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: a7b1f9b1bb
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a7b1f9b1bb'
${ noResults }
OpenBLAS/kernel/x86_64/dtrsm_kernel_RN_haswell.c

674 lines
20 kB
Raw Blame History 

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

  2. /* Copyright 2009, 2010 The University of Texas at Austin.           */

  3. /* All rights reserved.                                              */

  4. /*                                                                   */

  5. /* Redistribution and use in source and binary forms, with or        */

  6. /* without modification, are permitted provided that the following   */

  7. /* conditions are met:                                               */

  8. /*                                                                   */

  9. /*   1. Redistributions of source code must retain the above         */

  10. /*      copyright notice, this list of conditions and the following  */

  11. /*      disclaimer.                                                  */

  12. /*                                                                   */

  13. /*   2. Redistributions in binary form must reproduce the above      */

  14. /*      copyright notice, this list of conditions and the following  */

  15. /*      disclaimer in the documentation and/or other materials       */

  16. /*      provided with the distribution.                              */

  17. /*                                                                   */

  18. /*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */

  19. /*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */

  20. /*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */

  21. /*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */

  22. /*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */

  23. /*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */

  24. /*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */

  25. /*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */

  26. /*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */

  27. /*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */

  28. /*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */

  29. /*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */

  30. /*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */

  31. /*    POSSIBILITY OF SUCH DAMAGE.                                    */

  32. /*                                                                   */

  33. /* The views and conclusions contained in the software and           */

  34. /* documentation are those of the authors and should not be          */

  35. /* interpreted as representing official policies, either expressed   */

  36. /* or implied, of The University of Texas at Austin.                 */

  37. /*********************************************************************/

  38. 

  39. #include "common.h"

  40. 

  41. static FLOAT dm1 = -1.;

  42. 

  43. #ifdef CONJ

  44. #define GEMM_KERNEL   GEMM_KERNEL_R

  45. #else

  46. #define GEMM_KERNEL   GEMM_KERNEL_N

  47. #endif

  48. 

  49. #if GEMM_DEFAULT_UNROLL_M == 1

  50. #define GEMM_UNROLL_M_SHIFT 0

  51. #endif

  52. 

  53. #if GEMM_DEFAULT_UNROLL_M == 2

  54. #define GEMM_UNROLL_M_SHIFT 1

  55. #endif

  56. 

  57. #if GEMM_DEFAULT_UNROLL_M == 4

  58. #define GEMM_UNROLL_M_SHIFT 2

  59. #endif

  60. 

  61. #if GEMM_DEFAULT_UNROLL_M == 6

  62. #define GEMM_UNROLL_M_SHIFT 2

  63. #endif

  64. 

  65. #if GEMM_DEFAULT_UNROLL_M == 8

  66. #define GEMM_UNROLL_M_SHIFT 3

  67. #endif

  68. 

  69. #if GEMM_DEFAULT_UNROLL_M == 16

  70. #define GEMM_UNROLL_M_SHIFT 4

  71. #endif

  72. 

  73. #if GEMM_DEFAULT_UNROLL_N == 1

  74. #define GEMM_UNROLL_N_SHIFT 0

  75. #endif

  76. 

  77. #if GEMM_DEFAULT_UNROLL_N == 2

  78. #define GEMM_UNROLL_N_SHIFT 1

  79. #endif

  80. 

  81. #if GEMM_DEFAULT_UNROLL_N == 4

  82. #define GEMM_UNROLL_N_SHIFT 2

  83. #endif

  84. 

  85. #if GEMM_DEFAULT_UNROLL_N == 8

  86. #define GEMM_UNROLL_N_SHIFT 3

  87. #endif

  88. 

  89. #if GEMM_DEFAULT_UNROLL_N == 16

  90. #define GEMM_UNROLL_N_SHIFT 4

  91. #endif

  92. 

  93. 

  94. 

  95. static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, FLOAT *as, FLOAT *bs)  __attribute__ ((noinline));

  96. 

  97. static void dtrsm_RN_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, FLOAT *as, FLOAT *bs)

  98. {

  99. 

  100. 	FLOAT *c3 = c + ldc + ldc*2 ;

  101. 	FLOAT *c6 = c + ldc*4 + ldc*2 ;

  102. 	ldc = ldc *8;

  103. 	BLASLONG n1 = n * 8;

  104. 	BLASLONG i=0;

  105. 

