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.
31386 Commits
32 Branches
283MB
Tree: 3b2d285afb
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '3b2d285afb'
${ noResults }
FFmpeg/libavcodec/x86/dsputil_mmx.c

2891 lines
121KB
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 "libavutil/cpu.h"

  26. #include "libavutil/x86_cpu.h"

  27. #include "libavcodec/dsputil.h"

  28. #include "libavcodec/h264dsp.h"

  29. #include "libavcodec/mpegvideo.h"

  30. #include "libavcodec/simple_idct.h"

  31. #include "libavcodec/ac3dec.h"

  32. #include "dsputil_mmx.h"

  33. #include "idct_xvid.h"

  34. 

  35. //#undef NDEBUG

  36. //#include <assert.h>

  37. 

  38. /* pixel operations */

  39. DECLARE_ALIGNED(8,  const uint64_t, ff_bone) = 0x0101010101010101ULL;

  40. DECLARE_ALIGNED(8,  const uint64_t, ff_wtwo) = 0x0002000200020002ULL;

  41. 

  42. DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =

  43. {0x8000000080000000ULL, 0x8000000080000000ULL};

  44. 

  45. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_1  ) = {0x0001000100010001ULL, 0x0001000100010001ULL};

  46. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_2  ) = {0x0002000200020002ULL, 0x0002000200020002ULL};

  47. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_3  ) = {0x0003000300030003ULL, 0x0003000300030003ULL};

  48. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_4  ) = {0x0004000400040004ULL, 0x0004000400040004ULL};

  49. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};

  50. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_8  ) = {0x0008000800080008ULL, 0x0008000800080008ULL};

  51. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_9  ) = {0x0009000900090009ULL, 0x0009000900090009ULL};

  52. DECLARE_ALIGNED(8,  const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;

  53. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_16 ) = {0x0010001000100010ULL, 0x0010001000100010ULL};

  54. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_17 ) = {0x0011001100110011ULL, 0x0011001100110011ULL};

  55. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_18 ) = {0x0012001200120012ULL, 0x0012001200120012ULL};

  56. DECLARE_ALIGNED(8,  const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;

  57. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_27 ) = {0x001B001B001B001BULL, 0x001B001B001B001BULL};

  58. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_28 ) = {0x001C001C001C001CULL, 0x001C001C001C001CULL};

  59. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_32 ) = {0x0020002000200020ULL, 0x0020002000200020ULL};

  60. DECLARE_ALIGNED(8,  const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;

  61. DECLARE_ALIGNED(8,  const uint64_t, ff_pw_53 ) = 0x0035003500350035ULL;

  62. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_63 ) = {0x003F003F003F003FULL, 0x003F003F003F003FULL};

  63. DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_64 ) = {0x0040004000400040ULL, 0x0040004000400040ULL};

  64. DECLARE_ALIGNED(8,  const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;

  65. DECLARE_ALIGNED(8,  const uint64_t, ff_pw_128) = 0x0080008000800080ULL;

  66. DECLARE_ALIGNED(8,  const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;

  67. 

  68. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_0  ) = {0x0000000000000000ULL, 0x0000000000000000ULL};

  69. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_1  ) = {0x0101010101010101ULL, 0x0101010101010101ULL};

  70. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_3  ) = {0x0303030303030303ULL, 0x0303030303030303ULL};

  71. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_4  ) = {0x0404040404040404ULL, 0x0404040404040404ULL};

  72. DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;

  73. DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;

  74. DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;

  75. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_80 ) = {0x8080808080808080ULL, 0x8080808080808080ULL};

  76. DECLARE_ALIGNED(8,  const uint64_t, ff_pb_81 ) = 0x8181818181818181ULL;

  77. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_A1 ) = {0xA1A1A1A1A1A1A1A1ULL, 0xA1A1A1A1A1A1A1A1ULL};

  78. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_F8 ) = {0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL};

  79. DECLARE_ALIGNED(8,  const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;

  80. DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_FE ) = {0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL};

  81. 

  82. DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };

  83. DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };

  84. 

  85. #define JUMPALIGN() __asm__ volatile (".p2align 3"::)

  86. #define MOVQ_ZERO(regd)  __asm__ volatile ("pxor %%" #regd ", %%" #regd ::)

  87. 

  88. #define MOVQ_BFE(regd) \

  89.     __asm__ volatile ( \

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

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

  92. 

  93. #ifndef PIC

  94. #define MOVQ_BONE(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))

  95. #define MOVQ_WTWO(regd)  __asm__ volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))

  96. #else

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

  98. // pcmpeqd -> -1

  99. #define MOVQ_BONE(regd) \

  100.     __asm__ volatile ( \

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

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

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

  104. 

  105. #define MOVQ_WTWO(regd) \

  106.     __asm__ volatile ( \

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

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

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

  110. 

  111. #endif

  112. 

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

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

  115. // regfe is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

  123. 

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

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

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

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

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

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

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

  131. 

  132. // mm6 is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

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

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

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

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

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

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

  146. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  160. 

  161. /***********************************/

  162. /* MMX no rounding */

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

  164. #define SET_RND  MOVQ_WONE

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

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

  167. #define OP_AVG(a, b, c, e)              PAVGB_MMX(a, b, c, e)

  168. 

  169. #include "dsputil_mmx_rnd_template.c"

  170. 

  171. #undef DEF

  172. #undef SET_RND

  173. #undef PAVGBP

  174. #undef PAVGB

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

  176. /* MMX rounding */

  177. 

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

  179. #define SET_RND  MOVQ_WTWO

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

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

  182. 

  183. #include "dsputil_mmx_rnd_template.c"

  184. 

  185. #undef DEF

  186. #undef SET_RND

  187. #undef PAVGBP

  188. #undef PAVGB

  189. #undef OP_AVG

  190. 

  191. /***********************************/

  192. /* 3Dnow specific */

  193. 

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

  195. #define PAVGB "pavgusb"

  196. #define OP_AVG PAVGB

  197. 

  198. #include "dsputil_mmx_avg_template.c"

  199. 

  200. #undef DEF

  201. #undef PAVGB

  202. #undef OP_AVG

  203. 

  204. /***********************************/

  205. /* MMX2 specific */

  206. 

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

  208. 

  209. /* Introduced only in MMX2 set */

  210. #define PAVGB "pavgb"

  211. #define OP_AVG PAVGB

  212. 

  213. #include "dsputil_mmx_avg_template.c"

  214. 

  215. #undef DEF

  216. #undef PAVGB

  217. #undef OP_AVG

  218. 

  219. #define put_no_rnd_pixels16_mmx put_pixels16_mmx

  220. #define put_no_rnd_pixels8_mmx put_pixels8_mmx

  221. #define put_pixels16_mmx2 put_pixels16_mmx

  222. #define put_pixels8_mmx2 put_pixels8_mmx

  223. #define put_pixels4_mmx2 put_pixels4_mmx

  224. #define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx

  225. #define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx

  226. #define put_pixels16_3dnow put_pixels16_mmx

  227. #define put_pixels8_3dnow put_pixels8_mmx

  228. #define put_pixels4_3dnow put_pixels4_mmx

  229. #define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx

  230. #define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx

  231. 

  232. /***********************************/

  233. /* standard MMX */

  234. 

  235. void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)

  236. {

  237.     const DCTELEM *p;

  238.     uint8_t *pix;

  239. 

  240.     /* read the pixels */

  241.     p = block;

  242.     pix = pixels;

  243.     /* unrolled loop */

  244.         __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  261.                 ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "m"(*p)

  262.                 :"memory");

  263.         pix += line_size*4;

  264.         p += 32;

  265. 

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

  267.     // compiler would generate some very strange code

  268.     // thus using "r"

  269.     __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  286.             ::"r" (pix), "r" ((x86_reg)line_size), "r" ((x86_reg)line_size*3), "r"(p)

  287.             :"memory");

  288. }

  289. 

  290. #define put_signed_pixels_clamped_mmx_half(off) \

  291.             "movq    "#off"(%2), %%mm1          \n\t"\

  292.             "movq 16+"#off"(%2), %%mm2          \n\t"\

  293.             "movq 32+"#off"(%2), %%mm3          \n\t"\

  294.             "movq 48+"#off"(%2), %%mm4          \n\t"\

  295.             "packsswb  8+"#off"(%2), %%mm1      \n\t"\

  296.             "packsswb 24+"#off"(%2), %%mm2      \n\t"\

  297.             "packsswb 40+"#off"(%2), %%mm3      \n\t"\

  298.             "packsswb 56+"#off"(%2), %%mm4      \n\t"\

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

  300.             "paddb %%mm0, %%mm2                 \n\t"\

  301.             "paddb %%mm0, %%mm3                 \n\t"\

  302.             "paddb %%mm0, %%mm4                 \n\t"\

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

  304.             "movq %%mm2, (%0, %3)               \n\t"\

  305.             "movq %%mm3, (%0, %3, 2)            \n\t"\

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

  307. 

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

  309. {

  310.     x86_reg line_skip = line_size;

  311.     x86_reg line_skip3;

  312. 

  313.     __asm__ volatile (

  314.             "movq "MANGLE(ff_pb_80)", %%mm0 \n\t"

  315.             "lea (%3, %3, 2), %1                \n\t"

  316.             put_signed_pixels_clamped_mmx_half(0)

  317.             "lea (%0, %3, 4), %0                \n\t"

  318.             put_signed_pixels_clamped_mmx_half(64)

  319.             :"+&r" (pixels), "=&r" (line_skip3)

  320.             :"r" (block), "r"(line_skip)

  321.             :"memory");

  322. }

  323. 

  324. void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)

  325. {

  326.     const DCTELEM *p;

  327.     uint8_t *pix;

  328.     int i;

  329. 

  330.     /* read the pixels */

  331.     p = block;

  332.     pix = pixels;

  333.     MOVQ_ZERO(mm7);

  334.     i = 4;

  335.     do {

  336.         __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  358.                 :"r"(p)

  359.                 :"memory");

  360.         pix += line_size*2;

  361.         p += 16;

  362.     } while (--i);

  363. }

  364. 

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

  366. {

  367.     __asm__ volatile(

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

  369.          ".p2align 3                    \n\t"

  370.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  384.          "jnz 1b                        \n\t"

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

  386.          : "r"((x86_reg)line_size)

  387.          : "%"REG_a, "memory"

  388.         );

  389. }

  390. 

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

  392. {

  393.     __asm__ volatile(

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

  395.          ".p2align 3                    \n\t"

  396.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  410.          "jnz 1b                        \n\t"

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

  412.          : "r"((x86_reg)line_size)

  413.          : "%"REG_a, "memory"

  414.         );

  415. }

  416. 

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

  418. {

  419.     __asm__ volatile(

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

  421.          ".p2align 3                    \n\t"

  422.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  444.          "jnz 1b                        \n\t"

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

  446.          : "r"((x86_reg)line_size)

  447.          : "%"REG_a, "memory"

  448.         );

  449. }

  450. 

  451. static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  452. {

  453.     __asm__ volatile(

  454.          "1:                            \n\t"

  455.          "movdqu (%1), %%xmm0           \n\t"

  456.          "movdqu (%1,%3), %%xmm1        \n\t"

  457.          "movdqu (%1,%3,2), %%xmm2      \n\t"

  458.          "movdqu (%1,%4), %%xmm3        \n\t"

  459.          "movdqa %%xmm0, (%2)           \n\t"

  460.          "movdqa %%xmm1, (%2,%3)        \n\t"

  461.          "movdqa %%xmm2, (%2,%3,2)      \n\t"

  462.          "movdqa %%xmm3, (%2,%4)        \n\t"

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

  464.          "lea (%1,%3,4), %1             \n\t"

  465.          "lea (%2,%3,4), %2             \n\t"

  466.          "jnz 1b                        \n\t"

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

  468.          : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)

  469.          : "memory"

  470.         );

  471. }

  472. 

  473. static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  474. {

  475.     __asm__ volatile(

  476.          "1:                            \n\t"

  477.          "movdqu (%1), %%xmm0           \n\t"

  478.          "movdqu (%1,%3), %%xmm1        \n\t"

  479.          "movdqu (%1,%3,2), %%xmm2      \n\t"

  480.          "movdqu (%1,%4), %%xmm3        \n\t"

  481.          "pavgb  (%2), %%xmm0           \n\t"

  482.          "pavgb  (%2,%3), %%xmm1        \n\t"

  483.          "pavgb  (%2,%3,2), %%xmm2      \n\t"

  484.          "pavgb  (%2,%4), %%xmm3        \n\t"

  485.          "movdqa %%xmm0, (%2)           \n\t"

  486.          "movdqa %%xmm1, (%2,%3)        \n\t"

  487.          "movdqa %%xmm2, (%2,%3,2)      \n\t"

  488.          "movdqa %%xmm3, (%2,%4)        \n\t"

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

  490.          "lea (%1,%3,4), %1             \n\t"

  491.          "lea (%2,%3,4), %2             \n\t"

  492.          "jnz 1b                        \n\t"

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

  494.          : "r"((x86_reg)line_size), "r"((x86_reg)3L*line_size)

  495.          : "memory"

  496.         );

  497. }

  498. 

