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.
3514 Commits
32 Branches
289MB
Tree: 0dfd33c3f1
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '0dfd33c3f1'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx.c

2724 lines
94KB
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_16 attribute_used __attribute__ ((aligned(8))) = 0x0010001000100010ULL;

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

  44. 

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

  46. 

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

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

  49. 

  50. #define MOVQ_WONE(regd) \

  51.     __asm __volatile ( \

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

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

  54. 

  55. #define MOVQ_BFE(regd) \

  56.     __asm __volatile ( \

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

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

  59. 

  60. #ifndef PIC

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

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

  63. #else

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

  65. // pcmpeqd -> -1

  66. #define MOVQ_BONE(regd) \

  67.     __asm __volatile ( \

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

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

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

  71. 

  72. #define MOVQ_WTWO(regd) \

  73.     __asm __volatile ( \

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

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

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

  77. 

  78. #endif

  79. 

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

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

  82. // regfe is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

  90. 

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

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

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

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

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

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

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

  98. 

  99. // mm6 is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

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

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

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

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

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

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

  113. 

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

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

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

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

  118.     "por  " #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, " #regd "	\n\t"\

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

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

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

  127. 

  128. /***********************************/

  129. /* MMX no rounding */

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

  131. #define SET_RND  MOVQ_WONE

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

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

  134. 

  135. #include "dsputil_mmx_rnd.h"

  136. 

  137. #undef DEF

  138. #undef SET_RND

  139. #undef PAVGBP

  140. #undef PAVGB

  141. /***********************************/

  142. /* MMX rounding */

  143. 

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

  145. #define SET_RND  MOVQ_WTWO

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

  147. #define PAVGB(a, b, c, e)		PAVGB_MMX(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. /***********************************/

  157. /* 3Dnow specific */

  158. 

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

  160. /* for Athlons PAVGUSB is prefered */

  161. #define PAVGB "pavgusb"

  162. 

  163. #include "dsputil_mmx_avg.h"

  164. 

  165. #undef DEF

  166. #undef PAVGB

  167. 

  168. /***********************************/

  169. /* MMX2 specific */

  170. 

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

  172. 

  173. /* Introduced only in MMX2 set */

  174. #define PAVGB "pavgb"

  175. 

  176. #include "dsputil_mmx_avg.h"

  177. 

  178. #undef DEF

  179. #undef PAVGB

  180. 

  181. /***********************************/

  182. /* standard MMX */

  183. 

  184. #ifdef CONFIG_ENCODERS

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

  186. {

  187.     asm volatile(

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

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

  190.         ".balign 16		\n\t"

  191.         "1:			\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  206.         "js 1b			\n\t"

  207.         : "+r" (pixels)

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

  209.         : "%eax"

  210.     );

  211. }

  212. 

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

  214. {

  215.     asm volatile(

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

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

  218.         ".balign 16		\n\t"

  219.         "1:			\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  235.         "jnz 1b			\n\t"

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

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

  238.         : "%eax"

  239.     );

  240. }

  241. #endif //CONFIG_ENCODERS

  242. 

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

  244. {

  245.     const DCTELEM *p;

  246.     uint8_t *pix;

  247. 

  248.     /* read the pixels */

  249.     p = block;

  250.     pix = pixels;

  251.     /* unrolled loop */

  252. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  270. 		:"memory");

  271.         pix += line_size*4;

  272.         p += 32;

  273. 

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

  275.     // compiler would generate some very strange code

  276.     // thus using "r"

  277.     __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  295. 	    :"memory");

  296. }

  297. 

  298. static unsigned char __align8 vector128[8] =

  299.   { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };

  300. 

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

  302. {

  303.     int i;

  304. 

  305.     movq_m2r(*vector128, mm1);

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

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

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

  309.         block += 8;

  310.         paddb_r2r(mm1, mm0);

  311.         movq_r2m(mm0, *pixels);

  312.         pixels += line_size;

  313.     }

  314. }

  315. 

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

  317. {

  318.     const DCTELEM *p;

  319.     uint8_t *pix;

  320.     int i;

  321. 

  322.     /* read the pixels */

  323.     p = block;

  324.     pix = pixels;

  325.     MOVQ_ZERO(mm7);

  326.     i = 4;

  327.     do {

  328. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  350. 		:"r"(p)

  351. 		:"memory");

  352.         pix += line_size*2;

  353.         p += 16;

  354.     } while (--i);

  355. }

  356. 

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

  358. {

  359.     __asm __volatile(

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

  361. 	 ".balign 8			\n\t"

  362. 	 "1:				\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. 	 "movq (%1), %%mm0		\n\t"

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

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

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

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

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

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

  376. 	 "jnz 1b			\n\t"

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

  378. 	 : "r"(line_size)

  379. 	 : "%eax", "memory"

  380. 	);

  381. }

  382. 

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

  384. {

  385.     __asm __volatile(

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

  387. 	 ".balign 8			\n\t"

  388. 	 "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  410. 	 "jnz 1b			\n\t"

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

  412. 	 : "r"(line_size)

  413. 	 : "%eax", "memory"

  414. 	);

  415. }

  416. 

  417. static void clear_blocks_mmx(DCTELEM *blocks)

  418. {

  419.     __asm __volatile(

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

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

  422.                 "1:				\n\t"

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

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

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

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

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

  428.                 " js 1b				\n\t"

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

  430.                 : "%eax"

  431.         );

  432. }

  433. 

  434. #ifdef CONFIG_ENCODERS

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

  436.     const int h=16;

  437.     int sum;

  438.     int index= -line_size*h;

  439. 

  440.     __asm __volatile(

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

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

  443.                 "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  457.                 " js 1b				\n\t"

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

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

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

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

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

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

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

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

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

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

  468.         );

  469. 

  470.         return sum;

  471. }

  472. #endif //CONFIG_ENCODERS

  473. 

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

  475.     int i=0;

  476.     asm volatile(

  477.         "1:				\n\t"

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

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

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

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

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

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

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

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

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

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

  488.         " jb 1b				\n\t"

  489.         : "+r" (i)

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

  491.     );

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

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

  494. }

  495. 

  496. #define H263_LOOP_FILTER \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  566. 

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

  568.     const int strength= ff_h263_loop_filter_strength[qscale];

  569. 

  570.     asm volatile(

  571.     

  572.         H263_LOOP_FILTER

  573.         

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

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

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

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

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

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

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

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

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

  583.     );

  584. }

  585. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  603.         

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

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

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

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

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

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

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

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

  612.     );

  613. }

  614. 

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

  616.     const int strength= ff_h263_loop_filter_strength[qscale];

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

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

  619.     

  620.     src -= 2;

  621. 

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

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

  624.     asm volatile(

  625.         H263_LOOP_FILTER // 5 3 4 6

  626.         

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

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

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

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

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

  632.     );

  633. 

  634.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

  647.         "movd %%mm5, %0			\n\t"

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

  649.         "movd %%mm5, %1			\n\t"

  650.         "movd %%mm3, %2			\n\t"

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

  652.         "movd %%mm3, %3			\n\t"

  653.         "movd %%mm1, %4			\n\t"

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

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

  656.         "movd %%mm6, %6			\n\t"

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

  658.         "movd %%mm6, %7			\n\t"

  659.         : "=m" (*(uint32_t*)(src + 0*stride)),

  660.           "=m" (*(uint32_t*)(src + 1*stride)),

  661.           "=m" (*(uint32_t*)(src + 2*stride)),

  662.           "=m" (*(uint32_t*)(src + 3*stride)),

  663.           "=m" (*(uint32_t*)(src + 4*stride)),

  664.           "=m" (*(uint32_t*)(src + 5*stride)),

  665.           "=m" (*(uint32_t*)(src + 6*stride)),

  666.           "=m" (*(uint32_t*)(src + 7*stride))

  667.     );

  668. }

  669. 

  670. #ifdef CONFIG_ENCODERS

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

  672.     int tmp;

  673.   asm volatile (

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

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

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

  677.       "1:\n"

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

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

  680. 

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

  682. 

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

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

  685. 

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

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

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

  689. 

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

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

  692. 

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

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

  695. 

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

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

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

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

  700. 

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

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

  703.       "dec %%ecx\n"

  704.       "jnz 1b\n"

  705. 

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

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

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

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

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

  711.     return tmp;

  712. }

  713. 

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

  715.     int tmp;

  716.   asm volatile (

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

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

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

  720.       "1:\n"

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

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

  723. 

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

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

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

  727. 

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

  729. 

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

  731. 

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

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

  734. 

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

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

  737. 

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

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

  740. 

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

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

  743. 

  744.       "decl %%ecx\n"

  745.       "jnz 1b\n"

  746. 

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

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

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

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

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

  752.       : "r" (line_size) , "m" (h)

  753.       : "%ecx");

  754.     return tmp;

  755. }

  756. 

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

  758.     int tmp;

  759.   asm volatile (

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

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

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

  763.       "1:\n"

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

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

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

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

  768. 

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

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

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

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

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

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

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

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

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

  778. 

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

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

  781. 

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

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

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

  785. 

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

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

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

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

  790. 

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

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

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

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

  795. 

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

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

  798. 

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

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

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

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

  803. 

  804.       "decl %%ecx\n"

  805.       "jnz 1b\n"

  806. 

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

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

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

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

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

  812.       : "r" (line_size) , "m" (h)

  813.       : "%ecx");

  814.     return tmp;

  815. }

  816. 

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

  818.     int tmp;

  819.   asm volatile (

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

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

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

  823.       

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

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

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

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

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

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

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

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

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

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

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

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

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

  837.       

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

  839.       

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  865. 

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

  867.       "1:\n"

  868.   

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  894.       

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

  896.       

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  922. 

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

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

  925.       " jnz 1b\n"

  926. 

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

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

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

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

  931.       

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

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

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

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

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

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

  938.       : "%ecx");

  939.       return tmp;

  940. }

  941. 

