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.
4353 Commits
32 Branches
844MB
Tree: c64d476ccd
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c64d476ccd'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx.c

3312 lines
119KB
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. 

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

  34. 

  35. /* pixel operations */

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

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

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

  39. 

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

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

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

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

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

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

  46. 

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

  48. 

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

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

  51. 

  52. #define MOVQ_WONE(regd) \

  53.     __asm __volatile ( \

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

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

  56. 

  57. #define MOVQ_BFE(regd) \

  58.     __asm __volatile ( \

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

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

  61. 

  62. #ifndef PIC

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

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

  65. #else

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

  67. // pcmpeqd -> -1

  68. #define MOVQ_BONE(regd) \

  69.     __asm __volatile ( \

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

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

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

  73. 

  74. #define MOVQ_WTWO(regd) \

  75.     __asm __volatile ( \

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

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

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

  79. 

  80. #endif

  81. 

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

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

  84. // regfe is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

  92. 

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

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

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

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

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

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

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

  100. 

  101. // mm6 is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

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

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

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

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

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

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

  115. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  129. 

  130. /***********************************/

  131. /* MMX no rounding */

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

  133. #define SET_RND  MOVQ_WONE

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

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

  136. 

  137. #include "dsputil_mmx_rnd.h"

  138. 

  139. #undef DEF

  140. #undef SET_RND

  141. #undef PAVGBP

  142. #undef PAVGB

  143. /***********************************/

  144. /* MMX rounding */

  145. 

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

  147. #define SET_RND  MOVQ_WTWO

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

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

  150. 

  151. #include "dsputil_mmx_rnd.h"

  152. 

  153. #undef DEF

  154. #undef SET_RND

  155. #undef PAVGBP

  156. #undef PAVGB

  157. 

  158. /***********************************/

  159. /* 3Dnow specific */

  160. 

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

  162. /* for Athlons PAVGUSB is prefered */

  163. #define PAVGB "pavgusb"

  164. 

  165. #include "dsputil_mmx_avg.h"

  166. 

  167. #undef DEF

  168. #undef PAVGB

  169. 

  170. /***********************************/

  171. /* MMX2 specific */

  172. 

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

  174. 

  175. /* Introduced only in MMX2 set */

  176. #define PAVGB "pavgb"

  177. 

  178. #include "dsputil_mmx_avg.h"

  179. 

  180. #undef DEF

  181. #undef PAVGB

  182. 

  183. /***********************************/

  184. /* standard MMX */

  185. 

  186. #ifdef CONFIG_ENCODERS

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

  188. {

  189.     asm volatile(

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

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

  192.         ".balign 16		\n\t"

  193.         "1:			\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  208.         "js 1b			\n\t"

  209.         : "+r" (pixels)

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

  211.         : "%"REG_a

  212.     );

  213. }

  214. 

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

  216. {

  217.     asm volatile(

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

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

  220.         ".balign 16		\n\t"

  221.         "1:			\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  237.         "jnz 1b			\n\t"

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

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

  240.         : "%"REG_a

  241.     );

  242. }

  243. #endif //CONFIG_ENCODERS

  244. 

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

  246. {

  247.     const DCTELEM *p;

  248.     uint8_t *pix;

  249. 

  250.     /* read the pixels */

  251.     p = block;

  252.     pix = pixels;

  253.     /* unrolled loop */

  254. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  272. 		:"memory");

  273.         pix += line_size*4;

  274.         p += 32;

  275. 

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

  277.     // compiler would generate some very strange code

  278.     // thus using "r"

  279.     __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  297. 	    :"memory");

  298. }

  299. 

  300. static unsigned char __align8 vector128[8] =

  301.   { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };

  302. 

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

  304. {

  305.     int i;

  306. 

  307.     movq_m2r(*vector128, mm1);

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

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

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

  311.         block += 8;

  312.         paddb_r2r(mm1, mm0);

  313.         movq_r2m(mm0, *pixels);

  314.         pixels += line_size;

  315.     }

  316. }

  317. 

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

  319. {

  320.     const DCTELEM *p;

  321.     uint8_t *pix;

  322.     int i;

  323. 

  324.     /* read the pixels */

  325.     p = block;

  326.     pix = pixels;

  327.     MOVQ_ZERO(mm7);

  328.     i = 4;

  329.     do {

  330. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  352. 		:"r"(p)

  353. 		:"memory");

  354.         pix += line_size*2;

  355.         p += 16;

  356.     } while (--i);

  357. }

  358. 

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

  360. {

  361.     __asm __volatile(

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

  363. 	 ".balign 8			\n\t"

  364. 	 "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  378. 	 "jnz 1b			\n\t"

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

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

  381. 	 : "%"REG_a, "memory"

  382. 	);

  383. }

  384. 

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

  386. {

  387.     __asm __volatile(

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

  389. 	 ".balign 8			\n\t"

  390. 	 "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  404. 	 "jnz 1b			\n\t"

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

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

  407. 	 : "%"REG_a, "memory"

  408. 	);

  409. }

  410. 

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

  412. {

  413.     __asm __volatile(

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

  415. 	 ".balign 8			\n\t"

  416. 	 "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  438. 	 "jnz 1b			\n\t"

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

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

  441. 	 : "%"REG_a, "memory"

  442. 	);

  443. }

  444. 

  445. static void clear_blocks_mmx(DCTELEM *blocks)

  446. {

  447.     __asm __volatile(

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

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

  450.                 "1:				\n\t"

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

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

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

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

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

  456.                 " js 1b				\n\t"

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

  458.                 : "%"REG_a

  459.         );

  460. }

  461. 

  462. #ifdef CONFIG_ENCODERS

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

  464.     const int h=16;

  465.     int sum;

  466.     long index= -line_size*h;

  467. 

  468.     __asm __volatile(

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

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

  471.                 "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  485.                 " js 1b				\n\t"

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

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

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

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

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

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

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

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

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

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

  496.         );

  497. 

  498.         return sum;

  499. }

  500. #endif //CONFIG_ENCODERS

  501. 

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

  503.     long i=0;

  504.     asm volatile(

  505.         "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

  516.         " jb 1b				\n\t"

  517.         : "+r" (i)

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

  519.     );

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

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

  522. }

  523. 

  524. #define H263_LOOP_FILTER \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  594. 

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

  596.     const int strength= ff_h263_loop_filter_strength[qscale];

  597. 

  598.     asm volatile(

  599.     

  600.         H263_LOOP_FILTER

  601.         

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

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

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

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

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

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

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

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

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

  611.     );

  612. }

  613. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  631.         

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

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

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

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

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

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

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

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

  640.     );

  641. }

  642. 

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

  644.     const int strength= ff_h263_loop_filter_strength[qscale];

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

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

  647.     

  648.     src -= 2;

  649. 

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

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

  652.     asm volatile(

  653.         H263_LOOP_FILTER // 5 3 4 6

  654.         

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

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

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

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

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

  660.     );

  661. 

  662.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  687.         :: "r" (src),

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

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

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

  691.     );

  692. }

  693. 

  694. #ifdef CONFIG_ENCODERS

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

  696.     int tmp;

  697.   asm volatile (

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

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

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

  701.       "1:\n"

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

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

  704. 

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

  706. 

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

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

  709. 

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

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

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

  713. 

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

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

  716. 

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

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

  719. 

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

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

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

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

  724. 

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

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

  727.       "dec %%ecx\n"

  728.       "jnz 1b\n"

  729. 

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

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

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

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

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

  735.     return tmp;

  736. }

  737. 

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

  739.     int tmp;

  740.   asm volatile (

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

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

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

  744.       "1:\n"

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

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

  747. 

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

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

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

  751. 

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

  753. 

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

  755. 

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

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

  758. 

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

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

  761. 

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

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

  764. 

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

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

  767. 

  768.       "decl %%ecx\n"

  769.       "jnz 1b\n"

  770. 

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

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

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

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

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

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

  777.       : "%ecx");

  778.     return tmp;

  779. }

  780. 

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

  782.     int tmp;

  783.   asm volatile (

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

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

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

  787.       "1:\n"

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

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

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

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

  792. 

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

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

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

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

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

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

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

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

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

  802. 

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

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

  805. 

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

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

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

  809. 

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

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

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

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

  814. 

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

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

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

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

  819. 

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

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

  822. 

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

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

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

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

  827. 

  828.       "decl %%ecx\n"

  829.       "jnz 1b\n"

  830. 

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

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

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

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

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

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

  837.       : "%ecx");

  838.     return tmp;

  839. }

  840. 

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

  842.     int tmp;

  843.   asm volatile (

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

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

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

  847.       

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

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

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

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

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

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

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

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

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

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

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

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

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

  861.       

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

  863.       

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  889. 

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

  891.       "1:\n"

  892.   

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  918.       

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

  920.       

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  946. 

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

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

  949.       " jnz 1b\n"

  950. 

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

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

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

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

  955.       

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

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

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

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

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

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

  962.       : "%ecx");

  963.       return tmp;

  964. }

  965. 

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

  967.     int tmp;

  968.     uint8_t * pix= pix1;

  969.   asm volatile (

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

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

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

  973.       

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

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

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

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

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

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

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

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

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

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

  984.       

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

  986.       

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

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

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

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

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

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

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

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

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

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

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

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

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

  1000.       "pxor %%mm1, %%mm1\n"

  1001.       "pcmpgtw %%mm0, %%mm3\n\t"

  1002.       "pcmpgtw %%mm2, %%mm1\n\t"

  1003.       "pxor %%mm3, %%mm0\n"

  1004.       "pxor %%mm1, %%mm2\n"

  1005.       "psubw %%mm3, %%mm0\n" 

  1006.       "psubw %%mm1, %%mm2\n"

  1007.       "paddw %%mm0, %%mm2\n"

  1008.       "paddw %%mm2, %%mm6\n"

  1009. 

  1010.       "add %2,%0\n"

  1011.       "1:\n"

  1012.   

  1013.       "movq (%0),%%mm0\n"

  1014.       "movq 1(%0),%%mm1\n"

  1015.       "movq %%mm0, %%mm2\n"

  1016.       "movq %%mm1, %%mm3\n"

  1017.       "punpcklbw %%mm7,%%mm0\n"

  1018.       "punpcklbw %%mm7,%%mm1\n"

  1019.       "punpckhbw %%mm7,%%mm2\n"

  1020.       "punpckhbw %%mm7,%%mm3\n"

  1021.       "psubw %%mm1, %%mm0\n"

  1022.       "psubw %%mm3, %%mm2\n"

  1023.       "psubw %%mm0, %%mm4\n"

  1024.       "psubw %%mm2, %%mm5\n"

  1025.       "pxor %%mm3, %%mm3\n"

  1026.       "pxor %%mm1, %%mm1\n"

  1027.       "pcmpgtw %%mm4, %%mm3\n\t"

  1028.       "pcmpgtw %%mm5, %%mm1\n\t"

  1029.       "pxor %%mm3, %%mm4\n"

  1030.       "pxor %%mm1, %%mm5\n"

  1031.       "psubw %%mm3, %%mm4\n"

  1032.       "psubw %%mm1, %%mm5\n"

  1033.       "paddw %%mm4, %%mm5\n"

  1034.       "paddw %%mm5, %%mm6\n"

  1035.       

  1036.       "add %2,%0\n"

  1037.       

  1038.       "movq (%0),%%mm4\n"

  1039.       "movq 1(%0),%%mm1\n"

  1040.       "movq %%mm4, %%mm5\n"

  1041.       "movq %%mm1, %%mm3\n"

  1042.       "punpcklbw %%mm7,%%mm4\n"

  1043.       "punpcklbw %%mm7,%%mm1\n"

  1044.       "punpckhbw %%mm7,%%mm5\n"

  1045.       "punpckhbw %%mm7,%%mm3\n"

  1046.       "psubw %%mm1, %%mm4\n"

  1047.       "psubw %%mm3, %%mm5\n"

  1048.       "psubw %%mm4, %%mm0\n"

  1049.       "psubw %%mm5, %%mm2\n"

  1050.       "pxor %%mm3, %%mm3\n"

  1051.       "pxor %%mm1, %%mm1\n"

  1052.       "pcmpgtw %%mm0, %%mm3\n\t"

  1053.       "pcmpgtw %%mm2, %%mm1\n\t"

  1054.       "pxor %%mm3, %%mm0\n"

  1055.       "pxor %%mm1, %%mm2\n"

  1056.       "psubw %%mm3, %%mm0\n" 

  1057.       "psubw %%mm1, %%mm2\n"

  1058.       "paddw %%mm0, %%mm2\n"

  1059.       "paddw %%mm2, %%mm6\n"

  1060. 

  1061.       "add %2,%0\n"

  1062.       "subl $2, %%ecx\n"

  1063.       " jnz 1b\n"

  1064. 