  106.         __asm__  __volatile__

  107.         (

  108. 	"	vzeroupper							\n\t"

  109. 

  110. 	"	vxorpd	%%ymm8 , %%ymm8 , %%ymm8				\n\t"

  111. 	"	vxorpd	%%ymm9 , %%ymm9 , %%ymm9				\n\t"

  112. 	"	vxorpd	%%ymm10, %%ymm10, %%ymm10				\n\t"

  113. 	"	vxorpd	%%ymm11, %%ymm11, %%ymm11				\n\t"

  114. 	"	vxorpd	%%ymm12, %%ymm12, %%ymm12				\n\t"

  115. 	"	vxorpd	%%ymm13, %%ymm13, %%ymm13				\n\t"

  116. 	"	vxorpd	%%ymm14, %%ymm14, %%ymm14				\n\t"

  117. 	"	vxorpd	%%ymm15, %%ymm15, %%ymm15				\n\t"

  118. 

  119. 	"	cmpq	       $0, %0						\n\t"

  120. 	"	je	       4f						\n\t"

  121. 

  122. 	"	vmovups         (%8,%1,4), %%ymm0				\n\t"	// read a

  123. 	"	vmovups         (%9,%1,8), %%ymm1				\n\t"	// read b0

  124. 	"	vmovups       32(%9,%1,8), %%ymm2				\n\t"	// read b1

  125. 

  126. 

  127. 	"	addq		$8, %1						\n\t"

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

  129. 	"	je	       21f						\n\t"

  130. 

  131. 	"	.p2align 4							\n\t"

  132. 	"1:									\n\t"

  133. 

  134. 	"	vmovups         (%8,%1,4), %%ymm4				\n\t"	// read a

  135.         "       vpermilpd         $0x05  , %%ymm0 , %%ymm3                	\n\t"   // was vpermpd 0xb1

  136. 

  137. 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm8			\n\t"

  138. 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm12			\n\t"

  139. 

  140. 	"	vmovups         (%9,%1,8), %%ymm5				\n\t"	// read b0

  141. 	"	vfmadd231pd	%%ymm3 , %%ymm1 , %%ymm9			\n\t"

  142. 	"	vfmadd231pd	%%ymm3 , %%ymm2 , %%ymm13			\n\t"

  143. 

  144.         "       vpermpd         $0x1b  , %%ymm3 , %%ymm0                	\n\t"

  145. 	"	vmovups       32(%9,%1,8), %%ymm6				\n\t"	// read b1

  146.         "       vpermilpd         $0x05  , %%ymm0 , %%ymm3                	\n\t"

  147. 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm10			\n\t"

  148. 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm14			\n\t"

  149. 

  150. 	"	addq		$8, %1						\n\t"

  151. 	"	vfmadd231pd	%%ymm3 , %%ymm1 , %%ymm11			\n\t"

  152. 	"	vfmadd231pd	%%ymm3 , %%ymm2 , %%ymm15			\n\t"

  153. 

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

  155. 

  156. 	"	jz		22f						\n\t"

  157. 

  158. 	"	vmovups         (%8,%1,4), %%ymm0				\n\t"	// read a

  159. 

  160. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm8			\n\t"

  161. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm12			\n\t"

  162. 

  163.         "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"

  164. 	"	vmovups         (%9,%1,8), %%ymm1				\n\t"	// read b0

  165. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm9			\n\t"

  166. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm13			\n\t"

  167. 

  168.         "       vpermpd         $0x1b  , %%ymm4 , %%ymm4                	\n\t"

  169. 	"	vmovups       32(%9,%1,8), %%ymm2				\n\t"	// read b1

  170. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm10			\n\t"

  171. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm14			\n\t"

  172. 

  173.         "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"

  174. 	"	addq		$8, %1						\n\t"

  175. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm11			\n\t"

  176. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm15			\n\t"

  177. 

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

  179. 

  180. 	"	jnz		1b						\n\t"

  181. 

  182. 

  183. 	"21:									\n\t"

  184. 

  185. 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm8			\n\t"

  186. 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm12			\n\t"

  187. 

  188.         "       vpermilpd         $0x05  , %%ymm0 , %%ymm0                	\n\t"

  189. 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm9			\n\t"

  190. 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm13			\n\t"

  191. 

