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: c5e7339c9e
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c5e7339c9e'
${ noResults }
OpenBLAS/kernel/arm/zdot_vfp.S

310 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. * 2013/11/11 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. * [fp, #-128] - [fp, #-64] is reserved

  47. * for store and restore of floating point

  48. * registers

  49. *******************************************************/

  50. 

  51. #if !defined(__ARM_PCS_VFP)

  52. #define	OLD_RETURN_ADDR	r0

  53. #define	OLD_N		r1

  54. #define	OLD_X		r2

  55. #define	OLD_INC_X	r3

  56. #define	OLD_Y		[fp, #0 ]

  57. #define	OLD_INC_Y	[fp, #4 ]

  58. #define	RETURN_ADDR	r8

  59. #else

  60. #define	OLD_Y		r3

  61. #define	OLD_INC_Y	[fp, #0 ]

  62. #endif

  63. 

  64. #define I	r5

  65. #define Y	r6

  66. #define INC_Y	r7

  67. 

  68. #define X_PRE	512

  69. 

  70. /**************************************************************************************

  71. * Macro definitions

  72. **************************************************************************************/

  73. 

  74. .macro KERNEL_F4

  75. 

  76. 	pld	[ X, #X_PRE  ]

  77. 	pld	[ Y, #X_PRE  ]

  78. 

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

  80. 	vldmia.f64	Y!, { d8 - d9 }

  81. 	fmacd   d0  , d4,  d8

  82. 	fmacd   d1  , d4,  d9

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

  84. 	fmacd   d2  , d5,  d9

  85. 	fmacd   d3  , d5,  d8

  86. 

  87. 	vldmia.f64	Y!, { d10 - d11 }

  88. 	fmacd   d0  , d6,  d10

  89. 	fmacd   d1  , d6,  d11

  90. 	pld	[ X, #X_PRE  ]

  91. 	fmacd   d2  , d7,  d11

  92. 	fmacd   d3  , d7,  d10

  93. 

  94. 	pld	[ Y, #X_PRE  ]

  95. 

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

  97. 	vldmia.f64	Y!, { d8 - d9 }

  98. 	fmacd   d0  , d4,  d8

  99. 	fmacd   d1  , d4,  d9

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

  101. 	fmacd   d2  , d5,  d9

  102. 	fmacd   d3  , d5,  d8

  103. 

  104. 	vldmia.f64	Y!, { d10 - d11 }

  105. 	fmacd   d0  , d6,  d10

  106. 	fmacd   d1  , d6,  d11

  107. 	fmacd   d2  , d7,  d11

  108. 	fmacd   d3  , d7,  d10

  109. 

  110. .endm

  111. 

  112. .macro KERNEL_F1

  113. 

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

  115. 	vldmia.f64	Y!, { d8 - d9 }

  116. 	fmacd   d0  , d4,  d8

  117. 	fmacd   d1  , d4,  d9

  118. 	fmacd   d2  , d5,  d9

  119. 	fmacd   d3  , d5,  d8

  120. 

  121. .endm

  122. 

  123. 

  124. /*************************************************************************************************************************/

  125. 

  126. .macro KERNEL_S4

  127. 

  128. 	nop

  129. 

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

  131. 	vldmia.f64	Y, { d8 - d9 }

  132. 	fmacd   d0  , d4,  d8

  133. 	fmacd   d1  , d4,  d9

  134. 	fmacd   d2  , d5,  d9

  135. 	fmacd   d3  , d5,  d8

  136. 	add	X, X, INC_X

  137. 	add	Y, Y, INC_Y

  138. 

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

  140. 	vldmia.f64	Y, { d8 - d9 }

  141. 	fmacd   d0  , d4,  d8

  142. 	fmacd   d1  , d4,  d9

  143. 	fmacd   d2  , d5,  d9

  144. 	fmacd   d3  , d5,  d8

  145. 	add	X, X, INC_X

  146. 	add	Y, Y, INC_Y

  147. 

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

  149. 	vldmia.f64	Y, { d8 - d9 }

  150. 	fmacd   d0  , d4,  d8

  151. 	fmacd   d1  , d4,  d9

  152. 	fmacd   d2  , d5,  d9

  153. 	fmacd   d3  , d5,  d8

  154. 	add	X, X, INC_X

  155. 	add	Y, Y, INC_Y

  156. 

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

  158. 	vldmia.f64	Y, { d8 - d9 }

  159. 	fmacd   d0  , d4,  d8

  160. 	fmacd   d1  , d4,  d9

  161. 	fmacd   d2  , d5,  d9

  162. 	fmacd   d3  , d5,  d8

  163. 	add	X, X, INC_X

  164. 	add	Y, Y, INC_Y

  165. 

  166. .endm

  167. 

