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: f4b82d7bc4
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'f4b82d7bc4'
${ noResults }
OpenBLAS/kernel/zarch/cgemv_t_4.c

672 lines
22 kB
Raw 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. #include "common.h"

  29. 

  30. #define NBMAX 2048

  31. 

  32. static void cgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)

  33. {

  34.     __asm__ volatile (

  35. 		"vzero  %%v16                      \n\t"

  36. 		"vzero  %%v17                      \n\t"

  37. 		"vzero  %%v18                      \n\t"

  38. 		"vzero  %%v19                      \n\t"

  39.         "xgr   %%r1,%%r1                   \n\t"

  40.         "srlg  %%r0,%0,1                   \n\t"

  41.         "0:                                \n\t"

  42.         "pfd 1,1024(%%r1,%1)               \n\t"

  43.         "pfd 1,1024(%%r1,%2)               \n\t"

  44.         "pfd 1,1024(%%r1,%3)               \n\t"

  45.         "pfd 1,1024(%%r1,%4)               \n\t"

  46. 		"pfd 1,1024(%%r1,%5)               \n\t"

  47. 

  48. 		"vl     %%v20,0(%%r1,%5)           \n\t"

  49. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  50.         "vlef   %%v21,4(%%r1,%5),0         \n\t"

  51. 		"vlef   %%v21,12(%%r1,%5),2        \n\t"

  52.         "vflcsb %%v21,%%v21                \n\t"

  53.         "vlef   %%v21,0(%%r1,%5),1         \n\t"

  54. 		"vlef   %%v21,8(%%r1,%5),3         \n\t"

  55. #else

  56.         "vlef   %%v21,0(%%r1,%5),1         \n\t"

  57. 		"vlef   %%v21,8(%%r1,%5),3         \n\t"

  58.         "vflcsb %%v21,%%v21                \n\t"

  59.         "vlef   %%v21,4(%%r1,%5),0         \n\t"

  60. 		"vlef   %%v21,12(%%r1,%5),2        \n\t"

  61. #endif

  62. 

  63. 		"vlef   %%v22,0(%%r1,%1),0         \n\t"

  64. 		"vlef   %%v22,0(%%r1,%1),1         \n\t"

  65. 		"vlef   %%v22,8(%%r1,%1),2         \n\t"

  66. 		"vlef   %%v22,8(%%r1,%1),3         \n\t"

  67. 		"vlef   %%v23,4(%%r1,%1),0         \n\t"

  68. 		"vlef   %%v23,4(%%r1,%1),1         \n\t"

  69. 		"vlef   %%v23,12(%%r1,%1),2        \n\t"

  70. 		"vlef   %%v23,12(%%r1,%1),3        \n\t"

  71. 		"vlef   %%v24,0(%%r1,%2),0         \n\t"

  72. 		"vlef   %%v24,0(%%r1,%2),1         \n\t"

  73. 		"vlef   %%v24,8(%%r1,%2),2         \n\t"

  74. 		"vlef   %%v24,8(%%r1,%2),3         \n\t"

  75. 		"vlef   %%v25,4(%%r1,%2),0         \n\t"

  76. 		"vlef   %%v25,4(%%r1,%2),1         \n\t"

  77. 		"vlef   %%v25,12(%%r1,%2),2        \n\t"

  78. 		"vlef   %%v25,12(%%r1,%2),3        \n\t"

  79. 

  80.         "vfmasb   %%v16,%%v22,%%v20,%%v16  \n\t"

  81.         "vfmasb   %%v16,%%v23,%%v21,%%v16  \n\t"

  82.         "vfmasb   %%v17,%%v24,%%v20,%%v17  \n\t"

  83.         "vfmasb   %%v17,%%v25,%%v21,%%v17  \n\t"

  84. 

