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.
4870 Commits
32 Branches
1.2GB
Tree: cd7af76d9e
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'cd7af76d9e'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx.c

2884 lines
102KB
Raw Blame History 

  1. /*

  2.  * MMX optimized DSP utils

  3.  * Copyright (c) 2000, 2001 Fabrice Bellard.

  4.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  5.  *

  6.  * This library is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2 of the License, or (at your option) any later version.

  10.  *

  11.  * This library is distributed in the hope that it will be useful,

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

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with this library; if not, write to the Free Software

  18.  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  19.  *

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

  21.  */

  22. 

  23. #include "../dsputil.h"

  24. #include "../simple_idct.h"

  25. #include "../mpegvideo.h"

  26. #include "mmx.h"

  27. 

  28. //#undef NDEBUG

  29. //#include <assert.h>

  30. 

  31. extern const uint8_t ff_h263_loop_filter_strength[32];

  32. extern void ff_idct_xvid_mmx(short *block);

  33. extern void ff_idct_xvid_mmx2(short *block);

  34. 

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

  36. 

  37. /* pixel operations */

  38. static const uint64_t mm_bone attribute_used __attribute__ ((aligned(8))) = 0x0101010101010101ULL;

  39. static const uint64_t mm_wone attribute_used __attribute__ ((aligned(8))) = 0x0001000100010001ULL;

  40. static const uint64_t mm_wtwo attribute_used __attribute__ ((aligned(8))) = 0x0002000200020002ULL;

  41. 

  42. static const uint64_t ff_pw_20 attribute_used __attribute__ ((aligned(8))) = 0x0014001400140014ULL;

  43. static const uint64_t ff_pw_3  attribute_used __attribute__ ((aligned(8))) = 0x0003000300030003ULL;

  44. static const uint64_t ff_pw_4  attribute_used __attribute__ ((aligned(8))) = 0x0004000400040004ULL;

  45. static const uint64_t ff_pw_5  attribute_used __attribute__ ((aligned(8))) = 0x0005000500050005ULL;

  46. static const uint64_t ff_pw_16 attribute_used __attribute__ ((aligned(8))) = 0x0010001000100010ULL;

  47. static const uint64_t ff_pw_32 attribute_used __attribute__ ((aligned(8))) = 0x0020002000200020ULL;

  48. static const uint64_t ff_pw_64 attribute_used __attribute__ ((aligned(8))) = 0x0040004000400040ULL;

  49. static const uint64_t ff_pw_15 attribute_used __attribute__ ((aligned(8))) = 0x000F000F000F000FULL;

  50. 

  51. static const uint64_t ff_pb_3F attribute_used __attribute__ ((aligned(8))) = 0x3F3F3F3F3F3F3F3FULL;

  52. static const uint64_t ff_pb_FC attribute_used __attribute__ ((aligned(8))) = 0xFCFCFCFCFCFCFCFCULL;

  53. 

  54. #define JUMPALIGN() __asm __volatile (".balign 8"::)

  55. #define MOVQ_ZERO(regd)  __asm __volatile ("pxor %%" #regd ", %%" #regd ::)

  56. 

  57. #define MOVQ_WONE(regd) \

  58.     __asm __volatile ( \

  59.     "pcmpeqd %%" #regd ", %%" #regd " \n\t" \

  60.     "psrlw $15, %%" #regd ::)

  61. 

  62. #define MOVQ_BFE(regd) \

  63.     __asm __volatile ( \

  64.     "pcmpeqd %%" #regd ", %%" #regd " \n\t"\

  65.     "paddb %%" #regd ", %%" #regd " \n\t" ::)

  66. 

  67. #ifndef PIC

  68. #define MOVQ_BONE(regd)  __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_bone))

  69. #define MOVQ_WTWO(regd)  __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_wtwo))

  70. #else

  71. // for shared library it's better to use this way for accessing constants

  72. // pcmpeqd -> -1

  73. #define MOVQ_BONE(regd) \

  74.     __asm __volatile ( \

  75.     "pcmpeqd %%" #regd ", %%" #regd " \n\t" \

  76.     "psrlw $15, %%" #regd " \n\t" \

  77.     "packuswb %%" #regd ", %%" #regd " \n\t" ::)

  78. 

  79. #define MOVQ_WTWO(regd) \

  80.     __asm __volatile ( \

  81.     "pcmpeqd %%" #regd ", %%" #regd " \n\t" \

  82.     "psrlw $15, %%" #regd " \n\t" \

  83.     "psllw $1, %%" #regd " \n\t"::)

  84. 

  85. #endif

  86. 

  87. // using regr as temporary and for the output result

  88. // first argument is unmodifed and second is trashed

  89. // regfe is supposed to contain 0xfefefefefefefefe

  90. #define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \

  91.     "movq " #rega ", " #regr "	\n\t"\

  92.     "pand " #regb ", " #regr "	\n\t"\

  93.     "pxor " #rega ", " #regb "	\n\t"\

  94.     "pand " #regfe "," #regb "	\n\t"\

  95.     "psrlq $1, " #regb " 	\n\t"\

  96.     "paddb " #regb ", " #regr "	\n\t"

  97. 

  98. #define PAVGB_MMX(rega, regb, regr, regfe) \

  99.     "movq " #rega ", " #regr "	\n\t"\

  100.     "por  " #regb ", " #regr "	\n\t"\

  101.     "pxor " #rega ", " #regb "	\n\t"\

  102.     "pand " #regfe "," #regb "	\n\t"\

  103.     "psrlq $1, " #regb "	\n\t"\

  104.     "psubb " #regb ", " #regr "	\n\t"

  105. 

  106. // mm6 is supposed to contain 0xfefefefefefefefe

  107. #define PAVGBP_MMX_NO_RND(rega, regb, regr,  regc, regd, regp) \

  108.     "movq " #rega ", " #regr "	\n\t"\

  109.     "movq " #regc ", " #regp "	\n\t"\

  110.     "pand " #regb ", " #regr "	\n\t"\

  111.     "pand " #regd ", " #regp "	\n\t"\

  112.     "pxor " #rega ", " #regb "	\n\t"\

  113.     "pxor " #regc ", " #regd "	\n\t"\

  114.     "pand %%mm6, " #regb "	\n\t"\

  115.     "pand %%mm6, " #regd "	\n\t"\

  116.     "psrlq $1, " #regb " 	\n\t"\

  117.     "psrlq $1, " #regd " 	\n\t"\

  118.     "paddb " #regb ", " #regr "	\n\t"\

  119.     "paddb " #regd ", " #regp "	\n\t"

  120. 

  121. #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \

  122.     "movq " #rega ", " #regr "	\n\t"\

  123.     "movq " #regc ", " #regp "	\n\t"\

  124.     "por  " #regb ", " #regr "	\n\t"\

  125.     "por  " #regd ", " #regp "	\n\t"\

  126.     "pxor " #rega ", " #regb "	\n\t"\

  127.     "pxor " #regc ", " #regd "	\n\t"\

  128.     "pand %%mm6, " #regb "     	\n\t"\

  129.     "pand %%mm6, " #regd "     	\n\t"\

  130.     "psrlq $1, " #regd "	\n\t"\

  131.     "psrlq $1, " #regb "	\n\t"\

  132.     "psubb " #regb ", " #regr "	\n\t"\

  133.     "psubb " #regd ", " #regp "	\n\t"

  134. 

  135. /***********************************/

  136. /* MMX no rounding */

  137. #define DEF(x, y) x ## _no_rnd_ ## y ##_mmx

  138. #define SET_RND  MOVQ_WONE

  139. #define PAVGBP(a, b, c, d, e, f)	PAVGBP_MMX_NO_RND(a, b, c, d, e, f)

  140. #define PAVGB(a, b, c, e)		PAVGB_MMX_NO_RND(a, b, c, e)

  141. 

  142. #include "dsputil_mmx_rnd.h"

  143. 

  144. #undef DEF

  145. #undef SET_RND

  146. #undef PAVGBP

  147. #undef PAVGB

  148. /***********************************/

  149. /* MMX rounding */

  150. 

  151. #define DEF(x, y) x ## _ ## y ##_mmx

  152. #define SET_RND  MOVQ_WTWO

  153. #define PAVGBP(a, b, c, d, e, f)	PAVGBP_MMX(a, b, c, d, e, f)

  154. #define PAVGB(a, b, c, e)		PAVGB_MMX(a, b, c, e)

  155. 

  156. #include "dsputil_mmx_rnd.h"

  157. 

  158. #undef DEF

  159. #undef SET_RND

  160. #undef PAVGBP

  161. #undef PAVGB

  162. 

  163. /***********************************/

  164. /* 3Dnow specific */

  165. 

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

  167. /* for Athlons PAVGUSB is prefered */

  168. #define PAVGB "pavgusb"

  169. 

  170. #include "dsputil_mmx_avg.h"

  171. 

  172. #undef DEF

  173. #undef PAVGB

  174. 

  175. /***********************************/

  176. /* MMX2 specific */

  177. 

  178. #define DEF(x) x ## _mmx2

  179. 

  180. /* Introduced only in MMX2 set */

  181. #define PAVGB "pavgb"

  182. 

  183. #include "dsputil_mmx_avg.h"

  184. 

  185. #undef DEF

  186. #undef PAVGB

  187. 

  188. /***********************************/

  189. /* standard MMX */

  190. 

  191. #ifdef CONFIG_ENCODERS

  192. static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)

  193. {

  194.     asm volatile(

  195.         "mov $-128, %%"REG_a"	\n\t"

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

  197.         ".balign 16		\n\t"

  198.         "1:			\n\t"

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

  200.         "movq (%0, %2), %%mm2	\n\t"

  201.         "movq %%mm0, %%mm1	\n\t"

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

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

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

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

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

  207.         "movq %%mm0, (%1, %%"REG_a")\n\t"

  208.         "movq %%mm1, 8(%1, %%"REG_a")\n\t"

  209.         "movq %%mm2, 16(%1, %%"REG_a")\n\t"

  210.         "movq %%mm3, 24(%1, %%"REG_a")\n\t"

  211.         "add %3, %0		\n\t"

  212.         "add $32, %%"REG_a"	\n\t"

  213.         "js 1b			\n\t"

  214.         : "+r" (pixels)

  215.         : "r" (block+64), "r" ((long)line_size), "r" ((long)line_size*2)

  216.         : "%"REG_a

  217.     );

  218. }

  219. 

  220. static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)

  221. {

  222.     asm volatile(

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

  224.         "mov $-128, %%"REG_a"	\n\t"

  225.         ".balign 16		\n\t"

  226.         "1:			\n\t"

  227.         "movq (%0), %%mm0	\n\t"

  228.         "movq (%1), %%mm2	\n\t"

  229.         "movq %%mm0, %%mm1	\n\t"

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

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

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

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

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

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

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

  237.         "movq %%mm0, (%2, %%"REG_a")\n\t"

  238.         "movq %%mm1, 8(%2, %%"REG_a")\n\t"

  239.         "add %3, %0		\n\t"

  240.         "add %3, %1		\n\t"

  241.         "add $16, %%"REG_a"	\n\t"

  242.         "jnz 1b			\n\t"

  243.         : "+r" (s1), "+r" (s2)

  244.         : "r" (block+64), "r" ((long)stride)

  245.         : "%"REG_a

  246.     );

  247. }

  248. #endif //CONFIG_ENCODERS

  249. 

  250. void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)

  251. {

  252.     const DCTELEM *p;

  253.     uint8_t *pix;

  254. 

  255.     /* read the pixels */

  256.     p = block;

  257.     pix = pixels;

  258.     /* unrolled loop */

  259. 	__asm __volatile(

  260. 		"movq	%3, %%mm0\n\t"

  261. 		"movq	8%3, %%mm1\n\t"

  262. 		"movq	16%3, %%mm2\n\t"

  263. 		"movq	24%3, %%mm3\n\t"

  264. 		"movq	32%3, %%mm4\n\t"

  265. 		"movq	40%3, %%mm5\n\t"

  266. 		"movq	48%3, %%mm6\n\t"

  267. 		"movq	56%3, %%mm7\n\t"

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

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

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

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

  272. 		"movq	%%mm0, (%0)\n\t"

  273. 		"movq	%%mm2, (%0, %1)\n\t"

  274. 		"movq	%%mm4, (%0, %1, 2)\n\t"

  275. 		"movq	%%mm6, (%0, %2)\n\t"

  276. 		::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "m"(*p)

  277. 		:"memory");

  278.         pix += line_size*4;

  279.         p += 32;

  280. 

  281.     // if here would be an exact copy of the code above

  282.     // compiler would generate some very strange code

  283.     // thus using "r"

  284.     __asm __volatile(

  285. 	    "movq	(%3), %%mm0\n\t"

  286. 	    "movq	8(%3), %%mm1\n\t"

  287. 	    "movq	16(%3), %%mm2\n\t"

  288. 	    "movq	24(%3), %%mm3\n\t"

  289. 	    "movq	32(%3), %%mm4\n\t"

  290. 	    "movq	40(%3), %%mm5\n\t"

  291. 	    "movq	48(%3), %%mm6\n\t"

  292. 	    "movq	56(%3), %%mm7\n\t"

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

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

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

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

  297. 	    "movq	%%mm0, (%0)\n\t"

  298. 	    "movq	%%mm2, (%0, %1)\n\t"

  299. 	    "movq	%%mm4, (%0, %1, 2)\n\t"

  300. 	    "movq	%%mm6, (%0, %2)\n\t"

  301. 	    ::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "r"(p)

  302. 	    :"memory");

  303. }

  304. 

  305. static const unsigned char __align8 vector128[8] =

  306.   { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };

  307. 

  308. void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)

  309. {

  310.     int i;

  311. 

  312.     movq_m2r(*vector128, mm1);

  313.     for (i = 0; i < 8; i++) {

  314.         movq_m2r(*(block), mm0);

  315.         packsswb_m2r(*(block + 4), mm0);

  316.         block += 8;

  317.         paddb_r2r(mm1, mm0);

  318.         movq_r2m(mm0, *pixels);

  319.         pixels += line_size;

  320.     }

  321. }

  322. 

  323. void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)

  324. {

  325.     const DCTELEM *p;

  326.     uint8_t *pix;

  327.     int i;

  328. 

  329.     /* read the pixels */

  330.     p = block;

  331.     pix = pixels;

  332.     MOVQ_ZERO(mm7);

  333.     i = 4;

  334.     do {

  335. 	__asm __volatile(

  336. 		"movq	(%2), %%mm0\n\t"

  337. 		"movq	8(%2), %%mm1\n\t"

  338. 		"movq	16(%2), %%mm2\n\t"

  339. 		"movq	24(%2), %%mm3\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  357. 		:"r"(p)

  358. 		:"memory");

  359.         pix += line_size*2;

  360.         p += 16;

  361.     } while (--i);

  362. }

  363. 

