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: 549a9f1dbb
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '549a9f1dbb'
${ noResults }
OpenBLAS/kernel/x86_64/cscal_microk_steamroller-2.c

350 lines
11 kB
Raw Blame History 

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

  2. Copyright (c) 2014-2015, 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. #define HAVE_KERNEL_16 1

  30. 

  31. static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));

  32. 

  33. static void cscal_kernel_16( BLASLONG n, FLOAT *alpha, FLOAT *x)

  34. {

  35. 

  36. 

  37. 	__asm__  __volatile__

  38. 	(

  39. 	"vbroadcastss		(%2), %%ymm0		    \n\t"  // da_r	

  40. 	"vbroadcastss          4(%2), %%ymm1		    \n\t"  // da_i 	

  41. 

  42. 	"addq	$128, %1				    \n\t"

  43. 

  44. 	"vmovups	-128(%1), %%ymm4		    \n\t"

  45. 	"vmovups	 -96(%1), %%ymm5		    \n\t"

  46. 	"vmovups	 -64(%1), %%ymm6		    \n\t"

  47. 	"vmovups	 -32(%1), %%ymm7		    \n\t"

  48. 

  49. 	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"

  50. 	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"

  51. 	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"

  52. 	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"

  53. 

  54. 	"subq	        $16, %0			            \n\t"		

  55. 	"jz	2f					    \n\t"

  56. 

  57. 	".align 16				            \n\t"

  58. 	"1:				            	    \n\t"

  59. 

  60. 	"prefetcht0     512(%1)				    \n\t"

  61. 	// ".align 2				            \n\t"

  62. 

  63. 	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1

  64. 	"vmovups	   0(%1), %%ymm4		    \n\t"

  65. 	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"

  66. 	"vmovups	  32(%1), %%ymm5		    \n\t"

  67. 	"prefetcht0     768(%1)				    \n\t"

  68. 	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 

  69. 	"vmovups	  64(%1), %%ymm6		    \n\t"

  70. 	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 

  71. 	"vmovups	  96(%1), %%ymm7		    \n\t"

  72. 

  73. 	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0

  74. 	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 

  75. 	"vaddsubps	%%ymm12 , %%ymm8 , %%ymm8	    \n\t"

  76. 	"vaddsubps	%%ymm13 , %%ymm9 , %%ymm9	    \n\t"

  77. 	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 

  78. 	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 

  79. 	"vaddsubps	%%ymm14 , %%ymm10, %%ymm10	    \n\t"

  80. 	"vaddsubps	%%ymm15 , %%ymm11, %%ymm11	    \n\t"

  81. 

  82. 	"vmovups	%%ymm8 , -128(%1)		    \n\t"

  83. 	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"

  84. 	"vmovups	%%ymm10,  -64(%1)		    \n\t"

  85. 	"vmovups	%%ymm11,  -32(%1)		    \n\t"

  86. 	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"

  87. 	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"

  88. 	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"

  89. 	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"

  90. 

  91. 	"addq		$128 ,%1  	 	            \n\t"

  92. 	"subq	        $16, %0			            \n\t"		

  93. 	"jnz		1b		             	    \n\t"

  94. 

  95. 	"2:				            	    \n\t"

  96. 

  97. 

  98. 	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1

  99. 	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"

  100. 	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 

  101. 	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 

  102. 

  103. 	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0

  104. 	"vaddsubps	%%ymm12 , %%ymm8 , %%ymm8	    \n\t"

  105. 	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 

  106. 	"vaddsubps	%%ymm13 , %%ymm9 , %%ymm9	    \n\t"

  107. 	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 

  108. 	"vaddsubps	%%ymm14 , %%ymm10, %%ymm10	    \n\t"

  109. 	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 

  110. 	"vaddsubps	%%ymm15 , %%ymm11, %%ymm11	    \n\t"

  111. 

  112. 	"vmovups	%%ymm8 , -128(%1)		    \n\t"

  113. 	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"

  114. 	"vmovups	%%ymm10,  -64(%1)		    \n\t"

  115. 	"vmovups	%%ymm11,  -32(%1)		    \n\t"

  116. 

  117. 	"vzeroupper					    \n\t"

  118. 

  119. 	:

  120. 	  "+r" (n),  	// 0

  121.           "+r" (x)      // 1

  122. 	:

  123.           "r" (alpha)   // 2

  124. 	: "cc",

  125. 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 

  126. 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 

  127. 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 

  128. 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",

  129. 	  "memory"

  130. 	);

  131. 

  132. } 

  133. 

  134. static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));

  135. 

  136. static void cscal_kernel_16_zero_r( BLASLONG n, FLOAT *alpha, FLOAT *x)

  137. {

  138. 

  139. 