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

  943.     int tmp;

  944.     uint8_t * pix= pix1;

  945.   asm volatile (

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

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

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

  949.       

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

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

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

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

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

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

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

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

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

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

  960.       

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

  962.       

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  985. 

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

  987.       "1:\n"

  988.   

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

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

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

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

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

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

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

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

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

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

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

  1000.       "psubw %%mm2, %%mm5\n"

  1001.       "pxor %%mm3, %%mm3\n"

  1002.       "pxor %%mm1, %%mm1\n"

  1003.       "pcmpgtw %%mm4, %%mm3\n\t"

  1004.       "pcmpgtw %%mm5, %%mm1\n\t"

  1005.       "pxor %%mm3, %%mm4\n"

  1006.       "pxor %%mm1, %%mm5\n"

  1007.       "psubw %%mm3, %%mm4\n"

  1008.       "psubw %%mm1, %%mm5\n"

  1009.       "paddw %%mm4, %%mm5\n"

  1010.       "paddw %%mm5, %%mm6\n"

  1011.       

  1012.       "addl %2,%0\n"

  1013.       

  1014.       "movq (%0),%%mm4\n"

  1015.       "movq 1(%0),%%mm1\n"

  1016.       "movq %%mm4, %%mm5\n"

  1017.       "movq %%mm1, %%mm3\n"

  1018.       "punpcklbw %%mm7,%%mm4\n"

  1019.       "punpcklbw %%mm7,%%mm1\n"

  1020.       "punpckhbw %%mm7,%%mm5\n"

  1021.       "punpckhbw %%mm7,%%mm3\n"

  1022.       "psubw %%mm1, %%mm4\n"

  1023.       "psubw %%mm3, %%mm5\n"

  1024.       "psubw %%mm4, %%mm0\n"

  1025.       "psubw %%mm5, %%mm2\n"

  1026.       "pxor %%mm3, %%mm3\n"

  1027.       "pxor %%mm1, %%mm1\n"

  1028.       "pcmpgtw %%mm0, %%mm3\n\t"

  1029.       "pcmpgtw %%mm2, %%mm1\n\t"

  1030.       "pxor %%mm3, %%mm0\n"

  1031.       "pxor %%mm1, %%mm2\n"

  1032.       "psubw %%mm3, %%mm0\n" 

  1033.       "psubw %%mm1, %%mm2\n"

  1034.       "paddw %%mm0, %%mm2\n"

  1035.       "paddw %%mm2, %%mm6\n"

  1036. 

  1037.       "addl %2,%0\n"

  1038.       "subl $2, %%ecx\n"

  1039.       " jnz 1b\n"

  1040. 

  1041.       "movq %%mm6, %%mm0\n"

  1042.       "punpcklwd %%mm7,%%mm0\n"

  1043.       "punpckhwd %%mm7,%%mm6\n"

  1044.       "paddd %%mm0, %%mm6\n"

  1045.       

  1046.       "movq %%mm6,%%mm0\n"

  1047.       "psrlq $32, %%mm6\n"

  1048.       "paddd %%mm6,%%mm0\n"

  1049.       "movd %%mm0,%1\n"

  1050.       : "+r" (pix1), "=r"(tmp) 

  1051.       : "r" (line_size) , "g" (h-2)

  1052.       : "%ecx");

  1053.       return tmp + hf_noise8_mmx(pix+8, line_size, h);

  1054. }

  1055. 

  1056. static int nsse16_mmx(MpegEncContext *c, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1057.     int score1= sse16_mmx(c, pix1, pix2, line_size, h);

  1058.     int score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

  1059. 

  1060.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1061.     else  return score1 + ABS(score2)*8;

  1062. }

  1063. 

  1064. static int nsse8_mmx(MpegEncContext *c, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1065.     int score1= sse8_mmx(c, pix1, pix2, line_size, h);

  1066.     int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

  1067. 

  1068.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1069.     else  return score1 + ABS(score2)*8;

  1070. }

  1071. 

  1072. static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1073.     int tmp;

  1074.     

  1075.     assert( (((int)pix) & 7) == 0);

  1076.     assert((line_size &7) ==0);

  1077.     

  1078. #define SUM(in0, in1, out0, out1) \

  1079.       "movq (%0), %%mm2\n"\

  1080.       "movq 8(%0), %%mm3\n"\

  1081.       "addl %2,%0\n"\

  1082.       "movq %%mm2, " #out0 "\n"\

  1083.       "movq %%mm3, " #out1 "\n"\

  1084.       "psubusb " #in0 ", %%mm2\n"\

  1085.       "psubusb " #in1 ", %%mm3\n"\

  1086.       "psubusb " #out0 ", " #in0 "\n"\

  1087.       "psubusb " #out1 ", " #in1 "\n"\

  1088.       "por %%mm2, " #in0 "\n"\

  1089.       "por %%mm3, " #in1 "\n"\

  1090.       "movq " #in0 ", %%mm2\n"\

  1091.       "movq " #in1 ", %%mm3\n"\

  1092.       "punpcklbw %%mm7, " #in0 "\n"\

  1093.       "punpcklbw %%mm7, " #in1 "\n"\

  1094.       "punpckhbw %%mm7, %%mm2\n"\

  1095.       "punpckhbw %%mm7, %%mm3\n"\

  1096.       "paddw " #in1 ", " #in0 "\n"\

  1097.       "paddw %%mm3, %%mm2\n"\

  1098.       "paddw %%mm2, " #in0 "\n"\

  1099.       "paddw " #in0 ", %%mm6\n"

  1100. 

  1101.     

  1102.   asm volatile (

  1103.       "movl %3,%%ecx\n"

  1104.       "pxor %%mm6,%%mm6\n"

  1105.       "pxor %%mm7,%%mm7\n"

  1106.       "movq (%0),%%mm0\n"

  1107.       "movq 8(%0),%%mm1\n"

  1108.       "addl %2,%0\n"

  1109.       "subl $2, %%ecx\n"

  1110.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1111.       "1:\n"

  1112.       

  1113.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1114.       

  1115.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1116.       

  1117.       "subl $2, %%ecx\n"

  1118.       "jnz 1b\n"

  1119. 

  1120.       "movq %%mm6,%%mm0\n"

  1121.       "psrlq $32, %%mm6\n"

  1122.       "paddw %%mm6,%%mm0\n"

  1123.       "movq %%mm0,%%mm6\n"

  1124.       "psrlq $16, %%mm0\n"

  1125.       "paddw %%mm6,%%mm0\n"

  1126.       "movd %%mm0,%1\n"

  1127.       : "+r" (pix), "=r"(tmp) 

  1128.       : "r" (line_size) , "m" (h)

  1129.       : "%ecx");

  1130.     return tmp & 0xFFFF;

  1131. }

  1132. #undef SUM

  1133. 

  1134. static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1135.     int tmp;

  1136.     

  1137.     assert( (((int)pix) & 7) == 0);

  1138.     assert((line_size &7) ==0);

  1139.     

  1140. #define SUM(in0, in1, out0, out1) \

  1141.       "movq (%0), " #out0 "\n"\

  1142.       "movq 8(%0), " #out1 "\n"\

  1143.       "addl %2,%0\n"\

  1144.       "psadbw " #out0 ", " #in0 "\n"\

  1145.       "psadbw " #out1 ", " #in1 "\n"\

  1146.       "paddw " #in1 ", " #in0 "\n"\

  1147.       "paddw " #in0 ", %%mm6\n"

  1148. 

  1149.   asm volatile (

  1150.       "movl %3,%%ecx\n"

  1151.       "pxor %%mm6,%%mm6\n"

  1152.       "pxor %%mm7,%%mm7\n"

  1153.       "movq (%0),%%mm0\n"

  1154.       "movq 8(%0),%%mm1\n"

  1155.       "addl %2,%0\n"

  1156.       "subl $2, %%ecx\n"

  1157.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1158.       "1:\n"

  1159.       

  1160.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1161.       

  1162.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1163.       

  1164.       "subl $2, %%ecx\n"

  1165.       "jnz 1b\n"

  1166. 

  1167.       "movd %%mm6,%1\n"

  1168.       : "+r" (pix), "=r"(tmp) 

  1169.       : "r" (line_size) , "m" (h)

  1170.       : "%ecx");

  1171.     return tmp;

  1172. }

  1173. #undef SUM

  1174. 

  1175. static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1176.     int tmp;

  1177.     

  1178.     assert( (((int)pix1) & 7) == 0);

  1179.     assert( (((int)pix2) & 7) == 0);

  1180.     assert((line_size &7) ==0);

  1181.     

  1182. #define SUM(in0, in1, out0, out1) \

  1183.       "movq (%0),%%mm2\n"\

  1184.       "movq (%1)," #out0 "\n"\

  1185.       "movq 8(%0),%%mm3\n"\

  1186.       "movq 8(%1)," #out1 "\n"\

  1187.       "addl %3,%0\n"\

  1188.       "addl %3,%1\n"\

  1189.       "psubb " #out0 ", %%mm2\n"\

  1190.       "psubb " #out1 ", %%mm3\n"\

  1191.       "pxor %%mm7, %%mm2\n"\

  1192.       "pxor %%mm7, %%mm3\n"\

  1193.       "movq %%mm2, " #out0 "\n"\

  1194.       "movq %%mm3, " #out1 "\n"\

  1195.       "psubusb " #in0 ", %%mm2\n"\

  1196.       "psubusb " #in1 ", %%mm3\n"\

  1197.       "psubusb " #out0 ", " #in0 "\n"\

  1198.       "psubusb " #out1 ", " #in1 "\n"\

  1199.       "por %%mm2, " #in0 "\n"\

  1200.       "por %%mm3, " #in1 "\n"\

  1201.       "movq " #in0 ", %%mm2\n"\

  1202.       "movq " #in1 ", %%mm3\n"\

  1203.       "punpcklbw %%mm7, " #in0 "\n"\

  1204.       "punpcklbw %%mm7, " #in1 "\n"\

  1205.       "punpckhbw %%mm7, %%mm2\n"\

  1206.       "punpckhbw %%mm7, %%mm3\n"\

  1207.       "paddw " #in1 ", " #in0 "\n"\

  1208.       "paddw %%mm3, %%mm2\n"\

  1209.       "paddw %%mm2, " #in0 "\n"\

  1210.       "paddw " #in0 ", %%mm6\n"

  1211. 