  364. static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  365. {

  366.     __asm __volatile(

  367. 	 "lea (%3, %3), %%"REG_a"	\n\t"

  368. 	 ".balign 8			\n\t"

  369. 	 "1:				\n\t"

  370. 	 "movd (%1), %%mm0		\n\t"

  371. 	 "movd (%1, %3), %%mm1		\n\t"

  372. 	 "movd %%mm0, (%2)		\n\t"

  373. 	 "movd %%mm1, (%2, %3)		\n\t"

  374. 	 "add %%"REG_a", %1		\n\t"

  375. 	 "add %%"REG_a", %2		\n\t"

  376. 	 "movd (%1), %%mm0		\n\t"

  377. 	 "movd (%1, %3), %%mm1		\n\t"

  378. 	 "movd %%mm0, (%2)		\n\t"

  379. 	 "movd %%mm1, (%2, %3)		\n\t"

  380. 	 "add %%"REG_a", %1		\n\t"

  381. 	 "add %%"REG_a", %2		\n\t"

  382. 	 "subl $4, %0			\n\t"

  383. 	 "jnz 1b			\n\t"

  384. 	 : "+g"(h), "+r" (pixels),  "+r" (block)

  385. 	 : "r"((long)line_size)

  386. 	 : "%"REG_a, "memory"

  387. 	);

  388. }

  389. 

  390. static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  391. {

  392.     __asm __volatile(

  393. 	 "lea (%3, %3), %%"REG_a"	\n\t"

  394. 	 ".balign 8			\n\t"

  395. 	 "1:				\n\t"

  396. 	 "movq (%1), %%mm0		\n\t"

  397. 	 "movq (%1, %3), %%mm1		\n\t"

  398.      	 "movq %%mm0, (%2)		\n\t"

  399. 	 "movq %%mm1, (%2, %3)		\n\t"

  400. 	 "add %%"REG_a", %1		\n\t"

  401. 	 "add %%"REG_a", %2		\n\t"

  402. 	 "movq (%1), %%mm0		\n\t"

  403. 	 "movq (%1, %3), %%mm1		\n\t"

  404. 	 "movq %%mm0, (%2)		\n\t"

  405. 	 "movq %%mm1, (%2, %3)		\n\t"

  406. 	 "add %%"REG_a", %1		\n\t"

  407. 	 "add %%"REG_a", %2		\n\t"

  408. 	 "subl $4, %0			\n\t"

  409. 	 "jnz 1b			\n\t"

  410. 	 : "+g"(h), "+r" (pixels),  "+r" (block)

  411. 	 : "r"((long)line_size)

  412. 	 : "%"REG_a, "memory"

  413. 	);

  414. }

  415. 

  416. static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  417. {

  418.     __asm __volatile(

  419. 	 "lea (%3, %3), %%"REG_a"	\n\t"

  420. 	 ".balign 8			\n\t"

  421. 	 "1:				\n\t"

  422. 	 "movq (%1), %%mm0		\n\t"

  423. 	 "movq 8(%1), %%mm4		\n\t"

  424. 	 "movq (%1, %3), %%mm1		\n\t"

  425. 	 "movq 8(%1, %3), %%mm5		\n\t"

  426.      	 "movq %%mm0, (%2)		\n\t"

  427.      	 "movq %%mm4, 8(%2)		\n\t"

  428. 	 "movq %%mm1, (%2, %3)		\n\t"

  429. 	 "movq %%mm5, 8(%2, %3)		\n\t"

  430. 	 "add %%"REG_a", %1		\n\t"

  431. 	 "add %%"REG_a", %2       	\n\t"

  432. 	 "movq (%1), %%mm0		\n\t"

  433. 	 "movq 8(%1), %%mm4		\n\t"

  434. 	 "movq (%1, %3), %%mm1		\n\t"

  435. 	 "movq 8(%1, %3), %%mm5		\n\t"

  436. 	 "movq %%mm0, (%2)		\n\t"

  437. 	 "movq %%mm4, 8(%2)		\n\t"

  438. 	 "movq %%mm1, (%2, %3)		\n\t"

  439. 	 "movq %%mm5, 8(%2, %3)		\n\t"

  440. 	 "add %%"REG_a", %1		\n\t"

  441. 	 "add %%"REG_a", %2       	\n\t"

  442. 	 "subl $4, %0			\n\t"

  443. 	 "jnz 1b			\n\t"

  444. 	 : "+g"(h), "+r" (pixels),  "+r" (block)

  445. 	 : "r"((long)line_size)

  446. 	 : "%"REG_a, "memory"

  447. 	);

  448. }

  449. 

  450. static void clear_blocks_mmx(DCTELEM *blocks)

  451. {

  452.     __asm __volatile(

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

  454.                 "mov $-128*6, %%"REG_a"	\n\t"

  455.                 "1:				\n\t"

  456.                 "movq %%mm7, (%0, %%"REG_a")	\n\t"

  457.                 "movq %%mm7, 8(%0, %%"REG_a")	\n\t"

  458.                 "movq %%mm7, 16(%0, %%"REG_a")	\n\t"

  459.                 "movq %%mm7, 24(%0, %%"REG_a")	\n\t"

  460.                 "add $32, %%"REG_a"		\n\t"

  461.                 " js 1b				\n\t"

  462.                 : : "r" (((uint8_t *)blocks)+128*6)

  463.                 : "%"REG_a

  464.         );

  465. }

  466. 

  467. #ifdef CONFIG_ENCODERS

  468. static int pix_sum16_mmx(uint8_t * pix, int line_size){

  469.     const int h=16;

  470.     int sum;

  471.     long index= -line_size*h;

  472. 

  473.     __asm __volatile(

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

  475.                 "pxor %%mm6, %%mm6		\n\t"

  476.                 "1:				\n\t"

  477.                 "movq (%2, %1), %%mm0		\n\t"

  478.                 "movq (%2, %1), %%mm1		\n\t"

  479.                 "movq 8(%2, %1), %%mm2		\n\t"

  480.                 "movq 8(%2, %1), %%mm3		\n\t"

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

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

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

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

  485.                 "paddw %%mm0, %%mm1		\n\t"

  486.                 "paddw %%mm2, %%mm3		\n\t"

  487.                 "paddw %%mm1, %%mm3		\n\t"

  488.                 "paddw %%mm3, %%mm6		\n\t"

  489.                 "add %3, %1			\n\t"

  490.                 " js 1b				\n\t"

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

  492.                 "psrlq $32, %%mm6		\n\t"

  493.                 "paddw %%mm5, %%mm6		\n\t"

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

  495.                 "psrlq $16, %%mm6		\n\t"

  496.                 "paddw %%mm5, %%mm6		\n\t"

  497.                 "movd %%mm6, %0			\n\t"

  498.                 "andl $0xFFFF, %0		\n\t"

  499.                 : "=&r" (sum), "+r" (index)

  500.                 : "r" (pix - index), "r" ((long)line_size)

  501.         );

  502. 

  503.         return sum;

  504. }

  505. #endif //CONFIG_ENCODERS

  506. 

  507. static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){

  508.     long i=0;

  509.     asm volatile(

  510.         "1:				\n\t"

  511.         "movq  (%1, %0), %%mm0		\n\t"

  512.         "movq  (%2, %0), %%mm1		\n\t"

  513.         "paddb %%mm0, %%mm1		\n\t"

  514.         "movq %%mm1, (%2, %0)		\n\t"

  515.         "movq 8(%1, %0), %%mm0		\n\t"

  516.         "movq 8(%2, %0), %%mm1		\n\t"

  517.         "paddb %%mm0, %%mm1		\n\t"

  518.         "movq %%mm1, 8(%2, %0)		\n\t"

  519.         "add $16, %0			\n\t"

  520.         "cmp %3, %0			\n\t"

  521.         " jb 1b				\n\t"

  522.         : "+r" (i)

  523.         : "r"(src), "r"(dst), "r"((long)w-15)

  524.     );

  525.     for(; i<w; i++)

  526.         dst[i+0] += src[i+0];

  527. }

  528. 

  529. #define H263_LOOP_FILTER \

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

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

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

  533.         "movq  %3, %%mm2		\n\t"\

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

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

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

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

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

  539.         "psubw %%mm2, %%mm0		\n\t"\

  540.         "psubw %%mm3, %%mm1		\n\t"\

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

  542.         "movq  %1, %%mm3		\n\t"\

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

  544.         "movq  %2, %%mm5		\n\t"\

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

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

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

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

  549.         "psubw %%mm2, %%mm4		\n\t"\

  550.         "psubw %%mm3, %%mm5		\n\t"\

  551.         "psllw $2, %%mm4		\n\t"\

  552.         "psllw $2, %%mm5		\n\t"\

  553.         "paddw %%mm0, %%mm4		\n\t"\

  554.         "paddw %%mm1, %%mm5		\n\t"\

  555.         "pxor %%mm6, %%mm6		\n\t"\

  556.         "pcmpgtw %%mm4, %%mm6		\n\t"\

  557.         "pcmpgtw %%mm5, %%mm7		\n\t"\

  558.         "pxor %%mm6, %%mm4		\n\t"\

  559.         "pxor %%mm7, %%mm5		\n\t"\

  560.         "psubw %%mm6, %%mm4		\n\t"\

  561.         "psubw %%mm7, %%mm5		\n\t"\

  562.         "psrlw $3, %%mm4		\n\t"\

  563.         "psrlw $3, %%mm5		\n\t"\

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

  565.         "packsswb %%mm7, %%mm6		\n\t"\

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

  567.         "movd %4, %%mm2			\n\t"\

  568.         "punpcklbw %%mm2, %%mm2		\n\t"\

  569.         "punpcklbw %%mm2, %%mm2		\n\t"\

  570.         "punpcklbw %%mm2, %%mm2		\n\t"\

  571.         "psubusb %%mm4, %%mm2		\n\t"\

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

  573.         "psubusb %%mm4, %%mm3		\n\t"\

  574.         "psubb %%mm3, %%mm2		\n\t"\

  575.         "movq %1, %%mm3			\n\t"\

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

  577.         "pxor %%mm6, %%mm3		\n\t"\

  578.         "pxor %%mm6, %%mm4		\n\t"\

  579.         "paddusb %%mm2, %%mm3		\n\t"\

  580.         "psubusb %%mm2, %%mm4		\n\t"\

  581.         "pxor %%mm6, %%mm3		\n\t"\

  582.         "pxor %%mm6, %%mm4		\n\t"\

  583.         "paddusb %%mm2, %%mm2		\n\t"\

  584.         "packsswb %%mm1, %%mm0		\n\t"\

  585.         "pcmpgtb %%mm0, %%mm7		\n\t"\

  586.         "pxor %%mm7, %%mm0		\n\t"\

  587.         "psubb %%mm7, %%mm0		\n\t"\

  588.         "movq %%mm0, %%mm1		\n\t"\

  589.         "psubusb %%mm2, %%mm0		\n\t"\

  590.         "psubb %%mm0, %%mm1		\n\t"\

  591.         "pand %5, %%mm1			\n\t"\

  592.         "psrlw $2, %%mm1		\n\t"\

  593.         "pxor %%mm7, %%mm1		\n\t"\

  594.         "psubb %%mm7, %%mm1		\n\t"\

  595.         "movq %0, %%mm5			\n\t"\

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

  597.         "psubb %%mm1, %%mm5		\n\t"\

  598.         "paddb %%mm1, %%mm6		\n\t"

  599. 

  600. static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){

  601.     const int strength= ff_h263_loop_filter_strength[qscale];

  602. 

  603.     asm volatile(

  604.     

  605.         H263_LOOP_FILTER

  606.         

  607.         "movq %%mm3, %1			\n\t"

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

  609.         "movq %%mm5, %0			\n\t"

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

  611.         : "+m" (*(uint64_t*)(src - 2*stride)),

  612.           "+m" (*(uint64_t*)(src - 1*stride)),

  613.           "+m" (*(uint64_t*)(src + 0*stride)),

  614.           "+m" (*(uint64_t*)(src + 1*stride))

  615.         : "g" (2*strength), "m"(ff_pb_FC)

  616.     );

  617. }

  618. 

  619. static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){

  620.     asm volatile( //FIXME could save 1 instruction if done as 8x4 ...

  621.         "movd  %4, %%mm0		\n\t"

  622.         "movd  %5, %%mm1		\n\t"

  623.         "movd  %6, %%mm2		\n\t"

  624.         "movd  %7, %%mm3		\n\t"

  625.         "punpcklbw %%mm1, %%mm0		\n\t"

  626.         "punpcklbw %%mm3, %%mm2		\n\t"

  627.         "movq %%mm0, %%mm1		\n\t"

  628.         "punpcklwd %%mm2, %%mm0		\n\t"

  629.         "punpckhwd %%mm2, %%mm1		\n\t"

  630.         "movd  %%mm0, %0		\n\t"

  631.         "punpckhdq %%mm0, %%mm0		\n\t"

  632.         "movd  %%mm0, %1		\n\t"

  633.         "movd  %%mm1, %2		\n\t"

  634.         "punpckhdq %%mm1, %%mm1		\n\t"

  635.         "movd  %%mm1, %3		\n\t"

  636.         

  637.         : "=m" (*(uint32_t*)(dst + 0*dst_stride)),

  638.           "=m" (*(uint32_t*)(dst + 1*dst_stride)),

  639.           "=m" (*(uint32_t*)(dst + 2*dst_stride)),

  640.           "=m" (*(uint32_t*)(dst + 3*dst_stride))

  641.         :  "m" (*(uint32_t*)(src + 0*src_stride)),

  642.            "m" (*(uint32_t*)(src + 1*src_stride)),

  643.            "m" (*(uint32_t*)(src + 2*src_stride)),

  644.            "m" (*(uint32_t*)(src + 3*src_stride))

  645.     );

  646. }

  647. 

  648. static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){

  649.     const int strength= ff_h263_loop_filter_strength[qscale];

  650.     uint64_t temp[4] __attribute__ ((aligned(8)));

  651.     uint8_t *btemp= (uint8_t*)temp;

  652.     

  653.     src -= 2;

  654. 

  655.     transpose4x4(btemp  , src           , 8, stride);

  656.     transpose4x4(btemp+4, src + 4*stride, 8, stride);

  657.     asm volatile(

  658.         H263_LOOP_FILTER // 5 3 4 6

  659.         

  660.         : "+m" (temp[0]),

  661.           "+m" (temp[1]),

  662.           "+m" (temp[2]),

  663.           "+m" (temp[3])

  664.         : "g" (2*strength), "m"(ff_pb_FC)

  665.     );

  666. 

