falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
42 Commits
32 Branches
444MB
Tree: fe97c87d65
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'fe97c87d65'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx.c

1072 lines
26KB
Raw Blame History 

  1. /*

  2.  * MMX optimized DSP utils

  3.  * Copyright (c) 2000, 2001 Gerard Lantau.

  4.  *

  5.  * This program is free software; you can redistribute it and/or modify

  6.  * it under the terms of the GNU General Public License as published by

  7.  * the Free Software Foundation; either version 2 of the License, or

  8.  * (at your option) any later version.

  9.  *

  10.  * This program is distributed in the hope that it will be useful,

  11.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  12.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

  13.  * GNU General Public License for more details.

  14.  *

  15.  * You should have received a copy of the GNU General Public License

  16.  * along with this program; if not, write to the Free Software

  17.  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

  18.  *

  19.  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>

  20.  */

  21. 

  22. #include "../dsputil.h"

  23. 

  24. int mm_flags; /* multimedia extension flags */

  25. 

  26. int pix_abs16x16_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);

  27. int pix_abs16x16_sse(UINT8 *blk1, UINT8 *blk2, int lx, int h);

  28. int pix_abs16x16_x2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);

  29. int pix_abs16x16_y2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);

  30. int pix_abs16x16_xy2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);

  31. 

  32. #ifdef USE_MMX_IDCT

  33. /* external functions, defined in libmpeg2 */

  34. void mmx_idct(DCTELEM *block);

  35. void mmxext_idct(DCTELEM *block);

  36. /* this should be in dsputil.h?  -- A'rpi   */

  37. extern UINT8 ff_alternate_horizontal_scan[64];

  38. extern UINT8 ff_alternate_vertical_scan[64];

  39. extern UINT8 zigzag_direct[64];

  40. #endif

  41. 

  42. /* pixel operations */

  43. static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001;

  44. static const unsigned long long int mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002;

  45. //static const unsigned short mm_wone[4] __attribute__ ((aligned(8))) = { 0x1, 0x1, 0x1, 0x1 };

  46. //static const unsigned short mm_wtwo[4] __attribute__ ((aligned(8))) = { 0x2, 0x2, 0x2, 0x2 };

  47. 

  48. /***********************************/

  49. /* 3Dnow specific */

  50. 

  51. #define DEF(x) x ## _3dnow

  52. /* for Athlons PAVGUSB is prefered */

  53. #define PAVGB "pavgusb"

  54. 

  55. #include "dsputil_mmx_avg.h"

  56. 

  57. #undef DEF

  58. #undef PAVGB

  59. 

  60. /***********************************/

  61. /* MMX2 specific */

  62. 

  63. #define DEF(x) x ## _sse

  64. 

  65. /* Introduced only in MMX2 set */

  66. #define PAVGB "pavgb"

  67. 

  68. #include "dsputil_mmx_avg.h"

  69. 

  70. #undef DEF

  71. #undef PAVGB

  72. 

  73. /***********************************/

  74. /* standard MMX */

  75. 

  76. static void get_pixels_mmx(DCTELEM *block, const UINT8 *pixels, int line_size)

  77. {

  78.     DCTELEM *p;

  79.     const UINT8 *pix;

  80.     int i;

  81. 

  82.     /* read the pixels */

  83.     p = block;

  84.     pix = pixels;

  85.     __asm __volatile("pxor %%mm7, %%mm7":::"memory");

  86.     for(i=0;i<4;i++) {

  87. 	__asm __volatile(

  88. 		"movq	%1, %%mm0\n\t"

  89. 		"movq	%2, %%mm1\n\t"

  90. 		"movq	%%mm0, %%mm2\n\t"

  91. 		"movq	%%mm1, %%mm3\n\t"

  92. 		"punpcklbw %%mm7, %%mm0\n\t"

  93. 		"punpckhbw %%mm7, %%mm2\n\t"

  94. 		"punpcklbw %%mm7, %%mm1\n\t"

  95. 		"punpckhbw %%mm7, %%mm3\n\t"

  96. 		"movq	%%mm0, %0\n\t"

  97. 		"movq	%%mm2, 8%0\n\t"

  98. 		"movq	%%mm1, 16%0\n\t"

  99. 		"movq	%%mm3, 24%0\n\t"

  100. 		:"=m"(*p)

  101. 		:"m"(*pix), "m"(*(pix+line_size))

  102. 		:"memory");

  103.         pix += line_size*2;

  104.         p += 16;

  105.     }

  106. }

  107. 

  108. static void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size)

  109. {

  110.     const DCTELEM *p;

  111.     UINT8 *pix;

  112.     int i;

  113. 

  114.     /* read the pixels */

  115.     p = block;

  116.     pix = pixels;

  117.     for(i=0;i<2;i++) {

  118. 	__asm __volatile(

  119. 		"movq	%4, %%mm0\n\t"

  120. 		"movq	8%4, %%mm1\n\t"

  121. 		"movq	16%4, %%mm2\n\t"

  122. 		"movq	24%4, %%mm3\n\t"

  123. 		"movq	32%4, %%mm4\n\t"

  124. 		"movq	40%4, %%mm5\n\t"

  125. 		"movq	48%4, %%mm6\n\t"

  126. 		"movq	56%4, %%mm7\n\t"

  127. 		"packuswb %%mm1, %%mm0\n\t"

  128. 		"packuswb %%mm3, %%mm2\n\t"

  129. 		"packuswb %%mm5, %%mm4\n\t"

  130. 		"packuswb %%mm7, %%mm6\n\t"

  131. 		"movq	%%mm0, %0\n\t"

  132. 		"movq	%%mm2, %1\n\t"

  133. 		"movq	%%mm4, %2\n\t"

  134. 		"movq	%%mm6, %3\n\t"

  135. 		:"=m"(*pix), "=m"(*(pix+line_size))

  136. 		,"=m"(*(pix+line_size*2)), "=m"(*(pix+line_size*3))

  137. 		:"m"(*p)

  138. 		:"memory");

  139.         pix += line_size*4;

  140.         p += 32;

  141.     }

  142. }

  143. 

  144. static void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size)

  145. {

  146.     const DCTELEM *p;

  147.     UINT8 *pix;

  148.     int i;

  149. 

