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.

amin_hummer.S 9.9 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539
  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. #define ASSEMBLER
  39. #include "common.h"
  40. #define N r3
  41. #define X r4
  42. #define INCX r5
  43. #define INCX2 r6
  44. #define X2 r7
  45. #define C1 f1
  46. #define C2 f0
  47. #define C3 f2
  48. #define C4 f3
  49. #define A1 f4
  50. #define A2 f5
  51. #define A3 f6
  52. #define A4 f7
  53. #define A5 f8
  54. #define A6 f9
  55. #define A7 f10
  56. #define A8 f11
  57. #define F1 f12
  58. #define F2 f13
  59. #define F3 f14
  60. #define F4 f15
  61. #define F5 f16
  62. #define F6 f17
  63. #define F7 f18
  64. #define F8 f19
  65. #define T1 f20
  66. #define T2 f21
  67. #define T3 f22
  68. #define T4 f23
  69. #define T5 f24
  70. #define T6 f25
  71. #define T7 f26
  72. #define T8 f27
  73. PROLOGUE
  74. PROFCODE
  75. li r10, -16
  76. stfpdux f14, SP, r10
  77. stfpdux f15, SP, r10
  78. stfpdux f16, SP, r10
  79. stfpdux f17, SP, r10
  80. stfpdux f18, SP, r10
  81. stfpdux f19, SP, r10
  82. stfpdux f20, SP, r10
  83. stfpdux f21, SP, r10
  84. stfpdux f22, SP, r10
  85. stfpdux f23, SP, r10
  86. stfpdux f24, SP, r10
  87. stfpdux f25, SP, r10
  88. stfpdux f26, SP, r10
  89. stfpdux f27, SP, r10
  90. li r10, 0
  91. stwu r10, -4(SP)
  92. stwu r10, -4(SP)
  93. stwu r10, -4(SP)
  94. stwu r10, -4(SP)
  95. #ifdef F_INTERFACE
  96. LDINT N, 0(N)
  97. LDINT INCX, 0(INCX)
  98. #endif
  99. lfpdx C1, SP, r10 # Zero clear
  100. slwi INCX, INCX, BASE_SHIFT
  101. add INCX2, INCX, INCX
  102. cmpwi cr0, N, 0
  103. ble LL(999)
  104. cmpwi cr0, INCX, 0
  105. ble LL(999)
  106. LFD C1, 0 * SIZE(X)
  107. add X, X, INCX
  108. addi N, N, -1
  109. cmpwi cr0, N, 0
  110. fabs C1, C1
  111. ble LL(999)
  112. fsmfp C1, C1
  113. fpmr C2, C1
  114. fpmr C3, C1
  115. fpmr C4, C1
  116. cmpwi cr0, INCX, SIZE
  117. bne LL(100)
  118. andi. r0, X, 2 * SIZE - 1
  119. beq LL(05)
  120. LFD C2, 0 * SIZE(X)
  121. add X, X, INCX
  122. addi N, N, -1
  123. cmpwi cr0, N, 0
  124. fabs C2, C2
  125. ble LL(998)
  126. .align 4
  127. LL(05):
  128. sub X, X, INCX2
  129. srawi. r0, N, 4
  130. mtspr CTR, r0
  131. beq- LL(15)
  132. LFPDUX A1, X, INCX2
  133. LFPDUX A2, X, INCX2
  134. LFPDUX A3, X, INCX2
  135. LFPDUX A4, X, INCX2
  136. LFPDUX A5, X, INCX2
  137. fpabs T1, A1
  138. LFPDUX A6, X, INCX2
  139. fpabs T2, A2
  140. LFPDUX A7, X, INCX2
  141. fpabs T3, A3
  142. LFPDUX A8, X, INCX2
  143. fpabs T4, A4
  144. bdz LL(13)
  145. .align 4
  146. LL(12):
  147. fpsub F1, T1, C1
  148. LFPDUX A1, X, INCX2
  149. fpsub F2, T2, C2
  150. LFPDUX A2, X, INCX2
  151. fpsub F3, T3, C3
  152. LFPDUX A3, X, INCX2
  153. fpsub F4, T4, C4
  154. LFPDUX A4, X, INCX2
  155. fpabs T5, A5
  156. fpabs T6, A6
  157. fpabs T7, A7
  158. fpabs T8, A8
  159. fpsel C1, F1, C1, T1
  160. LFPDUX A5, X, INCX2
  161. fpsel C2, F2, C2, T2
  162. LFPDUX A6, X, INCX2
  163. fpsel C3, F3, C3, T3
  164. LFPDUX A7, X, INCX2
  165. fpsel C4, F4, C4, T4
  166. LFPDUX A8, X, INCX2
  167. fpsub F5, T5, C1
  168. fpsub F6, T6, C2
