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.
1651 Commits
32 Branches
850MB
Tree: a33c7159e1
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a33c7159e1'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx.c

1708 lines
66KB
Raw Blame History 

  1. /*

  2.  * MMX optimized DSP utils

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

  4.  *

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

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

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

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

  9.  *

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

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

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

  13.  * Lesser General Public License for more details.

  14.  *

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

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

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

  18.  *

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

  20.  */

  21. 

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

  23. 

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

  25. /* FIXME use them in static form */

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

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

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

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

  30. 

  31. int pix_abs16x16_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  32. int pix_abs16x16_x2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  33. int pix_abs16x16_y2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  34. int pix_abs16x16_xy2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  35. 

  36. int pix_abs8x8_mmx(UINT8 *blk1, UINT8 *blk2, int lx);

  37. int pix_abs8x8_x2_mmx(UINT8 *blk1, UINT8 *blk2, int lx);

  38. int pix_abs8x8_y2_mmx(UINT8 *blk1, UINT8 *blk2, int lx);

  39. int pix_abs8x8_xy2_mmx(UINT8 *blk1, UINT8 *blk2, int lx);

  40. 

  41. int pix_abs8x8_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  42. int pix_abs8x8_x2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  43. int pix_abs8x8_y2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  44. int pix_abs8x8_xy2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);

  45. 

  46. int sad16x16_mmx(void *s, UINT8 *blk1, UINT8 *blk2, int lx);

  47. int sad8x8_mmx(void *s, UINT8 *blk1, UINT8 *blk2, int lx);

  48. int sad16x16_mmx2(void *s, UINT8 *blk1, UINT8 *blk2, int lx);

  49. int sad8x8_mmx2(void *s, UINT8 *blk1, UINT8 *blk2, int lx);

  50. 

  51. /* pixel operations */

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

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

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

  55. 

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

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

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

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

  60. 

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

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

  63. 

  64. #define MOVQ_WONE(regd) \

  65.     __asm __volatile ( \

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

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

  68. 

  69. #define MOVQ_BFE(regd) \

  70.     __asm __volatile ( \

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

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

  73. 

  74. #ifndef PIC

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

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

  77. #else

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

  79. // pcmpeqd -> -1

  80. #define MOVQ_BONE(regd) \

  81.     __asm __volatile ( \

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

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

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

  85. 

  86. #define MOVQ_WTWO(regd) \

  87.     __asm __volatile ( \

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

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

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

  91. 

  92. #endif

  93. 

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

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

  96. // regfe is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

  104. 

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

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

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

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

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

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

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

  112. 

  113. // mm6 is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

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

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

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

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

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

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

  127. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  141. 

  142. /***********************************/

  143. /* MMX no rounding */

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

  145. #define SET_RND  MOVQ_WONE

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

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

  148. 

  149. #include "dsputil_mmx_rnd.h"

  150. 

  151. #undef DEF

  152. #undef SET_RND

  153. #undef PAVGBP

  154. #undef PAVGB

  155. /***********************************/

  156. /* MMX rounding */

  157. 

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

  159. #define SET_RND  MOVQ_WTWO

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

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

  162. 

  163. #include "dsputil_mmx_rnd.h"

  164. 

  165. #undef DEF

  166. #undef SET_RND

  167. #undef PAVGBP

  168. #undef PAVGB

  169. 

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

  171. /* 3Dnow specific */

  172. 

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

  174. /* for Athlons PAVGUSB is prefered */

  175. #define PAVGB "pavgusb"

  176. 

  177. #include "dsputil_mmx_avg.h"

  178. 

  179. #undef DEF

  180. #undef PAVGB

  181. 

  182. /***********************************/

  183. /* MMX2 specific */

  184. 

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

  186. 

  187. /* Introduced only in MMX2 set */

  188. #define PAVGB "pavgb"

  189. 

  190. #include "dsputil_mmx_avg.h"

  191. 

  192. #undef DEF

  193. #undef PAVGB

  194. 

  195. /***********************************/

  196. /* standard MMX */

  197. 

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

  199. {

  200.     asm volatile(

  201.         "movl $-128, %%eax	\n\t"

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

  203.         ".balign 16		\n\t"

  204.         "1:			\n\t"

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

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

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

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

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

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

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

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

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

  214.         "movq %%mm1, 8(%1, %%eax)\n\t"

  215.         "movq %%mm2, 16(%1, %%eax)\n\t"

  216.         "movq %%mm3, 24(%1, %%eax)\n\t"

  217.         "addl %3, %0		\n\t"

  218.         "addl $32, %%eax	\n\t"

  219.         "js 1b			\n\t"

  220.         : "+r" (pixels)

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

  222.         : "%eax"

  223.     );

  224. }

  225. 

  226. static inline void diff_pixels_mmx(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride)

  227. {

  228.     asm volatile(

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

  230.         "movl $-128, %%eax	\n\t"

  231.         ".balign 16		\n\t"

  232.         "1:			\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

  245.         "addl %3, %0		\n\t"

  246.         "addl %3, %1		\n\t"

  247.         "addl $16, %%eax	\n\t"

  248.         "jnz 1b			\n\t"

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

  250.         : "r" (block+64), "r" (stride)

  251.         : "%eax"

  252.     );

  253. }

  254. 

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

  256. {

  257.     const DCTELEM *p;

  258.     UINT8 *pix;

  259. 

  260.     /* read the pixels */

  261.     p = block;

  262.     pix = pixels;

  263.     /* unrolled loop */

  264. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  282. 		:"memory");

  283.         pix += line_size*4;

  284.         p += 32;

  285. 

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

  287.     // compiler would generate some very strange code

  288.     // thus using "r"

  289.     __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  307. 	    :"memory");

  308. }

  309. 

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

  311. {

  312.     const DCTELEM *p;

  313.     UINT8 *pix;

  314.     int i;

  315. 

  316.     /* read the pixels */

  317.     p = block;

  318.     pix = pixels;

  319.     MOVQ_ZERO(mm7);

  320.     i = 4;

  321.     do {

  322. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  344. 		:"r"(p)

  345. 		:"memory");

  346.         pix += line_size*2;

  347.         p += 16;

  348.     } while (--i);

  349. }

  350. 