  150.     /* read the pixels */

  151.     p = block;

  152.     pix = pixels;

  153. 	__asm __volatile("pxor	%%mm7, %%mm7":::"memory");

  154.     for(i=0;i<4;i++) {

  155. 	__asm __volatile(

  156. 		"movq	%2, %%mm0\n\t"

  157. 		"movq	8%2, %%mm1\n\t"

  158. 		"movq	16%2, %%mm2\n\t"

  159. 		"movq	24%2, %%mm3\n\t"

  160. 		"movq	%0, %%mm4\n\t"

  161. 		"movq	%1, %%mm6\n\t"

  162. 		"movq	%%mm4, %%mm5\n\t"

  163. 		"punpcklbw %%mm7, %%mm4\n\t"

  164. 		"punpckhbw %%mm7, %%mm5\n\t"

  165. 		"paddsw	%%mm4, %%mm0\n\t"

  166. 		"paddsw	%%mm5, %%mm1\n\t"

  167. 		"movq	%%mm6, %%mm5\n\t"

  168. 		"punpcklbw %%mm7, %%mm6\n\t"

  169. 		"punpckhbw %%mm7, %%mm5\n\t"

  170. 		"paddsw	%%mm6, %%mm2\n\t"

  171. 		"paddsw	%%mm5, %%mm3\n\t"

  172. 		"packuswb %%mm1, %%mm0\n\t"

  173. 		"packuswb %%mm3, %%mm2\n\t"

  174. 		"movq	%%mm0, %0\n\t"

  175. 		"movq	%%mm2, %1\n\t"

  176. 		:"=m"(*pix), "=m"(*(pix+line_size))

  177. 		:"m"(*p)

  178. 		:"memory");

  179.         pix += line_size*2;

  180.         p += 16;

  181.     }

  182. }

  183. 

  184. static void put_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  185. {

  186.     int dh, hh;

  187.     UINT8 *p;

  188.     const UINT8 *pix;

  189.     p   = block;

  190.     pix = pixels;

  191.     hh=h>>2;

  192.     dh=h&3;

  193.     while(hh--) {

  194.     __asm __volatile(

  195. 	"movq	%4, %%mm0\n\t"

  196. 	"movq	%5, %%mm1\n\t"

  197. 	"movq	%6, %%mm2\n\t"

  198. 	"movq	%7, %%mm3\n\t"

  199. 	"movq	%%mm0, %0\n\t"

  200. 	"movq	%%mm1, %1\n\t"

  201. 	"movq	%%mm2, %2\n\t"

  202. 	"movq	%%mm3, %3\n\t"

  203. 	:"=m"(*p), "=m"(*(p+line_size)), "=m"(*(p+line_size*2)), "=m"(*(p+line_size*3))

  204. 	:"m"(*pix), "m"(*(pix+line_size)), "m"(*(pix+line_size*2)), "m"(*(pix+line_size*3))

  205. 	:"memory");

  206.         pix = pix + line_size*4;

  207.         p =   p   + line_size*4;

  208.     }

  209.     while(dh--) {

  210.      __asm __volatile(

  211. 	"movq	%1, %%mm0\n\t"

  212. 	"movq	%%mm0, %0\n\t"

  213. 	:"=m"(*p)

  214. 	:"m"(*pix)

  215. 	:"memory");

  216.         pix = pix + line_size;

  217.         p =   p   + line_size;

  218.     }

  219. }

  220. 

  221. static void put_pixels_x2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  222. {

  223.   UINT8 *p;

  224.   const UINT8 *pix;

  225.   p = block;

  226.   pix = pixels;

  227.   __asm __volatile(

  228. 	"pxor	%%mm7, %%mm7\n\t"

  229. 	"movq	%0, %%mm4\n\t"

  230. 	::"m"(mm_wone):"memory");

  231.   do {

  232.     __asm __volatile(

  233. 	"movq	%1, %%mm0\n\t"

  234. 	"movq	1%1, %%mm1\n\t"

  235. 	"movq	%%mm0, %%mm2\n\t"

  236. 	"movq	%%mm1, %%mm3\n\t"

  237. 	"punpcklbw %%mm7, %%mm0\n\t"

  238. 	"punpcklbw %%mm7, %%mm1\n\t"

  239. 	"punpckhbw %%mm7, %%mm2\n\t"

  240. 	"punpckhbw %%mm7, %%mm3\n\t"

  241. 	"paddusw %%mm1, %%mm0\n\t"

  242. 	"paddusw %%mm3, %%mm2\n\t"

  243. 	"paddusw %%mm4, %%mm0\n\t"

  244. 	"paddusw %%mm4, %%mm2\n\t"

  245. 	"psrlw	$1, %%mm0\n\t"

  246. 	"psrlw	$1, %%mm2\n\t"

  247. 	"packuswb  %%mm2, %%mm0\n\t"

  248. 	"movq	%%mm0, %0\n\t"

  249. 	:"=m"(*p)

  250. 	:"m"(*pix)

  251. 	:"memory");

  252.    pix += line_size; p += line_size;

  253.   } while (--h);

  254. }

  255. 

  256. static void put_pixels_y2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  257. {

  258.   UINT8 *p;

  259.   const UINT8 *pix;

  260.   p = block;

  261.   pix = pixels;

  262.   __asm __volatile(

  263. 	"pxor	%%mm7, %%mm7\n\t"

  264. 	"movq	%0, %%mm4\n\t"

  265. 	::"m"(mm_wone):"memory");

  266.   do {

  267.     __asm __volatile(

  268. 	"movq	%1, %%mm0\n\t"

  269. 	"movq	%2, %%mm1\n\t"

  270. 	"movq	%%mm0, %%mm2\n\t"

  271. 	"movq	%%mm1, %%mm3\n\t"

  272. 	"punpcklbw %%mm7, %%mm0\n\t"

  273. 	"punpcklbw %%mm7, %%mm1\n\t"

  274. 	"punpckhbw %%mm7, %%mm2\n\t"

  275. 	"punpckhbw %%mm7, %%mm3\n\t"

  276. 	"paddusw %%mm1, %%mm0\n\t"

  277. 	"paddusw %%mm3, %%mm2\n\t"

  278. 	"paddusw %%mm4, %%mm0\n\t"

  279. 	"paddusw %%mm4, %%mm2\n\t"

  280. 	"psrlw	$1, %%mm0\n\t"

  281. 	"psrlw	$1, %%mm2\n\t"

  282. 	"packuswb  %%mm2, %%mm0\n\t"

  283. 	"movq	%%mm0, %0\n\t"

  284. 	:"=m"(*p)

  285. 	:"m"(*pix),

  286. 	 "m"(*(pix+line_size))

  287. 	:"memory");

  288.    pix += line_size;

  289.    p += line_size;

  290.   } while (--h);

  291. }

  292. 