  1212.     

  1213.   asm volatile (

  1214.       "movl %4,%%ecx\n"

  1215.       "pxor %%mm6,%%mm6\n"

  1216.       "pcmpeqw %%mm7,%%mm7\n"

  1217.       "psllw $15, %%mm7\n"

  1218.       "packsswb %%mm7, %%mm7\n"

  1219.       "movq (%0),%%mm0\n"

  1220.       "movq (%1),%%mm2\n"

  1221.       "movq 8(%0),%%mm1\n"

  1222.       "movq 8(%1),%%mm3\n"

  1223.       "addl %3,%0\n"

  1224.       "addl %3,%1\n"

  1225.       "subl $2, %%ecx\n"

  1226.       "psubb %%mm2, %%mm0\n"

  1227.       "psubb %%mm3, %%mm1\n"

  1228.       "pxor %%mm7, %%mm0\n"

  1229.       "pxor %%mm7, %%mm1\n"

  1230.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1231.       "1:\n"

  1232.       

  1233.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1234.       

  1235.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1236.       

  1237.       "subl $2, %%ecx\n"

  1238.       "jnz 1b\n"

  1239. 

  1240.       "movq %%mm6,%%mm0\n"

  1241.       "psrlq $32, %%mm6\n"

  1242.       "paddw %%mm6,%%mm0\n"

  1243.       "movq %%mm0,%%mm6\n"

  1244.       "psrlq $16, %%mm0\n"

  1245.       "paddw %%mm6,%%mm0\n"

  1246.       "movd %%mm0,%2\n"

  1247.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  1248.       : "r" (line_size) , "m" (h)

  1249.       : "%ecx");

  1250.     return tmp & 0x7FFF;

  1251. }

  1252. #undef SUM

  1253. 

  1254. static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1255.     int tmp;

  1256.     

  1257.     assert( (((int)pix1) & 7) == 0);

  1258.     assert( (((int)pix2) & 7) == 0);

  1259.     assert((line_size &7) ==0);

  1260.     

  1261. #define SUM(in0, in1, out0, out1) \

  1262.       "movq (%0)," #out0 "\n"\

  1263.       "movq (%1),%%mm2\n"\

  1264.       "movq 8(%0)," #out1 "\n"\

  1265.       "movq 8(%1),%%mm3\n"\

  1266.       "addl %3,%0\n"\

  1267.       "addl %3,%1\n"\

  1268.       "psubb %%mm2, " #out0 "\n"\

  1269.       "psubb %%mm3, " #out1 "\n"\

  1270.       "pxor %%mm7, " #out0 "\n"\

  1271.       "pxor %%mm7, " #out1 "\n"\

  1272.       "psadbw " #out0 ", " #in0 "\n"\

  1273.       "psadbw " #out1 ", " #in1 "\n"\

  1274.       "paddw " #in1 ", " #in0 "\n"\

  1275.       "paddw " #in0 ", %%mm6\n"

  1276. 

  1277.   asm volatile (

  1278.       "movl %4,%%ecx\n"

  1279.       "pxor %%mm6,%%mm6\n"

  1280.       "pcmpeqw %%mm7,%%mm7\n"

  1281.       "psllw $15, %%mm7\n"

  1282.       "packsswb %%mm7, %%mm7\n"

  1283.       "movq (%0),%%mm0\n"

  1284.       "movq (%1),%%mm2\n"

  1285.       "movq 8(%0),%%mm1\n"

  1286.       "movq 8(%1),%%mm3\n"

  1287.       "addl %3,%0\n"

  1288.       "addl %3,%1\n"

  1289.       "subl $2, %%ecx\n"

  1290.       "psubb %%mm2, %%mm0\n"

  1291.       "psubb %%mm3, %%mm1\n"

  1292.       "pxor %%mm7, %%mm0\n"

  1293.       "pxor %%mm7, %%mm1\n"

  1294.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1295.       "1:\n"

  1296.       

  1297.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1298.       

  1299.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1300.       

  1301.       "subl $2, %%ecx\n"

  1302.       "jnz 1b\n"

  1303. 

  1304.       "movd %%mm6,%2\n"

  1305.       : "+r" (pix1), "+r" (pix2), "=r"(tmp) 

  1306.       : "r" (line_size) , "m" (h)

  1307.       : "%ecx");

  1308.     return tmp;

  1309. }

  1310. #undef SUM

  1311. 

  1312. static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){

  1313.     int i=0;

  1314.     asm volatile(

  1315.         "1:				\n\t"

  1316.         "movq  (%2, %0), %%mm0		\n\t"

  1317.         "movq  (%1, %0), %%mm1		\n\t"

  1318.         "psubb %%mm0, %%mm1		\n\t"

  1319.         "movq %%mm1, (%3, %0)		\n\t"

  1320.         "movq 8(%2, %0), %%mm0		\n\t"

  1321.         "movq 8(%1, %0), %%mm1		\n\t"

  1322.         "psubb %%mm0, %%mm1		\n\t"

  1323.         "movq %%mm1, 8(%3, %0)		\n\t"

  1324.         "addl $16, %0			\n\t"

  1325.         "cmpl %4, %0			\n\t"

  1326.         " jb 1b				\n\t"

  1327.         : "+r" (i)

  1328.         : "r"(src1), "r"(src2), "r"(dst), "r"(w-15)

  1329.     );

  1330.     for(; i<w; i++)

  1331.         dst[i+0] = src1[i+0]-src2[i+0];

  1332. }

  1333. 

  1334. static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){

  1335.     int i=0;

  1336.     uint8_t l, lt;

  1337. 

  1338.     asm volatile(

  1339.         "1:				\n\t"

  1340.         "movq  -1(%1, %0), %%mm0	\n\t" // LT

  1341.         "movq  (%1, %0), %%mm1		\n\t" // T

  1342.         "movq  -1(%2, %0), %%mm2	\n\t" // L

  1343.         "movq  (%2, %0), %%mm3		\n\t" // X

  1344.         "movq %%mm2, %%mm4		\n\t" // L

  1345.         "psubb %%mm0, %%mm2		\n\t"

  1346.         "paddb %%mm1, %%mm2		\n\t" // L + T - LT

  1347.         "movq %%mm4, %%mm5		\n\t" // L

  1348.         "pmaxub %%mm1, %%mm4		\n\t" // max(T, L)

  1349.         "pminub %%mm5, %%mm1		\n\t" // min(T, L)

  1350.         "pminub %%mm2, %%mm4		\n\t" 

  1351.         "pmaxub %%mm1, %%mm4		\n\t"

  1352.         "psubb %%mm4, %%mm3		\n\t" // dst - pred

  1353.         "movq %%mm3, (%3, %0)		\n\t"

  1354.         "addl $8, %0			\n\t"

  1355.         "cmpl %4, %0			\n\t"

  1356.         " jb 1b				\n\t"

  1357.         : "+r" (i)

  1358.         : "r"(src1), "r"(src2), "r"(dst), "r"(w)

  1359.     );

  1360. 

  1361.     l= *left;

  1362.     lt= *left_top;

  1363.     

  1364.     dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);

  1365.     

  1366.     *left_top= src1[w-1];

  1367.     *left    = src2[w-1];

  1368. }

  1369. 

  1370. #define LBUTTERFLY2(a1,b1,a2,b2)\

  1371.     "paddw " #b1 ", " #a1 "		\n\t"\

  1372.     "paddw " #b2 ", " #a2 "		\n\t"\

  1373.     "paddw " #b1 ", " #b1 "		\n\t"\

  1374.     "paddw " #b2 ", " #b2 "		\n\t"\

  1375.     "psubw " #a1 ", " #b1 "		\n\t"\

  1376.     "psubw " #a2 ", " #b2 "		\n\t"

  1377. 

  1378. #define HADAMARD48\

  1379.         LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\

  1380.         LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\

  1381.         LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\

  1382.         LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\

  1383.         LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\

  1384.         LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\

  1385. 

  1386. #define MMABS(a,z)\

  1387.     "pxor " #z ", " #z "		\n\t"\

  1388.     "pcmpgtw " #a ", " #z "		\n\t"\

  1389.     "pxor " #z ", " #a "		\n\t"\

  1390.     "psubw " #z ", " #a "		\n\t"

  1391. 

  1392. #define MMABS_SUM(a,z, sum)\

  1393.     "pxor " #z ", " #z "		\n\t"\

  1394.     "pcmpgtw " #a ", " #z "		\n\t"\

  1395.     "pxor " #z ", " #a "		\n\t"\

  1396.     "psubw " #z ", " #a "		\n\t"\

  1397.     "paddusw " #a ", " #sum "		\n\t"

  1398. 

  1399. #define MMABS_MMX2(a,z)\

  1400.     "pxor " #z ", " #z "		\n\t"\

  1401.     "psubw " #a ", " #z "		\n\t"\

  1402.     "pmaxsw " #z ", " #a "		\n\t"

  1403. 

  1404. #define MMABS_SUM_MMX2(a,z, sum)\

  1405.     "pxor " #z ", " #z "		\n\t"\

  1406.     "psubw " #a ", " #z "		\n\t"\

  1407.     "pmaxsw " #z ", " #a "		\n\t"\

  1408.     "paddusw " #a ", " #sum "		\n\t"

  1409.         

