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: ceaa931e48
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ceaa931e48'
${ noResults }
OpenBLAS/kernel/arm/scal_vfp.S

377 lines
6.7 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. * 2013/11/15 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	OLD_INC_X	[sp, #0 ]

  42. 

  43. 

  44. #define	N	r0

  45. #define	INC_X	r1

  46. #define	X	r3

  47. 

  48. #define I	r12

  49. 

  50. #define X_PRE	512

  51. 

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

  53. * Macro definitions

  54. **************************************************************************************/

  55. 

  56. /*****************************************************************************************/

  57. 

  58. 

  59. 

  60. #if	!defined(COMPLEX)

  61. 

  62. #if	defined(DOUBLE)

  63. 

  64. .macro KERNEL_F4

  65. 

  66. 	pld	[ X, #X_PRE ]

  67. 	fldmiad	X,  { d4 - d7 }

  68. 	vmul.f64    d4, d4, d0

  69. 	vmul.f64    d5, d5, d0

  70. 	vmul.f64    d6, d6, d0

  71. 	fstmiad	X!, { d4 - d5 }

  72. 	vmul.f64    d7, d7, d0

  73. 	fstmiad	X!, { d6 - d7 }

  74. 

  75. .endm

  76. 

  77. 

  78. .macro KERNEL_F1

  79. 

  80. 	fldmiad	X,  { d4 }

  81. 	vmul.f64    d4, d4, d0

  82. 	fstmiad	X!, { d4 }

  83. 

  84. .endm

  85. 

  86. .macro KERNEL_S1

  87. 

  88. 	fldmiad	X,  { d4 }

  89. 	vmul.f64    d4, d4, d0

  90. 	fstmiad	X,  { d4 }

  91. 	add	X, X, INC_X

  92. 

  93. .endm

  94. 

  95. #else

  96. 

  97. .macro KERNEL_F4

  98. 

  99. 	fldmias	X,  { s4 - s7 }

  100. 	vmul.f32    s4, s4, s0

  101. 	vmul.f32    s5, s5, s0

  102. 	vmul.f32    s6, s6, s0

  103. 	fstmias	X!, { s4 - s5 }

  104. 	vmul.f32    s7, s7, s0

  105. 	fstmias	X!, { s6 - s7 }

  106. 

  107. .endm

  108. 

  109. 

  110. .macro KERNEL_F1

  111. 

  112. 	fldmias	X,  { s4 }

  113. 	vmul.f32    s4, s4, s0

  114. 	fstmias	X!, { s4 }

  115. 

  116. .endm

  117. 

  118. .macro KERNEL_S1

  119. 

  120. 	fldmias	X,  { s4 }

  121. 	vmul.f32    s4, s4, s0

  122. 	fstmias	X,  { s4 }

  123. 	add	X, X, INC_X

  124. 

  125. .endm

  126. 

  127. 

  128. 

  129. #endif

  130. 

  131. #else

  132. 

  133. #if	defined(DOUBLE)

  134. 

  135. .macro KERNEL_F4

  136. 

  137. 	pld	[ X, #X_PRE ]

  138. 

  139. 	fldmiad	X,  { d4 - d5 }

  140. 	vmul.f64    d2, d0, d4

  141. 	fnmacd     d2, d1, d5

  142. 	vmul.f64    d3, d0, d5

  143. 	fmacd      d3, d1, d4

  144. 	fstmiad	X!, { d2 - d3 }

  145. 

  146. 	fldmiad	X,  { d4 - d5 }

  147. 	vmul.f64    d2, d0, d4

  148. 	fnmacd     d2, d1, d5

  149. 	vmul.f64    d3, d0, d5

  150. 	fmacd      d3, d1, d4

  151. 	fstmiad	X!, { d2 - d3 }

  152. 

  153. 	pld	[ X, #X_PRE ]

  154. 

  155. 	fldmiad	X,  { d4 - d5 }

  156. 	vmul.f64    d2, d0, d4

  157. 	fnmacd     d2, d1, d5

  158. 	vmul.f64    d3, d0, d5

  159. 	fmacd      d3, d1, d4

  160. 	fstmiad	X!, { d2 - d3 }

  161. 

  162. 	fldmiad	X,  { d4 - d5 }

  163. 	vmul.f64    d2, d0, d4

  164. 	fnmacd     d2, d1, d5

  165. 	vmul.f64    d3, d0, d5

  166. 	fmacd      d3, d1, d4

  167. 	fstmiad	X!, { d2 - d3 }

  168. 

  169. .endm

  170. 

  171. 

  172. .macro KERNEL_F1

  173. 

  174. 	fldmiad	X,  { d4 - d5 }

  175. 	vmul.f64    d2, d0, d4

  176. 	fnmacd     d2, d1, d5

  177. 	vmul.f64    d3, d0, d5

  178. 	fmacd      d3, d1, d4

  179. 	fstmiad	X!, { d2 - d3 }

  180. 

  181. .endm

  182. 

  183. .macro KERNEL_S1

  184. 

  185. 	fldmiad	X,  { d4 - d5 }

  186. 	vmul.f64    d2, d0, d4

  187. 	fnmacd     d2, d1, d5

  188. 	vmul.f64    d3, d0, d5

  189. 	fmacd      d3, d1, d4

  190. 	fstmiad	X, { d2 - d3 }

  191. 	add	X, X, INC_X

  192. 

