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.
Branch: develop
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'develop'
${ noResults }
OpenBLAS/kernel/x86_64/ssymv_L_microk_sandy-2.c

244 lines
9.5 kB
Raw Permalink Blame History 

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

  2. Copyright (c) 2014, 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. #define HAVE_KERNEL_4x4 1

  29. static void ssymv_kernel_4x4( BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2) __attribute__ ((noinline));

  30. 

  31. static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)

  32. {

  33. 

  34.   if ( ( to - from ) & 4 )

  35.   {

  36. 	__asm__  __volatile__

  37. 	(

  38. 	"vzeroupper				     \n\t"

  39. 	"vxorps		%%xmm0 , %%xmm0 , %%xmm0     \n\t"	// temp2[0]

  40. 	"vxorps		%%xmm1 , %%xmm1 , %%xmm1     \n\t"	// temp2[1]

  41. 	"vxorps		%%xmm2 , %%xmm2 , %%xmm2     \n\t"	// temp2[2]

  42. 	"vxorps		%%xmm3 , %%xmm3 , %%xmm3     \n\t"	// temp2[3]

  43. 	"vbroadcastss   (%8),    %%xmm4	             \n\t"	// temp1[0]

  44. 	"vbroadcastss  4(%8),    %%xmm5	             \n\t"	// temp1[1]

  45. 	"vbroadcastss  8(%8),    %%xmm6	             \n\t"	// temp1[1]

  46. 	"vbroadcastss 12(%8),    %%xmm7	             \n\t"	// temp1[1]

  47. 

  48. 	".p2align 4				     \n\t"

  49. 	"1:				     \n\t"

  50. 

  51. 	"vmovups	(%3,%0,4), %%xmm9	           \n\t"  // 2 * y

  52. 	"vmovups	(%2,%0,4), %%xmm8	           \n\t"  // 2 * x

  53. 

  54. 	"vmovups	(%4,%0,4), %%xmm12	           \n\t"  // 2 * a

  55. 	"vmovups	(%5,%0,4), %%xmm13	           \n\t"  // 2 * a

  56. 	"vmovups	(%6,%0,4), %%xmm14	           \n\t"  // 2 * a

  57. 	"vmovups	(%7,%0,4), %%xmm15	           \n\t"  // 2 * a

  58. 

  59.         "vmulps         %%xmm4, %%xmm12 , %%xmm10          \n\t"

  60.         "vmulps         %%xmm8, %%xmm12 , %%xmm11          \n\t"

  61.         "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"

  62.         "vaddps         %%xmm0, %%xmm11 , %%xmm0           \n\t"

  63. 

  64.         "vmulps         %%xmm5, %%xmm13 , %%xmm10          \n\t"

  65.         "vmulps         %%xmm8, %%xmm13 , %%xmm11          \n\t"

  66.         "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"

  67.         "vaddps         %%xmm1, %%xmm11 , %%xmm1           \n\t"

  68. 

  69.         "vmulps         %%xmm6, %%xmm14 , %%xmm10          \n\t"

  70.         "vmulps         %%xmm8, %%xmm14 , %%xmm11          \n\t"

  71.         "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"

  72.         "vaddps         %%xmm2, %%xmm11 , %%xmm2           \n\t"

  73. 

  74.         "vmulps         %%xmm7, %%xmm15 , %%xmm10          \n\t"

  75.         "vmulps         %%xmm8, %%xmm15 , %%xmm11          \n\t"

  76.         "vaddps         %%xmm9, %%xmm10 , %%xmm9           \n\t"

  77.         "vaddps         %%xmm3, %%xmm11 , %%xmm3           \n\t"

  78. 

  79. 	"vmovups	%%xmm9 ,  (%3,%0,4)		   \n\t"

  80. 

  81. 	"addq		$4 , %0	  	 	      \n\t"

  82. 	"cmpq		%0 , %1			      \n\t"

  83. 	"jnz		1b		      \n\t"

  84. 

  85. 	"vmovss		  (%9), %%xmm4		      \n\t"

  86. 	"vmovss		 4(%9), %%xmm5		      \n\t"

  87. 	"vmovss		 8(%9), %%xmm6		      \n\t"

  88. 	"vmovss		12(%9), %%xmm7		      \n\t"

  89. 

