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.
6940 Commits
32 Branches
325MB
Tree: b78e7197a8
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b78e7197a8'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx.c

3531 lines
134KB
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 file is part of FFmpeg.

  7.  *

  8.  * FFmpeg is free software; you can redistribute it and/or

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

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

  11.  * version 2.1 of the License, or (at your option) any later version.

  12.  *

  13.  * FFmpeg is distributed in the hope that it will be useful,

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

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

  16.  * Lesser General Public License for more details.

  17.  *

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

  19.  * License along with FFmpeg; if not, write to the Free Software

  20.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  21.  *

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

  23.  */

  24. 

  25. #include "../dsputil.h"

  26. #include "../simple_idct.h"

  27. #include "../mpegvideo.h"

  28. #include "x86_cpu.h"

  29. #include "mmx.h"

  30. 

  31. //#undef NDEBUG

  32. //#include <assert.h>

  33. 

  34. extern void ff_idct_xvid_mmx(short *block);

  35. extern void ff_idct_xvid_mmx2(short *block);

  36. 

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

  38. 

  39. /* pixel operations */

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

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

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

  43. 

  44. static const uint64_t ff_pdw_80000000[2] attribute_used __attribute__ ((aligned(16))) =

  45. {0x8000000080000000ULL, 0x8000000080000000ULL};

  46. 

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

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

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

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

  51. static const uint64_t ff_pw_8  attribute_used __attribute__ ((aligned(8))) = 0x0008000800080008ULL;

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

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

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

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

  56. 

  57. static const uint64_t ff_pb_1  attribute_used __attribute__ ((aligned(8))) = 0x0101010101010101ULL;

  58. static const uint64_t ff_pb_3  attribute_used __attribute__ ((aligned(8))) = 0x0303030303030303ULL;

  59. static const uint64_t ff_pb_7  attribute_used __attribute__ ((aligned(8))) = 0x0707070707070707ULL;

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

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

  62. 

  63. #define JUMPALIGN() __asm __volatile (ASMALIGN(3)::)

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

  65. 

  66. #define MOVQ_WONE(regd) \

  67.     __asm __volatile ( \

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

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

  70. 

  71. #define MOVQ_BFE(regd) \

  72.     __asm __volatile ( \

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

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

  75. 

  76. #ifndef PIC

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

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

  79. #else

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

  81. // pcmpeqd -> -1

  82. #define MOVQ_BONE(regd) \

  83.     __asm __volatile ( \

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

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

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

  87. 

  88. #define MOVQ_WTWO(regd) \

  89.     __asm __volatile ( \

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

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

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

  93. 

  94. #endif

  95. 

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

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

  98. // regfe is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

  106. 

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

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

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

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

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

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

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

  114. 

  115. // mm6 is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

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

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

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

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

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

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

  129. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  143. 

  144. /***********************************/

  145. /* MMX no rounding */

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

  147. #define SET_RND  MOVQ_WONE

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

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

  150. 

  151. #include "dsputil_mmx_rnd.h"

  152. 

  153. #undef DEF

  154. #undef SET_RND

  155. #undef PAVGBP

  156. #undef PAVGB

  157. /***********************************/

  158. /* MMX rounding */

  159. 

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

  161. #define SET_RND  MOVQ_WTWO

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

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

  164. 

  165. #include "dsputil_mmx_rnd.h"

  166. 

  167. #undef DEF

  168. #undef SET_RND

  169. #undef PAVGBP

  170. #undef PAVGB

  171. 

  172. /***********************************/

  173. /* 3Dnow specific */

  174. 

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

  176. /* for Athlons PAVGUSB is prefered */

  177. #define PAVGB "pavgusb"

  178. 

  179. #include "dsputil_mmx_avg.h"

  180. 

  181. #undef DEF

  182. #undef PAVGB

  183. 

  184. /***********************************/

  185. /* MMX2 specific */

  186. 

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

  188. 

  189. /* Introduced only in MMX2 set */

  190. #define PAVGB "pavgb"

  191. 

  192. #include "dsputil_mmx_avg.h"

  193. 

  194. #undef DEF

  195. #undef PAVGB

  196. 

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

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

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

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

  201. 

  202. /***********************************/

  203. /* standard MMX */

  204. 

  205. #ifdef CONFIG_ENCODERS

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

  207. {

  208.     asm volatile(

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

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

  211.         ASMALIGN(4)

  212.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  227.         "js 1b                          \n\t"

  228.         : "+r" (pixels)

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

  230.         : "%"REG_a

  231.     );

  232. }

  233. 

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

  235. {

  236.     asm volatile(

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

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

  239.         ASMALIGN(4)

  240.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  256.         "jnz 1b                         \n\t"

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

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

  259.         : "%"REG_a

  260.     );

  261. }

  262. #endif //CONFIG_ENCODERS

  263. 

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

  265. {

  266.     const DCTELEM *p;

  267.     uint8_t *pix;

  268. 

  269.     /* read the pixels */

  270.     p = block;

  271.     pix = pixels;

  272.     /* unrolled loop */

  273.         __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  291.                 :"memory");

  292.         pix += line_size*4;

  293.         p += 32;

  294. 

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

  296.     // compiler would generate some very strange code

  297.     // thus using "r"

  298.     __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  316.             :"memory");

  317. }

  318. 

  319. static DECLARE_ALIGNED_8(const unsigned char, vector128[8]) =

  320.   { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };

  321. 

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

  323. {

  324.     int i;

  325. 

  326.     movq_m2r(*vector128, mm1);

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

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

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

  330.         block += 8;

  331.         paddb_r2r(mm1, mm0);

  332.         movq_r2m(mm0, *pixels);

  333.         pixels += line_size;

  334.     }

  335. }

  336. 

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

  338. {

  339.     const DCTELEM *p;

  340.     uint8_t *pix;

  341.     int i;

  342. 

  343.     /* read the pixels */

  344.     p = block;

  345.     pix = pixels;

  346.     MOVQ_ZERO(mm7);

  347.     i = 4;

  348.     do {

  349.         __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  371.                 :"r"(p)

  372.                 :"memory");

  373.         pix += line_size*2;

  374.         p += 16;

  375.     } while (--i);

  376. }

  377. 

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

  379. {

  380.     __asm __volatile(

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

  382.          ASMALIGN(3)

  383.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  397.          "jnz 1b                        \n\t"

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

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

  400.          : "%"REG_a, "memory"

  401.         );

  402. }

  403. 

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

  405. {

  406.     __asm __volatile(

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

  408.          ASMALIGN(3)

  409.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  423.          "jnz 1b                        \n\t"

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

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

  426.          : "%"REG_a, "memory"

  427.         );

  428. }

  429. 

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

  431. {

  432.     __asm __volatile(

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

  434.          ASMALIGN(3)

  435.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  457.          "jnz 1b                        \n\t"

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

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

  460.          : "%"REG_a, "memory"

  461.         );

  462. }

  463. 

  464. static void clear_blocks_mmx(DCTELEM *blocks)

  465. {

  466.     __asm __volatile(

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

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

  469.                 "1:                             \n\t"

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

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

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

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

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

  475.                 " js 1b                         \n\t"

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

  477.                 : "%"REG_a

  478.         );

  479. }

  480. 

  481. #ifdef CONFIG_ENCODERS

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

  483.     const int h=16;

  484.     int sum;

  485.     long index= -line_size*h;

  486. 

  487.     __asm __volatile(

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

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

  490.                 "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  504.                 " js 1b                         \n\t"

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

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

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

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

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

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

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

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

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

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

  515.         );

  516. 

  517.         return sum;

  518. }

  519. #endif //CONFIG_ENCODERS

  520. 

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

  522.     long i=0;

  523.     asm volatile(

  524.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

  535.         " jb 1b                         \n\t"

  536.         : "+r" (i)

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

  538.     );

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

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

  541. }

  542. 

  543. #define H263_LOOP_FILTER \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  613. 

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

  615.     const int strength= ff_h263_loop_filter_strength[qscale];

  616. 

  617.     asm volatile(

  618. 

  619.         H263_LOOP_FILTER

  620. 

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

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

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

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

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

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

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

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

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

  630.     );

  631. }

  632. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  650. 

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

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

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

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

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

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

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

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

  659.     );

  660. }

  661. 

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

  663.     const int strength= ff_h263_loop_filter_strength[qscale];

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

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

  666. 

  667.     src -= 2;

  668. 

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

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

  671.     asm volatile(

  672.         H263_LOOP_FILTER // 5 3 4 6

  673. 

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

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

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

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

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

  679.     );

  680. 

  681.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  706.         :: "r" (src),

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

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

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

  710.     );

  711. }

  712. 

  713. #ifdef CONFIG_ENCODERS

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

  715.     int tmp;

  716.   asm volatile (

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

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

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

  720.       "1:\n"

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

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

  723. 

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

  725. 

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

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

  728. 

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

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

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

  732. 

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

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

  735. 

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

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

  738. 

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

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

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

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

  743. 

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

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

  746.       "dec %%ecx\n"

  747.       "jnz 1b\n"

  748. 

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

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

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

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

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

  754.     return tmp;

  755. }

  756. 

  757. static int sse8_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.       "shr $1,%%ecx\n"

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

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

  764.       "1:\n"

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

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

  767.       "movq (%0,%3),%%mm3\n"    /* mm3 = pix1[1][0-7] */

  768.       "movq (%1,%3),%%mm4\n"    /* mm4 = pix2[1][0-7] */

  769. 

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

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

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

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

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

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

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

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

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

  779. 

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

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

  782. 

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

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

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

  786. 

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

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

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

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

  791. 

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

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

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

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

  796. 

  797.       "lea (%0,%3,2), %0\n"     /* pix1 += 2*line_size */

  798.       "lea (%1,%3,2), %1\n"     /* pix2 += 2*line_size */

  799. 

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

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

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

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

  804. 

  805.       "decl %%ecx\n"

  806.       "jnz 1b\n"

  807. 

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

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

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

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

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

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

  814.       : "%ecx");

  815.     return tmp;

  816. }

  817. 

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

  819.     int tmp;

  820.   asm volatile (

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

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

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

  824.       "1:\n"

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

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

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

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

  829. 

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

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

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

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

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

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

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

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

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

  839. 

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

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

  842. 

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

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

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

  846. 

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

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

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

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

  851. 

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

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

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

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

  856. 

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

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

  859. 

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

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

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

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

  864. 

  865.       "decl %%ecx\n"

  866.       "jnz 1b\n"

  867. 

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

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

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

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

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

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

  874.       : "%ecx");

  875.     return tmp;

  876. }

  877. 

  878. static int sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  879.     int tmp;

  880.   asm volatile (

  881.       "shr $1,%2\n"

  882.       "pxor %%xmm0,%%xmm0\n"    /* mm0 = 0 */

  883.       "pxor %%xmm7,%%xmm7\n"    /* mm7 holds the sum */

  884.       "1:\n"

  885.       "movdqu (%0),%%xmm1\n"    /* mm1 = pix1[0][0-15] */

  886.       "movdqu (%1),%%xmm2\n"    /* mm2 = pix2[0][0-15] */

  887.       "movdqu (%0,%4),%%xmm3\n" /* mm3 = pix1[1][0-15] */

  888.       "movdqu (%1,%4),%%xmm4\n" /* mm4 = pix2[1][0-15] */

  889. 

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

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

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

  893.       "movdqa %%xmm1,%%xmm5\n"

  894.       "movdqa %%xmm3,%%xmm6\n"

  895.       "psubusb %%xmm2,%%xmm1\n"

  896.       "psubusb %%xmm4,%%xmm3\n"

  897.       "psubusb %%xmm5,%%xmm2\n"

  898.       "psubusb %%xmm6,%%xmm4\n"

  899. 

  900.       "por %%xmm1,%%xmm2\n"

  901.       "por %%xmm3,%%xmm4\n"

  902. 

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

  904.       "movdqa %%xmm2,%%xmm1\n"

  905.       "movdqa %%xmm4,%%xmm3\n"

  906. 

  907.       "punpckhbw %%xmm0,%%xmm2\n"

  908.       "punpckhbw %%xmm0,%%xmm4\n"

  909.       "punpcklbw %%xmm0,%%xmm1\n"  /* mm1 now spread over (mm1,mm2) */

  910.       "punpcklbw %%xmm0,%%xmm3\n"  /* mm4 now spread over (mm3,mm4) */

  911. 

  912.       "pmaddwd %%xmm2,%%xmm2\n"

  913.       "pmaddwd %%xmm4,%%xmm4\n"

  914.       "pmaddwd %%xmm1,%%xmm1\n"

  915.       "pmaddwd %%xmm3,%%xmm3\n"

  916. 

  917.       "lea (%0,%4,2), %0\n"        /* pix1 += 2*line_size */

  918.       "lea (%1,%4,2), %1\n"        /* pix2 += 2*line_size */

  919. 

  920.       "paddd %%xmm2,%%xmm1\n"

  921.       "paddd %%xmm4,%%xmm3\n"

  922.       "paddd %%xmm1,%%xmm7\n"

  923.       "paddd %%xmm3,%%xmm7\n"

  924. 

  925.       "decl %2\n"

  926.       "jnz 1b\n"

  927. 

  928.       "movdqa %%xmm7,%%xmm1\n"

  929.       "psrldq $8, %%xmm7\n"        /* shift hi qword to lo */

  930.       "paddd %%xmm1,%%xmm7\n"

  931.       "movdqa %%xmm7,%%xmm1\n"

  932.       "psrldq $4, %%xmm7\n"        /* shift hi dword to lo */

  933.       "paddd %%xmm1,%%xmm7\n"

  934.       "movd %%xmm7,%3\n"

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

  936.       : "r" ((long)line_size));

  937.     return tmp;

  938. }

  939. 

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

  941.     int tmp;

  942.   asm volatile (

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

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

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

  946. 

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

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

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

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

  951.       "psrlq $8, %%mm0\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.       "add %2,%0\n"

  962. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  988. 

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

  990.       "1:\n"

  991. 

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

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

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

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

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

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

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

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

  1000.       "punpcklbw %%mm7,%%mm1\n"

  1001.       "punpckhbw %%mm7,%%mm2\n"

  1002.       "punpckhbw %%mm7,%%mm3\n"

  1003.       "psubw %%mm1, %%mm0\n"

  1004.       "psubw %%mm3, %%mm2\n"

  1005.       "psubw %%mm0, %%mm4\n"

  1006.       "psubw %%mm2, %%mm5\n"

  1007.       "pxor %%mm3, %%mm3\n"

  1008.       "pxor %%mm1, %%mm1\n"

  1009.       "pcmpgtw %%mm4, %%mm3\n\t"

  1010.       "pcmpgtw %%mm5, %%mm1\n\t"

  1011.       "pxor %%mm3, %%mm4\n"

  1012.       "pxor %%mm1, %%mm5\n"

  1013.       "psubw %%mm3, %%mm4\n"

  1014.       "psubw %%mm1, %%mm5\n"

  1015.       "paddw %%mm4, %%mm5\n"

  1016.       "paddw %%mm5, %%mm6\n"

  1017. 