  192.         "       vpermpd         $0x1b  , %%ymm0 , %%ymm0                	\n\t"

  193. 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm10			\n\t"

  194. 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm14			\n\t"

  195. 

  196.         "       vpermilpd         $0x05  , %%ymm0 , %%ymm0                	\n\t"

  197. 	"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm11			\n\t"

  198. 	"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm15			\n\t"

  199. 

  200. 	"	jmp	3f							\n\t"

  201. 

  202. 	"22:									\n\t"

  203. 	

  204. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm8			\n\t"

  205. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm12			\n\t"

  206. 

  207.         "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"

  208. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm9			\n\t"

  209. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm13			\n\t"

  210. 

  211.         "       vpermpd         $0x1b  , %%ymm4 , %%ymm4                	\n\t"

  212. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm10			\n\t"

  213. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm14			\n\t"

  214. 

  215.         "       vpermilpd         $0x05  , %%ymm4 , %%ymm4                	\n\t"

  216. 	"	vfmadd231pd	%%ymm4 , %%ymm5 , %%ymm11			\n\t"

  217. 	"	vfmadd231pd	%%ymm4 , %%ymm6 , %%ymm15			\n\t"

  218. 

  219. 	"3:								\n\t"	

  220. 

  221.         "       vpermilpd         $0x05  , %%ymm9 , %%ymm9                \n\t"

  222.         "       vpermilpd         $0x05  , %%ymm11, %%ymm11               \n\t"

  223. 

  224.         "       vblendpd        $0x0a  , %%ymm9 , %%ymm8 , %%ymm0       \n\t"

  225.         "       vblendpd        $0x05  , %%ymm9 , %%ymm8 , %%ymm1       \n\t"

  226.         "       vblendpd        $0x0a  , %%ymm11, %%ymm10, %%ymm2       \n\t"

  227.         "       vblendpd        $0x05  , %%ymm11, %%ymm10, %%ymm3       \n\t"

  228. 

  229.         "       vperm2f128         $0x01  , %%ymm2 , %%ymm2 , %%ymm2    \n\t"

  230.         "       vperm2f128         $0x01  , %%ymm3 , %%ymm3 , %%ymm3    \n\t"

  231. 

  232.         "       vblendpd        $0x03  , %%ymm0 , %%ymm2 , %%ymm8       \n\t"

  233.         "       vblendpd        $0x03  , %%ymm1 , %%ymm3 , %%ymm9       \n\t"

  234.         "       vblendpd        $0x03  , %%ymm2 , %%ymm0 , %%ymm10      \n\t"

  235.         "       vblendpd        $0x03  , %%ymm3 , %%ymm1 , %%ymm11      \n\t"

  236. 

  237.         "       vpermilpd         $0x05  , %%ymm13, %%ymm13               \n\t"

  238.         "       vpermilpd         $0x05  , %%ymm15, %%ymm15               \n\t"

  239. 

  240.         "       vblendpd        $0x0a  , %%ymm13, %%ymm12, %%ymm0       \n\t"

  241.         "       vblendpd        $0x05  , %%ymm13, %%ymm12, %%ymm1       \n\t"

  242.         "       vblendpd        $0x0a  , %%ymm15, %%ymm14, %%ymm2       \n\t"

  243.         "       vblendpd        $0x05  , %%ymm15, %%ymm14, %%ymm3       \n\t"

  244. 

  245.         "       vperm2f128         $0x01  , %%ymm2 , %%ymm2 , %%ymm2       \n\t"

  246.         "       vperm2f128         $0x01  , %%ymm3 , %%ymm3 , %%ymm3       \n\t"

  247. 

  248.         "       vblendpd        $0x03  , %%ymm0 , %%ymm2 , %%ymm12      \n\t"

  249.         "       vblendpd        $0x03  , %%ymm1 , %%ymm3 , %%ymm13      \n\t"

  250.         "       vblendpd        $0x03  , %%ymm2 , %%ymm0 , %%ymm14      \n\t"

  251.         "       vblendpd        $0x03  , %%ymm3 , %%ymm1 , %%ymm15      \n\t"

  252. 

  253. 

  254. 	"4:								\n\t"	

  255. 