  667.     asm volatile(

  668.         "movq %%mm5, %%mm1		\n\t"

  669.         "movq %%mm4, %%mm0		\n\t"

  670.         "punpcklbw %%mm3, %%mm5		\n\t"

  671.         "punpcklbw %%mm6, %%mm4		\n\t"

  672.         "punpckhbw %%mm3, %%mm1		\n\t"

  673.         "punpckhbw %%mm6, %%mm0		\n\t"

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

  675.         "movq %%mm1, %%mm6		\n\t"

  676.         "punpcklwd %%mm4, %%mm5		\n\t"

  677.         "punpcklwd %%mm0, %%mm1		\n\t"

  678.         "punpckhwd %%mm4, %%mm3		\n\t"

  679.         "punpckhwd %%mm0, %%mm6		\n\t"

  680.         "movd %%mm5, (%0)		\n\t"

  681.         "punpckhdq %%mm5, %%mm5		\n\t"

  682.         "movd %%mm5, (%0,%2)		\n\t"

  683.         "movd %%mm3, (%0,%2,2)		\n\t"

  684.         "punpckhdq %%mm3, %%mm3		\n\t"

  685.         "movd %%mm3, (%0,%3)		\n\t"

  686.         "movd %%mm1, (%1)		\n\t"

  687.         "punpckhdq %%mm1, %%mm1		\n\t"

  688.         "movd %%mm1, (%1,%2)		\n\t"

  689.         "movd %%mm6, (%1,%2,2)		\n\t"

  690.         "punpckhdq %%mm6, %%mm6		\n\t"

  691.         "movd %%mm6, (%1,%3)		\n\t"

  692.         :: "r" (src),

  693.            "r" (src + 4*stride),

  694.            "r" ((long)   stride ),

  695.            "r" ((long)(3*stride))

  696.     );

  697. }

  698. 

  699. #ifdef CONFIG_ENCODERS

  700. static int pix_norm1_mmx(uint8_t *pix, int line_size) {

  701.     int tmp;

  702.   asm volatile (

  703.       "movl $16,%%ecx\n"

  704.       "pxor %%mm0,%%mm0\n"

  705.       "pxor %%mm7,%%mm7\n"

  706.       "1:\n"

  707.       "movq (%0),%%mm2\n"	/* mm2 = pix[0-7] */

  708.       "movq 8(%0),%%mm3\n"	/* mm3 = pix[8-15] */

  709. 

  710.       "movq %%mm2,%%mm1\n"	/* mm1 = mm2 = pix[0-7] */

  711. 

  712.       "punpckhbw %%mm0,%%mm1\n"	/* mm1 = [pix4-7] */

  713.       "punpcklbw %%mm0,%%mm2\n"	/* mm2 = [pix0-3] */

  714. 

  715.       "movq %%mm3,%%mm4\n"	/* mm4 = mm3 = pix[8-15] */

  716.       "punpckhbw %%mm0,%%mm3\n"	/* mm3 = [pix12-15] */

  717.       "punpcklbw %%mm0,%%mm4\n"	/* mm4 = [pix8-11] */

  718. 

  719.       "pmaddwd %%mm1,%%mm1\n"	/* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */

  720.       "pmaddwd %%mm2,%%mm2\n"	/* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */

  721. 

  722.       "pmaddwd %%mm3,%%mm3\n"

  723.       "pmaddwd %%mm4,%%mm4\n"

  724. 

  725.       "paddd %%mm1,%%mm2\n"	/* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,

  726. 					  pix2^2+pix3^2+pix6^2+pix7^2) */

  727.       "paddd %%mm3,%%mm4\n"

  728.       "paddd %%mm2,%%mm7\n"

  729. 

  730.       "add %2, %0\n"

  731.       "paddd %%mm4,%%mm7\n"

  732.       "dec %%ecx\n"

  733.       "jnz 1b\n"

  734. 

  735.       "movq %%mm7,%%mm1\n"

  736.       "psrlq $32, %%mm7\n"	/* shift hi dword to lo */

  737.       "paddd %%mm7,%%mm1\n"

  738.       "movd %%mm1,%1\n"

  739.       : "+r" (pix), "=r"(tmp) : "r" ((long)line_size) : "%ecx" );

  740.     return tmp;

  741. }

  742. 

  743. static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  744.     int tmp;

  745.   asm volatile (

  746.       "movl %4,%%ecx\n"

  747.       "pxor %%mm0,%%mm0\n"	/* mm0 = 0 */

  748.       "pxor %%mm7,%%mm7\n"	/* mm7 holds the sum */

  749.       "1:\n"

  750.       "movq (%0),%%mm1\n"	/* mm1 = pix1[0-7] */

  751.       "movq (%1),%%mm2\n"	/* mm2 = pix2[0-7] */

  752. 

  753.       "movq %%mm1,%%mm5\n"

  754.       "psubusb %%mm2,%%mm1\n"

  755.       "psubusb %%mm5,%%mm2\n"

  756. 

  757.       "por %%mm1,%%mm2\n"

  758. 

  759.       "movq %%mm2,%%mm1\n"

  760. 

  761.       "punpckhbw %%mm0,%%mm2\n"

  762.       "punpcklbw %%mm0,%%mm1\n"	/* mm1 now spread over (mm1,mm2) */

  763. 

  764.       "pmaddwd %%mm2,%%mm2\n"

  765.       "pmaddwd %%mm1,%%mm1\n"

  766. 

  767.       "add %3,%0\n"

  768.       "add %3,%1\n"

  769. 

  770.       "paddd %%mm2,%%mm1\n"

  771.       "paddd %%mm1,%%mm7\n"

  772. 

  773.       "decl %%ecx\n"

  774.       "jnz 1b\n"

  775. 

  776.       "movq %%mm7,%%mm1\n"

  777.       "psrlq $32, %%mm7\n"	/* shift hi dword to lo */

  778.       "paddd %%mm7,%%mm1\n"

  779.       "movd %%mm1,%2\n"

  780.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  781.       : "r" ((long)line_size) , "m" (h)

  782.       : "%ecx");

  783.     return tmp;

  784. }

  785. 

  786. static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  787.     int tmp;

  788.   asm volatile (

  789.       "movl %4,%%ecx\n"

  790.       "pxor %%mm0,%%mm0\n"	/* mm0 = 0 */

  791.       "pxor %%mm7,%%mm7\n"	/* mm7 holds the sum */

  792.       "1:\n"

  793.       "movq (%0),%%mm1\n"	/* mm1 = pix1[0-7] */

  794.       "movq (%1),%%mm2\n"	/* mm2 = pix2[0-7] */

  795.       "movq 8(%0),%%mm3\n"	/* mm3 = pix1[8-15] */

  796.       "movq 8(%1),%%mm4\n"	/* mm4 = pix2[8-15] */

  797. 

  798.       /* todo: mm1-mm2, mm3-mm4 */

  799.       /* algo: substract mm1 from mm2 with saturation and vice versa */

  800.       /*       OR the results to get absolute difference */

  801.       "movq %%mm1,%%mm5\n"

  802.       "movq %%mm3,%%mm6\n"

  803.       "psubusb %%mm2,%%mm1\n"

  804.       "psubusb %%mm4,%%mm3\n"

  805.       "psubusb %%mm5,%%mm2\n"

  806.       "psubusb %%mm6,%%mm4\n"

  807. 

  808.       "por %%mm1,%%mm2\n"

  809.       "por %%mm3,%%mm4\n"

  810. 

  811.       /* now convert to 16-bit vectors so we can square them */

  812.       "movq %%mm2,%%mm1\n"

  813.       "movq %%mm4,%%mm3\n"

  814. 

  815.       "punpckhbw %%mm0,%%mm2\n"

  816.       "punpckhbw %%mm0,%%mm4\n"

  817.       "punpcklbw %%mm0,%%mm1\n"	/* mm1 now spread over (mm1,mm2) */

  818.       "punpcklbw %%mm0,%%mm3\n"	/* mm4 now spread over (mm3,mm4) */

  819. 

  820.       "pmaddwd %%mm2,%%mm2\n"

  821.       "pmaddwd %%mm4,%%mm4\n"

  822.       "pmaddwd %%mm1,%%mm1\n"

  823.       "pmaddwd %%mm3,%%mm3\n"

  824. 

  825.       "add %3,%0\n"

  826.       "add %3,%1\n"

  827. 

  828.       "paddd %%mm2,%%mm1\n"

  829.       "paddd %%mm4,%%mm3\n"

  830.       "paddd %%mm1,%%mm7\n"

  831.       "paddd %%mm3,%%mm7\n"

  832. 

  833.       "decl %%ecx\n"

  834.       "jnz 1b\n"

  835. 

  836.       "movq %%mm7,%%mm1\n"

  837.       "psrlq $32, %%mm7\n"	/* shift hi dword to lo */

  838.       "paddd %%mm7,%%mm1\n"

  839.       "movd %%mm1,%2\n"

  840.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  841.       : "r" ((long)line_size) , "m" (h)

  842.       : "%ecx");

  843.     return tmp;

  844. }

  845. 

  846. static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {

  847.     int tmp;

  848.   asm volatile (

  849.       "movl %3,%%ecx\n"

  850.       "pxor %%mm7,%%mm7\n"

  851.       "pxor %%mm6,%%mm6\n"

  852.       

  853.       "movq (%0),%%mm0\n"

  854.       "movq %%mm0, %%mm1\n"

  855.       "psllq $8, %%mm0\n"

  856.       "psrlq $8, %%mm1\n"

  857.       "psrlq $8, %%mm0\n"

  858.       "movq %%mm0, %%mm2\n"

  859.       "movq %%mm1, %%mm3\n"

  860.       "punpcklbw %%mm7,%%mm0\n"

  861.       "punpcklbw %%mm7,%%mm1\n"

  862.       "punpckhbw %%mm7,%%mm2\n"

  863.       "punpckhbw %%mm7,%%mm3\n"

  864.       "psubw %%mm1, %%mm0\n"

  865.       "psubw %%mm3, %%mm2\n"

  866.       

  867.       "add %2,%0\n"

  868.       

  869.       "movq (%0),%%mm4\n"

  870.       "movq %%mm4, %%mm1\n"

  871.       "psllq $8, %%mm4\n"

  872.       "psrlq $8, %%mm1\n"

  873.       "psrlq $8, %%mm4\n"

  874.       "movq %%mm4, %%mm5\n"

  875.       "movq %%mm1, %%mm3\n"

  876.       "punpcklbw %%mm7,%%mm4\n"

  877.       "punpcklbw %%mm7,%%mm1\n"

  878.       "punpckhbw %%mm7,%%mm5\n"

  879.       "punpckhbw %%mm7,%%mm3\n"

  880.       "psubw %%mm1, %%mm4\n"

  881.       "psubw %%mm3, %%mm5\n"

  882.       "psubw %%mm4, %%mm0\n"

  883.       "psubw %%mm5, %%mm2\n"

  884.       "pxor %%mm3, %%mm3\n"

  885.       "pxor %%mm1, %%mm1\n"

  886.       "pcmpgtw %%mm0, %%mm3\n\t"

  887.       "pcmpgtw %%mm2, %%mm1\n\t"

  888.       "pxor %%mm3, %%mm0\n"

  889.       "pxor %%mm1, %%mm2\n"

  890.       "psubw %%mm3, %%mm0\n" 

  891.       "psubw %%mm1, %%mm2\n"

  892.       "paddw %%mm0, %%mm2\n"

  893.       "paddw %%mm2, %%mm6\n"

  894. 

  895.       "add %2,%0\n"

  896.       "1:\n"

  897.   

  898.       "movq (%0),%%mm0\n"

  899.       "movq %%mm0, %%mm1\n"

  900.       "psllq $8, %%mm0\n"

  901.       "psrlq $8, %%mm1\n"

  902.       "psrlq $8, %%mm0\n"

  903.       "movq %%mm0, %%mm2\n"

  904.       "movq %%mm1, %%mm3\n"

  905.       "punpcklbw %%mm7,%%mm0\n"

  906.       "punpcklbw %%mm7,%%mm1\n"

  907.       "punpckhbw %%mm7,%%mm2\n"

  908.       "punpckhbw %%mm7,%%mm3\n"

  909.       "psubw %%mm1, %%mm0\n"

  910.       "psubw %%mm3, %%mm2\n"

  911.       "psubw %%mm0, %%mm4\n"

  912.       "psubw %%mm2, %%mm5\n"

  913.       "pxor %%mm3, %%mm3\n"

  914.       "pxor %%mm1, %%mm1\n"

  915.       "pcmpgtw %%mm4, %%mm3\n\t"

  916.       "pcmpgtw %%mm5, %%mm1\n\t"

  917.       "pxor %%mm3, %%mm4\n"

  918.       "pxor %%mm1, %%mm5\n"

  919.       "psubw %%mm3, %%mm4\n" 

  920.       "psubw %%mm1, %%mm5\n"

  921.       "paddw %%mm4, %%mm5\n"

  922.       "paddw %%mm5, %%mm6\n"

  923.       

  924.       "add %2,%0\n"

  925.       

  926.       "movq (%0),%%mm4\n"

  927.       "movq %%mm4, %%mm1\n"

  928.       "psllq $8, %%mm4\n"

  929.       "psrlq $8, %%mm1\n"

  930.       "psrlq $8, %%mm4\n"

  931.       "movq %%mm4, %%mm5\n"

  932.       "movq %%mm1, %%mm3\n"

  933.       "punpcklbw %%mm7,%%mm4\n"

  934.       "punpcklbw %%mm7,%%mm1\n"

  935.       "punpckhbw %%mm7,%%mm5\n"

  936.       "punpckhbw %%mm7,%%mm3\n"

  937.       "psubw %%mm1, %%mm4\n"

  938.       "psubw %%mm3, %%mm5\n"

  939.       "psubw %%mm4, %%mm0\n"

  940.       "psubw %%mm5, %%mm2\n"

  941.       "pxor %%mm3, %%mm3\n"

  942.       "pxor %%mm1, %%mm1\n"

  943.       "pcmpgtw %%mm0, %%mm3\n\t"

  944.       "pcmpgtw %%mm2, %%mm1\n\t"

  945.       "pxor %%mm3, %%mm0\n"

  946.       "pxor %%mm1, %%mm2\n"

  947.       "psubw %%mm3, %%mm0\n" 

  948.       "psubw %%mm1, %%mm2\n"

  949.       "paddw %%mm0, %%mm2\n"

  950.       "paddw %%mm2, %%mm6\n"

  951. 

  952.       "add %2,%0\n"

  953.       "subl $2, %%ecx\n"

  954.       " jnz 1b\n"

  955. 

  956.       "movq %%mm6, %%mm0\n"

  957.       "punpcklwd %%mm7,%%mm0\n"

  958.       "punpckhwd %%mm7,%%mm6\n"

  959.       "paddd %%mm0, %%mm6\n"

  960.       

  961.       "movq %%mm6,%%mm0\n"

  962.       "psrlq $32, %%mm6\n"

  963.       "paddd %%mm6,%%mm0\n"

  964.       "movd %%mm0,%1\n"

  965.       : "+r" (pix1), "=r"(tmp) 

  966.       : "r" ((long)line_size) , "g" (h-2)

  967.       : "%ecx");

  968.       return tmp;

  969. }

  970. 