  1018.       "add %2,%0\n"

  1019. 

  1020.       "movq (%0),%%mm4\n"

  1021.       "movq %%mm4, %%mm1\n"

  1022.       "psllq $8, %%mm4\n"

  1023.       "psrlq $8, %%mm1\n"

  1024.       "psrlq $8, %%mm4\n"

  1025.       "movq %%mm4, %%mm5\n"

  1026.       "movq %%mm1, %%mm3\n"

  1027.       "punpcklbw %%mm7,%%mm4\n"

  1028.       "punpcklbw %%mm7,%%mm1\n"

  1029.       "punpckhbw %%mm7,%%mm5\n"

  1030.       "punpckhbw %%mm7,%%mm3\n"

  1031.       "psubw %%mm1, %%mm4\n"

  1032.       "psubw %%mm3, %%mm5\n"

  1033.       "psubw %%mm4, %%mm0\n"

  1034.       "psubw %%mm5, %%mm2\n"

  1035.       "pxor %%mm3, %%mm3\n"

  1036.       "pxor %%mm1, %%mm1\n"

  1037.       "pcmpgtw %%mm0, %%mm3\n\t"

  1038.       "pcmpgtw %%mm2, %%mm1\n\t"

  1039.       "pxor %%mm3, %%mm0\n"

  1040.       "pxor %%mm1, %%mm2\n"

  1041.       "psubw %%mm3, %%mm0\n"

  1042.       "psubw %%mm1, %%mm2\n"

  1043.       "paddw %%mm0, %%mm2\n"

  1044.       "paddw %%mm2, %%mm6\n"

  1045. 

  1046.       "add %2,%0\n"

  1047.       "subl $2, %%ecx\n"

  1048.       " jnz 1b\n"

  1049. 

  1050.       "movq %%mm6, %%mm0\n"

  1051.       "punpcklwd %%mm7,%%mm0\n"

  1052.       "punpckhwd %%mm7,%%mm6\n"

  1053.       "paddd %%mm0, %%mm6\n"

  1054. 

  1055.       "movq %%mm6,%%mm0\n"

  1056.       "psrlq $32, %%mm6\n"

  1057.       "paddd %%mm6,%%mm0\n"

  1058.       "movd %%mm0,%1\n"

  1059.       : "+r" (pix1), "=r"(tmp)

  1060.       : "r" ((long)line_size) , "g" (h-2)

  1061.       : "%ecx");

  1062.       return tmp;

  1063. }

  1064. 

  1065. static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {

  1066.     int tmp;

  1067.     uint8_t * pix= pix1;

  1068.   asm volatile (

  1069.       "movl %3,%%ecx\n"

  1070.       "pxor %%mm7,%%mm7\n"

  1071.       "pxor %%mm6,%%mm6\n"

  1072. 

  1073.       "movq (%0),%%mm0\n"

  1074.       "movq 1(%0),%%mm1\n"

  1075.       "movq %%mm0, %%mm2\n"

  1076.       "movq %%mm1, %%mm3\n"

  1077.       "punpcklbw %%mm7,%%mm0\n"

  1078.       "punpcklbw %%mm7,%%mm1\n"

  1079.       "punpckhbw %%mm7,%%mm2\n"

  1080.       "punpckhbw %%mm7,%%mm3\n"

  1081.       "psubw %%mm1, %%mm0\n"

  1082.       "psubw %%mm3, %%mm2\n"

  1083. 

  1084.       "add %2,%0\n"

  1085. 

  1086.       "movq (%0),%%mm4\n"

  1087.       "movq 1(%0),%%mm1\n"

  1088.       "movq %%mm4, %%mm5\n"

  1089.       "movq %%mm1, %%mm3\n"

  1090.       "punpcklbw %%mm7,%%mm4\n"

  1091.       "punpcklbw %%mm7,%%mm1\n"

  1092.       "punpckhbw %%mm7,%%mm5\n"

  1093.       "punpckhbw %%mm7,%%mm3\n"

  1094.       "psubw %%mm1, %%mm4\n"

  1095.       "psubw %%mm3, %%mm5\n"

  1096.       "psubw %%mm4, %%mm0\n"

  1097.       "psubw %%mm5, %%mm2\n"

  1098.       "pxor %%mm3, %%mm3\n"

  1099.       "pxor %%mm1, %%mm1\n"

  1100.       "pcmpgtw %%mm0, %%mm3\n\t"

  1101.       "pcmpgtw %%mm2, %%mm1\n\t"

  1102.       "pxor %%mm3, %%mm0\n"

  1103.       "pxor %%mm1, %%mm2\n"

  1104.       "psubw %%mm3, %%mm0\n"

  1105.       "psubw %%mm1, %%mm2\n"

  1106.       "paddw %%mm0, %%mm2\n"

  1107.       "paddw %%mm2, %%mm6\n"

  1108. 

  1109.       "add %2,%0\n"

  1110.       "1:\n"

  1111. 

  1112.       "movq (%0),%%mm0\n"

  1113.       "movq 1(%0),%%mm1\n"

  1114.       "movq %%mm0, %%mm2\n"

  1115.       "movq %%mm1, %%mm3\n"

  1116.       "punpcklbw %%mm7,%%mm0\n"

  1117.       "punpcklbw %%mm7,%%mm1\n"

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

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

  1120.       "psubw %%mm1, %%mm0\n"

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

  1122.       "psubw %%mm0, %%mm4\n"

  1123.       "psubw %%mm2, %%mm5\n"

  1124.       "pxor %%mm3, %%mm3\n"

  1125.       "pxor %%mm1, %%mm1\n"

  1126.       "pcmpgtw %%mm4, %%mm3\n\t"

  1127.       "pcmpgtw %%mm5, %%mm1\n\t"

  1128.       "pxor %%mm3, %%mm4\n"

  1129.       "pxor %%mm1, %%mm5\n"

  1130.       "psubw %%mm3, %%mm4\n"

  1131.       "psubw %%mm1, %%mm5\n"

  1132.       "paddw %%mm4, %%mm5\n"

  1133.       "paddw %%mm5, %%mm6\n"

  1134. 

  1135.       "add %2,%0\n"

  1136. 

  1137.       "movq (%0),%%mm4\n"

  1138.       "movq 1(%0),%%mm1\n"

  1139.       "movq %%mm4, %%mm5\n"

  1140.       "movq %%mm1, %%mm3\n"

  1141.       "punpcklbw %%mm7,%%mm4\n"

  1142.       "punpcklbw %%mm7,%%mm1\n"

  1143.       "punpckhbw %%mm7,%%mm5\n"

  1144.       "punpckhbw %%mm7,%%mm3\n"

  1145.       "psubw %%mm1, %%mm4\n"

  1146.       "psubw %%mm3, %%mm5\n"

  1147.       "psubw %%mm4, %%mm0\n"

  1148.       "psubw %%mm5, %%mm2\n"

  1149.       "pxor %%mm3, %%mm3\n"

  1150.       "pxor %%mm1, %%mm1\n"

  1151.       "pcmpgtw %%mm0, %%mm3\n\t"

  1152.       "pcmpgtw %%mm2, %%mm1\n\t"

  1153.       "pxor %%mm3, %%mm0\n"

  1154.       "pxor %%mm1, %%mm2\n"

  1155.       "psubw %%mm3, %%mm0\n"

  1156.       "psubw %%mm1, %%mm2\n"

  1157.       "paddw %%mm0, %%mm2\n"

  1158.       "paddw %%mm2, %%mm6\n"

  1159. 

  1160.       "add %2,%0\n"

  1161.       "subl $2, %%ecx\n"

  1162.       " jnz 1b\n"

  1163. 

  1164.       "movq %%mm6, %%mm0\n"

  1165.       "punpcklwd %%mm7,%%mm0\n"

  1166.       "punpckhwd %%mm7,%%mm6\n"

  1167.       "paddd %%mm0, %%mm6\n"

  1168. 

  1169.       "movq %%mm6,%%mm0\n"

  1170.       "psrlq $32, %%mm6\n"

  1171.       "paddd %%mm6,%%mm0\n"

  1172.       "movd %%mm0,%1\n"

  1173.       : "+r" (pix1), "=r"(tmp)

  1174.       : "r" ((long)line_size) , "g" (h-2)

  1175.       : "%ecx");

  1176.       return tmp + hf_noise8_mmx(pix+8, line_size, h);

  1177. }

  1178. 

  1179. static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1180.     MpegEncContext *c = p;

  1181.     int score1, score2;

  1182. 

  1183.     if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);

  1184.     else  score1 = sse16_mmx(c, pix1, pix2, line_size, h);

  1185.     score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

  1186. 

  1187.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1188.     else  return score1 + ABS(score2)*8;

  1189. }

  1190. 

  1191. static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1192.     MpegEncContext *c = p;

  1193.     int score1= sse8_mmx(c, pix1, pix2, line_size, h);

  1194.     int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

  1195. 

  1196.     if(c) return score1 + ABS(score2)*c->avctx->nsse_weight;

  1197.     else  return score1 + ABS(score2)*8;

  1198. }

  1199. 

  1200. static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1201.     int tmp;

  1202. 

  1203.     assert( (((int)pix) & 7) == 0);

  1204.     assert((line_size &7) ==0);

  1205. 

  1206. #define SUM(in0, in1, out0, out1) \

  1207.       "movq (%0), %%mm2\n"\

  1208.       "movq 8(%0), %%mm3\n"\

  1209.       "add %2,%0\n"\

  1210.       "movq %%mm2, " #out0 "\n"\

  1211.       "movq %%mm3, " #out1 "\n"\

  1212.       "psubusb " #in0 ", %%mm2\n"\

  1213.       "psubusb " #in1 ", %%mm3\n"\

  1214.       "psubusb " #out0 ", " #in0 "\n"\

  1215.       "psubusb " #out1 ", " #in1 "\n"\

  1216.       "por %%mm2, " #in0 "\n"\

  1217.       "por %%mm3, " #in1 "\n"\

  1218.       "movq " #in0 ", %%mm2\n"\

  1219.       "movq " #in1 ", %%mm3\n"\

  1220.       "punpcklbw %%mm7, " #in0 "\n"\

  1221.       "punpcklbw %%mm7, " #in1 "\n"\

  1222.       "punpckhbw %%mm7, %%mm2\n"\

  1223.       "punpckhbw %%mm7, %%mm3\n"\

  1224.       "paddw " #in1 ", " #in0 "\n"\

  1225.       "paddw %%mm3, %%mm2\n"\

  1226.       "paddw %%mm2, " #in0 "\n"\

  1227.       "paddw " #in0 ", %%mm6\n"

  1228. 

  1229. 

  1230.   asm volatile (

  1231.       "movl %3,%%ecx\n"

  1232.       "pxor %%mm6,%%mm6\n"

  1233.       "pxor %%mm7,%%mm7\n"

  1234.       "movq (%0),%%mm0\n"

  1235.       "movq 8(%0),%%mm1\n"

  1236.       "add %2,%0\n"

  1237.       "subl $2, %%ecx\n"

  1238.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1239.       "1:\n"

  1240. 

  1241.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1242. 

  1243.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1244. 

  1245.       "subl $2, %%ecx\n"

  1246.       "jnz 1b\n"

  1247. 

  1248.       "movq %%mm6,%%mm0\n"

  1249.       "psrlq $32, %%mm6\n"

  1250.       "paddw %%mm6,%%mm0\n"

  1251.       "movq %%mm0,%%mm6\n"

  1252.       "psrlq $16, %%mm0\n"

  1253.       "paddw %%mm6,%%mm0\n"

  1254.       "movd %%mm0,%1\n"

  1255.       : "+r" (pix), "=r"(tmp)

  1256.       : "r" ((long)line_size) , "m" (h)

  1257.       : "%ecx");

  1258.     return tmp & 0xFFFF;

  1259. }

  1260. #undef SUM

  1261. 

  1262. static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1263.     int tmp;

  1264. 

  1265.     assert( (((int)pix) & 7) == 0);

  1266.     assert((line_size &7) ==0);

  1267. 

  1268. #define SUM(in0, in1, out0, out1) \

  1269.       "movq (%0), " #out0 "\n"\

  1270.       "movq 8(%0), " #out1 "\n"\

  1271.       "add %2,%0\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 %3,%%ecx\n"

  1279.       "pxor %%mm6,%%mm6\n"

  1280.       "pxor %%mm7,%%mm7\n"

  1281.       "movq (%0),%%mm0\n"

  1282.       "movq 8(%0),%%mm1\n"

  1283.       "add %2,%0\n"

  1284.       "subl $2, %%ecx\n"

  1285.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1286.       "1:\n"

  1287. 

  1288.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1289. 

  1290.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1291. 

  1292.       "subl $2, %%ecx\n"

  1293.       "jnz 1b\n"

  1294. 

  1295.       "movd %%mm6,%1\n"

  1296.       : "+r" (pix), "=r"(tmp)

  1297.       : "r" ((long)line_size) , "m" (h)

  1298.       : "%ecx");

  1299.     return tmp;

  1300. }

  1301. #undef SUM

  1302. 

  1303. static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1304.     int tmp;

  1305. 

  1306.     assert( (((int)pix1) & 7) == 0);

  1307.     assert( (((int)pix2) & 7) == 0);

  1308.     assert((line_size &7) ==0);

  1309. 

  1310. #define SUM(in0, in1, out0, out1) \

  1311.       "movq (%0),%%mm2\n"\

  1312.       "movq (%1)," #out0 "\n"\

  1313.       "movq 8(%0),%%mm3\n"\

  1314.       "movq 8(%1)," #out1 "\n"\

  1315.       "add %3,%0\n"\

  1316.       "add %3,%1\n"\

  1317.       "psubb " #out0 ", %%mm2\n"\

  1318.       "psubb " #out1 ", %%mm3\n"\

  1319.       "pxor %%mm7, %%mm2\n"\

  1320.       "pxor %%mm7, %%mm3\n"\

  1321.       "movq %%mm2, " #out0 "\n"\

  1322.       "movq %%mm3, " #out1 "\n"\

  1323.       "psubusb " #in0 ", %%mm2\n"\

  1324.       "psubusb " #in1 ", %%mm3\n"\

  1325.       "psubusb " #out0 ", " #in0 "\n"\

  1326.       "psubusb " #out1 ", " #in1 "\n"\

  1327.       "por %%mm2, " #in0 "\n"\

  1328.       "por %%mm3, " #in1 "\n"\

  1329.       "movq " #in0 ", %%mm2\n"\

  1330.       "movq " #in1 ", %%mm3\n"\

  1331.       "punpcklbw %%mm7, " #in0 "\n"\

  1332.       "punpcklbw %%mm7, " #in1 "\n"\

  1333.       "punpckhbw %%mm7, %%mm2\n"\

  1334.       "punpckhbw %%mm7, %%mm3\n"\

  1335.       "paddw " #in1 ", " #in0 "\n"\

  1336.       "paddw %%mm3, %%mm2\n"\

  1337.       "paddw %%mm2, " #in0 "\n"\

  1338.       "paddw " #in0 ", %%mm6\n"

  1339. 