  1065.       "movq %%mm6, %%mm0\n"

  1066.       "punpcklwd %%mm7,%%mm0\n"

  1067.       "punpckhwd %%mm7,%%mm6\n"

  1068.       "paddd %%mm0, %%mm6\n"

  1069.       

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

  1071.       "psrlq $32, %%mm6\n"

  1072.       "paddd %%mm6,%%mm0\n"

  1073.       "movd %%mm0,%1\n"

  1074.       : "+r" (pix1), "=r"(tmp) 

  1075.       : "r" ((long)line_size) , "g" (h-2)

  1076.       : "%ecx");

  1077.       return tmp + hf_noise8_mmx(pix+8, line_size, h);

  1078. }

  1079. 

  1080. static int nsse16_mmx(MpegEncContext *c, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1081.     int score1= sse16_mmx(c, pix1, pix2, line_size, h);

  1082.     int score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

  1083. 

  1084.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1085.     else  return score1 + ABS(score2)*8;

  1086. }

  1087. 

  1088. static int nsse8_mmx(MpegEncContext *c, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1089.     int score1= sse8_mmx(c, pix1, pix2, line_size, h);

  1090.     int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

  1091. 

  1092.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1093.     else  return score1 + ABS(score2)*8;

  1094. }

  1095. 

  1096. static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1097.     int tmp;

  1098.     

  1099.     assert( (((int)pix) & 7) == 0);

  1100.     assert((line_size &7) ==0);

  1101.     

  1102. #define SUM(in0, in1, out0, out1) \

  1103.       "movq (%0), %%mm2\n"\

  1104.       "movq 8(%0), %%mm3\n"\

  1105.       "add %2,%0\n"\

  1106.       "movq %%mm2, " #out0 "\n"\

  1107.       "movq %%mm3, " #out1 "\n"\

  1108.       "psubusb " #in0 ", %%mm2\n"\

  1109.       "psubusb " #in1 ", %%mm3\n"\

  1110.       "psubusb " #out0 ", " #in0 "\n"\

  1111.       "psubusb " #out1 ", " #in1 "\n"\

  1112.       "por %%mm2, " #in0 "\n"\

  1113.       "por %%mm3, " #in1 "\n"\

  1114.       "movq " #in0 ", %%mm2\n"\

  1115.       "movq " #in1 ", %%mm3\n"\

  1116.       "punpcklbw %%mm7, " #in0 "\n"\

  1117.       "punpcklbw %%mm7, " #in1 "\n"\

  1118.       "punpckhbw %%mm7, %%mm2\n"\

  1119.       "punpckhbw %%mm7, %%mm3\n"\

  1120.       "paddw " #in1 ", " #in0 "\n"\

  1121.       "paddw %%mm3, %%mm2\n"\

  1122.       "paddw %%mm2, " #in0 "\n"\

  1123.       "paddw " #in0 ", %%mm6\n"

  1124. 

  1125.     

  1126.   asm volatile (

  1127.       "movl %3,%%ecx\n"

  1128.       "pxor %%mm6,%%mm6\n"

  1129.       "pxor %%mm7,%%mm7\n"

  1130.       "movq (%0),%%mm0\n"

  1131.       "movq 8(%0),%%mm1\n"

  1132.       "add %2,%0\n"

  1133.       "subl $2, %%ecx\n"

  1134.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1135.       "1:\n"

  1136.       

  1137.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1138.       

  1139.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1140.       

  1141.       "subl $2, %%ecx\n"

  1142.       "jnz 1b\n"

  1143. 

  1144.       "movq %%mm6,%%mm0\n"

  1145.       "psrlq $32, %%mm6\n"

  1146.       "paddw %%mm6,%%mm0\n"

  1147.       "movq %%mm0,%%mm6\n"

  1148.       "psrlq $16, %%mm0\n"

  1149.       "paddw %%mm6,%%mm0\n"

  1150.       "movd %%mm0,%1\n"

  1151.       : "+r" (pix), "=r"(tmp) 

  1152.       : "r" ((long)line_size) , "m" (h)

  1153.       : "%ecx");

  1154.     return tmp & 0xFFFF;

  1155. }

  1156. #undef SUM

  1157. 

  1158. static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1159.     int tmp;

  1160.     

  1161.     assert( (((int)pix) & 7) == 0);

  1162.     assert((line_size &7) ==0);

  1163.     

  1164. #define SUM(in0, in1, out0, out1) \

  1165.       "movq (%0), " #out0 "\n"\

  1166.       "movq 8(%0), " #out1 "\n"\

  1167.       "add %2,%0\n"\

  1168.       "psadbw " #out0 ", " #in0 "\n"\

  1169.       "psadbw " #out1 ", " #in1 "\n"\

  1170.       "paddw " #in1 ", " #in0 "\n"\

  1171.       "paddw " #in0 ", %%mm6\n"

  1172. 

  1173.   asm volatile (

  1174.       "movl %3,%%ecx\n"

  1175.       "pxor %%mm6,%%mm6\n"

  1176.       "pxor %%mm7,%%mm7\n"

  1177.       "movq (%0),%%mm0\n"

  1178.       "movq 8(%0),%%mm1\n"

  1179.       "add %2,%0\n"

  1180.       "subl $2, %%ecx\n"

  1181.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1182.       "1:\n"

  1183.       

  1184.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1185.       

  1186.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1187.       

  1188.       "subl $2, %%ecx\n"

  1189.       "jnz 1b\n"

  1190. 

  1191.       "movd %%mm6,%1\n"

  1192.       : "+r" (pix), "=r"(tmp) 

  1193.       : "r" ((long)line_size) , "m" (h)

  1194.       : "%ecx");

  1195.     return tmp;

  1196. }

  1197. #undef SUM

  1198. 

  1199. static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1200.     int tmp;

  1201.     

  1202.     assert( (((int)pix1) & 7) == 0);

  1203.     assert( (((int)pix2) & 7) == 0);

  1204.     assert((line_size &7) ==0);

  1205.     

  1206. #define SUM(in0, in1, out0, out1) \

  1207.       "movq (%0),%%mm2\n"\

  1208.       "movq (%1)," #out0 "\n"\

  1209.       "movq 8(%0),%%mm3\n"\

  1210.       "movq 8(%1)," #out1 "\n"\

  1211.       "add %3,%0\n"\

  1212.       "add %3,%1\n"\

  1213.       "psubb " #out0 ", %%mm2\n"\

  1214.       "psubb " #out1 ", %%mm3\n"\

  1215.       "pxor %%mm7, %%mm2\n"\

  1216.       "pxor %%mm7, %%mm3\n"\

  1217.       "movq %%mm2, " #out0 "\n"\

  1218.       "movq %%mm3, " #out1 "\n"\

  1219.       "psubusb " #in0 ", %%mm2\n"\

  1220.       "psubusb " #in1 ", %%mm3\n"\

  1221.       "psubusb " #out0 ", " #in0 "\n"\

  1222.       "psubusb " #out1 ", " #in1 "\n"\

  1223.       "por %%mm2, " #in0 "\n"\

  1224.       "por %%mm3, " #in1 "\n"\

  1225.       "movq " #in0 ", %%mm2\n"\

  1226.       "movq " #in1 ", %%mm3\n"\

  1227.       "punpcklbw %%mm7, " #in0 "\n"\

  1228.       "punpcklbw %%mm7, " #in1 "\n"\

  1229.       "punpckhbw %%mm7, %%mm2\n"\

  1230.       "punpckhbw %%mm7, %%mm3\n"\

  1231.       "paddw " #in1 ", " #in0 "\n"\

  1232.       "paddw %%mm3, %%mm2\n"\

  1233.       "paddw %%mm2, " #in0 "\n"\

  1234.       "paddw " #in0 ", %%mm6\n"

  1235. 

  1236.     

  1237.   asm volatile (

  1238.       "movl %4,%%ecx\n"

  1239.       "pxor %%mm6,%%mm6\n"

  1240.       "pcmpeqw %%mm7,%%mm7\n"

  1241.       "psllw $15, %%mm7\n"

  1242.       "packsswb %%mm7, %%mm7\n"

  1243.       "movq (%0),%%mm0\n"

  1244.       "movq (%1),%%mm2\n"

  1245.       "movq 8(%0),%%mm1\n"

  1246.       "movq 8(%1),%%mm3\n"

  1247.       "add %3,%0\n"

  1248.       "add %3,%1\n"

  1249.       "subl $2, %%ecx\n"

  1250.       "psubb %%mm2, %%mm0\n"

  1251.       "psubb %%mm3, %%mm1\n"

  1252.       "pxor %%mm7, %%mm0\n"

  1253.       "pxor %%mm7, %%mm1\n"

  1254.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1255.       "1:\n"

  1256.       

  1257.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1258.       

  1259.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1260.       

  1261.       "subl $2, %%ecx\n"

  1262.       "jnz 1b\n"

  1263. 

  1264.       "movq %%mm6,%%mm0\n"

  1265.       "psrlq $32, %%mm6\n"

  1266.       "paddw %%mm6,%%mm0\n"

  1267.       "movq %%mm0,%%mm6\n"

  1268.       "psrlq $16, %%mm0\n"

  1269.       "paddw %%mm6,%%mm0\n"

  1270.       "movd %%mm0,%2\n"

  1271.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  1272.       : "r" ((long)line_size) , "m" (h)

  1273.       : "%ecx");

  1274.     return tmp & 0x7FFF;

  1275. }

  1276. #undef SUM

  1277. 

  1278. static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1279.     int tmp;

  1280.     

  1281.     assert( (((int)pix1) & 7) == 0);

  1282.     assert( (((int)pix2) & 7) == 0);

  1283.     assert((line_size &7) ==0);

  1284.     

  1285. #define SUM(in0, in1, out0, out1) \

  1286.       "movq (%0)," #out0 "\n"\

  1287.       "movq (%1),%%mm2\n"\

  1288.       "movq 8(%0)," #out1 "\n"\

  1289.       "movq 8(%1),%%mm3\n"\

  1290.       "add %3,%0\n"\

  1291.       "add %3,%1\n"\

  1292.       "psubb %%mm2, " #out0 "\n"\

  1293.       "psubb %%mm3, " #out1 "\n"\

  1294.       "pxor %%mm7, " #out0 "\n"\

  1295.       "pxor %%mm7, " #out1 "\n"\

  1296.       "psadbw " #out0 ", " #in0 "\n"\

  1297.       "psadbw " #out1 ", " #in1 "\n"\

  1298.       "paddw " #in1 ", " #in0 "\n"\

  1299.       "paddw " #in0 ", %%mm6\n"

  1300. 

  1301.   asm volatile (

  1302.       "movl %4,%%ecx\n"

  1303.       "pxor %%mm6,%%mm6\n"

  1304.       "pcmpeqw %%mm7,%%mm7\n"

  1305.       "psllw $15, %%mm7\n"

  1306.       "packsswb %%mm7, %%mm7\n"

  1307.       "movq (%0),%%mm0\n"

  1308.       "movq (%1),%%mm2\n"

  1309.       "movq 8(%0),%%mm1\n"

  1310.       "movq 8(%1),%%mm3\n"

  1311.       "add %3,%0\n"

  1312.       "add %3,%1\n"

  1313.       "subl $2, %%ecx\n"

  1314.       "psubb %%mm2, %%mm0\n"

  1315.       "psubb %%mm3, %%mm1\n"

  1316.       "pxor %%mm7, %%mm0\n"

  1317.       "pxor %%mm7, %%mm1\n"

  1318.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1319.       "1:\n"

  1320.       

  1321.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1322.       

  1323.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1324.       

  1325.       "subl $2, %%ecx\n"

  1326.       "jnz 1b\n"

  1327. 

  1328.       "movd %%mm6,%2\n"

  1329.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  1330.       : "r" ((long)line_size) , "m" (h)

  1331.       : "%ecx");

  1332.     return tmp;

  1333. }

  1334. #undef SUM

  1335. 

  1336. static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){

  1337.     long i=0;

  1338.     asm volatile(

  1339.         "1:				\n\t"

  1340.         "movq  (%2, %0), %%mm0		\n\t"

  1341.         "movq  (%1, %0), %%mm1		\n\t"

  1342.         "psubb %%mm0, %%mm1		\n\t"

  1343.         "movq %%mm1, (%3, %0)		\n\t"

  1344.         "movq 8(%2, %0), %%mm0		\n\t"

  1345.         "movq 8(%1, %0), %%mm1		\n\t"

  1346.         "psubb %%mm0, %%mm1		\n\t"

  1347.         "movq %%mm1, 8(%3, %0)		\n\t"

  1348.         "add $16, %0			\n\t"

  1349.         "cmp %4, %0			\n\t"

  1350.         " jb 1b				\n\t"

  1351.         : "+r" (i)

  1352.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15)

  1353.     );

  1354.     for(; i<w; i++)

  1355.         dst[i+0] = src1[i+0]-src2[i+0];

  1356. }

  1357. 