  351. static void put_pixels8_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  352. {

  353.     __asm __volatile(

  354. 	 "lea (%3, %3), %%eax		\n\t"

  355. 	 ".balign 8			\n\t"

  356. 	 "1:				\n\t"

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

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

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

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

  361. 	 "addl %%eax, %1		\n\t"

  362.          "addl %%eax, %2       		\n\t"

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

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

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

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

  367. 	 "addl %%eax, %1		\n\t"

  368. 	 "addl %%eax, %2       		\n\t"

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

  370. 	 "jnz 1b			\n\t"

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

  372. 	 : "r"(line_size)

  373. 	 : "%eax", "memory"

  374. 	);

  375. }

  376. 

  377. static void put_pixels16_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  378. {

  379.     __asm __volatile(

  380. 	 "lea (%3, %3), %%eax		\n\t"

  381. 	 ".balign 8			\n\t"

  382. 	 "1:				\n\t"

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

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

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

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

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

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

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

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

  391. 	 "addl %%eax, %1		\n\t"

  392.          "addl %%eax, %2       		\n\t"

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

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

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

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

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

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

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

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

  401. 	 "addl %%eax, %1		\n\t"

  402. 	 "addl %%eax, %2       		\n\t"

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

  404. 	 "jnz 1b			\n\t"

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

  406. 	 : "r"(line_size)

  407. 	 : "%eax", "memory"

  408. 	);

  409. }

  410. 

  411. static void clear_blocks_mmx(DCTELEM *blocks)

  412. {

  413.     __asm __volatile(

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

  415.                 "movl $-128*6, %%eax		\n\t"

  416.                 "1:				\n\t"

  417.                 "movq %%mm7, (%0, %%eax)	\n\t"

  418.                 "movq %%mm7, 8(%0, %%eax)	\n\t"

  419.                 "movq %%mm7, 16(%0, %%eax)	\n\t"

  420.                 "movq %%mm7, 24(%0, %%eax)	\n\t"

  421.                 "addl $32, %%eax		\n\t"

  422.                 " js 1b				\n\t"

  423.                 : : "r" (((int)blocks)+128*6)

  424.                 : "%eax"

  425.         );

  426. }

  427. 

  428. static int pix_sum16_mmx(UINT8 * pix, int line_size){

  429.     const int h=16;

  430.     int sum;

  431.     int index= -line_size*h;

  432. 

  433.     __asm __volatile(

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

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

  436.                 "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

  449.                 "addl %3, %1			\n\t"

  450.                 " js 1b				\n\t"

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

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

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

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

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

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

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

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

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

  460.                 : "r" (pix - index), "r" (line_size)

  461.         );

  462. 

  463.         return sum;

  464. }

  465. 

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

  467.     int i=0;

  468.     asm volatile(

  469.         "1:				\n\t"

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

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

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

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

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

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

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

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

  478.         "addl $16, %0			\n\t"

  479.         "cmpl %3, %0			\n\t"

  480.         " jb 1b				\n\t"

  481.         : "+r" (i)

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

  483.     );

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

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

  486. }

  487. 

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

  489.     int tmp;

  490.   asm volatile (

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

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

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

  494.       "1:\n"

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

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

  497. 

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

  499. 

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

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

  502. 

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

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

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

  506. 

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

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

  509. 

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

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

  512. 

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

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

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

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

  517. 

  518.       "addl %2, %0\n"

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

  520.       "dec %%ecx\n"

  521.       "jnz 1b\n"

  522. 

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

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

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

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

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

  528.     return tmp;

  529. }

  530. 

  531. static int sse16_mmx(void *v, UINT8 * pix1, UINT8 * pix2, int line_size) {

  532.     int tmp;

  533.   asm volatile (

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

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

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

  537.       "1:\n"

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

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

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

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

  542. 

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

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

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

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

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

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

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

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

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

  552. 

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

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

  555. 

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

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

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

  559. 

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

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

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

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

  564. 

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

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

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

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

  569. 

  570.       "addl %3,%0\n"

  571.       "addl %3,%1\n"

  572. 

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

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

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

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

  577. 

  578.       "decl %%ecx\n"

  579.       "jnz 1b\n"

  580. 

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

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

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

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

  585.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) : "r" (line_size) : "ecx");

  586.     return tmp;

  587. }

  588. 

  589. static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){

  590.     int i=0;

  591.     asm volatile(

  592.         "1:				\n\t"

  593.         "movq  (%2, %0), %%mm0		\n\t"

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

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

  596.         "movq %%mm1, (%3, %0)		\n\t"

  597.         "movq 8(%2, %0), %%mm0		\n\t"

  598.         "movq 8(%1, %0), %%mm1		\n\t"

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

  600.         "movq %%mm1, 8(%3, %0)		\n\t"

  601.         "addl $16, %0			\n\t"

  602.         "cmpl %4, %0			\n\t"

  603.         " jb 1b				\n\t"

  604.         : "+r" (i)

  605.         : "r"(src1), "r"(src2), "r"(dst), "r"(w-15)

  606.     );

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

  608.         dst[i+0] = src1[i+0]-src2[i+0];

  609. }

  610. #define LBUTTERFLY(a,b)\

  611.     "paddw " #b ", " #a "		\n\t"\

  612.     "paddw " #b ", " #b "		\n\t"\

  613.     "psubw " #a ", " #b "		\n\t"

  614. 

  615. #define HADAMARD48\

  616.         LBUTTERFLY(%%mm0, %%mm1)\

  617.         LBUTTERFLY(%%mm2, %%mm3)\

  618.         LBUTTERFLY(%%mm4, %%mm5)\

  619.         LBUTTERFLY(%%mm6, %%mm7)\

  620.         \

  621.         LBUTTERFLY(%%mm0, %%mm2)\

  622.         LBUTTERFLY(%%mm1, %%mm3)\

  623.         LBUTTERFLY(%%mm4, %%mm6)\

  624.         LBUTTERFLY(%%mm5, %%mm7)\

  625.         \

  626.         LBUTTERFLY(%%mm0, %%mm4)\

  627.         LBUTTERFLY(%%mm1, %%mm5)\

  628.         LBUTTERFLY(%%mm2, %%mm6)\

  629.         LBUTTERFLY(%%mm3, %%mm7)

  630. 

  631. #define MMABS(a,z)\

  632.     "pxor " #z ", " #z "		\n\t"\

  633.     "pcmpgtw " #a ", " #z "		\n\t"\

  634.     "pxor " #z ", " #a "		\n\t"\

  635.     "psubw " #z ", " #a "		\n\t"

  636. 