  1410. #define SBUTTERFLY(a,b,t,n)\

  1411.     "movq " #a ", " #t "		\n\t" /* abcd */\

  1412.     "punpckl" #n " " #b ", " #a "	\n\t" /* aebf */\

  1413.     "punpckh" #n " " #b ", " #t "	\n\t" /* cgdh */\

  1414. 

  1415. #define TRANSPOSE4(a,b,c,d,t)\

  1416.     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\

  1417.     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\

  1418.     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\

  1419.     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

  1420. 

  1421. #define LOAD4(o, a, b, c, d)\

  1422.         "movq "#o"(%1), " #a "		\n\t"\

  1423.         "movq "#o"+16(%1), " #b "	\n\t"\

  1424.         "movq "#o"+32(%1), " #c "	\n\t"\

  1425.         "movq "#o"+48(%1), " #d "	\n\t"

  1426. 

  1427. #define STORE4(o, a, b, c, d)\

  1428.         "movq "#a", "#o"(%1)		\n\t"\

  1429.         "movq "#b", "#o"+16(%1)		\n\t"\

  1430.         "movq "#c", "#o"+32(%1)		\n\t"\

  1431.         "movq "#d", "#o"+48(%1)		\n\t"\

  1432. 

  1433. static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1434.     uint64_t temp[16] __align8;

  1435.     int sum=0;

  1436.     

  1437.     assert(h==8);

  1438. 

  1439.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1440. 

  1441.     asm volatile(

  1442.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1443.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1444.         

  1445.         HADAMARD48

  1446.         

  1447.         "movq %%mm7, 112(%1)		\n\t"

  1448.         

  1449.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1450.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1451.         

  1452.         "movq 112(%1), %%mm7 		\n\t"

  1453.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1454.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1455. 

  1456.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1457.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1458.         

  1459.         HADAMARD48

  1460.         

  1461.         "movq %%mm7, 120(%1)		\n\t"

  1462.         

  1463.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1464.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1465.         

  1466.         "movq 120(%1), %%mm7 		\n\t"

  1467.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1468.         "movq %%mm7, %%mm5		\n\t"//FIXME remove

  1469.         "movq %%mm6, %%mm7		\n\t"

  1470.         "movq %%mm0, %%mm6		\n\t"

  1471. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1472.         

  1473.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1474. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1475.         

  1476.         HADAMARD48

  1477.         "movq %%mm7, 64(%1)		\n\t"

  1478.         MMABS(%%mm0, %%mm7)

  1479.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1480.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1481.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1482.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1483.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1484.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1485.         "movq 64(%1), %%mm1		\n\t"

  1486.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1487.         "movq %%mm0, 64(%1)		\n\t"

  1488.         

  1489.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1490.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1491.         

  1492.         HADAMARD48

  1493.         "movq %%mm7, (%1)		\n\t"

  1494.         MMABS(%%mm0, %%mm7)

  1495.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1496.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1497.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1498.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1499.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1500.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1501.         "movq (%1), %%mm1		\n\t"

  1502.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1503.         "movq 64(%1), %%mm1		\n\t"

  1504.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1505.         

  1506.         "movq %%mm0, %%mm1		\n\t"

  1507.         "psrlq $32, %%mm0		\n\t"

  1508.         "paddusw %%mm1, %%mm0		\n\t"

  1509.         "movq %%mm0, %%mm1		\n\t"

  1510.         "psrlq $16, %%mm0		\n\t"

  1511.         "paddusw %%mm1, %%mm0		\n\t"

  1512.         "movd %%mm0, %0			\n\t"

  1513.                 

  1514.         : "=r" (sum)

  1515.         : "r"(temp)

  1516.     );

  1517.     return sum&0xFFFF;

  1518. }

  1519. 

  1520. static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1521.     uint64_t temp[16] __align8;

  1522.     int sum=0;

  1523.     

  1524.     assert(h==8);

  1525. 

  1526.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1527. 

  1528.     asm volatile(

  1529.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1530.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1531.         

  1532.         HADAMARD48

  1533.         

  1534.         "movq %%mm7, 112(%1)		\n\t"

  1535.         

  1536.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1537.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1538.         

  1539.         "movq 112(%1), %%mm7 		\n\t"

  1540.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1541.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1542. 

  1543.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1544.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1545.         

  1546.         HADAMARD48

  1547.         

  1548.         "movq %%mm7, 120(%1)		\n\t"

  1549.         

  1550.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1551.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1552.         

  1553.         "movq 120(%1), %%mm7 		\n\t"

  1554.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1555.         "movq %%mm7, %%mm5		\n\t"//FIXME remove

  1556.         "movq %%mm6, %%mm7		\n\t"

  1557.         "movq %%mm0, %%mm6		\n\t"

  1558. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1559.         

  1560.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1561. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1562.         

  1563.         HADAMARD48

  1564.         "movq %%mm7, 64(%1)		\n\t"

  1565.         MMABS_MMX2(%%mm0, %%mm7)

  1566.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1567.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1568.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1569.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1570.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1571.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1572.         "movq 64(%1), %%mm1		\n\t"

  1573.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1574.         "movq %%mm0, 64(%1)		\n\t"

  1575.         

  1576.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1577.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1578.         

  1579.         HADAMARD48

  1580.         "movq %%mm7, (%1)		\n\t"

  1581.         MMABS_MMX2(%%mm0, %%mm7)

  1582.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1583.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1584.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1585.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1586.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1587.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1588.         "movq (%1), %%mm1		\n\t"

  1589.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1590.         "movq 64(%1), %%mm1		\n\t"

  1591.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1592.         

  1593.         "movq %%mm0, %%mm1		\n\t"

  1594.         "psrlq $32, %%mm0		\n\t"

  1595.         "paddusw %%mm1, %%mm0		\n\t"

  1596.         "movq %%mm0, %%mm1		\n\t"

  1597.         "psrlq $16, %%mm0		\n\t"

  1598.         "paddusw %%mm1, %%mm0		\n\t"

  1599.         "movd %%mm0, %0			\n\t"

  1600.                 

  1601.         : "=r" (sum)

  1602.         : "r"(temp)

  1603.     );

  1604.     return sum&0xFFFF;

  1605. }

  1606. 

  1607. 

  1608. WARPER8_16_SQ(hadamard8_diff_mmx, hadamard8_diff16_mmx)

  1609. WARPER8_16_SQ(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)

  1610. #endif //CONFIG_ENCODERS

  1611. 

  1612. #define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)

  1613. #define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

  1614. 

  1615. #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

  1616.         "paddw " #m4 ", " #m3 "		\n\t" /* x1 */\

  1617.         "movq "MANGLE(ff_pw_20)", %%mm4		\n\t" /* 20 */\

  1618.         "pmullw " #m3 ", %%mm4		\n\t" /* 20x1 */\

  1619.         "movq "#in7", " #m3 "		\n\t" /* d */\

  1620.         "movq "#in0", %%mm5		\n\t" /* D */\

  1621.         "paddw " #m3 ", %%mm5		\n\t" /* x4 */\

  1622.         "psubw %%mm5, %%mm4		\n\t" /* 20x1 - x4 */\

  1623.         "movq "#in1", %%mm5		\n\t" /* C */\

  1624.         "movq "#in2", %%mm6		\n\t" /* B */\

  1625.         "paddw " #m6 ", %%mm5		\n\t" /* x3 */\

  1626.         "paddw " #m5 ", %%mm6		\n\t" /* x2 */\

  1627.         "paddw %%mm6, %%mm6		\n\t" /* 2x2 */\

  1628.         "psubw %%mm6, %%mm5		\n\t" /* -2x2 + x3 */\

  1629.         "pmullw "MANGLE(ff_pw_3)", %%mm5	\n\t" /* -6x2 + 3x3 */\

  1630.         "paddw " #rnd ", %%mm4		\n\t" /* x2 */\

  1631.         "paddw %%mm4, %%mm5		\n\t" /* 20x1 - 6x2 + 3x3 - x4 */\

  1632.         "psraw $5, %%mm5		\n\t"\

  1633.         "packuswb %%mm5, %%mm5		\n\t"\

  1634.         OP(%%mm5, out, %%mm7, d)

  1635. 

  1636. #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\

  1637. static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1638.     uint64_t temp;\

  1639. \

  1640.     asm volatile(\

  1641.         "pxor %%mm7, %%mm7		\n\t"\

  1642.         "1:				\n\t"\

  1643.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  1644.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  1645.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  1646.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  1647.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  1648.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  1649.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  1650.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  1651.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  1652.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  1653.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  1654.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  1655.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  1656.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  1657.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  1658.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  1659.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  1660.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1661.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  1662.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  1663.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  1664.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  1665.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  1666.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  1667.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  1668.         "paddw %6, %%mm6		\n\t"\

  1669.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1670.         "psraw $5, %%mm0		\n\t"\

  1671.         "movq %%mm0, %5			\n\t"\

  1672.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1673.         \

  1674.         "movq 5(%0), %%mm0		\n\t" /* FGHIJKLM */\

  1675.         "movq %%mm0, %%mm5		\n\t" /* FGHIJKLM */\

  1676.         "movq %%mm0, %%mm6		\n\t" /* FGHIJKLM */\

  1677.         "psrlq $8, %%mm0		\n\t" /* GHIJKLM0 */\

  1678.         "psrlq $16, %%mm5		\n\t" /* HIJKLM00 */\

  1679.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0G0H0I0J */\

  1680.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0H0I0J0K */\

  1681.         "paddw %%mm0, %%mm2		\n\t" /* b */\

  1682.         "paddw %%mm5, %%mm3		\n\t" /* c */\

  1683.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  1684.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  1685.         "movq %%mm6, %%mm2		\n\t" /* FGHIJKLM */\

  1686.         "psrlq $24, %%mm6		\n\t" /* IJKLM000 */\

  1687.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0F0G0H0I */\

  1688.         "punpcklbw %%mm7, %%mm6		\n\t" /* 0I0J0K0L */\

  1689.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  1690.         "paddw %%mm2, %%mm1		\n\t" /* a */\