  1358. static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){

  1359.     long i=0;

  1360.     uint8_t l, lt;

  1361.     

  1362.     asm volatile(

  1363.         "1:				\n\t"

  1364.         "movq  -1(%1, %0), %%mm0	\n\t" // LT

  1365.         "movq  (%1, %0), %%mm1		\n\t" // T

  1366.         "movq  -1(%2, %0), %%mm2	\n\t" // L

  1367.         "movq  (%2, %0), %%mm3		\n\t" // X

  1368.         "movq %%mm2, %%mm4		\n\t" // L

  1369.         "psubb %%mm0, %%mm2		\n\t"

  1370.         "paddb %%mm1, %%mm2		\n\t" // L + T - LT

  1371.         "movq %%mm4, %%mm5		\n\t" // L

  1372.         "pmaxub %%mm1, %%mm4		\n\t" // max(T, L)

  1373.         "pminub %%mm5, %%mm1		\n\t" // min(T, L)

  1374.         "pminub %%mm2, %%mm4		\n\t" 

  1375.         "pmaxub %%mm1, %%mm4		\n\t"

  1376.         "psubb %%mm4, %%mm3		\n\t" // dst - pred

  1377.         "movq %%mm3, (%3, %0)		\n\t"

  1378.         "add $8, %0			\n\t"

  1379.         "cmp %4, %0			\n\t"

  1380.         " jb 1b				\n\t"

  1381.         : "+r" (i)

  1382.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w)

  1383.     );

  1384. 

  1385.     l= *left;

  1386.     lt= *left_top;

  1387.     

  1388.     dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);

  1389.     

  1390.     *left_top= src1[w-1];

  1391.     *left    = src2[w-1];

  1392. }

  1393. 

  1394. #define LBUTTERFLY2(a1,b1,a2,b2)\

  1395.     "paddw " #b1 ", " #a1 "		\n\t"\

  1396.     "paddw " #b2 ", " #a2 "		\n\t"\

  1397.     "paddw " #b1 ", " #b1 "		\n\t"\

  1398.     "paddw " #b2 ", " #b2 "		\n\t"\

  1399.     "psubw " #a1 ", " #b1 "		\n\t"\

  1400.     "psubw " #a2 ", " #b2 "		\n\t"

  1401. 

  1402. #define HADAMARD48\

  1403.         LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\

  1404.         LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\

  1405.         LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\

  1406.         LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\

  1407.         LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\

  1408.         LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\

  1409. 

  1410. #define MMABS(a,z)\

  1411.     "pxor " #z ", " #z "		\n\t"\

  1412.     "pcmpgtw " #a ", " #z "		\n\t"\

  1413.     "pxor " #z ", " #a "		\n\t"\

  1414.     "psubw " #z ", " #a "		\n\t"

  1415. 

  1416. #define MMABS_SUM(a,z, sum)\

  1417.     "pxor " #z ", " #z "		\n\t"\

  1418.     "pcmpgtw " #a ", " #z "		\n\t"\

  1419.     "pxor " #z ", " #a "		\n\t"\

  1420.     "psubw " #z ", " #a "		\n\t"\

  1421.     "paddusw " #a ", " #sum "		\n\t"

  1422. 

  1423. #define MMABS_MMX2(a,z)\

  1424.     "pxor " #z ", " #z "		\n\t"\

  1425.     "psubw " #a ", " #z "		\n\t"\

  1426.     "pmaxsw " #z ", " #a "		\n\t"

  1427. 

  1428. #define MMABS_SUM_MMX2(a,z, sum)\

  1429.     "pxor " #z ", " #z "		\n\t"\

  1430.     "psubw " #a ", " #z "		\n\t"\

  1431.     "pmaxsw " #z ", " #a "		\n\t"\

  1432.     "paddusw " #a ", " #sum "		\n\t"

  1433.         

  1434. #define SBUTTERFLY(a,b,t,n)\

  1435.     "movq " #a ", " #t "		\n\t" /* abcd */\

  1436.     "punpckl" #n " " #b ", " #a "	\n\t" /* aebf */\

  1437.     "punpckh" #n " " #b ", " #t "	\n\t" /* cgdh */\

  1438. 

  1439. #define TRANSPOSE4(a,b,c,d,t)\

  1440.     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\

  1441.     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\

  1442.     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\

  1443.     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

  1444. 

  1445. #define LOAD4(o, a, b, c, d)\

  1446.         "movq "#o"(%1), " #a "		\n\t"\

  1447.         "movq "#o"+16(%1), " #b "	\n\t"\

  1448.         "movq "#o"+32(%1), " #c "	\n\t"\

  1449.         "movq "#o"+48(%1), " #d "	\n\t"

  1450. 

  1451. #define STORE4(o, a, b, c, d)\

  1452.         "movq "#a", "#o"(%1)		\n\t"\

  1453.         "movq "#b", "#o"+16(%1)		\n\t"\

  1454.         "movq "#c", "#o"+32(%1)		\n\t"\

  1455.         "movq "#d", "#o"+48(%1)		\n\t"\

  1456. 

  1457. static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1458.     uint64_t temp[16] __align8;

  1459.     int sum=0;

  1460.     

  1461.     assert(h==8);

  1462. 

  1463.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1464. 

  1465.     asm volatile(

  1466.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1467.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1468.         

  1469.         HADAMARD48

  1470.         

  1471.         "movq %%mm7, 112(%1)		\n\t"

  1472.         

  1473.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1474.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1475.         

  1476.         "movq 112(%1), %%mm7 		\n\t"

  1477.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1478.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1479. 

  1480.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1481.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1482.         

  1483.         HADAMARD48

  1484.         

  1485.         "movq %%mm7, 120(%1)		\n\t"

  1486.         

  1487.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1488.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1489.         

  1490.         "movq 120(%1), %%mm7 		\n\t"

  1491.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1492.         "movq %%mm7, %%mm5		\n\t"//FIXME remove

  1493.         "movq %%mm6, %%mm7		\n\t"

  1494.         "movq %%mm0, %%mm6		\n\t"

  1495. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1496.         

  1497.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1498. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1499.         

  1500.         HADAMARD48

  1501.         "movq %%mm7, 64(%1)		\n\t"

  1502.         MMABS(%%mm0, %%mm7)

  1503.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1504.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1505.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1506.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1507.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1508.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1509.         "movq 64(%1), %%mm1		\n\t"

  1510.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1511.         "movq %%mm0, 64(%1)		\n\t"

  1512.         

  1513.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1514.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1515.         

  1516.         HADAMARD48

  1517.         "movq %%mm7, (%1)		\n\t"

  1518.         MMABS(%%mm0, %%mm7)

  1519.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1520.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1521.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1522.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1523.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1524.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1525.         "movq (%1), %%mm1		\n\t"

  1526.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1527.         "movq 64(%1), %%mm1		\n\t"

  1528.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1529.         

  1530.         "movq %%mm0, %%mm1		\n\t"

  1531.         "psrlq $32, %%mm0		\n\t"

  1532.         "paddusw %%mm1, %%mm0		\n\t"

  1533.         "movq %%mm0, %%mm1		\n\t"

  1534.         "psrlq $16, %%mm0		\n\t"

  1535.         "paddusw %%mm1, %%mm0		\n\t"

  1536.         "movd %%mm0, %0			\n\t"

  1537.                 

  1538.         : "=r" (sum)

  1539.         : "r"(temp)

  1540.     );

  1541.     return sum&0xFFFF;

  1542. }

  1543. 

  1544. static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1545.     uint64_t temp[16] __align8;

  1546.     int sum=0;

  1547.     

  1548.     assert(h==8);

  1549. 

  1550.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1551. 

  1552.     asm volatile(

  1553.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1554.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1555.         

  1556.         HADAMARD48

  1557.         

  1558.         "movq %%mm7, 112(%1)		\n\t"

  1559.         

  1560.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1561.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1562.         

  1563.         "movq 112(%1), %%mm7 		\n\t"

  1564.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1565.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1566. 

  1567.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1568.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1569.         

  1570.         HADAMARD48

  1571.         

  1572.         "movq %%mm7, 120(%1)		\n\t"

  1573.         

  1574.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1575.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1576.         

  1577.         "movq 120(%1), %%mm7 		\n\t"

  1578.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1579.         "movq %%mm7, %%mm5		\n\t"//FIXME remove

  1580.         "movq %%mm6, %%mm7		\n\t"

  1581.         "movq %%mm0, %%mm6		\n\t"

  1582. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1583.         

  1584.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1585. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1586.         

  1587.         HADAMARD48

  1588.         "movq %%mm7, 64(%1)		\n\t"

  1589.         MMABS_MMX2(%%mm0, %%mm7)

  1590.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1591.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1592.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1593.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1594.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1595.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1596.         "movq 64(%1), %%mm1		\n\t"

  1597.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1598.         "movq %%mm0, 64(%1)		\n\t"

  1599.         

  1600.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1601.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1602.         

  1603.         HADAMARD48

  1604.         "movq %%mm7, (%1)		\n\t"

  1605.         MMABS_MMX2(%%mm0, %%mm7)

  1606.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1607.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1608.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1609.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1610.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1611.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1612.         "movq (%1), %%mm1		\n\t"

  1613.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1614.         "movq 64(%1), %%mm1		\n\t"

  1615.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1616.         

  1617.         "movq %%mm0, %%mm1		\n\t"

  1618.         "psrlq $32, %%mm0		\n\t"

  1619.         "paddusw %%mm1, %%mm0		\n\t"

  1620.         "movq %%mm0, %%mm1		\n\t"

  1621.         "psrlq $16, %%mm0		\n\t"

  1622.         "paddusw %%mm1, %%mm0		\n\t"

  1623.         "movd %%mm0, %0			\n\t"

  1624.                 

  1625.         : "=r" (sum)

  1626.         : "r"(temp)

  1627.     );

  1628.     return sum&0xFFFF;

  1629. }

  1630. 

  1631. 

  1632. WARPER8_16_SQ(hadamard8_diff_mmx, hadamard8_diff16_mmx)

  1633. WARPER8_16_SQ(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)

  1634. #endif //CONFIG_ENCODERS

  1635. 

  1636. #define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)

  1637. #define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

  1638. 

  1639. #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

  1640.         "paddw " #m4 ", " #m3 "		\n\t" /* x1 */\

  1641.         "movq "MANGLE(ff_pw_20)", %%mm4		\n\t" /* 20 */\

  1642.         "pmullw " #m3 ", %%mm4		\n\t" /* 20x1 */\

  1643.         "movq "#in7", " #m3 "		\n\t" /* d */\

  1644.         "movq "#in0", %%mm5		\n\t" /* D */\

  1645.         "paddw " #m3 ", %%mm5		\n\t" /* x4 */\

  1646.         "psubw %%mm5, %%mm4		\n\t" /* 20x1 - x4 */\

  1647.         "movq "#in1", %%mm5		\n\t" /* C */\

  1648.         "movq "#in2", %%mm6		\n\t" /* B */\

  1649.         "paddw " #m6 ", %%mm5		\n\t" /* x3 */\

  1650.         "paddw " #m5 ", %%mm6		\n\t" /* x2 */\

  1651.         "paddw %%mm6, %%mm6		\n\t" /* 2x2 */\

  1652.         "psubw %%mm6, %%mm5		\n\t" /* -2x2 + x3 */\

  1653.         "pmullw "MANGLE(ff_pw_3)", %%mm5	\n\t" /* -6x2 + 3x3 */\

  1654.         "paddw " #rnd ", %%mm4		\n\t" /* x2 */\

  1655.         "paddw %%mm4, %%mm5		\n\t" /* 20x1 - 6x2 + 3x3 - x4 */\

  1656.         "psraw $5, %%mm5		\n\t"\

  1657.         "packuswb %%mm5, %%mm5		\n\t"\

  1658.         OP(%%mm5, out, %%mm7, d)

  1659. 