  499. #define CLEAR_BLOCKS(name,n) \

  500. static void name(DCTELEM *blocks)\

  501. {\

  502.     __asm__ volatile(\

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

  504.                 "mov     %1, %%"REG_a"          \n\t"\

  505.                 "1:                             \n\t"\

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

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

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

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

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

  511.                 " js 1b                         \n\t"\

  512.                 : : "r" (((uint8_t *)blocks)+128*n),\

  513.                     "i" (-128*n)\

  514.                 : "%"REG_a\

  515.         );\

  516. }

  517. CLEAR_BLOCKS(clear_blocks_mmx, 6)

  518. CLEAR_BLOCKS(clear_block_mmx, 1)

  519. 

  520. static void clear_block_sse(DCTELEM *block)

  521. {

  522.     __asm__ volatile(

  523.         "xorps  %%xmm0, %%xmm0  \n"

  524.         "movaps %%xmm0,    (%0) \n"

  525.         "movaps %%xmm0,  16(%0) \n"

  526.         "movaps %%xmm0,  32(%0) \n"

  527.         "movaps %%xmm0,  48(%0) \n"

  528.         "movaps %%xmm0,  64(%0) \n"

  529.         "movaps %%xmm0,  80(%0) \n"

  530.         "movaps %%xmm0,  96(%0) \n"

  531.         "movaps %%xmm0, 112(%0) \n"

  532.         :: "r"(block)

  533.         : "memory"

  534.     );

  535. }

  536. 

  537. static void clear_blocks_sse(DCTELEM *blocks)

  538. {\

  539.     __asm__ volatile(

  540.         "xorps  %%xmm0, %%xmm0  \n"

  541.         "mov     %1, %%"REG_a"  \n"

  542.         "1:                     \n"

  543.         "movaps %%xmm0,    (%0, %%"REG_a") \n"

  544.         "movaps %%xmm0,  16(%0, %%"REG_a") \n"

  545.         "movaps %%xmm0,  32(%0, %%"REG_a") \n"

  546.         "movaps %%xmm0,  48(%0, %%"REG_a") \n"

  547.         "movaps %%xmm0,  64(%0, %%"REG_a") \n"

  548.         "movaps %%xmm0,  80(%0, %%"REG_a") \n"

  549.         "movaps %%xmm0,  96(%0, %%"REG_a") \n"

  550.         "movaps %%xmm0, 112(%0, %%"REG_a") \n"

  551.         "add $128, %%"REG_a"    \n"

  552.         " js 1b                 \n"

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

  554.             "i" (-128*6)

  555.         : "%"REG_a

  556.     );

  557. }

  558. 

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

  560.     x86_reg i=0;

  561.     __asm__ volatile(

  562.         "jmp 2f                         \n\t"

  563.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

  573.         "2:                             \n\t"

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

  575.         " js 1b                         \n\t"

  576.         : "+r" (i)

  577.         : "r"(src), "r"(dst), "r"((x86_reg)w-15)

  578.     );

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

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

  581. }

  582. 

  583. #if HAVE_7REGS

  584. static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top) {

  585.     x86_reg w2 = -w;

  586.     x86_reg x;

  587.     int l = *left & 0xff;

  588.     int tl = *left_top & 0xff;

  589.     int t;

  590.     __asm__ volatile(

  591.         "mov    %7, %3 \n"

  592.         "1: \n"

  593.         "movzbl (%3,%4), %2 \n"

  594.         "mov    %2, %k3 \n"

  595.         "sub   %b1, %b3 \n"

  596.         "add   %b0, %b3 \n"

  597.         "mov    %2, %1 \n"

  598.         "cmp    %0, %2 \n"

  599.         "cmovg  %0, %2 \n"

  600.         "cmovg  %1, %0 \n"

  601.         "cmp   %k3, %0 \n"

  602.         "cmovg %k3, %0 \n"

  603.         "mov    %7, %3 \n"

  604.         "cmp    %2, %0 \n"

  605.         "cmovl  %2, %0 \n"

  606.         "add (%6,%4), %b0 \n"

  607.         "mov   %b0, (%5,%4) \n"

  608.         "inc    %4 \n"

  609.         "jl 1b \n"

  610.         :"+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)

  611.         :"r"(dst+w), "r"(diff+w), "rm"(top+w)

  612.     );

  613.     *left = l;

  614.     *left_top = tl;

  615. }

  616. #endif

  617. 

  618. #define H263_LOOP_FILTER \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  688. 

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

  690.     if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {

  691.     const int strength= ff_h263_loop_filter_strength[qscale];

  692. 

  693.     __asm__ volatile(

  694. 

  695.         H263_LOOP_FILTER

  696. 

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

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

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

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

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

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

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

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

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

  706.     );

  707.     }

  708. }

  709. 

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

  711.     if(CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {

  712.     const int strength= ff_h263_loop_filter_strength[qscale];

  713.     DECLARE_ALIGNED(8, uint64_t, temp)[4];

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

  715. 

  716.     src -= 2;

  717. 

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

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

  720.     __asm__ volatile(

  721.         H263_LOOP_FILTER // 5 3 4 6

  722. 

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

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

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

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

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

  728.     );

  729. 

  730.     __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  755.         :: "r" (src),

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

  757.            "r" ((x86_reg)   stride ),

  758.            "r" ((x86_reg)(3*stride))

  759.     );

  760.     }

  761. }

  762. 

  763. /* draw the edges of width 'w' of an image of size width, height

  764.    this mmx version can only handle w==8 || w==16 */

  765. static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides)

  766. {

  767.     uint8_t *ptr, *last_line;

  768.     int i;

  769. 

  770.     last_line = buf + (height - 1) * wrap;

  771.     /* left and right */

  772.     ptr = buf;

  773.     if(w==8)

  774.     {

  775.         __asm__ volatile(

  776.                 "1:                             \n\t"

  777.                 "movd (%0), %%mm0               \n\t"

  778.                 "punpcklbw %%mm0, %%mm0         \n\t"

  779.                 "punpcklwd %%mm0, %%mm0         \n\t"

  780.                 "punpckldq %%mm0, %%mm0         \n\t"

  781.                 "movq %%mm0, -8(%0)             \n\t"

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

  783.                 "punpckhbw %%mm1, %%mm1         \n\t"

  784.                 "punpckhwd %%mm1, %%mm1         \n\t"

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

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

  787.                 "add %1, %0                     \n\t"

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

  789.                 " jb 1b                         \n\t"

  790.                 : "+r" (ptr)

  791.                 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)

  792.         );

  793.     }

  794.     else

  795.     {

  796.         __asm__ volatile(

  797.                 "1:                             \n\t"

  798.                 "movd (%0), %%mm0               \n\t"

  799.                 "punpcklbw %%mm0, %%mm0         \n\t"

  800.                 "punpcklwd %%mm0, %%mm0         \n\t"

  801.                 "punpckldq %%mm0, %%mm0         \n\t"

  802.                 "movq %%mm0, -8(%0)             \n\t"

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

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

  805.                 "punpckhbw %%mm1, %%mm1         \n\t"

  806.                 "punpckhwd %%mm1, %%mm1         \n\t"

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

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

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

  810.                 "add %1, %0                     \n\t"

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

  812.                 " jb 1b                         \n\t"

  813.                 : "+r" (ptr)

  814.                 : "r" ((x86_reg)wrap), "r" ((x86_reg)width), "r" (ptr + wrap*height)

  815.         );

  816.     }

  817. 

  818.     /* top and bottom (and hopefully also the corners) */

  819.     if (sides&EDGE_TOP) {

  820.         for(i = 0; i < h; i += 4) {

  821.             ptr= buf - (i + 1) * wrap - w;

  822.             __asm__ volatile(

  823.                     "1:                             \n\t"

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

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

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

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

  828.                     "movq %%mm0, (%0, %3)           \n\t"

  829.                     "add $8, %0                     \n\t"

  830.                     "cmp %4, %0                     \n\t"

  831.                     " jb 1b                         \n\t"

  832.                     : "+r" (ptr)

  833.                     : "r" ((x86_reg)buf - (x86_reg)ptr - w), "r" ((x86_reg)-wrap), "r" ((x86_reg)-wrap*3), "r" (ptr+width+2*w)

  834.             );

  835.         }

  836.     }

  837. 

  838.     if (sides&EDGE_BOTTOM) {

  839.         for(i = 0; i < w; i += 4) {

  840.             ptr= last_line + (i + 1) * wrap - w;

  841.             __asm__ volatile(

  842.                     "1:                             \n\t"

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

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

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

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

  847.                     "movq %%mm0, (%0, %3)           \n\t"

  848.                     "add $8, %0                     \n\t"

  849.                     "cmp %4, %0                     \n\t"

  850.                     " jb 1b                         \n\t"

  851.                     : "+r" (ptr)

  852.                     : "r" ((x86_reg)last_line - (x86_reg)ptr - w), "r" ((x86_reg)wrap), "r" ((x86_reg)wrap*3), "r" (ptr+width+2*w)

  853.             );

  854.         }

  855.     }

  856. }

  857. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  878. 

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

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

  881.     uint64_t temp;\

  882. \

  883.     __asm__ volatile(\

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

  885.         "1:                               \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  916.         \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  944.         \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  971.         \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  989.         \

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

  991.         "add %4, %1                       \n\t"\

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

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

  994.         : "+a"(src), "+c"(dst), "+D"(h)\

  995.         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  996.         : "memory"\

  997.     );\

  998. }\

  999. \

  1000. static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1001.     int i;\

  1002.     int16_t temp[16];\

  1003.     /* quick HACK, XXX FIXME MUST be optimized */\

  1004.     for(i=0; i<h; i++)\

  1005.     {\

  1006.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1007.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1008.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1009.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1010.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1011.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\

  1012.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\

  1013.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\

  1014.         temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\

  1015.         temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\

  1016.         temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\

  1017.         temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\

  1018.         temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\

  1019.         temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\

  1020.         temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\

  1021.         temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\

  1022.         __asm__ volatile(\

  1023.             "movq (%0), %%mm0               \n\t"\

  1024.             "movq 8(%0), %%mm1              \n\t"\

  1025.             "paddw %2, %%mm0                \n\t"\

  1026.             "paddw %2, %%mm1                \n\t"\

  1027.             "psraw $5, %%mm0                \n\t"\

  1028.             "psraw $5, %%mm1                \n\t"\

  1029.             "packuswb %%mm1, %%mm0          \n\t"\

  1030.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1031.             "movq 16(%0), %%mm0             \n\t"\

  1032.             "movq 24(%0), %%mm1             \n\t"\

  1033.             "paddw %2, %%mm0                \n\t"\

  1034.             "paddw %2, %%mm1                \n\t"\

  1035.             "psraw $5, %%mm0                \n\t"\

  1036.             "psraw $5, %%mm1                \n\t"\

  1037.             "packuswb %%mm1, %%mm0          \n\t"\

  1038.             OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\

  1039.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1040.             : "memory"\

  1041.         );\

  1042.         dst+=dstStride;\

  1043.         src+=srcStride;\

  1044.     }\

  1045. }\

  1046. \

  1047. static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1048.     __asm__ volatile(\

  1049.         "pxor %%mm7, %%mm7                \n\t"\

  1050.         "1:                               \n\t"\

  1051.         "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\

  1052.         "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\

  1053.         "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\

  1054.         "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\

  1055.         "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\

  1056.         "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\

  1057.         "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\

  1058.         "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\

  1059.         "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\

  1060.         "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\

  1061.         "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\

  1062.         "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\

  1063.         "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\

  1064.         "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\

  1065.         "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\

  1066.         "paddw %%mm3, %%mm5               \n\t" /* b */\

  1067.         "paddw %%mm2, %%mm6               \n\t" /* c */\

  1068.         "paddw %%mm5, %%mm5               \n\t" /* 2b */\

  1069.         "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\

  1070.         "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\

  1071.         "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\

  1072.         "paddw %%mm4, %%mm0               \n\t" /* a */\

  1073.         "paddw %%mm1, %%mm5               \n\t" /* d */\

  1074.         "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\

  1075.         "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\

  1076.         "paddw %5, %%mm6                  \n\t"\

  1077.         "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\

  1078.         "psraw $5, %%mm0                  \n\t"\

  1079.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1080.         \

  1081.         "movd 5(%0), %%mm5                \n\t" /* FGHI */\

  1082.         "punpcklbw %%mm7, %%mm5           \n\t" /* 0F0G0H0I */\

  1083.         "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0G0H0I0I */\

  1084.         "paddw %%mm5, %%mm1               \n\t" /* a */\

  1085.         "paddw %%mm6, %%mm2               \n\t" /* b */\

  1086.         "pshufw $0xBE, %%mm5, %%mm6       \n\t" /* 0H0I0I0H */\

  1087.         "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0I0I0H0G */\

  1088.         "paddw %%mm6, %%mm3               \n\t" /* c */\

  1089.         "paddw %%mm5, %%mm4               \n\t" /* d */\

  1090.         "paddw %%mm2, %%mm2               \n\t" /* 2b */\

  1091.         "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\

  1092.         "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\

  1093.         "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\

  1094.         "psubw %%mm4, %%mm3               \n\t" /* -6b + 3c - d */\

  1095.         "paddw %5, %%mm1                  \n\t"\

  1096.         "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\

  1097.         "psraw $5, %%mm3                  \n\t"\

  1098.         "packuswb %%mm3, %%mm0            \n\t"\

  1099.         OP_MMX2(%%mm0, (%1), %%mm4, q)\

  1100.         \

  1101.         "add %3, %0                       \n\t"\

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

  1103.         "decl %2                          \n\t"\

  1104.         " jnz 1b                          \n\t"\

  1105.         : "+a"(src), "+c"(dst), "+d"(h)\

  1106.         : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER)\

  1107.         : "memory"\

  1108.     );\

  1109. }\

  1110. \

  1111. static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1112.     int i;\