  293. static void put_pixels_xy2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  294. {

  295.   UINT8 *p;

  296.   const UINT8 *pix;

  297.   p = block;

  298.   pix = pixels;

  299.   __asm __volatile(

  300. 	"pxor	%%mm7, %%mm7\n\t"

  301. 	"movq	%0, %%mm6\n\t"

  302. 	::"m"(mm_wtwo):"memory");

  303.   do {

  304.     __asm __volatile(

  305. 	"movq	%1, %%mm0\n\t"

  306. 	"movq	%2, %%mm1\n\t"

  307. 	"movq	1%1, %%mm4\n\t"

  308. 	"movq	1%2, %%mm5\n\t"

  309. 	"movq	%%mm0, %%mm2\n\t"

  310. 	"movq	%%mm1, %%mm3\n\t"

  311. 	"punpcklbw %%mm7, %%mm0\n\t"

  312. 	"punpcklbw %%mm7, %%mm1\n\t"

  313. 	"punpckhbw %%mm7, %%mm2\n\t"

  314. 	"punpckhbw %%mm7, %%mm3\n\t"

  315. 	"paddusw %%mm1, %%mm0\n\t"

  316. 	"paddusw %%mm3, %%mm2\n\t"

  317. 	"movq	%%mm4, %%mm1\n\t"

  318. 	"movq	%%mm5, %%mm3\n\t"

  319. 	"punpcklbw %%mm7, %%mm4\n\t"

  320. 	"punpcklbw %%mm7, %%mm5\n\t"

  321. 	"punpckhbw %%mm7, %%mm1\n\t"

  322. 	"punpckhbw %%mm7, %%mm3\n\t"

  323. 	"paddusw %%mm5, %%mm4\n\t"

  324. 	"paddusw %%mm3, %%mm1\n\t"

  325. 	"paddusw %%mm6, %%mm4\n\t"

  326. 	"paddusw %%mm6, %%mm1\n\t"

  327. 	"paddusw %%mm4, %%mm0\n\t"

  328. 	"paddusw %%mm1, %%mm2\n\t"

  329. 	"psrlw	$2, %%mm0\n\t"

  330. 	"psrlw	$2, %%mm2\n\t"

  331. 	"packuswb  %%mm2, %%mm0\n\t"

  332. 	"movq	%%mm0, %0\n\t"

  333. 	:"=m"(*p)

  334. 	:"m"(*pix),

  335. 	 "m"(*(pix+line_size))

  336. 	:"memory");

  337.    pix += line_size;

  338.    p += line_size;

  339.   } while(--h);

  340. }

  341. 

  342. static void   put_no_rnd_pixels_x2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  343. {

  344.   UINT8  *p;

  345.   const UINT8 *pix;

  346.   p = block;

  347.   pix = pixels;

  348.   __asm __volatile("pxor %%mm7, %%mm7\n\t":::"memory");

  349.   do {

  350.     __asm __volatile(

  351. 	"movq	%1, %%mm0\n\t"

  352. 	"movq	1%1, %%mm1\n\t"

  353. 	"movq	%%mm0, %%mm2\n\t"

  354. 	"movq	%%mm1, %%mm3\n\t"

  355. 	"punpcklbw %%mm7, %%mm0\n\t"

  356. 	"punpcklbw %%mm7, %%mm1\n\t"

  357. 	"punpckhbw %%mm7, %%mm2\n\t"

  358. 	"punpckhbw %%mm7, %%mm3\n\t"

  359. 	"paddusw %%mm1, %%mm0\n\t"

  360. 	"paddusw %%mm3, %%mm2\n\t"

  361. 	"psrlw	$1, %%mm0\n\t"

  362. 	"psrlw	$1, %%mm2\n\t"

  363. 	"packuswb  %%mm2, %%mm0\n\t"

  364. 	"movq	%%mm0, %0\n\t"

  365. 	:"=m"(*p)

  366. 	:"m"(*pix)

  367. 	:"memory");

  368.    pix += line_size;

  369.    p +=   line_size;

  370.   } while (--h);

  371. }

  372. 

  373. static void put_no_rnd_pixels_y2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  374. {

  375.   UINT8  *p;

  376.   const UINT8 *pix;

  377.   p = block;

  378.   pix = pixels;

  379.   __asm __volatile("pxor %%mm7, %%mm7\n\t":::"memory");

  380.   do {

  381.     __asm __volatile(

  382. 	"movq	%1, %%mm0\n\t"

  383. 	"movq	%2, %%mm1\n\t"

  384. 	"movq	%%mm0, %%mm2\n\t"

  385. 	"movq	%%mm1, %%mm3\n\t"

  386. 	"punpcklbw %%mm7, %%mm0\n\t"

  387. 	"punpcklbw %%mm7, %%mm1\n\t"

  388. 	"punpckhbw %%mm7, %%mm2\n\t"

  389. 	"punpckhbw %%mm7, %%mm3\n\t"

  390. 	"paddusw %%mm1, %%mm0\n\t"

  391. 	"paddusw %%mm3, %%mm2\n\t"

  392. 	"psrlw	$1, %%mm0\n\t"

  393. 	"psrlw	$1, %%mm2\n\t"

  394. 	"packuswb  %%mm2, %%mm0\n\t"

  395. 	"movq	%%mm0, %0\n\t"

  396. 	:"=m"(*p)

  397. 	:"m"(*pix),

  398. 	 "m"(*(pix+line_size))

  399. 	:"memory");

  400.    pix += line_size;

  401.    p +=   line_size;

  402.   } while(--h);

  403. }

  404. 

  405. static void   put_no_rnd_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  406. {

  407.   UINT8  *p;

  408.   const UINT8 *pix;

  409.   p = block;

  410.   pix = pixels;

  411.   __asm __volatile(

  412. 	"pxor	%%mm7, %%mm7\n\t"

  413. 	"movq	%0, %%mm6\n\t"

  414. 	::"m"(mm_wone):"memory");

  415.   do {

  416.     __asm __volatile(

  417. 	"movq	%1, %%mm0\n\t"

  418. 	"movq	%2, %%mm1\n\t"

  419. 	"movq	1%1, %%mm4\n\t"

  420. 	"movq	1%2, %%mm5\n\t"

  421. 	"movq	%%mm0, %%mm2\n\t"

  422. 	"movq	%%mm1, %%mm3\n\t"

  423. 	"punpcklbw %%mm7, %%mm0\n\t"

  424. 	"punpcklbw %%mm7, %%mm1\n\t"

  425. 	"punpckhbw %%mm7, %%mm2\n\t"

  426. 	"punpckhbw %%mm7, %%mm3\n\t"

  427. 	"paddusw %%mm1, %%mm0\n\t"

  428. 	"paddusw %%mm3, %%mm2\n\t"

  429. 	"movq	%%mm4, %%mm1\n\t"

  430. 	"movq	%%mm5, %%mm3\n\t"