  637. #define MMABS_SUM(a,z, sum)\

  638.     "pxor " #z ", " #z "		\n\t"\

  639.     "pcmpgtw " #a ", " #z "		\n\t"\

  640.     "pxor " #z ", " #a "		\n\t"\

  641.     "psubw " #z ", " #a "		\n\t"\

  642.     "paddusw " #a ", " #sum "		\n\t"

  643. 

  644.     

  645. #define SBUTTERFLY(a,b,t,n)\

  646.     "movq " #a ", " #t "		\n\t" /* abcd */\

  647.     "punpckl" #n " " #b ", " #a "	\n\t" /* aebf */\

  648.     "punpckh" #n " " #b ", " #t "	\n\t" /* cgdh */\

  649.     

  650. #define TRANSPOSE4(a,b,c,d,t)\

  651.     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\

  652.     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\

  653.     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\

  654.     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

  655. 

  656. #define LOAD4(o, a, b, c, d)\

  657.         "movq "#o"(%1), " #a "		\n\t"\

  658.         "movq "#o"+16(%1), " #b "	\n\t"\

  659.         "movq "#o"+32(%1), " #c "	\n\t"\

  660.         "movq "#o"+48(%1), " #d "	\n\t"

  661. 

  662. #define STORE4(o, a, b, c, d)\

  663.         "movq "#a", "#o"(%1)		\n\t"\

  664.         "movq "#b", "#o"+16(%1)		\n\t"\

  665.         "movq "#c", "#o"+32(%1)		\n\t"\

  666.         "movq "#d", "#o"+48(%1)		\n\t"\

  667. 

  668. static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride){

  669.     uint64_t temp[16] __align8;

  670.     int sum=0;

  671. 

  672.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  673. 

  674.     asm volatile(

  675.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  676.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  677.         

  678.         HADAMARD48

  679.         

  680.         "movq %%mm7, 112(%1)		\n\t"

  681.         

  682.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  683.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  684.         

  685.         "movq 112(%1), %%mm7 		\n\t"

  686.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  687.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  688. 

  689.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  690.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  691.         

  692.         HADAMARD48

  693.         

  694.         "movq %%mm7, 120(%1)		\n\t"

  695.         

  696.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  697.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  698.         

  699.         "movq 120(%1), %%mm7 		\n\t"

  700.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  701.         "movq %%mm7, %%mm5		\n\t"//FIXME remove

  702.         "movq %%mm6, %%mm7		\n\t"

  703.         "movq %%mm0, %%mm6		\n\t"

  704. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  705.         

  706.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  707. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  708.         

  709.         HADAMARD48

  710.         "movq %%mm7, 64(%1)		\n\t"

  711.         MMABS(%%mm0, %%mm7)

  712.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  713.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  714.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  715.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  716.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  717.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  718.         "movq 64(%1), %%mm1		\n\t"

  719.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  720.         "movq %%mm0, 64(%1)		\n\t"

  721.         

  722.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  723.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  724.         

  725.         HADAMARD48

  726.         "movq %%mm7, (%1)		\n\t"

  727.         MMABS(%%mm0, %%mm7)

  728.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  729.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  730.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  731.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  732.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  733.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

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

  735.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  736.         "movq 64(%1), %%mm1		\n\t"

  737.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  738.         

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

  740.         "psrlq $32, %%mm0		\n\t"

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

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

  743.         "psrlq $16, %%mm0		\n\t"

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

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

  746.                 

  747.         : "=r" (sum)

  748.         : "r"(temp)

  749.     );

  750.     return sum&0xFFFF;

  751. }

  752. 

  753. WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx)

  754. 

  755. #define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)

  756. #define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

  757. 

  758. #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

  759.         "paddw " #m4 ", " #m3 "		\n\t" /* x1 */\

  760.         "movq "MANGLE(ff_pw_20)", %%mm4		\n\t" /* 20 */\

  761.         "pmullw " #m3 ", %%mm4		\n\t" /* 20x1 */\

  762.         "movq "#in7", " #m3 "		\n\t" /* d */\

  763.         "movq "#in0", %%mm5		\n\t" /* D */\

  764.         "paddw " #m3 ", %%mm5		\n\t" /* x4 */\

  765.         "psubw %%mm5, %%mm4		\n\t" /* 20x1 - x4 */\

  766.         "movq "#in1", %%mm5		\n\t" /* C */\

  767.         "movq "#in2", %%mm6		\n\t" /* B */\

  768.         "paddw " #m6 ", %%mm5		\n\t" /* x3 */\

  769.         "paddw " #m5 ", %%mm6		\n\t" /* x2 */\

  770.         "paddw %%mm6, %%mm6		\n\t" /* 2x2 */\

  771.         "psubw %%mm6, %%mm5		\n\t" /* -2x2 + x3 */\

  772.         "pmullw "MANGLE(ff_pw_3)", %%mm5	\n\t" /* -6x2 + 3x3 */\

  773.         "paddw " #rnd ", %%mm4		\n\t" /* x2 */\

  774.         "paddw %%mm4, %%mm5		\n\t" /* 20x1 - 6x2 + 3x3 - x4 */\

  775.         "psraw $5, %%mm5		\n\t"\

  776.         "packuswb %%mm5, %%mm5		\n\t"\

  777.         OP(%%mm5, out, %%mm7, d)

  778. 

  779. #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\

  780. void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  781.     uint64_t temp;\

  782. \

  783.     asm volatile(\

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

  785.         "1:				\n\t"\

  786.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  787.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  788.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  789.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  790.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  791.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  792.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  793.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  794.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  795.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  796.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  797.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  798.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  799.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  800.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  801.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  802.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  803.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  804.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  805.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  806.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  807.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  808.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  809.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  810.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  811.         "paddw %6, %%mm6		\n\t"\

  812.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  813.         "psraw $5, %%mm0		\n\t"\