  169. fpsub F7, T7, C3
  170. fpsub F8, T8, C4
  171. fpabs T1, A1
  172. fpabs T2, A2
  173. fpabs T3, A3
  174. fpabs T4, A4
  175. fpsel C1, F5, C1, T5
  176. fpsel C2, F6, C2, T6
  177. fpsel C3, F7, C3, T7
  178. fpsel C4, F8, C4, T8
  179. bdnz LL(12)
  180. .align 4
  181. LL(13):
  182. fpabs T5, A5
  183. fpabs T6, A6
  184. fpabs T7, A7
  185. fpabs T8, A8
  186. fpsub F1, T1, C1
  187. fpsub F2, T2, C2
  188. fpsub F3, T3, C3
  189. fpsub F4, T4, C4
  190. fpsel C1, F1, C1, T1
  191. fpsel C2, F2, C2, T2
  192. fpsel C3, F3, C3, T3
  193. fpsel C4, F4, C4, T4
  194. fpsub F5, T5, C1
  195. fpsub F6, T6, C2
  196. fpsub F7, T7, C3
  197. fpsub F8, T8, C4
  198. fpsel C1, F5, C1, T5
  199. fpsel C2, F6, C2, T6
  200. fpsel C3, F7, C3, T7
  201. fpsel C4, F8, C4, T8
  202. .align 4
  203. LL(15):
  204. andi. r0, N, 15
  205. beq LL(998)
  206. andi. r0, N, 8
  207. beq LL(16)
  208. LFPDUX A1, X, INCX2
  209. LFPDUX A2, X, INCX2
  210. LFPDUX A3, X, INCX2
  211. LFPDUX A4, X, INCX2
  212. fpabs A1, A1
  213. fpabs A2, A2
  214. fpabs A3, A3
  215. fpabs A4, A4
  216. fpsub F1, A1, C1
  217. fpsub F2, A2, C2
  218. fpsub F3, A3, C3
  219. fpsub F4, A4, C4
  220. fpsel C1, F1, C1, A1
  221. fpsel C2, F2, C2, A2
  222. fpsel C3, F3, C3, A3
  223. fpsel C4, F4, C4, A4
  224. .align 4
  225. LL(16):
  226. andi. r0, N, 4
  227. beq LL(17)
  228. LFPDUX A1, X, INCX2
  229. LFPDUX A2, X, INCX2
  230. fpabs A1, A1
  231. fpabs A2, A2
  232. fpsub F1, A1, C1
  233. fpsub F2, A2, C2
  234. fpsel C1, F1, C1, A1
  235. fpsel C2, F2, C2, A2
  236. .align 4
  237. LL(17):
  238. andi. r0, N, 2
  239. beq LL(18)
  240. LFPDUX A1, X, INCX2
  241. fpabs A1, A1
  242. fpsub F1, A1, C1
  243. fpsel C1, F1, C1, A1
  244. .align 4
  245. LL(18):
  246. andi. r0, N, 1
  247. beq LL(998)
  248. LFDUX A1, X, INCX2
  249. fabs A1, A1
  250. fsub F1, A1, C1
  251. fsel C1, F1, C1, A1
  252. b LL(998)
  253. .align 4
  254. LL(100):
  255. sub X, X, INCX
  256. srawi. r0, N, 4
  257. mtspr CTR, r0
  258. beq- LL(105)
  259. LFDUX A1, X, INCX
  260. LFDUX A2, X, INCX
  261. LFDUX A3, X, INCX
  262. LFDUX A4, X, INCX
  263. LFSDUX A1, X, INCX
  264. LFSDUX A2, X, INCX
  265. LFSDUX A3, X, INCX
  266. LFSDUX A4, X, INCX
  267. LFDUX A5, X, INCX
  268. LFDUX A6, X, INCX
  269. LFDUX A7, X, INCX
  270. LFDUX A8, X, INCX
  271. LFSDUX A5, X, INCX
  272. fpabs T1, A1
  273. LFSDUX A6, X, INCX
  274. fpabs T2, A2
  275. LFSDUX A7, X, INCX
  276. fpabs T3, A3
  277. LFSDUX A8, X, INCX
  278. fpabs T4, A4
  279. bdz LL(103)
  280. .align 4
  281. LL(102):
  282. fpsub F1, T1, C1
  283. LFDUX A1, X, INCX
  284. fpsub F2, T2, C2
  285. LFDUX A2, X, INCX
  286. fpsub F3, T3, C3
  287. LFDUX A3, X, INCX
  288. fpsub F4, T4, C4
  289. LFDUX A4, X, INCX
  290. fpabs T5, A5
  291. LFSDUX A1, X, INCX
  292. fpabs T6, A6
  293. LFSDUX A2, X, INCX
  294. fpabs T7, A7
  295. LFSDUX A3, X, INCX
  296. fpabs T8, A8
  297. LFSDUX A4, X, INCX
  298. fpsel C1, F1, C1, T1
  299. LFDUX A5, X, INCX
  300. fpsel C2, F2, C2, T2
  301. LFDUX A6, X, INCX
  302. fpsel C3, F3, C3, T3
  303. LFDUX A7, X, INCX
  304. fpsel C4, F4, C4, T4
  305. LFDUX A8, X, INCX
