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/power/dgemv_n_microk_power10.c

481 lines
15 kB
Raw Permalink 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. #define HAVE_KERNEL_4x4 1

  29. 

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

  31. {

  32.   double *a0;

  33.   double *a1;

  34.   double *a2;

  35.   double *a3;

  36. 

  37.   __asm__

  38.     (

  39.        "lxvp	40, 0(%10)	\n\t"	// x0, x1

  40.        XXSPLTD_S(32,%x9,0)	// alpha, alpha

  41. 

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

  43. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

  44.        "xvmuldp     34, 40, 32  \n\t"   // x0 * alpha, x1 * alpha

  45.        "xvmuldp     35, 41, 32  \n\t"	// x2 * alpha, x3 * alpha

  46. #else

  47.        "xvmuldp		34, 41, 32	\n\t"	// x0 * alpha, x1 * alpha

  48.        "xvmuldp		35, 40, 32	\n\t"	// x2 * alpha, x3 * alpha

  49. #endif

  50. 

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

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

  53. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

  54.        XXSPLTD_S(32,34,0)   // x0 * alpha, x0 * alpha

  55.        XXSPLTD_S(33,34,1)   // x1 * alpha, x1 * alpha

  56.        XXSPLTD_S(34,35,0)   // x2 * alpha, x2 * alpha

  57.        XXSPLTD_S(35,35,1)   // x3 * alpha, x3 * alpha

  58. #else

  59.        XXSPLTD_S(32,34,1)	// x0 * alpha, x0 * alpha

  60.        XXSPLTD_S(33,34,0)	// x1 * alpha, x1 * alpha

  61.        XXSPLTD_S(34,35,1)	// x2 * alpha, x2 * alpha

  62.        XXSPLTD_S(35,35,0)	// x3 * alpha, x3 * alpha

  63. #endif

  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.        "lxvp		40, 0(%3)	\n\t"	// a0[0], a0[1]

  73. 

  74.        "lxvp		42, 0(%4)	\n\t"	// a1[0], a1[1]

  75. 

  76.        "lxvp		44, 0(%5)	\n\t"	// a2[0], a2[1]

  77. 

  78.        "lxvp		46, 0(%6)	\n\t"	// a3[0], a3[1]

  79. 

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

  81. 

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

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

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

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

  86. 

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

  88.        "ble		two%=		\n\t"

  89. 

  90.        ".align	5		\n"

  91.      "one%=:				\n\t"

  92. 

  93.        "lxvp		36, 0(%2)	\n\t"	// y0, y1

  94. 

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

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

  97. 

  98.        "lxvp		40, 0(%3)	\n\t"	// a0[0], a0[1]

  99. 

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

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

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

  103. 

  104.        "lxvp		42, 0(%4)	\n\t"	// a1[0], a1[1]

  105. 

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

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

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

  109. 

  110.        "lxvp		44, 0(%5)	\n\t"	// a2[0], a2[1]

  111. 

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

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

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

  115. 

  116.        "stxvp		36, 0(%2)	\n\t"	// y0, y1

  117. 

  118.        "lxvp		46, 0(%6)	\n\t"	// a3[0], a3[1]

  119. 

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

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

  122. 

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

  124.        "ble		two%=		\n\t"

  125. 

  126. 

  127.        "lxvp		36, 0(%2)	\n\t"	// y0, y1

  128. 

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

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

  131. 

  132.        "lxvp		40, 0(%3)	\n\t"	// a0[0], a0[1]

  133. 

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

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

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

  137. 

  138.        "lxvp		42, 0(%4)	\n\t"	// a1[0], a1[1]

  139. 

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

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

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

  143. 

  144.        "lxvp		44, 0(%5)	\n\t"	// a2[0], a2[1]

  145. 

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

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

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

  149. 

  150.        "stxvp		36, 0(%2)	\n\t"	// y0, y1

  151. 

  152.        "lxvp	46, 0(%6)	\n\t"	// a3[0], a3[1]

  153. 

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

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

  156. 

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

  158.        "ble		two%=		\n\t"

  159. 

  160. 

  161.        "lxvp		36, 0(%2)	\n\t"	// y0, y1

  162. 