  1340. 

  1341.   asm volatile (

  1342.       "movl %4,%%ecx\n"

  1343.       "pxor %%mm6,%%mm6\n"

  1344.       "pcmpeqw %%mm7,%%mm7\n"

  1345.       "psllw $15, %%mm7\n"

  1346.       "packsswb %%mm7, %%mm7\n"

  1347.       "movq (%0),%%mm0\n"

  1348.       "movq (%1),%%mm2\n"

  1349.       "movq 8(%0),%%mm1\n"

  1350.       "movq 8(%1),%%mm3\n"

  1351.       "add %3,%0\n"

  1352.       "add %3,%1\n"

  1353.       "subl $2, %%ecx\n"

  1354.       "psubb %%mm2, %%mm0\n"

  1355.       "psubb %%mm3, %%mm1\n"

  1356.       "pxor %%mm7, %%mm0\n"

  1357.       "pxor %%mm7, %%mm1\n"

  1358.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1359.       "1:\n"

  1360. 

  1361.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1362. 

  1363.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1364. 

  1365.       "subl $2, %%ecx\n"

  1366.       "jnz 1b\n"

  1367. 

  1368.       "movq %%mm6,%%mm0\n"

  1369.       "psrlq $32, %%mm6\n"

  1370.       "paddw %%mm6,%%mm0\n"

  1371.       "movq %%mm0,%%mm6\n"

  1372.       "psrlq $16, %%mm0\n"

  1373.       "paddw %%mm6,%%mm0\n"

  1374.       "movd %%mm0,%2\n"

  1375.       : "+r" (pix1), "+r" (pix2), "=r"(tmp)

  1376.       : "r" ((long)line_size) , "m" (h)

  1377.       : "%ecx");

  1378.     return tmp & 0x7FFF;

  1379. }

  1380. #undef SUM

  1381. 

  1382. static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1383.     int tmp;

  1384. 

  1385.     assert( (((int)pix1) & 7) == 0);

  1386.     assert( (((int)pix2) & 7) == 0);

  1387.     assert((line_size &7) ==0);

  1388. 

  1389. #define SUM(in0, in1, out0, out1) \

  1390.       "movq (%0)," #out0 "\n"\

  1391.       "movq (%1),%%mm2\n"\

  1392.       "movq 8(%0)," #out1 "\n"\

  1393.       "movq 8(%1),%%mm3\n"\

  1394.       "add %3,%0\n"\

  1395.       "add %3,%1\n"\

  1396.       "psubb %%mm2, " #out0 "\n"\

  1397.       "psubb %%mm3, " #out1 "\n"\

  1398.       "pxor %%mm7, " #out0 "\n"\

  1399.       "pxor %%mm7, " #out1 "\n"\

  1400.       "psadbw " #out0 ", " #in0 "\n"\

  1401.       "psadbw " #out1 ", " #in1 "\n"\

  1402.       "paddw " #in1 ", " #in0 "\n"\

  1403.       "paddw " #in0 ", %%mm6\n"

  1404. 

  1405.   asm volatile (

  1406.       "movl %4,%%ecx\n"

  1407.       "pxor %%mm6,%%mm6\n"

  1408.       "pcmpeqw %%mm7,%%mm7\n"

  1409.       "psllw $15, %%mm7\n"

  1410.       "packsswb %%mm7, %%mm7\n"

  1411.       "movq (%0),%%mm0\n"

  1412.       "movq (%1),%%mm2\n"

  1413.       "movq 8(%0),%%mm1\n"

  1414.       "movq 8(%1),%%mm3\n"

  1415.       "add %3,%0\n"

  1416.       "add %3,%1\n"

  1417.       "subl $2, %%ecx\n"

  1418.       "psubb %%mm2, %%mm0\n"

  1419.       "psubb %%mm3, %%mm1\n"

  1420.       "pxor %%mm7, %%mm0\n"

  1421.       "pxor %%mm7, %%mm1\n"

  1422.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1423.       "1:\n"

  1424. 

  1425.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1426. 

  1427.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1428. 

  1429.       "subl $2, %%ecx\n"

  1430.       "jnz 1b\n"

  1431. 

  1432.       "movd %%mm6,%2\n"

  1433.       : "+r" (pix1), "+r" (pix2), "=r"(tmp)

  1434.       : "r" ((long)line_size) , "m" (h)

  1435.       : "%ecx");

  1436.     return tmp;

  1437. }

  1438. #undef SUM

  1439. 

  1440. static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){

  1441.     long i=0;

  1442.     asm volatile(

  1443.         "1:                             \n\t"

  1444.         "movq  (%2, %0), %%mm0          \n\t"

  1445.         "movq  (%1, %0), %%mm1          \n\t"

  1446.         "psubb %%mm0, %%mm1             \n\t"

  1447.         "movq %%mm1, (%3, %0)           \n\t"

  1448.         "movq 8(%2, %0), %%mm0          \n\t"

  1449.         "movq 8(%1, %0), %%mm1          \n\t"

  1450.         "psubb %%mm0, %%mm1             \n\t"

  1451.         "movq %%mm1, 8(%3, %0)          \n\t"

  1452.         "add $16, %0                    \n\t"

  1453.         "cmp %4, %0                     \n\t"

  1454.         " jb 1b                         \n\t"

  1455.         : "+r" (i)

  1456.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15)

  1457.     );

  1458.     for(; i<w; i++)

  1459.         dst[i+0] = src1[i+0]-src2[i+0];

  1460. }

  1461. 

  1462. static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){

  1463.     long i=0;

  1464.     uint8_t l, lt;

  1465. 

  1466.     asm volatile(

  1467.         "1:                             \n\t"

  1468.         "movq  -1(%1, %0), %%mm0        \n\t" // LT

  1469.         "movq  (%1, %0), %%mm1          \n\t" // T

  1470.         "movq  -1(%2, %0), %%mm2        \n\t" // L

  1471.         "movq  (%2, %0), %%mm3          \n\t" // X

  1472.         "movq %%mm2, %%mm4              \n\t" // L

  1473.         "psubb %%mm0, %%mm2             \n\t"

  1474.         "paddb %%mm1, %%mm2             \n\t" // L + T - LT

  1475.         "movq %%mm4, %%mm5              \n\t" // L

  1476.         "pmaxub %%mm1, %%mm4            \n\t" // max(T, L)

  1477.         "pminub %%mm5, %%mm1            \n\t" // min(T, L)

  1478.         "pminub %%mm2, %%mm4            \n\t"

  1479.         "pmaxub %%mm1, %%mm4            \n\t"

  1480.         "psubb %%mm4, %%mm3             \n\t" // dst - pred

  1481.         "movq %%mm3, (%3, %0)           \n\t"

  1482.         "add $8, %0                     \n\t"

  1483.         "cmp %4, %0                     \n\t"

  1484.         " jb 1b                         \n\t"

  1485.         : "+r" (i)

  1486.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w)

  1487.     );

  1488. 

  1489.     l= *left;

  1490.     lt= *left_top;

  1491. 

  1492.     dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);

  1493. 

  1494.     *left_top= src1[w-1];

  1495.     *left    = src2[w-1];

  1496. }

  1497. 

  1498. #define LBUTTERFLY2(a1,b1,a2,b2)\

  1499.     "paddw " #b1 ", " #a1 "           \n\t"\

  1500.     "paddw " #b2 ", " #a2 "           \n\t"\

  1501.     "paddw " #b1 ", " #b1 "           \n\t"\

  1502.     "paddw " #b2 ", " #b2 "           \n\t"\

  1503.     "psubw " #a1 ", " #b1 "           \n\t"\

  1504.     "psubw " #a2 ", " #b2 "           \n\t"

  1505. 

  1506. #define HADAMARD48\

  1507.         LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\

  1508.         LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\

  1509.         LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\

  1510.         LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\

  1511.         LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\

  1512.         LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\

  1513. 

  1514. #define MMABS(a,z)\

  1515.     "pxor " #z ", " #z "              \n\t"\

  1516.     "pcmpgtw " #a ", " #z "           \n\t"\

  1517.     "pxor " #z ", " #a "              \n\t"\

  1518.     "psubw " #z ", " #a "             \n\t"

  1519. 

  1520. #define MMABS_SUM(a,z, sum)\

  1521.     "pxor " #z ", " #z "              \n\t"\

  1522.     "pcmpgtw " #a ", " #z "           \n\t"\

  1523.     "pxor " #z ", " #a "              \n\t"\

  1524.     "psubw " #z ", " #a "             \n\t"\

  1525.     "paddusw " #a ", " #sum "         \n\t"

  1526. 

  1527. #define MMABS_MMX2(a,z)\

  1528.     "pxor " #z ", " #z "              \n\t"\

  1529.     "psubw " #a ", " #z "             \n\t"\

  1530.     "pmaxsw " #z ", " #a "            \n\t"

  1531. 

  1532. #define MMABS_SUM_MMX2(a,z, sum)\

  1533.     "pxor " #z ", " #z "              \n\t"\

  1534.     "psubw " #a ", " #z "             \n\t"\

  1535.     "pmaxsw " #z ", " #a "            \n\t"\

  1536.     "paddusw " #a ", " #sum "         \n\t"

  1537. 

  1538. #define TRANSPOSE4(a,b,c,d,t)\

  1539.     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\

  1540.     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\

  1541.     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\

  1542.     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

  1543. 

  1544. #define LOAD4(o, a, b, c, d)\

  1545.         "movq "#o"(%1), " #a "        \n\t"\

  1546.         "movq "#o"+16(%1), " #b "     \n\t"\

  1547.         "movq "#o"+32(%1), " #c "     \n\t"\

  1548.         "movq "#o"+48(%1), " #d "     \n\t"

  1549. 

  1550. #define STORE4(o, a, b, c, d)\

  1551.         "movq "#a", "#o"(%1)          \n\t"\

  1552.         "movq "#b", "#o"+16(%1)       \n\t"\

  1553.         "movq "#c", "#o"+32(%1)       \n\t"\

  1554.         "movq "#d", "#o"+48(%1)       \n\t"\

  1555. 

  1556. static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1557.     DECLARE_ALIGNED_8(uint64_t, temp[16]);

  1558.     int sum=0;

  1559. 

  1560.     assert(h==8);

  1561. 

  1562.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1563. 

  1564.     asm volatile(

  1565.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1566.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1567. 

  1568.         HADAMARD48

  1569. 

  1570.         "movq %%mm7, 112(%1)            \n\t"

  1571. 

  1572.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1573.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1574. 

  1575.         "movq 112(%1), %%mm7            \n\t"

  1576.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1577.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1578. 

  1579.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1580.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1581. 

  1582.         HADAMARD48

  1583. 

  1584.         "movq %%mm7, 120(%1)            \n\t"

  1585. 

  1586.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1587.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1588. 

  1589.         "movq 120(%1), %%mm7            \n\t"

  1590.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1591.         "movq %%mm7, %%mm5              \n\t"//FIXME remove

  1592.         "movq %%mm6, %%mm7              \n\t"

  1593.         "movq %%mm0, %%mm6              \n\t"

  1594. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1595. 

  1596.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1597. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1598. 

  1599.         HADAMARD48

  1600.         "movq %%mm7, 64(%1)             \n\t"

  1601.         MMABS(%%mm0, %%mm7)

  1602.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1603.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1604.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1605.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1606.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1607.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1608.         "movq 64(%1), %%mm1             \n\t"

  1609.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1610.         "movq %%mm0, 64(%1)             \n\t"

  1611. 

  1612.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1613.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1614. 

  1615.         HADAMARD48

  1616.         "movq %%mm7, (%1)               \n\t"

  1617.         MMABS(%%mm0, %%mm7)

  1618.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1619.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1620.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1621.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1622.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1623.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1624.         "movq (%1), %%mm1               \n\t"

  1625.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1626.         "movq 64(%1), %%mm1             \n\t"

  1627.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1628. 

  1629.         "movq %%mm0, %%mm1              \n\t"

  1630.         "psrlq $32, %%mm0               \n\t"

  1631.         "paddusw %%mm1, %%mm0           \n\t"

  1632.         "movq %%mm0, %%mm1              \n\t"

  1633.         "psrlq $16, %%mm0               \n\t"

  1634.         "paddusw %%mm1, %%mm0           \n\t"

  1635.         "movd %%mm0, %0                 \n\t"

  1636. 

  1637.         : "=r" (sum)

  1638.         : "r"(temp)

  1639.     );

  1640.     return sum&0xFFFF;

  1641. }

  1642. 

  1643. static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1644.     DECLARE_ALIGNED_8(uint64_t, temp[16]);

  1645.     int sum=0;

  1646. 

  1647.     assert(h==8);

  1648. 

  1649.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1650. 

  1651.     asm volatile(

  1652.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1653.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1654. 

  1655.         HADAMARD48

  1656. 

  1657.         "movq %%mm7, 112(%1)            \n\t"

  1658. 

  1659.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1660.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1661. 

  1662.         "movq 112(%1), %%mm7            \n\t"

  1663.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1664.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1665. 

  1666.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1667.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1668. 

  1669.         HADAMARD48

  1670. 

  1671.         "movq %%mm7, 120(%1)            \n\t"

  1672. 

  1673.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1674.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1675. 

  1676.         "movq 120(%1), %%mm7            \n\t"

  1677.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1678.         "movq %%mm7, %%mm5              \n\t"//FIXME remove

  1679.         "movq %%mm6, %%mm7              \n\t"

  1680.         "movq %%mm0, %%mm6              \n\t"

  1681. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1682. 

  1683.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1684. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1685. 

  1686.         HADAMARD48

  1687.         "movq %%mm7, 64(%1)             \n\t"

  1688.         MMABS_MMX2(%%mm0, %%mm7)

  1689.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1690.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1691.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1692.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1693.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1694.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1695.         "movq 64(%1), %%mm1             \n\t"

  1696.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1697.         "movq %%mm0, 64(%1)             \n\t"

  1698. 

