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: 670ec6f757
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '670ec6f757'
${ noResults }
OpenBLAS/kernel/arm/asum_vfp.S

490 lines
8.8 kB
Raw Blame History 

  1. /***************************************************************************

  2. Copyright (c) 2013, The OpenBLAS Project

  3. All rights reserved.

  4. Redistribution and use in source and binary forms, with or without

  5. modification, are permitted provided that the following conditions are

  6. met:

  7. 1. Redistributions of source code must retain the above copyright

  8. notice, this list of conditions and the following disclaimer.

  9. 2. Redistributions in binary form must reproduce the above copyright

  10. notice, this list of conditions and the following disclaimer in

  11. the documentation and/or other materials provided with the

  12. distribution.

  13. 3. Neither the name of the OpenBLAS project nor the names of

  14. its contributors may be used to endorse or promote products

  15. derived from this software without specific prior written permission.

  16. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"

  17. AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE

  18. IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE

  19. ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE

  20. LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL

  21. DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR

  22. SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER

  23. CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,

  24. OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE

  25. USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

  26. *****************************************************************************/

  27. 

  28. /**************************************************************************************

  29. * 2013/11/11 Saar

  30. * 	 BLASTEST 		: OK

  31. * 	 CTEST			: OK

  32. * 	 TEST			: OK

  33. *

  34. **************************************************************************************/

  35. 

  36. #define ASSEMBLER

  37. #include "common.h"

  38. 

  39. #define STACKSIZE 256

  40. 

  41. #define	N	r0

  42. #define	X	r1

  43. #define	INC_X	r2

  44. 

  45. 

  46. #define I	r12

  47. 

  48. #define X_PRE	512

  49. 

  50. /**************************************************************************************

  51. * Macro definitions

  52. **************************************************************************************/

  53. 

  54. #if	!defined(COMPLEX)

  55. 

  56. #if	defined(DOUBLE)

  57. 

  58. .macro KERNEL_F4

  59. 

  60. 	pld	[ X, #X_PRE  ]

  61. 	vldmia.f64	X!, { d4 - d5 }

  62. 	vabs.f64   d4,  d4

  63. 	vadd.f64   d0  , d0,  d4

  64. 	vabs.f64   d5,  d5

  65. 	vldmia.f64	X!, { d6 - d7 }

  66. 	vabs.f64   d6,  d6

  67. 	vadd.f64   d1  , d1,  d5

  68. 	vabs.f64   d7,  d7

  69. 	vadd.f64   d0  , d0,  d6

  70. 	vadd.f64   d1  , d1,  d7

  71. 

  72. .endm

  73. 

  74. .macro KERNEL_F1

  75. 

  76. 	vldmia.f64	X!, { d4 }

  77. 	vabs.f64   d4,  d4

  78. 	vadd.f64   d0  , d0,  d4

  79. 

  80. .endm

  81. 

  82. 

  83. .macro KERNEL_S4

  84. 

  85. 	vldmia.f64	X, { d4 }

  86. 	vabs.f64   d4,  d4

  87. 	vadd.f64   d0  , d0,  d4

  88. 	add	X, X, INC_X

  89. 

  90. 	vldmia.f64	X, { d4 }

  91. 	vabs.f64   d4,  d4

  92. 	vadd.f64   d0  , d0,  d4

  93. 	add	X, X, INC_X

  94. 

  95. 	vldmia.f64	X, { d4 }

  96. 	vabs.f64   d4,  d4

  97. 	vadd.f64   d0  , d0,  d4

  98. 	add	X, X, INC_X

  99. 

  100. 	vldmia.f64	X, { d4 }

  101. 	vabs.f64   d4,  d4

  102. 	vadd.f64   d0  , d0,  d4

  103. 	add	X, X, INC_X

  104. 

  105. .endm

  106. 

  107. 

  108. .macro KERNEL_S1

  109. 

  110. 	vldmia.f64	X, { d4 }

  111. 	vabs.f64   d4,  d4

  112. 	vadd.f64   d0  , d0,  d4

  113. 	add	X, X, INC_X

  114. 

  115. .endm

  116. 

  117. #else

  118. 

  119. .macro KERNEL_F4

  120. 