  90. 	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"

  91. 	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"

  92. 	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"

  93. 	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"

  94. 

  95. 	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"

  96. 	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"

  97. 	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"

  98. 	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"

  99. 

  100. 	"vaddss		%%xmm4, %%xmm0, %%xmm0  \n\t"

  101. 	"vaddss		%%xmm5, %%xmm1, %%xmm1  \n\t"

  102. 	"vaddss		%%xmm6, %%xmm2, %%xmm2  \n\t"

  103. 	"vaddss		%%xmm7, %%xmm3, %%xmm3  \n\t"

  104. 

  105. 	"vmovss         %%xmm0 ,  (%9)		\n\t"	// save temp2

  106. 	"vmovss         %%xmm1 , 4(%9)		\n\t"	// save temp2

  107. 	"vmovss         %%xmm2 , 8(%9)		\n\t"	// save temp2

  108. 	"vmovss         %%xmm3 ,12(%9)		\n\t"	// save temp2

  109. 	"vzeroupper				     \n\t"

  110. 

  111. 	:

  112.           "+r" (from)	// 0	

  113.         :

  114. 	  "r" (to),  	// 1

  115.           "r" (x),      // 2

  116.           "r" (y),      // 3

  117.           "r" (a[0]),	// 4

  118.           "r" (a[1]),	// 5

  119.           "r" (a[2]),	// 6

  120.           "r" (a[3]),	// 8

  121.           "r" (temp1),  // 8

  122.           "r" (temp2)   // 9

  123. 	: "cc", 

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

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

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

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

  128. 	  "memory"

  129. 	);

  130. 	return;

  131.   }

  132. 

  133. 

  134. 	__asm__  __volatile__

  135. 	(

  136. 	"vzeroupper				     \n\t"

  137. 	"vxorps		%%ymm0 , %%ymm0 , %%ymm0     \n\t"	// temp2[0]

  138. 	"vxorps		%%ymm1 , %%ymm1 , %%ymm1     \n\t"	// temp2[1]

  139. 	"vxorps		%%ymm2 , %%ymm2 , %%ymm2     \n\t"	// temp2[2]

  140. 	"vxorps		%%ymm3 , %%ymm3 , %%ymm3     \n\t"	// temp2[3]

  141. 	"vbroadcastss   (%8),    %%ymm4	             \n\t"	// temp1[0]

  142. 	"vbroadcastss  4(%8),    %%ymm5	             \n\t"	// temp1[1]

  143. 	"vbroadcastss  8(%8),    %%ymm6	             \n\t"	// temp1[1]

  144. 	"vbroadcastss 12(%8),    %%ymm7	             \n\t"	// temp1[1]

  145. 

  146. 	".p2align 4				     \n\t"

  147. 	"1:				     \n\t"

  148. 

  149. 	"vmovups	(%3,%0,4), %%ymm9	           \n\t"  // 2 * y

  150. 	"vmovups	(%2,%0,4), %%ymm8	           \n\t"  // 2 * x

  151. 

  152. 	"vmovups	(%4,%0,4), %%ymm12	           \n\t"  // 2 * a

  153. 	"vmovups	(%5,%0,4), %%ymm13	           \n\t"  // 2 * a

  154. 	"vmovups	(%6,%0,4), %%ymm14	           \n\t"  // 2 * a

  155. 	"vmovups	(%7,%0,4), %%ymm15	           \n\t"  // 2 * a

  156. 

  157.         "vmulps         %%ymm4, %%ymm12 , %%ymm10          \n\t"

  158.         "vmulps         %%ymm8, %%ymm12 , %%ymm11          \n\t"

  159.         "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"

  160.         "vaddps         %%ymm0, %%ymm11 , %%ymm0           \n\t"

  161. 

  162.         "vmulps         %%ymm5, %%ymm13 , %%ymm10          \n\t"

  163.         "vmulps         %%ymm8, %%ymm13 , %%ymm11          \n\t"

  164.         "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"

  165.         "vaddps         %%ymm1, %%ymm11 , %%ymm1           \n\t"

  166. 