  1660. #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\

  1661. static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1662.     uint64_t temp;\

  1663. \

  1664.     asm volatile(\

  1665.         "pxor %%mm7, %%mm7		\n\t"\

  1666.         "1:				\n\t"\

  1667.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  1668.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  1669.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  1670.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  1671.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  1672.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  1673.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  1674.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  1675.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  1676.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  1677.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  1678.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  1679.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  1680.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  1681.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  1682.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  1683.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  1684.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1685.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  1686.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  1687.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  1688.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  1689.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  1690.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  1691.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  1692.         "paddw %6, %%mm6		\n\t"\

  1693.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1694.         "psraw $5, %%mm0		\n\t"\

  1695.         "movq %%mm0, %5			\n\t"\

  1696.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1697.         \

  1698.         "movq 5(%0), %%mm0		\n\t" /* FGHIJKLM */\

  1699.         "movq %%mm0, %%mm5		\n\t" /* FGHIJKLM */\

  1700.         "movq %%mm0, %%mm6		\n\t" /* FGHIJKLM */\

  1701.         "psrlq $8, %%mm0		\n\t" /* GHIJKLM0 */\

  1702.         "psrlq $16, %%mm5		\n\t" /* HIJKLM00 */\

  1703.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0G0H0I0J */\

  1704.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0H0I0J0K */\

  1705.         "paddw %%mm0, %%mm2		\n\t" /* b */\

  1706.         "paddw %%mm5, %%mm3		\n\t" /* c */\

  1707.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  1708.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  1709.         "movq %%mm6, %%mm2		\n\t" /* FGHIJKLM */\

  1710.         "psrlq $24, %%mm6		\n\t" /* IJKLM000 */\

  1711.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0F0G0H0I */\

  1712.         "punpcklbw %%mm7, %%mm6		\n\t" /* 0I0J0K0L */\

  1713.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  1714.         "paddw %%mm2, %%mm1		\n\t" /* a */\

  1715.         "paddw %%mm6, %%mm4		\n\t" /* d */\

  1716.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  1717.         "psubw %%mm4, %%mm3		\n\t" /* - 6b +3c - d */\

  1718.         "paddw %6, %%mm1		\n\t"\

  1719.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b +3c - d */\

  1720.         "psraw $5, %%mm3		\n\t"\

  1721.         "movq %5, %%mm1			\n\t"\

  1722.         "packuswb %%mm3, %%mm1		\n\t"\

  1723.         OP_MMX2(%%mm1, (%1),%%mm4, q)\

  1724.         /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\

  1725.         \

  1726.         "movq 9(%0), %%mm1		\n\t" /* JKLMNOPQ */\

  1727.         "movq %%mm1, %%mm4		\n\t" /* JKLMNOPQ */\

  1728.         "movq %%mm1, %%mm3		\n\t" /* JKLMNOPQ */\

  1729.         "psrlq $8, %%mm1		\n\t" /* KLMNOPQ0 */\

  1730.         "psrlq $16, %%mm4		\n\t" /* LMNOPQ00 */\

  1731.         "punpcklbw %%mm7, %%mm1		\n\t" /* 0K0L0M0N */\

  1732.         "punpcklbw %%mm7, %%mm4		\n\t" /* 0L0M0N0O */\

  1733.         "paddw %%mm1, %%mm5		\n\t" /* b */\

  1734.         "paddw %%mm4, %%mm0		\n\t" /* c */\

  1735.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1736.         "psubw %%mm5, %%mm0		\n\t" /* c - 2b */\

  1737.         "movq %%mm3, %%mm5		\n\t" /* JKLMNOPQ */\

  1738.         "psrlq $24, %%mm3		\n\t" /* MNOPQ000 */\

  1739.         "pmullw "MANGLE(ff_pw_3)", %%mm0		\n\t" /* 3c - 6b */\

  1740.         "punpcklbw %%mm7, %%mm3		\n\t" /* 0M0N0O0P */\

  1741.         "paddw %%mm3, %%mm2		\n\t" /* d */\

  1742.         "psubw %%mm2, %%mm0		\n\t" /* -6b + 3c - d */\

  1743.         "movq %%mm5, %%mm2		\n\t" /* JKLMNOPQ */\

  1744.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0J0K0L0M */\

  1745.         "punpckhbw %%mm7, %%mm5		\n\t" /* 0N0O0P0Q */\

  1746.         "paddw %%mm2, %%mm6		\n\t" /* a */\

  1747.         "pmullw "MANGLE(ff_pw_20)", %%mm6		\n\t" /* 20a */\

  1748.         "paddw %6, %%mm0		\n\t"\

  1749.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1750.         "psraw $5, %%mm0		\n\t"\

  1751.         /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\

  1752.         \

  1753.         "paddw %%mm5, %%mm3		\n\t" /* a */\

  1754.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0O0P0Q0Q */\

  1755.         "paddw %%mm4, %%mm6		\n\t" /* b */\

  1756.         "pshufw $0xBE, %%mm5, %%mm4	\n\t" /* 0P0Q0Q0P */\

  1757.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0Q0Q0P0O */\

  1758.         "paddw %%mm1, %%mm4		\n\t" /* c */\

  1759.         "paddw %%mm2, %%mm5		\n\t" /* d */\

  1760.         "paddw %%mm6, %%mm6		\n\t" /* 2b */\

  1761.         "psubw %%mm6, %%mm4		\n\t" /* c - 2b */\

  1762.         "pmullw "MANGLE(ff_pw_20)", %%mm3		\n\t" /* 20a */\

  1763.         "pmullw "MANGLE(ff_pw_3)", %%mm4		\n\t" /* 3c - 6b */\

  1764.         "psubw %%mm5, %%mm3		\n\t" /* -6b + 3c - d */\

  1765.         "paddw %6, %%mm4		\n\t"\

  1766.         "paddw %%mm3, %%mm4		\n\t" /* 20a - 6b + 3c - d */\

  1767.         "psraw $5, %%mm4		\n\t"\

  1768.         "packuswb %%mm4, %%mm0		\n\t"\

  1769.         OP_MMX2(%%mm0, 8(%1), %%mm4, q)\

  1770.         \

  1771.         "add %3, %0			\n\t"\

  1772.         "add %4, %1			\n\t"\

  1773.         "decl %2			\n\t"\

  1774.         " jnz 1b				\n\t"\

  1775.         : "+a"(src), "+c"(dst), "+m"(h)\

  1776.         : "d"((long)srcStride), "S"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1777.         : "memory"\

  1778.     );\

  1779. }\

  1780. \

  1781. static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1782.     int i;\

  1783.     int16_t temp[16];\

  1784.     /* quick HACK, XXX FIXME MUST be optimized */\

  1785.     for(i=0; i<h; i++)\

  1786.     {\

  1787.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1788.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1789.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1790.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1791.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1792.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\

  1793.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\

  1794.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\

  1795.         temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\

  1796.         temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\

  1797.         temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\

  1798.         temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\

  1799.         temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\

  1800.         temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\

  1801.         temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\

  1802.         temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\

  1803.         asm volatile(\

  1804.             "movq (%0), %%mm0		\n\t"\

  1805.             "movq 8(%0), %%mm1		\n\t"\

  1806.             "paddw %2, %%mm0		\n\t"\

  1807.             "paddw %2, %%mm1		\n\t"\

  1808.             "psraw $5, %%mm0		\n\t"\

  1809.             "psraw $5, %%mm1		\n\t"\

  1810.             "packuswb %%mm1, %%mm0	\n\t"\

  1811.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1812.             "movq 16(%0), %%mm0		\n\t"\

  1813.             "movq 24(%0), %%mm1		\n\t"\

  1814.             "paddw %2, %%mm0		\n\t"\

  1815.             "paddw %2, %%mm1		\n\t"\

  1816.             "psraw $5, %%mm0		\n\t"\

  1817.             "psraw $5, %%mm1		\n\t"\

  1818.             "packuswb %%mm1, %%mm0	\n\t"\

  1819.             OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\

  1820.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1821.             : "memory"\

  1822.         );\

  1823.         dst+=dstStride;\

  1824.         src+=srcStride;\

  1825.     }\

  1826. }\

  1827. \

  1828. static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1829.     uint64_t temp;\

  1830. \

  1831.     asm volatile(\

  1832.         "pxor %%mm7, %%mm7		\n\t"\

  1833.         "1:				\n\t"\

  1834.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  1835.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  1836.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  1837.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  1838.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  1839.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  1840.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  1841.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  1842.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  1843.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  1844.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  1845.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  1846.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  1847.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  1848.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  1849.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  1850.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  1851.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1852.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  1853.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  1854.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  1855.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  1856.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  1857.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  1858.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  1859.         "paddw %6, %%mm6		\n\t"\

  1860.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1861.         "psraw $5, %%mm0		\n\t"\

  1862.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1863.         \

  1864.         "movd 5(%0), %%mm5		\n\t" /* FGHI */\

  1865.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0F0G0H0I */\

  1866.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0G0H0I0I */\

  1867.         "paddw %%mm5, %%mm1		\n\t" /* a */\

  1868.         "paddw %%mm6, %%mm2		\n\t" /* b */\

  1869.         "pshufw $0xBE, %%mm5, %%mm6	\n\t" /* 0H0I0I0H */\

  1870.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0I0I0H0G */\

  1871.         "paddw %%mm6, %%mm3		\n\t" /* c */\

  1872.         "paddw %%mm5, %%mm4		\n\t" /* d */\

  1873.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  1874.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  1875.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  1876.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  1877.         "psubw %%mm4, %%mm3		\n\t" /* -6b + 3c - d */\

  1878.         "paddw %6, %%mm1		\n\t"\

  1879.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b + 3c - d */\

  1880.         "psraw $5, %%mm3		\n\t"\

  1881.         "packuswb %%mm3, %%mm0		\n\t"\

  1882.         OP_MMX2(%%mm0, (%1), %%mm4, q)\

  1883.         \

  1884.         "add %3, %0			\n\t"\

  1885.         "add %4, %1			\n\t"\

  1886.         "decl %2			\n\t"\

  1887.         " jnz 1b			\n\t"\

  1888.         : "+a"(src), "+c"(dst), "+m"(h)\

  1889.         : "S"((long)srcStride), "D"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1890.         : "memory"\

  1891.     );\

  1892. }\

  1893. \

  1894. static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1895.     int i;\

  1896.     int16_t temp[8];\

  1897.     /* quick HACK, XXX FIXME MUST be optimized */\

  1898.     for(i=0; i<h; i++)\

  1899.     {\

  1900.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1901.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1902.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1903.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1904.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1905.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\

  1906.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\

  1907.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\

  1908.         asm volatile(\

  1909.             "movq (%0), %%mm0		\n\t"\

  1910.             "movq 8(%0), %%mm1		\n\t"\

  1911.             "paddw %2, %%mm0		\n\t"\

  1912.             "paddw %2, %%mm1		\n\t"\

  1913.             "psraw $5, %%mm0		\n\t"\

  1914.             "psraw $5, %%mm1		\n\t"\

  1915.             "packuswb %%mm1, %%mm0	\n\t"\

  1916.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1917.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1918.             :"memory"\

  1919.         );\

  1920.         dst+=dstStride;\

  1921.         src+=srcStride;\

  1922.     }\

  1923. }

  1924. 

  1925. #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\

  1926. \

  1927. static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1928.     uint64_t temp[17*4];\

  1929.     uint64_t *temp_ptr= temp;\

  1930.     int count= 17;\

  1931. \

  1932.     /*FIXME unroll */\

  1933.     asm volatile(\

  1934.         "pxor %%mm7, %%mm7		\n\t"\

  1935.         "1:				\n\t"\

  1936.         "movq (%0), %%mm0		\n\t"\

  1937.         "movq (%0), %%mm1		\n\t"\

  1938.         "movq 8(%0), %%mm2		\n\t"\

  1939.         "movq 8(%0), %%mm3		\n\t"\

  1940.         "punpcklbw %%mm7, %%mm0		\n\t"\

  1941.         "punpckhbw %%mm7, %%mm1		\n\t"\

  1942.         "punpcklbw %%mm7, %%mm2		\n\t"\

  1943.         "punpckhbw %%mm7, %%mm3		\n\t"\

  1944.         "movq %%mm0, (%1)		\n\t"\

  1945.         "movq %%mm1, 17*8(%1)		\n\t"\

  1946.         "movq %%mm2, 2*17*8(%1)		\n\t"\

  1947.         "movq %%mm3, 3*17*8(%1)		\n\t"\

  1948.         "add $8, %1			\n\t"\

  1949.         "add %3, %0			\n\t"\

  1950.         "decl %2			\n\t"\

  1951.         " jnz 1b			\n\t"\

  1952.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  1953.         : "r" ((long)srcStride)\

  1954.         : "memory"\

  1955.     );\

  1956.     \

  1957.     temp_ptr= temp;\

  1958.     count=4;\

  1959.     \

  1960. /*FIXME reorder for speed */\

  1961.     asm volatile(\

  1962.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  1963.         "1:				\n\t"\

  1964.         "movq (%0), %%mm0		\n\t"\

  1965.         "movq 8(%0), %%mm1		\n\t"\

  1966.         "movq 16(%0), %%mm2		\n\t"\

  1967.         "movq 24(%0), %%mm3		\n\t"\

  1968.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  1969.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  1970.         "add %4, %1			\n\t"\

  1971.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  1972.         \

  1973.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  1974.         "add %4, %1			\n\t"\