  1691.         "paddw %%mm6, %%mm4		\n\t" /* d */\

  1692.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  1693.         "psubw %%mm4, %%mm3		\n\t" /* - 6b +3c - d */\

  1694.         "paddw %6, %%mm1		\n\t"\

  1695.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b +3c - d */\

  1696.         "psraw $5, %%mm3		\n\t"\

  1697.         "movq %5, %%mm1			\n\t"\

  1698.         "packuswb %%mm3, %%mm1		\n\t"\

  1699.         OP_MMX2(%%mm1, (%1),%%mm4, q)\

  1700.         /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\

  1701.         \

  1702.         "movq 9(%0), %%mm1		\n\t" /* JKLMNOPQ */\

  1703.         "movq %%mm1, %%mm4		\n\t" /* JKLMNOPQ */\

  1704.         "movq %%mm1, %%mm3		\n\t" /* JKLMNOPQ */\

  1705.         "psrlq $8, %%mm1		\n\t" /* KLMNOPQ0 */\

  1706.         "psrlq $16, %%mm4		\n\t" /* LMNOPQ00 */\

  1707.         "punpcklbw %%mm7, %%mm1		\n\t" /* 0K0L0M0N */\

  1708.         "punpcklbw %%mm7, %%mm4		\n\t" /* 0L0M0N0O */\

  1709.         "paddw %%mm1, %%mm5		\n\t" /* b */\

  1710.         "paddw %%mm4, %%mm0		\n\t" /* c */\

  1711.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1712.         "psubw %%mm5, %%mm0		\n\t" /* c - 2b */\

  1713.         "movq %%mm3, %%mm5		\n\t" /* JKLMNOPQ */\

  1714.         "psrlq $24, %%mm3		\n\t" /* MNOPQ000 */\

  1715.         "pmullw "MANGLE(ff_pw_3)", %%mm0		\n\t" /* 3c - 6b */\

  1716.         "punpcklbw %%mm7, %%mm3		\n\t" /* 0M0N0O0P */\

  1717.         "paddw %%mm3, %%mm2		\n\t" /* d */\

  1718.         "psubw %%mm2, %%mm0		\n\t" /* -6b + 3c - d */\

  1719.         "movq %%mm5, %%mm2		\n\t" /* JKLMNOPQ */\

  1720.         "punpcklbw %%mm7, %%mm2		\n\t" /* 0J0K0L0M */\

  1721.         "punpckhbw %%mm7, %%mm5		\n\t" /* 0N0O0P0Q */\

  1722.         "paddw %%mm2, %%mm6		\n\t" /* a */\

  1723.         "pmullw "MANGLE(ff_pw_20)", %%mm6		\n\t" /* 20a */\

  1724.         "paddw %6, %%mm0		\n\t"\

  1725.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1726.         "psraw $5, %%mm0		\n\t"\

  1727.         /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\

  1728.         \

  1729.         "paddw %%mm5, %%mm3		\n\t" /* a */\

  1730.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0O0P0Q0Q */\

  1731.         "paddw %%mm4, %%mm6		\n\t" /* b */\

  1732.         "pshufw $0xBE, %%mm5, %%mm4	\n\t" /* 0P0Q0Q0P */\

  1733.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0Q0Q0P0O */\

  1734.         "paddw %%mm1, %%mm4		\n\t" /* c */\

  1735.         "paddw %%mm2, %%mm5		\n\t" /* d */\

  1736.         "paddw %%mm6, %%mm6		\n\t" /* 2b */\

  1737.         "psubw %%mm6, %%mm4		\n\t" /* c - 2b */\

  1738.         "pmullw "MANGLE(ff_pw_20)", %%mm3		\n\t" /* 20a */\

  1739.         "pmullw "MANGLE(ff_pw_3)", %%mm4		\n\t" /* 3c - 6b */\

  1740.         "psubw %%mm5, %%mm3		\n\t" /* -6b + 3c - d */\

  1741.         "paddw %6, %%mm4		\n\t"\

  1742.         "paddw %%mm3, %%mm4		\n\t" /* 20a - 6b + 3c - d */\

  1743.         "psraw $5, %%mm4		\n\t"\

  1744.         "packuswb %%mm4, %%mm0		\n\t"\

  1745.         OP_MMX2(%%mm0, 8(%1), %%mm4, q)\

  1746.         \

  1747.         "addl %3, %0			\n\t"\

  1748.         "addl %4, %1			\n\t"\

  1749.         "decl %2			\n\t"\

  1750.         " jnz 1b				\n\t"\

  1751.         : "+a"(src), "+c"(dst), "+m"(h)\

  1752.         : "d"(srcStride), "S"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1753.         : "memory"\

  1754.     );\

  1755. }\

  1756. \

  1757. static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1758.     int i;\

  1759.     int16_t temp[16];\

  1760.     /* quick HACK, XXX FIXME MUST be optimized */\

  1761.     for(i=0; i<h; i++)\

  1762.     {\

  1763.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1764.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1765.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1766.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1767.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1768.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\

  1769.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\

  1770.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\

  1771.         temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\

  1772.         temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\

  1773.         temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\

  1774.         temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\

  1775.         temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\

  1776.         temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\

  1777.         temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\

  1778.         temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\

  1779.         asm volatile(\

  1780.             "movq (%0), %%mm0		\n\t"\

  1781.             "movq 8(%0), %%mm1		\n\t"\

  1782.             "paddw %2, %%mm0		\n\t"\

  1783.             "paddw %2, %%mm1		\n\t"\

  1784.             "psraw $5, %%mm0		\n\t"\

  1785.             "psraw $5, %%mm1		\n\t"\

  1786.             "packuswb %%mm1, %%mm0	\n\t"\

  1787.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1788.             "movq 16(%0), %%mm0		\n\t"\

  1789.             "movq 24(%0), %%mm1		\n\t"\

  1790.             "paddw %2, %%mm0		\n\t"\

  1791.             "paddw %2, %%mm1		\n\t"\

  1792.             "psraw $5, %%mm0		\n\t"\

  1793.             "psraw $5, %%mm1		\n\t"\

  1794.             "packuswb %%mm1, %%mm0	\n\t"\

  1795.             OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\

  1796.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1797.             : "memory"\

  1798.         );\

  1799.         dst+=dstStride;\

  1800.         src+=srcStride;\

  1801.     }\

  1802. }\

  1803. \

  1804. static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1805.     uint64_t temp;\

  1806. \

  1807.     asm volatile(\

  1808.         "pxor %%mm7, %%mm7		\n\t"\

  1809.         "1:				\n\t"\

  1810.         "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\

  1811.         "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\

  1812.         "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\

  1813.         "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\

  1814.         "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\

  1815.         "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\

  1816.         "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\

  1817.         "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\

  1818.         "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\

  1819.         "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\

  1820.         "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\

  1821.         "psllq $24, %%mm4		\n\t" /* 000ABCDE */\

  1822.         "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\

  1823.         "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\

  1824.         "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\

  1825.         "paddw %%mm3, %%mm5		\n\t" /* b */\

  1826.         "paddw %%mm2, %%mm6		\n\t" /* c */\

  1827.         "paddw %%mm5, %%mm5		\n\t" /* 2b */\

  1828.         "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\

  1829.         "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\

  1830.         "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\

  1831.         "paddw %%mm4, %%mm0		\n\t" /* a */\

  1832.         "paddw %%mm1, %%mm5		\n\t" /* d */\

  1833.         "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\

  1834.         "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\

  1835.         "paddw %6, %%mm6		\n\t"\

  1836.         "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\

  1837.         "psraw $5, %%mm0		\n\t"\

  1838.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1839.         \

  1840.         "movd 5(%0), %%mm5		\n\t" /* FGHI */\

  1841.         "punpcklbw %%mm7, %%mm5		\n\t" /* 0F0G0H0I */\

  1842.         "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0G0H0I0I */\

  1843.         "paddw %%mm5, %%mm1		\n\t" /* a */\

  1844.         "paddw %%mm6, %%mm2		\n\t" /* b */\

  1845.         "pshufw $0xBE, %%mm5, %%mm6	\n\t" /* 0H0I0I0H */\

  1846.         "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0I0I0H0G */\

  1847.         "paddw %%mm6, %%mm3		\n\t" /* c */\

  1848.         "paddw %%mm5, %%mm4		\n\t" /* d */\

  1849.         "paddw %%mm2, %%mm2		\n\t" /* 2b */\

  1850.         "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\

  1851.         "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\

  1852.         "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\

  1853.         "psubw %%mm4, %%mm3		\n\t" /* -6b + 3c - d */\

  1854.         "paddw %6, %%mm1		\n\t"\

  1855.         "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b + 3c - d */\

  1856.         "psraw $5, %%mm3		\n\t"\

  1857.         "packuswb %%mm3, %%mm0		\n\t"\

  1858.         OP_MMX2(%%mm0, (%1), %%mm4, q)\

  1859.         \

  1860.         "addl %3, %0			\n\t"\

  1861.         "addl %4, %1			\n\t"\

  1862.         "decl %2			\n\t"\

  1863.         " jnz 1b			\n\t"\

  1864.         : "+a"(src), "+c"(dst), "+m"(h)\

  1865.         : "S"(srcStride), "D"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1866.         : "memory"\

  1867.     );\

  1868. }\

  1869. \

  1870. static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1871.     int i;\

  1872.     int16_t temp[8];\

  1873.     /* quick HACK, XXX FIXME MUST be optimized */\

  1874.     for(i=0; i<h; i++)\

  1875.     {\

  1876.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1877.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1878.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1879.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1880.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1881.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\

  1882.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\

  1883.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\

  1884.         asm volatile(\

  1885.             "movq (%0), %%mm0		\n\t"\

  1886.             "movq 8(%0), %%mm1		\n\t"\

  1887.             "paddw %2, %%mm0		\n\t"\

  1888.             "paddw %2, %%mm1		\n\t"\

  1889.             "psraw $5, %%mm0		\n\t"\

  1890.             "psraw $5, %%mm1		\n\t"\

  1891.             "packuswb %%mm1, %%mm0	\n\t"\

  1892.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1893.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1894.             :"memory"\

  1895.         );\

  1896.         dst+=dstStride;\

  1897.         src+=srcStride;\

  1898.     }\

  1899. }

  1900. 