  85. 		"vlef   %%v26,0(%%r1,%3),0         \n\t"

  86. 		"vlef   %%v26,0(%%r1,%3),1         \n\t"

  87. 		"vlef   %%v26,8(%%r1,%3),2         \n\t"

  88. 		"vlef   %%v26,8(%%r1,%3),3         \n\t"

  89. 		"vlef   %%v27,4(%%r1,%3),0         \n\t"

  90. 		"vlef   %%v27,4(%%r1,%3),1         \n\t"

  91. 		"vlef   %%v27,12(%%r1,%3),2        \n\t"

  92. 		"vlef   %%v27,12(%%r1,%3),3        \n\t"

  93. 		"vlef   %%v28,0(%%r1,%4),0         \n\t"

  94. 		"vlef   %%v28,0(%%r1,%4),1         \n\t"

  95. 		"vlef   %%v28,8(%%r1,%4),2         \n\t"

  96. 		"vlef   %%v28,8(%%r1,%4),3         \n\t"

  97. 		"vlef   %%v29,4(%%r1,%4),0         \n\t"

  98. 		"vlef   %%v29,4(%%r1,%4),1         \n\t"

  99. 		"vlef   %%v29,12(%%r1,%4),2        \n\t"

  100. 		"vlef   %%v29,12(%%r1,%4),3        \n\t"

  101.         

  102.         "vfmasb   %%v18,%%v26,%%v20,%%v18  \n\t"

  103.         "vfmasb   %%v18,%%v27,%%v21,%%v18  \n\t"

  104.         "vfmasb   %%v19,%%v28,%%v20,%%v19  \n\t"

  105.         "vfmasb   %%v19,%%v29,%%v21,%%v19  \n\t"

  106. 

  107.         "agfi   %%r1,16                    \n\t"

  108.         "brctg  %%r0,0b                    \n\t"

  109. 

  110. 		"vrepg  %%v20,%%v16,1              \n\t"

  111. 		"vrepg  %%v21,%%v17,1              \n\t"

  112. 		"vrepg  %%v22,%%v18,1              \n\t"

  113. 		"vrepg  %%v23,%%v19,1              \n\t"

  114. 		"vfasb  %%v16,%%v16,%%v20          \n\t"

  115. 		"vfasb  %%v17,%%v17,%%v21          \n\t"

  116. 		"vfasb  %%v18,%%v18,%%v22          \n\t"

  117. 		"vfasb  %%v19,%%v19,%%v23          \n\t"

  118. 		"vmrhg  %%v16,%%v16,%%v17          \n\t"

  119. 		"vmrhg  %%v17,%%v18,%%v19          \n\t"

  120. 		"verllg %%v18,%%v16,32             \n\t"

  121.         "verllg %%v19,%%v17,32             \n\t"

  122. #if !defined(XCONJ)

  123. 		"vlrepf %%v20,0(%7)                \n\t"

  124. 		"vlef   %%v21,4(%7),0              \n\t"

  125. 		"vlef   %%v21,4(%7),2              \n\t"

  126.         "vflcsb %%v21,%%v21                \n\t"

  127.         "vlef   %%v21,4(%7),1              \n\t"

  128. 		"vlef   %%v21,4(%7),3              \n\t"

  129. #else

  130. 		"vlef   %%v20,0(%7),1              \n\t"

  131. 		"vlef   %%v20,0(%7),3              \n\t"

  132.         "vflcsb %%v20,%%v20                \n\t"

  133.         "vlef   %%v20,0(%7),0              \n\t"

  134. 		"vlef   %%v20,0(%7),2              \n\t"

  135. 		"vlrepf %%v21,4(%7)                \n\t"

  136. #endif

  137. 		"vl  %%v22,0(%6)                   \n\t"

  138. 		"vl  %%v23,16(%6)                  \n\t"

  139. 		"vfmasb   %%v22,%%v16,%%v20,%%v22  \n\t"

  140.         "vfmasb   %%v22,%%v18,%%v21,%%v22  \n\t"

  141. 		"vfmasb   %%v23,%%v17,%%v20,%%v23  \n\t"

  142.         "vfmasb   %%v23,%%v19,%%v21,%%v23  \n\t"

  143. 		"vst  %%v22,0(%6)                  \n\t"

  144. 		"vst  %%v23,16(%6)                     "

  145.         :

  146.         :"r"(n),"ZR"((const FLOAT (*)[n * 2])ap[0]),"ZR"((const FLOAT (*)[n * 2])ap[1]),"ZR"((const FLOAT (*)[n * 2])ap[2]),"ZR"((const FLOAT (*)[n * 2])ap[3]),"ZR"((const FLOAT (*)[n * 2])x),"ZQ"((FLOAT (*)[8])y),"ZQ"((const FLOAT (*)[2])alpha)

  147.         :"memory","cc","r0","r1","v16","v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29"

  148.     );

  149. }

  150. 