  1113.     int16_t temp[8];\

  1114.     /* quick HACK, XXX FIXME MUST be optimized */\

  1115.     for(i=0; i<h; i++)\

  1116.     {\

  1117.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1118.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1119.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1120.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1121.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1122.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\

  1123.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\

  1124.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\

  1125.         __asm__ volatile(\

  1126.             "movq (%0), %%mm0           \n\t"\

  1127.             "movq 8(%0), %%mm1          \n\t"\

  1128.             "paddw %2, %%mm0            \n\t"\

  1129.             "paddw %2, %%mm1            \n\t"\

  1130.             "psraw $5, %%mm0            \n\t"\

  1131.             "psraw $5, %%mm1            \n\t"\

  1132.             "packuswb %%mm1, %%mm0      \n\t"\

  1133.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1134.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1135.             :"memory"\

  1136.         );\

  1137.         dst+=dstStride;\

  1138.         src+=srcStride;\

  1139.     }\

  1140. }

  1141. 

  1142. #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\

  1143. \

  1144. static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1145.     uint64_t temp[17*4];\

  1146.     uint64_t *temp_ptr= temp;\

  1147.     int count= 17;\

  1148. \

  1149.     /*FIXME unroll */\

  1150.     __asm__ volatile(\

  1151.         "pxor %%mm7, %%mm7              \n\t"\

  1152.         "1:                             \n\t"\

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

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

  1155.         "movq 8(%0), %%mm2              \n\t"\

  1156.         "movq 8(%0), %%mm3              \n\t"\

  1157.         "punpcklbw %%mm7, %%mm0         \n\t"\

  1158.         "punpckhbw %%mm7, %%mm1         \n\t"\

  1159.         "punpcklbw %%mm7, %%mm2         \n\t"\

  1160.         "punpckhbw %%mm7, %%mm3         \n\t"\

  1161.         "movq %%mm0, (%1)               \n\t"\

  1162.         "movq %%mm1, 17*8(%1)           \n\t"\

  1163.         "movq %%mm2, 2*17*8(%1)         \n\t"\

  1164.         "movq %%mm3, 3*17*8(%1)         \n\t"\

  1165.         "add $8, %1                     \n\t"\

  1166.         "add %3, %0                     \n\t"\

  1167.         "decl %2                        \n\t"\

  1168.         " jnz 1b                        \n\t"\

  1169.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  1170.         : "r" ((x86_reg)srcStride)\

  1171.         : "memory"\

  1172.     );\

  1173.     \

  1174.     temp_ptr= temp;\

  1175.     count=4;\

  1176.     \

  1177. /*FIXME reorder for speed */\

  1178.     __asm__ volatile(\

  1179.         /*"pxor %%mm7, %%mm7              \n\t"*/\

  1180.         "1:                             \n\t"\

  1181.         "movq (%0), %%mm0               \n\t"\

  1182.         "movq 8(%0), %%mm1              \n\t"\

  1183.         "movq 16(%0), %%mm2             \n\t"\

  1184.         "movq 24(%0), %%mm3             \n\t"\

  1185.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  1186.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  1187.         "add %4, %1                     \n\t"\

  1188.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  1189.         \

  1190.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  1191.         "add %4, %1                     \n\t"\

  1192.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  1193.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\

  1194.         "add %4, %1                     \n\t"\

  1195.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\

  1196.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\

  1197.         "add %4, %1                     \n\t"\

  1198.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\

  1199.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\

  1200.         "add %4, %1                     \n\t"\

  1201.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\

  1202.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\

  1203.         "add %4, %1                     \n\t"\

  1204.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\

  1205.         \

  1206.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\

  1207.         "add %4, %1                     \n\t"  \

  1208.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\

  1209.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\

  1210.         \

  1211.         "add $136, %0                   \n\t"\

  1212.         "add %6, %1                     \n\t"\

  1213.         "decl %2                        \n\t"\

  1214.         " jnz 1b                        \n\t"\

  1215.         \

  1216.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  1217.         : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(x86_reg)dstStride)\

  1218.         :"memory"\

  1219.     );\

  1220. }\

  1221. \

  1222. static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1223.     uint64_t temp[9*2];\

  1224.     uint64_t *temp_ptr= temp;\

  1225.     int count= 9;\

  1226. \

  1227.     /*FIXME unroll */\

  1228.     __asm__ volatile(\

  1229.         "pxor %%mm7, %%mm7              \n\t"\

  1230.         "1:                             \n\t"\

  1231.         "movq (%0), %%mm0               \n\t"\

  1232.         "movq (%0), %%mm1               \n\t"\

  1233.         "punpcklbw %%mm7, %%mm0         \n\t"\

  1234.         "punpckhbw %%mm7, %%mm1         \n\t"\

  1235.         "movq %%mm0, (%1)               \n\t"\

  1236.         "movq %%mm1, 9*8(%1)            \n\t"\

  1237.         "add $8, %1                     \n\t"\

  1238.         "add %3, %0                     \n\t"\

  1239.         "decl %2                        \n\t"\

  1240.         " jnz 1b                        \n\t"\

  1241.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  1242.         : "r" ((x86_reg)srcStride)\

  1243.         : "memory"\

  1244.     );\

  1245.     \

  1246.     temp_ptr= temp;\

  1247.     count=2;\

  1248.     \

  1249. /*FIXME reorder for speed */\

  1250.     __asm__ volatile(\

  1251.         /*"pxor %%mm7, %%mm7              \n\t"*/\

  1252.         "1:                             \n\t"\

  1253.         "movq (%0), %%mm0               \n\t"\

  1254.         "movq 8(%0), %%mm1              \n\t"\

  1255.         "movq 16(%0), %%mm2             \n\t"\

  1256.         "movq 24(%0), %%mm3             \n\t"\

  1257.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  1258.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  1259.         "add %4, %1                     \n\t"\

  1260.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  1261.         \

  1262.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  1263.         "add %4, %1                     \n\t"\

  1264.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  1265.         \

  1266.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\

  1267.         "add %4, %1                     \n\t"\

  1268.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\

  1269.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\

  1270.                 \

  1271.         "add $72, %0                    \n\t"\

  1272.         "add %6, %1                     \n\t"\

  1273.         "decl %2                        \n\t"\

  1274.         " jnz 1b                        \n\t"\

  1275.          \

  1276.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  1277.         : "r"((x86_reg)dstStride), "r"(2*(x86_reg)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(x86_reg)dstStride)\

  1278.         : "memory"\

  1279.    );\

  1280. }\

  1281. \

  1282. static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  1283.     OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);\

  1284. }\

  1285. \

  1286. static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1287.     uint64_t temp[8];\

  1288.     uint8_t * const half= (uint8_t*)temp;\

  1289.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  1290.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  1291. }\

  1292. \

  1293. static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1294.     OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\

  1295. }\

  1296. \

  1297. static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1298.     uint64_t temp[8];\

  1299.     uint8_t * const half= (uint8_t*)temp;\

  1300.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  1301.     OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\

  1302. }\

  1303. \

  1304. static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1305.     uint64_t temp[8];\

  1306.     uint8_t * const half= (uint8_t*)temp;\

  1307.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  1308.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  1309. }\

  1310. \

  1311. static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1312.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\

  1313. }\

  1314. \

  1315. static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1316.     uint64_t temp[8];\

  1317.     uint8_t * const half= (uint8_t*)temp;\

  1318.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  1319.     OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\

  1320. }\

  1321. static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1322.     uint64_t half[8 + 9];\

  1323.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1324.     uint8_t * const halfHV= ((uint8_t*)half);\

  1325.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1326.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  1327.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1328.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  1329. }\

  1330. static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1331.     uint64_t half[8 + 9];\

  1332.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1333.     uint8_t * const halfHV= ((uint8_t*)half);\

  1334.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1335.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  1336.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1337.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  1338. }\

  1339. static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1340.     uint64_t half[8 + 9];\

  1341.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1342.     uint8_t * const halfHV= ((uint8_t*)half);\

  1343.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1344.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  1345.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1346.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  1347. }\

  1348. static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1349.     uint64_t half[8 + 9];\

  1350.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1351.     uint8_t * const halfHV= ((uint8_t*)half);\

  1352.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1353.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  1354.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1355.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  1356. }\

  1357. static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1358.     uint64_t half[8 + 9];\

  1359.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1360.     uint8_t * const halfHV= ((uint8_t*)half);\

  1361.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1362.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1363.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  1364. }\

  1365. static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1366.     uint64_t half[8 + 9];\

  1367.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  1368.     uint8_t * const halfHV= ((uint8_t*)half);\

  1369.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1370.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  1371.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  1372. }\

  1373. static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1374.     uint64_t half[8 + 9];\

  1375.     uint8_t * const halfH= ((uint8_t*)half);\

  1376.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1377.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  1378.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  1379. }\

  1380. static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1381.     uint64_t half[8 + 9];\

  1382.     uint8_t * const halfH= ((uint8_t*)half);\

  1383.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1384.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  1385.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  1386. }\

  1387. static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1388.     uint64_t half[9];\

  1389.     uint8_t * const halfH= ((uint8_t*)half);\

  1390.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  1391.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  1392. }\

  1393. static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  1394.     OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);\

  1395. }\

  1396. \

  1397. static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1398.     uint64_t temp[32];\

  1399.     uint8_t * const half= (uint8_t*)temp;\

  1400.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  1401.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  1402. }\

  1403. \

  1404. static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1405.     OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\

  1406. }\

  1407. \

  1408. static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1409.     uint64_t temp[32];\

  1410.     uint8_t * const half= (uint8_t*)temp;\

  1411.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  1412.     OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\

  1413. }\

  1414. \

  1415. static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1416.     uint64_t temp[32];\

  1417.     uint8_t * const half= (uint8_t*)temp;\

  1418.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  1419.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  1420. }\

  1421. \

  1422. static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1423.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\

  1424. }\

  1425. \

  1426. static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1427.     uint64_t temp[32];\

  1428.     uint8_t * const half= (uint8_t*)temp;\

  1429.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  1430.     OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\

  1431. }\

  1432. static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1433.     uint64_t half[16*2 + 17*2];\

  1434.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1435.     uint8_t * const halfHV= ((uint8_t*)half);\

  1436.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1437.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  1438.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1439.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  1440. }\

  1441. static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1442.     uint64_t half[16*2 + 17*2];\

  1443.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1444.     uint8_t * const halfHV= ((uint8_t*)half);\

  1445.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1446.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  1447.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1448.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  1449. }\

  1450. static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1451.     uint64_t half[16*2 + 17*2];\

  1452.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1453.     uint8_t * const halfHV= ((uint8_t*)half);\

  1454.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1455.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  1456.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1457.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  1458. }\

  1459. static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1460.     uint64_t half[16*2 + 17*2];\

  1461.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1462.     uint8_t * const halfHV= ((uint8_t*)half);\

  1463.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1464.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  1465.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1466.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  1467. }\

  1468. static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1469.     uint64_t half[16*2 + 17*2];\

  1470.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1471.     uint8_t * const halfHV= ((uint8_t*)half);\

  1472.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1473.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1474.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  1475. }\

  1476. static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1477.     uint64_t half[16*2 + 17*2];\

  1478.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  1479.     uint8_t * const halfHV= ((uint8_t*)half);\

  1480.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1481.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  1482.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  1483. }\

  1484. static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1485.     uint64_t half[17*2];\

  1486.     uint8_t * const halfH= ((uint8_t*)half);\

  1487.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1488.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  1489.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  1490. }\

  1491. static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1492.     uint64_t half[17*2];\

  1493.     uint8_t * const halfH= ((uint8_t*)half);\

  1494.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1495.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  1496.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  1497. }\

  1498. static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1499.     uint64_t half[17*2];\

  1500.     uint8_t * const halfH= ((uint8_t*)half);\

  1501.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  1502.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  1503. }

  1504. 