  1901. #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\

  1902. \

  1903. static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1904.     uint64_t temp[17*4];\

  1905.     uint64_t *temp_ptr= temp;\

  1906.     int count= 17;\

  1907. \

  1908.     /*FIXME unroll */\

  1909.     asm volatile(\

  1910.         "pxor %%mm7, %%mm7		\n\t"\

  1911.         "1:				\n\t"\

  1912.         "movq (%0), %%mm0		\n\t"\

  1913.         "movq (%0), %%mm1		\n\t"\

  1914.         "movq 8(%0), %%mm2		\n\t"\

  1915.         "movq 8(%0), %%mm3		\n\t"\

  1916.         "punpcklbw %%mm7, %%mm0		\n\t"\

  1917.         "punpckhbw %%mm7, %%mm1		\n\t"\

  1918.         "punpcklbw %%mm7, %%mm2		\n\t"\

  1919.         "punpckhbw %%mm7, %%mm3		\n\t"\

  1920.         "movq %%mm0, (%1)		\n\t"\

  1921.         "movq %%mm1, 17*8(%1)		\n\t"\

  1922.         "movq %%mm2, 2*17*8(%1)		\n\t"\

  1923.         "movq %%mm3, 3*17*8(%1)		\n\t"\

  1924.         "addl $8, %1			\n\t"\

  1925.         "addl %3, %0			\n\t"\

  1926.         "decl %2			\n\t"\

  1927.         " jnz 1b			\n\t"\

  1928.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  1929.         : "r" (srcStride)\

  1930.         : "memory"\

  1931.     );\

  1932.     \

  1933.     temp_ptr= temp;\

  1934.     count=4;\

  1935.     \

  1936. /*FIXME reorder for speed */\

  1937.     asm volatile(\

  1938.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  1939.         "1:				\n\t"\

  1940.         "movq (%0), %%mm0		\n\t"\

  1941.         "movq 8(%0), %%mm1		\n\t"\

  1942.         "movq 16(%0), %%mm2		\n\t"\

  1943.         "movq 24(%0), %%mm3		\n\t"\

  1944.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  1945.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  1946.         "addl %4, %1			\n\t"\

  1947.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  1948.         \

  1949.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  1950.         "addl %4, %1			\n\t"\

  1951.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  1952.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\

  1953.         "addl %4, %1			\n\t"\

  1954.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\

  1955.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\

  1956.         "addl %4, %1			\n\t"\

  1957.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\

  1958.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\

  1959.         "addl %4, %1			\n\t"\

  1960.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\

  1961.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\

  1962.         "addl %4, %1			\n\t"\

  1963.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\

  1964.         \

  1965.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\

  1966.         "addl %4, %1			\n\t"  \

  1967.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\

  1968.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\

  1969.         \

  1970.         "addl $136, %0			\n\t"\

  1971.         "addl %6, %1			\n\t"\

  1972.         "decl %2			\n\t"\

  1973.         " jnz 1b			\n\t"\

  1974.         \

  1975.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  1976.         : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*dstStride)\

  1977.         :"memory"\

  1978.     );\

  1979. }\

  1980. \

  1981. static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1982.     uint64_t temp[9*4];\

  1983.     uint64_t *temp_ptr= temp;\

  1984.     int count= 9;\

  1985. \

  1986.     /*FIXME unroll */\

  1987.     asm volatile(\

  1988.         "pxor %%mm7, %%mm7		\n\t"\

  1989.         "1:				\n\t"\

  1990.         "movq (%0), %%mm0		\n\t"\

  1991.         "movq (%0), %%mm1		\n\t"\

  1992.         "punpcklbw %%mm7, %%mm0		\n\t"\

  1993.         "punpckhbw %%mm7, %%mm1		\n\t"\

  1994.         "movq %%mm0, (%1)		\n\t"\

  1995.         "movq %%mm1, 9*8(%1)		\n\t"\

  1996.         "addl $8, %1			\n\t"\

  1997.         "addl %3, %0			\n\t"\

  1998.         "decl %2			\n\t"\

  1999.         " jnz 1b			\n\t"\

  2000.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  2001.         : "r" (srcStride)\

  2002.         : "memory"\

  2003.     );\

  2004.     \

  2005.     temp_ptr= temp;\

  2006.     count=2;\

  2007.     \

  2008. /*FIXME reorder for speed */\

  2009.     asm volatile(\

  2010.         /*"pxor %%mm7, %%mm7		\n\t"*/\

  2011.         "1:				\n\t"\

  2012.         "movq (%0), %%mm0		\n\t"\

  2013.         "movq 8(%0), %%mm1		\n\t"\

  2014.         "movq 16(%0), %%mm2		\n\t"\

  2015.         "movq 24(%0), %%mm3		\n\t"\

  2016.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  2017.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  2018.         "addl %4, %1			\n\t"\

  2019.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  2020.         \

  2021.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  2022.         "addl %4, %1			\n\t"\

  2023.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  2024.         \

  2025.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\

  2026.         "addl %4, %1			\n\t"\

  2027.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\

  2028.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\

  2029.                 \

  2030.         "addl $72, %0			\n\t"\

  2031.         "addl %6, %1			\n\t"\

  2032.         "decl %2			\n\t"\

  2033.         " jnz 1b			\n\t"\

  2034.          \

  2035.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2036.         : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*dstStride)\

  2037.         : "memory"\

  2038.    );\

  2039. }\

  2040. \

  2041. static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2042.     OPNAME ## pixels8_mmx(dst, src, stride, 8);\

  2043. }\

  2044. \

  2045. static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2046.     uint64_t temp[8];\

  2047.     uint8_t * const half= (uint8_t*)temp;\

  2048.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2049.     OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\

  2050. }\

  2051. \

  2052. static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2053.     OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\

  2054. }\

  2055. \

  2056. static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2057.     uint64_t temp[8];\

  2058.     uint8_t * const half= (uint8_t*)temp;\

  2059.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2060.     OPNAME ## pixels8_l2_mmx(dst, src+1, half, stride, stride, 8);\

  2061. }\

  2062. \

  2063. static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2064.     uint64_t temp[8];\

  2065.     uint8_t * const half= (uint8_t*)temp;\

  2066.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2067.     OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\

  2068. }\

  2069. \

  2070. static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2071.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2072. }\

  2073. \

  2074. static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2075.     uint64_t temp[8];\

  2076.     uint8_t * const half= (uint8_t*)temp;\

  2077.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2078.     OPNAME ## pixels8_l2_mmx(dst, src+stride, half, stride, stride, 8);\

  2079. }\

  2080. static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2081.     uint64_t half[8 + 9];\

  2082.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2083.     uint8_t * const halfHV= ((uint8_t*)half);\

  2084.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2085.     put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\

  2086.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2087.     OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\

  2088. }\

  2089. static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2090.     uint64_t half[8 + 9];\

  2091.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2092.     uint8_t * const halfHV= ((uint8_t*)half);\

  2093.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2094.     put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\

  2095.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2096.     OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\

  2097. }\

  2098. static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2099.     uint64_t half[8 + 9];\

  2100.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2101.     uint8_t * const halfHV= ((uint8_t*)half);\

  2102.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2103.     put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\

  2104.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2105.     OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\

  2106. }\

  2107. static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2108.     uint64_t half[8 + 9];\

  2109.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2110.     uint8_t * const halfHV= ((uint8_t*)half);\

  2111.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2112.     put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\

  2113.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2114.     OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\

  2115. }\

  2116. static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2117.     uint64_t half[8 + 9];\

  2118.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2119.     uint8_t * const halfHV= ((uint8_t*)half);\

  2120.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2121.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2122.     OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\

  2123. }\

  2124. static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2125.     uint64_t half[8 + 9];\

  2126.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2127.     uint8_t * const halfHV= ((uint8_t*)half);\

  2128.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2129.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2130.     OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\

  2131. }\

  2132. static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2133.     uint64_t half[8 + 9];\

  2134.     uint8_t * const halfH= ((uint8_t*)half);\

  2135.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2136.     put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\

  2137.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2138. }\

  2139. static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2140.     uint64_t half[8 + 9];\

  2141.     uint8_t * const halfH= ((uint8_t*)half);\

  2142.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2143.     put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\

  2144.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2145. }\

  2146. static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2147.     uint64_t half[9];\

  2148.     uint8_t * const halfH= ((uint8_t*)half);\

  2149.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2150.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2151. }\

  2152. static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2153.     OPNAME ## pixels16_mmx(dst, src, stride, 16);\

  2154. }\

  2155. \

  2156. static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2157.     uint64_t temp[32];\

  2158.     uint8_t * const half= (uint8_t*)temp;\

  2159.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2160.     OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\

  2161. }\

  2162. \

  2163. static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2164.     OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\

  2165. }\

  2166. \

  2167. static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2168.     uint64_t temp[32];\

  2169.     uint8_t * const half= (uint8_t*)temp;\

  2170.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2171.     OPNAME ## pixels16_l2_mmx(dst, src+1, half, stride, stride, 16);\

  2172. }\

  2173. \

  2174. static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2175.     uint64_t temp[32];\

  2176.     uint8_t * const half= (uint8_t*)temp;\

  2177.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2178.     OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\

  2179. }\

  2180. \

  2181. static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2182.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2183. }\