  151. static void cgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)

  152. {

  153.     __asm__ volatile (

  154. 		"vzero  %%v16                      \n\t"

  155. 		"vzero  %%v17                      \n\t"

  156.         "xgr   %%r1,%%r1                   \n\t"

  157.         "srlg  %%r0,%0,1                   \n\t"

  158.         "0:                                \n\t"

  159.         "pfd 1,1024(%%r1,%1)               \n\t"

  160.         "pfd 1,1024(%%r1,%2)               \n\t"

  161.         "pfd 1,1024(%%r1,%3)               \n\t"

  162. 

  163. 		"vl     %%v18,0(%%r1,%3)           \n\t"

  164. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  165.         "vlef   %%v19,4(%%r1,%3),0         \n\t"

  166. 		"vlef   %%v19,12(%%r1,%3),2        \n\t"

  167.         "vflcsb %%v19,%%v19                \n\t"

  168.         "vlef   %%v19,0(%%r1,%3),1         \n\t"

  169. 		"vlef   %%v19,8(%%r1,%3),3         \n\t"

  170. #else

  171.         "vlef   %%v19,0(%%r1,%3),1         \n\t"

  172. 		"vlef   %%v19,8(%%r1,%3),3         \n\t"

  173.         "vflcsb %%v19,%%v19                \n\t"

  174.         "vlef   %%v19,4(%%r1,%3),0         \n\t"

  175. 		"vlef   %%v19,12(%%r1,%3),2        \n\t"

  176. #endif

  177. 

  178. 		"vlef   %%v20,0(%%r1,%1),0         \n\t"

  179. 		"vlef   %%v20,0(%%r1,%1),1         \n\t"

  180. 		"vlef   %%v20,8(%%r1,%1),2         \n\t"

  181. 		"vlef   %%v20,8(%%r1,%1),3         \n\t"

  182. 		"vlef   %%v21,4(%%r1,%1),0         \n\t"

  183. 		"vlef   %%v21,4(%%r1,%1),1         \n\t"

  184. 		"vlef   %%v21,12(%%r1,%1),2        \n\t"

  185. 		"vlef   %%v21,12(%%r1,%1),3        \n\t"

  186. 		"vlef   %%v22,0(%%r1,%2),0         \n\t"

  187. 		"vlef   %%v22,0(%%r1,%2),1         \n\t"

  188. 		"vlef   %%v22,8(%%r1,%2),2         \n\t"

  189. 		"vlef   %%v22,8(%%r1,%2),3         \n\t"

  190. 		"vlef   %%v23,4(%%r1,%2),0         \n\t"

  191. 		"vlef   %%v23,4(%%r1,%2),1         \n\t"

  192. 		"vlef   %%v23,12(%%r1,%2),2        \n\t"

  193. 		"vlef   %%v23,12(%%r1,%2),3        \n\t"

  194. 

  195.         "vfmasb   %%v16,%%v20,%%v18,%%v16  \n\t"

  196.         "vfmasb   %%v16,%%v21,%%v19,%%v16  \n\t"

  197.         "vfmasb   %%v17,%%v22,%%v18,%%v17  \n\t"

  198.         "vfmasb   %%v17,%%v23,%%v19,%%v17  \n\t"

  199. 

  200.         "agfi   %%r1,16                    \n\t"

  201.         "brctg  %%r0,0b                    \n\t"

  202. 

  203. 		"vrepg  %%v18,%%v16,1              \n\t"

  204. 		"vrepg  %%v19,%%v17,1              \n\t"

  205. 		"vfasb  %%v16,%%v16,%%v18          \n\t"

  206. 		"vfasb  %%v17,%%v17,%%v19          \n\t"

  207. 		"vmrhg  %%v16,%%v16,%%v17          \n\t"

  208. 		"verllg %%v17,%%v16,32             \n\t"

  209. #if !defined(XCONJ)

  210. 		"vlrepf %%v18,0(%5)                \n\t"

  211. 		"vlef   %%v19,4(%5),0              \n\t"

  212. 		"vlef   %%v19,4(%5),2              \n\t"

  213.         "vflcsb %%v19,%%v19                \n\t"