  1975.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  1976.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\

  1977.         "add %4, %1			\n\t"\

  1978.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\

  1979.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\

  1980.         "add %4, %1			\n\t"\

  1981.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\

  1982.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\

  1983.         "add %4, %1			\n\t"\

  1984.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\

  1985.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\

  1986.         "add %4, %1			\n\t"\

  1987.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\

  1988.         \

  1989.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\

  1990.         "add %4, %1			\n\t"  \

  1991.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\

  1992.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\

  1993.         \

  1994.         "add $136, %0			\n\t"\

  1995.         "add %6, %1			\n\t"\

  1996.         "decl %2			\n\t"\

  1997.         " jnz 1b			\n\t"\

  1998.         \

  1999.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2000.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(long)dstStride)\

  2001.         :"memory"\

  2002.     );\

  2003. }\

  2004. \

  2005. static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2006.     uint64_t temp[9*2];\

  2007.     uint64_t *temp_ptr= temp;\

  2008.     int count= 9;\

  2009. \

  2010.     /*FIXME unroll */\

  2011.     asm volatile(\

  2012.         "pxor %%mm7, %%mm7		\n\t"\

  2013.         "1:				\n\t"\

  2014.         "movq (%0), %%mm0		\n\t"\

  2015.         "movq (%0), %%mm1		\n\t"\

  2016.         "punpcklbw %%mm7, %%mm0		\n\t"\

  2017.         "punpckhbw %%mm7, %%mm1		\n\t"\

  2018.         "movq %%mm0, (%1)		\n\t"\

  2019.         "movq %%mm1, 9*8(%1)		\n\t"\

  2020.         "add $8, %1			\n\t"\

  2021.         "add %3, %0			\n\t"\

  2022.         "decl %2			\n\t"\

  2023.         " jnz 1b			\n\t"\

  2024.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  2025.         : "r" ((long)srcStride)\

  2026.         : "memory"\

  2027.     );\

  2028.     \

  2029.     temp_ptr= temp;\

  2030.     count=2;\

  2031.     \

  2032. /*FIXME reorder for speed */\

  2033.     asm volatile(\

  2034.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  2035.         "1:				\n\t"\

  2036.         "movq (%0), %%mm0		\n\t"\

  2037.         "movq 8(%0), %%mm1		\n\t"\

  2038.         "movq 16(%0), %%mm2		\n\t"\

  2039.         "movq 24(%0), %%mm3		\n\t"\

  2040.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  2041.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  2042.         "add %4, %1			\n\t"\

  2043.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  2044.         \

  2045.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  2046.         "add %4, %1			\n\t"\

  2047.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  2048.         \

  2049.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\

  2050.         "add %4, %1			\n\t"\

  2051.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\

  2052.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\

  2053.                 \

  2054.         "add $72, %0			\n\t"\

  2055.         "add %6, %1			\n\t"\

  2056.         "decl %2			\n\t"\

  2057.         " jnz 1b			\n\t"\

  2058.          \

  2059.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2060.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(long)dstStride)\

  2061.         : "memory"\

  2062.    );\

  2063. }\

  2064. \

  2065. static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2066.     OPNAME ## pixels8_mmx(dst, src, stride, 8);\

  2067. }\

  2068. \

  2069. static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2070.     uint64_t temp[8];\

  2071.     uint8_t * const half= (uint8_t*)temp;\

  2072.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2073.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2074. }\

  2075. \

  2076. static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2077.     OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\

  2078. }\

  2079. \

  2080. static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2081.     uint64_t temp[8];\

  2082.     uint8_t * const half= (uint8_t*)temp;\

  2083.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2084.     OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\

  2085. }\

  2086. \

  2087. static void OPNAME ## qpel8_mc01_ ## 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_v_lowpass_ ## MMX(half, src, 8, stride);\

  2091.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2092. }\

  2093. \

  2094. static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2095.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2096. }\

  2097. \

  2098. static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2099.     uint64_t temp[8];\

  2100.     uint8_t * const half= (uint8_t*)temp;\

  2101.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2102.     OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\

  2103. }\

  2104. static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2105.     uint64_t half[8 + 9];\

  2106.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2107.     uint8_t * const halfHV= ((uint8_t*)half);\

  2108.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2109.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2110.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2111.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2112. }\

  2113. static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2114.     uint64_t half[8 + 9];\

  2115.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2116.     uint8_t * const halfHV= ((uint8_t*)half);\

  2117.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2118.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2119.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2120.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2121. }\

  2122. static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2123.     uint64_t half[8 + 9];\

  2124.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2125.     uint8_t * const halfHV= ((uint8_t*)half);\

  2126.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2127.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2128.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2129.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2130. }\

  2131. static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2132.     uint64_t half[8 + 9];\

  2133.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2134.     uint8_t * const halfHV= ((uint8_t*)half);\

  2135.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2136.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2137.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2138.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2139. }\

  2140. static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2141.     uint64_t half[8 + 9];\

  2142.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2143.     uint8_t * const halfHV= ((uint8_t*)half);\

  2144.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2145.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2146.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2147. }\

  2148. static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2149.     uint64_t half[8 + 9];\

  2150.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2151.     uint8_t * const halfHV= ((uint8_t*)half);\

  2152.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2153.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2154.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2155. }\

  2156. static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2157.     uint64_t half[8 + 9];\

  2158.     uint8_t * const halfH= ((uint8_t*)half);\

  2159.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2160.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2161.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2162. }\

  2163. static void OPNAME ## qpel8_mc32_ ## 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+1, halfH, 8, stride, 9);\

  2168.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2169. }\

  2170. static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2171.     uint64_t half[9];\

  2172.     uint8_t * const halfH= ((uint8_t*)half);\

  2173.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2174.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2175. }\

  2176. static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2177.     OPNAME ## pixels16_mmx(dst, src, stride, 16);\

  2178. }\

  2179. \

  2180. static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2181.     uint64_t temp[32];\

  2182.     uint8_t * const half= (uint8_t*)temp;\

  2183.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2184.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2185. }\

  2186. \

  2187. static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2188.     OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\

  2189. }\

  2190. \

  2191. static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2192.     uint64_t temp[32];\

  2193.     uint8_t * const half= (uint8_t*)temp;\

  2194.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2195.     OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\

  2196. }\

  2197. \

  2198. static void OPNAME ## qpel16_mc01_ ## 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_v_lowpass_ ## MMX(half, src, 16, stride);\

  2202.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2203. }\

  2204. \

  2205. static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2206.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2207. }\

  2208. \

  2209. static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2210.     uint64_t temp[32];\

  2211.     uint8_t * const half= (uint8_t*)temp;\

  2212.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2213.     OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\

  2214. }\

  2215. static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2216.     uint64_t half[16*2 + 17*2];\

  2217.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2218.     uint8_t * const halfHV= ((uint8_t*)half);\

  2219.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2220.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2221.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2222.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2223. }\

  2224. static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2225.     uint64_t half[16*2 + 17*2];\

  2226.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2227.     uint8_t * const halfHV= ((uint8_t*)half);\

  2228.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2229.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2230.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2231.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2232. }\

  2233. static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2234.     uint64_t half[16*2 + 17*2];\

  2235.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2236.     uint8_t * const halfHV= ((uint8_t*)half);\

  2237.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2238.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2239.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2240.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2241. }\

  2242. static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2243.     uint64_t half[16*2 + 17*2];\

  2244.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2245.     uint8_t * const halfHV= ((uint8_t*)half);\

  2246.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2247.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2248.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2249.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2250. }\

  2251. static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2252.     uint64_t half[16*2 + 17*2];\

  2253.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2254.     uint8_t * const halfHV= ((uint8_t*)half);\

  2255.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2256.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2257.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2258. }\

  2259. static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2260.     uint64_t half[16*2 + 17*2];\

  2261.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2262.     uint8_t * const halfHV= ((uint8_t*)half);\

  2263.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2264.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2265.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2266. }\

  2267. static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2268.     uint64_t half[17*2];\

  2269.     uint8_t * const halfH= ((uint8_t*)half);\

  2270.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2271.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2272.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2273. }\

  2274. static void OPNAME ## qpel16_mc32_ ## 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+1, halfH, 16, stride, 17);\

  2279.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2280. }\

  2281. static void OPNAME ## qpel16_mc22_ ## 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.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2286. }

  2287. 

  2288. #define QPEL_H264V(A,B,C,D,E,F,OP)\

  2289.         "movd (%0), "#F"		\n\t"\

  2290.         "movq "#C", %%mm6		\n\t"\

  2291.         "paddw "#D", %%mm6		\n\t"\

  2292.         "psllw $2, %%mm6		\n\t"\

  2293.         "psubw "#B", %%mm6		\n\t"\

  2294.         "psubw "#E", %%mm6		\n\t"\

  2295.         "pmullw %4, %%mm6		\n\t"\

  2296.         "add %2, %0			\n\t"\

  2297.         "punpcklbw %%mm7, "#F"		\n\t"\

  2298.         "paddw %5, "#A"			\n\t"\

  2299.         "paddw "#F", "#A"		\n\t"\

  2300.         "paddw "#A", %%mm6		\n\t"\

  2301.         "psraw $5, %%mm6		\n\t"\

  2302.         "packuswb %%mm6, %%mm6		\n\t"\

  2303.         OP(%%mm6, (%1), A, d)\

  2304.         "add %3, %1			\n\t"     

  2305. 

  2306. #define QPEL_H264HV(A,B,C,D,E,F,OF)\

  2307.         "movd (%0), "#F"		\n\t"\

  2308.         "movq "#C", %%mm6		\n\t"\

  2309.         "paddw "#D", %%mm6		\n\t"\

  2310.         "psllw $2, %%mm6		\n\t"\

  2311.         "psubw "#B", %%mm6		\n\t"\

  2312.         "psubw "#E", %%mm6		\n\t"\

  2313.         "pmullw %3, %%mm6		\n\t"\

  2314.         "add %2, %0			\n\t"\

  2315.         "punpcklbw %%mm7, "#F"		\n\t"\

  2316.         "paddw "#F", "#A"		\n\t"\

  2317.         "paddw "#A", %%mm6		\n\t"\

  2318.         "movq %%mm6, "#OF"(%1)		\n\t"

  2319.         

  2320. #define QPEL_H264(OPNAME, OP, MMX)\

  2321. static void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2322.     int h=4;\

  2323. \

  2324.     asm volatile(\

  2325.         "pxor %%mm7, %%mm7		\n\t"\

  2326.         "movq %5, %%mm4			\n\t"\

  2327.         "movq %6, %%mm5			\n\t"\

  2328.         "1:				\n\t"\

  2329.         "movd  -1(%0), %%mm1		\n\t"\

  2330.         "movd    (%0), %%mm2		\n\t"\

  2331.         "movd   1(%0), %%mm3		\n\t"\

  2332.         "movd   2(%0), %%mm0		\n\t"\

  2333.         "punpcklbw %%mm7, %%mm1		\n\t"\

  2334.         "punpcklbw %%mm7, %%mm2		\n\t"\

  2335.         "punpcklbw %%mm7, %%mm3		\n\t"\

  2336.         "punpcklbw %%mm7, %%mm0		\n\t"\

  2337.         "paddw %%mm0, %%mm1		\n\t"\

  2338.         "paddw %%mm3, %%mm2		\n\t"\

  2339.         "movd  -2(%0), %%mm0		\n\t"\

  2340.         "movd   3(%0), %%mm3		\n\t"\

  2341.         "punpcklbw %%mm7, %%mm0		\n\t"\

  2342.         "punpcklbw %%mm7, %%mm3		\n\t"\

  2343.         "paddw %%mm3, %%mm0		\n\t"\

  2344.         "psllw $2, %%mm2		\n\t"\

  2345.         "psubw %%mm1, %%mm2		\n\t"\

  2346.         "pmullw %%mm4, %%mm2		\n\t"\

  2347.         "paddw %%mm5, %%mm0		\n\t"\

  2348.         "paddw %%mm2, %%mm0		\n\t"\

  2349.         "psraw $5, %%mm0		\n\t"\

  2350.         "packuswb %%mm0, %%mm0		\n\t"\

  2351.         OP(%%mm0, (%1),%%mm6, d)\

  2352.         "add %3, %0			\n\t"\

  2353.         "add %4, %1			\n\t"\

  2354.         "decl %2			\n\t"\

  2355.         " jnz 1b			\n\t"\

  2356.         : "+a"(src), "+c"(dst), "+m"(h)\

  2357.         : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  2358.         : "memory"\

  2359.     );\

  2360. }\

  2361. static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2362.     src -= 2*srcStride;\