  971. static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {

  972.     int tmp;

  973.     uint8_t * pix= pix1;

  974.   asm volatile (

  975.       "movl %3,%%ecx\n"

  976.       "pxor %%mm7,%%mm7\n"

  977.       "pxor %%mm6,%%mm6\n"

  978.       

  979.       "movq (%0),%%mm0\n"

  980.       "movq 1(%0),%%mm1\n"

  981.       "movq %%mm0, %%mm2\n"

  982.       "movq %%mm1, %%mm3\n"

  983.       "punpcklbw %%mm7,%%mm0\n"

  984.       "punpcklbw %%mm7,%%mm1\n"

  985.       "punpckhbw %%mm7,%%mm2\n"

  986.       "punpckhbw %%mm7,%%mm3\n"

  987.       "psubw %%mm1, %%mm0\n"

  988.       "psubw %%mm3, %%mm2\n"

  989.       

  990.       "add %2,%0\n"

  991.       

  992.       "movq (%0),%%mm4\n"

  993.       "movq 1(%0),%%mm1\n"

  994.       "movq %%mm4, %%mm5\n"

  995.       "movq %%mm1, %%mm3\n"

  996.       "punpcklbw %%mm7,%%mm4\n"

  997.       "punpcklbw %%mm7,%%mm1\n"

  998.       "punpckhbw %%mm7,%%mm5\n"

  999.       "punpckhbw %%mm7,%%mm3\n"

  1000.       "psubw %%mm1, %%mm4\n"

  1001.       "psubw %%mm3, %%mm5\n"

  1002.       "psubw %%mm4, %%mm0\n"

  1003.       "psubw %%mm5, %%mm2\n"

  1004.       "pxor %%mm3, %%mm3\n"

  1005.       "pxor %%mm1, %%mm1\n"

  1006.       "pcmpgtw %%mm0, %%mm3\n\t"

  1007.       "pcmpgtw %%mm2, %%mm1\n\t"

  1008.       "pxor %%mm3, %%mm0\n"

  1009.       "pxor %%mm1, %%mm2\n"

  1010.       "psubw %%mm3, %%mm0\n" 

  1011.       "psubw %%mm1, %%mm2\n"

  1012.       "paddw %%mm0, %%mm2\n"

  1013.       "paddw %%mm2, %%mm6\n"

  1014. 

  1015.       "add %2,%0\n"

  1016.       "1:\n"

  1017.   

  1018.       "movq (%0),%%mm0\n"

  1019.       "movq 1(%0),%%mm1\n"

  1020.       "movq %%mm0, %%mm2\n"

  1021.       "movq %%mm1, %%mm3\n"

  1022.       "punpcklbw %%mm7,%%mm0\n"

  1023.       "punpcklbw %%mm7,%%mm1\n"

  1024.       "punpckhbw %%mm7,%%mm2\n"

  1025.       "punpckhbw %%mm7,%%mm3\n"

  1026.       "psubw %%mm1, %%mm0\n"

  1027.       "psubw %%mm3, %%mm2\n"

  1028.       "psubw %%mm0, %%mm4\n"

  1029.       "psubw %%mm2, %%mm5\n"

  1030.       "pxor %%mm3, %%mm3\n"

  1031.       "pxor %%mm1, %%mm1\n"

  1032.       "pcmpgtw %%mm4, %%mm3\n\t"

  1033.       "pcmpgtw %%mm5, %%mm1\n\t"

  1034.       "pxor %%mm3, %%mm4\n"

  1035.       "pxor %%mm1, %%mm5\n"

  1036.       "psubw %%mm3, %%mm4\n"

  1037.       "psubw %%mm1, %%mm5\n"

  1038.       "paddw %%mm4, %%mm5\n"

  1039.       "paddw %%mm5, %%mm6\n"

  1040.       

  1041.       "add %2,%0\n"

  1042.       

  1043.       "movq (%0),%%mm4\n"

  1044.       "movq 1(%0),%%mm1\n"

  1045.       "movq %%mm4, %%mm5\n"

  1046.       "movq %%mm1, %%mm3\n"

  1047.       "punpcklbw %%mm7,%%mm4\n"

  1048.       "punpcklbw %%mm7,%%mm1\n"

  1049.       "punpckhbw %%mm7,%%mm5\n"

  1050.       "punpckhbw %%mm7,%%mm3\n"

  1051.       "psubw %%mm1, %%mm4\n"

  1052.       "psubw %%mm3, %%mm5\n"

  1053.       "psubw %%mm4, %%mm0\n"

  1054.       "psubw %%mm5, %%mm2\n"

  1055.       "pxor %%mm3, %%mm3\n"

  1056.       "pxor %%mm1, %%mm1\n"

  1057.       "pcmpgtw %%mm0, %%mm3\n\t"

  1058.       "pcmpgtw %%mm2, %%mm1\n\t"

  1059.       "pxor %%mm3, %%mm0\n"

  1060.       "pxor %%mm1, %%mm2\n"

  1061.       "psubw %%mm3, %%mm0\n" 

  1062.       "psubw %%mm1, %%mm2\n"

  1063.       "paddw %%mm0, %%mm2\n"

  1064.       "paddw %%mm2, %%mm6\n"

  1065. 

  1066.       "add %2,%0\n"

  1067.       "subl $2, %%ecx\n"

  1068.       " jnz 1b\n"

  1069. 

  1070.       "movq %%mm6, %%mm0\n"

  1071.       "punpcklwd %%mm7,%%mm0\n"

  1072.       "punpckhwd %%mm7,%%mm6\n"

  1073.       "paddd %%mm0, %%mm6\n"

  1074.       

  1075.       "movq %%mm6,%%mm0\n"

  1076.       "psrlq $32, %%mm6\n"

  1077.       "paddd %%mm6,%%mm0\n"

  1078.       "movd %%mm0,%1\n"

  1079.       : "+r" (pix1), "=r"(tmp) 

  1080.       : "r" ((long)line_size) , "g" (h-2)

  1081.       : "%ecx");

  1082.       return tmp + hf_noise8_mmx(pix+8, line_size, h);

  1083. }

  1084. 

  1085. static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1086.     MpegEncContext *c = p;

  1087.     int score1= sse16_mmx(c, pix1, pix2, line_size, h);

  1088.     int score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

  1089. 

  1090.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1091.     else  return score1 + ABS(score2)*8;

  1092. }

  1093. 

  1094. static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1095.     MpegEncContext *c = p;

  1096.     int score1= sse8_mmx(c, pix1, pix2, line_size, h);

  1097.     int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

  1098. 

  1099.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1100.     else  return score1 + ABS(score2)*8;

  1101. }

  1102. 

  1103. static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1104.     int tmp;

  1105.     

  1106.     assert( (((int)pix) & 7) == 0);

  1107.     assert((line_size &7) ==0);

  1108.     

  1109. #define SUM(in0, in1, out0, out1) \

  1110.       "movq (%0), %%mm2\n"\

  1111.       "movq 8(%0), %%mm3\n"\

  1112.       "add %2,%0\n"\

  1113.       "movq %%mm2, " #out0 "\n"\

  1114.       "movq %%mm3, " #out1 "\n"\

  1115.       "psubusb " #in0 ", %%mm2\n"\

  1116.       "psubusb " #in1 ", %%mm3\n"\

  1117.       "psubusb " #out0 ", " #in0 "\n"\

  1118.       "psubusb " #out1 ", " #in1 "\n"\

  1119.       "por %%mm2, " #in0 "\n"\

  1120.       "por %%mm3, " #in1 "\n"\

  1121.       "movq " #in0 ", %%mm2\n"\

  1122.       "movq " #in1 ", %%mm3\n"\

  1123.       "punpcklbw %%mm7, " #in0 "\n"\

  1124.       "punpcklbw %%mm7, " #in1 "\n"\

  1125.       "punpckhbw %%mm7, %%mm2\n"\

  1126.       "punpckhbw %%mm7, %%mm3\n"\

  1127.       "paddw " #in1 ", " #in0 "\n"\

  1128.       "paddw %%mm3, %%mm2\n"\

  1129.       "paddw %%mm2, " #in0 "\n"\

  1130.       "paddw " #in0 ", %%mm6\n"

  1131. 

  1132.     

  1133.   asm volatile (

  1134.       "movl %3,%%ecx\n"

  1135.       "pxor %%mm6,%%mm6\n"

  1136.       "pxor %%mm7,%%mm7\n"

  1137.       "movq (%0),%%mm0\n"

  1138.       "movq 8(%0),%%mm1\n"

  1139.       "add %2,%0\n"

  1140.       "subl $2, %%ecx\n"

  1141.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1142.       "1:\n"

  1143.       

  1144.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1145.       

  1146.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1147.       

  1148.       "subl $2, %%ecx\n"

  1149.       "jnz 1b\n"

  1150. 

  1151.       "movq %%mm6,%%mm0\n"

  1152.       "psrlq $32, %%mm6\n"

  1153.       "paddw %%mm6,%%mm0\n"

  1154.       "movq %%mm0,%%mm6\n"

  1155.       "psrlq $16, %%mm0\n"

  1156.       "paddw %%mm6,%%mm0\n"

  1157.       "movd %%mm0,%1\n"

  1158.       : "+r" (pix), "=r"(tmp) 

  1159.       : "r" ((long)line_size) , "m" (h)

  1160.       : "%ecx");

  1161.     return tmp & 0xFFFF;

  1162. }

  1163. #undef SUM

  1164. 

  1165. static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1166.     int tmp;

  1167.     

  1168.     assert( (((int)pix) & 7) == 0);

  1169.     assert((line_size &7) ==0);

  1170.     

  1171. #define SUM(in0, in1, out0, out1) \

  1172.       "movq (%0), " #out0 "\n"\

  1173.       "movq 8(%0), " #out1 "\n"\

  1174.       "add %2,%0\n"\

  1175.       "psadbw " #out0 ", " #in0 "\n"\

  1176.       "psadbw " #out1 ", " #in1 "\n"\

  1177.       "paddw " #in1 ", " #in0 "\n"\

  1178.       "paddw " #in0 ", %%mm6\n"

  1179. 

  1180.   asm volatile (

  1181.       "movl %3,%%ecx\n"

  1182.       "pxor %%mm6,%%mm6\n"

  1183.       "pxor %%mm7,%%mm7\n"

  1184.       "movq (%0),%%mm0\n"

  1185.       "movq 8(%0),%%mm1\n"

  1186.       "add %2,%0\n"

  1187.       "subl $2, %%ecx\n"

  1188.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1189.       "1:\n"

  1190.       

  1191.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1192.       

  1193.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1194.       

  1195.       "subl $2, %%ecx\n"

  1196.       "jnz 1b\n"

  1197. 

  1198.       "movd %%mm6,%1\n"

  1199.       : "+r" (pix), "=r"(tmp) 

  1200.       : "r" ((long)line_size) , "m" (h)

  1201.       : "%ecx");

  1202.     return tmp;

  1203. }

  1204. #undef SUM

  1205. 

  1206. static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1207.     int tmp;

  1208.     

  1209.     assert( (((int)pix1) & 7) == 0);

  1210.     assert( (((int)pix2) & 7) == 0);

  1211.     assert((line_size &7) ==0);

  1212.     

  1213. #define SUM(in0, in1, out0, out1) \

  1214.       "movq (%0),%%mm2\n"\

  1215.       "movq (%1)," #out0 "\n"\

  1216.       "movq 8(%0),%%mm3\n"\

  1217.       "movq 8(%1)," #out1 "\n"\

  1218.       "add %3,%0\n"\

  1219.       "add %3,%1\n"\

  1220.       "psubb " #out0 ", %%mm2\n"\

  1221.       "psubb " #out1 ", %%mm3\n"\

  1222.       "pxor %%mm7, %%mm2\n"\

  1223.       "pxor %%mm7, %%mm3\n"\

  1224.       "movq %%mm2, " #out0 "\n"\

  1225.       "movq %%mm3, " #out1 "\n"\

  1226.       "psubusb " #in0 ", %%mm2\n"\

  1227.       "psubusb " #in1 ", %%mm3\n"\

  1228.       "psubusb " #out0 ", " #in0 "\n"\

  1229.       "psubusb " #out1 ", " #in1 "\n"\

  1230.       "por %%mm2, " #in0 "\n"\

  1231.       "por %%mm3, " #in1 "\n"\

  1232.       "movq " #in0 ", %%mm2\n"\

  1233.       "movq " #in1 ", %%mm3\n"\

  1234.       "punpcklbw %%mm7, " #in0 "\n"\

  1235.       "punpcklbw %%mm7, " #in1 "\n"\

  1236.       "punpckhbw %%mm7, %%mm2\n"\

  1237.       "punpckhbw %%mm7, %%mm3\n"\

  1238.       "paddw " #in1 ", " #in0 "\n"\

  1239.       "paddw %%mm3, %%mm2\n"\

  1240.       "paddw %%mm2, " #in0 "\n"\

  1241.       "paddw " #in0 ", %%mm6\n"

  1242. 

  1243.     

  1244.   asm volatile (

  1245.       "movl %4,%%ecx\n"

  1246.       "pxor %%mm6,%%mm6\n"

  1247.       "pcmpeqw %%mm7,%%mm7\n"

  1248.       "psllw $15, %%mm7\n"

  1249.       "packsswb %%mm7, %%mm7\n"

  1250.       "movq (%0),%%mm0\n"

  1251.       "movq (%1),%%mm2\n"

  1252.       "movq 8(%0),%%mm1\n"

  1253.       "movq 8(%1),%%mm3\n"

  1254.       "add %3,%0\n"

  1255.       "add %3,%1\n"

  1256.       "subl $2, %%ecx\n"

  1257.       "psubb %%mm2, %%mm0\n"

  1258.       "psubb %%mm3, %%mm1\n"

  1259.       "pxor %%mm7, %%mm0\n"

  1260.       "pxor %%mm7, %%mm1\n"

  1261.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1262.       "1:\n"

  1263.       

  1264.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1265.       

  1266.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1267.       

  1268.       "subl $2, %%ecx\n"

  1269.       "jnz 1b\n"

  1270. 

  1271.       "movq %%mm6,%%mm0\n"

  1272.       "psrlq $32, %%mm6\n"

  1273.       "paddw %%mm6,%%mm0\n"

  1274.       "movq %%mm0,%%mm6\n"

  1275.       "psrlq $16, %%mm0\n"

  1276.       "paddw %%mm6,%%mm0\n"

  1277.       "movd %%mm0,%2\n"

  1278.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  1279.       : "r" ((long)line_size) , "m" (h)

  1280.       : "%ecx");

  1281.     return tmp & 0x7FFF;

  1282. }

  1283. #undef SUM

  1284. 

  1285. static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1286.     int tmp;

  1287.     

  1288.     assert( (((int)pix1) & 7) == 0);

  1289.     assert( (((int)pix2) & 7) == 0);

  1290.     assert((line_size &7) ==0);

  1291.     