  814.         "movq %%mm0, %5			\n\t"\

  815.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  816.         \

  817.         "movq 5(%0), %%mm0		\n\t" /* FGHIJKLM */\

  818.         "movq %%mm0, %%mm5		\n\t" /* FGHIJKLM */\

  819.         "movq %%mm0, %%mm6		\n\t" /* FGHIJKLM */\

  820.         "psrlq $8, %%mm0		\n\t" /* GHIJKLM0 */\

  821.         "psrlq $16, %%mm5		\n\t" /* HIJKLM00 */\

  822.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0G0H0I0J */\

  823.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0H0I0J0K */\

  824.         "paddw %%mm0, %%mm2		\n\t" /* b */\

  825.         "paddw %%mm5, %%mm3		\n\t" /* c */\

  826.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  827.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  828.         "movq %%mm6, %%mm2		\n\t" /* FGHIJKLM */\

  829.         "psrlq $24, %%mm6		\n\t" /* IJKLM000 */\

  830.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0F0G0H0I */\

  831.         "punpcklbw %%mm7, %%mm6		\n\t" /* 0I0J0K0L */\

  832.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  833.         "paddw %%mm2, %%mm1		\n\t" /* a */\

  834.         "paddw %%mm6, %%mm4		\n\t" /* d */\

  835.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  836.         "psubw %%mm4, %%mm3		\n\t" /* - 6b +3c - d */\

  837.         "paddw %6, %%mm1		\n\t"\

  838.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b +3c - d */\

  839.         "psraw $5, %%mm3		\n\t"\

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

  841.         "packuswb %%mm3, %%mm1		\n\t"\

  842.         OP_MMX2(%%mm1, (%1),%%mm4, q)\

  843.         /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\

  844.         \

  845.         "movq 9(%0), %%mm1		\n\t" /* JKLMNOPQ */\

  846.         "movq %%mm1, %%mm4		\n\t" /* JKLMNOPQ */\

  847.         "movq %%mm1, %%mm3		\n\t" /* JKLMNOPQ */\

  848.         "psrlq $8, %%mm1		\n\t" /* KLMNOPQ0 */\

  849.         "psrlq $16, %%mm4		\n\t" /* LMNOPQ00 */\

  850.         "punpcklbw %%mm7, %%mm1		\n\t" /* 0K0L0M0N */\

  851.         "punpcklbw %%mm7, %%mm4		\n\t" /* 0L0M0N0O */\

  852.         "paddw %%mm1, %%mm5		\n\t" /* b */\

  853.         "paddw %%mm4, %%mm0		\n\t" /* c */\

  854.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  855.         "psubw %%mm5, %%mm0		\n\t" /* c - 2b */\

  856.         "movq %%mm3, %%mm5		\n\t" /* JKLMNOPQ */\

  857.         "psrlq $24, %%mm3		\n\t" /* MNOPQ000 */\

  858.         "pmullw "MANGLE(ff_pw_3)", %%mm0		\n\t" /* 3c - 6b */\

  859.         "punpcklbw %%mm7, %%mm3		\n\t" /* 0M0N0O0P */\

  860.         "paddw %%mm3, %%mm2		\n\t" /* d */\

  861.         "psubw %%mm2, %%mm0		\n\t" /* -6b + 3c - d */\

  862.         "movq %%mm5, %%mm2		\n\t" /* JKLMNOPQ */\

  863.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0J0K0L0M */\

  864.         "punpckhbw %%mm7, %%mm5		\n\t" /* 0N0O0P0Q */\

  865.         "paddw %%mm2, %%mm6		\n\t" /* a */\

  866.         "pmullw "MANGLE(ff_pw_20)", %%mm6		\n\t" /* 20a */\

  867.         "paddw %6, %%mm0		\n\t"\

  868.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  869.         "psraw $5, %%mm0		\n\t"\

  870.         /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\

  871.         \

  872.         "paddw %%mm5, %%mm3		\n\t" /* a */\

  873.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0O0P0Q0Q */\

  874.         "paddw %%mm4, %%mm6		\n\t" /* b */\

  875.         "pshufw $0xBE, %%mm5, %%mm4	\n\t" /* 0P0Q0Q0P */\

  876.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0Q0Q0P0O */\

  877.         "paddw %%mm1, %%mm4		\n\t" /* c */\

  878.         "paddw %%mm2, %%mm5		\n\t" /* d */\

  879.         "paddw %%mm6, %%mm6		\n\t" /* 2b */\

  880.         "psubw %%mm6, %%mm4		\n\t" /* c - 2b */\

  881.         "pmullw "MANGLE(ff_pw_20)", %%mm3		\n\t" /* 20a */\

  882.         "pmullw "MANGLE(ff_pw_3)", %%mm4		\n\t" /* 3c - 6b */\

  883.         "psubw %%mm5, %%mm3		\n\t" /* -6b + 3c - d */\

  884.         "paddw %6, %%mm4		\n\t"\

  885.         "paddw %%mm3, %%mm4		\n\t" /* 20a - 6b + 3c - d */\

  886.         "psraw $5, %%mm4		\n\t"\

  887.         "packuswb %%mm4, %%mm0		\n\t"\

  888.         OP_MMX2(%%mm0, 8(%1), %%mm4, q)\

  889.         \

  890.         "addl %3, %0			\n\t"\

  891.         "addl %4, %1			\n\t"\

  892.         "decl %2			\n\t"\

  893.         " jnz 1b				\n\t"\

  894.         : "+a"(src), "+c"(dst), "+m"(h)\

  895.         : "d"(srcStride), "S"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  896.         : "memory"\

  897.     );\

  898. }\

  899. \

  900. static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  901.     int i;\

  902.     int16_t temp[16];\

  903.     /* quick HACK, XXX FIXME MUST be optimized */\

  904.     for(i=0; i<h; i++)\

  905.     {\

  906.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  907.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  908.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  909.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  910.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  911.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\

  912.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\

  913.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\

  914.         temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\

  915.         temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\

  916.         temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\

  917.         temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\

  918.         temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\

  919.         temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\

  920.         temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\

  921.         temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\

  922.         asm volatile(\

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

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

  925.             "paddw %2, %%mm0		\n\t"\

  926.             "paddw %2, %%mm1		\n\t"\

  927.             "psraw $5, %%mm0		\n\t"\

  928.             "psraw $5, %%mm1		\n\t"\

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

  930.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  931.             "movq 16(%0), %%mm0		\n\t"\

  932.             "movq 24(%0), %%mm1		\n\t"\

  933.             "paddw %2, %%mm0		\n\t"\

  934.             "paddw %2, %%mm1		\n\t"\

  935.             "psraw $5, %%mm0		\n\t"\

  936.             "psraw $5, %%mm1		\n\t"\

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

  938.             OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\

  939.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  940.             : "memory"\