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

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

  165. 

  166.        "lxvp		40, 0(%3)	\n\t"	// a0[0], a0[1]

  167. 

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

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

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

  171. 

  172.        "lxvp		42, 0(%4)	\n\t"	// a1[0], a1[1]

  173. 

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

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

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

  177. 

  178.        "lxvp		44, 0(%5)	\n\t"	// a2[0], a2[1]

  179. 

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

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

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

  183. 

  184.        "stxvp		36, 0(%2)	\n\t"	// y0, y1

  185. 

  186.        "lxvp		46, 0(%6)	\n\t"	// a3[0], a3[1]

  187. 

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

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

  190. 

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

  192.        "ble		two%=		\n\t"

  193. 

  194. 

  195.        "lxvp		36, 0(%2)	\n\t"	// y0, y1

  196. 

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

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

  199. 

  200.        "lxvp		40, 0(%3)	\n\t"	// a0[0], a0[1]

  201. 

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

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

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

  205. 

  206.        "lxvp		42, 0(%4)	\n\t"	// a1[0], a1[1]

  207. 

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

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

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

  211. 

  212.        "lxvp		44, 0(%5)	\n\t"	// a2[0], a2[1]

  213. 

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

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

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

  217. 

  218.        "stxvp		36, 0(%2)	\n\t"	// y0, y1

  219. 

  220.        "lxvp		46, 0(%6)	\n\t"	// a3[0], a3[1]

  221. 

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

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

  224. 

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

  226.        "bgt		one%=		\n"

  227. 

  228.      "two%=:				\n\t"

  229. 

  230.        "lxvp		36, 0(%2)	\n\t"	// y0, y1

  231. 

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

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

  234. 

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

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

  237. 

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

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

  240. 

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

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

  243. 

  244.        "stxvp		36, 0(%2)	\n\t"	// y0, y1

  245. 

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

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

  248.      :

  249.        "+m" (*y),

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

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

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

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

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

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

  256.      :

  257.        "m" (*x),

  258.        "m" (*ap),

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

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

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

  262.        "3" (ap),	// 12

  263.        "4" (lda)	// 13

  264.      :

  265.        "cr0",

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

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

  268.      );

  269. }

  270. static void dgemv_kernel_4x8 (long n, double *ap, long lda, double *x, double *y, double alpha)

  271. {

  272. 

  273.   double *a0;

  274.   double *a1;

  275.   double *a2;

  276.   double *a3;

  277.   double *a4;

  278.   double *a5;

  279.   double *a6;

  280.   double *a7;

  281.   long tmp;

  282.   __asm__

  283.     (

  284.        "lxvp		34, 0( %15)	\n\t"	// x0, x1

  285.        "lxvp		38, 32( %15)	\n\t"	// x4, x5

  286. 

  287.        XXSPLTD_S(58,%x14,0)	// alpha, alpha

  288.        "sldi		%10, %17, 3	\n\t"	// lda * sizeof (double)

  289.        "xvmuldp         34, 34, 58      \n\t"   // x0 * alpha, x1 * alpha

  290.        "xvmuldp         35, 35, 58      \n\t"   // x2 * alpha, x3 * alpha

  291.        "xvmuldp         38, 38, 58      \n\t"   // x4 * alpha, x5 * alpha

  292.        "xvmuldp         39, 39, 58      \n\t"   // x6 * alpha, x7 * alpha

  293. 

  294.        "li		%11, 32   \n\t"

  295. 