  431. 	"punpcklbw %%mm7, %%mm4\n\t"

  432. 	"punpcklbw %%mm7, %%mm5\n\t"

  433. 	"punpckhbw %%mm7, %%mm1\n\t"

  434. 	"punpckhbw %%mm7, %%mm3\n\t"

  435. 	"paddusw %%mm5, %%mm4\n\t"

  436. 	"paddusw %%mm3, %%mm1\n\t"

  437. 	"paddusw %%mm6, %%mm4\n\t"

  438. 	"paddusw %%mm6, %%mm1\n\t"

  439. 	"paddusw %%mm4, %%mm0\n\t"

  440. 	"paddusw %%mm1, %%mm2\n\t"

  441. 	"psrlw	$2, %%mm0\n\t"

  442. 	"psrlw	$2, %%mm2\n\t"

  443. 	"packuswb  %%mm2, %%mm0\n\t"

  444. 	"movq	%%mm0, %0\n\t"

  445. 	:"=m"(*p)

  446. 	:"m"(*pix),

  447. 	 "m"(*(pix+line_size))

  448. 	:"memory");

  449.    pix += line_size;

  450.    p +=   line_size;

  451.   } while(--h);

  452. }

  453. 

  454. static void avg_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  455. {

  456.   UINT8  *p;

  457.   const UINT8 *pix;

  458.   p = block;

  459.   pix = pixels;

  460.   __asm __volatile(

  461. 	"pxor	%%mm7, %%mm7\n\t"

  462. 	"movq	%0, %%mm6\n\t"

  463. 	::"m"(mm_wone):"memory");

  464.   do {

  465.     __asm __volatile(

  466. 	"movq	%0, %%mm0\n\t"

  467. 	"movq	%1, %%mm1\n\t"

  468. 	"movq	%%mm0, %%mm2\n\t"

  469. 	"movq	%%mm1, %%mm3\n\t"

  470. 	"punpcklbw %%mm7, %%mm0\n\t"

  471. 	"punpcklbw %%mm7, %%mm1\n\t"

  472. 	"punpckhbw %%mm7, %%mm2\n\t"

  473. 	"punpckhbw %%mm7, %%mm3\n\t"

  474. 	"paddusw %%mm1, %%mm0\n\t"

  475. 	"paddusw %%mm3, %%mm2\n\t"

  476. 	"paddusw %%mm6, %%mm0\n\t"

  477. 	"paddusw %%mm6, %%mm2\n\t"

  478. 	"psrlw	$1, %%mm0\n\t"

  479. 	"psrlw	$1, %%mm2\n\t"

  480. 	"packuswb  %%mm2, %%mm0\n\t"

  481. 	"movq	%%mm0, %0\n\t"

  482. 	:"=m"(*p)

  483. 	:"m"(*pix)

  484. 	:"memory");

  485.    pix += line_size;

  486.    p +=   line_size;

  487.   }

  488.   while (--h);

  489. }

  490. 

  491. static void   avg_pixels_x2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  492. {

  493.   UINT8  *p;

  494.   const UINT8 *pix;

  495.   p = block;

  496.   pix = pixels;

  497.   __asm __volatile(

  498. 	"pxor	%%mm7, %%mm7\n\t"

  499. 	"movq	%0, %%mm6\n\t"

  500. 	::"m"(mm_wone):"memory");

  501.   do {

  502.     __asm __volatile(

  503. 	"movq	%1, %%mm1\n\t"

  504. 	"movq	%0, %%mm0\n\t"

  505. 	"movq	1%1, %%mm4\n\t"

  506. 	"movq	%%mm0, %%mm2\n\t"

  507. 	"movq	%%mm1, %%mm3\n\t"

  508. 	"movq	%%mm4, %%mm5\n\t"

  509. 	"punpcklbw %%mm7, %%mm1\n\t"

  510. 	"punpckhbw %%mm7, %%mm3\n\t"

  511. 	"punpcklbw %%mm7, %%mm4\n\t"

  512. 	"punpckhbw %%mm7, %%mm5\n\t"

  513. 	"punpcklbw %%mm7, %%mm0\n\t"

  514. 	"punpckhbw %%mm7, %%mm2\n\t"

  515. 	"paddusw %%mm4, %%mm1\n\t"

  516. 	"paddusw %%mm5, %%mm3\n\t"

  517. 	"paddusw %%mm6, %%mm1\n\t"

  518. 	"paddusw %%mm6, %%mm3\n\t"

  519. 	"psrlw	$1, %%mm1\n\t"

  520. 	"psrlw	$1, %%mm3\n\t"

  521. 	"paddusw %%mm6, %%mm0\n\t"

  522. 	"paddusw %%mm6, %%mm2\n\t"

  523. 	"paddusw %%mm1, %%mm0\n\t"

  524. 	"paddusw %%mm3, %%mm2\n\t"

  525. 	"psrlw	$1, %%mm0\n\t"

  526. 	"psrlw	$1, %%mm2\n\t"

  527. 	"packuswb  %%mm2, %%mm0\n\t"

  528. 	"movq	%%mm0, %0\n\t"

  529. 	:"=m"(*p)

  530. 	:"m"(*pix)

  531. 	:"memory");

  532.    pix += line_size;

  533.    p +=   line_size;

  534.   } while (--h);

  535. }

  536. 

  537. static void   avg_pixels_y2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  538. {

  539.   UINT8  *p;

  540.   const UINT8 *pix;

  541.   p = block;

  542.   pix = pixels;

  543.   __asm __volatile(

  544. 	"pxor	%%mm7, %%mm7\n\t"

  545. 	"movq	%0, %%mm6\n\t"

  546. 	::"m"(mm_wone):"memory");

  547.   do {

  548.     __asm __volatile(

  549. 	"movq	%1, %%mm1\n\t"

  550. 	"movq	%0, %%mm0\n\t"

  551. 	"movq	%2, %%mm4\n\t"

  552. 	"movq	%%mm0, %%mm2\n\t"

  553. 	"movq	%%mm1, %%mm3\n\t"

  554. 	"movq	%%mm4, %%mm5\n\t"

  555. 	"punpcklbw %%mm7, %%mm1\n\t"

  556. 	"punpckhbw %%mm7, %%mm3\n\t"

  557. 	"punpcklbw %%mm7, %%mm4\n\t"

  558. 	"punpckhbw %%mm7, %%mm5\n\t"

  559. 	"punpcklbw %%mm7, %%mm0\n\t"

  560. 	"punpckhbw %%mm7, %%mm2\n\t"

  561. 	"paddusw %%mm4, %%mm1\n\t"

  562. 	"paddusw %%mm5, %%mm3\n\t"