  941.         );\

  942.         dst+=dstStride;\

  943.         src+=srcStride;\

  944.     }\

  945. }\

  946. \

  947. void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  948.     uint64_t temp;\

  949. \

  950.     asm volatile(\

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

  952.         "1:				\n\t"\

  953.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  954.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  955.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  956.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  957.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  958.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  959.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  960.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  961.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  962.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  963.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  964.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  965.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  966.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  967.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  968.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  969.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  970.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  971.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  972.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  973.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  974.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  975.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  976.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  977.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  978.         "paddw %6, %%mm6		\n\t"\

  979.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  980.         "psraw $5, %%mm0		\n\t"\

  981.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  982.         \

  983.         "movd 5(%0), %%mm5		\n\t" /* FGHI */\

  984.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0F0G0H0I */\

  985.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0G0H0I0I */\

  986.         "paddw %%mm5, %%mm1		\n\t" /* a */\

  987.         "paddw %%mm6, %%mm2		\n\t" /* b */\

  988.         "pshufw $0xBE, %%mm5, %%mm6	\n\t" /* 0H0I0I0H */\

  989.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0I0I0H0G */\

  990.         "paddw %%mm6, %%mm3		\n\t" /* c */\

  991.         "paddw %%mm5, %%mm4		\n\t" /* d */\

  992.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  993.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  994.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  995.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  996.         "psubw %%mm4, %%mm3		\n\t" /* -6b + 3c - d */\

  997.         "paddw %6, %%mm1		\n\t"\

  998.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b + 3c - d */\

  999.         "psraw $5, %%mm3		\n\t"\

  1000.         "packuswb %%mm3, %%mm0		\n\t"\

  1001.         OP_MMX2(%%mm0, (%1), %%mm4, q)\

  1002.         \

  1003.         "addl %3, %0			\n\t"\

  1004.         "addl %4, %1			\n\t"\

  1005.         "decl %2			\n\t"\

  1006.         " jnz 1b			\n\t"\

  1007.         : "+a"(src), "+c"(dst), "+m"(h)\

  1008.         : "S"(srcStride), "D"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1009.         : "memory"\

  1010.     );\

  1011. }\

  1012. \

  1013. static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1014.     int i;\

  1015.     int16_t temp[8];\

  1016.     /* quick HACK, XXX FIXME MUST be optimized */\

  1017.     for(i=0; i<h; i++)\

  1018.     {\

  1019.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1020.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1021.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1022.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1023.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1024.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\

  1025.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\

  1026.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\

  1027.         asm volatile(\

  1028.             "movq (%0), %%mm0		\n\t"\

  1029.             "movq 8(%0), %%mm1		\n\t"\

  1030.             "paddw %2, %%mm0		\n\t"\

  1031.             "paddw %2, %%mm1		\n\t"\

  1032.             "psraw $5, %%mm0		\n\t"\

  1033.             "psraw $5, %%mm1		\n\t"\

  1034.             "packuswb %%mm1, %%mm0	\n\t"\

  1035.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1036.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1037.             :"memory"\

  1038.         );\

  1039.         dst+=dstStride;\

  1040.         src+=srcStride;\

  1041.     }\

  1042. }

  1043. 

  1044. #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\

  1045. \

  1046. static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1047.     uint64_t temp[17*4];\

  1048.     uint64_t *temp_ptr= temp;\

  1049.     int count= 17;\

  1050. \

  1051.     /*FIXME unroll */\

  1052.     asm volatile(\

  1053.         "pxor %%mm7, %%mm7		\n\t"\

  1054.         "1:				\n\t"\

  1055.         "movq (%0), %%mm0		\n\t"\

  1056.         "movq (%0), %%mm1		\n\t"\

  1057.         "movq 8(%0), %%mm2		\n\t"\

  1058.         "movq 8(%0), %%mm3		\n\t"\

  1059.         "punpcklbw %%mm7, %%mm0		\n\t"\

  1060.         "punpckhbw %%mm7, %%mm1		\n\t"\

  1061.         "punpcklbw %%mm7, %%mm2		\n\t"\

  1062.         "punpckhbw %%mm7, %%mm3		\n\t"\

  1063.         "movq %%mm0, (%1)		\n\t"\

  1064.         "movq %%mm1, 17*8(%1)		\n\t"\

  1065.         "movq %%mm2, 2*17*8(%1)		\n\t"\

  1066.         "movq %%mm3, 3*17*8(%1)		\n\t"\

  1067.         "addl $8, %1			\n\t"\

  1068.         "addl %3, %0			\n\t"\

  1069.         "decl %2			\n\t"\

  1070.         " jnz 1b			\n\t"\

  1071.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  1072.         : "r" (srcStride)\

  1073.         : "memory"\

  1074.     );\

  1075.     \

  1076.     temp_ptr= temp;\

  1077.     count=4;\

  1078.     \

  1079. /*FIXME reorder for speed */\

  1080.     asm volatile(\

  1081.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  1082.         "1:				\n\t"\

  1083.         "movq (%0), %%mm0		\n\t"\

  1084.         "movq 8(%0), %%mm1		\n\t"\

  1085.         "movq 16(%0), %%mm2		\n\t"\

  1086.         "movq 24(%0), %%mm3		\n\t"\

  1087.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  1088.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  1089.         "addl %4, %1			\n\t"\

  1090.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  1091.         \

  1092.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  1093.         "addl %4, %1			\n\t"\

  1094.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  1095.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\

  1096.         "addl %4, %1			\n\t"\

  1097.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\

  1098.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\

  1099.         "addl %4, %1			\n\t"\

  1100.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\

  1101.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\

  1102.         "addl %4, %1			\n\t"\

  1103.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\

  1104.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\

  1105.         "addl %4, %1			\n\t"\

  1106.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\

  1107.         \

  1108.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\

  1109.         "addl %4, %1			\n\t"  \

  1110.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\

  1111.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\

  1112.         \

  1113.         "addl $136, %0			\n\t"\

  1114.         "addl %6, %1			\n\t"\

  1115.         "decl %2			\n\t"\

  1116.         " jnz 1b			\n\t"\

  1117.         \

  1118.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  1119.         : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*dstStride)\

  1120.         :"memory"\

  1121.     );\

  1122. }\

  1123. \

  1124. void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1125.     uint64_t temp[9*4];\