  296.        "add		%4, %3, %10	\n\t"	// a0 = ap, a1 = a0 + lda

  297.        "add		%10, %10, %10	\n\t"	// 2 * lda

  298. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

  299.        XXSPLTD_S(32,34,0)       // x0 * alpha, x0 * alpha

  300.        XXSPLTD_S(33,34,1)       // x1 * alpha, x1 * alpha

  301.        XXSPLTD_S(34,35,0)       // x2 * alpha, x2 * alpha

  302.        XXSPLTD_S(35,35,1)       // x3 * alpha, x3 * alpha

  303.        XXSPLTD_S(48,39,0)       // x6 * alpha, x6 * alpha

  304.        XXSPLTD_S(49,39,1)       // x7 * alpha, x7 * alpha

  305.        XXSPLTD_S(39,38,1)       // x5 * alpha, x5 * alpha

  306.        XXSPLTD_S(38,38,0)       // x4 * alpha, x4 * alpha

  307. #else

  308.        XXSPLTD_S(32,34,1)       // x0 * alpha, x0 * alpha

  309.        XXSPLTD_S(33,34,0)       // x1 * alpha, x1 * alpha

  310.        XXSPLTD_S(34,35,1)       // x2 * alpha, x2 * alpha

  311.        XXSPLTD_S(35,35,0)       // x3 * alpha, x3 * alpha

  312.        XXSPLTD_S(48,39,1)       // x6 * alpha, x6 * alpha

  313.        XXSPLTD_S(49,39,0)       // x7 * alpha, x7 * alpha

  314.        XXSPLTD_S(39,38,0)       // x5 * alpha, x5 * alpha

  315.        XXSPLTD_S(38,38,1)       // x4 * alpha, x4 * alpha

  316. #endif

  317. 

  318.        "add		%5, %3, %10	\n\t"	// a2 = a0 + 2 * lda

  319.        "add		%6, %4, %10	\n\t"	// a3 = a1 + 2 * lda

  320.        "add		%7, %5, %10	\n\t"	// a4 = a2 + 2 * lda

  321.        "add		%8, %6, %10	\n\t"	// a5 = a3 + 2 * lda

  322.        "add		%9, %7, %10	\n\t"	// a6 = a4 + 2 * lda

  323.        "add		%10, %8, %10	\n\t"	// a7 = a5 + 2 * lda

  324. 

  325.        "lxvp		40, 0( %3)	\n\t"	// a0[0], a0[1]

  326.        "lxvp		42, 0( %4)	\n\t"	// a1[0], a1[1]

  327.        "lxvp		44, 0( %5)	\n\t"	// a2[0], a2[1]

  328.        "lxvp		46, 0( %6)	\n\t"	// a3[0], a3[1]

  329.        "lxvp		50, 0( %7)	\n\t"	// a4[0]

  330.        "lxvp		52, 0( %8)	\n\t"	// a5[0]

  331.        "lxvp		54, 0( %9)	\n\t"	// a6[0]

  332.        "lxvp		56, 0( %10)	\n\t"	// a7[0]

  333. 

  334. 

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

  336.        "ble		two%=		\n\t"

  337. 

  338.        ".align	5		\n"

  339.      "one%=:				\n\t"

  340. 

  341.        "lxvp		36, 0( %2)	\n\t"	// y0, y1

  342. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

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

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

  345. #else

  346.        "xvmaddadp       36, 40, 34      \n\t"

  347.        "xvmaddadp       37, 41, 34      \n\t"

  348. #endif

  349.        "lxvpx		40, %3, %11	\n\t"	// a0[0], a0[1]

  350. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

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

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

  353. #else

  354.        "xvmaddadp       36, 42, 35      \n\t"

  355.        "xvmaddadp       37, 43, 35      \n\t"

  356. #endif

  357.        "lxvpx		42, %4, %11	\n\t"	// a1[0], a1[1]

  358. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

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

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

  361. #else

  362.        "xvmaddadp       36, 44, 32      \n\t"

  363.        "xvmaddadp       37, 45, 32      \n\t"

  364. #endif

  365.        "lxvpx		44, %5, %11	\n\t"	// a2[0], a2[1]

  366. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

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

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

  369. #else

  370.        "xvmaddadp       36, 46, 33      \n\t"

  371.        "xvmaddadp       37, 47, 33      \n\t"

  372. #endif

  373.        "lxvpx		46, %6, %11	\n\t"	// a3[0], a3[1]

  374. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

  375.        "xvmaddadp       36, 50, 38      \n\t"

  376.        "xvmaddadp       37, 51, 38      \n\t"

  377. #else

  378.        "xvmaddadp       36, 50, 48      \n\t"

  379.        "xvmaddadp       37, 51, 48      \n\t"

  380. #endif

  381.        "lxvpx		50, %7, %11	\n\t"	// a4[0]

