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: 02634b549b
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '02634b549b'
${ noResults }
OpenBLAS/kernel/power/dgemv_n_microk_power8.c

300 lines
9.0 kB
Raw Blame History 

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

  2. Copyright (c) 2013-2016, 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. * 2016/03/30 Werner Saar (wernsaar@googlemail.com)

  30. * 	 BLASTEST 		: OK

  31. * 	 CTEST			: OK

  32. * 	 TEST			: OK

  33. *	 LAPACK-TEST		: OK

  34. **************************************************************************************/

  35. 

  36. #define HAVE_KERNEL_4x4 1

  37. 

  38. static void dgemv_kernel_4x4 (long n, double *ap, long lda, double *x, double *y, double alpha)

  39. {

  40.   double *a0;

  41.   double *a1;

  42.   double *a2;

  43.   double *a3;

  44. 

  45.   __asm__

  46.     (

  47.        "lxvd2x		34, 0, %10	\n\t"	// x0, x1

  48.        "lxvd2x		35, %11, %10	\n\t"	// x2, x3

  49.        "xxspltd		32, %x9, 0	\n\t"	// alpha, alpha

  50. 

  51.        "sldi		%6, %13, 3	\n\t"	// lda * sizeof (double)

  52. 

  53.        "xvmuldp		34, 34, 32	\n\t"	// x0 * alpha, x1 * alpha

  54.        "xvmuldp		35, 35, 32	\n\t"	// x2 * alpha, x3 * alpha

  55. 

  56.        "add		%4, %3, %6	\n\t"	// a0 = ap, a1 = a0 + lda

  57.        "add		%6, %6, %6	\n\t"	// 2 * lda

  58. 

  59.        "xxspltd		32, 34, 0	\n\t"	// x0 * alpha, x0 * alpha

  60.        "xxspltd		33, 34, 1	\n\t"	// x1 * alpha, x1 * alpha

  61.        "xxspltd		34, 35, 0	\n\t"	// x2 * alpha, x2 * alpha

  62.        "xxspltd		35, 35, 1	\n\t"	// x3 * alpha, x3 * alpha

  63. 

  64.        "add		%5, %3, %6	\n\t"	// a2 = a0 + 2 * lda

  65.        "add		%6, %4, %6	\n\t"	// a3 = a1 + 2 * lda

  66. 

  67.        "dcbt		0, %3		\n\t"

  68.        "dcbt		0, %4		\n\t"

  69.        "dcbt		0, %5		\n\t"

  70.        "dcbt		0, %6		\n\t"

  71. 

  72.        "lxvd2x		40, 0, %3	\n\t"	// a0[0], a0[1]

  73.        "lxvd2x		41, %11, %3	\n\t"	// a0[2], a0[3]

  74. 

  75.        "lxvd2x		42, 0, %4	\n\t"	// a1[0], a1[1]

  76.        "lxvd2x		43, %11, %4	\n\t"	// a1[2], a1[3]

  77. 

  78.        "lxvd2x		44, 0, %5	\n\t"	// a2[0], a2[1]

  79.        "lxvd2x		45, %11, %5	\n\t"	// a2[2], a2[3]

  80. 

  81.        "lxvd2x		46, 0, %6	\n\t"	// a3[0], a3[1]

  82.        "lxvd2x		47, %11, %6	\n\t"	// a3[2], a3[3]

  83. 

  84.        "dcbt		0, %2		\n\t"

  85. 

  86.        "addi		%3, %3, 32	\n\t"

  87.        "addi		%4, %4, 32	\n\t"

  88.        "addi		%5, %5, 32	\n\t"

  89.        "addi		%6, %6, 32	\n\t"

  90. 

  91.        "addic.		%1, %1, -4	\n\t"

  92.        "ble		2f		\n\t"

  93. 

  94.        ".p2align	5		\n"

  95.      "1:				\n\t"

  96. 

  97.        "lxvd2x		36, 0, %2	\n\t"	// y0, y1

  98.        "lxvd2x		37, %11, %2	\n\t"	// y2, y3

  99. 

  100.        "xvmaddadp 	36, 40, 32	\n\t"

  101.        "xvmaddadp 	37, 41, 32	\n\t"

  102. 

  103.        "lxvd2x		40, 0, %3	\n\t"	// a0[0], a0[1]

  104.        "lxvd2x		41, %11, %3	\n\t"	// a0[2], a0[3]

  105. 

  106.        "xvmaddadp 	36, 42, 33	\n\t"

  107.        "addi		%3, %3, 32	\n\t"

  108.        "xvmaddadp 	37, 43, 33	\n\t"

  109. 

  110.        "lxvd2x		42, 0, %4	\n\t"	// a1[0], a1[1]

  111.        "lxvd2x		43, %11, %4	\n\t"	// a1[2], a1[3]

  112. 