  1126.     uint64_t *temp_ptr= temp;\

  1127.     int count= 9;\

  1128. \

  1129.     /*FIXME unroll */\

  1130.     asm volatile(\

  1131.         "pxor %%mm7, %%mm7		\n\t"\

  1132.         "1:				\n\t"\

  1133.         "movq (%0), %%mm0		\n\t"\

  1134.         "movq (%0), %%mm1		\n\t"\

  1135.         "punpcklbw %%mm7, %%mm0		\n\t"\

  1136.         "punpckhbw %%mm7, %%mm1		\n\t"\

  1137.         "movq %%mm0, (%1)		\n\t"\

  1138.         "movq %%mm1, 9*8(%1)		\n\t"\

  1139.         "addl $8, %1			\n\t"\

  1140.         "addl %3, %0			\n\t"\

  1141.         "decl %2			\n\t"\

  1142.         " jnz 1b			\n\t"\

  1143.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  1144.         : "r" (srcStride)\

  1145.         : "memory"\

  1146.     );\

  1147.     \

  1148.     temp_ptr= temp;\

  1149.     count=2;\

  1150.     \

  1151. /*FIXME reorder for speed */\

  1152.     asm volatile(\

  1153.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  1154.         "1:				\n\t"\

  1155.         "movq (%0), %%mm0		\n\t"\

  1156.         "movq 8(%0), %%mm1		\n\t"\

  1157.         "movq 16(%0), %%mm2		\n\t"\

  1158.         "movq 24(%0), %%mm3		\n\t"\

  1159.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  1160.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  1161.         "addl %4, %1			\n\t"\

  1162.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  1163.         \

  1164.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  1165.         "addl %4, %1			\n\t"\

  1166.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  1167.         \

  1168.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\

  1169.         "addl %4, %1			\n\t"\

  1170.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\

  1171.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\

  1172.                 \

  1173.         "addl $72, %0			\n\t"\

  1174.         "addl %6, %1			\n\t"\

  1175.         "decl %2			\n\t"\

  1176.         " jnz 1b			\n\t"\

  1177.          \

  1178.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  1179.         : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*dstStride)\

  1180.         : "memory"\

  1181.    );\

  1182. }\

  1183. \

  1184. static void OPNAME ## qpel8_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\

  1185.     OPNAME ## pixels8_mmx(dst, src, stride, 8);\

  1186. }\

  1187. \

  1188. static void OPNAME ## qpel8_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1189.     uint64_t temp[8];\

  1190.     uint8_t * const half= (uint8_t*)temp;\

  1191.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  1192.     OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\

  1193. }\

  1194. \

  1195. static void OPNAME ## qpel8_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1196.     OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\

  1197. }\

  1198. \

  1199. static void OPNAME ## qpel8_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1200.     uint64_t temp[8];\

  1201.     uint8_t * const half= (uint8_t*)temp;\

  1202.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  1203.     OPNAME ## pixels8_l2_mmx(dst, src+1, half, stride, stride, 8);\

  1204. }\

  1205. \

  1206. static void OPNAME ## qpel8_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1207.     uint64_t temp[8];\

  1208.     uint8_t * const half= (uint8_t*)temp;\

  1209.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  1210.     OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\

  1211. }\

  1212. \

  1213. static void OPNAME ## qpel8_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1214.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\

  1215. }\

  1216. \

  1217. static void OPNAME ## qpel8_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1218.     uint64_t temp[8];\

  1219.     uint8_t * const half= (uint8_t*)temp;\

  1220.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  1221.     OPNAME ## pixels8_l2_mmx(dst, src+stride, half, stride, stride, 8);\

  1222. }\

  1223. static void OPNAME ## qpel8_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1224.     uint64_t half[8 + 9];\

  1225.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1226.     uint8_t * const halfHV= ((uint8_t*)half);\

  1227.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1228.     put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\

  1229.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1230.     OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\

  1231. }\

  1232. static void OPNAME ## qpel8_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1233.     uint64_t half[8 + 9];\

  1234.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1235.     uint8_t * const halfHV= ((uint8_t*)half);\

  1236.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1237.     put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\

  1238.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1239.     OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\

  1240. }\

  1241. static void OPNAME ## qpel8_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1242.     uint64_t half[8 + 9];\

  1243.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1244.     uint8_t * const halfHV= ((uint8_t*)half);\

  1245.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1246.     put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\

  1247.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1248.     OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\

  1249. }\

  1250. static void OPNAME ## qpel8_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1251.     uint64_t half[8 + 9];\

  1252.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1253.     uint8_t * const halfHV= ((uint8_t*)half);\

  1254.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1255.     put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\

  1256.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1257.     OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\

  1258. }\

  1259. static void OPNAME ## qpel8_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1260.     uint64_t half[8 + 9];\

  1261.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1262.     uint8_t * const halfHV= ((uint8_t*)half);\

  1263.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1264.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1265.     OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\

  1266. }\

  1267. static void OPNAME ## qpel8_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1268.     uint64_t half[8 + 9];\

  1269.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1270.     uint8_t * const halfHV= ((uint8_t*)half);\

  1271.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1272.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1273.     OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\

  1274. }\

  1275. static void OPNAME ## qpel8_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1276.     uint64_t half[8 + 9];\

  1277.     uint8_t * const halfH= ((uint8_t*)half);\

  1278.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1279.     put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\

  1280.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  1281. }\

  1282. static void OPNAME ## qpel8_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1283.     uint64_t half[8 + 9];\

  1284.     uint8_t * const halfH= ((uint8_t*)half);\

  1285.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1286.     put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\

  1287.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  1288. }\

  1289. static void OPNAME ## qpel8_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1290.     uint64_t half[9];\

  1291.     uint8_t * const halfH= ((uint8_t*)half);\

  1292.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1293.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  1294. }\

  1295. static void OPNAME ## qpel16_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\

  1296.     OPNAME ## pixels16_mmx(dst, src, stride, 16);\

  1297. }\

  1298. \

  1299. static void OPNAME ## qpel16_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1300.     uint64_t temp[32];\

  1301.     uint8_t * const half= (uint8_t*)temp;\

  1302.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  1303.     OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\

  1304. }\

  1305. \

  1306. static void OPNAME ## qpel16_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1307.     OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\

  1308. }\

  1309. \

  1310. static void OPNAME ## qpel16_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1311.     uint64_t temp[32];\

