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/snrm2.S

335 lines
6.8 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. #define c1	%f0

  48. #define c2	%f2

  49. #define c3	%f4

  50. #define c4	%f6

  51. #define t1	%f8

  52. #define t2	%f10

  53. #define t3	%f12

  54. #define t4	%f14

  55. 

  56. #define a1	%f16

  57. #define a2	%f18

  58. #define a3	%f20

  59. #define a4	%f22

  60. #define a5	%f24

  61. #define a6	%f26

  62. #define a7	%f28

  63. #define a8	%f30

  64. 

  65. 	PROLOGUE

  66. 	SAVESP

  67. 

  68. 	FCLR(0)

  69. 

  70. 	FMOV	c1, c2

  71. 	FMOV	c1, c3

  72. 	FMOV	c1, c4

  73. 	FMOV	c1, t1

  74. 	FMOV	c1, t2

  75. 	FMOV	c1, t3

  76. 	FMOV	c1, t4

  77. 

  78. 	cmp	INCX, 0

  79. 	beq	.LL20

  80. 	sll	INCX, BASE_SHIFT, INCX

  81. 

  82. 	cmp	N, 0

  83. 	ble	.LL20

  84. 	nop

  85. 

  86. 	cmp	INCX, SIZE

  87. 	bne	.LL50

  88. 	nop

  89. 

  90. 	sra	N, 3, I

  91. 	cmp	I, 0

  92. 	ble,pn	%icc, .LL15

  93. 	nop

  94. 

  95. 	ld	[X +  0 * SIZE], a1

  96. 	add	I, -1, I

  97. 	ld	[X +  1 * SIZE], a2

  98. 	cmp	I, 0

  99. 	ld	[X +  2 * SIZE], a3

  100. 	ld	[X +  3 * SIZE], a4

  101. 	ld	[X +  4 * SIZE], a5

  102. 	ld	[X +  5 * SIZE], a6

  103. 	ld	[X +  6 * SIZE], a7

  104. 	ld	[X +  7 * SIZE], a8

  105. 

  106. 	ble,pt	%icc, .LL12

  107. 	add	X, 8 * SIZE, X

  108. 

  109. #define PREFETCHSIZE 40

  110. 

  111. .LL11:

  112. 	faddd	c1, t1, c1

  113. 	fsmuld	a1, a1, t1

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

  115. 

  116. 	faddd	c2, t2, c2

  117. 	add	I, -1, I

  118. 	fsmuld	a2, a2, t2

  119. 	ld	[X +  0 * SIZE], a1

  120. 

  121. 	faddd	c3, t3, c3

  122. 	cmp	I, 0

  123. 	fsmuld	a3, a3, t3

  124. 	ld	[X +  1 * SIZE], a2

  125. 

  126. 	faddd	c4, t4, c4

  127. 	fsmuld	a4, a4, t4

  128. 	ld	[X +  2 * SIZE], a3

  129. 

  130. 	faddd	c1, t1, c1

  131. 	fsmuld	a5, a5, t1

  132. 	ld	[X +  3 * SIZE], a4

  133. 

  134. 	faddd	c2, t2, c2

  135. 	fsmuld	a6, a6, t2

  136. 	ld	[X +  4 * SIZE], a5

  137. 

  138. 	faddd	c3, t3, c3

  139. 	fsmuld	a7, a7, t3

  140. 	ld	[X +  5 * SIZE], a6

  141. 

  142. 	faddd	c4, t4, c4

  143. 	ld	[X +  6 * SIZE], a7

  144. 	fsmuld	a8, a8, t4

  145. 	add	X, 8 * SIZE, X

  146. 

  147. 	bg,pt	%icc, .LL11

  148. 	ld	[X -  1 * SIZE], a8

  149. 

  150. .LL12:

  151. 	faddd	c1, t1, c1

  152. 	fsmuld	a1, a1, t1

  153. 	faddd	c2, t2, c2

  154. 	fsmuld	a2, a2, t2

  155. 

  156. 	faddd	c3, t3, c3

  157. 	fsmuld	a3, a3, t3

  158. 	faddd	c4, t4, c4

  159. 	fsmuld	a4, a4, t4

  160. 

  161. 	faddd	c1, t1, c1

  162. 	fsmuld	a5, a5, t1

  163. 	faddd	c2, t2, c2

  164. 	fsmuld	a6, a6, t2

  165. 

  166. 	faddd	c3, t3, c3

  167. 	fsmuld	a7, a7, t3

  168. 	faddd	c4, t4, c4

  169. 	fsmuld	a8, a8, t4

  170. 

  171. .LL15:

  172. 	and	N, 7, I

  173. 	cmp	I,  0

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

  175. 	nop

  176. 

  177. .LL16:

  178. 	ld	[X +  0 * SIZE], a1

  179. 

  180. 	add	I, -1, I

  181. 	cmp	I, 0

  182. 	faddd	c1, t1, c1

  183. 	fsmuld	a1, a1, t1

  184. 

  185. 	bg,pt	%icc, .LL16

  186. 	add	X, 1 * SIZE, X

  187. 

  188. .LL19:

  189. 	faddd	c1, t1, c1

  190. 	faddd	c2, t2, c2

  191. 	faddd	c3, t3, c3

  192. 	faddd	c4, t4, c4

  193. 