  1699.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1700.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1701. 

  1702.         HADAMARD48

  1703.         "movq %%mm7, (%1)               \n\t"

  1704.         MMABS_MMX2(%%mm0, %%mm7)

  1705.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1706.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1707.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1708.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1709.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1710.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1711.         "movq (%1), %%mm1               \n\t"

  1712.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1713.         "movq 64(%1), %%mm1             \n\t"

  1714.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1715. 

  1716.         "pshufw $0x0E, %%mm0, %%mm1     \n\t"

  1717.         "paddusw %%mm1, %%mm0           \n\t"

  1718.         "pshufw $0x01, %%mm0, %%mm1     \n\t"

  1719.         "paddusw %%mm1, %%mm0           \n\t"

  1720.         "movd %%mm0, %0                 \n\t"

  1721. 

  1722.         : "=r" (sum)

  1723.         : "r"(temp)

  1724.     );

  1725.     return sum&0xFFFF;

  1726. }

  1727. 

  1728. 

  1729. WARPER8_16_SQ(hadamard8_diff_mmx, hadamard8_diff16_mmx)

  1730. WARPER8_16_SQ(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)

  1731. #endif //CONFIG_ENCODERS

  1732. 

  1733. #define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)

  1734. #define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

  1735. 

  1736. #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

  1737.         "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\

  1738.         "movq "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */\

  1739.         "pmullw " #m3 ", %%mm4            \n\t" /* 20x1 */\

  1740.         "movq "#in7", " #m3 "             \n\t" /* d */\

  1741.         "movq "#in0", %%mm5               \n\t" /* D */\

  1742.         "paddw " #m3 ", %%mm5             \n\t" /* x4 */\

  1743.         "psubw %%mm5, %%mm4               \n\t" /* 20x1 - x4 */\

  1744.         "movq "#in1", %%mm5               \n\t" /* C */\

  1745.         "movq "#in2", %%mm6               \n\t" /* B */\

  1746.         "paddw " #m6 ", %%mm5             \n\t" /* x3 */\

  1747.         "paddw " #m5 ", %%mm6             \n\t" /* x2 */\

  1748.         "paddw %%mm6, %%mm6               \n\t" /* 2x2 */\

  1749.         "psubw %%mm6, %%mm5               \n\t" /* -2x2 + x3 */\

  1750.         "pmullw "MANGLE(ff_pw_3)", %%mm5  \n\t" /* -6x2 + 3x3 */\

  1751.         "paddw " #rnd ", %%mm4            \n\t" /* x2 */\

  1752.         "paddw %%mm4, %%mm5               \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\

  1753.         "psraw $5, %%mm5                  \n\t"\

  1754.         "packuswb %%mm5, %%mm5            \n\t"\

  1755.         OP(%%mm5, out, %%mm7, d)

  1756. 

  1757. #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\

  1758. static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1759.     uint64_t temp;\

  1760. \

  1761.     asm volatile(\

  1762.         "pxor %%mm7, %%mm7                \n\t"\

  1763.         "1:                               \n\t"\

  1764.         "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\

  1765.         "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\

  1766.         "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\

  1767.         "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\

  1768.         "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\

  1769.         "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\

  1770.         "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\

  1771.         "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\

  1772.         "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\

  1773.         "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\

  1774.         "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\

  1775.         "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\

  1776.         "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\

  1777.         "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\

  1778.         "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\

  1779.         "paddw %%mm3, %%mm5               \n\t" /* b */\

  1780.         "paddw %%mm2, %%mm6               \n\t" /* c */\

  1781.         "paddw %%mm5, %%mm5               \n\t" /* 2b */\

  1782.         "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\

  1783.         "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\

  1784.         "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\

  1785.         "paddw %%mm4, %%mm0               \n\t" /* a */\

  1786.         "paddw %%mm1, %%mm5               \n\t" /* d */\

  1787.         "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\

  1788.         "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\

  1789.         "paddw %6, %%mm6                  \n\t"\

  1790.         "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\

  1791.         "psraw $5, %%mm0                  \n\t"\

  1792.         "movq %%mm0, %5                   \n\t"\

  1793.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1794.         \

  1795.         "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\

  1796.         "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\

  1797.         "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\

  1798.         "psrlq $8, %%mm0                  \n\t" /* GHIJKLM0 */\

  1799.         "psrlq $16, %%mm5                 \n\t" /* HIJKLM00 */\

  1800.         "punpcklbw %%mm7, %%mm0           \n\t" /* 0G0H0I0J */\

  1801.         "punpcklbw %%mm7, %%mm5           \n\t" /* 0H0I0J0K */\

  1802.         "paddw %%mm0, %%mm2               \n\t" /* b */\

  1803.         "paddw %%mm5, %%mm3               \n\t" /* c */\

  1804.         "paddw %%mm2, %%mm2               \n\t" /* 2b */\

  1805.         "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\

  1806.         "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\

  1807.         "psrlq $24, %%mm6                 \n\t" /* IJKLM000 */\

  1808.         "punpcklbw %%mm7, %%mm2           \n\t" /* 0F0G0H0I */\

  1809.         "punpcklbw %%mm7, %%mm6           \n\t" /* 0I0J0K0L */\

  1810.         "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\

  1811.         "paddw %%mm2, %%mm1               \n\t" /* a */\

  1812.         "paddw %%mm6, %%mm4               \n\t" /* d */\

  1813.         "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\

  1814.         "psubw %%mm4, %%mm3               \n\t" /* - 6b +3c - d */\

  1815.         "paddw %6, %%mm1                  \n\t"\

  1816.         "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b +3c - d */\

  1817.         "psraw $5, %%mm3                  \n\t"\

  1818.         "movq %5, %%mm1                   \n\t"\

  1819.         "packuswb %%mm3, %%mm1            \n\t"\

  1820.         OP_MMX2(%%mm1, (%1),%%mm4, q)\

  1821.         /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\

  1822.         \

  1823.         "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\

  1824.         "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\

  1825.         "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\

  1826.         "psrlq $8, %%mm1                  \n\t" /* KLMNOPQ0 */\

  1827.         "psrlq $16, %%mm4                 \n\t" /* LMNOPQ00 */\

  1828.         "punpcklbw %%mm7, %%mm1           \n\t" /* 0K0L0M0N */\

  1829.         "punpcklbw %%mm7, %%mm4           \n\t" /* 0L0M0N0O */\

  1830.         "paddw %%mm1, %%mm5               \n\t" /* b */\

  1831.         "paddw %%mm4, %%mm0               \n\t" /* c */\

  1832.         "paddw %%mm5, %%mm5               \n\t" /* 2b */\

  1833.         "psubw %%mm5, %%mm0               \n\t" /* c - 2b */\

  1834.         "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\

  1835.         "psrlq $24, %%mm3                 \n\t" /* MNOPQ000 */\

  1836.         "pmullw "MANGLE(ff_pw_3)", %%mm0  \n\t" /* 3c - 6b */\

  1837.         "punpcklbw %%mm7, %%mm3           \n\t" /* 0M0N0O0P */\

  1838.         "paddw %%mm3, %%mm2               \n\t" /* d */\

  1839.         "psubw %%mm2, %%mm0               \n\t" /* -6b + 3c - d */\

  1840.         "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\

  1841.         "punpcklbw %%mm7, %%mm2           \n\t" /* 0J0K0L0M */\

  1842.         "punpckhbw %%mm7, %%mm5           \n\t" /* 0N0O0P0Q */\

  1843.         "paddw %%mm2, %%mm6               \n\t" /* a */\

  1844.         "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\

  1845.         "paddw %6, %%mm0                  \n\t"\

  1846.         "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\

  1847.         "psraw $5, %%mm0                  \n\t"\

  1848.         /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\

  1849.         \

  1850.         "paddw %%mm5, %%mm3               \n\t" /* a */\

  1851.         "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */\

  1852.         "paddw %%mm4, %%mm6               \n\t" /* b */\

  1853.         "pshufw $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */\

  1854.         "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */\

  1855.         "paddw %%mm1, %%mm4               \n\t" /* c */\

  1856.         "paddw %%mm2, %%mm5               \n\t" /* d */\

  1857.         "paddw %%mm6, %%mm6               \n\t" /* 2b */\

  1858.         "psubw %%mm6, %%mm4               \n\t" /* c - 2b */\

  1859.         "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\

  1860.         "pmullw "MANGLE(ff_pw_3)", %%mm4  \n\t" /* 3c - 6b */\

  1861.         "psubw %%mm5, %%mm3               \n\t" /* -6b + 3c - d */\

  1862.         "paddw %6, %%mm4                  \n\t"\

  1863.         "paddw %%mm3, %%mm4               \n\t" /* 20a - 6b + 3c - d */\

  1864.         "psraw $5, %%mm4                  \n\t"\

  1865.         "packuswb %%mm4, %%mm0            \n\t"\

  1866.         OP_MMX2(%%mm0, 8(%1), %%mm4, q)\

  1867.         \

  1868.         "add %3, %0                       \n\t"\

  1869.         "add %4, %1                       \n\t"\

  1870.         "decl %2                          \n\t"\

  1871.         " jnz 1b                          \n\t"\

  1872.         : "+a"(src), "+c"(dst), "+m"(h)\

  1873.         : "d"((long)srcStride), "S"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1874.         : "memory"\

  1875.     );\

  1876. }\

  1877. \

  1878. static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1879.     int i;\

  1880.     int16_t temp[16];\

  1881.     /* quick HACK, XXX FIXME MUST be optimized */\

  1882.     for(i=0; i<h; i++)\

  1883.     {\

  1884.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1885.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1886.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1887.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1888.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1889.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\

  1890.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\

  1891.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\

  1892.         temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\

  1893.         temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\

  1894.         temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\

  1895.         temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\

  1896.         temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\

  1897.         temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\

  1898.         temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\

  1899.         temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\

  1900.         asm volatile(\

  1901.             "movq (%0), %%mm0               \n\t"\

  1902.             "movq 8(%0), %%mm1              \n\t"\

  1903.             "paddw %2, %%mm0                \n\t"\

  1904.             "paddw %2, %%mm1                \n\t"\

  1905.             "psraw $5, %%mm0                \n\t"\

  1906.             "psraw $5, %%mm1                \n\t"\

  1907.             "packuswb %%mm1, %%mm0          \n\t"\

  1908.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1909.             "movq 16(%0), %%mm0             \n\t"\

  1910.             "movq 24(%0), %%mm1             \n\t"\

  1911.             "paddw %2, %%mm0                \n\t"\

  1912.             "paddw %2, %%mm1                \n\t"\

  1913.             "psraw $5, %%mm0                \n\t"\

  1914.             "psraw $5, %%mm1                \n\t"\

  1915.             "packuswb %%mm1, %%mm0          \n\t"\

  1916.             OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\

  1917.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1918.             : "memory"\

  1919.         );\

  1920.         dst+=dstStride;\

  1921.         src+=srcStride;\

  1922.     }\

  1923. }\

  1924. \

  1925. static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1926.     uint64_t temp;\

  1927. \

  1928.     asm volatile(\

  1929.         "pxor %%mm7, %%mm7                \n\t"\

  1930.         "1:                               \n\t"\

  1931.         "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\

  1932.         "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\

  1933.         "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\

  1934.         "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\

  1935.         "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\

  1936.         "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\

  1937.         "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\

  1938.         "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\

  1939.         "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\

  1940.         "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\

  1941.         "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\

  1942.         "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\

  1943.         "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\

  1944.         "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\

  1945.         "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\

  1946.         "paddw %%mm3, %%mm5               \n\t" /* b */\

  1947.         "paddw %%mm2, %%mm6               \n\t" /* c */\

  1948.         "paddw %%mm5, %%mm5               \n\t" /* 2b */\

  1949.         "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\

  1950.         "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\

  1951.         "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\

  1952.         "paddw %%mm4, %%mm0               \n\t" /* a */\

  1953.         "paddw %%mm1, %%mm5               \n\t" /* d */\

  1954.         "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\

  1955.         "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\

  1956.         "paddw %6, %%mm6                  \n\t"\

  1957.         "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\

  1958.         "psraw $5, %%mm0                  \n\t"\

  1959.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1960.         \

  1961.         "movd 5(%0), %%mm5                \n\t" /* FGHI */\

  1962.         "punpcklbw %%mm7, %%mm5           \n\t" /* 0F0G0H0I */\

  1963.         "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0G0H0I0I */\

  1964.         "paddw %%mm5, %%mm1               \n\t" /* a */\

  1965.         "paddw %%mm6, %%mm2               \n\t" /* b */\

  1966.         "pshufw $0xBE, %%mm5, %%mm6       \n\t" /* 0H0I0I0H */\

  1967.         "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0I0I0H0G */\

  1968.         "paddw %%mm6, %%mm3               \n\t" /* c */\

  1969.         "paddw %%mm5, %%mm4               \n\t" /* d */\

  1970.         "paddw %%mm2, %%mm2               \n\t" /* 2b */\

  1971.         "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\

  1972.         "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\

  1973.         "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\

  1974.         "psubw %%mm4, %%mm3               \n\t" /* -6b + 3c - d */\

  1975.         "paddw %6, %%mm1                  \n\t"\

  1976.         "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\

  1977.         "psraw $5, %%mm3                  \n\t"\

  1978.         "packuswb %%mm3, %%mm0            \n\t"\

  1979.         OP_MMX2(%%mm0, (%1), %%mm4, q)\

  1980.         \

  1981.         "add %3, %0                       \n\t"\

  1982.         "add %4, %1                       \n\t"\

  1983.         "decl %2                          \n\t"\

  1984.         " jnz 1b                          \n\t"\

  1985.         : "+a"(src), "+c"(dst), "+m"(h)\

  1986.         : "S"((long)srcStride), "D"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1987.         : "memory"\

  1988.     );\

  1989. }\

  1990. \

  1991. static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1992.     int i;\

  1993.     int16_t temp[8];\

  1994.     /* quick HACK, XXX FIXME MUST be optimized */\

  1995.     for(i=0; i<h; i++)\

  1996.     {\

  1997.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1998.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1999.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  2000.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  2001.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  2002.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\

  2003.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\

  2004.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\

  2005.         asm volatile(\

  2006.             "movq (%0), %%mm0           \n\t"\

  2007.             "movq 8(%0), %%mm1          \n\t"\

  2008.             "paddw %2, %%mm0            \n\t"\

  2009.             "paddw %2, %%mm1            \n\t"\

  2010.             "psraw $5, %%mm0            \n\t"\

  2011.             "psraw $5, %%mm1            \n\t"\

  2012.             "packuswb %%mm1, %%mm0      \n\t"\

  2013.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  2014.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  2015.             :"memory"\

  2016.         );\

  2017.         dst+=dstStride;\

  2018.         src+=srcStride;\

  2019.     }\

  2020. }

  2021. 