  1505. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"

  1506. #define AVG_3DNOW_OP(a,b,temp, size) \

  1507. "mov" #size " " #b ", " #temp "   \n\t"\

  1508. "pavgusb " #temp ", " #a "        \n\t"\

  1509. "mov" #size " " #a ", " #b "      \n\t"

  1510. #define AVG_MMX2_OP(a,b,temp, size) \

  1511. "mov" #size " " #b ", " #temp "   \n\t"\

  1512. "pavgb " #temp ", " #a "          \n\t"\

  1513. "mov" #size " " #a ", " #b "      \n\t"

  1514. 

  1515. QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)

  1516. QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)

  1517. QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)

  1518. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)

  1519. QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)

  1520. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)

  1521. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)

  1522. QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)

  1523. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

  1524. 

  1525. /***********************************/

  1526. /* bilinear qpel: not compliant to any spec, only for -lavdopts fast */

  1527. 

  1528. #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\

  1529. static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1530.     OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\

  1531. }

  1532. #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\

  1533. static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1534.     OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\

  1535. }

  1536. 

  1537. #define QPEL_2TAP(OPNAME, SIZE, MMX)\

  1538. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\

  1539. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\

  1540. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\

  1541. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\

  1542.                           OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\

  1543. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\

  1544.                           OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\

  1545. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\

  1546.                           OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\

  1547. static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1548.     OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\

  1549. }\

  1550. static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1551.     OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\

  1552. }\

  1553. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,         1,       0)\

  1554. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,        -1,       0)\

  1555. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,         stride,  0)\

  1556. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,   -stride,  0)\

  1557. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,         stride,  1)\

  1558. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,         stride, -1)\

  1559. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,   -stride,  1)\

  1560. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\

  1561. 

  1562. QPEL_2TAP(put_, 16, mmx2)

  1563. QPEL_2TAP(avg_, 16, mmx2)

  1564. QPEL_2TAP(put_,  8, mmx2)

  1565. QPEL_2TAP(avg_,  8, mmx2)

  1566. QPEL_2TAP(put_, 16, 3dnow)

  1567. QPEL_2TAP(avg_, 16, 3dnow)

  1568. QPEL_2TAP(put_,  8, 3dnow)

  1569. QPEL_2TAP(avg_,  8, 3dnow)

  1570. 

  1571. 

  1572. #if 0

  1573. static void just_return(void) { return; }

  1574. #endif

  1575. 

  1576. #if HAVE_YASM

  1577. typedef void emu_edge_core_func (uint8_t *buf, const uint8_t *src,

  1578.                                  x86_reg linesize, x86_reg start_y,

  1579.                                  x86_reg end_y, x86_reg block_h,

  1580.                                  x86_reg start_x, x86_reg end_x,

  1581.                                  x86_reg block_w);

  1582. extern emu_edge_core_func ff_emu_edge_core_mmx;

  1583. extern emu_edge_core_func ff_emu_edge_core_sse;

  1584. 

  1585. static av_always_inline

  1586. void emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,

  1587.                       int block_w, int block_h,

  1588.                       int src_x, int src_y, int w, int h,

  1589.                       emu_edge_core_func *core_fn)

  1590. {

  1591.     int start_y, start_x, end_y, end_x, src_y_add=0;

  1592. 

  1593.     if(src_y>= h){

  1594.         src_y_add = h-1-src_y;

  1595.         src_y=h-1;

  1596.     }else if(src_y<=-block_h){

  1597.         src_y_add = 1-block_h-src_y;

  1598.         src_y=1-block_h;

  1599.     }

  1600.     if(src_x>= w){

  1601.         src+= (w-1-src_x);

  1602.         src_x=w-1;

  1603.     }else if(src_x<=-block_w){

  1604.         src+= (1-block_w-src_x);

  1605.         src_x=1-block_w;

  1606.     }

  1607. 

  1608.     start_y= FFMAX(0, -src_y);

  1609.     start_x= FFMAX(0, -src_x);

  1610.     end_y= FFMIN(block_h, h-src_y);

  1611.     end_x= FFMIN(block_w, w-src_x);

  1612.     assert(start_x < end_x && block_w > 0);

  1613.     assert(start_y < end_y && block_h > 0);

  1614. 

  1615.     // fill in the to-be-copied part plus all above/below

  1616.     src += (src_y_add+start_y)*linesize + start_x;

  1617.     buf += start_x;

  1618.     core_fn(buf, src, linesize, start_y, end_y, block_h, start_x, end_x, block_w);

  1619. }

  1620. 

  1621. #if ARCH_X86_32

  1622. static av_noinline

  1623. void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src, int linesize,

  1624.                           int block_w, int block_h,

  1625.                           int src_x, int src_y, int w, int h)

  1626. {

  1627.     emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,

  1628.                      w, h, &ff_emu_edge_core_mmx);

  1629. }

  1630. #endif

  1631. static av_noinline

  1632. void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src, int linesize,

  1633.                           int block_w, int block_h,

  1634.                           int src_x, int src_y, int w, int h)

  1635. {

  1636.     emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,

  1637.                      w, h, &ff_emu_edge_core_sse);

  1638. }

  1639. #endif /* HAVE_YASM */

  1640. 

  1641. typedef void emulated_edge_mc_func (uint8_t *dst, const uint8_t *src,

  1642.                                     int linesize, int block_w, int block_h,

  1643.                                     int src_x, int src_y, int w, int h);

  1644. 

  1645. static av_always_inline

  1646. void gmc(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  1647.          int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height,

  1648.          emulated_edge_mc_func *emu_edge_fn)

  1649. {

  1650.     const int w = 8;

  1651.     const int ix = ox>>(16+shift);

  1652.     const int iy = oy>>(16+shift);

  1653.     const int oxs = ox>>4;

  1654.     const int oys = oy>>4;

  1655.     const int dxxs = dxx>>4;

  1656.     const int dxys = dxy>>4;

  1657.     const int dyxs = dyx>>4;

  1658.     const int dyys = dyy>>4;

  1659.     const uint16_t r4[4] = {r,r,r,r};

  1660.     const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};

  1661.     const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};

  1662.     const uint64_t shift2 = 2*shift;

  1663.     uint8_t edge_buf[(h+1)*stride];

  1664.     int x, y;

  1665. 

  1666.     const int dxw = (dxx-(1<<(16+shift)))*(w-1);

  1667.     const int dyh = (dyy-(1<<(16+shift)))*(h-1);

  1668.     const int dxh = dxy*(h-1);

  1669.     const int dyw = dyx*(w-1);

  1670.     if( // non-constant fullpel offset (3% of blocks)

  1671.         ((ox^(ox+dxw)) | (ox^(ox+dxh)) | (ox^(ox+dxw+dxh)) |

  1672.          (oy^(oy+dyw)) | (oy^(oy+dyh)) | (oy^(oy+dyw+dyh))) >> (16+shift)

  1673.         // uses more than 16 bits of subpel mv (only at huge resolution)

  1674.         || (dxx|dxy|dyx|dyy)&15 )

  1675.     {

  1676.         //FIXME could still use mmx for some of the rows

  1677.         ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);

  1678.         return;

  1679.     }

  1680. 

  1681.     src += ix + iy*stride;

  1682.     if( (unsigned)ix >= width-w ||

  1683.         (unsigned)iy >= height-h )

  1684.     {

  1685.         emu_edge_fn(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);

  1686.         src = edge_buf;

  1687.     }

  1688. 

  1689.     __asm__ volatile(

  1690.         "movd         %0, %%mm6 \n\t"

  1691.         "pxor      %%mm7, %%mm7 \n\t"

  1692.         "punpcklwd %%mm6, %%mm6 \n\t"

  1693.         "punpcklwd %%mm6, %%mm6 \n\t"

  1694.         :: "r"(1<<shift)

  1695.     );

  1696. 

  1697.     for(x=0; x<w; x+=4){

  1698.         uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),

  1699.                             oxs - dxys + dxxs*(x+1),

  1700.                             oxs - dxys + dxxs*(x+2),

  1701.                             oxs - dxys + dxxs*(x+3) };

  1702.         uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),

  1703.                             oys - dyys + dyxs*(x+1),

  1704.                             oys - dyys + dyxs*(x+2),

  1705.                             oys - dyys + dyxs*(x+3) };

  1706. 

  1707.         for(y=0; y<h; y++){

  1708.             __asm__ volatile(

  1709.                 "movq   %0,  %%mm4 \n\t"

  1710.                 "movq   %1,  %%mm5 \n\t"

  1711.                 "paddw  %2,  %%mm4 \n\t"

  1712.                 "paddw  %3,  %%mm5 \n\t"

  1713.                 "movq   %%mm4, %0  \n\t"

  1714.                 "movq   %%mm5, %1  \n\t"

  1715.                 "psrlw  $12, %%mm4 \n\t"

  1716.                 "psrlw  $12, %%mm5 \n\t"

  1717.                 : "+m"(*dx4), "+m"(*dy4)

  1718.                 : "m"(*dxy4), "m"(*dyy4)

  1719.             );

  1720. 

  1721.             __asm__ volatile(

  1722.                 "movq   %%mm6, %%mm2 \n\t"

  1723.                 "movq   %%mm6, %%mm1 \n\t"

  1724.                 "psubw  %%mm4, %%mm2 \n\t"

  1725.                 "psubw  %%mm5, %%mm1 \n\t"

  1726.                 "movq   %%mm2, %%mm0 \n\t"

  1727.                 "movq   %%mm4, %%mm3 \n\t"

  1728.                 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)

  1729.                 "pmullw %%mm5, %%mm3 \n\t" // dx*dy

  1730.                 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy

  1731.                 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)

  1732. 

  1733.                 "movd   %4,    %%mm5 \n\t"

  1734.                 "movd   %3,    %%mm4 \n\t"

  1735.                 "punpcklbw %%mm7, %%mm5 \n\t"

  1736.                 "punpcklbw %%mm7, %%mm4 \n\t"

  1737.                 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy

  1738.                 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy

  1739. 

  1740.                 "movd   %2,    %%mm5 \n\t"

  1741.                 "movd   %1,    %%mm4 \n\t"

  1742.                 "punpcklbw %%mm7, %%mm5 \n\t"

  1743.                 "punpcklbw %%mm7, %%mm4 \n\t"

  1744.                 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)

  1745.                 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)

  1746.                 "paddw  %5,    %%mm1 \n\t"

  1747.                 "paddw  %%mm3, %%mm2 \n\t"

  1748.                 "paddw  %%mm1, %%mm0 \n\t"

  1749.                 "paddw  %%mm2, %%mm0 \n\t"

  1750. 

  1751.                 "psrlw    %6,    %%mm0 \n\t"

  1752.                 "packuswb %%mm0, %%mm0 \n\t"

  1753.                 "movd     %%mm0, %0    \n\t"

  1754. 

  1755.                 : "=m"(dst[x+y*stride])

  1756.                 : "m"(src[0]), "m"(src[1]),

  1757.                   "m"(src[stride]), "m"(src[stride+1]),

  1758.                   "m"(*r4), "m"(shift2)

  1759.             );

  1760.             src += stride;

  1761.         }

  1762.         src += 4-h*stride;

  1763.     }

  1764. }

  1765. 

  1766. #if HAVE_YASM

  1767. #if ARCH_X86_32

  1768. static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  1769.                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)

  1770. {

  1771.     gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,

  1772.         width, height, &emulated_edge_mc_mmx);

  1773. }

  1774. #endif

  1775. static void gmc_sse(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  1776.                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)

  1777. {

  1778.     gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,

  1779.         width, height, &emulated_edge_mc_sse);

  1780. }

  1781. #else

  1782. static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  1783.                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)

  1784. {

  1785.     gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,

  1786.         width, height, &ff_emulated_edge_mc);

  1787. }

  1788. #endif

  1789. 

  1790. #define PREFETCH(name, op) \

  1791. static void name(void *mem, int stride, int h){\

  1792.     const uint8_t *p= mem;\

  1793.     do{\

  1794.         __asm__ volatile(#op" %0" :: "m"(*p));\

  1795.         p+= stride;\

  1796.     }while(--h);\

  1797. }

  1798. PREFETCH(prefetch_mmx2,  prefetcht0)

  1799. PREFETCH(prefetch_3dnow, prefetch)

  1800. #undef PREFETCH

  1801. 

  1802. #include "h264_qpel_mmx.c"

  1803. 

  1804. void ff_put_h264_chroma_mc8_mmx_rnd   (uint8_t *dst, uint8_t *src,

  1805.                                        int stride, int h, int x, int y);

  1806. void ff_put_rv40_chroma_mc8_mmx       (uint8_t *dst, uint8_t *src,

  1807.                                        int stride, int h, int x, int y);

  1808. void ff_avg_h264_chroma_mc8_mmx2_rnd  (uint8_t *dst, uint8_t *src,

  1809.                                        int stride, int h, int x, int y);

  1810. void ff_avg_rv40_chroma_mc8_mmx2      (uint8_t *dst, uint8_t *src,

  1811.                                        int stride, int h, int x, int y);

  1812. void ff_avg_h264_chroma_mc8_3dnow_rnd (uint8_t *dst, uint8_t *src,

  1813.                                        int stride, int h, int x, int y);

  1814. void ff_avg_rv40_chroma_mc8_3dnow     (uint8_t *dst, uint8_t *src,

  1815.                                        int stride, int h, int x, int y);

  1816. 