  214.         "vlef   %%v19,4(%5),1              \n\t"

  215. 		"vlef   %%v19,4(%5),3              \n\t"

  216. #else

  217. 		"vlef   %%v18,0(%5),1              \n\t"

  218. 		"vlef   %%v18,0(%5),3              \n\t"

  219.         "vflcsb %%v18,%%v18                \n\t"

  220.         "vlef   %%v18,0(%5),0              \n\t"

  221. 		"vlef   %%v18,0(%5),2              \n\t"

  222. 		"vlrepf %%v19,4(%5)                \n\t"

  223. #endif

  224. 		"vl  %%v20,0(%4)                   \n\t"

  225. 		"vfmasb   %%v20,%%v16,%%v18,%%v20  \n\t"

  226.         "vfmasb   %%v20,%%v17,%%v19,%%v20  \n\t"

  227. 		"vst  %%v20,0(%4)                      "

  228.         :

  229.         :"r"(n),"ZR"((const FLOAT (*)[n * 2])ap[0]),"ZR"((const FLOAT (*)[n * 2])ap[1]),"ZR"((const FLOAT (*)[n * 2])x),"ZQ"((FLOAT (*)[4])y),"ZQ"((const FLOAT (*)[2])alpha)

  230.         :"memory","cc","r0","r1","v16","v17","v18","v19","v20","v21","v22","v23"

  231.     );

  232. }

  233. 

  234. static void cgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *alpha)

  235. {

  236.     __asm__ volatile (

  237. 		"vzero  %%v16                      \n\t"

  238.         "xgr   %%r1,%%r1                   \n\t"

  239.         "srlg  %%r0,%0,1                   \n\t"

  240.         "0:                                \n\t"

  241.         "pfd 1,1024(%%r1,%1)               \n\t"

  242.         "pfd 1,1024(%%r1,%2)               \n\t"

  243. 

  244. 		"vl     %%v17,0(%%r1,%2)           \n\t"

  245. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  246.         "vlef   %%v18,4(%%r1,%2),0         \n\t"

  247. 		"vlef   %%v18,12(%%r1,%2),2        \n\t"

  248.         "vflcsb %%v18,%%v18                \n\t"

  249.         "vlef   %%v18,0(%%r1,%2),1         \n\t"

  250. 		"vlef   %%v18,8(%%r1,%2),3         \n\t"

  251. #else

  252.         "vlef   %%v18,0(%%r1,%2),1         \n\t"

  253. 		"vlef   %%v18,8(%%r1,%2),3         \n\t"

  254.         "vflcsb %%v18,%%v18                \n\t"

  255.         "vlef   %%v18,4(%%r1,%2),0         \n\t"

  256. 		"vlef   %%v18,12(%%r1,%2),2        \n\t"

  257. #endif

  258. 

  259. 		"vlef   %%v19,0(%%r1,%1),0         \n\t"

  260. 		"vlef   %%v19,0(%%r1,%1),1         \n\t"

  261. 		"vlef   %%v19,8(%%r1,%1),2         \n\t"

  262. 		"vlef   %%v19,8(%%r1,%1),3         \n\t"

  263. 		"vlef   %%v20,4(%%r1,%1),0         \n\t"

  264. 		"vlef   %%v20,4(%%r1,%1),1         \n\t"

  265. 		"vlef   %%v20,12(%%r1,%1),2        \n\t"

  266. 		"vlef   %%v20,12(%%r1,%1),3        \n\t"

  267. 

  268.         "vfmasb   %%v16,%%v19,%%v17,%%v16  \n\t"

  269.         "vfmasb   %%v16,%%v20,%%v18,%%v16  \n\t"

  270. 

  271.         "agfi   %%r1,16                    \n\t"

  272.         "brctg  %%r0,0b                    \n\t"

  273. 