  256. 	"	vmovups		  (%4)      , %%ymm0			\n\t"	// read c0

  257. 	"	vmovups		  (%4,%7,1) , %%ymm1			\n\t"	// read c1

  258. 	"	vmovups		  (%4,%7,2) , %%ymm2			\n\t"	// read c2

  259. 	"	vmovups		  (%5)      , %%ymm3			\n\t"	// read c3

  260. 

  261. 	"	vmovups		  (%5,%7,1) , %%ymm4			\n\t"	// read c4

  262. 	"	vmovups		  (%5,%7,2) , %%ymm5			\n\t"	// read c5

  263. 	"	vmovups		  (%6)      , %%ymm6			\n\t"	// read c6

  264. 	"	vmovups		  (%6,%7,1) , %%ymm7			\n\t"	// read c7

  265. 

  266. 	"	vsubpd		%%ymm8 , %%ymm0 , %%ymm8		\n\t"

  267. 	"	vmovups           (%3),  %%ymm0				\n\t"

  268. 	"	vsubpd		%%ymm9 , %%ymm1 , %%ymm9		\n\t"

  269. 	"	vpermpd		$0x55 ,  %%ymm0 , %%ymm1		\n\t"

  270. 	"	vsubpd		%%ymm10, %%ymm2 , %%ymm10		\n\t"

  271. 	"	vpermpd		$0xaa ,  %%ymm0 , %%ymm2		\n\t"

  272. 	"	vsubpd		%%ymm11, %%ymm3 , %%ymm11		\n\t"

  273. 	"	vpermpd		$0xff ,  %%ymm0 , %%ymm3		\n\t"

  274. 	"	vpermpd		$0x00 ,  %%ymm0 , %%ymm0		\n\t"

  275. 

  276. 	"	vsubpd		%%ymm12, %%ymm4 , %%ymm12		\n\t"

  277. 	"	vmovups         32(%3),  %%ymm4				\n\t"

  278. 	"	vsubpd		%%ymm13, %%ymm5 , %%ymm13		\n\t"

  279. 	"	vpermpd		$0x55 ,  %%ymm4 , %%ymm5		\n\t"

  280. 	"	vsubpd		%%ymm14, %%ymm6 , %%ymm14		\n\t"

  281. 	"	vpermpd		$0xaa ,  %%ymm4 , %%ymm6		\n\t"

  282. 	"	vsubpd		%%ymm15, %%ymm7 , %%ymm15		\n\t"

  283. 	"	vpermpd		$0xff ,  %%ymm4 , %%ymm7		\n\t"

  284. 	"	vpermpd		$0x00 ,  %%ymm4 , %%ymm4		\n\t"

  285. 

  286. 

  287. 	"5:								\n\t"	// i = 0

  288. 

  289. 	"	addq	$64, %3						\n\t"	// b=b+8

  290. 

  291. 	"	vmulpd		%%ymm8 , %%ymm0, %%ymm8			\n\t"	// a *bb

  292. 	"	vmovups           (%3),  %%ymm0				\n\t"

  293. 	"	vmovups		%%ymm8 , (%2)				\n\t"	// write a

  294. 	"	vmovups		%%ymm8 , (%4)				\n\t"	// write c

  295. 

  296. 	"	vfnmadd231pd	%%ymm8 , %%ymm1 , %%ymm9		\n\t"

  297. 	"	vmovups         32(%3),  %%ymm1				\n\t"

  298. 	"	vfnmadd231pd	%%ymm8 , %%ymm2 , %%ymm10		\n\t"

  299. 	"	vpermpd		$0xaa ,  %%ymm0 , %%ymm2		\n\t"

  300. 	"	vfnmadd231pd	%%ymm8 , %%ymm3 , %%ymm11		\n\t"

  301. 	"	vpermpd		$0xff ,  %%ymm0 , %%ymm3		\n\t"

  302. 	"	vfnmadd231pd	%%ymm8 , %%ymm4 , %%ymm12		\n\t"

  303. 	"	vpermpd		$0x55 ,  %%ymm0 , %%ymm0		\n\t"

  304. 	"	vfnmadd231pd	%%ymm8 , %%ymm5 , %%ymm13		\n\t"

  305. 	"	vpermpd		$0x55 ,  %%ymm1 , %%ymm5		\n\t"

  306. 	"	vfnmadd231pd	%%ymm8 , %%ymm6 , %%ymm14		\n\t"

  307. 	"	vpermpd		$0xaa ,  %%ymm1 , %%ymm6		\n\t"

  308. 	"	vfnmadd231pd	%%ymm8 , %%ymm7 , %%ymm15		\n\t"

  309. 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"

  310. 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm4		\n\t"

  311. 