  563. 	"paddusw %%mm6, %%mm1\n\t"

  564. 	"paddusw %%mm6, %%mm3\n\t"

  565. 	"psrlw	$1, %%mm1\n\t"

  566. 	"psrlw	$1, %%mm3\n\t"

  567. 	"paddusw %%mm6, %%mm0\n\t"

  568. 	"paddusw %%mm6, %%mm2\n\t"

  569. 	"paddusw %%mm1, %%mm0\n\t"

  570. 	"paddusw %%mm3, %%mm2\n\t"

  571. 	"psrlw	$1, %%mm0\n\t"

  572. 	"psrlw	$1, %%mm2\n\t"

  573. 	"packuswb  %%mm2, %%mm0\n\t"

  574. 	"movq	%%mm0, %0\n\t"

  575. 	:"=m"(*p)

  576. 	:"m"(*pix), "m"(*(pix+line_size))

  577. 	:"memory");

  578.    pix += line_size;

  579.    p +=   line_size ;

  580.   } while(--h);

  581. }

  582. 

  583. static void   avg_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  584. {

  585.   UINT8  *p;

  586.   const UINT8 *pix;

  587.   p = block;

  588.   pix = pixels;

  589.   __asm __volatile(

  590. 	"pxor	%%mm7, %%mm7\n\t"

  591. 	"movq	%0, %%mm6\n\t"

  592. 	::"m"(mm_wtwo):"memory");

  593.   do {

  594.     __asm __volatile(

  595. 	"movq	%1, %%mm0\n\t"

  596. 	"movq	%2, %%mm1\n\t"

  597. 	"movq	1%1, %%mm4\n\t"

  598. 	"movq	1%2, %%mm5\n\t"

  599. 	"movq	%%mm0, %%mm2\n\t"

  600. 	"movq	%%mm1, %%mm3\n\t"

  601. 	"punpcklbw %%mm7, %%mm0\n\t"

  602. 	"punpcklbw %%mm7, %%mm1\n\t"

  603. 	"punpckhbw %%mm7, %%mm2\n\t"

  604. 	"punpckhbw %%mm7, %%mm3\n\t"

  605. 	"paddusw %%mm1, %%mm0\n\t"

  606. 	"paddusw %%mm3, %%mm2\n\t"

  607. 	"movq	%%mm4, %%mm1\n\t"

  608. 	"movq	%%mm5, %%mm3\n\t"

  609. 	"punpcklbw %%mm7, %%mm4\n\t"

  610. 	"punpcklbw %%mm7, %%mm5\n\t"

  611. 	"punpckhbw %%mm7, %%mm1\n\t"

  612. 	"punpckhbw %%mm7, %%mm3\n\t"

  613. 	"paddusw %%mm5, %%mm4\n\t"

  614. 	"paddusw %%mm3, %%mm1\n\t"

  615. 	"paddusw %%mm6, %%mm4\n\t"

  616. 	"paddusw %%mm6, %%mm1\n\t"

  617. 	"paddusw %%mm4, %%mm0\n\t"

  618. 	"paddusw %%mm1, %%mm2\n\t"

  619. 	"movq	%3, %%mm5\n\t"

  620. 	"psrlw	$2, %%mm0\n\t"

  621. 	"movq	%0, %%mm1\n\t"

  622. 	"psrlw	$2, %%mm2\n\t"

  623. 	"movq	%%mm1, %%mm3\n\t"

  624. 	"punpcklbw %%mm7, %%mm1\n\t"

  625. 	"punpckhbw %%mm7, %%mm3\n\t"

  626. 	"paddusw %%mm1, %%mm0\n\t"

  627. 	"paddusw %%mm3, %%mm2\n\t"

  628. 	"paddusw %%mm5, %%mm0\n\t"

  629. 	"paddusw %%mm5, %%mm2\n\t"

  630. 	"psrlw	$1, %%mm0\n\t"

  631. 	"psrlw	$1, %%mm2\n\t"

  632. 	"packuswb  %%mm2, %%mm0\n\t"

  633. 	"movq	%%mm0, %0\n\t"

  634. 	:"=m"(*p)

  635. 	:"m"(*pix),

  636. 	 "m"(*(pix+line_size)), "m"(mm_wone)

  637. 	:"memory");

  638.    pix += line_size;

  639.    p +=   line_size ;

  640.   } while(--h);

  641. }

  642. 

  643. static void avg_no_rnd_pixels_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  644. {

  645.   UINT8  *p;

  646.   const UINT8 *pix;

  647.   p = block;

  648.   pix = pixels;

  649.   __asm __volatile("pxor %%mm7, %%mm7\n\t":::"memory");

  650.   do {

  651.     __asm __volatile(

  652. 	"movq	%1, %%mm0\n\t"

  653. 	"movq	%0, %%mm1\n\t"

  654. 	"movq	%%mm0, %%mm2\n\t"

  655. 	"movq	%%mm1, %%mm3\n\t"

  656. 	"punpcklbw %%mm7, %%mm0\n\t"

  657. 	"punpcklbw %%mm7, %%mm1\n\t"

  658. 	"punpckhbw %%mm7, %%mm2\n\t"

  659. 	"punpckhbw %%mm7, %%mm3\n\t"

  660. 	"paddusw %%mm1, %%mm0\n\t"

  661. 	"paddusw %%mm3, %%mm2\n\t"

  662. 	"psrlw	$1, %%mm0\n\t"

  663. 	"psrlw	$1, %%mm2\n\t"

  664. 	"packuswb  %%mm2, %%mm0\n\t"

  665. 	"movq	%%mm0, %0\n\t"

  666. 	:"=m"(*p)

  667. 	:"m"(*pix)

  668. 	:"memory");

  669.    pix += line_size;

  670.    p +=   line_size ;

  671.   } while (--h);

  672. }

  673. 