  2022. #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\

  2023. \

  2024. static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2025.     uint64_t temp[17*4];\

  2026.     uint64_t *temp_ptr= temp;\

  2027.     int count= 17;\

  2028. \

  2029.     /*FIXME unroll */\

  2030.     asm volatile(\

  2031.         "pxor %%mm7, %%mm7              \n\t"\

  2032.         "1:                             \n\t"\

  2033.         "movq (%0), %%mm0               \n\t"\

  2034.         "movq (%0), %%mm1               \n\t"\

  2035.         "movq 8(%0), %%mm2              \n\t"\

  2036.         "movq 8(%0), %%mm3              \n\t"\

  2037.         "punpcklbw %%mm7, %%mm0         \n\t"\

  2038.         "punpckhbw %%mm7, %%mm1         \n\t"\

  2039.         "punpcklbw %%mm7, %%mm2         \n\t"\

  2040.         "punpckhbw %%mm7, %%mm3         \n\t"\

  2041.         "movq %%mm0, (%1)               \n\t"\

  2042.         "movq %%mm1, 17*8(%1)           \n\t"\

  2043.         "movq %%mm2, 2*17*8(%1)         \n\t"\

  2044.         "movq %%mm3, 3*17*8(%1)         \n\t"\

  2045.         "add $8, %1                     \n\t"\

  2046.         "add %3, %0                     \n\t"\

  2047.         "decl %2                        \n\t"\

  2048.         " jnz 1b                        \n\t"\

  2049.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  2050.         : "r" ((long)srcStride)\

  2051.         : "memory"\

  2052.     );\

  2053.     \

  2054.     temp_ptr= temp;\

  2055.     count=4;\

  2056.     \

  2057. /*FIXME reorder for speed */\

  2058.     asm volatile(\

  2059.         /*"pxor %%mm7, %%mm7              \n\t"*/\

  2060.         "1:                             \n\t"\

  2061.         "movq (%0), %%mm0               \n\t"\

  2062.         "movq 8(%0), %%mm1              \n\t"\

  2063.         "movq 16(%0), %%mm2             \n\t"\

  2064.         "movq 24(%0), %%mm3             \n\t"\

  2065.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  2066.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  2067.         "add %4, %1                     \n\t"\

  2068.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  2069.         \

  2070.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  2071.         "add %4, %1                     \n\t"\

  2072.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  2073.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\

  2074.         "add %4, %1                     \n\t"\

  2075.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\

  2076.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\

  2077.         "add %4, %1                     \n\t"\

  2078.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\

  2079.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\

  2080.         "add %4, %1                     \n\t"\

  2081.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\

  2082.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\

  2083.         "add %4, %1                     \n\t"\

  2084.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\

  2085.         \

  2086.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\

  2087.         "add %4, %1                     \n\t"  \

  2088.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\

  2089.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\

  2090.         \

  2091.         "add $136, %0                   \n\t"\

  2092.         "add %6, %1                     \n\t"\

  2093.         "decl %2                        \n\t"\

  2094.         " jnz 1b                        \n\t"\

  2095.         \

  2096.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2097.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(long)dstStride)\

  2098.         :"memory"\

  2099.     );\

  2100. }\

  2101. \

  2102. static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2103.     uint64_t temp[9*2];\

  2104.     uint64_t *temp_ptr= temp;\

  2105.     int count= 9;\

  2106. \

  2107.     /*FIXME unroll */\

  2108.     asm volatile(\

  2109.         "pxor %%mm7, %%mm7              \n\t"\

  2110.         "1:                             \n\t"\

  2111.         "movq (%0), %%mm0               \n\t"\

  2112.         "movq (%0), %%mm1               \n\t"\

  2113.         "punpcklbw %%mm7, %%mm0         \n\t"\

  2114.         "punpckhbw %%mm7, %%mm1         \n\t"\

  2115.         "movq %%mm0, (%1)               \n\t"\

  2116.         "movq %%mm1, 9*8(%1)            \n\t"\

  2117.         "add $8, %1                     \n\t"\

  2118.         "add %3, %0                     \n\t"\

  2119.         "decl %2                        \n\t"\

  2120.         " jnz 1b                        \n\t"\

  2121.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  2122.         : "r" ((long)srcStride)\

  2123.         : "memory"\

  2124.     );\

  2125.     \

  2126.     temp_ptr= temp;\

  2127.     count=2;\

  2128.     \

  2129. /*FIXME reorder for speed */\

  2130.     asm volatile(\

  2131.         /*"pxor %%mm7, %%mm7              \n\t"*/\

  2132.         "1:                             \n\t"\

  2133.         "movq (%0), %%mm0               \n\t"\

  2134.         "movq 8(%0), %%mm1              \n\t"\

  2135.         "movq 16(%0), %%mm2             \n\t"\

  2136.         "movq 24(%0), %%mm3             \n\t"\

  2137.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  2138.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  2139.         "add %4, %1                     \n\t"\

  2140.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  2141.         \

  2142.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  2143.         "add %4, %1                     \n\t"\

  2144.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  2145.         \

  2146.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\

  2147.         "add %4, %1                     \n\t"\

  2148.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\

  2149.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\

  2150.                 \

  2151.         "add $72, %0                    \n\t"\

  2152.         "add %6, %1                     \n\t"\

  2153.         "decl %2                        \n\t"\

  2154.         " jnz 1b                        \n\t"\

  2155.          \

  2156.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2157.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(long)dstStride)\

  2158.         : "memory"\

  2159.    );\

  2160. }\

  2161. \

  2162. static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2163.     OPNAME ## pixels8_mmx(dst, src, stride, 8);\

  2164. }\

  2165. \

  2166. static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2167.     uint64_t temp[8];\

  2168.     uint8_t * const half= (uint8_t*)temp;\

  2169.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2170.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2171. }\

  2172. \

  2173. static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2174.     OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\

  2175. }\

  2176. \

  2177. static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2178.     uint64_t temp[8];\

  2179.     uint8_t * const half= (uint8_t*)temp;\

  2180.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2181.     OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\

  2182. }\

  2183. \

  2184. static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2185.     uint64_t temp[8];\

  2186.     uint8_t * const half= (uint8_t*)temp;\

  2187.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2188.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2189. }\

  2190. \

  2191. static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2192.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2193. }\

  2194. \

  2195. static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2196.     uint64_t temp[8];\

  2197.     uint8_t * const half= (uint8_t*)temp;\

  2198.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2199.     OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\

  2200. }\

  2201. static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2202.     uint64_t half[8 + 9];\

  2203.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2204.     uint8_t * const halfHV= ((uint8_t*)half);\

  2205.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2206.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2207.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2208.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2209. }\

  2210. static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2211.     uint64_t half[8 + 9];\

  2212.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2213.     uint8_t * const halfHV= ((uint8_t*)half);\

  2214.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2215.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2216.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2217.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2218. }\

  2219. static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2220.     uint64_t half[8 + 9];\

  2221.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2222.     uint8_t * const halfHV= ((uint8_t*)half);\

  2223.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2224.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2225.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2226.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2227. }\

  2228. static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2229.     uint64_t half[8 + 9];\

  2230.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2231.     uint8_t * const halfHV= ((uint8_t*)half);\

  2232.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2233.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2234.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2235.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2236. }\

  2237. static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2238.     uint64_t half[8 + 9];\

  2239.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2240.     uint8_t * const halfHV= ((uint8_t*)half);\

  2241.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2242.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2243.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2244. }\

  2245. static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2246.     uint64_t half[8 + 9];\

  2247.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2248.     uint8_t * const halfHV= ((uint8_t*)half);\

  2249.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2250.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2251.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2252. }\

  2253. static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2254.     uint64_t half[8 + 9];\

  2255.     uint8_t * const halfH= ((uint8_t*)half);\

  2256.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2257.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2258.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2259. }\

  2260. static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2261.     uint64_t half[8 + 9];\

  2262.     uint8_t * const halfH= ((uint8_t*)half);\

  2263.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2264.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2265.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2266. }\

  2267. static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2268.     uint64_t half[9];\

  2269.     uint8_t * const halfH= ((uint8_t*)half);\

  2270.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2271.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2272. }\

  2273. static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2274.     OPNAME ## pixels16_mmx(dst, src, stride, 16);\

  2275. }\

  2276. \

  2277. static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2278.     uint64_t temp[32];\

  2279.     uint8_t * const half= (uint8_t*)temp;\

  2280.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2281.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2282. }\

  2283. \

  2284. static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2285.     OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\

  2286. }\

  2287. \

  2288. static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2289.     uint64_t temp[32];\

  2290.     uint8_t * const half= (uint8_t*)temp;\

  2291.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2292.     OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\

  2293. }\

  2294. \

  2295. static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2296.     uint64_t temp[32];\

  2297.     uint8_t * const half= (uint8_t*)temp;\

  2298.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2299.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2300. }\

  2301. \

  2302. static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2303.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2304. }\

  2305. \

  2306. static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2307.     uint64_t temp[32];\

  2308.     uint8_t * const half= (uint8_t*)temp;\

  2309.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2310.     OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\

  2311. }\

  2312. static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2313.     uint64_t half[16*2 + 17*2];\

  2314.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2315.     uint8_t * const halfHV= ((uint8_t*)half);\

  2316.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2317.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2318.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2319.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2320. }\

  2321. static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2322.     uint64_t half[16*2 + 17*2];\

  2323.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2324.     uint8_t * const halfHV= ((uint8_t*)half);\

  2325.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2326.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2327.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2328.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2329. }\

  2330. static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2331.     uint64_t half[16*2 + 17*2];\

  2332.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2333.     uint8_t * const halfHV= ((uint8_t*)half);\

  2334.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2335.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2336.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2337.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2338. }\

  2339. static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2340.     uint64_t half[16*2 + 17*2];\

  2341.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2342.     uint8_t * const halfHV= ((uint8_t*)half);\

  2343.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2344.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2345.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2346.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2347. }\

  2348. static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2349.     uint64_t half[16*2 + 17*2];\

  2350.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2351.     uint8_t * const halfHV= ((uint8_t*)half);\

  2352.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2353.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2354.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2355. }\

  2356. static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2357.     uint64_t half[16*2 + 17*2];\

  2358.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2359.     uint8_t * const halfHV= ((uint8_t*)half);\

  2360.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2361.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2362.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2363. }\

  2364. static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2365.     uint64_t half[17*2];\

  2366.     uint8_t * const halfH= ((uint8_t*)half);\

  2367.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2368.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2369.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2370. }\

  2371. static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2372.     uint64_t half[17*2];\

  2373.     uint8_t * const halfH= ((uint8_t*)half);\

  2374.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2375.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2376.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2377. }\

  2378. static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2379.     uint64_t half[17*2];\

  2380.     uint8_t * const halfH= ((uint8_t*)half);\

  2381.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2382.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2383. }

  2384. 

  2385. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"

  2386. #define AVG_3DNOW_OP(a,b,temp, size) \

  2387. "mov" #size " " #b ", " #temp "   \n\t"\

  2388. "pavgusb " #temp ", " #a "        \n\t"\

  2389. "mov" #size " " #a ", " #b "      \n\t"

  2390. #define AVG_MMX2_OP(a,b,temp, size) \

  2391. "mov" #size " " #b ", " #temp "   \n\t"\

  2392. "pavgb " #temp ", " #a "          \n\t"\

  2393. "mov" #size " " #a ", " #b "      \n\t"

  2394. 

  2395. QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)

  2396. QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)

  2397. QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)

  2398. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)

  2399. QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)

  2400. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)

  2401. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)

  2402. QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)

  2403. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

  2404. 

  2405. /***********************************/

  2406. /* bilinear qpel: not compliant to any spec, only for -lavdopts fast */

  2407. 

  2408. #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\

  2409. static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2410.     OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\

  2411. }

  2412. #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\

  2413. static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2414.     OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\

  2415. }

  2416. 

  2417. #define QPEL_2TAP(OPNAME, SIZE, MMX)\

  2418. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\

  2419. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\

  2420. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\

  2421. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\

  2422.                           OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\

  2423. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\

  2424.                           OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\

  2425. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\

  2426.                           OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\

  2427. static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2428.     OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\

  2429. }\

  2430. static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2431.     OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\

  2432. }\

  2433. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,         1,       0)\

  2434. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,        -1,       0)\

  2435. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,         stride,  0)\

  2436. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,   -stride,  0)\

  2437. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,         stride,  1)\

  2438. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,         stride, -1)\

  2439. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,   -stride,  1)\

  2440. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\

  2441. 

  2442. QPEL_2TAP(put_, 16, mmx2)

  2443. QPEL_2TAP(avg_, 16, mmx2)

  2444. QPEL_2TAP(put_,  8, mmx2)

  2445. QPEL_2TAP(avg_,  8, mmx2)

  2446. QPEL_2TAP(put_, 16, 3dnow)

  2447. QPEL_2TAP(avg_, 16, 3dnow)

  2448. QPEL_2TAP(put_,  8, 3dnow)

  2449. QPEL_2TAP(avg_,  8, 3dnow)

  2450. 

  2451. 

  2452. #if 0

  2453. static void just_return() { return; }

  2454. #endif

  2455. 

  2456. #define SET_QPEL_FUNC(postfix1, postfix2) \

  2457.     c->put_ ## postfix1 = put_ ## postfix2;\

  2458.     c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\

  2459.     c->avg_ ## postfix1 = avg_ ## postfix2;

  2460. 

  2461. static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  2462.                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){

  2463.     const int w = 8;

  2464.     const int ix = ox>>(16+shift);

  2465.     const int iy = oy>>(16+shift);

  2466.     const int oxs = ox>>4;

  2467.     const int oys = oy>>4;

  2468.     const int dxxs = dxx>>4;

  2469.     const int dxys = dxy>>4;

  2470.     const int dyxs = dyx>>4;

  2471.     const int dyys = dyy>>4;

  2472.     const uint16_t r4[4] = {r,r,r,r};

  2473.     const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};

  2474.     const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};

  2475.     const uint64_t shift2 = 2*shift;

  2476.     uint8_t edge_buf[(h+1)*stride];

  2477.     int x, y;

  2478. 