  113.        "xvmaddadp 	36, 44, 34	\n\t"

  114.        "addi		%4, %4, 32	\n\t"

  115.        "xvmaddadp 	37, 45, 34	\n\t"

  116. 

  117.        "lxvd2x		44, 0, %5	\n\t"	// a2[0], a2[1]

  118.        "lxvd2x		45, %11, %5	\n\t"	// a2[2], a2[3]

  119. 

  120.        "xvmaddadp 	36, 46, 35	\n\t"

  121.        "addi		%5, %5, 32	\n\t"

  122.        "xvmaddadp 	37, 47, 35	\n\t"

  123. 

  124.        "stxvd2x		36, 0, %2	\n\t"	// y0, y1

  125.        "stxvd2x		37, %11, %2	\n\t"	// y2, y3

  126. 

  127.        "lxvd2x		46, 0, %6	\n\t"	// a3[0], a3[1]

  128.        "lxvd2x		47, %11, %6	\n\t"	// a3[2], a3[3]

  129. 

  130.        "addi		%6, %6, 32	\n\t"

  131.        "addi		%2, %2, 32	\n\t"

  132. 

  133.        "addic.		%1, %1, -4	\n\t"

  134.        "ble		2f		\n\t"

  135. 

  136. 

  137.        "lxvd2x		36, 0, %2	\n\t"	// y0, y1

  138.        "lxvd2x		37, %11, %2	\n\t"	// y2, y3

  139. 

  140.        "xvmaddadp 	36, 40, 32	\n\t"

  141.        "xvmaddadp 	37, 41, 32	\n\t"

  142. 

  143.        "lxvd2x		40, 0, %3	\n\t"	// a0[0], a0[1]

  144.        "lxvd2x		41, %11, %3	\n\t"	// a0[2], a0[3]

  145. 

  146.        "xvmaddadp 	36, 42, 33	\n\t"

  147.        "addi		%3, %3, 32	\n\t"

  148.        "xvmaddadp 	37, 43, 33	\n\t"

  149. 

  150.        "lxvd2x		42, 0, %4	\n\t"	// a1[0], a1[1]

  151.        "lxvd2x		43, %11, %4	\n\t"	// a1[2], a1[3]

  152. 

  153.        "xvmaddadp 	36, 44, 34	\n\t"

  154.        "addi		%4, %4, 32	\n\t"

  155.        "xvmaddadp 	37, 45, 34	\n\t"

  156. 

  157.        "lxvd2x		44, 0, %5	\n\t"	// a2[0], a2[1]

  158.        "lxvd2x		45, %11, %5	\n\t"	// a2[2], a2[3]

  159. 

  160.        "xvmaddadp 	36, 46, 35	\n\t"

  161.        "addi		%5, %5, 32	\n\t"

  162.        "xvmaddadp 	37, 47, 35	\n\t"

  163. 

  164.        "stxvd2x		36, 0, %2	\n\t"	// y0, y1

  165.        "stxvd2x		37, %11, %2	\n\t"	// y2, y3

  166. 

  167.        "lxvd2x		46, 0, %6	\n\t"	// a3[0], a3[1]

  168.        "lxvd2x		47, %11, %6	\n\t"	// a3[2], a3[3]

  169. 

  170.        "addi		%6, %6, 32	\n\t"

  171.        "addi		%2, %2, 32	\n\t"

  172. 

  173.        "addic.		%1, %1, -4	\n\t"

  174.        "ble		2f		\n\t"

  175. 

  176. 

  177.        "lxvd2x		36, 0, %2	\n\t"	// y0, y1

  178.        "lxvd2x		37, %11, %2	\n\t"	// y2, y3

  179. 

  180.        "xvmaddadp 	36, 40, 32	\n\t"

  181.        "xvmaddadp 	37, 41, 32	\n\t"

  182. 

  183.        "lxvd2x		40, 0, %3	\n\t"	// a0[0], a0[1]

  184.        "lxvd2x		41, %11, %3	\n\t"	// a0[2], a0[3]

  185. 

  186.        "xvmaddadp 	36, 42, 33	\n\t"

  187.        "addi		%3, %3, 32	\n\t"

  188.        "xvmaddadp 	37, 43, 33	\n\t"

  189. 

  190.        "lxvd2x		42, 0, %4	\n\t"	// a1[0], a1[1]

  191.        "lxvd2x		43, %11, %4	\n\t"	// a1[2], a1[3]

  192. 

  193.        "xvmaddadp 	36, 44, 34	\n\t"

  194.        "addi		%4, %4, 32	\n\t"

  195.        "xvmaddadp 	37, 45, 34	\n\t"

  196. 

  197.        "lxvd2x		44, 0, %5	\n\t"	// a2[0], a2[1]

  198.        "lxvd2x		45, %11, %5	\n\t"	// a2[2], a2[3]

  199. 

  200.        "xvmaddadp 	36, 46, 35	\n\t"

  201.        "addi		%5, %5, 32	\n\t"

  202.        "xvmaddadp 	37, 47, 35	\n\t"

  203. 