  2184. \

  2185. static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2186.     uint64_t temp[32];\

  2187.     uint8_t * const half= (uint8_t*)temp;\

  2188.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2189.     OPNAME ## pixels16_l2_mmx(dst, src+stride, half, stride, stride, 16);\

  2190. }\

  2191. static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2192.     uint64_t half[16*2 + 17*2];\

  2193.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2194.     uint8_t * const halfHV= ((uint8_t*)half);\

  2195.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2196.     put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\

  2197.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2198.     OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\

  2199. }\

  2200. static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2201.     uint64_t half[16*2 + 17*2];\

  2202.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2203.     uint8_t * const halfHV= ((uint8_t*)half);\

  2204.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2205.     put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\

  2206.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2207.     OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\

  2208. }\

  2209. static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2210.     uint64_t half[16*2 + 17*2];\

  2211.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2212.     uint8_t * const halfHV= ((uint8_t*)half);\

  2213.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2214.     put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\

  2215.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2216.     OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\

  2217. }\

  2218. static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2219.     uint64_t half[16*2 + 17*2];\

  2220.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2221.     uint8_t * const halfHV= ((uint8_t*)half);\

  2222.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2223.     put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\

  2224.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2225.     OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\

  2226. }\

  2227. static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2228.     uint64_t half[16*2 + 17*2];\

  2229.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2230.     uint8_t * const halfHV= ((uint8_t*)half);\

  2231.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2232.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2233.     OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\

  2234. }\

  2235. static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2236.     uint64_t half[16*2 + 17*2];\

  2237.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2238.     uint8_t * const halfHV= ((uint8_t*)half);\

  2239.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2240.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2241.     OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\

  2242. }\

  2243. static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2244.     uint64_t half[17*2];\

  2245.     uint8_t * const halfH= ((uint8_t*)half);\

  2246.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2247.     put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\

  2248.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2249. }\

  2250. static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2251.     uint64_t half[17*2];\

  2252.     uint8_t * const halfH= ((uint8_t*)half);\

  2253.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2254.     put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\

  2255.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2256. }\

  2257. static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2258.     uint64_t half[17*2];\

  2259.     uint8_t * const halfH= ((uint8_t*)half);\

  2260.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2261.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2262. }

  2263. 

  2264. 

  2265. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "	\n\t"

  2266. #define AVG_3DNOW_OP(a,b,temp, size) \

  2267. "mov" #size " " #b ", " #temp "	\n\t"\

  2268. "pavgusb " #temp ", " #a "	\n\t"\

  2269. "mov" #size " " #a ", " #b "	\n\t"

  2270. #define AVG_MMX2_OP(a,b,temp, size) \

  2271. "mov" #size " " #b ", " #temp "	\n\t"\

  2272. "pavgb " #temp ", " #a "	\n\t"\

  2273. "mov" #size " " #a ", " #b "	\n\t"

  2274. 

  2275. QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)

  2276. QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)

  2277. QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)

  2278. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)

  2279. QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)

  2280. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)

  2281. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)

  2282. QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)

  2283. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

  2284. 

  2285. #if 0

  2286. static void just_return() { return; }

  2287. #endif

  2288. 

  2289. #define SET_QPEL_FUNC(postfix1, postfix2) \

  2290.     c->put_ ## postfix1 = put_ ## postfix2;\

  2291.     c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\

  2292.     c->avg_ ## postfix1 = avg_ ## postfix2;

  2293. 

  2294. static int try_8x8basis_mmx(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){

  2295.     int i=0;

  2296.     

  2297.     assert(ABS(scale) < 256);

  2298.     scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2299. 

  2300.     asm volatile(

  2301.         "pcmpeqw %%mm6, %%mm6		\n\t" // -1w

  2302.         "psrlw $15, %%mm6		\n\t" //  1w

  2303.         "pxor %%mm7, %%mm7		\n\t"

  2304.         "movd  %4, %%mm5		\n\t" 

  2305.         "punpcklwd %%mm5, %%mm5		\n\t" 

  2306.         "punpcklwd %%mm5, %%mm5		\n\t" 

  2307.         "1:				\n\t"

  2308.         "movq  (%1, %0), %%mm0		\n\t" 

  2309.         "movq  8(%1, %0), %%mm1		\n\t"

  2310.         "pmulhw %%mm5, %%mm0		\n\t"

  2311.         "pmulhw %%mm5, %%mm1		\n\t"

  2312.         "paddw %%mm6, %%mm0		\n\t"

  2313.         "paddw %%mm6, %%mm1		\n\t"

  2314.         "psraw $1, %%mm0		\n\t"

  2315.         "psraw $1, %%mm1		\n\t"

  2316.         "paddw (%2, %0), %%mm0		\n\t"

  2317.         "paddw 8(%2, %0), %%mm1		\n\t"

  2318.         "psraw $6, %%mm0		\n\t"

  2319.         "psraw $6, %%mm1		\n\t"

  2320.         "pmullw (%3, %0), %%mm0		\n\t"

  2321.         "pmullw 8(%3, %0), %%mm1	\n\t"

  2322.         "pmaddwd %%mm0, %%mm0		\n\t"

  2323.         "pmaddwd %%mm1, %%mm1		\n\t"

  2324.         "paddd %%mm1, %%mm0		\n\t"

  2325.         "psrld $4, %%mm0		\n\t"

  2326.         "paddd %%mm0, %%mm7		\n\t"

  2327.         "addl $16, %0			\n\t"

  2328.         "cmpl $128, %0			\n\t" //FIXME optimize & bench

  2329.         " jb 1b				\n\t"

  2330.         "movq %%mm7, %%mm6		\n\t"

  2331.         "psrlq $32, %%mm7		\n\t"

  2332.         "paddd %%mm6, %%mm7		\n\t"

  2333.         "psrld $2, %%mm7		\n\t"

  2334.         "movd %%mm7, %0			\n\t"

  2335.         

  2336.         : "+r" (i)

  2337.         : "r"(basis), "r"(rem), "r"(weight), "g"(scale)

  2338.     );

  2339.     return i;

  2340. }

  2341. 

  2342. static void add_8x8basis_mmx(int16_t rem[64], int16_t basis[64], int scale){

  2343.     int i=0;

  2344.     

  2345.     if(ABS(scale) < 256){

  2346.         scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2347.         asm volatile(

  2348.                 "pcmpeqw %%mm6, %%mm6		\n\t" // -1w

  2349.                 "psrlw $15, %%mm6		\n\t" //  1w

  2350.                 "movd  %3, %%mm5		\n\t" 

  2351.                 "punpcklwd %%mm5, %%mm5		\n\t" 

  2352.                 "punpcklwd %%mm5, %%mm5		\n\t" 

  2353.                 "1:				\n\t"

  2354.                 "movq  (%1, %0), %%mm0		\n\t" 

  2355.                 "movq  8(%1, %0), %%mm1		\n\t"

  2356.                 "pmulhw %%mm5, %%mm0		\n\t"

  2357.                 "pmulhw %%mm5, %%mm1		\n\t"

  2358.                 "paddw %%mm6, %%mm0		\n\t" 

  2359.                 "paddw %%mm6, %%mm1		\n\t"

  2360.                 "psraw $1, %%mm0		\n\t"

  2361.                 "psraw $1, %%mm1		\n\t"

  2362.                 "paddw (%2, %0), %%mm0		\n\t"

  2363.                 "paddw 8(%2, %0), %%mm1		\n\t"

  2364.                 "movq %%mm0, (%2, %0)		\n\t"

  2365.                 "movq %%mm1, 8(%2, %0)		\n\t"

  2366.                 "addl $16, %0			\n\t"

  2367.                 "cmpl $128, %0			\n\t" //FIXME optimize & bench

  2368.                 " jb 1b				\n\t"

  2369.                 

  2370.                 : "+r" (i)

  2371.                 : "r"(basis), "r"(rem), "g"(scale)

  2372.         );

  2373.     }else{

  2374.         for(i=0; i<8*8; i++){

  2375.             rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);

  2376.         }    

  2377.     }

  2378. }

  2379.     

  2380. /* external functions, from idct_mmx.c */

  2381. void ff_mmx_idct(DCTELEM *block);

  2382. void ff_mmxext_idct(DCTELEM *block);

  2383. 

  2384. /* XXX: those functions should be suppressed ASAP when all IDCTs are

  2385.    converted */

  2386. static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2387. {

  2388.     ff_mmx_idct (block);

  2389.     put_pixels_clamped_mmx(block, dest, line_size);

  2390. }

  2391. static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2392. {

  2393.     ff_mmx_idct (block);

  2394.     add_pixels_clamped_mmx(block, dest, line_size);

  2395. }

  2396. static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2397. {

  2398.     ff_mmxext_idct (block);

  2399.     put_pixels_clamped_mmx(block, dest, line_size);

  2400. }

  2401. static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2402. {

  2403.     ff_mmxext_idct (block);

  2404.     add_pixels_clamped_mmx(block, dest, line_size);

  2405. }

  2406.     

  2407. void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)

  2408. {

  2409.     mm_flags = mm_support();

  2410. 

  2411.     if (avctx->dsp_mask) {

  2412. 	if (avctx->dsp_mask & FF_MM_FORCE)

  2413. 	    mm_flags |= (avctx->dsp_mask & 0xffff);

  2414. 	else

  2415. 	    mm_flags &= ~(avctx->dsp_mask & 0xffff);

  2416.     }

  2417. 

  2418. #if 0

  2419.     av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");

  2420.     if (mm_flags & MM_MMX)

  2421.         av_log(avctx, AV_LOG_INFO, " mmx");

  2422.     if (mm_flags & MM_MMXEXT)

  2423.         av_log(avctx, AV_LOG_INFO, " mmxext");

  2424.     if (mm_flags & MM_3DNOW)

  2425.         av_log(avctx, AV_LOG_INFO, " 3dnow");

  2426.     if (mm_flags & MM_SSE)

  2427.         av_log(avctx, AV_LOG_INFO, " sse");

  2428.     if (mm_flags & MM_SSE2)

  2429.         av_log(avctx, AV_LOG_INFO, " sse2");

  2430.     av_log(avctx, AV_LOG_INFO, "\n");

  2431. #endif

  2432. 