  2479.     const int dxw = (dxx-(1<<(16+shift)))*(w-1);

  2480.     const int dyh = (dyy-(1<<(16+shift)))*(h-1);

  2481.     const int dxh = dxy*(h-1);

  2482.     const int dyw = dyx*(w-1);

  2483.     if( // non-constant fullpel offset (3% of blocks)

  2484.         (ox^(ox+dxw) | ox^(ox+dxh) | ox^(ox+dxw+dxh) |

  2485.          oy^(oy+dyw) | oy^(oy+dyh) | oy^(oy+dyw+dyh)) >> (16+shift)

  2486.         // uses more than 16 bits of subpel mv (only at huge resolution)

  2487.         || (dxx|dxy|dyx|dyy)&15 )

  2488.     {

  2489.         //FIXME could still use mmx for some of the rows

  2490.         ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);

  2491.         return;

  2492.     }

  2493. 

  2494.     src += ix + iy*stride;

  2495.     if( (unsigned)ix >= width-w ||

  2496.         (unsigned)iy >= height-h )

  2497.     {

  2498.         ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);

  2499.         src = edge_buf;

  2500.     }

  2501. 

  2502.     asm volatile(

  2503.         "movd         %0, %%mm6 \n\t"

  2504.         "pxor      %%mm7, %%mm7 \n\t"

  2505.         "punpcklwd %%mm6, %%mm6 \n\t"

  2506.         "punpcklwd %%mm6, %%mm6 \n\t"

  2507.         :: "r"(1<<shift)

  2508.     );

  2509. 

  2510.     for(x=0; x<w; x+=4){

  2511.         uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),

  2512.                             oxs - dxys + dxxs*(x+1),

  2513.                             oxs - dxys + dxxs*(x+2),

  2514.                             oxs - dxys + dxxs*(x+3) };

  2515.         uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),

  2516.                             oys - dyys + dyxs*(x+1),

  2517.                             oys - dyys + dyxs*(x+2),

  2518.                             oys - dyys + dyxs*(x+3) };

  2519. 

  2520.         for(y=0; y<h; y++){

  2521.             asm volatile(

  2522.                 "movq   %0,  %%mm4 \n\t"

  2523.                 "movq   %1,  %%mm5 \n\t"

  2524.                 "paddw  %2,  %%mm4 \n\t"

  2525.                 "paddw  %3,  %%mm5 \n\t"

  2526.                 "movq   %%mm4, %0  \n\t"

  2527.                 "movq   %%mm5, %1  \n\t"

  2528.                 "psrlw  $12, %%mm4 \n\t"

  2529.                 "psrlw  $12, %%mm5 \n\t"

  2530.                 : "+m"(*dx4), "+m"(*dy4)

  2531.                 : "m"(*dxy4), "m"(*dyy4)

  2532.             );

  2533. 

  2534.             asm volatile(

  2535.                 "movq   %%mm6, %%mm2 \n\t"

  2536.                 "movq   %%mm6, %%mm1 \n\t"

  2537.                 "psubw  %%mm4, %%mm2 \n\t"

  2538.                 "psubw  %%mm5, %%mm1 \n\t"

  2539.                 "movq   %%mm2, %%mm0 \n\t"

  2540.                 "movq   %%mm4, %%mm3 \n\t"

  2541.                 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)

  2542.                 "pmullw %%mm5, %%mm3 \n\t" // dx*dy

  2543.                 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy

  2544.                 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)

  2545. 

  2546.                 "movd   %4,    %%mm5 \n\t"

  2547.                 "movd   %3,    %%mm4 \n\t"

  2548.                 "punpcklbw %%mm7, %%mm5 \n\t"

  2549.                 "punpcklbw %%mm7, %%mm4 \n\t"

  2550.                 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy

  2551.                 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy

  2552. 

  2553.                 "movd   %2,    %%mm5 \n\t"

  2554.                 "movd   %1,    %%mm4 \n\t"

  2555.                 "punpcklbw %%mm7, %%mm5 \n\t"

  2556.                 "punpcklbw %%mm7, %%mm4 \n\t"

  2557.                 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)

  2558.                 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)

  2559.                 "paddw  %5,    %%mm1 \n\t"

  2560.                 "paddw  %%mm3, %%mm2 \n\t"

  2561.                 "paddw  %%mm1, %%mm0 \n\t"

  2562.                 "paddw  %%mm2, %%mm0 \n\t"

  2563. 

  2564.                 "psrlw    %6,    %%mm0 \n\t"

  2565.                 "packuswb %%mm0, %%mm0 \n\t"

  2566.                 "movd     %%mm0, %0    \n\t"

  2567. 

  2568.                 : "=m"(dst[x+y*stride])

  2569.                 : "m"(src[0]), "m"(src[1]),

  2570.                   "m"(src[stride]), "m"(src[stride+1]),

  2571.                   "m"(*r4), "m"(shift2)

  2572.             );

  2573.             src += stride;

  2574.         }

  2575.         src += 4-h*stride;

  2576.     }

  2577. }

  2578. 

  2579. #ifdef CONFIG_ENCODERS

  2580. static int try_8x8basis_mmx(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){

  2581.     long i=0;

  2582. 

  2583.     assert(ABS(scale) < 256);

  2584.     scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2585. 

  2586.     asm volatile(

  2587.         "pcmpeqw %%mm6, %%mm6           \n\t" // -1w

  2588.         "psrlw $15, %%mm6               \n\t" //  1w

  2589.         "pxor %%mm7, %%mm7              \n\t"

  2590.         "movd  %4, %%mm5                \n\t"

  2591.         "punpcklwd %%mm5, %%mm5         \n\t"

  2592.         "punpcklwd %%mm5, %%mm5         \n\t"

  2593.         "1:                             \n\t"

  2594.         "movq  (%1, %0), %%mm0          \n\t"

  2595.         "movq  8(%1, %0), %%mm1         \n\t"

  2596.         "pmulhw %%mm5, %%mm0            \n\t"

  2597.         "pmulhw %%mm5, %%mm1            \n\t"

  2598.         "paddw %%mm6, %%mm0             \n\t"

  2599.         "paddw %%mm6, %%mm1             \n\t"

  2600.         "psraw $1, %%mm0                \n\t"

  2601.         "psraw $1, %%mm1                \n\t"

  2602.         "paddw (%2, %0), %%mm0          \n\t"

  2603.         "paddw 8(%2, %0), %%mm1         \n\t"

  2604.         "psraw $6, %%mm0                \n\t"

  2605.         "psraw $6, %%mm1                \n\t"

  2606.         "pmullw (%3, %0), %%mm0         \n\t"

  2607.         "pmullw 8(%3, %0), %%mm1        \n\t"

  2608.         "pmaddwd %%mm0, %%mm0           \n\t"

  2609.         "pmaddwd %%mm1, %%mm1           \n\t"

  2610.         "paddd %%mm1, %%mm0             \n\t"

  2611.         "psrld $4, %%mm0                \n\t"

  2612.         "paddd %%mm0, %%mm7             \n\t"

  2613.         "add $16, %0                    \n\t"

  2614.         "cmp $128, %0                   \n\t" //FIXME optimize & bench

  2615.         " jb 1b                         \n\t"

  2616.         "movq %%mm7, %%mm6              \n\t"

  2617.         "psrlq $32, %%mm7               \n\t"

  2618.         "paddd %%mm6, %%mm7             \n\t"

  2619.         "psrld $2, %%mm7                \n\t"

  2620.         "movd %%mm7, %0                 \n\t"

  2621. 

  2622.         : "+r" (i)

  2623.         : "r"(basis), "r"(rem), "r"(weight), "g"(scale)

  2624.     );

  2625.     return i;

  2626. }

  2627. 

  2628. static void add_8x8basis_mmx(int16_t rem[64], int16_t basis[64], int scale){

  2629.     long i=0;

  2630. 

  2631.     if(ABS(scale) < 256){

  2632.         scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2633.         asm volatile(

  2634.                 "pcmpeqw %%mm6, %%mm6   \n\t" // -1w

  2635.                 "psrlw $15, %%mm6       \n\t" //  1w

  2636.                 "movd  %3, %%mm5        \n\t"

  2637.                 "punpcklwd %%mm5, %%mm5 \n\t"

  2638.                 "punpcklwd %%mm5, %%mm5 \n\t"

  2639.                 "1:                     \n\t"

  2640.                 "movq  (%1, %0), %%mm0  \n\t"

  2641.                 "movq  8(%1, %0), %%mm1 \n\t"

  2642.                 "pmulhw %%mm5, %%mm0    \n\t"

  2643.                 "pmulhw %%mm5, %%mm1    \n\t"

  2644.                 "paddw %%mm6, %%mm0     \n\t"

  2645.                 "paddw %%mm6, %%mm1     \n\t"

  2646.                 "psraw $1, %%mm0        \n\t"

  2647.                 "psraw $1, %%mm1        \n\t"

  2648.                 "paddw (%2, %0), %%mm0  \n\t"

  2649.                 "paddw 8(%2, %0), %%mm1 \n\t"

  2650.                 "movq %%mm0, (%2, %0)   \n\t"

  2651.                 "movq %%mm1, 8(%2, %0)  \n\t"

  2652.                 "add $16, %0            \n\t"

  2653.                 "cmp $128, %0           \n\t" //FIXME optimize & bench

  2654.                 " jb 1b                 \n\t"

  2655. 

  2656.                 : "+r" (i)

  2657.                 : "r"(basis), "r"(rem), "g"(scale)

  2658.         );

  2659.     }else{

  2660.         for(i=0; i<8*8; i++){

  2661.             rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);

  2662.         }

  2663.     }

  2664. }

  2665. #endif /* CONFIG_ENCODERS */

  2666. 

  2667. #define PREFETCH(name, op) \

  2668. void name(void *mem, int stride, int h){\

  2669.     const uint8_t *p= mem;\

  2670.     do{\

  2671.         asm volatile(#op" %0" :: "m"(*p));\

  2672.         p+= stride;\

  2673.     }while(--h);\

  2674. }

  2675. PREFETCH(prefetch_mmx2,  prefetcht0)

  2676. PREFETCH(prefetch_3dnow, prefetch)

  2677. #undef PREFETCH

  2678. 

  2679. #include "h264dsp_mmx.c"

  2680. 

  2681. /* AVS specific */

  2682. void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx);

  2683. 

  2684. void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2685.     put_pixels8_mmx(dst, src, stride, 8);

  2686. }

  2687. void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2688.     avg_pixels8_mmx(dst, src, stride, 8);

  2689. }

  2690. void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2691.     put_pixels16_mmx(dst, src, stride, 16);

  2692. }

  2693. void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2694.     avg_pixels16_mmx(dst, src, stride, 16);

  2695. }

  2696. 

  2697. /* external functions, from idct_mmx.c */

  2698. void ff_mmx_idct(DCTELEM *block);

  2699. void ff_mmxext_idct(DCTELEM *block);

  2700. 

  2701. void ff_vp3_idct_sse2(int16_t *input_data);

  2702. void ff_vp3_idct_mmx(int16_t *data);

  2703. void ff_vp3_dsp_init_mmx(void);

  2704. 

  2705. /* XXX: those functions should be suppressed ASAP when all IDCTs are

  2706.    converted */

  2707. static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2708. {

  2709.     ff_mmx_idct (block);

  2710.     put_pixels_clamped_mmx(block, dest, line_size);

  2711. }

  2712. static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2713. {

  2714.     ff_mmx_idct (block);

  2715.     add_pixels_clamped_mmx(block, dest, line_size);

  2716. }

  2717. static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2718. {

  2719.     ff_mmxext_idct (block);

  2720.     put_pixels_clamped_mmx(block, dest, line_size);

  2721. }

  2722. static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2723. {

  2724.     ff_mmxext_idct (block);

  2725.     add_pixels_clamped_mmx(block, dest, line_size);

  2726. }

  2727. static void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block)

  2728. {

  2729.     ff_vp3_idct_sse2(block);

  2730.     put_signed_pixels_clamped_mmx(block, dest, line_size);

  2731. }

  2732. static void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block)

  2733. {

  2734.     ff_vp3_idct_sse2(block);

  2735.     add_pixels_clamped_mmx(block, dest, line_size);

  2736. }

  2737. static void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)

  2738. {

  2739.     ff_vp3_idct_mmx(block);

  2740.     put_signed_pixels_clamped_mmx(block, dest, line_size);

  2741. }

  2742. static void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)

  2743. {

  2744.     ff_vp3_idct_mmx(block);

  2745.     add_pixels_clamped_mmx(block, dest, line_size);

  2746. }

  2747. static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)

  2748. {

  2749.     ff_idct_xvid_mmx (block);

  2750.     put_pixels_clamped_mmx(block, dest, line_size);

  2751. }

  2752. static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)

  2753. {

  2754.     ff_idct_xvid_mmx (block);

  2755.     add_pixels_clamped_mmx(block, dest, line_size);

  2756. }

  2757. static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)

  2758. {

  2759.     ff_idct_xvid_mmx2 (block);

  2760.     put_pixels_clamped_mmx(block, dest, line_size);

  2761. }

  2762. static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)

  2763. {

  2764.     ff_idct_xvid_mmx2 (block);

  2765.     add_pixels_clamped_mmx(block, dest, line_size);

  2766. }

  2767. 

  2768. static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)

  2769. {

  2770.     int i;

  2771.     asm volatile("pxor %%mm7, %%mm7":);

  2772.     for(i=0; i<blocksize; i+=2) {

  2773.         asm volatile(

  2774.             "movq    %0,    %%mm0 \n\t"

  2775.             "movq    %1,    %%mm1 \n\t"

  2776.             "movq    %%mm0, %%mm2 \n\t"

  2777.             "movq    %%mm1, %%mm3 \n\t"

  2778.             "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0

  2779.             "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0

  2780.             "pslld   $31,   %%mm2 \n\t" // keep only the sign bit

  2781.             "pxor    %%mm2, %%mm1 \n\t"

  2782.             "movq    %%mm3, %%mm4 \n\t"

  2783.             "pand    %%mm1, %%mm3 \n\t"

  2784.             "pandn   %%mm1, %%mm4 \n\t"

  2785.             "pfadd   %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))

  2786.             "pfsub   %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))

  2787.             "movq    %%mm3, %1    \n\t"

  2788.             "movq    %%mm0, %0    \n\t"

  2789.             :"+m"(mag[i]), "+m"(ang[i])

  2790.             ::"memory"

  2791.         );

  2792.     }

  2793.     asm volatile("femms");

  2794. }

  2795. static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)

  2796. {

  2797.     int i;

  2798. 