  121. 	vldmia.f32	X!, { s4 - s5 }

  122. 	vabs.f32   s4,  s4

  123. 	vadd.f32   s0  , s0,  s4

  124. 	vabs.f32   s5,  s5

  125. 	vldmia.f32	X!, { s6 - s7 }

  126. 	vabs.f32   s6,  s6

  127. 	vadd.f32   s1  , s1,  s5

  128. 	vabs.f32   s7,  s7

  129. 	vadd.f32   s0  , s0,  s6

  130. 	vadd.f32   s1  , s1,  s7

  131. 

  132. .endm

  133. 

  134. .macro KERNEL_F1

  135. 

  136. 	vldmia.f32	X!, { s4 }

  137. 	vabs.f32   s4,  s4

  138. 	vadd.f32   s0  , s0,  s4

  139. 

  140. .endm

  141. 

  142. 

  143. .macro KERNEL_S4

  144. 

  145. 	vldmia.f32	X, { s4 }

  146. 	vabs.f32   s4,  s4

  147. 	vadd.f32   s0  , s0,  s4

  148. 	add	X, X, INC_X

  149. 

  150. 	vldmia.f32	X, { s4 }

  151. 	vabs.f32   s4,  s4

  152. 	vadd.f32   s0  , s0,  s4

  153. 	add	X, X, INC_X

  154. 

  155. 	vldmia.f32	X, { s4 }

  156. 	vabs.f32   s4,  s4

  157. 	vadd.f32   s0  , s0,  s4

  158. 	add	X, X, INC_X

  159. 

  160. 	vldmia.f32	X, { s4 }

  161. 	vabs.f32   s4,  s4

  162. 	vadd.f32   s0  , s0,  s4

  163. 	add	X, X, INC_X

  164. 

  165. .endm

  166. 

  167. 

  168. .macro KERNEL_S1

  169. 

  170. 	vldmia.f32	X, { s4 }

  171. 	vabs.f32   s4,  s4

  172. 	vadd.f32   s0  , s0,  s4

  173. 	add	X, X, INC_X

  174. 

  175. .endm

  176. 

  177. 

  178. #endif

  179. 

  180. #else

  181. 

  182. #if	defined(DOUBLE)

  183. 

  184. .macro KERNEL_F4

  185. 

  186. 	pld	[ X, #X_PRE  ]

  187. 	vldmia.f64	X!, { d4 - d5 }

  188. 	vabs.f64   d4,  d4

  189. 	vadd.f64   d0  , d0,  d4

  190. 	vabs.f64   d5,  d5

  191. 	vldmia.f64	X!, { d6 - d7 }

  192. 	vabs.f64   d6,  d6

  193. 	vadd.f64   d1  , d1,  d5

  194. 	vabs.f64   d7,  d7

  195. 	vadd.f64   d0  , d0,  d6

  196. 	vadd.f64   d1  , d1,  d7

  197. 

  198. 	pld	[ X, #X_PRE  ]

  199. 	vldmia.f64	X!, { d4 - d5 }

  200. 	vabs.f64   d4,  d4

  201. 	vadd.f64   d0  , d0,  d4

  202. 	vabs.f64   d5,  d5

  203. 	vldmia.f64	X!, { d6 - d7 }

  204. 	vabs.f64   d6,  d6

  205. 	vadd.f64   d1  , d1,  d5

  206. 	vabs.f64   d7,  d7

  207. 	vadd.f64   d0  , d0,  d6

  208. 	vadd.f64   d1  , d1,  d7

  209. 

  210. 

  211. .endm

  212. 

  213. .macro KERNEL_F1

  214. 

  215. 	vldmia.f64	X!, { d4 }

  216. 	vabs.f64   d4,  d4

  217. 	vadd.f64   d0  , d0,  d4

  218. 

  219. 	vldmia.f64	X!, { d4 }

  220. 	vabs.f64   d4,  d4

  221. 	vadd.f64   d0  , d0,  d4

  222. 

  223. 

  224. .endm

  225. 

  226. 

  227. .macro KERNEL_S4

  228. 

  229. 	vldmia.f64	X, { d4 -d5 }

  230. 	vabs.f64   d4,  d4

  231. 	vadd.f64   d0  , d0,  d4

  232. 	vabs.f64   d5,  d5

  233. 	vadd.f64   d0  , d0,  d5

  234. 	add	X, X, INC_X

  235. 

  236. 	vldmia.f64	X, { d4 -d5 }

  237. 	vabs.f64   d4,  d4

  238. 	vadd.f64   d0  , d0,  d4

  239. 	vabs.f64   d5,  d5

  240. 	vadd.f64   d0  , d0,  d5

  241. 	add	X, X, INC_X

  242. 