  2363.     asm volatile(\

  2364.         "pxor %%mm7, %%mm7		\n\t"\

  2365.         "movd (%0), %%mm0		\n\t"\

  2366.         "add %2, %0			\n\t"\

  2367.         "movd (%0), %%mm1		\n\t"\

  2368.         "add %2, %0			\n\t"\

  2369.         "movd (%0), %%mm2		\n\t"\

  2370.         "add %2, %0			\n\t"\

  2371.         "movd (%0), %%mm3		\n\t"\

  2372.         "add %2, %0			\n\t"\

  2373.         "movd (%0), %%mm4		\n\t"\

  2374.         "add %2, %0			\n\t"\

  2375.         "punpcklbw %%mm7, %%mm0		\n\t"\

  2376.         "punpcklbw %%mm7, %%mm1		\n\t"\

  2377.         "punpcklbw %%mm7, %%mm2		\n\t"\

  2378.         "punpcklbw %%mm7, %%mm3		\n\t"\

  2379.         "punpcklbw %%mm7, %%mm4		\n\t"\

  2380.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  2381.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  2382.         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  2383.         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

  2384.          \

  2385.         : "+a"(src), "+c"(dst)\

  2386.         : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  2387.         : "memory"\

  2388.     );\

  2389. }\

  2390. static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  2391.     int h=4;\

  2392.     int w=3;\

  2393.     src -= 2*srcStride+2;\

  2394.     while(w--){\

  2395.         asm volatile(\

  2396.             "pxor %%mm7, %%mm7			\n\t"\

  2397.             "movd (%0), %%mm0			\n\t"\

  2398.             "add %2, %0				\n\t"\

  2399.             "movd (%0), %%mm1			\n\t"\

  2400.             "add %2, %0				\n\t"\

  2401.             "movd (%0), %%mm2			\n\t"\

  2402.             "add %2, %0				\n\t"\

  2403.             "movd (%0), %%mm3			\n\t"\

  2404.             "add %2, %0				\n\t"\

  2405.             "movd (%0), %%mm4			\n\t"\

  2406.             "add %2, %0				\n\t"\

  2407.             "punpcklbw %%mm7, %%mm0		\n\t"\

  2408.             "punpcklbw %%mm7, %%mm1		\n\t"\

  2409.             "punpcklbw %%mm7, %%mm2		\n\t"\

  2410.             "punpcklbw %%mm7, %%mm3		\n\t"\

  2411.             "punpcklbw %%mm7, %%mm4		\n\t"\

  2412.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\

  2413.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\

  2414.             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\

  2415.             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\

  2416.              \

  2417.             : "+a"(src)\

  2418.             : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  2419.             : "memory"\

  2420.         );\

  2421.         tmp += 4;\

  2422.         src += 4 - 9*srcStride;\

  2423.     }\

  2424.     tmp -= 3*4;\

  2425.     asm volatile(\

  2426.         "movq %4, %%mm6			\n\t"\

  2427.         "1:				\n\t"\

  2428.         "movq     (%0), %%mm0		\n\t"\

  2429.         "paddw  10(%0), %%mm0		\n\t"\

  2430.         "movq    2(%0), %%mm1		\n\t"\

  2431.         "paddw   8(%0), %%mm1		\n\t"\

  2432.         "movq    4(%0), %%mm2		\n\t"\

  2433.         "paddw   6(%0), %%mm2		\n\t"\

  2434.         "psubw %%mm1, %%mm0		\n\t"/*a-b   (abccba)*/\

  2435.         "psraw $2, %%mm0		\n\t"/*(a-b)/4 */\

  2436.         "psubw %%mm1, %%mm0		\n\t"/*(a-b)/4-b */\

  2437.         "paddsw %%mm2, %%mm0		\n\t"\

  2438.         "psraw $2, %%mm0		\n\t"/*((a-b)/4-b)/4 */\

  2439.         "paddw %%mm6, %%mm2		\n\t"\

  2440.         "paddw %%mm2, %%mm0		\n\t"\

  2441.         "psraw $6, %%mm0		\n\t"\

  2442.         "packuswb %%mm0, %%mm0		\n\t"\

  2443.         OP(%%mm0, (%1),%%mm7, d)\

  2444.         "add $24, %0			\n\t"\

  2445.         "add %3, %1			\n\t"\

  2446.         "decl %2			\n\t"\

  2447.         " jnz 1b			\n\t"\

  2448.         : "+a"(tmp), "+c"(dst), "+m"(h)\

  2449.         : "S"((long)dstStride), "m"(ff_pw_32)\

  2450.         : "memory"\

  2451.     );\

  2452. }\

  2453. \

  2454. static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2455.     int h=8;\

  2456.     asm volatile(\

  2457.         "pxor %%mm7, %%mm7		\n\t"\

  2458.         "movq %5, %%mm6			\n\t"\

  2459.         "1:				\n\t"\

  2460.         "movq    (%0), %%mm0		\n\t"\

  2461.         "movq   1(%0), %%mm2		\n\t"\

  2462.         "movq %%mm0, %%mm1		\n\t"\

  2463.         "movq %%mm2, %%mm3		\n\t"\

  2464.         "punpcklbw %%mm7, %%mm0		\n\t"\

  2465.         "punpckhbw %%mm7, %%mm1		\n\t"\

  2466.         "punpcklbw %%mm7, %%mm2		\n\t"\

  2467.         "punpckhbw %%mm7, %%mm3		\n\t"\

  2468.         "paddw %%mm2, %%mm0		\n\t"\

  2469.         "paddw %%mm3, %%mm1		\n\t"\

  2470.         "psllw $2, %%mm0		\n\t"\

  2471.         "psllw $2, %%mm1		\n\t"\

  2472.         "movq   -1(%0), %%mm2		\n\t"\

  2473.         "movq    2(%0), %%mm4		\n\t"\

  2474.         "movq %%mm2, %%mm3		\n\t"\

  2475.         "movq %%mm4, %%mm5		\n\t"\

  2476.         "punpcklbw %%mm7, %%mm2		\n\t"\

  2477.         "punpckhbw %%mm7, %%mm3		\n\t"\

  2478.         "punpcklbw %%mm7, %%mm4		\n\t"\

  2479.         "punpckhbw %%mm7, %%mm5		\n\t"\

  2480.         "paddw %%mm4, %%mm2		\n\t"\

  2481.         "paddw %%mm3, %%mm5		\n\t"\

  2482.         "psubw %%mm2, %%mm0		\n\t"\

  2483.         "psubw %%mm5, %%mm1		\n\t"\

  2484.         "pmullw %%mm6, %%mm0		\n\t"\

  2485.         "pmullw %%mm6, %%mm1		\n\t"\

  2486.         "movd   -2(%0), %%mm2		\n\t"\

  2487.         "movd    7(%0), %%mm5		\n\t"\

  2488.         "punpcklbw %%mm7, %%mm2		\n\t"\

  2489.         "punpcklbw %%mm7, %%mm5		\n\t"\

  2490.         "paddw %%mm3, %%mm2		\n\t"\

  2491.         "paddw %%mm5, %%mm4		\n\t"\

  2492.         "movq %6, %%mm5			\n\t"\

  2493.         "paddw %%mm5, %%mm2		\n\t"\

  2494.         "paddw %%mm5, %%mm4		\n\t"\

  2495.         "paddw %%mm2, %%mm0		\n\t"\

  2496.         "paddw %%mm4, %%mm1		\n\t"\

  2497.         "psraw $5, %%mm0		\n\t"\

  2498.         "psraw $5, %%mm1		\n\t"\

  2499.         "packuswb %%mm1, %%mm0		\n\t"\

  2500.         OP(%%mm0, (%1),%%mm5, q)\

  2501.         "add %3, %0			\n\t"\

  2502.         "add %4, %1			\n\t"\

  2503.         "decl %2			\n\t"\

  2504.         " jnz 1b			\n\t"\

  2505.         : "+a"(src), "+c"(dst), "+m"(h)\

  2506.         : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  2507.         : "memory"\

  2508.     );\

  2509. }\

  2510. \

  2511. static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2512.     int h= 2;\

  2513.     src -= 2*srcStride;\

  2514.     \

  2515.     while(h--){\

  2516.       asm volatile(\

  2517.         "pxor %%mm7, %%mm7		\n\t"\

  2518.         "movd (%0), %%mm0		\n\t"\

  2519.         "add %2, %0			\n\t"\

  2520.         "movd (%0), %%mm1		\n\t"\

  2521.         "add %2, %0			\n\t"\

  2522.         "movd (%0), %%mm2		\n\t"\

  2523.         "add %2, %0			\n\t"\

  2524.         "movd (%0), %%mm3		\n\t"\

  2525.         "add %2, %0			\n\t"\

  2526.         "movd (%0), %%mm4		\n\t"\

  2527.         "add %2, %0			\n\t"\

  2528.         "punpcklbw %%mm7, %%mm0		\n\t"\

  2529.         "punpcklbw %%mm7, %%mm1		\n\t"\

  2530.         "punpcklbw %%mm7, %%mm2		\n\t"\

  2531.         "punpcklbw %%mm7, %%mm3		\n\t"\

  2532.         "punpcklbw %%mm7, %%mm4		\n\t"\

  2533.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  2534.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  2535.         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  2536.         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

  2537.         QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\

  2538.         QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\

  2539.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  2540.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  2541.          \

  2542.         : "+a"(src), "+c"(dst)\

  2543.         : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  2544.         : "memory"\

  2545.      );\

  2546.      src += 4-13*srcStride;\

  2547.      dst +=  4-8*dstStride;\

  2548.    }\

  2549. }\

  2550. static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  2551.     int h=8;\

  2552.     int w=4;\

  2553.     src -= 2*srcStride+2;\

  2554.     while(w--){\

  2555.         asm volatile(\

  2556.             "pxor %%mm7, %%mm7			\n\t"\

  2557.             "movd (%0), %%mm0			\n\t"\

  2558.             "add %2, %0				\n\t"\

  2559.             "movd (%0), %%mm1			\n\t"\

  2560.             "add %2, %0				\n\t"\

  2561.             "movd (%0), %%mm2			\n\t"\

  2562.             "add %2, %0				\n\t"\

  2563.             "movd (%0), %%mm3			\n\t"\

  2564.             "add %2, %0				\n\t"\

  2565.             "movd (%0), %%mm4			\n\t"\

  2566.             "add %2, %0				\n\t"\

  2567.             "punpcklbw %%mm7, %%mm0		\n\t"\

  2568.             "punpcklbw %%mm7, %%mm1		\n\t"\

  2569.             "punpcklbw %%mm7, %%mm2		\n\t"\

  2570.             "punpcklbw %%mm7, %%mm3		\n\t"\

  2571.             "punpcklbw %%mm7, %%mm4		\n\t"\

  2572.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*4)\

  2573.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*4)\

  2574.             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*4)\

  2575.             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*4)\

  2576.             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*8*4)\

  2577.             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*8*4)\

  2578.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*8*4)\

  2579.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*8*4)\

  2580.              \

  2581.             : "+a"(src)\

  2582.             : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  2583.             : "memory"\

  2584.         );\

  2585.         tmp += 4;\

  2586.         src += 4 - 13*srcStride;\

  2587.     }\

  2588.     tmp -= 4*4;\

  2589.     asm volatile(\

  2590.         "movq %4, %%mm6			\n\t"\

  2591.         "1:				\n\t"\

  2592.         "movq     (%0), %%mm0		\n\t"\

  2593.         "movq    8(%0), %%mm3		\n\t"\

  2594.         "movq    2(%0), %%mm1		\n\t"\

  2595.         "movq   10(%0), %%mm4		\n\t"\

  2596.         "paddw   %%mm4, %%mm0		\n\t"\

  2597.         "paddw   %%mm3, %%mm1		\n\t"\

  2598.         "paddw  18(%0), %%mm3		\n\t"\

  2599.         "paddw  16(%0), %%mm4		\n\t"\

  2600.         "movq    4(%0), %%mm2		\n\t"\

  2601.         "movq   12(%0), %%mm5		\n\t"\

  2602.         "paddw   6(%0), %%mm2		\n\t"\

  2603.         "paddw  14(%0), %%mm5		\n\t"\

  2604.         "psubw %%mm1, %%mm0		\n\t"\

  2605.         "psubw %%mm4, %%mm3		\n\t"\

  2606.         "psraw $2, %%mm0		\n\t"\

  2607.         "psraw $2, %%mm3		\n\t"\

  2608.         "psubw %%mm1, %%mm0		\n\t"\

  2609.         "psubw %%mm4, %%mm3		\n\t"\

  2610.         "paddsw %%mm2, %%mm0		\n\t"\

  2611.         "paddsw %%mm5, %%mm3		\n\t"\

  2612.         "psraw $2, %%mm0		\n\t"\

  2613.         "psraw $2, %%mm3		\n\t"\

  2614.         "paddw %%mm6, %%mm2		\n\t"\

  2615.         "paddw %%mm6, %%mm5		\n\t"\

  2616.         "paddw %%mm2, %%mm0		\n\t"\

  2617.         "paddw %%mm5, %%mm3		\n\t"\

  2618.         "psraw $6, %%mm0		\n\t"\

  2619.         "psraw $6, %%mm3		\n\t"\

  2620.         "packuswb %%mm3, %%mm0		\n\t"\

  2621.         OP(%%mm0, (%1),%%mm7, q)\

  2622.         "add $32, %0			\n\t"\

  2623.         "add %3, %1			\n\t"\

  2624.         "decl %2			\n\t"\

  2625.         " jnz 1b			\n\t"\

  2626.         : "+a"(tmp), "+c"(dst), "+m"(h)\

  2627.         : "S"((long)dstStride), "m"(ff_pw_32)\

  2628.         : "memory"\

  2629.     );\

  2630. }\

  2631. static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2632.     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  2633.     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  2634.     src += 8*srcStride;\

  2635.     dst += 8*dstStride;\

  2636.     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  2637.     OPNAME ## h264_qpel8_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  2638. }\

  2639. \

  2640. static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2641.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  2642.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  2643.     src += 8*srcStride;\

  2644.     dst += 8*dstStride;\

  2645.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  2646.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  2647. }\

  2648. \

  2649. static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  2650.     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride);\

  2651.     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp  , src+8, dstStride, tmpStride, srcStride);\

  2652.     src += 8*srcStride;\

  2653.     dst += 8*dstStride;\

  2654.     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride);\

  2655.     OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(dst+8, tmp  , src+8, dstStride, tmpStride, srcStride);\

  2656. }\

  2657. 