  1817. void ff_put_h264_chroma_mc4_mmx       (uint8_t *dst, uint8_t *src,

  1818.                                        int stride, int h, int x, int y);

  1819. void ff_put_rv40_chroma_mc4_mmx       (uint8_t *dst, uint8_t *src,

  1820.                                        int stride, int h, int x, int y);

  1821. void ff_avg_h264_chroma_mc4_mmx2      (uint8_t *dst, uint8_t *src,

  1822.                                        int stride, int h, int x, int y);

  1823. void ff_avg_rv40_chroma_mc4_mmx2      (uint8_t *dst, uint8_t *src,

  1824.                                        int stride, int h, int x, int y);

  1825. void ff_avg_h264_chroma_mc4_3dnow     (uint8_t *dst, uint8_t *src,

  1826.                                        int stride, int h, int x, int y);

  1827. void ff_avg_rv40_chroma_mc4_3dnow     (uint8_t *dst, uint8_t *src,

  1828.                                        int stride, int h, int x, int y);

  1829. 

  1830. void ff_put_h264_chroma_mc2_mmx2      (uint8_t *dst, uint8_t *src,

  1831.                                        int stride, int h, int x, int y);

  1832. void ff_avg_h264_chroma_mc2_mmx2      (uint8_t *dst, uint8_t *src,

  1833.                                        int stride, int h, int x, int y);

  1834. 

  1835. void ff_put_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,

  1836.                                        int stride, int h, int x, int y);

  1837. void ff_put_h264_chroma_mc4_ssse3     (uint8_t *dst, uint8_t *src,

  1838.                                        int stride, int h, int x, int y);

  1839. 

  1840. void ff_avg_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,

  1841.                                        int stride, int h, int x, int y);

  1842. void ff_avg_h264_chroma_mc4_ssse3     (uint8_t *dst, uint8_t *src,

  1843.                                        int stride, int h, int x, int y);

  1844. 

  1845. #define CHROMA_MC(OP, NUM, DEPTH, OPT) \

  1846. void ff_ ## OP ## _h264_chroma_mc ## NUM ## _ ## DEPTH ## _ ## OPT \

  1847.                                       (uint8_t *dst, uint8_t *src,\

  1848.                                        int stride, int h, int x, int y);

  1849. 

  1850. CHROMA_MC(put, 2, 10, mmxext)

  1851. CHROMA_MC(avg, 2, 10, mmxext)

  1852. CHROMA_MC(put, 4, 10, mmxext)

  1853. CHROMA_MC(avg, 4, 10, mmxext)

  1854. CHROMA_MC(put, 8, 10, sse2)

  1855. CHROMA_MC(avg, 8, 10, sse2)

  1856. CHROMA_MC(put, 8, 10, avx)

  1857. CHROMA_MC(avg, 8, 10, avx)

  1858. 

  1859. /* CAVS specific */

  1860. void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  1861.     put_pixels8_mmx(dst, src, stride, 8);

  1862. }

  1863. void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  1864.     avg_pixels8_mmx(dst, src, stride, 8);

  1865. }

  1866. void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  1867.     put_pixels16_mmx(dst, src, stride, 16);

  1868. }

  1869. void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  1870.     avg_pixels16_mmx(dst, src, stride, 16);

  1871. }

  1872. 

  1873. /* VC1 specific */

  1874. void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {

  1875.     put_pixels8_mmx(dst, src, stride, 8);

  1876. }

  1877. void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd) {

  1878.     avg_pixels8_mmx2(dst, src, stride, 8);

  1879. }

  1880. 

  1881. /* XXX: those functions should be suppressed ASAP when all IDCTs are

  1882.    converted */

  1883. #if CONFIG_GPL

  1884. static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  1885. {

  1886.     ff_mmx_idct (block);

  1887.     ff_put_pixels_clamped_mmx(block, dest, line_size);

  1888. }

  1889. static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  1890. {

  1891.     ff_mmx_idct (block);

  1892.     ff_add_pixels_clamped_mmx(block, dest, line_size);

  1893. }

  1894. static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  1895. {

  1896.     ff_mmxext_idct (block);

  1897.     ff_put_pixels_clamped_mmx(block, dest, line_size);

  1898. }

  1899. static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  1900. {

  1901.     ff_mmxext_idct (block);

  1902.     ff_add_pixels_clamped_mmx(block, dest, line_size);

  1903. }

  1904. #endif

  1905. static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)

  1906. {

  1907.     ff_idct_xvid_mmx (block);

  1908.     ff_put_pixels_clamped_mmx(block, dest, line_size);

  1909. }

  1910. static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)

  1911. {

  1912.     ff_idct_xvid_mmx (block);

  1913.     ff_add_pixels_clamped_mmx(block, dest, line_size);

  1914. }

  1915. static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)

  1916. {

  1917.     ff_idct_xvid_mmx2 (block);

  1918.     ff_put_pixels_clamped_mmx(block, dest, line_size);

  1919. }

  1920. static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)

  1921. {

  1922.     ff_idct_xvid_mmx2 (block);

  1923.     ff_add_pixels_clamped_mmx(block, dest, line_size);

  1924. }

  1925. 

  1926. static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)

  1927. {

  1928.     int i;

  1929.     __asm__ volatile("pxor %%mm7, %%mm7":);

  1930.     for(i=0; i<blocksize; i+=2) {

  1931.         __asm__ volatile(

  1932.             "movq    %0,    %%mm0 \n\t"

  1933.             "movq    %1,    %%mm1 \n\t"

  1934.             "movq    %%mm0, %%mm2 \n\t"

  1935.             "movq    %%mm1, %%mm3 \n\t"

  1936.             "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0

  1937.             "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0

  1938.             "pslld   $31,   %%mm2 \n\t" // keep only the sign bit

  1939.             "pxor    %%mm2, %%mm1 \n\t"

  1940.             "movq    %%mm3, %%mm4 \n\t"

  1941.             "pand    %%mm1, %%mm3 \n\t"

  1942.             "pandn   %%mm1, %%mm4 \n\t"

  1943.             "pfadd   %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))

  1944.             "pfsub   %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))

  1945.             "movq    %%mm3, %1    \n\t"

  1946.             "movq    %%mm0, %0    \n\t"

  1947.             :"+m"(mag[i]), "+m"(ang[i])

  1948.             ::"memory"

  1949.         );

  1950.     }

  1951.     __asm__ volatile("femms");

  1952. }

  1953. static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)

  1954. {

  1955.     int i;

  1956. 

  1957.     __asm__ volatile(

  1958.             "movaps  %0,     %%xmm5 \n\t"

  1959.         ::"m"(ff_pdw_80000000[0])

  1960.     );

  1961.     for(i=0; i<blocksize; i+=4) {

  1962.         __asm__ volatile(

  1963.             "movaps  %0,     %%xmm0 \n\t"

  1964.             "movaps  %1,     %%xmm1 \n\t"

  1965.             "xorps   %%xmm2, %%xmm2 \n\t"

  1966.             "xorps   %%xmm3, %%xmm3 \n\t"

  1967.             "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0

  1968.             "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0

  1969.             "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit

  1970.             "xorps   %%xmm2, %%xmm1 \n\t"

  1971.             "movaps  %%xmm3, %%xmm4 \n\t"

  1972.             "andps   %%xmm1, %%xmm3 \n\t"

  1973.             "andnps  %%xmm1, %%xmm4 \n\t"

  1974.             "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))

  1975.             "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))

  1976.             "movaps  %%xmm3, %1     \n\t"

  1977.             "movaps  %%xmm0, %0     \n\t"

  1978.             :"+m"(mag[i]), "+m"(ang[i])

  1979.             ::"memory"

  1980.         );

  1981.     }

  1982. }

  1983. 

  1984. #define IF1(x) x

  1985. #define IF0(x)

  1986. 

  1987. #define MIX5(mono,stereo)\

  1988.     __asm__ volatile(\

  1989.         "movss          0(%2), %%xmm5 \n"\

  1990.         "movss          8(%2), %%xmm6 \n"\

  1991.         "movss         24(%2), %%xmm7 \n"\

  1992.         "shufps    $0, %%xmm5, %%xmm5 \n"\

  1993.         "shufps    $0, %%xmm6, %%xmm6 \n"\

  1994.         "shufps    $0, %%xmm7, %%xmm7 \n"\

  1995.         "1: \n"\

  1996.         "movaps       (%0,%1), %%xmm0 \n"\

  1997.         "movaps  0x400(%0,%1), %%xmm1 \n"\

  1998.         "movaps  0x800(%0,%1), %%xmm2 \n"\

  1999.         "movaps  0xc00(%0,%1), %%xmm3 \n"\

  2000.         "movaps 0x1000(%0,%1), %%xmm4 \n"\

  2001.         "mulps         %%xmm5, %%xmm0 \n"\

  2002.         "mulps         %%xmm6, %%xmm1 \n"\

  2003.         "mulps         %%xmm5, %%xmm2 \n"\

  2004.         "mulps         %%xmm7, %%xmm3 \n"\

  2005.         "mulps         %%xmm7, %%xmm4 \n"\

  2006.  stereo("addps         %%xmm1, %%xmm0 \n")\

  2007.         "addps         %%xmm1, %%xmm2 \n"\

  2008.         "addps         %%xmm3, %%xmm0 \n"\

  2009.         "addps         %%xmm4, %%xmm2 \n"\

  2010.    mono("addps         %%xmm2, %%xmm0 \n")\

  2011.         "movaps  %%xmm0,      (%0,%1) \n"\

  2012.  stereo("movaps  %%xmm2, 0x400(%0,%1) \n")\

  2013.         "add $16, %0 \n"\

  2014.         "jl 1b \n"\

  2015.         :"+&r"(i)\

  2016.         :"r"(samples[0]+len), "r"(matrix)\

  2017.         :XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \

  2018.                       "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\

  2019.          "memory"\

  2020.     );

  2021. 

  2022. #define MIX_MISC(stereo)\

  2023.     __asm__ volatile(\

  2024.         "1: \n"\

  2025.         "movaps  (%3,%0), %%xmm0 \n"\

  2026.  stereo("movaps   %%xmm0, %%xmm1 \n")\

  2027.         "mulps    %%xmm4, %%xmm0 \n"\

  2028.  stereo("mulps    %%xmm5, %%xmm1 \n")\

  2029.         "lea 1024(%3,%0), %1 \n"\

  2030.         "mov %5, %2 \n"\

  2031.         "2: \n"\

  2032.         "movaps   (%1),   %%xmm2 \n"\

  2033.  stereo("movaps   %%xmm2, %%xmm3 \n")\

  2034.         "mulps   (%4,%2), %%xmm2 \n"\

  2035.  stereo("mulps 16(%4,%2), %%xmm3 \n")\

  2036.         "addps    %%xmm2, %%xmm0 \n"\

  2037.  stereo("addps    %%xmm3, %%xmm1 \n")\

  2038.         "add $1024, %1 \n"\

  2039.         "add $32, %2 \n"\

  2040.         "jl 2b \n"\

  2041.         "movaps   %%xmm0,     (%3,%0) \n"\

  2042.  stereo("movaps   %%xmm1, 1024(%3,%0) \n")\

  2043.         "add $16, %0 \n"\

  2044.         "jl 1b \n"\

  2045.         :"+&r"(i), "=&r"(j), "=&r"(k)\

  2046.         :"r"(samples[0]+len), "r"(matrix_simd+in_ch), "g"((intptr_t)-32*(in_ch-1))\

  2047.         :"memory"\

  2048.     );

  2049. 

  2050. static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)

  2051. {

  2052.     int (*matrix_cmp)[2] = (int(*)[2])matrix;

  2053.     intptr_t i,j,k;

  2054. 

  2055.     i = -len*sizeof(float);

  2056.     if(in_ch == 5 && out_ch == 2 && !(matrix_cmp[0][1]|matrix_cmp[2][0]|matrix_cmp[3][1]|matrix_cmp[4][0]|(matrix_cmp[1][0]^matrix_cmp[1][1])|(matrix_cmp[0][0]^matrix_cmp[2][1]))) {

  2057.         MIX5(IF0,IF1);

  2058.     } else if(in_ch == 5 && out_ch == 1 && matrix_cmp[0][0]==matrix_cmp[2][0] && matrix_cmp[3][0]==matrix_cmp[4][0]) {

  2059.         MIX5(IF1,IF0);

  2060.     } else {

  2061.         DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];

  2062.         j = 2*in_ch*sizeof(float);

  2063.         __asm__ volatile(

  2064.             "1: \n"

  2065.             "sub $8, %0 \n"

  2066.             "movss     (%2,%0), %%xmm4 \n"

  2067.             "movss    4(%2,%0), %%xmm5 \n"

  2068.             "shufps $0, %%xmm4, %%xmm4 \n"

  2069.             "shufps $0, %%xmm5, %%xmm5 \n"

  2070.             "movaps %%xmm4,   (%1,%0,4) \n"

  2071.             "movaps %%xmm5, 16(%1,%0,4) \n"

  2072.             "jg 1b \n"

  2073.             :"+&r"(j)

  2074.             :"r"(matrix_simd), "r"(matrix)

  2075.             :"memory"

  2076.         );

  2077.         if(out_ch == 2) {

  2078.             MIX_MISC(IF1);

  2079.         } else {

  2080.             MIX_MISC(IF0);

  2081.         }

  2082.     }

  2083. }

  2084. 