  243. 	vldmia.f64	X, { d4 -d5 }

  244. 	vabs.f64   d4,  d4

  245. 	vadd.f64   d0  , d0,  d4

  246. 	vabs.f64   d5,  d5

  247. 	vadd.f64   d0  , d0,  d5

  248. 	add	X, X, INC_X

  249. 

  250. 	vldmia.f64	X, { d4 -d5 }

  251. 	vabs.f64   d4,  d4

  252. 	vadd.f64   d0  , d0,  d4

  253. 	vabs.f64   d5,  d5

  254. 	vadd.f64   d0  , d0,  d5

  255. 	add	X, X, INC_X

  256. 

  257. .endm

  258. 

  259. 

  260. .macro KERNEL_S1

  261. 

  262. 	vldmia.f64	X, { d4 -d5 }

  263. 	vabs.f64   d4,  d4

  264. 	vadd.f64   d0  , d0,  d4

  265. 	vabs.f64   d5,  d5

  266. 	vadd.f64   d0  , d0,  d5

  267. 	add	X, X, INC_X

  268. 

  269. .endm

  270. 

  271. #else

  272. 

  273. .macro KERNEL_F4

  274. 

  275. 	pld	[ X, #X_PRE  ]

  276. 	vldmia.f32	X!, { s4 - s5 }

  277. 	vabs.f32   s4,  s4

  278. 	vadd.f32   s0  , s0,  s4

  279. 	vabs.f32   s5,  s5

  280. 	vldmia.f32	X!, { s6 - s7 }

  281. 	vabs.f32   s6,  s6

  282. 	vadd.f32   s1  , s1,  s5

  283. 	vabs.f32   s7,  s7

  284. 	vadd.f32   s0  , s0,  s6

  285. 	vadd.f32   s1  , s1,  s7

  286. 

  287. 	vldmia.f32	X!, { s4 - s5 }

  288. 	vabs.f32   s4,  s4

  289. 	vadd.f32   s0  , s0,  s4

  290. 	vabs.f32   s5,  s5

  291. 	vldmia.f32	X!, { s6 - s7 }

  292. 	vabs.f32   s6,  s6

  293. 	vadd.f32   s1  , s1,  s5

  294. 	vabs.f32   s7,  s7

  295. 	vadd.f32   s0  , s0,  s6

  296. 	vadd.f32   s1  , s1,  s7

  297. 

  298. 

  299. .endm

  300. 

  301. .macro KERNEL_F1

  302. 

  303. 	vldmia.f32	X!, { s4 }

  304. 	vabs.f32   s4,  s4

  305. 	vadd.f32   s0  , s0,  s4

  306. 

  307. 	vldmia.f32	X!, { s4 }

  308. 	vabs.f32   s4,  s4

  309. 	vadd.f32   s0  , s0,  s4

  310. 

  311. .endm

  312. 

  313. 

  314. .macro KERNEL_S4

  315. 

  316. 	vldmia.f32	X, { s4 -s5 }

  317. 	vabs.f32   s4,  s4

  318. 	vadd.f32   s0  , s0,  s4

  319. 	vabs.f32   s5,  s5

  320. 	vadd.f32   s0  , s0,  s5

  321. 	add	X, X, INC_X

  322. 

  323. 	vldmia.f32	X, { s4 -s5 }

  324. 	vabs.f32   s4,  s4

  325. 	vadd.f32   s0  , s0,  s4

  326. 	vabs.f32   s5,  s5

  327. 	vadd.f32   s0  , s0,  s5

  328. 	add	X, X, INC_X

  329. 

  330. 	vldmia.f32	X, { s4 -s5 }

  331. 	vabs.f32   s4,  s4

  332. 	vadd.f32   s0  , s0,  s4

  333. 	vabs.f32   s5,  s5

  334. 	vadd.f32   s0  , s0,  s5

  335. 	add	X, X, INC_X

  336. 

  337. 	vldmia.f32	X, { s4 -s5 }

  338. 	vabs.f32   s4,  s4

  339. 	vadd.f32   s0  , s0,  s4

  340. 	vabs.f32   s5,  s5

  341. 	vadd.f32   s0  , s0,  s5

  342. 	add	X, X, INC_X

  343. 

  344. .endm

  345. 