  2658. #define H264_MC(OPNAME, SIZE, MMX) \

  2659. static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2660.     OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\

  2661. }\

  2662. \

  2663. static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2664.     uint64_t temp[SIZE*SIZE/8];\

  2665.     uint8_t * const half= (uint8_t*)temp;\

  2666.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\

  2667.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\

  2668. }\

  2669. \

  2670. static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2671.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\

  2672. }\

  2673. \

  2674. static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2675.     uint64_t temp[SIZE*SIZE/8];\

  2676.     uint8_t * const half= (uint8_t*)temp;\

  2677.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\

  2678.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+1, half, stride, stride, SIZE);\

  2679. }\

  2680. \

  2681. static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2682.     uint64_t temp[SIZE*SIZE/8];\

  2683.     uint8_t * const half= (uint8_t*)temp;\

  2684.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  2685.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\

  2686. }\

  2687. \

  2688. static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2689.     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\

  2690. }\

  2691. \

  2692. static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2693.     uint64_t temp[SIZE*SIZE/8];\

  2694.     uint8_t * const half= (uint8_t*)temp;\

  2695.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  2696.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\

  2697. }\

  2698. \

  2699. static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2700.     uint64_t temp[SIZE*SIZE/4];\

  2701.     uint8_t * const halfH= (uint8_t*)temp;\

  2702.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  2703.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\

  2704.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  2705.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  2706. }\

  2707. \

  2708. static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2709.     uint64_t temp[SIZE*SIZE/4];\

  2710.     uint8_t * const halfH= (uint8_t*)temp;\

  2711.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  2712.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\

  2713.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  2714.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  2715. }\

  2716. \

  2717. static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2718.     uint64_t temp[SIZE*SIZE/4];\

  2719.     uint8_t * const halfH= (uint8_t*)temp;\

  2720.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  2721.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\

  2722.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  2723.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  2724. }\

  2725. \

  2726. static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2727.     uint64_t temp[SIZE*SIZE/4];\

  2728.     uint8_t * const halfH= (uint8_t*)temp;\

  2729.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  2730.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\

  2731.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  2732.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  2733. }\

  2734. \

  2735. static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2736.     uint64_t temp[SIZE*(SIZE+8)/4];\

  2737.     int16_t * const tmp= (int16_t*)temp;\

  2738.     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\

  2739. }\

  2740. \

  2741. static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2742.     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\

  2743.     uint8_t * const halfH= (uint8_t*)temp;\

  2744.     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\

  2745.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\

  2746.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\

  2747.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  2748.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\

  2749. }\

  2750. \

  2751. static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2752.     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\

  2753.     uint8_t * const halfH= (uint8_t*)temp;\

  2754.     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\

  2755.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\

  2756.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\

  2757.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  2758.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\

  2759. }\

  2760. \

  2761. static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2762.     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\

  2763.     uint8_t * const halfV= (uint8_t*)temp;\

  2764.     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\

  2765.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\

  2766.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  2767.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  2768.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\

  2769. }\

  2770. \

  2771. static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2772.     uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4];\

  2773.     uint8_t * const halfV= (uint8_t*)temp;\

  2774.     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\

  2775.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\

  2776.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  2777.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  2778.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfV, halfHV, stride, SIZE, SIZE);\

  2779. }\

  2780. 

  2781. 

  2782. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "	\n\t"

  2783. #define AVG_3DNOW_OP(a,b,temp, size) \

  2784. "mov" #size " " #b ", " #temp "	\n\t"\

  2785. "pavgusb " #temp ", " #a "	\n\t"\

  2786. "mov" #size " " #a ", " #b "	\n\t"

  2787. #define AVG_MMX2_OP(a,b,temp, size) \

  2788. "mov" #size " " #b ", " #temp "	\n\t"\

  2789. "pavgb " #temp ", " #a "	\n\t"\

  2790. "mov" #size " " #a ", " #b "	\n\t"

  2791. 

  2792. QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)

  2793. QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)

  2794. QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)

  2795. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)

  2796. QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)

  2797. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)

  2798. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)

  2799. QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)

  2800. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

  2801. 

  2802. QPEL_H264(put_       ,       PUT_OP, 3dnow)

  2803. QPEL_H264(avg_       , AVG_3DNOW_OP, 3dnow)

  2804. QPEL_H264(put_       ,       PUT_OP, mmx2)

  2805. QPEL_H264(avg_       ,  AVG_MMX2_OP, mmx2)

  2806. 

  2807. H264_MC(put_, 4, 3dnow)

  2808. H264_MC(put_, 8, 3dnow)

  2809. H264_MC(put_, 16,3dnow)

  2810. H264_MC(avg_, 4, 3dnow)

  2811. H264_MC(avg_, 8, 3dnow)

  2812. H264_MC(avg_, 16,3dnow)

  2813. H264_MC(put_, 4, mmx2)

  2814. H264_MC(put_, 8, mmx2)

  2815. H264_MC(put_, 16,mmx2)

  2816. H264_MC(avg_, 4, mmx2)

  2817. H264_MC(avg_, 8, mmx2)

  2818. H264_MC(avg_, 16,mmx2)

  2819. 

  2820. #if 0

  2821. static void just_return() { return; }

  2822. #endif

  2823. 

  2824. #define SET_QPEL_FUNC(postfix1, postfix2) \

  2825.     c->put_ ## postfix1 = put_ ## postfix2;\

  2826.     c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\

  2827.     c->avg_ ## postfix1 = avg_ ## postfix2;

  2828. 

  2829. static int try_8x8basis_mmx(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){

  2830.     long i=0;

  2831.     

  2832.     assert(ABS(scale) < 256);

  2833.     scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2834. 

  2835.     asm volatile(

  2836.         "pcmpeqw %%mm6, %%mm6		\n\t" // -1w

  2837.         "psrlw $15, %%mm6		\n\t" //  1w

  2838.         "pxor %%mm7, %%mm7		\n\t"

  2839.         "movd  %4, %%mm5		\n\t" 

  2840.         "punpcklwd %%mm5, %%mm5		\n\t" 

  2841.         "punpcklwd %%mm5, %%mm5		\n\t" 

  2842.         "1:				\n\t"

  2843.         "movq  (%1, %0), %%mm0		\n\t" 

  2844.         "movq  8(%1, %0), %%mm1		\n\t"

  2845.         "pmulhw %%mm5, %%mm0		\n\t"

  2846.         "pmulhw %%mm5, %%mm1		\n\t"

  2847.         "paddw %%mm6, %%mm0		\n\t"

  2848.         "paddw %%mm6, %%mm1		\n\t"

  2849.         "psraw $1, %%mm0		\n\t"

  2850.         "psraw $1, %%mm1		\n\t"

  2851.         "paddw (%2, %0), %%mm0		\n\t"

  2852.         "paddw 8(%2, %0), %%mm1		\n\t"

  2853.         "psraw $6, %%mm0		\n\t"

  2854.         "psraw $6, %%mm1		\n\t"

  2855.         "pmullw (%3, %0), %%mm0		\n\t"

  2856.         "pmullw 8(%3, %0), %%mm1	\n\t"

  2857.         "pmaddwd %%mm0, %%mm0		\n\t"

  2858.         "pmaddwd %%mm1, %%mm1		\n\t"

  2859.         "paddd %%mm1, %%mm0		\n\t"

  2860.         "psrld $4, %%mm0		\n\t"

  2861.         "paddd %%mm0, %%mm7		\n\t"

  2862.         "add $16, %0			\n\t"

  2863.         "cmp $128, %0			\n\t" //FIXME optimize & bench

  2864.         " jb 1b				\n\t"

  2865.         "movq %%mm7, %%mm6		\n\t"

  2866.         "psrlq $32, %%mm7		\n\t"

  2867.         "paddd %%mm6, %%mm7		\n\t"

  2868.         "psrld $2, %%mm7		\n\t"

  2869.         "movd %%mm7, %0			\n\t"

  2870.         

  2871.         : "+r" (i)

  2872.         : "r"(basis), "r"(rem), "r"(weight), "g"(scale)

  2873.     );

  2874.     return i;

  2875. }

  2876. 

  2877. static void add_8x8basis_mmx(int16_t rem[64], int16_t basis[64], int scale){

  2878.     long i=0;

  2879.     

  2880.     if(ABS(scale) < 256){

  2881.         scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2882.         asm volatile(

  2883.                 "pcmpeqw %%mm6, %%mm6		\n\t" // -1w

  2884.                 "psrlw $15, %%mm6		\n\t" //  1w

  2885.                 "movd  %3, %%mm5		\n\t" 

  2886.                 "punpcklwd %%mm5, %%mm5		\n\t" 

  2887.                 "punpcklwd %%mm5, %%mm5		\n\t" 

  2888.                 "1:				\n\t"

  2889.                 "movq  (%1, %0), %%mm0		\n\t" 

  2890.                 "movq  8(%1, %0), %%mm1		\n\t"

  2891.                 "pmulhw %%mm5, %%mm0		\n\t"

  2892.                 "pmulhw %%mm5, %%mm1		\n\t"

  2893.                 "paddw %%mm6, %%mm0		\n\t" 

  2894.                 "paddw %%mm6, %%mm1		\n\t"

  2895.                 "psraw $1, %%mm0		\n\t"

  2896.                 "psraw $1, %%mm1		\n\t"

  2897.                 "paddw (%2, %0), %%mm0		\n\t"

  2898.                 "paddw 8(%2, %0), %%mm1		\n\t"

  2899.                 "movq %%mm0, (%2, %0)		\n\t"

  2900.                 "movq %%mm1, 8(%2, %0)		\n\t"

  2901.                 "add $16, %0			\n\t"

  2902.                 "cmp $128, %0			\n\t" //FIXME optimize & bench

  2903.                 " jb 1b				\n\t"

  2904.                 

  2905.                 : "+r" (i)

  2906.                 : "r"(basis), "r"(rem), "g"(scale)

  2907.         );

  2908.     }else{

  2909.         for(i=0; i<8*8; i++){

  2910.             rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);

  2911.         }    

  2912.     }

  2913. }

  2914.     

  2915. /* external functions, from idct_mmx.c */

  2916. void ff_mmx_idct(DCTELEM *block);

  2917. void ff_mmxext_idct(DCTELEM *block);

  2918. 

  2919. /* XXX: those functions should be suppressed ASAP when all IDCTs are

  2920.    converted */

  2921. static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2922. {

  2923.     ff_mmx_idct (block);

  2924.     put_pixels_clamped_mmx(block, dest, line_size);

  2925. }

  2926. static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2927. {

  2928.     ff_mmx_idct (block);

  2929.     add_pixels_clamped_mmx(block, dest, line_size);

  2930. }

  2931. static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2932. {

  2933.     ff_mmxext_idct (block);

  2934.     put_pixels_clamped_mmx(block, dest, line_size);

  2935. }

  2936. static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2937. {

  2938.     ff_mmxext_idct (block);

  2939.     add_pixels_clamped_mmx(block, dest, line_size);

  2940. }

  2941.     

  2942. void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)

  2943. {

  2944.     mm_flags = mm_support();

  2945. 