  204.        "stxvd2x		36, 0, %2	\n\t"	// y0, y1

  205.        "stxvd2x		37, %11, %2	\n\t"	// y2, y3

  206. 

  207.        "lxvd2x		46, 0, %6	\n\t"	// a3[0], a3[1]

  208.        "lxvd2x		47, %11, %6	\n\t"	// a3[2], a3[3]

  209. 

  210.        "addi		%6, %6, 32	\n\t"

  211.        "addi		%2, %2, 32	\n\t"

  212. 

  213.        "addic.		%1, %1, -4	\n\t"

  214.        "ble		2f		\n\t"

  215. 

  216. 

  217.        "lxvd2x		36, 0, %2	\n\t"	// y0, y1

  218.        "lxvd2x		37, %11, %2	\n\t"	// y2, y3

  219. 

  220.        "xvmaddadp 	36, 40, 32	\n\t"

  221.        "xvmaddadp 	37, 41, 32	\n\t"

  222. 

  223.        "lxvd2x		40, 0, %3	\n\t"	// a0[0], a0[1]

  224.        "lxvd2x		41, %11, %3	\n\t"	// a0[2], a0[3]

  225. 

  226.        "xvmaddadp 	36, 42, 33	\n\t"

  227.        "addi		%3, %3, 32	\n\t"

  228.        "xvmaddadp 	37, 43, 33	\n\t"

  229. 

  230.        "lxvd2x		42, 0, %4	\n\t"	// a1[0], a1[1]

  231.        "lxvd2x		43, %11, %4	\n\t"	// a1[2], a1[3]

  232. 

  233.        "xvmaddadp 	36, 44, 34	\n\t"

  234.        "addi		%4, %4, 32	\n\t"

  235.        "xvmaddadp 	37, 45, 34	\n\t"

  236. 

  237.        "lxvd2x		44, 0, %5	\n\t"	// a2[0], a2[1]

  238.        "lxvd2x		45, %11, %5	\n\t"	// a2[2], a2[3]

  239. 

  240.        "xvmaddadp 	36, 46, 35	\n\t"

  241.        "addi		%5, %5, 32	\n\t"

  242.        "xvmaddadp 	37, 47, 35	\n\t"

  243. 

  244.        "stxvd2x		36, 0, %2	\n\t"	// y0, y1

  245.        "stxvd2x		37, %11, %2	\n\t"	// y2, y3

  246. 

  247.        "lxvd2x		46, 0, %6	\n\t"	// a3[0], a3[1]

  248.        "lxvd2x		47, %11, %6	\n\t"	// a3[2], a3[3]

  249. 

  250.        "addi		%6, %6, 32	\n\t"

  251.        "addi		%2, %2, 32	\n\t"

  252. 

  253.        "addic.		%1, %1, -4	\n\t"

  254.        "bgt		1b		\n"

  255. 

  256.      "2:				\n\t"

  257. 

  258.        "lxvd2x		36, 0, %2	\n\t"	// y0, y1

  259.        "lxvd2x		37, %11, %2	\n\t"	// y2, y3

  260. 

  261.        "xvmaddadp 	36, 40, 32	\n\t"

  262.        "xvmaddadp 	37, 41, 32	\n\t"

  263. 

  264.        "xvmaddadp 	36, 42, 33	\n\t"

  265.        "xvmaddadp 	37, 43, 33	\n\t"

  266. 

  267.        "xvmaddadp 	36, 44, 34	\n\t"

  268.        "xvmaddadp 	37, 45, 34	\n\t"

  269. 

  270.        "xvmaddadp 	36, 46, 35	\n\t"

  271.        "xvmaddadp 	37, 47, 35	\n\t"

  272. 

  273.        "stxvd2x		36, 0, %2	\n\t"	// y0, y1

  274.        "stxvd2x		37, %11, %2	\n"	// y2, y3

  275. 

  276.      "#n=%1 ap=%8=%12 lda=%13 x=%7=%10 y=%0=%2 alpha=%9 o16=%11\n"

  277.      "#a0=%3 a1=%4 a2=%5 a3=%6"

  278.      :

  279.        "+m" (*y),

  280.        "+r" (n),	// 1

  281.        "+b" (y),	// 2

  282.        "=b" (a0),	// 3

  283.        "=b" (a1),	// 4

  284.        "=&b" (a2),	// 5

  285.        "=&b" (a3)	// 6

  286.      :

  287.        "m" (*x),

  288.        "m" (*ap),

  289.        "d" (alpha),	// 9

  290.        "r" (x),		// 10

  291.        "b" (16),	// 11

  292.        "3" (ap),	// 12

  293.        "4" (lda)	// 13

  294.      :

  295.        "cr0",

  296.        "vs32","vs33","vs34","vs35","vs36","vs37",

  297.        "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47"

  298.      );

  299. }