  674. static void   avg_no_rnd_pixels_x2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  675. {

  676.   UINT8  *p;

  677.   const UINT8 *pix;

  678.   p = block;

  679.   pix = pixels;

  680.   __asm __volatile(

  681.       "pxor	%%mm7, %%mm7\n\t":::"memory");

  682.   do {

  683.     __asm __volatile(

  684. 	"movq	%1, %%mm0\n\t"

  685. 	"movq	1%1, %%mm1\n\t"

  686. 	"movq	%0, %%mm4\n\t"

  687. 	"movq	%%mm0, %%mm2\n\t"

  688. 	"movq	%%mm1, %%mm3\n\t"

  689. 	"movq	%%mm4, %%mm5\n\t"

  690. 	"punpcklbw %%mm7, %%mm0\n\t"

  691. 	"punpcklbw %%mm7, %%mm1\n\t"

  692. 	"punpckhbw %%mm7, %%mm2\n\t"

  693. 	"punpckhbw %%mm7, %%mm3\n\t"

  694. 	"punpcklbw %%mm7, %%mm4\n\t"

  695. 	"punpckhbw %%mm7, %%mm5\n\t"

  696. 	"paddusw %%mm1, %%mm0\n\t"

  697. 	"paddusw %%mm3, %%mm2\n\t"

  698. 	"psrlw	$1, %%mm0\n\t"

  699. 	"psrlw	$1, %%mm2\n\t"

  700. 	"paddusw %%mm4, %%mm0\n\t"

  701. 	"paddusw %%mm5, %%mm2\n\t"

  702. 	"psrlw	$1, %%mm0\n\t"

  703. 	"psrlw	$1, %%mm2\n\t"

  704. 	"packuswb  %%mm2, %%mm0\n\t"

  705. 	"movq	%%mm0, %0\n\t"

  706. 	:"=m"(*p)

  707. 	:"m"(*pix)

  708. 	:"memory");

  709.    pix += line_size;

  710.    p +=   line_size;

  711.  } while (--h);

  712. }

  713. 

  714. static void   avg_no_rnd_pixels_y2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  715. {

  716.   UINT8  *p;

  717.   const UINT8 *pix;

  718.   p = block;

  719.   pix = pixels;

  720.   __asm __volatile(

  721.       "pxor	%%mm7, %%mm7\n\t":::"memory");

  722.   do {

  723.     __asm __volatile(

  724. 	"movq	%1, %%mm0\n\t"

  725. 	"movq	%2, %%mm1\n\t"

  726. 	"movq	%0, %%mm4\n\t"

  727. 	"movq	%%mm0, %%mm2\n\t"

  728. 	"movq	%%mm1, %%mm3\n\t"

  729. 	"movq	%%mm4, %%mm5\n\t"

  730. 	"punpcklbw %%mm7, %%mm0\n\t"

  731. 	"punpcklbw %%mm7, %%mm1\n\t"

  732. 	"punpckhbw %%mm7, %%mm2\n\t"

  733. 	"punpckhbw %%mm7, %%mm3\n\t"

  734. 	"punpcklbw %%mm7, %%mm4\n\t"

  735. 	"punpckhbw %%mm7, %%mm5\n\t"

  736. 	"paddusw %%mm1, %%mm0\n\t"

  737. 	"paddusw %%mm3, %%mm2\n\t"

  738. 	"psrlw	$1, %%mm0\n\t"

  739. 	"psrlw	$1, %%mm2\n\t"

  740. 	"paddusw %%mm4, %%mm0\n\t"

  741. 	"paddusw %%mm5, %%mm2\n\t"

  742. 	"psrlw	$1, %%mm0\n\t"

  743. 	"psrlw	$1, %%mm2\n\t"

  744. 	"packuswb  %%mm2, %%mm0\n\t"

  745. 	"movq	%%mm0, %0\n\t"

  746. 	:"=m"(*p)

  747. 	:"m"(*pix), "m"(*(pix+line_size))

  748. 	:"memory");

  749.    pix += line_size;

  750.    p +=   line_size ;

  751.   } while(--h);

  752. }

  753. 

  754. static void   avg_no_rnd_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)

  755. {

  756.   UINT8  *p;

  757.   const UINT8 *pix;

  758.   p = block;

  759.   pix = pixels;

  760.   __asm __volatile(

  761. 	"pxor	%%mm7, %%mm7\n\t"

  762. 	"movq	%0, %%mm6\n\t"

  763. 	::"m"(mm_wone):"memory");

  764.   do {

  765.     __asm __volatile(

  766. 	"movq	%1, %%mm0\n\t"

  767. 	"movq	%2, %%mm1\n\t"

  768. 	"movq	1%1, %%mm4\n\t"

  769. 	"movq	1%2, %%mm5\n\t"

  770. 	"movq	%%mm0, %%mm2\n\t"

  771. 	"movq	%%mm1, %%mm3\n\t"

  772. 	"punpcklbw %%mm7, %%mm0\n\t"

  773. 	"punpcklbw %%mm7, %%mm1\n\t"

  774. 	"punpckhbw %%mm7, %%mm2\n\t"

  775. 	"punpckhbw %%mm7, %%mm3\n\t"

  776. 	"paddusw %%mm1, %%mm0\n\t"

  777. 	"paddusw %%mm3, %%mm2\n\t"

  778. 	"movq	%%mm4, %%mm1\n\t"

  779. 	"movq	%%mm5, %%mm3\n\t"

  780. 	"punpcklbw %%mm7, %%mm4\n\t"

  781. 	"punpcklbw %%mm7, %%mm5\n\t"

  782. 	"punpckhbw %%mm7, %%mm1\n\t"

  783. 	"punpckhbw %%mm7, %%mm3\n\t"

  784. 	"paddusw %%mm5, %%mm4\n\t"

  785. 	"paddusw %%mm3, %%mm1\n\t"

  786. 	"paddusw %%mm6, %%mm4\n\t"

  787. 	"paddusw %%mm6, %%mm1\n\t"

  788. 	"paddusw %%mm4, %%mm0\n\t"

  789. 	"paddusw %%mm1, %%mm2\n\t"

  790. 	"movq	%0, %%mm1\n\t"

  791. 	"psrlw	$2, %%mm0\n\t"

  792. 	"movq	%%mm1, %%mm3\n\t"

  793. 	"psrlw	$2, %%mm2\n\t"

  794. 	"punpcklbw %%mm7, %%mm1\n\t"

  795. 	"punpckhbw %%mm7, %%mm3\n\t"

  796. 	"paddusw %%mm1, %%mm0\n\t"

  797. 	"paddusw %%mm3, %%mm2\n\t"

  798. 	"psrlw	$1, %%mm0\n\t"

  799. 	"psrlw	$1, %%mm2\n\t"

  800. 	"packuswb  %%mm2, %%mm0\n\t"

  801. 	"movq	%%mm0, %0\n\t"

  802. 	:"=m"(*p)

  803. 	:"m"(*pix),

  804. 	 "m"(*(pix+line_size))

  805. 	:"memory");

  806.    pix += line_size;

  807.    p += line_size;

  808.   } while(--h);

  809. }

  810. 