  2946.     if (avctx->dsp_mask) {

  2947. 	if (avctx->dsp_mask & FF_MM_FORCE)

  2948. 	    mm_flags |= (avctx->dsp_mask & 0xffff);

  2949. 	else

  2950. 	    mm_flags &= ~(avctx->dsp_mask & 0xffff);

  2951.     }

  2952. 

  2953. #if 0

  2954.     av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");

  2955.     if (mm_flags & MM_MMX)

  2956.         av_log(avctx, AV_LOG_INFO, " mmx");

  2957.     if (mm_flags & MM_MMXEXT)

  2958.         av_log(avctx, AV_LOG_INFO, " mmxext");

  2959.     if (mm_flags & MM_3DNOW)

  2960.         av_log(avctx, AV_LOG_INFO, " 3dnow");

  2961.     if (mm_flags & MM_SSE)

  2962.         av_log(avctx, AV_LOG_INFO, " sse");

  2963.     if (mm_flags & MM_SSE2)

  2964.         av_log(avctx, AV_LOG_INFO, " sse2");

  2965.     av_log(avctx, AV_LOG_INFO, "\n");

  2966. #endif

  2967. 

  2968.     if (mm_flags & MM_MMX) {

  2969.         const int idct_algo= avctx->idct_algo;

  2970. 

  2971. #ifdef CONFIG_ENCODERS

  2972.         const int dct_algo = avctx->dct_algo;

  2973.         if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){

  2974.             if(mm_flags & MM_SSE2){

  2975.                 c->fdct = ff_fdct_sse2;

  2976. 	    }else if(mm_flags & MM_MMXEXT){

  2977.                 c->fdct = ff_fdct_mmx2;

  2978.             }else{

  2979.                 c->fdct = ff_fdct_mmx;

  2980.             }

  2981.         }

  2982. #endif //CONFIG_ENCODERS

  2983.         if(avctx->lowres==0){

  2984.             if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){

  2985.                 c->idct_put= ff_simple_idct_put_mmx;

  2986.                 c->idct_add= ff_simple_idct_add_mmx;

  2987.                 c->idct    = ff_simple_idct_mmx;

  2988.                 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;

  2989.             }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){

  2990.                 if(mm_flags & MM_MMXEXT){

  2991.                     c->idct_put= ff_libmpeg2mmx2_idct_put;

  2992.                     c->idct_add= ff_libmpeg2mmx2_idct_add;

  2993.                     c->idct    = ff_mmxext_idct;

  2994.                 }else{

  2995.                     c->idct_put= ff_libmpeg2mmx_idct_put;

  2996.                     c->idct_add= ff_libmpeg2mmx_idct_add;

  2997.                     c->idct    = ff_mmx_idct;

  2998.                 }

  2999.                 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;

  3000.             }

  3001.         }

  3002. 

  3003.         /* VP3 optimized DSP functions */

  3004.         if (mm_flags & MM_SSE2) {

  3005.             c->vp3_dsp_init = vp3_dsp_init_sse2;

  3006.             c->vp3_idct = vp3_idct_sse2;

  3007.         } else {

  3008.             c->vp3_dsp_init = vp3_dsp_init_mmx;

  3009.             c->vp3_idct = vp3_idct_mmx;

  3010.         }

  3011. 

  3012. #ifdef CONFIG_ENCODERS

  3013.         c->get_pixels = get_pixels_mmx;

  3014.         c->diff_pixels = diff_pixels_mmx;

  3015. #endif //CONFIG_ENCODERS

  3016.         c->put_pixels_clamped = put_pixels_clamped_mmx;

  3017.         c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;

  3018.         c->add_pixels_clamped = add_pixels_clamped_mmx;

  3019.         c->clear_blocks = clear_blocks_mmx;

  3020. #ifdef CONFIG_ENCODERS

  3021.         c->pix_sum = pix_sum16_mmx;

  3022. #endif //CONFIG_ENCODERS

  3023. 

  3024.         c->put_pixels_tab[0][0] = put_pixels16_mmx;

  3025.         c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;

  3026.         c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;

  3027.         c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

  3028. 

  3029.         c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;

  3030.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;

  3031.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;

  3032.         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

  3033. 

  3034.         c->avg_pixels_tab[0][0] = avg_pixels16_mmx;

  3035.         c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;

  3036.         c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;

  3037.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

  3038. 

  3039.         c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;

  3040.         c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;

  3041.         c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;

  3042.         c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

  3043. 

  3044.         c->put_pixels_tab[1][0] = put_pixels8_mmx;

  3045.         c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;

  3046.         c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;

  3047.         c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

  3048. 

  3049.         c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;

  3050.         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;

  3051.         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;

  3052.         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

  3053. 

  3054.         c->avg_pixels_tab[1][0] = avg_pixels8_mmx;

  3055.         c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;

  3056.         c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;

  3057.         c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

  3058. 

  3059.         c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;

  3060.         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;

  3061.         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;

  3062.         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;

  3063.                 

  3064.         c->add_bytes= add_bytes_mmx;

  3065. #ifdef CONFIG_ENCODERS

  3066.         c->diff_bytes= diff_bytes_mmx;

  3067.         

  3068.         c->hadamard8_diff[0]= hadamard8_diff16_mmx;

  3069.         c->hadamard8_diff[1]= hadamard8_diff_mmx;

  3070.         

  3071. 	c->pix_norm1 = pix_norm1_mmx;

  3072. 	c->sse[0] = sse16_mmx;

  3073.   	c->sse[1] = sse8_mmx;

  3074.         c->vsad[4]= vsad_intra16_mmx;

  3075. 

  3076. 	c->nsse[0] = nsse16_mmx;

  3077. 	c->nsse[1] = nsse8_mmx;

  3078.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3079.             c->vsad[0] = vsad16_mmx;

  3080.         }

  3081.         

  3082.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3083.             c->try_8x8basis= try_8x8basis_mmx;

  3084.         }

  3085.         c->add_8x8basis= add_8x8basis_mmx;

  3086.         

  3087. #endif //CONFIG_ENCODERS

  3088. 

  3089.         c->h263_v_loop_filter= h263_v_loop_filter_mmx;

  3090.         c->h263_h_loop_filter= h263_h_loop_filter_mmx;        

  3091.         

  3092.         if (mm_flags & MM_MMXEXT) {

  3093.             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;

  3094.             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

  3095. 

  3096.             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;

  3097.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;

  3098.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

  3099. 

  3100.             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;

  3101.             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

  3102. 

  3103.             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;

  3104.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;

  3105.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

  3106. 

  3107. #ifdef CONFIG_ENCODERS

  3108.             c->hadamard8_diff[0]= hadamard8_diff16_mmx2;

  3109.             c->hadamard8_diff[1]= hadamard8_diff_mmx2;

  3110.             c->vsad[4]= vsad_intra16_mmx2;

  3111. #endif //CONFIG_ENCODERS

  3112. 

  3113.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3114.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;

  3115.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

  3116.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

  3117.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

  3118.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

  3119.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;

  3120. #ifdef CONFIG_ENCODERS

  3121.                 c->vsad[0] = vsad16_mmx2;

  3122. #endif //CONFIG_ENCODERS

  3123.             }

  3124. 

  3125. #if 1

  3126.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)

  3127.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)

  3128.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)

  3129.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)

  3130.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)

  3131.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)

  3132.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)

  3133.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)

  3134.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)

  3135.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)

  3136.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)

  3137.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)

  3138.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)

  3139.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)

  3140.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)

  3141.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)

  3142.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)

  3143.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)

  3144.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)

  3145.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)

  3146.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)

  3147.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)

  3148.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)

  3149.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)

  3150.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)

  3151.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)

  3152.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)

  3153.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)

  3154.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)

  3155.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)

  3156.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)

  3157.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)

  3158. #endif

  3159. 

  3160. //FIXME 3dnow too

  3161. #define dspfunc(PFX, IDX, NUM) \

  3162.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_mmx2; \

  3163.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_mmx2; \

  3164.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_mmx2; \

  3165.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_mmx2; \

  3166.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_mmx2; \

  3167.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_mmx2; \

  3168.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_mmx2; \

  3169.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_mmx2; \

  3170.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_mmx2; \

  3171.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_mmx2; \

  3172.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_mmx2; \

  3173.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_mmx2; \

  3174.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_mmx2; \

  3175.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_mmx2; \

  3176.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_mmx2; \

  3177.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_mmx2

  3178. 

  3179.             dspfunc(put_h264_qpel, 0, 16);

  3180.             dspfunc(put_h264_qpel, 1, 8);

  3181.             dspfunc(put_h264_qpel, 2, 4);

  3182.             dspfunc(avg_h264_qpel, 0, 16);

  3183.             dspfunc(avg_h264_qpel, 1, 8);

  3184.             dspfunc(avg_h264_qpel, 2, 4);

  3185. #undef dspfunc

  3186. 

  3187. #ifdef CONFIG_ENCODERS

  3188.             c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;

  3189. #endif //CONFIG_ENCODERS

  3190.         } else if (mm_flags & MM_3DNOW) {

  3191.             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;

  3192.             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  3193. 

  3194.             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  3195.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  3196.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  3197. 

  3198.             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;

  3199.             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

  3200. 

  3201.             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;

  3202.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;

  3203.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

  3204. 

  3205.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3206.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  3207.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  3208.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;

  3209.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

  3210.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  3211.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;

  3212.             }

  3213. 

  3214.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)

  3215.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)

  3216.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)

  3217.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)

  3218.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)

  3219.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)

  3220.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)

  3221.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)

  3222.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)

  3223.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)

  3224.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)

  3225.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)

  3226.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)

  3227.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)

  3228.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)

  3229.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)

  3230.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)

  3231.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)

  3232.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)

  3233.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)

  3234.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)

  3235.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)

  3236.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)

  3237.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)

  3238.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)

  3239.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)

  3240.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)

  3241.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)

  3242.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)

  3243.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)

  3244.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)

  3245.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)

  3246. 

  3247. #define dspfunc(PFX, IDX, NUM) \

  3248.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_3dnow; \

  3249.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_3dnow; \

  3250.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_3dnow; \

  3251.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_3dnow; \

  3252.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_3dnow; \

  3253.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_3dnow; \

  3254.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_3dnow; \

  3255.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_3dnow; \

  3256.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_3dnow; \

  3257.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_3dnow; \

  3258.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_3dnow; \

  3259.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_3dnow; \

  3260.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_3dnow; \

  3261.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_3dnow; \

  3262.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_3dnow; \

  3263.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_3dnow

  3264. 

  3265.             dspfunc(put_h264_qpel, 0, 16);

  3266.             dspfunc(put_h264_qpel, 1, 8);

  3267.             dspfunc(put_h264_qpel, 2, 4);

  3268.             dspfunc(avg_h264_qpel, 0, 16);

  3269.             dspfunc(avg_h264_qpel, 1, 8);

  3270.             dspfunc(avg_h264_qpel, 2, 4);

  3271.         }

  3272.     }

  3273.         

  3274. #ifdef CONFIG_ENCODERS

  3275.     dsputil_init_pix_mmx(c, avctx);

  3276. #endif //CONFIG_ENCODERS

  3277. #if 0

  3278.     // for speed testing

  3279.     get_pixels = just_return;

  3280.     put_pixels_clamped = just_return;

  3281.     add_pixels_clamped = just_return;

  3282. 

  3283.     pix_abs16x16 = just_return;

  3284.     pix_abs16x16_x2 = just_return;

  3285.     pix_abs16x16_y2 = just_return;

  3286.     pix_abs16x16_xy2 = just_return;

  3287. 

  3288.     put_pixels_tab[0] = just_return;

  3289.     put_pixels_tab[1] = just_return;

  3290.     put_pixels_tab[2] = just_return;

  3291.     put_pixels_tab[3] = just_return;

  3292. 

  3293.     put_no_rnd_pixels_tab[0] = just_return;

  3294.     put_no_rnd_pixels_tab[1] = just_return;

  3295.     put_no_rnd_pixels_tab[2] = just_return;

  3296.     put_no_rnd_pixels_tab[3] = just_return;

  3297. 

  3298.     avg_pixels_tab[0] = just_return;

  3299.     avg_pixels_tab[1] = just_return;

  3300.     avg_pixels_tab[2] = just_return;

  3301.     avg_pixels_tab[3] = just_return;

  3302. 

  3303.     avg_no_rnd_pixels_tab[0] = just_return;

  3304.     avg_no_rnd_pixels_tab[1] = just_return;

  3305.     avg_no_rnd_pixels_tab[2] = just_return;

  3306.     avg_no_rnd_pixels_tab[3] = just_return;

  3307. 

  3308.     //av_fdct = just_return;

  3309.     //ff_idct = just_return;

  3310. #endif

  3311. }