  346. 

  347. .macro KERNEL_S1

  348. 

  349. 	vldmia.f32	X, { s4 -s5 }

  350. 	vabs.f32   s4,  s4

  351. 	vadd.f32   s0  , s0,  s4

  352. 	vabs.f32   s5,  s5

  353. 	vadd.f32   s0  , s0,  s5

  354. 	add	X, X, INC_X

  355. 

  356. .endm

  357. 

  358. #endif

  359. 

  360. #endif

  361. 

  362. /**************************************************************************************

  363. * End of macro definitions

  364. **************************************************************************************/

  365. 

  366. 	PROLOGUE

  367. 

  368. 	.align 5

  369. 

  370.         movs    r12, #0                                          // clear floating point register

  371.         vmov    s0, r12

  372.         vmov    s1, r12

  373. #if     defined(DOUBLE)

  374.         vcvt.f64.f32    d0, s0

  375.         vcvt.f64.f32    d1, s1

  376. #endif

  377. 

  378. 	cmp	N, #0

  379. 	ble	asum_kernel_L999

  380. 

  381. 	cmp	INC_X, #0

  382. 	beq	asum_kernel_L999

  383. 

  384. 	cmp	INC_X, #1

  385. 	bne	asum_kernel_S_BEGIN

  386. 

  387. 

  388. asum_kernel_F_BEGIN:

  389. 

  390. 	asrs	I, N, #2					// I = N / 4

  391. 	ble	asum_kernel_F1

  392. 

  393. 	.align 5

  394. 

  395. asum_kernel_F4:

  396. 

  397. #if !defined(DOUBLE) && !defined(COMPLEX)

  398. 	pld	[ X, #X_PRE  ]

  399. #endif

  400. 	KERNEL_F4

  401. 

  402. 	subs	I, I, #1

  403. 	ble	asum_kernel_F1

  404. 

  405. 	KERNEL_F4

  406. 

  407. 	subs	I, I, #1

  408. 	bne	asum_kernel_F4

  409. 

  410. asum_kernel_F1:

  411. 

  412. 	ands	I, N, #3

  413. 	ble	asum_kernel_L999

  414. 

  415. asum_kernel_F10:

  416. 

  417. 	KERNEL_F1

  418. 

  419. 	subs    I, I, #1

  420.         bne     asum_kernel_F10

  421. 

  422. 	b	asum_kernel_L999

  423. 

  424. asum_kernel_S_BEGIN:

  425. 

  426. #if defined(COMPLEX)

  427. 

  428. #if defined(DOUBLE)

  429. 	lsl	INC_X, INC_X, #4				// INC_X * SIZE * 2

  430. #else

  431. 	lsl	INC_X, INC_X, #3				// INC_X * SIZE * 2

  432. #endif

  433. 

  434. #else

  435. 

  436. #if defined(DOUBLE)

  437. 	lsl	INC_X, INC_X, #3				// INC_X * SIZE

  438. #else

  439. 	lsl	INC_X, INC_X, #2				// INC_X * SIZE

  440. #endif

  441. 

  442. #endif

  443. 

  444. 	asrs	I, N, #2					// I = N / 4

  445. 	ble	asum_kernel_S1

  446. 

  447. 	.align 5

  448. 

  449. asum_kernel_S4:

  450. 

  451. 	KERNEL_S4

  452. 

  453. 	subs	I, I, #1

  454. 	bne	asum_kernel_S4

  455. 

  456. asum_kernel_S1:

  457. 

  458. 	ands	I, N, #3

  459. 	ble	asum_kernel_L999

  460. 

  461. asum_kernel_S10:

  462. 

  463. 	KERNEL_S1

  464. 

  465. 	subs    I, I, #1

  466.         bne     asum_kernel_S10

  467. 

  468. 

  469. asum_kernel_L999:

  470. 

  471. 

  472. #if defined(DOUBLE)

  473. 	vadd.f64	d0 , d0, d1				// set return value

  474. #else

  475. 	vadd.f32	s0 , s0, s1				// set return value

  476. #endif

  477. 

  478. #if !defined(__ARM_PCS_VFP)

  479. #if !defined(DOUBLE)

  480. 	vmov	r0, s0

  481. #else

  482. 	vmov	r0, r1, d0

  483. #endif

  484. #endif

  485. 

  486. 	bx	lr

  487. 

  488. 	EPILOGUE