  811. static void sub_pixels_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)

  812. {

  813.   DCTELEM  *p;

  814.   const UINT8 *pix;

  815.   p = block;

  816.   pix = pixels;

  817.   __asm __volatile("pxor %%mm7, %%mm7":::"memory");

  818.   do {

  819.     __asm __volatile(

  820. 	"movq	%0, %%mm0\n\t"

  821. 	"movq	%1, %%mm2\n\t"

  822. 	"movq	8%0, %%mm1\n\t"

  823. 	"movq	%%mm2, %%mm3\n\t"

  824. 	"punpcklbw %%mm7, %%mm2\n\t"

  825. 	"punpckhbw %%mm7, %%mm3\n\t"

  826. 	"psubsw %%mm2, %%mm0\n\t"

  827. 	"psubsw %%mm3, %%mm1\n\t"

  828. 	"movq	%%mm0, %0\n\t"

  829. 	"movq	%%mm1, 8%0\n\t"

  830. 	:"=m"(*p)

  831. 	:"m"(*pix)

  832. 	:"memory");

  833.    pix += line_size;

  834.    p +=   8;

  835.   } while (--h);

  836. }

  837. 

  838. static void sub_pixels_x2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)

  839. {

  840.   DCTELEM  *p;

  841.   const UINT8 *pix;

  842.   p = block;

  843.   pix = pixels;

  844.   __asm __volatile(

  845.       "pxor	%%mm7, %%mm7\n\t"

  846.       "movq	%0, %%mm6"

  847.       ::"m"(mm_wone):"memory");

  848.   do {

  849.     __asm __volatile(

  850. 	"movq	%0, %%mm0\n\t"

  851. 	"movq	%1, %%mm2\n\t"

  852. 	"movq	8%0, %%mm1\n\t"

  853. 	"movq	1%1, %%mm4\n\t"

  854. 	"movq	%%mm2, %%mm3\n\t"

  855. 	"movq	%%mm4, %%mm5\n\t"

  856. 	"punpcklbw %%mm7, %%mm2\n\t"

  857. 	"punpckhbw %%mm7, %%mm3\n\t"

  858. 	"punpcklbw %%mm7, %%mm4\n\t"

  859. 	"punpckhbw %%mm7, %%mm5\n\t"

  860. 	"paddusw %%mm4, %%mm2\n\t"

  861. 	"paddusw %%mm5, %%mm3\n\t"

  862. 	"paddusw %%mm6, %%mm2\n\t"

  863. 	"paddusw %%mm6, %%mm3\n\t"

  864. 	"psrlw	$1, %%mm2\n\t"

  865. 	"psrlw	$1, %%mm3\n\t"

  866. 	"psubsw %%mm2, %%mm0\n\t"

  867. 	"psubsw %%mm3, %%mm1\n\t"

  868. 	"movq	%%mm0, %0\n\t"

  869. 	"movq	%%mm1, 8%0\n\t"

  870. 	:"=m"(*p)

  871. 	:"m"(*pix)

  872. 	:"memory");

  873.    pix += line_size;

  874.    p +=   8;

  875.  } while (--h);

  876. }

  877. 

  878. static void sub_pixels_y2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)

  879. {

  880.   DCTELEM  *p;

  881.   const UINT8 *pix;

  882.   p = block;

  883.   pix = pixels;

  884.   __asm __volatile(

  885.       "pxor	%%mm7, %%mm7\n\t"

  886.       "movq	%0, %%mm6"

  887.       ::"m"(mm_wone):"memory");

  888.   do {

  889.     __asm __volatile(

  890. 	"movq	%0, %%mm0\n\t"

  891. 	"movq	%1, %%mm2\n\t"

  892. 	"movq	8%0, %%mm1\n\t"

  893. 	"movq	%2, %%mm4\n\t"

  894. 	"movq	%%mm2, %%mm3\n\t"

  895. 	"movq	%%mm4, %%mm5\n\t"

  896. 	"punpcklbw %%mm7, %%mm2\n\t"

  897. 	"punpckhbw %%mm7, %%mm3\n\t"

  898. 	"punpcklbw %%mm7, %%mm4\n\t"

  899. 	"punpckhbw %%mm7, %%mm5\n\t"

  900. 	"paddusw %%mm4, %%mm2\n\t"

  901. 	"paddusw %%mm5, %%mm3\n\t"

  902. 	"paddusw %%mm6, %%mm2\n\t"

  903. 	"paddusw %%mm6, %%mm3\n\t"

  904. 	"psrlw	$1, %%mm2\n\t"

  905. 	"psrlw	$1, %%mm3\n\t"

  906. 	"psubsw %%mm2, %%mm0\n\t"

  907. 	"psubsw %%mm3, %%mm1\n\t"

  908. 	"movq	%%mm0, %0\n\t"

  909. 	"movq	%%mm1, 8%0\n\t"

  910. 	:"=m"(*p)

  911. 	:"m"(*pix), "m"(*(pix+line_size))

  912. 	:"memory");

  913.    pix += line_size;

  914.    p +=   8;

  915.  } while (--h);

  916. }

  917. 

  918. static void   sub_pixels_xy2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)

  919. {

  920.   DCTELEM  *p;

  921.   const UINT8 *pix;

  922.   p = block;

  923.   pix = pixels;

  924.   __asm __volatile(

  925. 	"pxor	%%mm7, %%mm7\n\t"

  926. 	"movq	%0, %%mm6\n\t"

  927. 	::"m"(mm_wtwo):"memory");

  928.   do {

  929.     __asm __volatile(

  930. 	"movq	%1, %%mm0\n\t"

  931. 	"movq	%2, %%mm1\n\t"

  932. 	"movq	1%1, %%mm4\n\t"

  933. 	"movq	1%2, %%mm5\n\t"

  934. 	"movq	%%mm0, %%mm2\n\t"

  935. 	"movq	%%mm1, %%mm3\n\t"

  936. 	"punpcklbw %%mm7, %%mm0\n\t"

  937. 	"punpcklbw %%mm7, %%mm1\n\t"

  938. 	"punpckhbw %%mm7, %%mm2\n\t"

  939. 	"punpckhbw %%mm7, %%mm3\n\t"

  940. 	"paddusw %%mm1, %%mm0\n\t"

  941. 	"paddusw %%mm3, %%mm2\n\t"

  942. 	"movq	%%mm4, %%mm1\n\t"

  943. 	"movq	%%mm5, %%mm3\n\t"

  944. 	"punpcklbw %%mm7, %%mm4\n\t"

  945. 	"punpcklbw %%mm7, %%mm5\n\t"

  946. 	"punpckhbw %%mm7, %%mm1\n\t"

  947. 	"punpckhbw %%mm7, %%mm3\n\t"

  948. 	"paddusw %%mm5, %%mm4\n\t"

  949. 	"paddusw %%mm3, %%mm1\n\t"

  950. 	"paddusw %%mm6, %%mm4\n\t"

  951. 	"paddusw %%mm6, %%mm1\n\t"

  952. 	"paddusw %%mm4, %%mm0\n\t"

  953. 	"paddusw %%mm1, %%mm2\n\t"

  954. 	"movq	%0, %%mm1\n\t"

  955. 	"movq	8%0, %%mm3\n\t"

  956. 	"psrlw	$2, %%mm0\n\t"

  957. 	"psrlw	$2, %%mm2\n\t"

  958. 	"psubsw %%mm0, %%mm1\n\t"

  959. 	"psubsw %%mm2, %%mm3\n\t"

  960. 	"movq	%%mm1, %0\n\t"

  961. 	"movq	%%mm3, 8%0\n\t"

  962. 	:"=m"(*p)

  963. 	:"m"(*pix),

  964. 	 "m"(*(pix+line_size))

  965. 	:"memory");

  966.    pix += line_size;

  967.    p +=   8 ;

  968.   } while(--h);

  969. }

  970. 