  1292. #define SUM(in0, in1, out0, out1) \

  1293.       "movq (%0)," #out0 "\n"\

  1294.       "movq (%1),%%mm2\n"\

  1295.       "movq 8(%0)," #out1 "\n"\

  1296.       "movq 8(%1),%%mm3\n"\

  1297.       "add %3,%0\n"\

  1298.       "add %3,%1\n"\

  1299.       "psubb %%mm2, " #out0 "\n"\

  1300.       "psubb %%mm3, " #out1 "\n"\

  1301.       "pxor %%mm7, " #out0 "\n"\

  1302.       "pxor %%mm7, " #out1 "\n"\

  1303.       "psadbw " #out0 ", " #in0 "\n"\

  1304.       "psadbw " #out1 ", " #in1 "\n"\

  1305.       "paddw " #in1 ", " #in0 "\n"\

  1306.       "paddw " #in0 ", %%mm6\n"

  1307. 

  1308.   asm volatile (

  1309.       "movl %4,%%ecx\n"

  1310.       "pxor %%mm6,%%mm6\n"

  1311.       "pcmpeqw %%mm7,%%mm7\n"

  1312.       "psllw $15, %%mm7\n"

  1313.       "packsswb %%mm7, %%mm7\n"

  1314.       "movq (%0),%%mm0\n"

  1315.       "movq (%1),%%mm2\n"

  1316.       "movq 8(%0),%%mm1\n"

  1317.       "movq 8(%1),%%mm3\n"

  1318.       "add %3,%0\n"

  1319.       "add %3,%1\n"

  1320.       "subl $2, %%ecx\n"

  1321.       "psubb %%mm2, %%mm0\n"

  1322.       "psubb %%mm3, %%mm1\n"

  1323.       "pxor %%mm7, %%mm0\n"

  1324.       "pxor %%mm7, %%mm1\n"

  1325.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1326.       "1:\n"

  1327.       

  1328.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1329.       

  1330.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1331.       

  1332.       "subl $2, %%ecx\n"

  1333.       "jnz 1b\n"

  1334. 

  1335.       "movd %%mm6,%2\n"

  1336.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  1337.       : "r" ((long)line_size) , "m" (h)

  1338.       : "%ecx");

  1339.     return tmp;

  1340. }

  1341. #undef SUM

  1342. 

  1343. static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){

  1344.     long i=0;

  1345.     asm volatile(

  1346.         "1:				\n\t"

  1347.         "movq  (%2, %0), %%mm0		\n\t"

  1348.         "movq  (%1, %0), %%mm1		\n\t"

  1349.         "psubb %%mm0, %%mm1		\n\t"

  1350.         "movq %%mm1, (%3, %0)		\n\t"

  1351.         "movq 8(%2, %0), %%mm0		\n\t"

  1352.         "movq 8(%1, %0), %%mm1		\n\t"

  1353.         "psubb %%mm0, %%mm1		\n\t"

  1354.         "movq %%mm1, 8(%3, %0)		\n\t"

  1355.         "add $16, %0			\n\t"

  1356.         "cmp %4, %0			\n\t"

  1357.         " jb 1b				\n\t"

  1358.         : "+r" (i)

  1359.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15)

  1360.     );

  1361.     for(; i<w; i++)

  1362.         dst[i+0] = src1[i+0]-src2[i+0];

  1363. }

  1364. 

  1365. static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){

  1366.     long i=0;

  1367.     uint8_t l, lt;

  1368.     

  1369.     asm volatile(

  1370.         "1:				\n\t"

  1371.         "movq  -1(%1, %0), %%mm0	\n\t" // LT

  1372.         "movq  (%1, %0), %%mm1		\n\t" // T

  1373.         "movq  -1(%2, %0), %%mm2	\n\t" // L

  1374.         "movq  (%2, %0), %%mm3		\n\t" // X

  1375.         "movq %%mm2, %%mm4		\n\t" // L

  1376.         "psubb %%mm0, %%mm2		\n\t"

  1377.         "paddb %%mm1, %%mm2		\n\t" // L + T - LT

  1378.         "movq %%mm4, %%mm5		\n\t" // L

  1379.         "pmaxub %%mm1, %%mm4		\n\t" // max(T, L)

  1380.         "pminub %%mm5, %%mm1		\n\t" // min(T, L)

  1381.         "pminub %%mm2, %%mm4		\n\t" 

  1382.         "pmaxub %%mm1, %%mm4		\n\t"

  1383.         "psubb %%mm4, %%mm3		\n\t" // dst - pred

  1384.         "movq %%mm3, (%3, %0)		\n\t"

  1385.         "add $8, %0			\n\t"

  1386.         "cmp %4, %0			\n\t"

  1387.         " jb 1b				\n\t"

  1388.         : "+r" (i)

  1389.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w)

  1390.     );

  1391. 

  1392.     l= *left;

  1393.     lt= *left_top;

  1394.     

  1395.     dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);

  1396.     

  1397.     *left_top= src1[w-1];

  1398.     *left    = src2[w-1];

  1399. }

  1400. 

  1401. #define LBUTTERFLY2(a1,b1,a2,b2)\

  1402.     "paddw " #b1 ", " #a1 "		\n\t"\

  1403.     "paddw " #b2 ", " #a2 "		\n\t"\

  1404.     "paddw " #b1 ", " #b1 "		\n\t"\

  1405.     "paddw " #b2 ", " #b2 "		\n\t"\

  1406.     "psubw " #a1 ", " #b1 "		\n\t"\

  1407.     "psubw " #a2 ", " #b2 "		\n\t"

  1408. 

  1409. #define HADAMARD48\

  1410.         LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\

  1411.         LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\

  1412.         LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\

  1413.         LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\

  1414.         LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\

  1415.         LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\

  1416. 

  1417. #define MMABS(a,z)\

  1418.     "pxor " #z ", " #z "		\n\t"\

  1419.     "pcmpgtw " #a ", " #z "		\n\t"\

  1420.     "pxor " #z ", " #a "		\n\t"\

  1421.     "psubw " #z ", " #a "		\n\t"

  1422. 

  1423. #define MMABS_SUM(a,z, sum)\

  1424.     "pxor " #z ", " #z "		\n\t"\

  1425.     "pcmpgtw " #a ", " #z "		\n\t"\

  1426.     "pxor " #z ", " #a "		\n\t"\

  1427.     "psubw " #z ", " #a "		\n\t"\

  1428.     "paddusw " #a ", " #sum "		\n\t"

  1429. 

  1430. #define MMABS_MMX2(a,z)\

  1431.     "pxor " #z ", " #z "		\n\t"\

  1432.     "psubw " #a ", " #z "		\n\t"\

  1433.     "pmaxsw " #z ", " #a "		\n\t"

  1434. 

  1435. #define MMABS_SUM_MMX2(a,z, sum)\

  1436.     "pxor " #z ", " #z "		\n\t"\

  1437.     "psubw " #a ", " #z "		\n\t"\

  1438.     "pmaxsw " #z ", " #a "		\n\t"\

  1439.     "paddusw " #a ", " #sum "		\n\t"

  1440.         

  1441. #define SBUTTERFLY(a,b,t,n)\

  1442.     "movq " #a ", " #t "		\n\t" /* abcd */\

  1443.     "punpckl" #n " " #b ", " #a "	\n\t" /* aebf */\

  1444.     "punpckh" #n " " #b ", " #t "	\n\t" /* cgdh */\

  1445. 

  1446. #define TRANSPOSE4(a,b,c,d,t)\

  1447.     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\

  1448.     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\

  1449.     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\

  1450.     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

  1451. 

  1452. #define LOAD4(o, a, b, c, d)\

  1453.         "movq "#o"(%1), " #a "		\n\t"\

  1454.         "movq "#o"+16(%1), " #b "	\n\t"\

  1455.         "movq "#o"+32(%1), " #c "	\n\t"\

  1456.         "movq "#o"+48(%1), " #d "	\n\t"

  1457. 

  1458. #define STORE4(o, a, b, c, d)\

  1459.         "movq "#a", "#o"(%1)		\n\t"\

  1460.         "movq "#b", "#o"+16(%1)		\n\t"\

  1461.         "movq "#c", "#o"+32(%1)		\n\t"\

  1462.         "movq "#d", "#o"+48(%1)		\n\t"\

  1463. 

  1464. static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1465.     uint64_t temp[16] __align8;

  1466.     int sum=0;

  1467.     

  1468.     assert(h==8);

  1469. 

  1470.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1471. 

  1472.     asm volatile(

  1473.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1474.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1475.         

  1476.         HADAMARD48

  1477.         

  1478.         "movq %%mm7, 112(%1)		\n\t"

  1479.         

  1480.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1481.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1482.         

  1483.         "movq 112(%1), %%mm7 		\n\t"

  1484.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1485.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1486. 

  1487.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1488.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1489.         

  1490.         HADAMARD48

  1491.         

  1492.         "movq %%mm7, 120(%1)		\n\t"

  1493.         

  1494.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1495.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1496.         

  1497.         "movq 120(%1), %%mm7 		\n\t"

  1498.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1499.         "movq %%mm7, %%mm5		\n\t"//FIXME remove

  1500.         "movq %%mm6, %%mm7		\n\t"

  1501.         "movq %%mm0, %%mm6		\n\t"

  1502. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1503.         

  1504.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1505. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1506.         

  1507.         HADAMARD48

  1508.         "movq %%mm7, 64(%1)		\n\t"

  1509.         MMABS(%%mm0, %%mm7)

  1510.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1511.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1512.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1513.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1514.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1515.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1516.         "movq 64(%1), %%mm1		\n\t"

  1517.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1518.         "movq %%mm0, 64(%1)		\n\t"

  1519.         

  1520.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1521.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1522.         

  1523.         HADAMARD48

  1524.         "movq %%mm7, (%1)		\n\t"

  1525.         MMABS(%%mm0, %%mm7)

  1526.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1527.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1528.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1529.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1530.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1531.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1532.         "movq (%1), %%mm1		\n\t"

  1533.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1534.         "movq 64(%1), %%mm1		\n\t"

  1535.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1536.         

  1537.         "movq %%mm0, %%mm1		\n\t"

  1538.         "psrlq $32, %%mm0		\n\t"

  1539.         "paddusw %%mm1, %%mm0		\n\t"

  1540.         "movq %%mm0, %%mm1		\n\t"

  1541.         "psrlq $16, %%mm0		\n\t"

  1542.         "paddusw %%mm1, %%mm0		\n\t"

  1543.         "movd %%mm0, %0			\n\t"

  1544.                 

  1545.         : "=r" (sum)

  1546.         : "r"(temp)

  1547.     );

  1548.     return sum&0xFFFF;

  1549. }

  1550. 

  1551. static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1552.     uint64_t temp[16] __align8;

  1553.     int sum=0;

  1554.     

  1555.     assert(h==8);

  1556. 

  1557.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1558. 

  1559.     asm volatile(

  1560.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1561.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1562.         

  1563.         HADAMARD48

  1564.         

  1565.         "movq %%mm7, 112(%1)		\n\t"

  1566.         

  1567.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1568.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1569.         

  1570.         "movq 112(%1), %%mm7 		\n\t"

  1571.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1572.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1573. 

  1574.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1575.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1576.         

  1577.         HADAMARD48

  1578.         

  1579.         "movq %%mm7, 120(%1)		\n\t"

  1580.         

  1581.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1582.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1583.         

  1584.         "movq 120(%1), %%mm7 		\n\t"

  1585.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1586.         "movq %%mm7, %%mm5		\n\t"//FIXME remove

  1587.         "movq %%mm6, %%mm7		\n\t"

  1588.         "movq %%mm0, %%mm6		\n\t"

  1589. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1590.         

  1591.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1592. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1593.         

  1594.         HADAMARD48

  1595.         "movq %%mm7, 64(%1)		\n\t"

  1596.         MMABS_MMX2(%%mm0, %%mm7)

  1597.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1598.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1599.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1600.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1601.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1602.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1603.         "movq 64(%1), %%mm1		\n\t"

  1604.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1605.         "movq %%mm0, 64(%1)		\n\t"

  1606.         

  1607.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1608.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1609.         

  1610.         HADAMARD48

  1611.         "movq %%mm7, (%1)		\n\t"

  1612.         MMABS_MMX2(%%mm0, %%mm7)

  1613.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1614.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1615.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1616.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1617.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1618.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1619.         "movq (%1), %%mm1		\n\t"

  1620.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1621.         "movq 64(%1), %%mm1		\n\t"

  1622.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1623.         

  1624.         "movq %%mm0, %%mm1		\n\t"

  1625.         "psrlq $32, %%mm0		\n\t"

  1626.         "paddusw %%mm1, %%mm0		\n\t"

  1627.         "movq %%mm0, %%mm1		\n\t"

  1628.         "psrlq $16, %%mm0		\n\t"

  1629.         "paddusw %%mm1, %%mm0		\n\t"

  1630.         "movd %%mm0, %0			\n\t"

  1631.                 

  1632.         : "=r" (sum)

  1633.         : "r"(temp)

  1634.     );

  1635.     return sum&0xFFFF;

  1636. }

  1637. 

  1638. 

  1639. WARPER8_16_SQ(hadamard8_diff_mmx, hadamard8_diff16_mmx)

  1640. WARPER8_16_SQ(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)

  1641. #endif //CONFIG_ENCODERS

  1642. 

  1643. #define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)

  1644. #define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

  1645. 

  1646. #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

  1647.         "paddw " #m4 ", " #m3 "		\n\t" /* x1 */\

  1648.         "movq "MANGLE(ff_pw_20)", %%mm4		\n\t" /* 20 */\

  1649.         "pmullw " #m3 ", %%mm4		\n\t" /* 20x1 */\

  1650.         "movq "#in7", " #m3 "		\n\t" /* d */\

  1651.         "movq "#in0", %%mm5		\n\t" /* D */\

  1652.         "paddw " #m3 ", %%mm5		\n\t" /* x4 */\

  1653.         "psubw %%mm5, %%mm4		\n\t" /* 20x1 - x4 */\

  1654.         "movq "#in1", %%mm5		\n\t" /* C */\

  1655.         "movq "#in2", %%mm6		\n\t" /* B */\

  1656.         "paddw " #m6 ", %%mm5		\n\t" /* x3 */\

  1657.         "paddw " #m5 ", %%mm6		\n\t" /* x2 */\

  1658.         "paddw %%mm6, %%mm6		\n\t" /* 2x2 */\

  1659.         "psubw %%mm6, %%mm5		\n\t" /* -2x2 + x3 */\

  1660.         "pmullw "MANGLE(ff_pw_3)", %%mm5	\n\t" /* -6x2 + 3x3 */\

  1661.         "paddw " #rnd ", %%mm4		\n\t" /* x2 */\

  1662.         "paddw %%mm4, %%mm5		\n\t" /* 20x1 - 6x2 + 3x3 - x4 */\

  1663.         "psraw $5, %%mm5		\n\t"\

  1664.         "packuswb %%mm5, %%mm5		\n\t"\

  1665.         OP(%%mm5, out, %%mm7, d)

  1666. 