  140. 	__asm__  __volatile__

  141. 	(

  142. 	"vxorps	           %%ymm0, %%ymm0, %%ymm0	    \n\t"

  143. 	"vbroadcastss          4(%2), %%ymm1		    \n\t"  // da_i 	

  144. 

  145. 	"addq	$128, %1				    \n\t"

  146. 

  147. 	"vmovups	-128(%1), %%ymm4		    \n\t"

  148. 	"vmovups	 -96(%1), %%ymm5		    \n\t"

  149. 	"vmovups	 -64(%1), %%ymm6		    \n\t"

  150. 	"vmovups	 -32(%1), %%ymm7		    \n\t"

  151. 

  152. 	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"

  153. 	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"

  154. 	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"

  155. 	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"

  156. 

  157. 	"subq	        $16, %0			            \n\t"		

  158. 	"jz	2f					    \n\t"

  159. 

  160. 	".align 16				            \n\t"

  161. 	"1:				            	    \n\t"

  162. 

  163. 	//"prefetcht0     128(%1)				    \n\t"

  164. 	// ".align 2				            \n\t"

  165. 

  166. 	"vmovups	   0(%1), %%ymm4		    \n\t"

  167. 	"vmovups	  32(%1), %%ymm5		    \n\t"

  168. 	"vmovups	  64(%1), %%ymm6		    \n\t"

  169. 	"vmovups	  96(%1), %%ymm7		    \n\t"

  170. 

  171. 	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0

  172. 	"vaddsubps	%%ymm12 , %%ymm0 , %%ymm8	    \n\t"

  173. 	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 

  174. 	"vaddsubps	%%ymm13 , %%ymm0 , %%ymm9	    \n\t"

  175. 	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 

  176. 	"vaddsubps	%%ymm14 , %%ymm0 , %%ymm10	    \n\t"

  177. 	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 

  178. 	"vaddsubps	%%ymm15 , %%ymm0 , %%ymm11	    \n\t"

  179. 

  180. 	"vmovups	%%ymm8 , -128(%1)		    \n\t"

  181. 	"vpermilps	$0xb1 , %%ymm4, %%ymm12		    \n\t"

  182. 	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"

  183. 	"vpermilps	$0xb1 , %%ymm5, %%ymm13		    \n\t"

  184. 	"vmovups	%%ymm10,  -64(%1)		    \n\t"

  185. 	"vpermilps	$0xb1 , %%ymm6, %%ymm14 	    \n\t"

  186. 	"vmovups	%%ymm11,  -32(%1)		    \n\t"

  187. 	"vpermilps	$0xb1 , %%ymm7, %%ymm15		    \n\t"

  188. 

  189. 	"addq		$128 ,%1  	 	            \n\t"

  190. 	"subq	        $16, %0			            \n\t"		

  191. 	"jnz		1b		             	    \n\t"

  192. 

  193. 	"2:				            	    \n\t"

  194. 

  195. 	"vmulps		%%ymm1, %%ymm12, %%ymm12	    \n\t" // da_i*x1 , da_i *x0

  196. 	"vaddsubps	%%ymm12 , %%ymm0 , %%ymm8	    \n\t"

  197. 	"vmulps		%%ymm1, %%ymm13, %%ymm13	    \n\t" 

  198. 	"vaddsubps	%%ymm13 , %%ymm0 , %%ymm9	    \n\t"

  199. 	"vmulps		%%ymm1, %%ymm14, %%ymm14	    \n\t" 

  200. 	"vaddsubps	%%ymm14 , %%ymm0 , %%ymm10	    \n\t"

  201. 	"vmulps		%%ymm1, %%ymm15, %%ymm15	    \n\t" 

  202. 	"vaddsubps	%%ymm15 , %%ymm0 , %%ymm11	    \n\t"

  203. 

  204. 	"vmovups	%%ymm8 , -128(%1)		    \n\t"

  205. 	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"

  206. 	"vmovups	%%ymm10,  -64(%1)		    \n\t"

  207. 	"vmovups	%%ymm11,  -32(%1)		    \n\t"

  208. 

  209. 	"vzeroupper					    \n\t"

  210. 

  211. 	: 

  212. 	  "+r" (n),  	// 0

  213.           "+r" (x)      // 1

  214. 	:

  215.           "r" (alpha)   // 2

  216. 	: "cc",

  217. 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 

  218. 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 

  219. 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 

  220. 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",

  221. 	  "memory"

  222. 	);

  223. 

  224. } 

  225. 

  226. 

  227. 

  228. static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));

  229. 

  230. static void cscal_kernel_16_zero_i( BLASLONG n, FLOAT *alpha, FLOAT *x)

  231. {

  232. 

  233. 

