 	FADD	c2, t2, c2

  309. 	add	I, -1, I

  310. 	FMOV	a1, t1

  311. 	FMOV	a2, t2

  312. 	cmp	I, 0

  313. 	bg,pt	%icc, .LL56

  314. 	add	X, INCX, X

  315. 

  316. .LL59:

  317. 	FADD	c1, t1, c1

  318. 	FADD	c2, t2, c2

  319. 	FADD	c1, t3, c1

  320. 	FADD	c2, t4, c2

  321. 

  322. 	FADD	c1, c2, c1

  323. 

  324. 	return	%i7 + 8

  325. 	clr	%o0

  326. 

  327. 	EPILOGUE