  312. 	"	addq	$64, %3						\n\t"	// b=b+8

  313. 	"	addq	$32, %2						\n\t"	// a=a+8

  314. 

  315. 

  316. 

  317. 	"	vmulpd		%%ymm9 , %%ymm0, %%ymm9			\n\t"	// a *bb

  318. 	"	vmovups           (%3),  %%ymm0				\n\t"

  319. 	"	vmovups         32(%3),  %%ymm1				\n\t"

  320. 	"	vmovups		%%ymm9 , (%2)				\n\t"	// write a

  321. 	"	vmovups		%%ymm9 , (%4,%7,1)			\n\t"	// write c

  322. 

  323. 	"	vfnmadd231pd	%%ymm9 , %%ymm2 , %%ymm10		\n\t"

  324. 	"	vfnmadd231pd	%%ymm9 , %%ymm3 , %%ymm11		\n\t"

  325. 	"	vpermpd		$0xff ,  %%ymm0 , %%ymm3		\n\t"

  326. 	"	vfnmadd231pd	%%ymm9 , %%ymm4 , %%ymm12		\n\t"

  327. 	"	vpermpd		$0xaa ,  %%ymm0 , %%ymm0		\n\t"

  328. 	"	vfnmadd231pd	%%ymm9 , %%ymm5 , %%ymm13		\n\t"

  329. 	"	vpermpd		$0x55 ,  %%ymm1 , %%ymm5		\n\t"

  330. 	"	vfnmadd231pd	%%ymm9 , %%ymm6 , %%ymm14		\n\t"

  331. 	"	vpermpd		$0xaa ,  %%ymm1 , %%ymm6		\n\t"

  332. 	"	vfnmadd231pd	%%ymm9 , %%ymm7 , %%ymm15		\n\t"

  333. 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"

  334. 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm4		\n\t"

  335. 

  336. 	"	addq	$64, %3						\n\t"	// b=b+8

  337. 	"	addq	$32, %2						\n\t"	// a=a+8

  338. 

  339. 	"	vmulpd		%%ymm10, %%ymm0, %%ymm10		\n\t"	// a *bb

  340. 	"	vmovups           (%3),  %%ymm0				\n\t"

  341. 	"	vmovups         32(%3),  %%ymm1				\n\t"

  342. 	"	vmovups		%%ymm10, (%2)				\n\t"	// write a

  343. 	"	vmovups		%%ymm10, (%4,%7,2)			\n\t"	// write c

  344. 

  345. 	"	vfnmadd231pd	%%ymm10, %%ymm3 , %%ymm11		\n\t"

  346. 	"	vpermpd		$0xff ,  %%ymm0 , %%ymm0		\n\t"

  347. 	"	vfnmadd231pd	%%ymm10, %%ymm4 , %%ymm12		\n\t"

  348. 	"	vfnmadd231pd	%%ymm10, %%ymm5 , %%ymm13		\n\t"

  349. 	"	vpermpd		$0x55 ,  %%ymm1 , %%ymm5		\n\t"

  350. 	"	vfnmadd231pd	%%ymm10, %%ymm6 , %%ymm14		\n\t"

  351. 	"	vpermpd		$0xaa ,  %%ymm1 , %%ymm6		\n\t"

  352. 	"	vfnmadd231pd	%%ymm10, %%ymm7 , %%ymm15		\n\t"

  353. 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"

  354. 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm4		\n\t"

  355. 

  356. 

  357. 	"	addq	$64, %3						\n\t"	// b=b+8

  358. 	"	addq	$32, %2						\n\t"	// a=a+8

  359. 

  360. 

  361. 

  362. 	"	vmulpd		%%ymm11, %%ymm0, %%ymm11		\n\t"	// a *bb

  363. 	"	vmovups         32(%3),  %%ymm1				\n\t"

  364. 	"	vmovups		%%ymm11, (%2)				\n\t"	// write a

  365. 	"	vmovups		%%ymm11, (%5)     			\n\t"	// write c

  366. 