  2085. static void vector_fmul_3dnow(float *dst, const float *src0, const float *src1, int len){

  2086.     x86_reg i = (len-4)*4;

  2087.     __asm__ volatile(

  2088.         "1: \n\t"

  2089.         "movq    (%2,%0), %%mm0 \n\t"

  2090.         "movq   8(%2,%0), %%mm1 \n\t"

  2091.         "pfmul   (%3,%0), %%mm0 \n\t"

  2092.         "pfmul  8(%3,%0), %%mm1 \n\t"

  2093.         "movq   %%mm0,  (%1,%0) \n\t"

  2094.         "movq   %%mm1, 8(%1,%0) \n\t"

  2095.         "sub  $16, %0 \n\t"

  2096.         "jge 1b \n\t"

  2097.         "femms  \n\t"

  2098.         :"+r"(i)

  2099.         :"r"(dst), "r"(src0), "r"(src1)

  2100.         :"memory"

  2101.     );

  2102. }

  2103. static void vector_fmul_sse(float *dst, const float *src0, const float *src1, int len){

  2104.     x86_reg i = (len-8)*4;

  2105.     __asm__ volatile(

  2106.         "1: \n\t"

  2107.         "movaps    (%2,%0), %%xmm0 \n\t"

  2108.         "movaps  16(%2,%0), %%xmm1 \n\t"

  2109.         "mulps     (%3,%0), %%xmm0 \n\t"

  2110.         "mulps   16(%3,%0), %%xmm1 \n\t"

  2111.         "movaps  %%xmm0,   (%1,%0) \n\t"

  2112.         "movaps  %%xmm1, 16(%1,%0) \n\t"

  2113.         "sub  $32, %0 \n\t"

  2114.         "jge 1b \n\t"

  2115.         :"+r"(i)

  2116.         :"r"(dst), "r"(src0), "r"(src1)

  2117.         :"memory"

  2118.     );

  2119. }

  2120. 

  2121. static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){

  2122.     x86_reg i = len*4-16;

  2123.     __asm__ volatile(

  2124.         "1: \n\t"

  2125.         "pswapd   8(%1), %%mm0 \n\t"

  2126.         "pswapd    (%1), %%mm1 \n\t"

  2127.         "pfmul  (%3,%0), %%mm0 \n\t"

  2128.         "pfmul 8(%3,%0), %%mm1 \n\t"

  2129.         "movq  %%mm0,  (%2,%0) \n\t"

  2130.         "movq  %%mm1, 8(%2,%0) \n\t"

  2131.         "add   $16, %1 \n\t"

  2132.         "sub   $16, %0 \n\t"

  2133.         "jge   1b \n\t"

  2134.         :"+r"(i), "+r"(src1)

  2135.         :"r"(dst), "r"(src0)

  2136.     );

  2137.     __asm__ volatile("femms");

  2138. }

  2139. static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){

  2140.     x86_reg i = len*4-32;

  2141.     __asm__ volatile(

  2142.         "1: \n\t"

  2143.         "movaps        16(%1), %%xmm0 \n\t"

  2144.         "movaps          (%1), %%xmm1 \n\t"

  2145.         "shufps $0x1b, %%xmm0, %%xmm0 \n\t"

  2146.         "shufps $0x1b, %%xmm1, %%xmm1 \n\t"

  2147.         "mulps        (%3,%0), %%xmm0 \n\t"

  2148.         "mulps      16(%3,%0), %%xmm1 \n\t"

  2149.         "movaps     %%xmm0,   (%2,%0) \n\t"

  2150.         "movaps     %%xmm1, 16(%2,%0) \n\t"

  2151.         "add    $32, %1 \n\t"

  2152.         "sub    $32, %0 \n\t"

  2153.         "jge    1b \n\t"

  2154.         :"+r"(i), "+r"(src1)

  2155.         :"r"(dst), "r"(src0)

  2156.     );

  2157. }

  2158. 

  2159. static void vector_fmul_add_3dnow(float *dst, const float *src0, const float *src1,

  2160.                                   const float *src2, int len){

  2161.     x86_reg i = (len-4)*4;

  2162.     __asm__ volatile(

  2163.         "1: \n\t"

  2164.         "movq    (%2,%0), %%mm0 \n\t"

  2165.         "movq   8(%2,%0), %%mm1 \n\t"

  2166.         "pfmul   (%3,%0), %%mm0 \n\t"

  2167.         "pfmul  8(%3,%0), %%mm1 \n\t"

  2168.         "pfadd   (%4,%0), %%mm0 \n\t"

  2169.         "pfadd  8(%4,%0), %%mm1 \n\t"

  2170.         "movq  %%mm0,   (%1,%0) \n\t"

  2171.         "movq  %%mm1,  8(%1,%0) \n\t"

  2172.         "sub  $16, %0 \n\t"

  2173.         "jge  1b \n\t"

  2174.         :"+r"(i)

  2175.         :"r"(dst), "r"(src0), "r"(src1), "r"(src2)

  2176.         :"memory"

  2177.     );

  2178.     __asm__ volatile("femms");

  2179. }

  2180. static void vector_fmul_add_sse(float *dst, const float *src0, const float *src1,

  2181.                                 const float *src2, int len){

  2182.     x86_reg i = (len-8)*4;

  2183.     __asm__ volatile(

  2184.         "1: \n\t"

  2185.         "movaps   (%2,%0), %%xmm0 \n\t"

  2186.         "movaps 16(%2,%0), %%xmm1 \n\t"

  2187.         "mulps    (%3,%0), %%xmm0 \n\t"

  2188.         "mulps  16(%3,%0), %%xmm1 \n\t"

  2189.         "addps    (%4,%0), %%xmm0 \n\t"

  2190.         "addps  16(%4,%0), %%xmm1 \n\t"

  2191.         "movaps %%xmm0,   (%1,%0) \n\t"

  2192.         "movaps %%xmm1, 16(%1,%0) \n\t"

  2193.         "sub  $32, %0 \n\t"

  2194.         "jge  1b \n\t"

  2195.         :"+r"(i)

  2196.         :"r"(dst), "r"(src0), "r"(src1), "r"(src2)

  2197.         :"memory"

  2198.     );

  2199. }

  2200. 

  2201. #if HAVE_6REGS

  2202. static void vector_fmul_window_3dnow2(float *dst, const float *src0, const float *src1,

  2203.                                       const float *win, int len){

  2204.     x86_reg i = -len*4;

  2205.     x86_reg j = len*4-8;

  2206.     __asm__ volatile(

  2207.         "1: \n"

  2208.         "pswapd  (%5,%1), %%mm1 \n"

  2209.         "movq    (%5,%0), %%mm0 \n"

  2210.         "pswapd  (%4,%1), %%mm5 \n"

  2211.         "movq    (%3,%0), %%mm4 \n"

  2212.         "movq      %%mm0, %%mm2 \n"

  2213.         "movq      %%mm1, %%mm3 \n"

  2214.         "pfmul     %%mm4, %%mm2 \n" // src0[len+i]*win[len+i]

  2215.         "pfmul     %%mm5, %%mm3 \n" // src1[    j]*win[len+j]

  2216.         "pfmul     %%mm4, %%mm1 \n" // src0[len+i]*win[len+j]

  2217.         "pfmul     %%mm5, %%mm0 \n" // src1[    j]*win[len+i]

  2218.         "pfadd     %%mm3, %%mm2 \n"

  2219.         "pfsub     %%mm0, %%mm1 \n"

  2220.         "pswapd    %%mm2, %%mm2 \n"

  2221.         "movq      %%mm1, (%2,%0) \n"

  2222.         "movq      %%mm2, (%2,%1) \n"

  2223.         "sub $8, %1 \n"

  2224.         "add $8, %0 \n"

  2225.         "jl 1b \n"

  2226.         "femms \n"

  2227.         :"+r"(i), "+r"(j)

  2228.         :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)

  2229.     );

  2230. }

  2231. 

  2232. static void vector_fmul_window_sse(float *dst, const float *src0, const float *src1,

  2233.                                    const float *win, int len){

  2234.     x86_reg i = -len*4;

  2235.     x86_reg j = len*4-16;

  2236.     __asm__ volatile(

  2237.         "1: \n"

  2238.         "movaps       (%5,%1), %%xmm1 \n"

  2239.         "movaps       (%5,%0), %%xmm0 \n"

  2240.         "movaps       (%4,%1), %%xmm5 \n"

  2241.         "movaps       (%3,%0), %%xmm4 \n"

  2242.         "shufps $0x1b, %%xmm1, %%xmm1 \n"

  2243.         "shufps $0x1b, %%xmm5, %%xmm5 \n"

  2244.         "movaps        %%xmm0, %%xmm2 \n"

  2245.         "movaps        %%xmm1, %%xmm3 \n"

  2246.         "mulps         %%xmm4, %%xmm2 \n" // src0[len+i]*win[len+i]

  2247.         "mulps         %%xmm5, %%xmm3 \n" // src1[    j]*win[len+j]

  2248.         "mulps         %%xmm4, %%xmm1 \n" // src0[len+i]*win[len+j]

  2249.         "mulps         %%xmm5, %%xmm0 \n" // src1[    j]*win[len+i]

  2250.         "addps         %%xmm3, %%xmm2 \n"

  2251.         "subps         %%xmm0, %%xmm1 \n"

  2252.         "shufps $0x1b, %%xmm2, %%xmm2 \n"

  2253.         "movaps        %%xmm1, (%2,%0) \n"

  2254.         "movaps        %%xmm2, (%2,%1) \n"

  2255.         "sub $16, %1 \n"

  2256.         "add $16, %0 \n"

  2257.         "jl 1b \n"

  2258.         :"+r"(i), "+r"(j)

  2259.         :"r"(dst+len), "r"(src0+len), "r"(src1), "r"(win+len)

  2260.     );

  2261. }

  2262. #endif /* HAVE_6REGS */

  2263. 

  2264. static void vector_clipf_sse(float *dst, const float *src, float min, float max,

  2265.                              int len)

  2266. {

  2267.     x86_reg i = (len-16)*4;

  2268.     __asm__ volatile(

  2269.         "movss  %3, %%xmm4 \n"

  2270.         "movss  %4, %%xmm5 \n"

  2271.         "shufps $0, %%xmm4, %%xmm4 \n"

  2272.         "shufps $0, %%xmm5, %%xmm5 \n"

  2273.         "1: \n\t"

  2274.         "movaps    (%2,%0), %%xmm0 \n\t" // 3/1 on intel

  2275.         "movaps  16(%2,%0), %%xmm1 \n\t"

  2276.         "movaps  32(%2,%0), %%xmm2 \n\t"

  2277.         "movaps  48(%2,%0), %%xmm3 \n\t"

  2278.         "maxps      %%xmm4, %%xmm0 \n\t"

  2279.         "maxps      %%xmm4, %%xmm1 \n\t"

  2280.         "maxps      %%xmm4, %%xmm2 \n\t"

  2281.         "maxps      %%xmm4, %%xmm3 \n\t"

  2282.         "minps      %%xmm5, %%xmm0 \n\t"

  2283.         "minps      %%xmm5, %%xmm1 \n\t"

  2284.         "minps      %%xmm5, %%xmm2 \n\t"

  2285.         "minps      %%xmm5, %%xmm3 \n\t"

  2286.         "movaps  %%xmm0,   (%1,%0) \n\t"

  2287.         "movaps  %%xmm1, 16(%1,%0) \n\t"

  2288.         "movaps  %%xmm2, 32(%1,%0) \n\t"

  2289.         "movaps  %%xmm3, 48(%1,%0) \n\t"

  2290.         "sub  $64, %0 \n\t"

  2291.         "jge 1b \n\t"

  2292.         :"+&r"(i)

  2293.         :"r"(dst), "r"(src), "m"(min), "m"(max)

  2294.         :"memory"

  2295.     );

  2296. }

  2297. 

  2298. void ff_vp3_idct_mmx(int16_t *input_data);

  2299. void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block);

  2300. void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block);

  2301. 

  2302. void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int line_size, const DCTELEM *block);

  2303. 

  2304. void ff_vp3_v_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);

  2305. void ff_vp3_h_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);

  2306. 

  2307. void ff_vp3_idct_sse2(int16_t *input_data);

  2308. void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block);

  2309. void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block);

  2310. 