  167.         "vmulps         %%ymm6, %%ymm14 , %%ymm10          \n\t"

  168.         "vmulps         %%ymm8, %%ymm14 , %%ymm11          \n\t"

  169.         "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"

  170.         "vaddps         %%ymm2, %%ymm11 , %%ymm2           \n\t"

  171. 

  172.         "vmulps         %%ymm7, %%ymm15 , %%ymm10          \n\t"

  173.         "vmulps         %%ymm8, %%ymm15 , %%ymm11          \n\t"

  174.         "vaddps         %%ymm9, %%ymm10 , %%ymm9           \n\t"

  175.         "vaddps         %%ymm3, %%ymm11 , %%ymm3           \n\t"

  176. 

  177. 	"vmovups	%%ymm9 ,  (%3,%0,4)		   \n\t"

  178. 

  179. 	"addq		$8 , %0	  	 	      \n\t"

  180. 	"cmpq		%0 , %1			      \n\t"

  181. 	"jnz		1b		      \n\t"

  182. 

  183. 	"vmovss		  (%9), %%xmm4		      \n\t"

  184. 	"vmovss		 4(%9), %%xmm5		      \n\t"

  185. 	"vmovss		 8(%9), %%xmm6		      \n\t"

  186. 	"vmovss		12(%9), %%xmm7		      \n\t"

  187. 

  188.         "vextractf128 $0x01, %%ymm0 , %%xmm12         \n\t"

  189.         "vextractf128 $0x01, %%ymm1 , %%xmm13         \n\t"

  190.         "vextractf128 $0x01, %%ymm2 , %%xmm14         \n\t"

  191.         "vextractf128 $0x01, %%ymm3 , %%xmm15         \n\t"

  192. 

  193.         "vaddps         %%xmm0, %%xmm12, %%xmm0       \n\t"

  194.         "vaddps         %%xmm1, %%xmm13, %%xmm1       \n\t"

  195.         "vaddps         %%xmm2, %%xmm14, %%xmm2       \n\t"

  196.         "vaddps         %%xmm3, %%xmm15, %%xmm3       \n\t"

  197. 

  198. 	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"

  199. 	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"

  200. 	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"

  201. 	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"

  202. 

  203. 	"vhaddps        %%xmm0, %%xmm0, %%xmm0  \n\t"

  204. 	"vhaddps        %%xmm1, %%xmm1, %%xmm1  \n\t"

  205. 	"vhaddps        %%xmm2, %%xmm2, %%xmm2  \n\t"

  206. 	"vhaddps        %%xmm3, %%xmm3, %%xmm3  \n\t"

  207. 

  208. 	"vaddss		%%xmm4, %%xmm0, %%xmm0  \n\t"

  209. 	"vaddss		%%xmm5, %%xmm1, %%xmm1  \n\t"

  210. 	"vaddss		%%xmm6, %%xmm2, %%xmm2  \n\t"

  211. 	"vaddss		%%xmm7, %%xmm3, %%xmm3  \n\t"

  212. 

  213. 	"vmovss         %%xmm0 ,  (%9)		\n\t"	// save temp2

  214. 	"vmovss         %%xmm1 , 4(%9)		\n\t"	// save temp2

  215. 	"vmovss         %%xmm2 , 8(%9)		\n\t"	// save temp2

  216. 	"vmovss         %%xmm3 ,12(%9)		\n\t"	// save temp2

  217. 	"vzeroupper				     \n\t"

  218. 

  219. 	:

  220.           "+r" (from)	// 0	

  221.         :

  222. 	  "r" (to),  	// 1

  223.           "r" (x),      // 2

  224.           "r" (y),      // 3

  225.           "r" (a[0]),	// 4

  226.           "r" (a[1]),	// 5

  227.           "r" (a[2]),	// 6

  228.           "r" (a[3]),	// 8

  229.           "r" (temp1),  // 8

  230.           "r" (temp2)   // 9

  231. 	: "cc", 

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

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

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

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

  236. 	  "memory"

  237. 	);

  238. 

  239. 

  240. 

  241. } 

  242.