  2799.     asm volatile(

  2800.             "movaps  %0,     %%xmm5 \n\t"

  2801.         ::"m"(ff_pdw_80000000[0])

  2802.     );

  2803.     for(i=0; i<blocksize; i+=4) {

  2804.         asm volatile(

  2805.             "movaps  %0,     %%xmm0 \n\t"

  2806.             "movaps  %1,     %%xmm1 \n\t"

  2807.             "xorps   %%xmm2, %%xmm2 \n\t"

  2808.             "xorps   %%xmm3, %%xmm3 \n\t"

  2809.             "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0

  2810.             "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0

  2811.             "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit

  2812.             "xorps   %%xmm2, %%xmm1 \n\t"

  2813.             "movaps  %%xmm3, %%xmm4 \n\t"

  2814.             "andps   %%xmm1, %%xmm3 \n\t"

  2815.             "andnps  %%xmm1, %%xmm4 \n\t"

  2816.             "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))

  2817.             "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))

  2818.             "movaps  %%xmm3, %1     \n\t"

  2819.             "movaps  %%xmm0, %0     \n\t"

  2820.             :"+m"(mag[i]), "+m"(ang[i])

  2821.             ::"memory"

  2822.         );

  2823.     }

  2824. }

  2825. 

  2826. static void vector_fmul_3dnow(float *dst, const float *src, int len){

  2827.     long i = (len-4)*4;

  2828.     asm volatile(

  2829.         "1: \n\t"

  2830.         "movq    (%1,%0), %%mm0 \n\t"

  2831.         "movq   8(%1,%0), %%mm1 \n\t"

  2832.         "pfmul   (%2,%0), %%mm0 \n\t"

  2833.         "pfmul  8(%2,%0), %%mm1 \n\t"

  2834.         "movq   %%mm0,  (%1,%0) \n\t"

  2835.         "movq   %%mm1, 8(%1,%0) \n\t"

  2836.         "sub  $16, %0 \n\t"

  2837.         "jge 1b \n\t"

  2838.         "femms  \n\t"

  2839.         :"+r"(i)

  2840.         :"r"(dst), "r"(src)

  2841.         :"memory"

  2842.     );

  2843. }

  2844. static void vector_fmul_sse(float *dst, const float *src, int len){

  2845.     long i = (len-8)*4;

  2846.     asm volatile(

  2847.         "1: \n\t"

  2848.         "movaps    (%1,%0), %%xmm0 \n\t"

  2849.         "movaps  16(%1,%0), %%xmm1 \n\t"

  2850.         "mulps     (%2,%0), %%xmm0 \n\t"

  2851.         "mulps   16(%2,%0), %%xmm1 \n\t"

  2852.         "movaps  %%xmm0,   (%1,%0) \n\t"

  2853.         "movaps  %%xmm1, 16(%1,%0) \n\t"

  2854.         "sub  $32, %0 \n\t"

  2855.         "jge 1b \n\t"

  2856.         :"+r"(i)

  2857.         :"r"(dst), "r"(src)

  2858.         :"memory"

  2859.     );

  2860. }

  2861. 

  2862. static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){

  2863.     long i = len*4-16;

  2864.     asm volatile(

  2865.         "1: \n\t"

  2866.         "pswapd   8(%1), %%mm0 \n\t"

  2867.         "pswapd    (%1), %%mm1 \n\t"

  2868.         "pfmul  (%3,%0), %%mm0 \n\t"

  2869.         "pfmul 8(%3,%0), %%mm1 \n\t"

  2870.         "movq  %%mm0,  (%2,%0) \n\t"

  2871.         "movq  %%mm1, 8(%2,%0) \n\t"

  2872.         "add   $16, %1 \n\t"

  2873.         "sub   $16, %0 \n\t"

  2874.         "jge   1b \n\t"

  2875.         :"+r"(i), "+r"(src1)

  2876.         :"r"(dst), "r"(src0)

  2877.     );

  2878.     asm volatile("femms");

  2879. }

  2880. static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){

  2881.     long i = len*4-32;

  2882.     asm volatile(

  2883.         "1: \n\t"

  2884.         "movaps        16(%1), %%xmm0 \n\t"

  2885.         "movaps          (%1), %%xmm1 \n\t"

  2886.         "shufps $0x1b, %%xmm0, %%xmm0 \n\t"

  2887.         "shufps $0x1b, %%xmm1, %%xmm1 \n\t"

  2888.         "mulps        (%3,%0), %%xmm0 \n\t"

  2889.         "mulps      16(%3,%0), %%xmm1 \n\t"

  2890.         "movaps     %%xmm0,   (%2,%0) \n\t"

  2891.         "movaps     %%xmm1, 16(%2,%0) \n\t"

  2892.         "add    $32, %1 \n\t"

  2893.         "sub    $32, %0 \n\t"

  2894.         "jge    1b \n\t"

  2895.         :"+r"(i), "+r"(src1)

  2896.         :"r"(dst), "r"(src0)

  2897.     );

  2898. }

  2899. 

  2900. static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1,

  2901.                                       const float *src2, int src3, int len, int step){

  2902.     long i = (len-4)*4;

  2903.     if(step == 2 && src3 == 0){

  2904.         dst += (len-4)*2;

  2905.         asm volatile(

  2906.             "1: \n\t"

  2907.             "movq   (%2,%0),  %%mm0 \n\t"

  2908.             "movq  8(%2,%0),  %%mm1 \n\t"

  2909.             "pfmul  (%3,%0),  %%mm0 \n\t"

  2910.             "pfmul 8(%3,%0),  %%mm1 \n\t"

  2911.             "pfadd  (%4,%0),  %%mm0 \n\t"

  2912.             "pfadd 8(%4,%0),  %%mm1 \n\t"

  2913.             "movd     %%mm0,   (%1) \n\t"

  2914.             "movd     %%mm1, 16(%1) \n\t"

  2915.             "psrlq      $32,  %%mm0 \n\t"

  2916.             "psrlq      $32,  %%mm1 \n\t"

  2917.             "movd     %%mm0,  8(%1) \n\t"

  2918.             "movd     %%mm1, 24(%1) \n\t"

  2919.             "sub  $32, %1 \n\t"

  2920.             "sub  $16, %0 \n\t"

  2921.             "jge  1b \n\t"

  2922.             :"+r"(i), "+r"(dst)

  2923.             :"r"(src0), "r"(src1), "r"(src2)

  2924.             :"memory"

  2925.         );

  2926.     }

  2927.     else if(step == 1 && src3 == 0){

  2928.         asm volatile(

  2929.             "1: \n\t"

  2930.             "movq    (%2,%0), %%mm0 \n\t"

  2931.             "movq   8(%2,%0), %%mm1 \n\t"

  2932.             "pfmul   (%3,%0), %%mm0 \n\t"

  2933.             "pfmul  8(%3,%0), %%mm1 \n\t"

  2934.             "pfadd   (%4,%0), %%mm0 \n\t"

  2935.             "pfadd  8(%4,%0), %%mm1 \n\t"

  2936.             "movq  %%mm0,   (%1,%0) \n\t"

  2937.             "movq  %%mm1,  8(%1,%0) \n\t"

  2938.             "sub  $16, %0 \n\t"

  2939.             "jge  1b \n\t"

  2940.             :"+r"(i)

  2941.             :"r"(dst), "r"(src0), "r"(src1), "r"(src2)

  2942.             :"memory"

  2943.         );

  2944.     }

  2945.     else

  2946.         ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);

  2947.     asm volatile("femms");

  2948. }

  2949. static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *src1,

  2950.                                     const float *src2, int src3, int len, int step){

  2951.     long i = (len-8)*4;

  2952.     if(step == 2 && src3 == 0){

  2953.         dst += (len-8)*2;

  2954.         asm volatile(

  2955.             "1: \n\t"

  2956.             "movaps   (%2,%0), %%xmm0 \n\t"

  2957.             "movaps 16(%2,%0), %%xmm1 \n\t"

  2958.             "mulps    (%3,%0), %%xmm0 \n\t"

  2959.             "mulps  16(%3,%0), %%xmm1 \n\t"

  2960.             "addps    (%4,%0), %%xmm0 \n\t"

  2961.             "addps  16(%4,%0), %%xmm1 \n\t"

  2962.             "movss     %%xmm0,   (%1) \n\t"

  2963.             "movss     %%xmm1, 32(%1) \n\t"

  2964.             "movhlps   %%xmm0, %%xmm2 \n\t"

  2965.             "movhlps   %%xmm1, %%xmm3 \n\t"

  2966.             "movss     %%xmm2, 16(%1) \n\t"

  2967.             "movss     %%xmm3, 48(%1) \n\t"

  2968.             "shufps $0xb1, %%xmm0, %%xmm0 \n\t"

  2969.             "shufps $0xb1, %%xmm1, %%xmm1 \n\t"

  2970.             "movss     %%xmm0,  8(%1) \n\t"

  2971.             "movss     %%xmm1, 40(%1) \n\t"

  2972.             "movhlps   %%xmm0, %%xmm2 \n\t"

  2973.             "movhlps   %%xmm1, %%xmm3 \n\t"

  2974.             "movss     %%xmm2, 24(%1) \n\t"

  2975.             "movss     %%xmm3, 56(%1) \n\t"

  2976.             "sub  $64, %1 \n\t"

  2977.             "sub  $32, %0 \n\t"

  2978.             "jge  1b \n\t"

  2979.             :"+r"(i), "+r"(dst)

  2980.             :"r"(src0), "r"(src1), "r"(src2)

  2981.             :"memory"

  2982.         );

  2983.     }

  2984.     else if(step == 1 && src3 == 0){

  2985.         asm volatile(

  2986.             "1: \n\t"

  2987.             "movaps   (%2,%0), %%xmm0 \n\t"

  2988.             "movaps 16(%2,%0), %%xmm1 \n\t"

  2989.             "mulps    (%3,%0), %%xmm0 \n\t"

  2990.             "mulps  16(%3,%0), %%xmm1 \n\t"

  2991.             "addps    (%4,%0), %%xmm0 \n\t"

  2992.             "addps  16(%4,%0), %%xmm1 \n\t"

  2993.             "movaps %%xmm0,   (%1,%0) \n\t"

  2994.             "movaps %%xmm1, 16(%1,%0) \n\t"

  2995.             "sub  $32, %0 \n\t"

  2996.             "jge  1b \n\t"

  2997.             :"+r"(i)

  2998.             :"r"(dst), "r"(src0), "r"(src1), "r"(src2)

  2999.             :"memory"

  3000.         );

  3001.     }

  3002.     else

  3003.         ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);

  3004. }

  3005. 

  3006. void float_to_int16_3dnow(int16_t *dst, const float *src, int len){

  3007.     // not bit-exact: pf2id uses different rounding than C and SSE

  3008.     int i;

  3009.     for(i=0; i<len; i+=4) {

  3010.         asm volatile(

  3011.             "pf2id       %1, %%mm0 \n\t"

  3012.             "pf2id       %2, %%mm1 \n\t"

  3013.             "packssdw %%mm1, %%mm0 \n\t"

  3014.             "movq     %%mm0, %0    \n\t"

  3015.             :"=m"(dst[i])

  3016.             :"m"(src[i]), "m"(src[i+2])

  3017.         );

  3018.     }

  3019.     asm volatile("femms");

  3020. }

  3021. void float_to_int16_sse(int16_t *dst, const float *src, int len){

  3022.     int i;

  3023.     for(i=0; i<len; i+=4) {

  3024.         asm volatile(

  3025.             "cvtps2pi    %1, %%mm0 \n\t"

  3026.             "cvtps2pi    %2, %%mm1 \n\t"

  3027.             "packssdw %%mm1, %%mm0 \n\t"

  3028.             "movq     %%mm0, %0    \n\t"

  3029.             :"=m"(dst[i])

  3030.             :"m"(src[i]), "m"(src[i+2])

  3031.         );

  3032.     }

  3033.     asm volatile("emms");

  3034. }

  3035. 

  3036. #ifdef CONFIG_SNOW_ENCODER

  3037. extern void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width);

  3038. extern void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width);

  3039. extern void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);

  3040. extern void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);

  3041. extern void ff_snow_inner_add_yblock_sse2(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,

  3042.                            int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);

  3043. extern void ff_snow_inner_add_yblock_mmx(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,

  3044.                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);

  3045. #endif

  3046. 

  3047. void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)

  3048. {

  3049.     mm_flags = mm_support();

  3050. 

  3051.     if (avctx->dsp_mask) {

  3052.         if (avctx->dsp_mask & FF_MM_FORCE)

  3053.             mm_flags |= (avctx->dsp_mask & 0xffff);

  3054.         else

  3055.             mm_flags &= ~(avctx->dsp_mask & 0xffff);

  3056.     }

  3057. 

  3058. #if 0

  3059.     av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");

  3060.     if (mm_flags & MM_MMX)

  3061.         av_log(avctx, AV_LOG_INFO, " mmx");

  3062.     if (mm_flags & MM_MMXEXT)

  3063.         av_log(avctx, AV_LOG_INFO, " mmxext");

  3064.     if (mm_flags & MM_3DNOW)

  3065.         av_log(avctx, AV_LOG_INFO, " 3dnow");

  3066.     if (mm_flags & MM_SSE)

  3067.         av_log(avctx, AV_LOG_INFO, " sse");

  3068.     if (mm_flags & MM_SSE2)

  3069.         av_log(avctx, AV_LOG_INFO, " sse2");

  3070.     av_log(avctx, AV_LOG_INFO, "\n");

  3071. #endif

  3072. 

  3073.     if (mm_flags & MM_MMX) {

  3074.         const int idct_algo= avctx->idct_algo;

  3075. 