  274. 		"vrepg  %%v17,%%v16,1              \n\t"

  275. 		"vfasb  %%v16,%%v16,%%v17          \n\t"

  276. 		"verllg %%v17,%%v16,32             \n\t"

  277. #if !defined(XCONJ)

  278. 		"vlrepf %%v18,0(%4)                \n\t"

  279. 		"vlef   %%v19,4(%4),0              \n\t"

  280.         "vflcsb %%v19,%%v19                \n\t"

  281.         "vlef   %%v19,4(%4),1              \n\t"

  282. #else

  283. 		"vlef   %%v18,0(%4),1              \n\t"

  284.         "vflcsb %%v18,%%v18                \n\t"

  285.         "vlef   %%v18,0(%4),0              \n\t"

  286. 		"vlrepf %%v19,4(%4)                \n\t"

  287. #endif

  288. 		"vleg     %%v20,0(%3),0            \n\t"

  289. 		"vfmasb   %%v20,%%v16,%%v18,%%v20  \n\t"

  290.         "vfmasb   %%v20,%%v17,%%v19,%%v20  \n\t"

  291. 		"vsteg    %%v20,0(%3),0                "

  292.         :

  293.         :"r"(n),"ZR"((const FLOAT (*)[n * 2])ap),"ZR"((const FLOAT (*)[n * 2])x),"ZQ"((FLOAT (*)[2])y),"ZQ"((const FLOAT (*)[2])alpha)

  294.         :"memory","cc","r0","r1","v16","v17","v18","v19","v20","v21","v22","v23"

  295.     );

  296. }

  297. 

  298. static void copy_x(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_src)

  299. {

  300.         BLASLONG i;

  301.         for ( i=0; i<n; i++ )

  302.         {

  303.                 *dest     = *src;

  304.                 *(dest+1) = *(src+1);

  305.                 dest+=2;

  306.                 src += inc_src;

  307.         }

  308. }

  309. 

  310. int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT alpha_r, FLOAT alpha_i, FLOAT *a, BLASLONG lda, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y, FLOAT *buffer)

  311. {

  312. 	BLASLONG i;

  313. 	BLASLONG j;

  314. 	FLOAT *a_ptr;

  315. 	FLOAT *x_ptr;

  316. 	FLOAT *y_ptr;

  317. 	FLOAT *ap[8];

  318. 	BLASLONG n1;

  319. 	BLASLONG m1;

  320. 	BLASLONG m2;

  321. 	BLASLONG m3;

  322. 	BLASLONG n2;

  323. 	BLASLONG lda4;

  324. 	FLOAT ybuffer[8],*xbuffer;

  325. 	FLOAT alpha[2];

  326. 

  327.         if ( m < 1 ) return(0);

  328.         if ( n < 1 ) return(0);

  329. 

  330.         inc_x <<= 1;

  331.         inc_y <<= 1;

  332.         lda   <<= 1;

  333. 	lda4    = lda << 2;

  334. 

  335. 	xbuffer = buffer;

  336. 	

  337. 	n1 = n  >> 2 ;

  338. 	n2 = n  &  3 ;

  339. 	

  340. 	m3 = m & 3 ;

  341. 	m1 = m - m3;

  342. 	m2 = (m & (NBMAX-1)) - m3 ;

  343. 	

  344. 	alpha[0] = alpha_r;

  345. 	alpha[1] = alpha_i;

  346. 

  347. 	BLASLONG NB = NBMAX;

  348. 

  349. 	while ( NB == NBMAX )

  350. 	{

  351. 		

  352. 		m1 -= NB;

  353. 		if ( m1 < 0)

  354. 		{

  355. 			if ( m2 == 0 ) break;	

  356. 			NB = m2;

  357. 		}

  358. 		

  359. 		y_ptr = y;

  360. 		a_ptr = a;

  361. 		x_ptr = x;

  362. 		ap[0] = a_ptr;

  363. 		ap[1] = a_ptr + lda;

  364. 		ap[2] = ap[1] + lda;

  365. 		ap[3] = ap[2] + lda;

  366. 		if ( inc_x != 2 )

  367. 			copy_x(NB,x_ptr,xbuffer,inc_x);

  368. 		else

  369. 			xbuffer = x_ptr;

  370. 		

  371. 		if ( inc_y == 2 )

  372. 		{

  373. 

  374. 			for( i = 0; i < n1 ; i++)

  375. 			{

  376. 				cgemv_kernel_4x4(NB,ap,xbuffer,y_ptr,alpha);

  377. 				ap[0] += lda4;

  378. 				ap[1] += lda4;

  379. 				ap[2] += lda4;

  380. 				ap[3] += lda4;

  381. 				a_ptr += lda4;

  382. 				y_ptr += 8;

  383. 				