  1312.     uint8_t * const half= (uint8_t*)temp;\

  1313.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  1314.     OPNAME ## pixels16_l2_mmx(dst, src+1, half, stride, stride, 16);\

  1315. }\

  1316. \

  1317. static void OPNAME ## qpel16_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1318.     uint64_t temp[32];\

  1319.     uint8_t * const half= (uint8_t*)temp;\

  1320.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  1321.     OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\

  1322. }\

  1323. \

  1324. static void OPNAME ## qpel16_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1325.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\

  1326. }\

  1327. \

  1328. static void OPNAME ## qpel16_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1329.     uint64_t temp[32];\

  1330.     uint8_t * const half= (uint8_t*)temp;\

  1331.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  1332.     OPNAME ## pixels16_l2_mmx(dst, src+stride, half, stride, stride, 16);\

  1333. }\

  1334. static void OPNAME ## qpel16_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1335.     uint64_t half[16*2 + 17*2];\

  1336.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1337.     uint8_t * const halfHV= ((uint8_t*)half);\

  1338.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1339.     put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\

  1340.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1341.     OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\

  1342. }\

  1343. static void OPNAME ## qpel16_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1344.     uint64_t half[16*2 + 17*2];\

  1345.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1346.     uint8_t * const halfHV= ((uint8_t*)half);\

  1347.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1348.     put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\

  1349.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1350.     OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\

  1351. }\

  1352. static void OPNAME ## qpel16_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1353.     uint64_t half[16*2 + 17*2];\

  1354.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1355.     uint8_t * const halfHV= ((uint8_t*)half);\

  1356.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1357.     put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\

  1358.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1359.     OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\

  1360. }\

  1361. static void OPNAME ## qpel16_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1362.     uint64_t half[16*2 + 17*2];\

  1363.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1364.     uint8_t * const halfHV= ((uint8_t*)half);\

  1365.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1366.     put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\

  1367.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1368.     OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\

  1369. }\

  1370. static void OPNAME ## qpel16_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1371.     uint64_t half[16*2 + 17*2];\

  1372.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1373.     uint8_t * const halfHV= ((uint8_t*)half);\

  1374.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1375.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1376.     OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\

  1377. }\

  1378. static void OPNAME ## qpel16_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1379.     uint64_t half[16*2 + 17*2];\

  1380.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1381.     uint8_t * const halfHV= ((uint8_t*)half);\

  1382.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1383.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1384.     OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\

  1385. }\

  1386. static void OPNAME ## qpel16_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1387.     uint64_t half[17*2];\

  1388.     uint8_t * const halfH= ((uint8_t*)half);\

  1389.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1390.     put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\

  1391.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  1392. }\

  1393. static void OPNAME ## qpel16_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1394.     uint64_t half[17*2];\

  1395.     uint8_t * const halfH= ((uint8_t*)half);\

  1396.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1397.     put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\

  1398.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  1399. }\

  1400. static void OPNAME ## qpel16_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\

  1401.     uint64_t half[17*2];\

  1402.     uint8_t * const halfH= ((uint8_t*)half);\

  1403.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1404.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  1405. }

  1406. 

  1407. 

  1408. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "	\n\t"

  1409. #define AVG_3DNOW_OP(a,b,temp, size) \

  1410. "mov" #size " " #b ", " #temp "	\n\t"\

  1411. "pavgusb " #temp ", " #a "	\n\t"\

  1412. "mov" #size " " #a ", " #b "	\n\t"

  1413. #define AVG_MMX2_OP(a,b,temp, size) \

  1414. "mov" #size " " #b ", " #temp "	\n\t"\

  1415. "pavgb " #temp ", " #a "	\n\t"\

  1416. "mov" #size " " #a ", " #b "	\n\t"

  1417. 

  1418. QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)

  1419. QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)

  1420. QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)

  1421. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)

  1422. QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)

  1423. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)

  1424. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)

  1425. QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)

  1426. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

  1427. 

  1428. #if 0

  1429. static void just_return() { return; }

  1430. #endif

  1431. 

  1432. #define SET_QPEL_FUNC(postfix1, postfix2) \

  1433.     c->put_ ## postfix1 = put_ ## postfix2;\

  1434.     c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\

  1435.     c->avg_ ## postfix1 = avg_ ## postfix2;

  1436.     

  1437. void dsputil_init_mmx(DSPContext* c, unsigned mask)

  1438. {

  1439.     mm_flags = mm_support();

  1440. #if 0

  1441.     fprintf(stderr, "libavcodec: CPU flags:");

  1442.     if (mm_flags & MM_MMX)

  1443.         fprintf(stderr, " mmx");

  1444.     if (mm_flags & MM_MMXEXT)

  1445.         fprintf(stderr, " mmxext");

  1446.     if (mm_flags & MM_3DNOW)

  1447.         fprintf(stderr, " 3dnow");

  1448.     if (mm_flags & MM_SSE)

  1449.         fprintf(stderr, " sse");

  1450.     if (mm_flags & MM_SSE2)

  1451.         fprintf(stderr, " sse2");

  1452.     fprintf(stderr, "\n");

  1453. #endif

  1454. 

  1455.     if (mm_flags & MM_MMX) {

  1456.         c->get_pixels = get_pixels_mmx;

  1457.         c->diff_pixels = diff_pixels_mmx;

  1458.         c->put_pixels_clamped = put_pixels_clamped_mmx;

  1459.         c->add_pixels_clamped = add_pixels_clamped_mmx;

  1460.         c->clear_blocks = clear_blocks_mmx;

  1461.         c->pix_sum = pix_sum16_mmx;

  1462. 

  1463.         c->pix_abs16x16     = pix_abs16x16_mmx;

  1464.         c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx;

  1465.         c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx;

  1466.         c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;

  1467.         c->pix_abs8x8     = pix_abs8x8_mmx;

  1468.         c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx;

  1469.         c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx;

  1470.         c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx;

  1471. 

  1472.         c->put_pixels_tab[0][0] = put_pixels16_mmx;

  1473.         c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;

  1474.         c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;

  1475.         c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

  1476. 

  1477.         c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;

  1478.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;

  1479.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;

  1480.         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

  1481. 

  1482.         c->avg_pixels_tab[0][0] = avg_pixels16_mmx;

  1483.         c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;

  1484.         c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;

  1485.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

  1486. 