  193. .endm

  194. 

  195. 

  196. #else

  197. 

  198. .macro KERNEL_F4

  199. 

  200. 	pld	[ X, #X_PRE ]

  201. 

  202. 	fldmias	X,  { s4 - s5 }

  203. 	vmul.f32    s2, s0, s4

  204. 	fnmacs     s2, s1, s5

  205. 	vmul.f32    s3, s0, s5

  206. 	fmacs      s3, s1, s4

  207. 	fstmias	X!, { s2 - s3 }

  208. 

  209. 	fldmias	X,  { s4 - s5 }

  210. 	vmul.f32    s2, s0, s4

  211. 	fnmacs     s2, s1, s5

  212. 	vmul.f32    s3, s0, s5

  213. 	fmacs      s3, s1, s4

  214. 	fstmias	X!, { s2 - s3 }

  215. 

  216. 	fldmias	X,  { s4 - s5 }

  217. 	vmul.f32    s2, s0, s4

  218. 	fnmacs     s2, s1, s5

  219. 	vmul.f32    s3, s0, s5

  220. 	fmacs      s3, s1, s4

  221. 	fstmias	X!, { s2 - s3 }

  222. 

  223. 	fldmias	X,  { s4 - s5 }

  224. 	vmul.f32    s2, s0, s4

  225. 	fnmacs     s2, s1, s5

  226. 	vmul.f32    s3, s0, s5

  227. 	fmacs      s3, s1, s4

  228. 	fstmias	X!, { s2 - s3 }

  229. 

  230. .endm

  231. 

  232. 

  233. .macro KERNEL_F1

  234. 

  235. 	fldmias	X,  { s4 - s5 }

  236. 	vmul.f32    s2, s0, s4

  237. 	fnmacs     s2, s1, s5

  238. 	vmul.f32    s3, s0, s5

  239. 	fmacs      s3, s1, s4

  240. 	fstmias	X!, { s2 - s3 }

  241. 

  242. .endm

  243. 

  244. .macro KERNEL_S1

  245. 

  246. 	fldmias	X,  { s4 - s5 }

  247. 	vmul.f32    s2, s0, s4

  248. 	fnmacs     s2, s1, s5

  249. 	vmul.f32    s3, s0, s5

  250. 	fmacs      s3, s1, s4

  251. 	fstmias	X, { s2 - s3 }

  252. 	add	X, X, INC_X

  253. 

  254. .endm

  255. 

  256. 

  257. 

  258. #endif

  259. 

  260. #endif

  261. 

  262. /**************************************************************************************

  263. * End of macro definitions

  264. **************************************************************************************/

  265. 

  266. 	PROLOGUE

  267. 

  268. 	.align 5

  269. 

  270. 	ldr    INC_X , OLD_INC_X

  271. 

  272. 	cmp	N, #0

  273. 	ble	scal_kernel_L999

  274. 

  275. 	cmp	INC_X, #0

  276. 	ble	scal_kernel_L999

  277. 

  278. 	cmp	INC_X, #1

  279. 	bne	scal_kernel_S_BEGIN

  280. 

  281. 

  282. scal_kernel_F_BEGIN:

  283. 

  284. 

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

  286. 	ble	scal_kernel_F1

  287. 

  288. 	.align 5

  289. 

  290. scal_kernel_F4:

  291. 

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

  293. 	pld	[ X, #X_PRE ]

  294. #endif

  295. 

  296. 	KERNEL_F4

  297. 

  298. 	subs	I, I, #1

  299. 	ble	scal_kernel_F1

  300. 

  301. 	KERNEL_F4

  302. 

  303. 	subs	I, I, #1

  304. 	bne	scal_kernel_F4

  305. 

  306. scal_kernel_F1:

  307. 

  308. 	ands	I, N, #3

  309. 	ble	scal_kernel_L999

  310. 

  311. scal_kernel_F10:

  312. 

  313. 	KERNEL_F1

  314. 

  315. 	subs    I, I, #1

  316.         bne     scal_kernel_F10

  317. 

  318. 	b	scal_kernel_L999

  319. 

  320. scal_kernel_S_BEGIN:

  321. 

  322. #if defined(COMPLEX)

  323. 

  324. #if defined(DOUBLE)

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

  326. #else

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

  328. #endif

  329. 

  330. #else

  331. 

  332. #if defined(DOUBLE)

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

  334. #else

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

  336. #endif

  337. 

  338. #endif

  339. 

  340. 

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

  342. 	ble	scal_kernel_S1

  343. 

  344. 	.align 5

  345. 

  346. scal_kernel_S4:

  347. 

  348. 	KERNEL_S1

  349. 	KERNEL_S1

  350. 	KERNEL_S1

  351. 	KERNEL_S1

  352. 

  353. 	subs	I, I, #1

  354. 	bne	scal_kernel_S4

  355. 

  356. scal_kernel_S1:

  357. 

  358. 	ands	I, N, #3

  359. 	ble	scal_kernel_L999

  360. 

  361. scal_kernel_S10:

  362. 

  363. 	KERNEL_S1

  364. 

  365. 	subs    I, I, #1

  366.         bne     scal_kernel_S10

  367. 

  368. 

  369. scal_kernel_L999:

  370. 

  371. 	mov	r0, #0		// set return value

  372. 

  373. 	bx	lr

  374. 

  375. 	EPILOGUE