  306. fpsub F5, T5, C1
  307. LFSDUX A5, X, INCX
  308. fpsub F6, T6, C2
  309. LFSDUX A6, X, INCX
  310. fpsub F7, T7, C3
  311. LFSDUX A7, X, INCX
  312. fpsub F8, T8, C4
  313. LFSDUX A8, X, INCX
  314. fpabs T1, A1
  315. fpabs T2, A2
  316. fpabs T3, A3
  317. fpabs T4, A4
  318. fpsel C1, F5, C1, T5
  319. fpsel C2, F6, C2, T6
  320. fpsel C3, F7, C3, T7
  321. fpsel C4, F8, C4, T8
  322. bdnz LL(102)
  323. .align 4
  324. LL(103):
  325. fpabs T5, A5
  326. fpabs T6, A6
  327. fpabs T7, A7
  328. fpabs T8, A8
  329. fpsub F1, T1, C1
  330. fpsub F2, T2, C2
  331. fpsub F3, T3, C3
  332. fpsub F4, T4, C4
  333. fpsel C1, F1, C1, T1
  334. fpsel C2, F2, C2, T2
  335. fpsel C3, F3, C3, T3
  336. fpsel C4, F4, C4, T4
  337. fpsub F5, T5, C1
  338. fpsub F6, T6, C2
  339. fpsub F7, T7, C3
  340. fpsub F8, T8, C4
  341. fpsel C1, F5, C1, T5
  342. fpsel C2, F6, C2, T6
  343. fpsel C3, F7, C3, T7
  344. fpsel C4, F8, C4, T8
  345. .align 4
  346. LL(105):
  347. andi. r0, N, 15
  348. beq LL(998)
  349. andi. r0, N, 8
  350. beq LL(106)
  351. LFDUX A1, X, INCX
  352. LFDUX A2, X, INCX
  353. LFDUX A3, X, INCX
  354. LFDUX A4, X, INCX
  355. LFSDUX A1, X, INCX
  356. LFSDUX A2, X, INCX
  357. LFSDUX A3, X, INCX
  358. LFSDUX A4, X, INCX
  359. fpabs A1, A1
  360. fpabs A2, A2
  361. fpabs A3, A3
  362. fpabs A4, A4
  363. fpsub F1, A1, C1
  364. fpsub F2, A2, C2
  365. fpsub F3, A3, C3
  366. fpsub F4, A4, C4
  367. fpsel C1, F1, C1, A1
  368. fpsel C2, F2, C2, A2
  369. fpsel C3, F3, C3, A3
  370. fpsel C4, F4, C4, A4
  371. .align 4
  372. LL(106):
  373. andi. r0, N, 4
  374. beq LL(107)
  375. LFDUX A1, X, INCX
  376. LFDUX A2, X, INCX
  377. LFDUX A3, X, INCX
  378. LFDUX A4, X, INCX
  379. fabs A1, A1
  380. fabs A2, A2
  381. fabs A3, A3
  382. fabs A4, A4
  383. fsub F1, A1, C1
  384. fsub F2, A2, C2
  385. fsub F3, A3, C3
  386. fsub F4, A4, C4
  387. fsel C1, F1, C1, A1
  388. fsel C2, F2, C2, A2
  389. fsel C3, F3, C3, A3
  390. fsel C4, F4, C4, A4
  391. .align 4
  392. LL(107):
  393. andi. r0, N, 2
  394. beq LL(108)
  395. LFDUX A1, X, INCX
  396. LFDUX A2, X, INCX
  397. fabs A1, A1
  398. fabs A2, A2
  399. fsub F1, A1, C1
  400. fsub F2, A2, C2
  401. fsel C1, F1, C1, A1
  402. fsel C2, F2, C2, A2
  403. .align 4
  404. LL(108):
  405. andi. r0, N, 1
  406. beq LL(998)
  407. LFDUX A1, X, INCX
  408. fabs A1, A1
  409. fsub F1, A1, C1
  410. fsel C1, F1, C1, A1
  411. .align 4
  412. LL(998):
  413. fpsub F1, C2, C1
  414. fpsub F2, C4, C3
  415. fpsel C1, F1, C1, C2
  416. fpsel C3, F2, C3, C4
  417. fpsub F1, C3, C1
  418. fpsel C1, F1, C1, C3
  419. fsmtp C2, C1
  420. fsub F1, C2, C1
  421. fsel C1, F1, C1, C2
  422. .align 4
  423. LL(999):
  424. li r10, 16
  425. lfpdux f27, SP, r10
  426. lfpdux f26, SP, r10
  427. lfpdux f25, SP, r10
  428. lfpdux f24, SP, r10
  429. lfpdux f23, SP, r10
  430. lfpdux f22, SP, r10
  431. lfpdux f21, SP, r10
  432. lfpdux f20, SP, r10
  433. lfpdux f19, SP, r10
  434. lfpdux f18, SP, r10
  435. lfpdux f17, SP, r10
  436. lfpdux f16, SP, r10
  437. lfpdux f15, SP, r10
  438. lfpdux f14, SP, r10
  439. addi SP, SP, 16
  440. blr
  441. EPILOGUE