  168. 

  169. .macro KERNEL_S1

  170. 

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

  172. 	vldmia.f64	Y, { d8 - d9 }

  173. 	fmacd   d0  , d4,  d8

  174. 	fmacd   d1  , d4,  d9

  175. 	fmacd   d2  , d5,  d9

  176. 	fmacd   d3  , d5,  d8

  177. 	add	X, X, INC_X

  178. 	add	Y, Y, INC_Y

  179. 

  180. .endm

  181. 

  182. 

  183. 

  184. /**************************************************************************************

  185. * End of macro definitions

  186. **************************************************************************************/

  187. 

  188. 	PROLOGUE

  189. 

  190. 	.align 5

  191. 

  192. 	push	{r4 - r9, fp}

  193. 	add	fp, sp, #28

  194. 	sub	sp, sp, #STACKSIZE				// reserve stack

  195. 

  196. 	sub	r4, fp, #128

  197. 	vstm	r4, { d8 - d15} 				// store floating point registers

  198. 

  199.         movs    r4, #0                                          // clear floating point register

  200.         vmov    s0, r4

  201.         vcvt.f64.f32    d0, s0

  202.         vcvt.f64.f32    d1, s0

  203.         vcvt.f64.f32    d2, s0

  204.         vcvt.f64.f32    d3, s0

  205. 

  206. #if !defined(__ARM_PCS_VFP)

  207. 	mov	RETURN_ADDR, OLD_RETURN_ADDR

  208. 	mov	N, OLD_N

  209. 	mov	X, OLD_X

  210. 	mov	INC_X, OLD_INC_X

  211. 	ldr	Y, OLD_Y

  212. 	ldr	INC_Y, OLD_INC_Y

  213. #else

  214. 	mov	Y, OLD_Y

  215. 	ldr	INC_Y, OLD_INC_Y

  216. #endif

  217. 

  218. 	cmp	N, #0

  219. 	ble	zdot_kernel_L999

  220. 

  221. #	cmp	INC_X, #0

  222. #	beq	zdot_kernel_L999

  223. 

  224. #	cmp	INC_Y, #0

  225. #	beq	zdot_kernel_L999

  226. 

  227. 	cmp	INC_X, #1

  228. 	bne	zdot_kernel_S_BEGIN

  229. 

  230. 	cmp	INC_Y, #1

  231. 	bne	zdot_kernel_S_BEGIN

  232. 

  233. zdot_kernel_F_BEGIN:

  234. 

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

  236. 	ble	zdot_kernel_F1

  237. 

  238. zdot_kernel_F4:

  239. 

  240. 	KERNEL_F4

  241. 

  242. 	subs	I, I, #1

  243. 	bne	zdot_kernel_F4

  244. 

  245. zdot_kernel_F1:

  246. 

  247. 	ands	I, N, #3

  248. 	ble	zdot_kernel_L999

  249. 

  250. zdot_kernel_F10:

  251. 

  252. 	KERNEL_F1

  253. 

  254. 	subs    I, I, #1

  255.         bne     zdot_kernel_F10

  256. 

  257. 	b	zdot_kernel_L999

  258. 

  259. zdot_kernel_S_BEGIN:

  260. 

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

  262. 	lsl	INC_Y, INC_Y, #4				// INC_Y * SIZE * 2

  263. 

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

  265. 	ble	zdot_kernel_S1

  266. 

  267. zdot_kernel_S4:

  268. 

  269. 	KERNEL_S4

  270. 

  271. 	subs	I, I, #1

  272. 	bne	zdot_kernel_S4

  273. 

  274. zdot_kernel_S1:

  275. 

  276. 	ands	I, N, #3

  277. 	ble	zdot_kernel_L999

  278. 

  279. zdot_kernel_S10:

  280. 

  281. 	KERNEL_S1

  282. 

  283. 	subs    I, I, #1

  284.         bne     zdot_kernel_S10

  285. 

  286. 

  287. 

  288. zdot_kernel_L999:

  289. 

  290. 	sub	r3, fp, #128

  291. 	vldm	r3, { d8 - d15}					// restore floating point registers

  292. 

  293. #if !defined(CONJ)

  294. 	vsub.f64	d0 , d0, d2

  295. 	vadd.f64	d1 , d1, d3

  296. #else

  297. 	vadd.f64	d0 , d0, d2

  298. 	vsub.f64	d1 , d1, d3

  299. #endif

  300. #if !defined(__ARM_PCS_VFP)

  301. 	vstm	RETURN_ADDR, {d0 - d1}

  302. #endif

  303. 

  304. 	sub	sp, fp, #28

  305. 	pop	{r4 - r9, fp}

  306. 	bx	lr

  307. 

  308. 	EPILOGUE