  3076. #ifdef CONFIG_ENCODERS

  3077.         const int dct_algo = avctx->dct_algo;

  3078.         if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){

  3079.             if(mm_flags & MM_SSE2){

  3080.                 c->fdct = ff_fdct_sse2;

  3081.             }else if(mm_flags & MM_MMXEXT){

  3082.                 c->fdct = ff_fdct_mmx2;

  3083.             }else{

  3084.                 c->fdct = ff_fdct_mmx;

  3085.             }

  3086.         }

  3087. #endif //CONFIG_ENCODERS

  3088.         if(avctx->lowres==0){

  3089.             if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){

  3090.                 c->idct_put= ff_simple_idct_put_mmx;

  3091.                 c->idct_add= ff_simple_idct_add_mmx;

  3092.                 c->idct    = ff_simple_idct_mmx;

  3093.                 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;

  3094. #ifdef CONFIG_GPL

  3095.             }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){

  3096.                 if(mm_flags & MM_MMXEXT){

  3097.                     c->idct_put= ff_libmpeg2mmx2_idct_put;

  3098.                     c->idct_add= ff_libmpeg2mmx2_idct_add;

  3099.                     c->idct    = ff_mmxext_idct;

  3100.                 }else{

  3101.                     c->idct_put= ff_libmpeg2mmx_idct_put;

  3102.                     c->idct_add= ff_libmpeg2mmx_idct_add;

  3103.                     c->idct    = ff_mmx_idct;

  3104.                 }

  3105.                 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;

  3106. #endif

  3107.             }else if(idct_algo==FF_IDCT_VP3 &&

  3108.                      avctx->codec->id!=CODEC_ID_THEORA &&

  3109.                      !(avctx->flags & CODEC_FLAG_BITEXACT)){

  3110.                 if(mm_flags & MM_SSE2){

  3111.                     c->idct_put= ff_vp3_idct_put_sse2;

  3112.                     c->idct_add= ff_vp3_idct_add_sse2;

  3113.                     c->idct    = ff_vp3_idct_sse2;

  3114.                     c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;

  3115.                 }else{

  3116.                     ff_vp3_dsp_init_mmx();

  3117.                     c->idct_put= ff_vp3_idct_put_mmx;

  3118.                     c->idct_add= ff_vp3_idct_add_mmx;

  3119.                     c->idct    = ff_vp3_idct_mmx;

  3120.                     c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;

  3121.                 }

  3122.             }else if(idct_algo==FF_IDCT_CAVS){

  3123.                     c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;

  3124.             }else if(idct_algo==FF_IDCT_XVIDMMX){

  3125.                 if(mm_flags & MM_MMXEXT){

  3126.                     c->idct_put= ff_idct_xvid_mmx2_put;

  3127.                     c->idct_add= ff_idct_xvid_mmx2_add;

  3128.                     c->idct    = ff_idct_xvid_mmx2;

  3129.                 }else{

  3130.                     c->idct_put= ff_idct_xvid_mmx_put;

  3131.                     c->idct_add= ff_idct_xvid_mmx_add;

  3132.                     c->idct    = ff_idct_xvid_mmx;

  3133.                 }

  3134.             }

  3135.         }

  3136. 

  3137. #ifdef CONFIG_ENCODERS

  3138.         c->get_pixels = get_pixels_mmx;

  3139.         c->diff_pixels = diff_pixels_mmx;

  3140. #endif //CONFIG_ENCODERS

  3141.         c->put_pixels_clamped = put_pixels_clamped_mmx;

  3142.         c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;

  3143.         c->add_pixels_clamped = add_pixels_clamped_mmx;

  3144.         c->clear_blocks = clear_blocks_mmx;

  3145. #ifdef CONFIG_ENCODERS

  3146.         c->pix_sum = pix_sum16_mmx;

  3147. #endif //CONFIG_ENCODERS

  3148. 

  3149.         c->put_pixels_tab[0][0] = put_pixels16_mmx;

  3150.         c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;

  3151.         c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;

  3152.         c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

  3153. 

  3154.         c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;

  3155.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;

  3156.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;

  3157.         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

  3158. 

  3159.         c->avg_pixels_tab[0][0] = avg_pixels16_mmx;

  3160.         c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;

  3161.         c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;

  3162.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

  3163. 

  3164.         c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;

  3165.         c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;

  3166.         c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;

  3167.         c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

  3168. 

  3169.         c->put_pixels_tab[1][0] = put_pixels8_mmx;

  3170.         c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;

  3171.         c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;

  3172.         c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

  3173. 

  3174.         c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;

  3175.         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;

  3176.         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;

  3177.         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

  3178. 

  3179.         c->avg_pixels_tab[1][0] = avg_pixels8_mmx;

  3180.         c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;

  3181.         c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;

  3182.         c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

  3183. 

  3184.         c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;

  3185.         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;

  3186.         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;

  3187.         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;

  3188. 

  3189.         c->gmc= gmc_mmx;

  3190. 

  3191.         c->add_bytes= add_bytes_mmx;

  3192. #ifdef CONFIG_ENCODERS

  3193.         c->diff_bytes= diff_bytes_mmx;

  3194. 

  3195.         c->hadamard8_diff[0]= hadamard8_diff16_mmx;

  3196.         c->hadamard8_diff[1]= hadamard8_diff_mmx;

  3197. 

  3198.         c->pix_norm1 = pix_norm1_mmx;

  3199.         c->sse[0] = (mm_flags & MM_SSE2) ? sse16_sse2 : sse16_mmx;

  3200.           c->sse[1] = sse8_mmx;

  3201.         c->vsad[4]= vsad_intra16_mmx;

  3202. 

  3203.         c->nsse[0] = nsse16_mmx;

  3204.         c->nsse[1] = nsse8_mmx;

  3205.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3206.             c->vsad[0] = vsad16_mmx;

  3207.         }

  3208. 

  3209.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3210.             c->try_8x8basis= try_8x8basis_mmx;

  3211.         }

  3212.         c->add_8x8basis= add_8x8basis_mmx;

  3213. 

  3214. #endif //CONFIG_ENCODERS

  3215. 

  3216.         c->h263_v_loop_filter= h263_v_loop_filter_mmx;

  3217.         c->h263_h_loop_filter= h263_h_loop_filter_mmx;

  3218.         c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx;

  3219.         c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx;

  3220. 

  3221.         c->h264_idct_dc_add=

  3222.         c->h264_idct_add= ff_h264_idct_add_mmx;

  3223.         c->h264_idct8_dc_add=

  3224.         c->h264_idct8_add= ff_h264_idct8_add_mmx;

  3225. 

  3226.         if (mm_flags & MM_MMXEXT) {

  3227.             c->prefetch = prefetch_mmx2;

  3228. 

  3229.             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;

  3230.             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

  3231. 

  3232.             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;

  3233.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;

  3234.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

  3235. 

  3236.             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;

  3237.             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

  3238. 

  3239.             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;

  3240.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;

  3241.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

  3242. 

  3243. #ifdef CONFIG_ENCODERS

  3244.             c->hadamard8_diff[0]= hadamard8_diff16_mmx2;

  3245.             c->hadamard8_diff[1]= hadamard8_diff_mmx2;

  3246.             c->vsad[4]= vsad_intra16_mmx2;

  3247. #endif //CONFIG_ENCODERS

  3248. 

  3249.             c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;

  3250.             c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;

  3251. 

  3252.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3253.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;

  3254.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

  3255.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

  3256.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

  3257.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

  3258.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;

  3259. #ifdef CONFIG_ENCODERS

  3260.                 c->vsad[0] = vsad16_mmx2;

  3261. #endif //CONFIG_ENCODERS

  3262.             }

  3263. 

  3264. #if 1

  3265.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)

  3266.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)

  3267.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)

  3268.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)

  3269.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)

  3270.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)

  3271.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)

  3272.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)

  3273.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)

  3274.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)

  3275.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)

  3276.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)

  3277.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)

  3278.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)

  3279.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)

  3280.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)

  3281.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)

  3282.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)

  3283.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)

  3284.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)

  3285.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)

  3286.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)

  3287.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)

  3288.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)

  3289.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)

  3290.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)

  3291.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)

  3292.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)

  3293.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)

  3294.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)

  3295.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)

  3296.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)

  3297. #endif

  3298. 

  3299. //FIXME 3dnow too

  3300. #define dspfunc(PFX, IDX, NUM) \

  3301.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_mmx2; \

  3302.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_mmx2; \

  3303.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_mmx2; \

  3304.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_mmx2; \

  3305.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_mmx2; \

  3306.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_mmx2; \

  3307.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_mmx2; \

  3308.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_mmx2; \

  3309.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_mmx2; \

  3310.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_mmx2; \

  3311.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_mmx2; \

  3312.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_mmx2; \

  3313.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_mmx2; \

  3314.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_mmx2; \

  3315.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_mmx2; \

  3316.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_mmx2

  3317. 

  3318.             dspfunc(put_h264_qpel, 0, 16);

  3319.             dspfunc(put_h264_qpel, 1, 8);

  3320.             dspfunc(put_h264_qpel, 2, 4);

  3321.             dspfunc(avg_h264_qpel, 0, 16);

  3322.             dspfunc(avg_h264_qpel, 1, 8);

  3323.             dspfunc(avg_h264_qpel, 2, 4);

  3324. 

  3325.             dspfunc(put_2tap_qpel, 0, 16);

  3326.             dspfunc(put_2tap_qpel, 1, 8);

  3327.             dspfunc(avg_2tap_qpel, 0, 16);

  3328.             dspfunc(avg_2tap_qpel, 1, 8);

  3329. #undef dspfunc

  3330. 

  3331.             c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2;

  3332.             c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2;

  3333.             c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2;

  3334.             c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2;

  3335.             c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;

  3336.             c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;

  3337.             c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;

  3338.             c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;

  3339.             c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;

  3340.             c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;

  3341.             c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;

  3342. 

  3343.             c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;

  3344.             c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;

  3345.             c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;

  3346.             c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;

  3347.             c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;

  3348.             c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;

  3349.             c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;

  3350.             c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;

  3351. 

  3352.             c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;

  3353.             c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;

  3354.             c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;

  3355.             c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;

  3356.             c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;

  3357.             c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;

  3358.             c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;

  3359.             c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;

  3360. 

  3361. #ifdef CONFIG_CAVS_DECODER

  3362.             ff_cavsdsp_init_mmx2(c, avctx);

  3363. #endif

  3364. 

  3365. #ifdef CONFIG_ENCODERS

  3366.             c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;

  3367. #endif //CONFIG_ENCODERS

  3368.         } else if (mm_flags & MM_3DNOW) {

  3369.             c->prefetch = prefetch_3dnow;

  3370. 

  3371.             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;

  3372.             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  3373. 

  3374.             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  3375.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  3376.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  3377. 

  3378.             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;

  3379.             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

  3380. 

  3381.             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;

  3382.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;

  3383.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

  3384. 

  3385.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3386.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  3387.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  3388.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;

  3389.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

  3390.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  3391.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;

  3392.             }

  3393. 

  3394.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)

  3395.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)

  3396.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)

  3397.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)

  3398.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)

  3399.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)

  3400.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)

  3401.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)

  3402.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)

  3403.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)

  3404.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)

  3405.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)

  3406.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)

  3407.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)

  3408.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)

  3409.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)

  3410.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)

  3411.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)

  3412.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)

  3413.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)

  3414.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)

  3415.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)

  3416.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)

  3417.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)

  3418.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)

  3419.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)

  3420.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)

  3421.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)

  3422.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)

  3423.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)

  3424.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)

  3425.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)

  3426. 

  3427. #define dspfunc(PFX, IDX, NUM) \

  3428.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_3dnow; \

  3429.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_3dnow; \

  3430.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_3dnow; \

  3431.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_3dnow; \

  3432.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_3dnow; \

  3433.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_3dnow; \

  3434.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_3dnow; \

  3435.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_3dnow; \

  3436.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_3dnow; \

  3437.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_3dnow; \

  3438.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_3dnow; \

  3439.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_3dnow; \

  3440.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_3dnow; \

  3441.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_3dnow; \

  3442.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_3dnow; \

  3443.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_3dnow

  3444. 

  3445.             dspfunc(put_h264_qpel, 0, 16);

  3446.             dspfunc(put_h264_qpel, 1, 8);

  3447.             dspfunc(put_h264_qpel, 2, 4);

  3448.             dspfunc(avg_h264_qpel, 0, 16);

  3449.             dspfunc(avg_h264_qpel, 1, 8);

  3450.             dspfunc(avg_h264_qpel, 2, 4);

  3451. 

  3452.             dspfunc(put_2tap_qpel, 0, 16);

  3453.             dspfunc(put_2tap_qpel, 1, 8);

  3454.             dspfunc(avg_2tap_qpel, 0, 16);

  3455.             dspfunc(avg_2tap_qpel, 1, 8);

  3456. 

  3457.             c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow;

  3458.             c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow;

  3459.         }

  3460. 

  3461. #ifdef CONFIG_SNOW_ENCODER

  3462.         if(mm_flags & MM_SSE2){

  3463.             c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;

  3464.             c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;

  3465.             c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;

  3466.         }

  3467.         else{

  3468.             c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;

  3469.             c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;

  3470.             c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;

  3471.         }

  3472. #endif

  3473. 

  3474.         if(mm_flags & MM_3DNOW){

  3475.             c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;

  3476.             c->vector_fmul = vector_fmul_3dnow;

  3477.             if(!(avctx->flags & CODEC_FLAG_BITEXACT))

  3478.                 c->float_to_int16 = float_to_int16_3dnow;

  3479.         }

  3480.         if(mm_flags & MM_3DNOWEXT)

  3481.             c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;

  3482.         if(mm_flags & MM_SSE){

  3483.             c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;

  3484.             c->vector_fmul = vector_fmul_sse;

  3485.             c->float_to_int16 = float_to_int16_sse;

  3486.             c->vector_fmul_reverse = vector_fmul_reverse_sse;

  3487.             c->vector_fmul_add_add = vector_fmul_add_add_sse;

  3488.         }

  3489.         if(mm_flags & MM_3DNOW)

  3490.             c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse

  3491.     }

  3492. 

  3493. #ifdef CONFIG_ENCODERS

  3494.     dsputil_init_pix_mmx(c, avctx);

  3495. #endif //CONFIG_ENCODERS

  3496. #if 0

  3497.     // for speed testing

  3498.     get_pixels = just_return;

  3499.     put_pixels_clamped = just_return;

  3500.     add_pixels_clamped = just_return;

  3501. 

  3502.     pix_abs16x16 = just_return;

  3503.     pix_abs16x16_x2 = just_return;

  3504.     pix_abs16x16_y2 = just_return;

  3505.     pix_abs16x16_xy2 = just_return;

  3506. 

  3507.     put_pixels_tab[0] = just_return;

  3508.     put_pixels_tab[1] = just_return;

  3509.     put_pixels_tab[2] = just_return;

  3510.     put_pixels_tab[3] = just_return;

  3511. 

  3512.     put_no_rnd_pixels_tab[0] = just_return;

  3513.     put_no_rnd_pixels_tab[1] = just_return;

  3514.     put_no_rnd_pixels_tab[2] = just_return;

  3515.     put_no_rnd_pixels_tab[3] = just_return;

  3516. 

  3517.     avg_pixels_tab[0] = just_return;

  3518.     avg_pixels_tab[1] = just_return;

  3519.     avg_pixels_tab[2] = just_return;

  3520.     avg_pixels_tab[3] = just_return;

  3521. 

  3522.     avg_no_rnd_pixels_tab[0] = just_return;

  3523.     avg_no_rnd_pixels_tab[1] = just_return;

  3524.     avg_no_rnd_pixels_tab[2] = just_return;

  3525.     avg_no_rnd_pixels_tab[3] = just_return;

  3526. 

  3527.     //av_fdct = just_return;

  3528.     //ff_idct = just_return;

  3529. #endif

  3530. }