  367. 	"	vfnmadd231pd	%%ymm11, %%ymm4 , %%ymm12		\n\t"

  368. 	"	vfnmadd231pd	%%ymm11, %%ymm5 , %%ymm13		\n\t"

  369. 	"	vpermpd		$0x55 ,  %%ymm1 , %%ymm5		\n\t"

  370. 	"	vfnmadd231pd	%%ymm11, %%ymm6 , %%ymm14		\n\t"

  371. 	"	vpermpd		$0xaa ,  %%ymm1 , %%ymm6		\n\t"

  372. 	"	vfnmadd231pd	%%ymm11, %%ymm7 , %%ymm15		\n\t"

  373. 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"

  374. 	"	vpermpd		$0x00 ,  %%ymm1 , %%ymm0		\n\t"

  375. 

  376. 

  377. 	"	addq	$64, %3						\n\t"	// b=b+8

  378. 	"	addq	$32, %2						\n\t"	// a=a+8

  379. 

  380. 

  381. 	"	vmulpd		%%ymm12, %%ymm0, %%ymm12		\n\t"	// a *bb

  382. 	"	vmovups         32(%3),  %%ymm1				\n\t"

  383. 	"	vmovups		%%ymm12, (%2)				\n\t"	// write a

  384. 	"	vmovups		%%ymm12, (%5,%7,1)			\n\t"	// write c

  385. 

  386. 	"	vfnmadd231pd	%%ymm12, %%ymm5 , %%ymm13		\n\t"

  387. 	"	vfnmadd231pd	%%ymm12, %%ymm6 , %%ymm14		\n\t"

  388. 	"	vpermpd		$0xaa ,  %%ymm1 , %%ymm6		\n\t"

  389. 	"	vfnmadd231pd	%%ymm12, %%ymm7 , %%ymm15		\n\t"

  390. 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"

  391. 	"	vpermpd		$0x55 ,  %%ymm1 , %%ymm0		\n\t"

  392. 

  393. 	"	addq	$64, %3						\n\t"	// b=b+8

  394. 	"	addq	$32, %2						\n\t"	// a=a+8

  395. 

  396. 	"	vmulpd		%%ymm13, %%ymm0, %%ymm13		\n\t"	// a *bb

  397. 	"	vmovups         32(%3),  %%ymm1				\n\t"

  398. 	"	vmovups		%%ymm13, (%2)				\n\t"	// write a

  399. 	"	vmovups		%%ymm13, (%5,%7,2)			\n\t"	// write c

  400. 

  401. 	"	vfnmadd231pd	%%ymm13, %%ymm6 , %%ymm14		\n\t"

  402. 	"	vfnmadd231pd	%%ymm13, %%ymm7 , %%ymm15		\n\t"

  403. 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm7		\n\t"

  404. 	"	vpermpd		$0xaa ,  %%ymm1 , %%ymm0		\n\t"

  405. 

  406. 

  407. 	"	addq	$64, %3						\n\t"	// b=b+8

  408. 	"	addq	$32, %2						\n\t"	// a=a+8

  409. 

  410. 

  411. 	"	vmulpd		%%ymm14, %%ymm0, %%ymm14		\n\t"	// a *bb

  412. 	"	vmovups         32(%3),  %%ymm1				\n\t"

  413. 	"	vmovups		%%ymm14, (%2)				\n\t"	// write a

  414. 	"	vmovups		%%ymm14, (%6)     			\n\t"	// write c

  415. 

  416. 	"	vfnmadd231pd	%%ymm14, %%ymm7 , %%ymm15		\n\t"

  417. 

  418. 	"	vpermpd		$0xff ,  %%ymm1 , %%ymm0		\n\t"

  419. 

  420. 	"	addq	$32, %2						\n\t"	// a=a+8

  421. 

  422. 	"	vmulpd		%%ymm15, %%ymm0, %%ymm15		\n\t"	// a *bb

  423. 	"	vmovups		%%ymm15, (%2)				\n\t"	// write a

  424. 	"	vmovups		%%ymm15, (%6,%7,1)			\n\t"	// write c

  425. 

  426. 	"	vzeroupper						\n\t"

  427. 

  428.         :

  429.           "+r" (n1),     // 0    

  430.           "+a" (i),      // 1    

  431.           "+r" (as),     // 2

  432.           "+r" (bs)      // 3

  433.         :

  434.           "r" (c),      // 4

  435.           "r" (c3),     // 5

  436.           "r" (c6),     // 6

  437.           "r" (ldc),    // 7

  438.           "r" (a),     // 8

  439.           "r" (b)      // 9

  440.         : "cc",

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

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

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

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

  445.           "memory"

  446.         );

  447. 