  2311. int32_t ff_scalarproduct_int16_mmx2(const int16_t *v1, const int16_t *v2, int order, int shift);

  2312. int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2, int order, int shift);

  2313. int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);

  2314. int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);

  2315. int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);

  2316. 

  2317. void ff_apply_window_int16_mmxext    (int16_t *output, const int16_t *input,

  2318.                                       const int16_t *window, unsigned int len);

  2319. void ff_apply_window_int16_mmxext_ba (int16_t *output, const int16_t *input,

  2320.                                       const int16_t *window, unsigned int len);

  2321. void ff_apply_window_int16_sse2      (int16_t *output, const int16_t *input,

  2322.                                       const int16_t *window, unsigned int len);

  2323. void ff_apply_window_int16_sse2_ba   (int16_t *output, const int16_t *input,

  2324.                                       const int16_t *window, unsigned int len);

  2325. void ff_apply_window_int16_ssse3     (int16_t *output, const int16_t *input,

  2326.                                       const int16_t *window, unsigned int len);

  2327. void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,

  2328.                                       const int16_t *window, unsigned int len);

  2329. 

  2330. void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);

  2331. int  ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, int w, int left);

  2332. int  ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, int w, int left);

  2333. 

  2334. float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);

  2335. 

  2336. void ff_vector_clip_int32_mmx     (int32_t *dst, const int32_t *src, int32_t min,

  2337.                                    int32_t max, unsigned int len);

  2338. void ff_vector_clip_int32_sse2    (int32_t *dst, const int32_t *src, int32_t min,

  2339.                                    int32_t max, unsigned int len);

  2340. void ff_vector_clip_int32_sse2_int(int32_t *dst, const int32_t *src, int32_t min,

  2341.                                    int32_t max, unsigned int len);

  2342. void ff_vector_clip_int32_sse41   (int32_t *dst, const int32_t *src, int32_t min,

  2343.                                    int32_t max, unsigned int len);

  2344. 

  2345. void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)

  2346. {

  2347.     int mm_flags = av_get_cpu_flags();

  2348.     const int high_bit_depth = avctx->codec_id == CODEC_ID_H264 && avctx->bits_per_raw_sample > 8;

  2349.     const int bit_depth = avctx->bits_per_raw_sample;

  2350. 

  2351.     if (avctx->dsp_mask) {

  2352.         if (avctx->dsp_mask & AV_CPU_FLAG_FORCE)

  2353.             mm_flags |= (avctx->dsp_mask & 0xffff);

  2354.         else

  2355.             mm_flags &= ~(avctx->dsp_mask & 0xffff);

  2356.     }

  2357. 

  2358. #if 0

  2359.     av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");

  2360.     if (mm_flags & AV_CPU_FLAG_MMX)

  2361.         av_log(avctx, AV_LOG_INFO, " mmx");

  2362.     if (mm_flags & AV_CPU_FLAG_MMX2)

  2363.         av_log(avctx, AV_LOG_INFO, " mmx2");

  2364.     if (mm_flags & AV_CPU_FLAG_3DNOW)

  2365.         av_log(avctx, AV_LOG_INFO, " 3dnow");

  2366.     if (mm_flags & AV_CPU_FLAG_SSE)

  2367.         av_log(avctx, AV_LOG_INFO, " sse");

  2368.     if (mm_flags & AV_CPU_FLAG_SSE2)

  2369.         av_log(avctx, AV_LOG_INFO, " sse2");

  2370.     av_log(avctx, AV_LOG_INFO, "\n");

  2371. #endif

  2372. 

  2373.     if (mm_flags & AV_CPU_FLAG_MMX) {

  2374.         const int idct_algo= avctx->idct_algo;

  2375. 

  2376.         if(avctx->lowres==0){

  2377.             if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){

  2378.                 c->idct_put= ff_simple_idct_put_mmx;

  2379.                 c->idct_add= ff_simple_idct_add_mmx;

  2380.                 c->idct    = ff_simple_idct_mmx;

  2381.                 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;

  2382. #if CONFIG_GPL

  2383.             }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){

  2384.                 if(mm_flags & AV_CPU_FLAG_MMX2){

  2385.                     c->idct_put= ff_libmpeg2mmx2_idct_put;

  2386.                     c->idct_add= ff_libmpeg2mmx2_idct_add;

  2387.                     c->idct    = ff_mmxext_idct;

  2388.                 }else{

  2389.                     c->idct_put= ff_libmpeg2mmx_idct_put;

  2390.                     c->idct_add= ff_libmpeg2mmx_idct_add;

  2391.                     c->idct    = ff_mmx_idct;

  2392.                 }

  2393.                 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;

  2394. #endif

  2395.             }else if((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER || CONFIG_VP6_DECODER) &&

  2396.                      idct_algo==FF_IDCT_VP3 && HAVE_YASM){

  2397.                 if(mm_flags & AV_CPU_FLAG_SSE2){

  2398.                     c->idct_put= ff_vp3_idct_put_sse2;

  2399.                     c->idct_add= ff_vp3_idct_add_sse2;

  2400.                     c->idct    = ff_vp3_idct_sse2;

  2401.                     c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;

  2402.                 }else{

  2403.                     c->idct_put= ff_vp3_idct_put_mmx;

  2404.                     c->idct_add= ff_vp3_idct_add_mmx;

  2405.                     c->idct    = ff_vp3_idct_mmx;

  2406.                     c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;

  2407.                 }

  2408.             }else if(idct_algo==FF_IDCT_CAVS){

  2409.                     c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;

  2410.             }else if(idct_algo==FF_IDCT_XVIDMMX){

  2411.                 if(mm_flags & AV_CPU_FLAG_SSE2){

  2412.                     c->idct_put= ff_idct_xvid_sse2_put;

  2413.                     c->idct_add= ff_idct_xvid_sse2_add;

  2414.                     c->idct    = ff_idct_xvid_sse2;

  2415.                     c->idct_permutation_type= FF_SSE2_IDCT_PERM;

  2416.                 }else if(mm_flags & AV_CPU_FLAG_MMX2){

  2417.                     c->idct_put= ff_idct_xvid_mmx2_put;

  2418.                     c->idct_add= ff_idct_xvid_mmx2_add;

  2419.                     c->idct    = ff_idct_xvid_mmx2;

  2420.                 }else{

  2421.                     c->idct_put= ff_idct_xvid_mmx_put;

  2422.                     c->idct_add= ff_idct_xvid_mmx_add;

  2423.                     c->idct    = ff_idct_xvid_mmx;

  2424.                 }

  2425.             }

  2426.         }

  2427. 

  2428.         c->put_pixels_clamped = ff_put_pixels_clamped_mmx;

  2429.         c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;

  2430.         c->add_pixels_clamped = ff_add_pixels_clamped_mmx;

  2431.         if (!high_bit_depth) {

  2432.         c->clear_block  = clear_block_mmx;

  2433.         c->clear_blocks = clear_blocks_mmx;

  2434.         if ((mm_flags & AV_CPU_FLAG_SSE) &&

  2435.             !(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)){

  2436.             /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */

  2437.             c->clear_block  = clear_block_sse;

  2438.             c->clear_blocks = clear_blocks_sse;

  2439.         }

  2440.         }

  2441. 

  2442. #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \

  2443.         c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \

  2444.         c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \

  2445.         c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \

  2446.         c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU

  2447. 

  2448.         if (!high_bit_depth) {

  2449.         SET_HPEL_FUNCS(put, 0, 16, mmx);

  2450.         SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);

  2451.         SET_HPEL_FUNCS(avg, 0, 16, mmx);

  2452.         SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);

  2453.         SET_HPEL_FUNCS(put, 1, 8, mmx);

  2454.         SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);

  2455.         SET_HPEL_FUNCS(avg, 1, 8, mmx);

  2456.         SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);

  2457.         }

  2458. 

  2459. #if ARCH_X86_32 || !HAVE_YASM

  2460.         c->gmc= gmc_mmx;

  2461. #endif

  2462. #if ARCH_X86_32 && HAVE_YASM

  2463.         if (!high_bit_depth)

  2464.         c->emulated_edge_mc = emulated_edge_mc_mmx;

  2465. #endif

  2466. 

  2467.         c->add_bytes= add_bytes_mmx;

  2468. 

  2469.         if (!high_bit_depth)

  2470.         c->draw_edges = draw_edges_mmx;

  2471. 

  2472.         if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {

  2473.             c->h263_v_loop_filter= h263_v_loop_filter_mmx;

  2474.             c->h263_h_loop_filter= h263_h_loop_filter_mmx;

  2475.         }

  2476. 

  2477. #if HAVE_YASM

  2478.         if (!high_bit_depth) {

  2479.         c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_mmx_rnd;

  2480.         c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_mmx;

  2481.         }

  2482. 

  2483.         c->put_rv40_chroma_pixels_tab[0]= ff_put_rv40_chroma_mc8_mmx;

  2484.         c->put_rv40_chroma_pixels_tab[1]= ff_put_rv40_chroma_mc4_mmx;

  2485. 

  2486.         c->vector_clip_int32 = ff_vector_clip_int32_mmx;

  2487. #endif

  2488. 

  2489.         if (mm_flags & AV_CPU_FLAG_MMX2) {

  2490.             c->prefetch = prefetch_mmx2;

  2491. 

  2492.             if (!high_bit_depth) {

  2493.             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;

  2494.             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

  2495. 

  2496.             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;

  2497.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;

  2498.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

  2499. 

  2500.             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;

  2501.             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

  2502. 

  2503.             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;

  2504.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;

  2505.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

  2506.             }

  2507. 

  2508.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2509.                 if (!high_bit_depth) {

  2510.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;

  2511.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

  2512.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

  2513.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

  2514.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

  2515.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;

  2516.                 }

  2517. 

  2518.                 if (CONFIG_VP3_DECODER && HAVE_YASM) {

  2519.                     c->vp3_v_loop_filter= ff_vp3_v_loop_filter_mmx2;

  2520.                     c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2;

  2521.                 }

  2522.             }

  2523.             if (CONFIG_VP3_DECODER && HAVE_YASM) {

  2524.                 c->vp3_idct_dc_add = ff_vp3_idct_dc_add_mmx2;

  2525.             }

  2526. 

  2527.             if (CONFIG_VP3_DECODER

  2528.                 && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {

  2529.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_mmx2;

  2530.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_mmx2;

  2531.             }

  2532. 

  2533. #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \

  2534.             c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \

  2535.             c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \

  2536.             c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \

  2537.             c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \

  2538.             c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \

  2539.             c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \

  2540.             c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \

  2541.             c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \

  2542.             c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \

  2543.             c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \

  2544.             c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \

  2545.             c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \

  2546.             c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \

  2547.             c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \

  2548.             c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \

  2549.             c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU

  2550. 

  2551.             SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2, );

  2552.             SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2, );

  2553.             SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2, );

  2554.             SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2, );

  2555.             SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2, );

  2556.             SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2, );

  2557. 

  2558.             if (!high_bit_depth) {

  2559.             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2, );

  2560.             SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2, );

  2561.             SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2, );

  2562.             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2, );

  2563.             SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2, );

  2564.             SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2, );

  2565.             }

  2566.             else if (bit_depth == 10) {

  2567. #if HAVE_YASM

  2568. #if !ARCH_X86_64

  2569.                 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);

  2570.                 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);

  2571.                 SET_QPEL_FUNCS(put_h264_qpel, 1, 8,  10_mmxext, ff_);

  2572.                 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8,  10_mmxext, ff_);

  2573. #endif

  2574.                 SET_QPEL_FUNCS(put_h264_qpel, 2, 4,  10_mmxext, ff_);

  2575.                 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4,  10_mmxext, ff_);

  2576. #endif

  2577.             }

  2578. 

  2579.             SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2, );

  2580.             SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2, );

  2581.             SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2, );

  2582.             SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2, );

  2583. 

  2584. #if HAVE_YASM

  2585.             c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_mmx2;

  2586.             c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_mmx2;

  2587. 

  2588.             if (!high_bit_depth) {

  2589.             c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_mmx2_rnd;

  2590.             c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_mmx2;

  2591.             c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_mmx2;

  2592.             c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_mmx2;

  2593.             }

  2594.             if (bit_depth == 10) {

  2595.                 c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_10_mmxext;

  2596.                 c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_10_mmxext;

  2597.                 c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_10_mmxext;

  2598.                 c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_10_mmxext;

  2599.             }

  2600. 

  2601.             c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;

  2602. #endif

  2603. #if HAVE_7REGS

  2604.             if( mm_flags&AV_CPU_FLAG_3DNOW )

  2605.                 c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;

  2606. #endif

  2607. 

  2608.         } else if (mm_flags & AV_CPU_FLAG_3DNOW) {

  2609.             c->prefetch = prefetch_3dnow;

  2610. 

  2611.             if (!high_bit_depth) {

  2612.             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;

  2613.             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  2614. 

  2615.             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  2616.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  2617.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  2618. 

  2619.             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;

  2620.             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

  2621. 

  2622.             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;

  2623.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;

  2624.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

  2625. 