  971. void dsputil_init_mmx(void)

  972. {

  973.     mm_flags = mm_support();

  974. #if 0

  975.     printf("CPU flags:");

  976.     if (mm_flags & MM_MMX)

  977.         printf(" mmx");

  978.     if (mm_flags & MM_MMXEXT)

  979.         printf(" mmxext");

  980.     if (mm_flags & MM_3DNOW)

  981.         printf(" 3dnow");

  982.     if (mm_flags & MM_SSE)

  983.         printf(" sse");

  984.     if (mm_flags & MM_SSE2)

  985.         printf(" sse2");

  986.     printf("\n");

  987. #endif

  988. 

  989.     if (mm_flags & MM_MMX) {

  990.         get_pixels = get_pixels_mmx;

  991.         put_pixels_clamped = put_pixels_clamped_mmx;

  992.         add_pixels_clamped = add_pixels_clamped_mmx;

  993.         

  994.         pix_abs16x16 = pix_abs16x16_mmx;

  995.         pix_abs16x16_x2 = pix_abs16x16_x2_mmx;

  996.         pix_abs16x16_y2 = pix_abs16x16_y2_mmx;

  997.         pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;

  998.         av_fdct = fdct_mmx;

  999. 

  1000.         put_pixels_tab[0] = put_pixels_mmx;

  1001.         put_pixels_tab[1] = put_pixels_x2_mmx;

  1002.         put_pixels_tab[2] = put_pixels_y2_mmx;

  1003.         put_pixels_tab[3] = put_pixels_xy2_mmx;

  1004. 

  1005.         put_no_rnd_pixels_tab[0] = put_pixels_mmx;

  1006.         put_no_rnd_pixels_tab[1] = put_no_rnd_pixels_x2_mmx;

  1007.         put_no_rnd_pixels_tab[2] = put_no_rnd_pixels_y2_mmx;

  1008.         put_no_rnd_pixels_tab[3] = put_no_rnd_pixels_xy2_mmx;

  1009.         

  1010.         avg_pixels_tab[0] = avg_pixels_mmx;

  1011.         avg_pixels_tab[1] = avg_pixels_x2_mmx;

  1012.         avg_pixels_tab[2] = avg_pixels_y2_mmx;

  1013.         avg_pixels_tab[3] = avg_pixels_xy2_mmx;

  1014. 

  1015.         avg_no_rnd_pixels_tab[0] = avg_no_rnd_pixels_mmx;

  1016.         avg_no_rnd_pixels_tab[1] = avg_no_rnd_pixels_x2_mmx;

  1017.         avg_no_rnd_pixels_tab[2] = avg_no_rnd_pixels_y2_mmx;

  1018.         avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels_xy2_mmx;

  1019.         

  1020.         sub_pixels_tab[0] = sub_pixels_mmx;

  1021.         sub_pixels_tab[1] = sub_pixels_x2_mmx;

  1022.         sub_pixels_tab[2] = sub_pixels_y2_mmx;

  1023.         sub_pixels_tab[3] = sub_pixels_xy2_mmx;

  1024. 

  1025.         if (mm_flags & MM_MMXEXT) {

  1026.             pix_abs16x16 = pix_abs16x16_sse;

  1027.         }

  1028. 

  1029.         if (mm_flags & MM_SSE) {

  1030.             put_pixels_tab[1] = put_pixels_x2_sse;

  1031.             put_pixels_tab[2] = put_pixels_y2_sse;

  1032.             

  1033.             avg_pixels_tab[0] = avg_pixels_sse;

  1034.             avg_pixels_tab[1] = avg_pixels_x2_sse;

  1035.             avg_pixels_tab[2] = avg_pixels_y2_sse;

  1036.             avg_pixels_tab[3] = avg_pixels_xy2_sse;

  1037. 

  1038.             sub_pixels_tab[1] = sub_pixels_x2_sse;

  1039.             sub_pixels_tab[2] = sub_pixels_y2_sse;

  1040.         } else if (mm_flags & MM_3DNOW) {

  1041.             put_pixels_tab[1] = put_pixels_x2_3dnow;

  1042.             put_pixels_tab[2] = put_pixels_y2_3dnow;

  1043.             

  1044.             avg_pixels_tab[0] = avg_pixels_3dnow;

  1045.             avg_pixels_tab[1] = avg_pixels_x2_3dnow;

  1046.             avg_pixels_tab[2] = avg_pixels_y2_3dnow;

  1047.             avg_pixels_tab[3] = avg_pixels_xy2_3dnow;

  1048. 

  1049.             sub_pixels_tab[1] = sub_pixels_x2_3dnow;

  1050.             sub_pixels_tab[2] = sub_pixels_y2_3dnow;

  1051.         }

  1052. 

  1053. #ifdef USE_MMX_IDCT

  1054. 	/* use MMX / MMXEXT iDCT code from libmpeg2 */

  1055. 	//printf("LIBAVCODEC: Using MMX%s iDCT code\n",(mm_flags & MM_MMXEXT)?"EXT":"");

  1056. 	ff_idct = (mm_flags & MM_MMXEXT) ? mmxext_idct : mmx_idct;

  1057. 	/* the mmx/mmxext idct uses a reordered input, so we patch scan tables */

  1058. 	{   int i,j;

  1059. 	    for (i = 0; i < 64; i++) {

  1060. 		j = zigzag_direct[i];

  1061. 		zigzag_direct[i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);

  1062. 		j = ff_alternate_horizontal_scan[i];

  1063. 		ff_alternate_horizontal_scan[i] =  (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);

  1064. 		j = ff_alternate_vertical_scan[i];

  1065. 		ff_alternate_vertical_scan[i] =  (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);

  1066. 	    }

  1067. 	}

  1068. #endif

  1069. 

  1070.     }

  1071. }