  1667. #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\

  1668. static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1669.     uint64_t temp;\

  1670. \

  1671.     asm volatile(\

  1672.         "pxor %%mm7, %%mm7		\n\t"\

  1673.         "1:				\n\t"\

  1674.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  1675.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  1676.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  1677.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  1678.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  1679.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  1680.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  1681.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  1682.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  1683.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  1684.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  1685.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  1686.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  1687.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  1688.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  1689.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  1690.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  1691.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1692.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  1693.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  1694.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  1695.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  1696.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  1697.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  1698.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  1699.         "paddw %6, %%mm6		\n\t"\

  1700.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1701.         "psraw $5, %%mm0		\n\t"\

  1702.         "movq %%mm0, %5			\n\t"\

  1703.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1704.         \

  1705.         "movq 5(%0), %%mm0		\n\t" /* FGHIJKLM */\

  1706.         "movq %%mm0, %%mm5		\n\t" /* FGHIJKLM */\

  1707.         "movq %%mm0, %%mm6		\n\t" /* FGHIJKLM */\

  1708.         "psrlq $8, %%mm0		\n\t" /* GHIJKLM0 */\

  1709.         "psrlq $16, %%mm5		\n\t" /* HIJKLM00 */\

  1710.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0G0H0I0J */\

  1711.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0H0I0J0K */\

  1712.         "paddw %%mm0, %%mm2		\n\t" /* b */\

  1713.         "paddw %%mm5, %%mm3		\n\t" /* c */\

  1714.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  1715.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  1716.         "movq %%mm6, %%mm2		\n\t" /* FGHIJKLM */\

  1717.         "psrlq $24, %%mm6		\n\t" /* IJKLM000 */\

  1718.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0F0G0H0I */\

  1719.         "punpcklbw %%mm7, %%mm6		\n\t" /* 0I0J0K0L */\

  1720.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  1721.         "paddw %%mm2, %%mm1		\n\t" /* a */\

  1722.         "paddw %%mm6, %%mm4		\n\t" /* d */\

  1723.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  1724.         "psubw %%mm4, %%mm3		\n\t" /* - 6b +3c - d */\

  1725.         "paddw %6, %%mm1		\n\t"\

  1726.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b +3c - d */\

  1727.         "psraw $5, %%mm3		\n\t"\

  1728.         "movq %5, %%mm1			\n\t"\

  1729.         "packuswb %%mm3, %%mm1		\n\t"\

  1730.         OP_MMX2(%%mm1, (%1),%%mm4, q)\

  1731.         /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\

  1732.         \

  1733.         "movq 9(%0), %%mm1		\n\t" /* JKLMNOPQ */\

  1734.         "movq %%mm1, %%mm4		\n\t" /* JKLMNOPQ */\

  1735.         "movq %%mm1, %%mm3		\n\t" /* JKLMNOPQ */\

  1736.         "psrlq $8, %%mm1		\n\t" /* KLMNOPQ0 */\

  1737.         "psrlq $16, %%mm4		\n\t" /* LMNOPQ00 */\

  1738.         "punpcklbw %%mm7, %%mm1		\n\t" /* 0K0L0M0N */\

  1739.         "punpcklbw %%mm7, %%mm4		\n\t" /* 0L0M0N0O */\

  1740.         "paddw %%mm1, %%mm5		\n\t" /* b */\

  1741.         "paddw %%mm4, %%mm0		\n\t" /* c */\

  1742.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1743.         "psubw %%mm5, %%mm0		\n\t" /* c - 2b */\

  1744.         "movq %%mm3, %%mm5		\n\t" /* JKLMNOPQ */\

  1745.         "psrlq $24, %%mm3		\n\t" /* MNOPQ000 */\

  1746.         "pmullw "MANGLE(ff_pw_3)", %%mm0		\n\t" /* 3c - 6b */\

  1747.         "punpcklbw %%mm7, %%mm3		\n\t" /* 0M0N0O0P */\

  1748.         "paddw %%mm3, %%mm2		\n\t" /* d */\

  1749.         "psubw %%mm2, %%mm0		\n\t" /* -6b + 3c - d */\

  1750.         "movq %%mm5, %%mm2		\n\t" /* JKLMNOPQ */\

  1751.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0J0K0L0M */\

  1752.         "punpckhbw %%mm7, %%mm5		\n\t" /* 0N0O0P0Q */\

  1753.         "paddw %%mm2, %%mm6		\n\t" /* a */\

  1754.         "pmullw "MANGLE(ff_pw_20)", %%mm6		\n\t" /* 20a */\

  1755.         "paddw %6, %%mm0		\n\t"\

  1756.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1757.         "psraw $5, %%mm0		\n\t"\

  1758.         /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\

  1759.         \

  1760.         "paddw %%mm5, %%mm3		\n\t" /* a */\

  1761.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0O0P0Q0Q */\

  1762.         "paddw %%mm4, %%mm6		\n\t" /* b */\

  1763.         "pshufw $0xBE, %%mm5, %%mm4	\n\t" /* 0P0Q0Q0P */\

  1764.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0Q0Q0P0O */\

  1765.         "paddw %%mm1, %%mm4		\n\t" /* c */\

  1766.         "paddw %%mm2, %%mm5		\n\t" /* d */\

  1767.         "paddw %%mm6, %%mm6		\n\t" /* 2b */\

  1768.         "psubw %%mm6, %%mm4		\n\t" /* c - 2b */\

  1769.         "pmullw "MANGLE(ff_pw_20)", %%mm3		\n\t" /* 20a */\

  1770.         "pmullw "MANGLE(ff_pw_3)", %%mm4		\n\t" /* 3c - 6b */\

  1771.         "psubw %%mm5, %%mm3		\n\t" /* -6b + 3c - d */\

  1772.         "paddw %6, %%mm4		\n\t"\

  1773.         "paddw %%mm3, %%mm4		\n\t" /* 20a - 6b + 3c - d */\

  1774.         "psraw $5, %%mm4		\n\t"\

  1775.         "packuswb %%mm4, %%mm0		\n\t"\

  1776.         OP_MMX2(%%mm0, 8(%1), %%mm4, q)\

  1777.         \

  1778.         "add %3, %0			\n\t"\

  1779.         "add %4, %1			\n\t"\

  1780.         "decl %2			\n\t"\

  1781.         " jnz 1b				\n\t"\

  1782.         : "+a"(src), "+c"(dst), "+m"(h)\

  1783.         : "d"((long)srcStride), "S"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1784.         : "memory"\

  1785.     );\

  1786. }\

  1787. \

  1788. static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1789.     int i;\

  1790.     int16_t temp[16];\

  1791.     /* quick HACK, XXX FIXME MUST be optimized */\

  1792.     for(i=0; i<h; i++)\

  1793.     {\

  1794.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1795.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1796.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1797.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1798.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1799.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\

  1800.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\

  1801.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\

  1802.         temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\

  1803.         temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\

  1804.         temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\

  1805.         temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\

  1806.         temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\

  1807.         temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\

  1808.         temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\

  1809.         temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\

  1810.         asm volatile(\

  1811.             "movq (%0), %%mm0		\n\t"\

  1812.             "movq 8(%0), %%mm1		\n\t"\

  1813.             "paddw %2, %%mm0		\n\t"\

  1814.             "paddw %2, %%mm1		\n\t"\

  1815.             "psraw $5, %%mm0		\n\t"\

  1816.             "psraw $5, %%mm1		\n\t"\

  1817.             "packuswb %%mm1, %%mm0	\n\t"\

  1818.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1819.             "movq 16(%0), %%mm0		\n\t"\

  1820.             "movq 24(%0), %%mm1		\n\t"\

  1821.             "paddw %2, %%mm0		\n\t"\

  1822.             "paddw %2, %%mm1		\n\t"\

  1823.             "psraw $5, %%mm0		\n\t"\

  1824.             "psraw $5, %%mm1		\n\t"\

  1825.             "packuswb %%mm1, %%mm0	\n\t"\

  1826.             OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\

  1827.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1828.             : "memory"\

  1829.         );\

  1830.         dst+=dstStride;\

  1831.         src+=srcStride;\

  1832.     }\

  1833. }\

  1834. \

  1835. static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1836.     uint64_t temp;\

  1837. \

  1838.     asm volatile(\

  1839.         "pxor %%mm7, %%mm7		\n\t"\

  1840.         "1:				\n\t"\

  1841.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  1842.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  1843.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  1844.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  1845.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  1846.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  1847.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  1848.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  1849.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  1850.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  1851.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  1852.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  1853.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  1854.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  1855.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  1856.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  1857.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  1858.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1859.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  1860.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  1861.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  1862.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  1863.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  1864.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  1865.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  1866.         "paddw %6, %%mm6		\n\t"\

  1867.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1868.         "psraw $5, %%mm0		\n\t"\

  1869.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1870.         \

  1871.         "movd 5(%0), %%mm5		\n\t" /* FGHI */\

  1872.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0F0G0H0I */\

  1873.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0G0H0I0I */\

  1874.         "paddw %%mm5, %%mm1		\n\t" /* a */\

  1875.         "paddw %%mm6, %%mm2		\n\t" /* b */\

  1876.         "pshufw $0xBE, %%mm5, %%mm6	\n\t" /* 0H0I0I0H */\

  1877.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0I0I0H0G */\

  1878.         "paddw %%mm6, %%mm3		\n\t" /* c */\

  1879.         "paddw %%mm5, %%mm4		\n\t" /* d */\

  1880.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  1881.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  1882.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  1883.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  1884.         "psubw %%mm4, %%mm3		\n\t" /* -6b + 3c - d */\

  1885.         "paddw %6, %%mm1		\n\t"\

  1886.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b + 3c - d */\

  1887.         "psraw $5, %%mm3		\n\t"\

  1888.         "packuswb %%mm3, %%mm0		\n\t"\

  1889.         OP_MMX2(%%mm0, (%1), %%mm4, q)\

  1890.         \

  1891.         "add %3, %0			\n\t"\

  1892.         "add %4, %1			\n\t"\

  1893.         "decl %2			\n\t"\

  1894.         " jnz 1b			\n\t"\

  1895.         : "+a"(src), "+c"(dst), "+m"(h)\

  1896.         : "S"((long)srcStride), "D"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1897.         : "memory"\

  1898.     );\

  1899. }\

  1900. \

  1901. static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1902.     int i;\

  1903.     int16_t temp[8];\

  1904.     /* quick HACK, XXX FIXME MUST be optimized */\

  1905.     for(i=0; i<h; i++)\

  1906.     {\

  1907.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1908.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1909.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1910.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1911.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1912.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\

  1913.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\

  1914.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\

  1915.         asm volatile(\

  1916.             "movq (%0), %%mm0		\n\t"\

  1917.             "movq 8(%0), %%mm1		\n\t"\

  1918.             "paddw %2, %%mm0		\n\t"\

  1919.             "paddw %2, %%mm1		\n\t"\

  1920.             "psraw $5, %%mm0		\n\t"\

  1921.             "psraw $5, %%mm1		\n\t"\

  1922.             "packuswb %%mm1, %%mm0	\n\t"\

  1923.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1924.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1925.             :"memory"\

  1926.         );\

  1927.         dst+=dstStride;\

  1928.         src+=srcStride;\

  1929.     }\

  1930. }

  1931. 

  1932. #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\

  1933. \

  1934. static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1935.     uint64_t temp[17*4];\

  1936.     uint64_t *temp_ptr= temp;\

  1937.     int count= 17;\

  1938. \

  1939.     /*FIXME unroll */\

  1940.     asm volatile(\

  1941.         "pxor %%mm7, %%mm7		\n\t"\

  1942.         "1:				\n\t"\

  1943.         "movq (%0), %%mm0		\n\t"\

  1944.         "movq (%0), %%mm1		\n\t"\

  1945.         "movq 8(%0), %%mm2		\n\t"\

  1946.         "movq 8(%0), %%mm3		\n\t"\

  1947.         "punpcklbw %%mm7, %%mm0		\n\t"\

  1948.         "punpckhbw %%mm7, %%mm1		\n\t"\

  1949.         "punpcklbw %%mm7, %%mm2		\n\t"\

  1950.         "punpckhbw %%mm7, %%mm3		\n\t"\

  1951.         "movq %%mm0, (%1)		\n\t"\

  1952.         "movq %%mm1, 17*8(%1)		\n\t"\

  1953.         "movq %%mm2, 2*17*8(%1)		\n\t"\

  1954.         "movq %%mm3, 3*17*8(%1)		\n\t"\

  1955.         "add $8, %1			\n\t"\

  1956.         "add %3, %0			\n\t"\

  1957.         "decl %2			\n\t"\

  1958.         " jnz 1b			\n\t"\

  1959.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  1960.         : "r" ((long)srcStride)\

  1961.         : "memory"\

  1962.     );\

  1963.     \

  1964.     temp_ptr= temp;\

  1965.     count=4;\

  1966.     \

  1967. /*FIXME reorder for speed */\

  1968.     asm volatile(\

  1969.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  1970.         "1:				\n\t"\

  1971.         "movq (%0), %%mm0		\n\t"\

  1972.         "movq 8(%0), %%mm1		\n\t"\

  1973.         "movq 16(%0), %%mm2		\n\t"\

  1974.         "movq 24(%0), %%mm3		\n\t"\

  1975.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  1976.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  1977.         "add %4, %1			\n\t"\

  1978.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  1979.         \

  1980.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  1981.         "add %4, %1			\n\t"\

  1982.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  1983.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\

  1984.         "add %4, %1			\n\t"\

  1985.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\

  1986.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\

  1987.         "add %4, %1			\n\t"\

  1988.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\

  1989.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\

  1990.         "add %4, %1			\n\t"\

  1991.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\

  1992.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\

  1993.         "add %4, %1			\n\t"\

  1994.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\

  1995.         \

  1996.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\

  1997.         "add %4, %1			\n\t"  \

  1998.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\

  1999.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\

  2000.         \

  2001.         "add $136, %0			\n\t"\

  2002.         "add %6, %1			\n\t"\

  2003.         "decl %2			\n\t"\

  2004.         " jnz 1b			\n\t"\

  2005.         \

  2006.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2007.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(long)dstStride)\

  2008.         :"memory"\

  2009.     );\

  2010. }\

  2011. \

  2012. static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2013.     uint64_t temp[9*2];\