  384. 			}

  385. 

  386. 			if ( n2 & 2 )

  387. 			{

  388. 				cgemv_kernel_4x2(NB,ap,xbuffer,y_ptr,alpha);

  389. 				a_ptr += lda * 2;

  390. 				y_ptr += 4;

  391. 

  392. 			}

  393. 

  394. 			if ( n2 & 1 )

  395. 			{

  396. 				cgemv_kernel_4x1(NB,a_ptr,xbuffer,y_ptr,alpha);

  397. 				/* a_ptr += lda;

  398. 				y_ptr += 2; */

  399. 

  400. 			}

  401. 

  402. 		}

  403. 		else

  404. 		{

  405. 

  406. 			for( i = 0; i < n1 ; i++)

  407. 			{

  408. 				memset(ybuffer,0,sizeof(ybuffer));

  409. 				cgemv_kernel_4x4(NB,ap,xbuffer,ybuffer,alpha);

  410. 				ap[0] += lda4;

  411. 				ap[1] += lda4;

  412. 				ap[2] += lda4;

  413. 				ap[3] += lda4;

  414. 				a_ptr += lda4;

  415. 

  416. 				y_ptr[0] += ybuffer[0];

  417. 				y_ptr[1] += ybuffer[1];

  418. 				y_ptr  += inc_y;

  419. 				y_ptr[0] += ybuffer[2];

  420. 				y_ptr[1] += ybuffer[3];

  421. 				y_ptr  += inc_y;

  422. 				y_ptr[0] += ybuffer[4];

  423. 				y_ptr[1] += ybuffer[5];

  424. 				y_ptr  += inc_y;

  425. 				y_ptr[0] += ybuffer[6];

  426. 				y_ptr[1] += ybuffer[7];

  427. 				y_ptr  += inc_y;

  428. 

  429. 			}

  430. 

  431. 			for( i = 0; i < n2 ; i++)

  432. 			{

  433. 				memset(ybuffer,0,sizeof(ybuffer));

  434. 				cgemv_kernel_4x1(NB,a_ptr,xbuffer,ybuffer,alpha);

  435. 				a_ptr += lda;

  436. 				y_ptr[0] += ybuffer[0];

  437. 				y_ptr[1] += ybuffer[1];

  438. 				y_ptr  += inc_y;

  439. 

  440. 			}

  441. 

  442. 		}

  443. 		a += 2 * NB;

  444. 		x += NB * inc_x;	

  445. 	}

  446. 

  447. 

  448. 

  449. 	if ( m3 == 0 ) return(0);

  450. 

  451.         x_ptr = x;

  452.         j=0;

  453.         a_ptr = a;

  454.         y_ptr = y;

  455. 

  456. 	if ( m3 == 3 )

  457. 	{

  458. 

  459.                 FLOAT temp_r ;

  460.                 FLOAT temp_i ;

  461. 		FLOAT x0 = x_ptr[0];

  462. 		FLOAT x1 = x_ptr[1];

  463. 		x_ptr += inc_x;

  464. 		FLOAT x2 = x_ptr[0];

  465. 		FLOAT x3 = x_ptr[1];

  466. 		x_ptr += inc_x;

  467. 		FLOAT x4 = x_ptr[0];

  468. 		FLOAT x5 = x_ptr[1];

  469. 	        while ( j < n)

  470.         	{

  471. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  472.                        	temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 

  473.                        	temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 

  474.                        	temp_r += a_ptr[2] * x2 - a_ptr[3] * x3; 

  475.                        	temp_i += a_ptr[2] * x3 + a_ptr[3] * x2; 

  476.                        	temp_r += a_ptr[4] * x4 - a_ptr[5] * x5;

  477.                        	temp_i += a_ptr[4] * x5 + a_ptr[5] * x4;

  478. #else

  479. 

  480.                        	temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 

  481.                        	temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 

  482.                        	temp_r += a_ptr[2] * x2 + a_ptr[3] * x3; 

  483.                        	temp_i += a_ptr[2] * x3 - a_ptr[3] * x2; 

  484.                        	temp_r += a_ptr[4] * x4 + a_ptr[5] * x5;

  485.                        	temp_i += a_ptr[4] * x5 - a_ptr[5] * x4;

  486. #endif

  487. 