  448. }

  449. 

  450. 

  451. 

  452. #ifndef COMPLEX

  453. 

  454. static inline void solve(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {

  455. 

  456.   FLOAT aa, bb;

  457. 

  458.   int i, j, k;

  459. 

  460.   for (i = 0; i < n; i++) {

  461. 

  462.     bb = *(b + i);

  463. 

  464.     for (j = 0; j < m; j ++) {

  465.       aa = *(c + j + i * ldc);

  466.       aa *= bb;

  467.       *a  = aa;

  468.       *(c + j + i * ldc) = aa;

  469.       a ++;

  470. 

  471.       for (k = i + 1; k < n; k ++){

  472. 	*(c + j + k * ldc) -= aa * *(b + k);

  473.       }

  474. 

  475.     }

  476.     b += n;

  477.   }

  478. }

  479. 

  480. #else

  481. 

  482. static inline void solve(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {

  483. 

  484.   FLOAT aa1, aa2;

  485.   FLOAT bb1, bb2;

  486.   FLOAT cc1, cc2;

  487. 

  488.   int i, j, k;

  489. 

  490.   ldc *= 2;

  491. 

  492.   for (i = 0; i < n; i++) {

  493. 

  494.     bb1 = *(b + i * 2 + 0);

  495.     bb2 = *(b + i * 2 + 1);

  496. 

  497.     for (j = 0; j < m; j ++) {

  498.       aa1 = *(c + j * 2 + 0 + i * ldc);

  499.       aa2 = *(c + j * 2 + 1 + i * ldc);

  500. 

  501. #ifndef CONJ

  502.       cc1 = aa1 * bb1 - aa2 * bb2;

  503.       cc2 = aa1 * bb2 + aa2 * bb1;

  504. #else

  505.       cc1 =  aa1 * bb1 + aa2 * bb2;

  506.       cc2 = -aa1 * bb2 + aa2 * bb1;

  507. #endif

  508. 

  509.       *(a + 0) = cc1;

  510.       *(a + 1) = cc2;

  511.       *(c + j * 2 + 0 + i * ldc) = cc1;

  512.       *(c + j * 2 + 1 + i * ldc) = cc2;

  513.       a += 2;

  514. 

  515.       for (k = i + 1; k < n; k ++){

  516. #ifndef CONJ

  517. 	*(c + j * 2 + 0 + k * ldc) -= cc1 * *(b + k * 2 + 0) - cc2 * *(b + k * 2 + 1);

  518. 	*(c + j * 2 + 1 + k * ldc) -= cc1 * *(b + k * 2 + 1) + cc2 * *(b + k * 2 + 0);

  519. #else

  520. 	*(c + j * 2 + 0 + k * ldc) -=   cc1 * *(b + k * 2 + 0) + cc2 * *(b + k * 2 + 1);

  521. 	*(c + j * 2 + 1 + k * ldc) -= - cc1 * *(b + k * 2 + 1) + cc2 * *(b + k * 2 + 0);

  522. #endif

  523.       }

  524. 

  525.     }

  526.     b += n * 2;

  527.   }

  528. }

  529. 

  530. #endif

  531. 

  532. 

  533. int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT dummy1,

  534. #ifdef COMPLEX

  535. 	   FLOAT dummy2,

  536. #endif

  537. 	   FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG offset){

  538. 

  539.   FLOAT *aa, *cc;

  540.   BLASLONG  kk;

  541.   BLASLONG i, j, jj;

  542. 

  543. #if 0

  544.   fprintf(stderr, "TRSM RN KERNEL m = %3ld  n = %3ld  k = %3ld offset = %3ld\n",

  545. 	  m, n, k, offset);

  546. #endif

  547. 

  548.   jj = 0;

  549.   j = (n >> GEMM_UNROLL_N_SHIFT);

  550.   kk = -offset;

  551. 

  552.   while (j > 0) {

  553. 

  554.     aa = a;

  555.     cc = c;

  556. 

  557.     i = (m >> GEMM_UNROLL_M_SHIFT);

  558. 

  559.     if (i > 0) {

  560.       do {

  561. 