  2014.     uint64_t *temp_ptr= temp;\

  2015.     int count= 9;\

  2016. \

  2017.     /*FIXME unroll */\

  2018.     asm volatile(\

  2019.         "pxor %%mm7, %%mm7		\n\t"\

  2020.         "1:				\n\t"\

  2021.         "movq (%0), %%mm0		\n\t"\

  2022.         "movq (%0), %%mm1		\n\t"\

  2023.         "punpcklbw %%mm7, %%mm0		\n\t"\

  2024.         "punpckhbw %%mm7, %%mm1		\n\t"\

  2025.         "movq %%mm0, (%1)		\n\t"\

  2026.         "movq %%mm1, 9*8(%1)		\n\t"\

  2027.         "add $8, %1			\n\t"\

  2028.         "add %3, %0			\n\t"\

  2029.         "decl %2			\n\t"\

  2030.         " jnz 1b			\n\t"\

  2031.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  2032.         : "r" ((long)srcStride)\

  2033.         : "memory"\

  2034.     );\

  2035.     \

  2036.     temp_ptr= temp;\

  2037.     count=2;\

  2038.     \

  2039. /*FIXME reorder for speed */\

  2040.     asm volatile(\

  2041.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  2042.         "1:				\n\t"\

  2043.         "movq (%0), %%mm0		\n\t"\

  2044.         "movq 8(%0), %%mm1		\n\t"\

  2045.         "movq 16(%0), %%mm2		\n\t"\

  2046.         "movq 24(%0), %%mm3		\n\t"\

  2047.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  2048.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  2049.         "add %4, %1			\n\t"\

  2050.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  2051.         \

  2052.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  2053.         "add %4, %1			\n\t"\

  2054.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  2055.         \

  2056.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\

  2057.         "add %4, %1			\n\t"\

  2058.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\

  2059.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\

  2060.                 \

  2061.         "add $72, %0			\n\t"\

  2062.         "add %6, %1			\n\t"\

  2063.         "decl %2			\n\t"\

  2064.         " jnz 1b			\n\t"\

  2065.          \

  2066.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2067.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(long)dstStride)\

  2068.         : "memory"\

  2069.    );\

  2070. }\

  2071. \

  2072. static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2073.     OPNAME ## pixels8_mmx(dst, src, stride, 8);\

  2074. }\

  2075. \

  2076. static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2077.     uint64_t temp[8];\

  2078.     uint8_t * const half= (uint8_t*)temp;\

  2079.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2080.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2081. }\

  2082. \

  2083. static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2084.     OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\

  2085. }\

  2086. \

  2087. static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2088.     uint64_t temp[8];\

  2089.     uint8_t * const half= (uint8_t*)temp;\

  2090.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2091.     OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\

  2092. }\

  2093. \

  2094. static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2095.     uint64_t temp[8];\

  2096.     uint8_t * const half= (uint8_t*)temp;\

  2097.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2098.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2099. }\

  2100. \

  2101. static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2102.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2103. }\

  2104. \

  2105. static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2106.     uint64_t temp[8];\

  2107.     uint8_t * const half= (uint8_t*)temp;\

  2108.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2109.     OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\

  2110. }\

  2111. static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2112.     uint64_t half[8 + 9];\

  2113.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2114.     uint8_t * const halfHV= ((uint8_t*)half);\

  2115.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2116.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2117.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2118.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2119. }\

  2120. static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2121.     uint64_t half[8 + 9];\

  2122.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2123.     uint8_t * const halfHV= ((uint8_t*)half);\

  2124.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2125.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2126.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2127.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2128. }\

  2129. static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2130.     uint64_t half[8 + 9];\

  2131.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2132.     uint8_t * const halfHV= ((uint8_t*)half);\

  2133.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2134.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2135.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2136.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2137. }\

  2138. static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2139.     uint64_t half[8 + 9];\

  2140.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2141.     uint8_t * const halfHV= ((uint8_t*)half);\

  2142.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2143.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2144.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2145.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2146. }\

  2147. static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2148.     uint64_t half[8 + 9];\

  2149.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2150.     uint8_t * const halfHV= ((uint8_t*)half);\

  2151.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2152.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2153.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2154. }\

  2155. static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2156.     uint64_t half[8 + 9];\

  2157.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2158.     uint8_t * const halfHV= ((uint8_t*)half);\

  2159.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2160.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2161.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2162. }\

  2163. static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2164.     uint64_t half[8 + 9];\

  2165.     uint8_t * const halfH= ((uint8_t*)half);\

  2166.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2167.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2168.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2169. }\

  2170. static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2171.     uint64_t half[8 + 9];\

  2172.     uint8_t * const halfH= ((uint8_t*)half);\

  2173.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2174.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2175.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2176. }\

  2177. static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2178.     uint64_t half[9];\

  2179.     uint8_t * const halfH= ((uint8_t*)half);\

  2180.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2181.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2182. }\

  2183. static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2184.     OPNAME ## pixels16_mmx(dst, src, stride, 16);\

  2185. }\

  2186. \

  2187. static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2188.     uint64_t temp[32];\

  2189.     uint8_t * const half= (uint8_t*)temp;\

  2190.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2191.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2192. }\

  2193. \

  2194. static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2195.     OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\

  2196. }\

  2197. \

  2198. static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2199.     uint64_t temp[32];\

  2200.     uint8_t * const half= (uint8_t*)temp;\

  2201.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2202.     OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\

  2203. }\

  2204. \

  2205. static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2206.     uint64_t temp[32];\

  2207.     uint8_t * const half= (uint8_t*)temp;\

  2208.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2209.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2210. }\

  2211. \

  2212. static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2213.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2214. }\

  2215. \

  2216. static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2217.     uint64_t temp[32];\

  2218.     uint8_t * const half= (uint8_t*)temp;\

  2219.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2220.     OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\

  2221. }\

  2222. static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2223.     uint64_t half[16*2 + 17*2];\

  2224.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2225.     uint8_t * const halfHV= ((uint8_t*)half);\

  2226.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2227.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2228.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2229.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2230. }\

  2231. static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2232.     uint64_t half[16*2 + 17*2];\

  2233.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2234.     uint8_t * const halfHV= ((uint8_t*)half);\

  2235.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2236.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2237.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2238.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2239. }\

  2240. static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2241.     uint64_t half[16*2 + 17*2];\

  2242.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2243.     uint8_t * const halfHV= ((uint8_t*)half);\

  2244.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2245.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2246.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2247.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2248. }\

  2249. static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2250.     uint64_t half[16*2 + 17*2];\

  2251.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2252.     uint8_t * const halfHV= ((uint8_t*)half);\

  2253.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2254.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2255.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2256.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2257. }\

  2258. static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2259.     uint64_t half[16*2 + 17*2];\

  2260.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2261.     uint8_t * const halfHV= ((uint8_t*)half);\

  2262.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2263.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2264.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2265. }\

  2266. static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2267.     uint64_t half[16*2 + 17*2];\

  2268.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2269.     uint8_t * const halfHV= ((uint8_t*)half);\

  2270.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2271.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2272.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2273. }\

  2274. static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2275.     uint64_t half[17*2];\

  2276.     uint8_t * const halfH= ((uint8_t*)half);\

  2277.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2278.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2279.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2280. }\

  2281. static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2282.     uint64_t half[17*2];\

  2283.     uint8_t * const halfH= ((uint8_t*)half);\

  2284.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2285.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2286.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2287. }\

  2288. static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2289.     uint64_t half[17*2];\

  2290.     uint8_t * const halfH= ((uint8_t*)half);\

  2291.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2292.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2293. }

  2294. 

  2295. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "	\n\t"

  2296. #define AVG_3DNOW_OP(a,b,temp, size) \

  2297. "mov" #size " " #b ", " #temp "	\n\t"\

  2298. "pavgusb " #temp ", " #a "	\n\t"\

  2299. "mov" #size " " #a ", " #b "	\n\t"

  2300. #define AVG_MMX2_OP(a,b,temp, size) \

  2301. "mov" #size " " #b ", " #temp "	\n\t"\

  2302. "pavgb " #temp ", " #a "	\n\t"\

  2303. "mov" #size " " #a ", " #b "	\n\t"

  2304. 

  2305. QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)

  2306. QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)

  2307. QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)

  2308. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)

  2309. QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)

  2310. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)

  2311. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)

  2312. QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)

  2313. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

  2314. 

  2315. #if 0

  2316. static void just_return() { return; }

  2317. #endif

  2318. 

  2319. #define SET_QPEL_FUNC(postfix1, postfix2) \

  2320.     c->put_ ## postfix1 = put_ ## postfix2;\

  2321.     c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\

  2322.     c->avg_ ## postfix1 = avg_ ## postfix2;

  2323. 

  2324. static int try_8x8basis_mmx(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){

  2325.     long i=0;

  2326.     

  2327.     assert(ABS(scale) < 256);

  2328.     scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2329. 

  2330.     asm volatile(

  2331.         "pcmpeqw %%mm6, %%mm6		\n\t" // -1w

  2332.         "psrlw $15, %%mm6		\n\t" //  1w

  2333.         "pxor %%mm7, %%mm7		\n\t"

  2334.         "movd  %4, %%mm5		\n\t" 

  2335.         "punpcklwd %%mm5, %%mm5		\n\t" 

  2336.         "punpcklwd %%mm5, %%mm5		\n\t" 

  2337.         "1:				\n\t"

  2338.         "movq  (%1, %0), %%mm0		\n\t" 

  2339.         "movq  8(%1, %0), %%mm1		\n\t"

  2340.         "pmulhw %%mm5, %%mm0		\n\t"

  2341.         "pmulhw %%mm5, %%mm1		\n\t"

  2342.         "paddw %%mm6, %%mm0		\n\t"

  2343.         "paddw %%mm6, %%mm1		\n\t"

  2344.         "psraw $1, %%mm0		\n\t"

  2345.         "psraw $1, %%mm1		\n\t"

  2346.         "paddw (%2, %0), %%mm0		\n\t"

  2347.         "paddw 8(%2, %0), %%mm1		\n\t"

  2348.         "psraw $6, %%mm0		\n\t"

  2349.         "psraw $6, %%mm1		\n\t"

  2350.         "pmullw (%3, %0), %%mm0		\n\t"

  2351.         "pmullw 8(%3, %0), %%mm1	\n\t"

  2352.         "pmaddwd %%mm0, %%mm0		\n\t"

  2353.         "pmaddwd %%mm1, %%mm1		\n\t"

  2354.         "paddd %%mm1, %%mm0		\n\t"

  2355.         "psrld $4, %%mm0		\n\t"

  2356.         "paddd %%mm0, %%mm7		\n\t"

  2357.         "add $16, %0			\n\t"

  2358.         "cmp $128, %0			\n\t" //FIXME optimize & bench

  2359.         " jb 1b				\n\t"

  2360.         "movq %%mm7, %%mm6		\n\t"

  2361.         "psrlq $32, %%mm7		\n\t"

  2362.         "paddd %%mm6, %%mm7		\n\t"

  2363.         "psrld $2, %%mm7		\n\t"

  2364.         "movd %%mm7, %0			\n\t"

  2365.         

  2366.         : "+r" (i)

  2367.         : "r"(basis), "r"(rem), "r"(weight), "g"(scale)

  2368.     );

  2369.     return i;

  2370. }

  2371. 

  2372. static void add_8x8basis_mmx(int16_t rem[64], int16_t basis[64], int scale){

  2373.     long i=0;

  2374.     

  2375.     if(ABS(scale) < 256){

  2376.         scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2377.         asm volatile(

  2378.                 "pcmpeqw %%mm6, %%mm6		\n\t" // -1w

  2379.                 "psrlw $15, %%mm6		\n\t" //  1w

  2380.                 "movd  %3, %%mm5		\n\t" 

  2381.                 "punpcklwd %%mm5, %%mm5		\n\t" 

  2382.                 "punpcklwd %%mm5, %%mm5		\n\t" 

  2383.                 "1:				\n\t"

  2384.                 "movq  (%1, %0), %%mm0		\n\t" 

  2385.                 "movq  8(%1, %0), %%mm1		\n\t"

  2386.                 "pmulhw %%mm5, %%mm0		\n\t"

  2387.                 "pmulhw %%mm5, %%mm1		\n\t"

  2388.                 "paddw %%mm6, %%mm0		\n\t" 

  2389.                 "paddw %%mm6, %%mm1		\n\t"

  2390.                 "psraw $1, %%mm0		\n\t"

  2391.                 "psraw $1, %%mm1		\n\t"

  2392.                 "paddw (%2, %0), %%mm0		\n\t"

  2393.                 "paddw 8(%2, %0), %%mm1		\n\t"

  2394.                 "movq %%mm0, (%2, %0)		\n\t"

  2395.                 "movq %%mm1, 8(%2, %0)		\n\t"

  2396.                 "add $16, %0			\n\t"

  2397.                 "cmp $128, %0			\n\t" //FIXME optimize & bench

  2398.                 " jb 1b				\n\t"

  2399.                 

  2400.                 : "+r" (i)

  2401.                 : "r"(basis), "r"(rem), "g"(scale)

  2402.         );

  2403.     }else{

  2404.         for(i=0; i<8*8; i++){

  2405.             rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);

  2406.         }    

  2407.     }

  2408. }

  2409. 

  2410. #include "h264dsp_mmx.c"

  2411.     

  2412. /* external functions, from idct_mmx.c */

  2413. void ff_mmx_idct(DCTELEM *block);

  2414. void ff_mmxext_idct(DCTELEM *block);

  2415. 

  2416. void ff_vp3_idct_sse2(int16_t *input_data);

  2417. void ff_vp3_idct_mmx(int16_t *data);

  2418. void ff_vp3_dsp_init_mmx(void);

  2419. 

  2420. /* XXX: those functions should be suppressed ASAP when all IDCTs are

  2421.    converted */

  2422. static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2423. {

  2424.     ff_mmx_idct (block);

  2425.     put_pixels_clamped_mmx(block, dest, line_size);

  2426. }

  2427. static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2428. {

  2429.     ff_mmx_idct (block);

  2430.     add_pixels_clamped_mmx(block, dest, line_size);

  2431. }

  2432. static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2433. {

  2434.     ff_mmxext_idct (block);

  2435.     put_pixels_clamped_mmx(block, dest, line_size);

  2436. }

  2437. static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2438. {

  2439.     ff_mmxext_idct (block);

  2440.     add_pixels_clamped_mmx(block, dest, line_size);

  2441. }

  2442. static void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block)

  2443. {

  2444.     ff_vp3_idct_sse2(block);

  2445.     put_signed_pixels_clamped_mmx(block, dest, line_size);

  2446. }

  2447. static void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block)

  2448. {

  2449.     ff_vp3_idct_sse2(block);

  2450.     add_pixels_clamped_mmx(block, dest, line_size);

  2451. }

  2452. static void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)

  2453. {

  2454.     ff_vp3_idct_mmx(block);

  2455.     put_signed_pixels_clamped_mmx(block, dest, line_size);

  2456. }

  2457. static void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)

  2458. {

  2459.     ff_vp3_idct_mmx(block);

  2460.     add_pixels_clamped_mmx(block, dest, line_size);

  2461. }

  2462. #ifdef CONFIG_GPL

  2463. static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)

  2464. {

  2465.     ff_idct_xvid_mmx (block);

  2466.     put_pixels_clamped_mmx(block, dest, line_size);

  2467. }

  2468. static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)

  2469. {

  2470.     ff_idct_xvid_mmx (block);

  2471.     add_pixels_clamped_mmx(block, dest, line_size);

  2472. }

  2473. static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)

  2474. {

  2475.     ff_idct_xvid_mmx2 (block);

  2476.     put_pixels_clamped_mmx(block, dest, line_size);

  2477. }

  2478. static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)

  2479. {

  2480.     ff_idct_xvid_mmx2 (block);

  2481.     add_pixels_clamped_mmx(block, dest, line_size);

  2482. }

  2483. #endif

  2484.     