  1487.         c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;

  1488.         c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;

  1489.         c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;

  1490.         c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

  1491. 

  1492.         c->put_pixels_tab[1][0] = put_pixels8_mmx;

  1493.         c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;

  1494.         c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;

  1495.         c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

  1496. 

  1497.         c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;

  1498.         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;

  1499.         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;

  1500.         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

  1501. 

  1502.         c->avg_pixels_tab[1][0] = avg_pixels8_mmx;

  1503.         c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;

  1504.         c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;

  1505.         c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

  1506. 

  1507.         c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;

  1508.         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;

  1509.         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;

  1510.         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;

  1511.                 

  1512.         c->add_bytes= add_bytes_mmx;

  1513.         c->diff_bytes= diff_bytes_mmx;

  1514.         

  1515.         c->hadamard8_diff[0]= hadamard8_diff16_mmx;

  1516.         c->hadamard8_diff[1]= hadamard8_diff_mmx;

  1517.         

  1518.         c->sad[0]= sad16x16_mmx;

  1519.         c->sad[1]= sad8x8_mmx;

  1520. 

  1521. 	c->pix_norm1 = pix_norm1_mmx;

  1522. 	c->sse[0] = sse16_mmx;

  1523.         

  1524.         if (mm_flags & MM_MMXEXT) {

  1525.             c->pix_abs16x16     = pix_abs16x16_mmx2;

  1526.             c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx2;

  1527.             c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx2;

  1528.             c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx2;

  1529. 

  1530.             c->pix_abs8x8     = pix_abs8x8_mmx2;

  1531.             c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx2;

  1532.             c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx2;

  1533.             c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx2;

  1534. 

  1535.             c->sad[0]= sad16x16_mmx2;

  1536.             c->sad[1]= sad8x8_mmx2;

  1537.             

  1538.             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;

  1539.             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

  1540.             c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;

  1541.             c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

  1542. 

  1543.             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;

  1544.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;

  1545.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

  1546.             c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

  1547. 

  1548.             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;

  1549.             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

  1550.             c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

  1551.             c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

  1552. 

  1553.             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;

  1554.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;

  1555.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

  1556.             c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;

  1557. 

  1558. #if 1

  1559.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)

  1560.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)

  1561.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)

  1562.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)

  1563.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)

  1564.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)

  1565.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)

  1566.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)

  1567.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)

  1568.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)

  1569.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)

  1570.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)

  1571.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)

  1572.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)

  1573.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)

  1574.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)

  1575.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)

  1576.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)

  1577.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)

  1578.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)

  1579.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)

  1580.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)

  1581.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)

  1582.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)

  1583.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)

  1584.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)

  1585.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)

  1586.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)

  1587.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)

  1588.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)

  1589.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)

  1590.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)

  1591. #endif

  1592.         } else if (mm_flags & MM_3DNOW) {

  1593.             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;

  1594.             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  1595.             c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  1596.             c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  1597. 

  1598.             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  1599.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  1600.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  1601.             c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  1602. 

  1603.             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;

  1604.             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

  1605.             c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;

  1606.             c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

  1607. 

  1608.             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;

  1609.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;

  1610.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

  1611.             c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;

  1612. 

  1613.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)

  1614.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)

  1615.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)

  1616.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)

  1617.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)

  1618.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)

  1619.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)

  1620.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)

  1621.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)

  1622.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)

  1623.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)

  1624.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)

  1625.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)

  1626.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)

  1627.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)

  1628.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)

  1629.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)

  1630.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)

  1631.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)

  1632.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)

  1633.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)

  1634.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)

  1635.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)

  1636.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)

  1637.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)

  1638.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)

  1639.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)

  1640.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)

  1641.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)

  1642.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)

  1643.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)

  1644.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)

  1645.         }

  1646.     }

  1647. 

  1648. #if 0

  1649.     // for speed testing

  1650.     get_pixels = just_return;

  1651.     put_pixels_clamped = just_return;

  1652.     add_pixels_clamped = just_return;

  1653. 

  1654.     pix_abs16x16 = just_return;

  1655.     pix_abs16x16_x2 = just_return;

  1656.     pix_abs16x16_y2 = just_return;

  1657.     pix_abs16x16_xy2 = just_return;

  1658. 

  1659.     put_pixels_tab[0] = just_return;

  1660.     put_pixels_tab[1] = just_return;

  1661.     put_pixels_tab[2] = just_return;

  1662.     put_pixels_tab[3] = just_return;

  1663. 

  1664.     put_no_rnd_pixels_tab[0] = just_return;

  1665.     put_no_rnd_pixels_tab[1] = just_return;

  1666.     put_no_rnd_pixels_tab[2] = just_return;

  1667.     put_no_rnd_pixels_tab[3] = just_return;

  1668. 

  1669.     avg_pixels_tab[0] = just_return;

  1670.     avg_pixels_tab[1] = just_return;

  1671.     avg_pixels_tab[2] = just_return;

  1672.     avg_pixels_tab[3] = just_return;

  1673. 

  1674.     avg_no_rnd_pixels_tab[0] = just_return;

  1675.     avg_no_rnd_pixels_tab[1] = just_return;

  1676.     avg_no_rnd_pixels_tab[2] = just_return;

  1677.     avg_no_rnd_pixels_tab[3] = just_return;

  1678. 

  1679.     //av_fdct = just_return;

  1680.     //ff_idct = just_return;

  1681. #endif

  1682. }

  1683. 

  1684. /* remove any non bit exact operation (testing purpose). NOTE that

  1685.    this function should be kept as small as possible because it is

  1686.    always difficult to test automatically non bit exact cases. */

  1687. void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask)

  1688. {

  1689.     if (mm_flags & MM_MMX) {

  1690.         /* MMX2 & 3DNOW */

  1691.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;

  1692.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;

  1693.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

  1694.         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;

  1695.         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;

  1696.         c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

  1697. 

  1698.         if (mm_flags & MM_MMXEXT) {

  1699.             c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx;

  1700.             c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx;

  1701.             c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;

  1702.             c->pix_abs8x8_x2 = pix_abs8x8_x2_mmx;

  1703.             c->pix_abs8x8_y2 = pix_abs8x8_y2_mmx;

  1704.             c->pix_abs8x8_xy2= pix_abs8x8_xy2_mmx;

  1705.         }

  1706.     }

  1707. }