  234. 	__asm__  __volatile__

  235. 	(

  236. 	"vbroadcastss		(%2), %%ymm0		    \n\t"  // da_r	

  237. 

  238. 	"addq	$128, %1				    \n\t"

  239. 

  240. 	"vmovups	-128(%1), %%ymm4		    \n\t"

  241. 	"vmovups	 -96(%1), %%ymm5		    \n\t"

  242. 	"vmovups	 -64(%1), %%ymm6		    \n\t"

  243. 	"vmovups	 -32(%1), %%ymm7		    \n\t"

  244. 

  245. 

  246. 	"subq	        $16, %0			            \n\t"		

  247. 	"jz	2f					    \n\t"

  248. 

  249. 	".align 16				            \n\t"

  250. 	"1:				            	    \n\t"

  251. 

  252. 	//"prefetcht0     128(%1)				    \n\t"

  253. 	// ".align 2				            \n\t"

  254. 

  255. 	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1

  256. 	"vmovups	   0(%1), %%ymm4		    \n\t"

  257. 	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"

  258. 	"vmovups	  32(%1), %%ymm5		    \n\t"

  259. 	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 

  260. 	"vmovups	  64(%1), %%ymm6		    \n\t"

  261. 	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 

  262. 	"vmovups	  96(%1), %%ymm7		    \n\t"

  263. 

  264. 	"vmovups	%%ymm8 , -128(%1)		    \n\t"

  265. 	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"

  266. 	"vmovups	%%ymm10,  -64(%1)		    \n\t"

  267. 	"vmovups	%%ymm11,  -32(%1)		    \n\t"

  268. 

  269. 	"addq		$128 ,%1  	 	            \n\t"

  270. 	"subq	        $16, %0			            \n\t"		

  271. 	"jnz		1b		             	    \n\t"

  272. 

  273. 	"2:				            	    \n\t"

  274. 

  275. 

  276. 	"vmulps		%%ymm0, %%ymm4 , %%ymm8		    \n\t" // da_r*x0 , da_r *x1

  277. 	"vmulps		%%ymm0, %%ymm5 , %%ymm9		    \n\t"

  278. 	"vmulps		%%ymm0, %%ymm6 , %%ymm10	    \n\t" 

  279. 	"vmulps		%%ymm0, %%ymm7 , %%ymm11	    \n\t" 

  280. 

  281. 	"vmovups	%%ymm8 , -128(%1)		    \n\t"

  282. 	"vmovups	%%ymm9 ,  -96(%1)		    \n\t"

  283. 	"vmovups	%%ymm10,  -64(%1)		    \n\t"

  284. 	"vmovups	%%ymm11,  -32(%1)		    \n\t"

  285. 

  286. 	"vzeroupper					    \n\t"

  287. 

  288. 	:

  289. 	  "+r" (n),  	// 0

  290.           "+r" (x)      // 1

  291. 	:

  292.           "r" (alpha)   // 2

  293. 	: "cc",

  294. 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 

  295. 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 

  296. 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 

  297. 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",

  298. 	  "memory"

  299. 	);

  300. 

  301. } 

  302. 

  303. 

  304. static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x) __attribute__ ((noinline));

  305. 

  306. static void cscal_kernel_16_zero( BLASLONG n, FLOAT *alpha, FLOAT *x)

  307. {

  308. 

  309. 

  310. 	__asm__  __volatile__

  311. 	(

  312. 	"vxorps	           %%ymm0, %%ymm0, %%ymm0	    \n\t"

  313. 

  314. 	"addq	$128, %1				    \n\t"

  315. 

  316. 	".align 16				            \n\t"

  317. 	"1:				            	    \n\t"

  318. 

  319. 	//"prefetcht0     128(%1)				    \n\t"

  320. 	// ".align 2				            \n\t"

  321. 

  322. 	"vmovups	%%ymm0 , -128(%1)		    \n\t"

  323. 	"vmovups	%%ymm0 ,  -96(%1)		    \n\t"

  324. 	"vmovups	%%ymm0 ,  -64(%1)		    \n\t"

  325. 	"vmovups	%%ymm0 ,  -32(%1)		    \n\t"

  326. 

  327. 	"addq		$128 ,%1  	 	            \n\t"

  328. 	"subq	        $16, %0			            \n\t"		

  329. 	"jnz		1b		             	    \n\t"

  330. 

  331. 	"vzeroupper					    \n\t"

  332. 

  333. 	:

  334. 	  "+r" (n),  	// 0

  335.           "+r" (x)      // 1

  336. 	:

  337.           "r" (alpha)   // 2

  338. 	: "cc",

  339. 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 

  340. 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 

  341. 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 

  342. 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",

  343. 	  "memory"

  344. 	);

  345. 

  346. } 

  347. 

  348.