  194. 	faddd	c1, c2, c1

  195. 	faddd	c3, c4, c3

  196. 	faddd	c1, c3, c1

  197. 

  198. 	fsqrtd	c1, c1

  199. 

  200. #if !defined(NEED_F2CCONV) || !defined(F_INTERFACE_F2C)

  201. 	fdtos	c1, c1

  202. #endif

  203. .LL20:

  204. 

  205. 	return	%i7 + 8

  206. 	clr	%g0

  207. 

  208. .LL50:

  209. 	sra	N, 3, I

  210. 	cmp	I, 0

  211. 	ble,pn	%icc, .LL55

  212. 	nop

  213. 

  214. 	ld	[X +  0 * SIZE], a1

  215. 	add	X, INCX, X

  216. 	ld	[X +  0 * SIZE], a2

  217. 	add	X, INCX, X

  218. 	ld	[X +  0 * SIZE], a3

  219. 	add	X, INCX, X

  220. 	ld	[X +  0 * SIZE], a4

  221. 	add	X, INCX, X

  222. 	ld	[X +  0 * SIZE], a5

  223. 	add	X, INCX, X

  224. 	ld	[X +  0 * SIZE], a6

  225. 	add	X, INCX, X

  226. 	add	I, -1, I

  227. 	ld	[X +  0 * SIZE], a7

  228. 	cmp	I, 0

  229. 	add	X, INCX, X

  230. 	ld	[X +  0 * SIZE], a8

  231. 

  232. 	ble,pt	%icc, .LL52

  233. 	add	X, INCX, X

  234. 

  235. .LL51:

  236. 	faddd	c1, t1, c1

  237. 	add	I, -1, I

  238. 	fsmuld	a1, a1, t1

  239. 	ld	[X +  0 * SIZE], a1

  240. 	add	X, INCX, X

  241. 

  242. 	faddd	c2, t2, c2

  243. 	cmp	I, 0

  244. 	fsmuld	a2, a2, t2

  245. 	ld	[X +  0 * SIZE], a2

  246. 	add	X, INCX, X

  247. 

  248. 	faddd	c3, t3, c3

  249. 	fsmuld	a3, a3, t3

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

  251. 	add	X, INCX, X

  252. 

  253. 	faddd	c4, t4, c4

  254. 	fsmuld	a4, a4, t4

  255. 	ld	[X +  0 * SIZE], a4

  256. 	add	X, INCX, X

  257. 

  258. 	faddd	c1, t1, c1

  259. 	fsmuld	a5, a5, t1

  260. 	ld	[X +  0 * SIZE], a5

  261. 	add	X, INCX, X

  262. 

  263. 	faddd	c2, t2, c2

  264. 	fsmuld	a6, a6, t2

  265. 	ld	[X +  0 * SIZE], a6

  266. 	add	X, INCX, X

  267. 

  268. 	faddd	c3, t3, c3

  269. 	fsmuld	a7, a7, t3

  270. 	ld	[X +  0 * SIZE], a7

  271. 	add	X, INCX, X

  272. 

  273. 	faddd	c4, t4, c4

  274. 	fsmuld	a8, a8, t4

  275. 	ld	[X +  0 * SIZE], a8

  276. 	bg,pt	%icc, .LL51

  277. 	add	X, INCX, X

  278. 

  279. .LL52:

  280. 	faddd	c1, t1, c1

  281. 	fsmuld	a1, a1, t1

  282. 	faddd	c2, t2, c2

  283. 	fsmuld	a2, a2, t2

  284. 

  285. 	faddd	c3, t3, c3

  286. 	fsmuld	a3, a3, t3

  287. 	faddd	c4, t4, c4

  288. 	fsmuld	a4, a4, t4

  289. 

  290. 	faddd	c1, t1, c1

  291. 	fsmuld	a5, a5, t1

  292. 	faddd	c2, t2, c2

  293. 	fsmuld	a6, a6, t2

  294. 

  295. 	faddd	c3, t3, c3

  296. 	fsmuld	a7, a7, t3

  297. 	faddd	c4, t4, c4

  298. 	fsmuld	a8, a8, t4

  299. 

  300. .LL55:

  301. 	and	N, 7, I

  302. 	cmp	I,  0

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

  304. 	nop

  305. 

  306. .LL56:

  307. 	ld	[X +  0 * SIZE], a1

  308. 	add	I, -1, I

  309. 	cmp	I, 0

  310. 	faddd	c1, t1, c1

  311. 	fsmuld	a1, a1, t1

  312. 	bg,pt	%icc, .LL56

  313. 	add	X, INCX, X

  314. 

  315. .LL59:

  316. 	faddd	c1, t1, c1

  317. 	faddd	c2, t2, c2

  318. 	faddd	c3, t3, c3

  319. 	faddd	c4, t4, c4

  320. 

  321. 	faddd	c1, c2, c1

  322. 	faddd	c3, c4, c3

  323. 	faddd	c1, c3, c1

  324. 

  325. 	fsqrtd	c1, c1

  326. 

  327. #if !defined(NEED_F2CCONV) || !defined(F_INTERFACE_F2C)

  328. 	fdtos	c1, c1

  329. #endif

  330. 

  331. 	return	%i7 + 8

  332. 	clr	%o0

  333. 

  334. 	EPILOGUE