  382. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

  383.        "xvmaddadp       36, 52, 39      \n\t"

  384.        "xvmaddadp       37, 53, 39      \n\t"

  385. #else

  386.        "xvmaddadp       36, 52, 49      \n\t"

  387.        "xvmaddadp       37, 53, 49      \n\t"

  388. #endif

  389.        "lxvpx		52, %8, %11	\n\t"	// a5[0]

  390. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

  391.        "xvmaddadp       36, 54, 48      \n\t"

  392.        "xvmaddadp       37, 55, 48      \n\t"

  393. #else

  394.        "xvmaddadp       36, 54, 38      \n\t"

  395.        "xvmaddadp       37, 55, 38      \n\t"

  396. #endif

  397.        "lxvpx		54, %9, %11	\n\t"	// a6[0]

  398. #if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

  399.        "xvmaddadp       36, 56, 49      \n\t"

  400.        "xvmaddadp       37, 57, 49      \n\t"

  401. #else

  402.        "xvmaddadp       36, 56, 39      \n\t"

  403.        "xvmaddadp       37, 57, 39      \n\t"

  404. #endif

  405.        "lxvpx		56, %10, %11	\n\t"	// a7[0]

  406.        "addi		%11, %11, 32    \n\t"

  407. 

  408.        "stxvp		36, 0( %2)	\n\t"	// y0, y1

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

  410. 

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

  412.        "bgt		one%=		\n"

  413. 

  414.      "two%=:				\n\t"

  415. 

  416.        "lxvp		36, 0( %2)	\n\t"	// y0, y1

  417. #if  (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)

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

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

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

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

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

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

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

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

  426.        "xvmaddadp       36, 50, 38      \n\t"

  427.        "xvmaddadp       37, 51, 38      \n\t"

  428.        "xvmaddadp       36, 52, 39      \n\t"

  429.        "xvmaddadp       37, 53, 39      \n\t"

  430.        "xvmaddadp       36, 54, 48      \n\t"

  431.        "xvmaddadp       37, 55, 48      \n\t"

  432.        "xvmaddadp       36, 56, 49      \n\t"

  433.        "xvmaddadp       37, 57, 49      \n\t"

  434. #else

  435.        "xvmaddadp       36, 40, 34      \n\t"

  436.        "xvmaddadp       37, 41, 34      \n\t"

  437.        "xvmaddadp       36, 42, 35      \n\t"

  438.        "xvmaddadp       37, 43, 35      \n\t"

  439.        "xvmaddadp       36, 44, 32      \n\t"

  440.        "xvmaddadp       37, 45, 32      \n\t"

  441.        "xvmaddadp       36, 46, 33      \n\t"

  442.        "xvmaddadp       37, 47, 33      \n\t"

  443.        "xvmaddadp       36, 50, 48      \n\t"

  444.        "xvmaddadp       37, 51, 48      \n\t"

  445.        "xvmaddadp       36, 52, 49      \n\t"

  446.        "xvmaddadp       37, 53, 49      \n\t"

  447.        "xvmaddadp       36, 54, 38      \n\t"

  448.        "xvmaddadp       37, 55, 38      \n\t"

  449.        "xvmaddadp       36, 56, 39      \n\t"

  450.        "xvmaddadp       37, 57, 39      \n\t"

  451. #endif

  452.        "stxvp		36, 0( %2)	\n\t"	// y0, y1

  453. 

  454.      :

  455.        "+m" (*y),

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

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

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

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

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

  461.        "=&b" (a3),	// 6

  462.        "=&b" (a4),	// 7

  463.        "=&b" (a5),	// 8

  464.        "=&b" (a6),	// 9

  465.        "=&b" (a7),	// 10

  466.        "=b" (tmp)

  467.      :

  468.        "m" (*x),

  469.        "m" (*ap),

  470.        "d" (alpha),	// 14

  471.        "r" (x),		// 15

  472.        "3" (ap),	// 16

  473.        "4" (lda)	// 17

  474.      :

  475.        "cr0",

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

  477.        "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47", "vs48",

  478.        "vs49","vs50","vs51","vs52","vs53","vs54","vs55","vs56", "vs57", "vs58"

  479.      );

  480. }