  2485. void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)

  2486. {

  2487.     mm_flags = mm_support();

  2488. 

  2489.     if (avctx->dsp_mask) {

  2490. 	if (avctx->dsp_mask & FF_MM_FORCE)

  2491. 	    mm_flags |= (avctx->dsp_mask & 0xffff);

  2492. 	else

  2493. 	    mm_flags &= ~(avctx->dsp_mask & 0xffff);

  2494.     }

  2495. 

  2496. #if 0

  2497.     av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");

  2498.     if (mm_flags & MM_MMX)

  2499.         av_log(avctx, AV_LOG_INFO, " mmx");

  2500.     if (mm_flags & MM_MMXEXT)

  2501.         av_log(avctx, AV_LOG_INFO, " mmxext");

  2502.     if (mm_flags & MM_3DNOW)

  2503.         av_log(avctx, AV_LOG_INFO, " 3dnow");

  2504.     if (mm_flags & MM_SSE)

  2505.         av_log(avctx, AV_LOG_INFO, " sse");

  2506.     if (mm_flags & MM_SSE2)

  2507.         av_log(avctx, AV_LOG_INFO, " sse2");

  2508.     av_log(avctx, AV_LOG_INFO, "\n");

  2509. #endif

  2510. 

  2511.     if (mm_flags & MM_MMX) {

  2512.         const int idct_algo= avctx->idct_algo;

  2513. 

  2514. #ifdef CONFIG_ENCODERS

  2515.         const int dct_algo = avctx->dct_algo;

  2516.         if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){

  2517.             if(mm_flags & MM_SSE2){

  2518.                 c->fdct = ff_fdct_sse2;

  2519. 	    }else if(mm_flags & MM_MMXEXT){

  2520.                 c->fdct = ff_fdct_mmx2;

  2521.             }else{

  2522.                 c->fdct = ff_fdct_mmx;

  2523.             }

  2524.         }

  2525. #endif //CONFIG_ENCODERS

  2526.         if(avctx->lowres==0){

  2527.             if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){

  2528.                 c->idct_put= ff_simple_idct_put_mmx;

  2529.                 c->idct_add= ff_simple_idct_add_mmx;

  2530.                 c->idct    = ff_simple_idct_mmx;

  2531.                 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;

  2532.             }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){

  2533.                 if(mm_flags & MM_MMXEXT){

  2534.                     c->idct_put= ff_libmpeg2mmx2_idct_put;

  2535.                     c->idct_add= ff_libmpeg2mmx2_idct_add;

  2536.                     c->idct    = ff_mmxext_idct;

  2537.                 }else{

  2538.                     c->idct_put= ff_libmpeg2mmx_idct_put;

  2539.                     c->idct_add= ff_libmpeg2mmx_idct_add;

  2540.                     c->idct    = ff_mmx_idct;

  2541.                 }

  2542.                 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;

  2543.             }else if(idct_algo==FF_IDCT_VP3){

  2544.                 if(mm_flags & MM_SSE2){

  2545.                     c->idct_put= ff_vp3_idct_put_sse2;

  2546.                     c->idct_add= ff_vp3_idct_add_sse2;

  2547.                     c->idct    = ff_vp3_idct_sse2;

  2548.                     c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;

  2549.                 }else{

  2550.                     ff_vp3_dsp_init_mmx();

  2551.                     c->idct_put= ff_vp3_idct_put_mmx;

  2552.                     c->idct_add= ff_vp3_idct_add_mmx;

  2553.                     c->idct    = ff_vp3_idct_mmx;

  2554.                     c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;

  2555.                 }

  2556. #ifdef CONFIG_GPL

  2557.             }else if(idct_algo==FF_IDCT_XVIDMMX){

  2558.                 if(mm_flags & MM_MMXEXT){

  2559.                     c->idct_put= ff_idct_xvid_mmx2_put;

  2560.                     c->idct_add= ff_idct_xvid_mmx2_add;

  2561.                     c->idct    = ff_idct_xvid_mmx2;

  2562.                 }else{

  2563.                     c->idct_put= ff_idct_xvid_mmx_put;

  2564.                     c->idct_add= ff_idct_xvid_mmx_add;

  2565.                     c->idct    = ff_idct_xvid_mmx;

  2566.                 }

  2567. #endif

  2568.             }

  2569.         }

  2570. 

  2571. #ifdef CONFIG_ENCODERS

  2572.         c->get_pixels = get_pixels_mmx;

  2573.         c->diff_pixels = diff_pixels_mmx;

  2574. #endif //CONFIG_ENCODERS

  2575.         c->put_pixels_clamped = put_pixels_clamped_mmx;

  2576.         c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;

  2577.         c->add_pixels_clamped = add_pixels_clamped_mmx;

  2578.         c->clear_blocks = clear_blocks_mmx;

  2579. #ifdef CONFIG_ENCODERS

  2580.         c->pix_sum = pix_sum16_mmx;

  2581. #endif //CONFIG_ENCODERS

  2582. 

  2583.         c->put_pixels_tab[0][0] = put_pixels16_mmx;

  2584.         c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;

  2585.         c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;

  2586.         c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

  2587. 

  2588.         c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;

  2589.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;

  2590.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;

  2591.         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

  2592. 

  2593.         c->avg_pixels_tab[0][0] = avg_pixels16_mmx;

  2594.         c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;

  2595.         c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;

  2596.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

  2597. 

  2598.         c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;

  2599.         c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;

  2600.         c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;

  2601.         c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

  2602. 

  2603.         c->put_pixels_tab[1][0] = put_pixels8_mmx;

  2604.         c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;

  2605.         c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;

  2606.         c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

  2607. 

  2608.         c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;

  2609.         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;

  2610.         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;

  2611.         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

  2612. 

  2613.         c->avg_pixels_tab[1][0] = avg_pixels8_mmx;

  2614.         c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;

  2615.         c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;

  2616.         c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

  2617. 

  2618.         c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;

  2619.         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;

  2620.         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;

  2621.         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;

  2622.                 

  2623.         c->add_bytes= add_bytes_mmx;

  2624. #ifdef CONFIG_ENCODERS

  2625.         c->diff_bytes= diff_bytes_mmx;

  2626.         

  2627.         c->hadamard8_diff[0]= hadamard8_diff16_mmx;

  2628.         c->hadamard8_diff[1]= hadamard8_diff_mmx;

  2629.         

  2630. 	c->pix_norm1 = pix_norm1_mmx;

  2631. 	c->sse[0] = sse16_mmx;

  2632.   	c->sse[1] = sse8_mmx;

  2633.         c->vsad[4]= vsad_intra16_mmx;

  2634. 

  2635. 	c->nsse[0] = nsse16_mmx;

  2636. 	c->nsse[1] = nsse8_mmx;

  2637.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2638.             c->vsad[0] = vsad16_mmx;

  2639.         }

  2640.         

  2641.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2642.             c->try_8x8basis= try_8x8basis_mmx;

  2643.         }

  2644.         c->add_8x8basis= add_8x8basis_mmx;

  2645.         

  2646. #endif //CONFIG_ENCODERS

  2647. 

  2648.         c->h263_v_loop_filter= h263_v_loop_filter_mmx;

  2649.         c->h263_h_loop_filter= h263_h_loop_filter_mmx;        

  2650. 	c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx;

  2651.         

  2652.         if (mm_flags & MM_MMXEXT) {

  2653.             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;

  2654.             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

  2655. 

  2656.             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;

  2657.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;

  2658.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

  2659. 

  2660.             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;

  2661.             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

  2662. 

  2663.             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;

  2664.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;

  2665.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

  2666. 

  2667. #ifdef CONFIG_ENCODERS

  2668.             c->hadamard8_diff[0]= hadamard8_diff16_mmx2;

  2669.             c->hadamard8_diff[1]= hadamard8_diff_mmx2;

  2670.             c->vsad[4]= vsad_intra16_mmx2;

  2671. #endif //CONFIG_ENCODERS

  2672. 

  2673.             c->h264_idct_add= ff_h264_idct_add_mmx2;

  2674. 

  2675.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2676.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;

  2677.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

  2678.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

  2679.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

  2680.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

  2681.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;

  2682. #ifdef CONFIG_ENCODERS

  2683.                 c->vsad[0] = vsad16_mmx2;

  2684. #endif //CONFIG_ENCODERS

  2685.             }

  2686. 

  2687. #if 1

  2688.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)

  2689.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)

  2690.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)

  2691.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)

  2692.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)

  2693.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)

  2694.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)

  2695.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)

  2696.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)

  2697.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)

  2698.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)

  2699.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)

  2700.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)

  2701.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)

  2702.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)

  2703.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)

  2704.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)

  2705.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)

  2706.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)

  2707.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)

  2708.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)

  2709.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)

  2710.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)

  2711.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)

  2712.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)

  2713.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)

  2714.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)

  2715.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)

  2716.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)

  2717.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)

  2718.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)

  2719.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)

  2720. #endif

  2721. 

  2722. //FIXME 3dnow too

  2723. #define dspfunc(PFX, IDX, NUM) \

  2724.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_mmx2; \

  2725.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_mmx2; \

  2726.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_mmx2; \

  2727.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_mmx2; \

  2728.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_mmx2; \

  2729.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_mmx2; \

  2730.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_mmx2; \

  2731.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_mmx2; \

  2732.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_mmx2; \

  2733.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_mmx2; \

  2734.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_mmx2; \

  2735.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_mmx2; \

  2736.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_mmx2; \

  2737.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_mmx2; \

  2738.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_mmx2; \

  2739.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_mmx2

  2740. 

  2741.             dspfunc(put_h264_qpel, 0, 16);

  2742.             dspfunc(put_h264_qpel, 1, 8);

  2743.             dspfunc(put_h264_qpel, 2, 4);

  2744.             dspfunc(avg_h264_qpel, 0, 16);

  2745.             dspfunc(avg_h264_qpel, 1, 8);

  2746.             dspfunc(avg_h264_qpel, 2, 4);

  2747. #undef dspfunc

  2748. 

  2749. 	    c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2;

  2750.             c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;

  2751.             c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;

  2752.             c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;

  2753.             c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;

  2754.             c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;

  2755.             c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;

  2756. 

  2757. #ifdef CONFIG_ENCODERS

  2758.             c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;

  2759. #endif //CONFIG_ENCODERS

  2760.         } else if (mm_flags & MM_3DNOW) {

  2761.             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;

  2762.             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  2763. 

  2764.             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  2765.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  2766.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  2767. 

  2768.             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;

  2769.             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

  2770. 

  2771.             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;

  2772.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;

  2773.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

  2774. 

  2775.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2776.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  2777.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  2778.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;

  2779.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

  2780.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  2781.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;

  2782.             }

  2783. 

  2784.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)

  2785.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)

  2786.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)

  2787.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)

  2788.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)

  2789.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)

  2790.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)

  2791.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)

  2792.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)

  2793.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)

  2794.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)

  2795.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)

  2796.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)

  2797.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)

  2798.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)

  2799.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)

  2800.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)

  2801.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)

  2802.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)

  2803.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)

  2804.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)

  2805.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)

  2806.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)

  2807.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)

  2808.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)

  2809.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)

  2810.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)

  2811.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)

  2812.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)

  2813.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)

  2814.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)

  2815.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)

  2816. 

  2817. #define dspfunc(PFX, IDX, NUM) \

  2818.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_3dnow; \

  2819.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_3dnow; \

  2820.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_3dnow; \

  2821.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_3dnow; \

  2822.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_3dnow; \

  2823.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_3dnow; \

  2824.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_3dnow; \

  2825.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_3dnow; \

  2826.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_3dnow; \

  2827.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_3dnow; \

  2828.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_3dnow; \

  2829.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_3dnow; \

  2830.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_3dnow; \

  2831.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_3dnow; \

  2832.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_3dnow; \

  2833.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_3dnow

  2834. 

  2835.             dspfunc(put_h264_qpel, 0, 16);

  2836.             dspfunc(put_h264_qpel, 1, 8);

  2837.             dspfunc(put_h264_qpel, 2, 4);

  2838.             dspfunc(avg_h264_qpel, 0, 16);

  2839.             dspfunc(avg_h264_qpel, 1, 8);

  2840.             dspfunc(avg_h264_qpel, 2, 4);

  2841. 

  2842. 	    c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow;

  2843.         }

  2844.     }

  2845.         

  2846. #ifdef CONFIG_ENCODERS

  2847.     dsputil_init_pix_mmx(c, avctx);

  2848. #endif //CONFIG_ENCODERS

  2849. #if 0

  2850.     // for speed testing

  2851.     get_pixels = just_return;

  2852.     put_pixels_clamped = just_return;

  2853.     add_pixels_clamped = just_return;

  2854. 

  2855.     pix_abs16x16 = just_return;

  2856.     pix_abs16x16_x2 = just_return;

  2857.     pix_abs16x16_y2 = just_return;

  2858.     pix_abs16x16_xy2 = just_return;

  2859. 

  2860.     put_pixels_tab[0] = just_return;

  2861.     put_pixels_tab[1] = just_return;

  2862.     put_pixels_tab[2] = just_return;

  2863.     put_pixels_tab[3] = just_return;

  2864. 

  2865.     put_no_rnd_pixels_tab[0] = just_return;

  2866.     put_no_rnd_pixels_tab[1] = just_return;

  2867.     put_no_rnd_pixels_tab[2] = just_return;

  2868.     put_no_rnd_pixels_tab[3] = just_return;

  2869. 

  2870.     avg_pixels_tab[0] = just_return;

  2871.     avg_pixels_tab[1] = just_return;

  2872.     avg_pixels_tab[2] = just_return;

  2873.     avg_pixels_tab[3] = just_return;

  2874. 

  2875.     avg_no_rnd_pixels_tab[0] = just_return;

  2876.     avg_no_rnd_pixels_tab[1] = just_return;

  2877.     avg_no_rnd_pixels_tab[2] = just_return;

  2878.     avg_no_rnd_pixels_tab[3] = just_return;

  2879. 

  2880.     //av_fdct = just_return;

  2881.     //ff_idct = just_return;

  2882. #endif

  2883. }