  488. #if !defined(XCONJ) 

  489.                 	y_ptr[0] += alpha_r * temp_r - alpha_i * temp_i;

  490.                 	y_ptr[1] += alpha_r * temp_i + alpha_i * temp_r;

  491. #else

  492.                 	y_ptr[0] += alpha_r * temp_r + alpha_i * temp_i;

  493.                 	y_ptr[1] -= alpha_r * temp_i - alpha_i * temp_r;

  494. #endif

  495. 

  496.                 	a_ptr += lda;

  497.                 	y_ptr += inc_y;

  498.                 	j++;

  499.         	}

  500.         	return(0);

  501. 	}

  502. 

  503. 

  504. 	if ( m3 == 2 )

  505. 	{

  506. 

  507.                 FLOAT temp_r ;

  508.                 FLOAT temp_i ;

  509.                 FLOAT temp_r1 ;

  510.                 FLOAT temp_i1 ;

  511. 		FLOAT x0 = x_ptr[0];

  512. 		FLOAT x1 = x_ptr[1];

  513. 		x_ptr += inc_x;

  514. 		FLOAT x2 = x_ptr[0];

  515. 		FLOAT x3 = x_ptr[1];

  516. 		FLOAT ar = alpha[0];

  517. 		FLOAT ai = alpha[1];

  518. 

  519. 	        while ( j < ( n & -2 ))

  520.         	{

  521. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  522.                        	temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 

  523.                        	temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 

  524.                        	temp_r += a_ptr[2] * x2 - a_ptr[3] * x3; 

  525.                        	temp_i += a_ptr[2] * x3 + a_ptr[3] * x2; 

  526.                 	a_ptr += lda;

  527.                        	temp_r1  = a_ptr[0] * x0 - a_ptr[1] * x1; 

  528.                        	temp_i1  = a_ptr[0] * x1 + a_ptr[1] * x0; 

  529.                        	temp_r1 += a_ptr[2] * x2 - a_ptr[3] * x3; 

  530.                        	temp_i1 += a_ptr[2] * x3 + a_ptr[3] * x2; 

  531. #else

  532. 

  533.                        	temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 

  534.                        	temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 

  535.                        	temp_r += a_ptr[2] * x2 + a_ptr[3] * x3; 

  536.                        	temp_i += a_ptr[2] * x3 - a_ptr[3] * x2; 

  537.                 	a_ptr += lda;

  538.                        	temp_r1  = a_ptr[0] * x0 + a_ptr[1] * x1; 

  539.                        	temp_i1  = a_ptr[0] * x1 - a_ptr[1] * x0; 

  540.                        	temp_r1 += a_ptr[2] * x2 + a_ptr[3] * x3; 

  541.                        	temp_i1 += a_ptr[2] * x3 - a_ptr[3] * x2; 

  542. #endif

  543. 

  544. #if !defined(XCONJ) 

  545.                 	y_ptr[0] += ar * temp_r - ai * temp_i;

  546.                 	y_ptr[1] += ar * temp_i + ai * temp_r;

  547.                 	y_ptr += inc_y;

  548.                 	y_ptr[0] += ar * temp_r1 - ai * temp_i1;

  549.                 	y_ptr[1] += ar * temp_i1 + ai * temp_r1;

  550. #else

  551.                 	y_ptr[0] += ar * temp_r + ai * temp_i;

  552.                 	y_ptr[1] -= ar * temp_i - ai * temp_r;

  553.                 	y_ptr += inc_y;

  554.                 	y_ptr[0] += ar * temp_r1 + ai * temp_i1;

  555.                 	y_ptr[1] -= ar * temp_i1 - ai * temp_r1;

  556. #endif

  557. 

  558.                 	a_ptr += lda;

  559.                 	y_ptr += inc_y;

  560.                 	j+=2;

  561.         	}

  562. 

  563. 

  564. 	        while ( j < n)

  565.         	{

  566. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  567.                        	temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 

  568.                        	temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 

  569.                        	temp_r += a_ptr[2] * x2 - a_ptr[3] * x3; 

  570.                        	temp_i += a_ptr[2] * x3 + a_ptr[3] * x2; 

  571. #else

  572. 