  562. 	dtrsm_RN_solve_opt(kk, aa, b, cc, ldc, aa + kk * GEMM_UNROLL_M * COMPSIZE, b + kk * GEMM_UNROLL_N * COMPSIZE);

  563. /* 

  564.         solve(GEMM_UNROLL_M, GEMM_UNROLL_N,

  565.               aa + kk * GEMM_UNROLL_M * COMPSIZE,

  566.               b  + kk * GEMM_UNROLL_N * COMPSIZE,

  567.               cc, ldc);

  568. */

  569. 	aa += GEMM_UNROLL_M * k * COMPSIZE;

  570. 	cc += GEMM_UNROLL_M     * COMPSIZE;

  571. 	i --;

  572.       } while (i > 0);

  573.     }

  574. 

  575. 

  576.     if (m & (GEMM_UNROLL_M - 1)) {

  577.       i = (GEMM_UNROLL_M >> 1);

  578.       while (i > 0) {

  579. 	if (m & i) {

  580. 	    if (kk > 0) {

  581. 	      GEMM_KERNEL(i, GEMM_UNROLL_N, kk, dm1,

  582. #ifdef COMPLEX

  583. 			  ZERO,

  584. #endif

  585. 			  aa, b, cc, ldc);

  586. 	    }

  587. 	  solve(i, GEMM_UNROLL_N,

  588. 		aa + kk * i             * COMPSIZE,

  589. 		b  + kk * GEMM_UNROLL_N * COMPSIZE,

  590. 		cc, ldc);

  591. 

  592. 	  aa += i * k * COMPSIZE;

  593. 	  cc += i     * COMPSIZE;

  594. 	}

  595. 	i >>= 1;

  596.       }

  597.     }

  598. 

  599.     kk += GEMM_UNROLL_N;

  600.     b += GEMM_UNROLL_N * k   * COMPSIZE;

  601.     c += GEMM_UNROLL_N * ldc * COMPSIZE;

  602.     j --;

  603.     jj += GEMM_UNROLL_M;

  604.   }

  605. 

  606.   if (n & (GEMM_UNROLL_N - 1)) {

  607. 

  608.     j = (GEMM_UNROLL_N >> 1);

  609.     while (j > 0) {

  610.       if (n & j) {

  611. 

  612. 	aa = a;

  613. 	cc = c;

  614. 

  615. 	i = (m >> GEMM_UNROLL_M_SHIFT);

  616. 

  617. 	while (i > 0) {

  618. 	  if (kk > 0) {

  619. 	    GEMM_KERNEL(GEMM_UNROLL_M, j, kk, dm1,

  620. #ifdef COMPLEX

  621. 			ZERO,

  622. #endif

  623. 			aa,

  624. 			b,

  625. 			cc,

  626. 			ldc);

  627. 	  }

  628. 

  629. 	  solve(GEMM_UNROLL_M, j,

  630. 		aa + kk * GEMM_UNROLL_M * COMPSIZE,

  631. 		b  + kk * j             * COMPSIZE, cc, ldc);

  632. 

  633. 	  aa += GEMM_UNROLL_M * k * COMPSIZE;

  634. 	  cc += GEMM_UNROLL_M     * COMPSIZE;

  635. 	  i --;

  636. 	}

  637. 

  638. 	if (m & (GEMM_UNROLL_M - 1)) {

  639. 	  i = (GEMM_UNROLL_M >> 1);

  640. 	  while (i > 0) {

  641. 	    if (m & i) {

  642. 	      if (kk > 0) {

  643. 		GEMM_KERNEL(i, j, kk, dm1,

  644. #ifdef COMPLEX

  645. 			    ZERO,

  646. #endif

  647. 			    aa,

  648. 			    b,

  649. 			    cc,

  650. 			    ldc);

  651. 	      }

  652. 

  653. 	      solve(i, j,

  654. 		    aa + kk * i * COMPSIZE,

  655. 		    b  + kk * j * COMPSIZE, cc, ldc);

  656. 

  657. 	      aa += i * k * COMPSIZE;

  658. 	      cc += i     * COMPSIZE;

  659. 	      }

  660. 	    i >>= 1;

  661. 	  }

  662. 	}

  663. 

  664. 	b += j * k   * COMPSIZE;

  665. 	c += j * ldc * COMPSIZE;

  666. 	kk += j;

  667.       }

  668.       j >>= 1;

  669.     }

  670.   }

  671. 

  672.   return 0;

  673. }