  2433.     if (mm_flags & MM_MMX) {

  2434.         const int idct_algo= avctx->idct_algo;

  2435. 

  2436. #ifdef CONFIG_ENCODERS

  2437.         const int dct_algo = avctx->dct_algo;

  2438.         if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){

  2439.             if(mm_flags & MM_SSE2){

  2440.                 c->fdct = ff_fdct_sse2;

  2441. 	    }else if(mm_flags & MM_MMXEXT){

  2442.                 c->fdct = ff_fdct_mmx2;

  2443.             }else{

  2444.                 c->fdct = ff_fdct_mmx;

  2445.             }

  2446.         }

  2447. #endif //CONFIG_ENCODERS

  2448. 

  2449.         if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){

  2450.             c->idct_put= ff_simple_idct_put_mmx;

  2451.             c->idct_add= ff_simple_idct_add_mmx;

  2452.             c->idct    = ff_simple_idct_mmx;

  2453.             c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;

  2454.         }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){

  2455.             if(mm_flags & MM_MMXEXT){

  2456.                 c->idct_put= ff_libmpeg2mmx2_idct_put;

  2457.                 c->idct_add= ff_libmpeg2mmx2_idct_add;

  2458.                 c->idct    = ff_mmxext_idct;

  2459.             }else{

  2460.                 c->idct_put= ff_libmpeg2mmx_idct_put;

  2461.                 c->idct_add= ff_libmpeg2mmx_idct_add;

  2462.                 c->idct    = ff_mmx_idct;

  2463.             }

  2464.             c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;

  2465.         }

  2466. 

  2467.         /* VP3 optimized DSP functions */

  2468.         if (mm_flags & MM_SSE2) {

  2469.             c->vp3_dsp_init = vp3_dsp_init_sse2;

  2470.             c->vp3_idct = vp3_idct_sse2;

  2471.         } else {

  2472.             c->vp3_dsp_init = vp3_dsp_init_mmx;

  2473.             c->vp3_idct = vp3_idct_mmx;

  2474.         }

  2475. 

  2476. #ifdef CONFIG_ENCODERS

  2477.         c->get_pixels = get_pixels_mmx;

  2478.         c->diff_pixels = diff_pixels_mmx;

  2479. #endif //CONFIG_ENCODERS

  2480.         c->put_pixels_clamped = put_pixels_clamped_mmx;

  2481.         c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;

  2482.         c->add_pixels_clamped = add_pixels_clamped_mmx;

  2483.         c->clear_blocks = clear_blocks_mmx;

  2484. #ifdef CONFIG_ENCODERS

  2485.         c->pix_sum = pix_sum16_mmx;

  2486. #endif //CONFIG_ENCODERS

  2487. 

  2488.         c->put_pixels_tab[0][0] = put_pixels16_mmx;

  2489.         c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;

  2490.         c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;

  2491.         c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

  2492. 

  2493.         c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;

  2494.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;

  2495.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;

  2496.         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

  2497. 

  2498.         c->avg_pixels_tab[0][0] = avg_pixels16_mmx;

  2499.         c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;

  2500.         c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;

  2501.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

  2502. 

  2503.         c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;

  2504.         c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;

  2505.         c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;

  2506.         c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

  2507. 

  2508.         c->put_pixels_tab[1][0] = put_pixels8_mmx;

  2509.         c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;

  2510.         c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;

  2511.         c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

  2512. 

  2513.         c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;

  2514.         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;

  2515.         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;

  2516.         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

  2517. 

  2518.         c->avg_pixels_tab[1][0] = avg_pixels8_mmx;

  2519.         c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;

  2520.         c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;

  2521.         c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

  2522. 

  2523.         c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;

  2524.         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;

  2525.         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;

  2526.         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;

  2527.                 

  2528.         c->add_bytes= add_bytes_mmx;

  2529. #ifdef CONFIG_ENCODERS

  2530.         c->diff_bytes= diff_bytes_mmx;

  2531.         

  2532.         c->hadamard8_diff[0]= hadamard8_diff16_mmx;

  2533.         c->hadamard8_diff[1]= hadamard8_diff_mmx;

  2534.         

  2535. 	c->pix_norm1 = pix_norm1_mmx;

  2536. 	c->sse[0] = sse16_mmx;

  2537.   	c->sse[1] = sse8_mmx;

  2538.         c->vsad[4]= vsad_intra16_mmx;

  2539. 

  2540. 	c->nsse[0] = nsse16_mmx;

  2541. 	c->nsse[1] = nsse8_mmx;

  2542.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2543.             c->vsad[0] = vsad16_mmx;

  2544.         }

  2545.         

  2546.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2547.             c->try_8x8basis= try_8x8basis_mmx;

  2548.         }

  2549.         c->add_8x8basis= add_8x8basis_mmx;

  2550.         

  2551. #endif //CONFIG_ENCODERS

  2552. 

  2553.         c->h263_v_loop_filter= h263_v_loop_filter_mmx;

  2554.         c->h263_h_loop_filter= h263_h_loop_filter_mmx;        

  2555.         

  2556.         if (mm_flags & MM_MMXEXT) {

  2557.             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;

  2558.             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

  2559. 

  2560.             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;

  2561.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;

  2562.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

  2563. 

  2564.             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;

  2565.             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

  2566. 

  2567.             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;

  2568.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;

  2569.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

  2570. 

  2571. #ifdef CONFIG_ENCODERS

  2572.             c->hadamard8_diff[0]= hadamard8_diff16_mmx2;

  2573.             c->hadamard8_diff[1]= hadamard8_diff_mmx2;

  2574.             c->vsad[4]= vsad_intra16_mmx2;

  2575. #endif //CONFIG_ENCODERS

  2576. 

  2577.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2578.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;

  2579.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

  2580.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

  2581.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

  2582.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

  2583.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;

  2584. #ifdef CONFIG_ENCODERS

  2585.                 c->vsad[0] = vsad16_mmx2;

  2586. #endif //CONFIG_ENCODERS

  2587.             }

  2588. 

  2589. #if 1

  2590.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)

  2591.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)

  2592.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)

  2593.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)

  2594.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)

  2595.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)

  2596.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)

  2597.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)

  2598.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)

  2599.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)

  2600.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)

  2601.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)

  2602.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)

  2603.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)

  2604.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)

  2605.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)

  2606.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)

  2607.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)

  2608.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)

  2609.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)

  2610.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)

  2611.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)

  2612.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)

  2613.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)

  2614.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)

  2615.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)

  2616.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)

  2617.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)

  2618.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)

  2619.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)

  2620.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)

  2621.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)

  2622. #endif

  2623. 

  2624. #ifdef CONFIG_ENCODERS

  2625.             c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;

  2626. #endif //CONFIG_ENCODERS

  2627.         } else if (mm_flags & MM_3DNOW) {

  2628.             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;

  2629.             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  2630. 

  2631.             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  2632.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  2633.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  2634. 

  2635.             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;

  2636.             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

  2637. 

  2638.             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;

  2639.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;

  2640.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

  2641. 

  2642.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2643.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  2644.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  2645.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;

  2646.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

  2647.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  2648.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;

  2649.             }

  2650. 

  2651.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)

  2652.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)

  2653.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)

  2654.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)

  2655.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)

  2656.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)

  2657.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)

  2658.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)

  2659.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)

  2660.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)

  2661.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)

  2662.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)

  2663.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)

  2664.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)

  2665.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)

  2666.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)

  2667.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)

  2668.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)

  2669.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)

  2670.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)

  2671.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)

  2672.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)

  2673.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)

  2674.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)

  2675.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)

  2676.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)

  2677.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)

  2678.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)

  2679.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)

  2680.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)

  2681.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)

  2682.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)

  2683.         }

  2684.     }

  2685.         

  2686. #ifdef CONFIG_ENCODERS

  2687.     dsputil_init_pix_mmx(c, avctx);

  2688. #endif //CONFIG_ENCODERS

  2689. #if 0

  2690.     // for speed testing

  2691.     get_pixels = just_return;

  2692.     put_pixels_clamped = just_return;

  2693.     add_pixels_clamped = just_return;

  2694. 

  2695.     pix_abs16x16 = just_return;

  2696.     pix_abs16x16_x2 = just_return;

  2697.     pix_abs16x16_y2 = just_return;

  2698.     pix_abs16x16_xy2 = just_return;

  2699. 

  2700.     put_pixels_tab[0] = just_return;

  2701.     put_pixels_tab[1] = just_return;

  2702.     put_pixels_tab[2] = just_return;

  2703.     put_pixels_tab[3] = just_return;

  2704. 

  2705.     put_no_rnd_pixels_tab[0] = just_return;

  2706.     put_no_rnd_pixels_tab[1] = just_return;

  2707.     put_no_rnd_pixels_tab[2] = just_return;

  2708.     put_no_rnd_pixels_tab[3] = just_return;

  2709. 

  2710.     avg_pixels_tab[0] = just_return;

  2711.     avg_pixels_tab[1] = just_return;

  2712.     avg_pixels_tab[2] = just_return;

  2713.     avg_pixels_tab[3] = just_return;

  2714. 

  2715.     avg_no_rnd_pixels_tab[0] = just_return;

  2716.     avg_no_rnd_pixels_tab[1] = just_return;

  2717.     avg_no_rnd_pixels_tab[2] = just_return;

  2718.     avg_no_rnd_pixels_tab[3] = just_return;

  2719. 

  2720.     //av_fdct = just_return;

  2721.     //ff_idct = just_return;

  2722. #endif

  2723. }