  573.                        	temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 

  574.                        	temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 

  575.                        	temp_r += a_ptr[2] * x2 + a_ptr[3] * x3; 

  576.                        	temp_i += a_ptr[2] * x3 - a_ptr[3] * x2; 

  577. #endif

  578. 

  579. #if !defined(XCONJ) 

  580.                 	y_ptr[0] += ar * temp_r - ai * temp_i;

  581.                 	y_ptr[1] += ar * temp_i + ai * temp_r;

  582. #else

  583.                 	y_ptr[0] += ar * temp_r + ai * temp_i;

  584.                 	y_ptr[1] -= ar * temp_i - ai * temp_r;

  585. #endif

  586. 

  587.                 	a_ptr += lda;

  588.                 	y_ptr += inc_y;

  589.                 	j++;

  590.         	}

  591. 

  592.         	return(0);

  593. 	}

  594. 

  595. 

  596. 	if ( m3 == 1 )

  597. 	{

  598. 

  599.                 FLOAT temp_r ;

  600.                 FLOAT temp_i ;

  601.                 FLOAT temp_r1 ;

  602.                 FLOAT temp_i1 ;

  603. 		FLOAT x0 = x_ptr[0];

  604. 		FLOAT x1 = x_ptr[1];

  605. 		FLOAT ar = alpha[0];

  606. 		FLOAT ai = alpha[1];

  607. 

  608. 	        while ( j < ( n & -2 ))

  609.         	{

  610. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  611.                        	temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 

  612.                        	temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 

  613.                 	a_ptr += lda;

  614.                        	temp_r1  = a_ptr[0] * x0 - a_ptr[1] * x1; 

  615.                        	temp_i1  = a_ptr[0] * x1 + a_ptr[1] * x0; 

  616. #else

  617. 

  618.                        	temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 

  619.                        	temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 

  620.                 	a_ptr += lda;

  621.                        	temp_r1  = a_ptr[0] * x0 + a_ptr[1] * x1; 

  622.                        	temp_i1  = a_ptr[0] * x1 - a_ptr[1] * x0; 

  623. #endif

  624. 

  625. #if !defined(XCONJ) 

  626.                 	y_ptr[0] += ar * temp_r - ai * temp_i;

  627.                 	y_ptr[1] += ar * temp_i + ai * temp_r;

  628.                 	y_ptr += inc_y;

  629.                 	y_ptr[0] += ar * temp_r1 - ai * temp_i1;

  630.                 	y_ptr[1] += ar * temp_i1 + ai * temp_r1;

  631. #else

  632.                 	y_ptr[0] += ar * temp_r + ai * temp_i;

  633.                 	y_ptr[1] -= ar * temp_i - ai * temp_r;

  634.                 	y_ptr += inc_y;

  635.                 	y_ptr[0] += ar * temp_r1 + ai * temp_i1;

  636.                 	y_ptr[1] -= ar * temp_i1 - ai * temp_r1;

  637. #endif

  638. 

  639.                 	a_ptr += lda;

  640.                 	y_ptr += inc_y;

  641.                 	j+=2;

  642.         	}

  643. 

  644. 	        while ( j < n)

  645.         	{

  646. #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )

  647.                        	temp_r  = a_ptr[0] * x0 - a_ptr[1] * x1; 

  648.                        	temp_i  = a_ptr[0] * x1 + a_ptr[1] * x0; 

  649. #else

  650. 

  651.                        	temp_r  = a_ptr[0] * x0 + a_ptr[1] * x1; 

  652.                        	temp_i  = a_ptr[0] * x1 - a_ptr[1] * x0; 

  653. #endif

  654. 

  655. #if !defined(XCONJ) 

  656.                 	y_ptr[0] += ar * temp_r - ai * temp_i;

  657.                 	y_ptr[1] += ar * temp_i + ai * temp_r;

  658. #else

  659.                 	y_ptr[0] += ar * temp_r + ai * temp_i;

  660.                 	y_ptr[1] -= ar * temp_i - ai * temp_r;

  661. #endif

  662. 

  663.                 	a_ptr += lda;

  664.                 	y_ptr += inc_y;

  665.                 	j++;

  666.         	}

  667.         	return(0);

  668. 	}

  669. 

  670. 	return(0);

  671. }