  2626.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  2627.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  2628.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  2629.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;

  2630.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

  2631.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  2632.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;

  2633.             }

  2634.             }

  2635. 

  2636.             if (CONFIG_VP3_DECODER

  2637.                 && (avctx->codec_id == CODEC_ID_VP3 || avctx->codec_id == CODEC_ID_THEORA)) {

  2638.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_3dnow;

  2639.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_3dnow;

  2640.             }

  2641. 

  2642.             SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow, );

  2643.             SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow, );

  2644.             SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow, );

  2645.             SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow, );

  2646.             SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow, );

  2647.             SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow, );

  2648. 

  2649.             if (!high_bit_depth) {

  2650.             SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow, );

  2651.             SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow, );

  2652.             SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow, );

  2653.             SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow, );

  2654.             SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow, );

  2655.             SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow, );

  2656.             }

  2657. 

  2658.             SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow, );

  2659.             SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow, );

  2660.             SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow, );

  2661.             SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow, );

  2662. 

  2663. #if HAVE_YASM

  2664.             if (!high_bit_depth) {

  2665.             c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_3dnow_rnd;

  2666.             c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_3dnow;

  2667.             }

  2668. 

  2669.             c->avg_rv40_chroma_pixels_tab[0]= ff_avg_rv40_chroma_mc8_3dnow;

  2670.             c->avg_rv40_chroma_pixels_tab[1]= ff_avg_rv40_chroma_mc4_3dnow;

  2671. #endif

  2672.         }

  2673. 

  2674. 

  2675. #define H264_QPEL_FUNCS(x, y, CPU)\

  2676.             c->put_h264_qpel_pixels_tab[0][x+y*4] = put_h264_qpel16_mc##x##y##_##CPU;\

  2677.             c->put_h264_qpel_pixels_tab[1][x+y*4] = put_h264_qpel8_mc##x##y##_##CPU;\

  2678.             c->avg_h264_qpel_pixels_tab[0][x+y*4] = avg_h264_qpel16_mc##x##y##_##CPU;\

  2679.             c->avg_h264_qpel_pixels_tab[1][x+y*4] = avg_h264_qpel8_mc##x##y##_##CPU;

  2680.         if((mm_flags & AV_CPU_FLAG_SSE2) && !(mm_flags & AV_CPU_FLAG_3DNOW)){

  2681.             // these functions are slower than mmx on AMD, but faster on Intel

  2682.             if (!high_bit_depth) {

  2683.             c->put_pixels_tab[0][0] = put_pixels16_sse2;

  2684.             c->put_no_rnd_pixels_tab[0][0] = put_pixels16_sse2;

  2685.             c->avg_pixels_tab[0][0] = avg_pixels16_sse2;

  2686.             H264_QPEL_FUNCS(0, 0, sse2);

  2687.             }

  2688.         }

  2689.         if(mm_flags & AV_CPU_FLAG_SSE2){

  2690.             if (!high_bit_depth) {

  2691.             H264_QPEL_FUNCS(0, 1, sse2);

  2692.             H264_QPEL_FUNCS(0, 2, sse2);

  2693.             H264_QPEL_FUNCS(0, 3, sse2);

  2694.             H264_QPEL_FUNCS(1, 1, sse2);

  2695.             H264_QPEL_FUNCS(1, 2, sse2);

  2696.             H264_QPEL_FUNCS(1, 3, sse2);

  2697.             H264_QPEL_FUNCS(2, 1, sse2);

  2698.             H264_QPEL_FUNCS(2, 2, sse2);

  2699.             H264_QPEL_FUNCS(2, 3, sse2);

  2700.             H264_QPEL_FUNCS(3, 1, sse2);

  2701.             H264_QPEL_FUNCS(3, 2, sse2);

  2702.             H264_QPEL_FUNCS(3, 3, sse2);

  2703.             }

  2704. #if HAVE_YASM

  2705. #define H264_QPEL_FUNCS_10(x, y, CPU)\

  2706.             c->put_h264_qpel_pixels_tab[0][x+y*4] = ff_put_h264_qpel16_mc##x##y##_10_##CPU;\

  2707.             c->put_h264_qpel_pixels_tab[1][x+y*4] = ff_put_h264_qpel8_mc##x##y##_10_##CPU;\

  2708.             c->avg_h264_qpel_pixels_tab[0][x+y*4] = ff_avg_h264_qpel16_mc##x##y##_10_##CPU;\

  2709.             c->avg_h264_qpel_pixels_tab[1][x+y*4] = ff_avg_h264_qpel8_mc##x##y##_10_##CPU;

  2710.             if (bit_depth == 10) {

  2711.                 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);

  2712.                 SET_QPEL_FUNCS(put_h264_qpel, 1, 8,  10_sse2, ff_);

  2713.                 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);

  2714.                 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8,  10_sse2, ff_);

  2715.                 H264_QPEL_FUNCS_10(1, 0, sse2_cache64)

  2716.                 H264_QPEL_FUNCS_10(2, 0, sse2_cache64)

  2717.                 H264_QPEL_FUNCS_10(3, 0, sse2_cache64)

  2718. 

  2719.                 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_sse2;

  2720.                 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_sse2;

  2721.             }

  2722. #endif

  2723.         }

  2724. #if HAVE_SSSE3

  2725.         if(mm_flags & AV_CPU_FLAG_SSSE3){

  2726.             if (!high_bit_depth) {

  2727.             H264_QPEL_FUNCS(1, 0, ssse3);

  2728.             H264_QPEL_FUNCS(1, 1, ssse3);

  2729.             H264_QPEL_FUNCS(1, 2, ssse3);

  2730.             H264_QPEL_FUNCS(1, 3, ssse3);

  2731.             H264_QPEL_FUNCS(2, 0, ssse3);

  2732.             H264_QPEL_FUNCS(2, 1, ssse3);

  2733.             H264_QPEL_FUNCS(2, 2, ssse3);

  2734.             H264_QPEL_FUNCS(2, 3, ssse3);

  2735.             H264_QPEL_FUNCS(3, 0, ssse3);

  2736.             H264_QPEL_FUNCS(3, 1, ssse3);

  2737.             H264_QPEL_FUNCS(3, 2, ssse3);

  2738.             H264_QPEL_FUNCS(3, 3, ssse3);

  2739.             }

  2740. #if HAVE_YASM

  2741.             else if (bit_depth == 10) {

  2742.                 H264_QPEL_FUNCS_10(1, 0, ssse3_cache64)

  2743.                 H264_QPEL_FUNCS_10(2, 0, ssse3_cache64)

  2744.                 H264_QPEL_FUNCS_10(3, 0, ssse3_cache64)

  2745.             }

  2746.             if (!high_bit_depth) {

  2747.             c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_ssse3_rnd;

  2748.             c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_ssse3_rnd;

  2749.             c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_ssse3;

  2750.             c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_ssse3;

  2751.             }

  2752.             c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;

  2753.             if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe

  2754.                 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;

  2755. #endif

  2756.         }

  2757. #endif

  2758. 

  2759.         if(mm_flags & AV_CPU_FLAG_3DNOW){

  2760.             c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;

  2761.             c->vector_fmul = vector_fmul_3dnow;

  2762.         }

  2763.         if(mm_flags & AV_CPU_FLAG_3DNOWEXT){

  2764.             c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;

  2765. #if HAVE_6REGS

  2766.             c->vector_fmul_window = vector_fmul_window_3dnow2;

  2767. #endif

  2768.         }

  2769.         if(mm_flags & AV_CPU_FLAG_MMX2){

  2770. #if HAVE_YASM

  2771.             c->scalarproduct_int16 = ff_scalarproduct_int16_mmx2;

  2772.             c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;

  2773.             if (avctx->flags & CODEC_FLAG_BITEXACT) {

  2774.                 c->apply_window_int16 = ff_apply_window_int16_mmxext_ba;

  2775.             } else {

  2776.                 c->apply_window_int16 = ff_apply_window_int16_mmxext;

  2777.             }

  2778. #endif

  2779.         }

  2780.         if(mm_flags & AV_CPU_FLAG_SSE){

  2781.             c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;

  2782.             c->ac3_downmix = ac3_downmix_sse;

  2783.             c->vector_fmul = vector_fmul_sse;

  2784.             c->vector_fmul_reverse = vector_fmul_reverse_sse;

  2785.             c->vector_fmul_add = vector_fmul_add_sse;

  2786. #if HAVE_6REGS

  2787.             c->vector_fmul_window = vector_fmul_window_sse;

  2788. #endif

  2789.             c->vector_clipf = vector_clipf_sse;

  2790. #if HAVE_YASM

  2791.             c->scalarproduct_float = ff_scalarproduct_float_sse;

  2792. #endif

  2793.         }

  2794.         if(mm_flags & AV_CPU_FLAG_3DNOW)

  2795.             c->vector_fmul_add = vector_fmul_add_3dnow; // faster than sse

  2796.         if(mm_flags & AV_CPU_FLAG_SSE2){

  2797. #if HAVE_YASM

  2798.             c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;

  2799.             c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;

  2800.             if (mm_flags & AV_CPU_FLAG_ATOM) {

  2801.                 c->vector_clip_int32 = ff_vector_clip_int32_sse2_int;

  2802.             } else {

  2803.                 c->vector_clip_int32 = ff_vector_clip_int32_sse2;

  2804.             }

  2805.             if (avctx->flags & CODEC_FLAG_BITEXACT) {

  2806.                 c->apply_window_int16 = ff_apply_window_int16_sse2_ba;

  2807.             } else {

  2808.                 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {

  2809.                     c->apply_window_int16 = ff_apply_window_int16_sse2;

  2810.                 }

  2811.             }

  2812. 

  2813.             if (!high_bit_depth)

  2814.             c->emulated_edge_mc = emulated_edge_mc_sse;

  2815.             c->gmc= gmc_sse;

  2816. #endif

  2817.         }

  2818.         if (mm_flags & AV_CPU_FLAG_SSSE3) {

  2819. #if HAVE_YASM

  2820.             if (mm_flags & AV_CPU_FLAG_ATOM) {

  2821.                 c->apply_window_int16 = ff_apply_window_int16_ssse3_atom;

  2822.             } else {

  2823.                 c->apply_window_int16 = ff_apply_window_int16_ssse3;

  2824.             }

  2825.             if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW))) { // cachesplit

  2826.                 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;

  2827.             }

  2828. #endif

  2829.         }

  2830. 

  2831.         if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {

  2832. #if HAVE_YASM

  2833.             c->vector_clip_int32 = ff_vector_clip_int32_sse41;

  2834. #endif

  2835.         }

  2836. 

  2837. #if HAVE_AVX && HAVE_YASM

  2838.         if (mm_flags & AV_CPU_FLAG_AVX) {

  2839.             if (bit_depth == 10) {

  2840.                 //AVX implies !cache64.

  2841.                 //TODO: Port cache(32|64) detection from x264.

  2842.                 H264_QPEL_FUNCS_10(1, 0, sse2)

  2843.                 H264_QPEL_FUNCS_10(2, 0, sse2)

  2844.                 H264_QPEL_FUNCS_10(3, 0, sse2)

  2845. 

  2846.                 c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_avx;

  2847.                 c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_avx;

  2848.             }

  2849.         }

  2850. #endif

  2851.     }

  2852. 

  2853.     if (CONFIG_ENCODERS)

  2854.         dsputilenc_init_mmx(c, avctx);

  2855. 

  2856. #if 0

  2857.     // for speed testing

  2858.     get_pixels = just_return;

  2859.     put_pixels_clamped = just_return;

  2860.     add_pixels_clamped = just_return;

  2861. 

  2862.     pix_abs16x16 = just_return;

  2863.     pix_abs16x16_x2 = just_return;

  2864.     pix_abs16x16_y2 = just_return;

  2865.     pix_abs16x16_xy2 = just_return;

  2866. 

  2867.     put_pixels_tab[0] = just_return;

  2868.     put_pixels_tab[1] = just_return;

  2869.     put_pixels_tab[2] = just_return;

  2870.     put_pixels_tab[3] = just_return;

  2871. 

  2872.     put_no_rnd_pixels_tab[0] = just_return;

  2873.     put_no_rnd_pixels_tab[1] = just_return;

  2874.     put_no_rnd_pixels_tab[2] = just_return;

  2875.     put_no_rnd_pixels_tab[3] = just_return;

  2876. 

  2877.     avg_pixels_tab[0] = just_return;

  2878.     avg_pixels_tab[1] = just_return;

  2879.     avg_pixels_tab[2] = just_return;

  2880.     avg_pixels_tab[3] = just_return;

  2881. 

  2882.     avg_no_rnd_pixels_tab[0] = just_return;

  2883.     avg_no_rnd_pixels_tab[1] = just_return;

  2884.     avg_no_rnd_pixels_tab[2] = just_return;

  2885.     avg_no_rnd_pixels_tab[3] = just_return;

  2886. 

  2887.     //av_fdct = just_return;

  2888.     //ff_idct = just_return;

  2889. #endif

  2890. }