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.
8579 Commits
32 Branches
759MB
Tree: 024c7378c3
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '024c7378c3'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx.c

3535 lines
135KB
Raw Blame History 

  1. /*

  2.  * MMX optimized DSP utils

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

  4.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  5.  *

  6.  * This file is part of FFmpeg.

  7.  *

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

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

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

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

  12.  *

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

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

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

  16.  * Lesser General Public License for more details.

  17.  *

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

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

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

  21.  *

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

  23.  */

  24. 

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

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

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

  28. #include "x86_cpu.h"

  29. #include "mmx.h"

  30. 

  31. //#undef NDEBUG

  32. //#include <assert.h>

  33. 

  34. extern void ff_idct_xvid_mmx(short *block);

  35. extern void ff_idct_xvid_mmx2(short *block);

  36. 

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

  38. 

  39. /* pixel operations */

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

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

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

  43. 

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

  45. {0x8000000080000000ULL, 0x8000000080000000ULL};

  46. 

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

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

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

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

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

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

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

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

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

  56. 

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

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

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

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

  61. static const uint64_t ff_pb_A1 attribute_used __attribute__ ((aligned(8))) = 0xA1A1A1A1A1A1A1A1ULL;

  62. static const uint64_t ff_pb_5F attribute_used __attribute__ ((aligned(8))) = 0x5F5F5F5F5F5F5F5FULL;

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

  64. 

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

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

  67. 

  68. #define MOVQ_WONE(regd) \

  69.     __asm __volatile ( \

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

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

  72. 

  73. #define MOVQ_BFE(regd) \

  74.     __asm __volatile ( \

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

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

  77. 

  78. #ifndef PIC

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

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

  81. #else

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

  83. // pcmpeqd -> -1

  84. #define MOVQ_BONE(regd) \

  85.     __asm __volatile ( \

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

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

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

  89. 

  90. #define MOVQ_WTWO(regd) \

  91.     __asm __volatile ( \

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

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

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

  95. 

  96. #endif

  97. 

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

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

  100. // regfe is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

  108. 

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

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

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

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

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

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

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

  116. 

  117. // mm6 is supposed to contain 0xfefefefefefefefe

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

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

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

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

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

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

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

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

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

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

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

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

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

  131. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  145. 

  146. /***********************************/

  147. /* MMX no rounding */

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

  149. #define SET_RND  MOVQ_WONE

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

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

  152. 

  153. #include "dsputil_mmx_rnd.h"

  154. 

  155. #undef DEF

  156. #undef SET_RND

  157. #undef PAVGBP

  158. #undef PAVGB

  159. /***********************************/

  160. /* MMX rounding */

  161. 

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

  163. #define SET_RND  MOVQ_WTWO

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

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

  166. 

  167. #include "dsputil_mmx_rnd.h"

  168. 

  169. #undef DEF

  170. #undef SET_RND

  171. #undef PAVGBP

  172. #undef PAVGB

  173. 

  174. /***********************************/

  175. /* 3Dnow specific */

  176. 

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

  178. /* for Athlons PAVGUSB is prefered */

  179. #define PAVGB "pavgusb"

  180. 

  181. #include "dsputil_mmx_avg.h"

  182. 

  183. #undef DEF

  184. #undef PAVGB

  185. 

  186. /***********************************/

  187. /* MMX2 specific */

  188. 

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

  190. 

  191. /* Introduced only in MMX2 set */

  192. #define PAVGB "pavgb"

  193. 

  194. #include "dsputil_mmx_avg.h"

  195. 

  196. #undef DEF

  197. #undef PAVGB

  198. 

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

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

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

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

  203. 

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

  205. /* standard MMX */

  206. 

  207. #ifdef CONFIG_ENCODERS

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

  209. {

  210.     asm volatile(

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

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

  213.         ASMALIGN(4)

  214.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  229.         "js 1b                          \n\t"

  230.         : "+r" (pixels)

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

  232.         : "%"REG_a

  233.     );

  234. }

  235. 

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

  237. {

  238.     asm volatile(

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

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

  241.         ASMALIGN(4)

  242.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  258.         "jnz 1b                         \n\t"

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

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

  261.         : "%"REG_a

  262.     );

  263. }

  264. #endif //CONFIG_ENCODERS

  265. 

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

  267. {

  268.     const DCTELEM *p;

  269.     uint8_t *pix;

  270. 

  271.     /* read the pixels */

  272.     p = block;

  273.     pix = pixels;

  274.     /* unrolled loop */

  275.         __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  293.                 :"memory");

  294.         pix += line_size*4;

  295.         p += 32;

  296. 

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

  298.     // compiler would generate some very strange code

  299.     // thus using "r"

  300.     __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  318.             :"memory");

  319. }

  320. 

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

  322.   { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };

  323. 

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

  325. {

  326.     int i;

  327. 

  328.     movq_m2r(*vector128, mm1);

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

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

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

  332.         block += 8;

  333.         paddb_r2r(mm1, mm0);

  334.         movq_r2m(mm0, *pixels);

  335.         pixels += line_size;

  336.     }

  337. }

  338. 

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

  340. {

  341.     const DCTELEM *p;

  342.     uint8_t *pix;

  343.     int i;

  344. 

  345.     /* read the pixels */

  346.     p = block;

  347.     pix = pixels;

  348.     MOVQ_ZERO(mm7);

  349.     i = 4;

  350.     do {

  351.         __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  373.                 :"r"(p)

  374.                 :"memory");

  375.         pix += line_size*2;

  376.         p += 16;

  377.     } while (--i);

  378. }

  379. 

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

  381. {

  382.     __asm __volatile(

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

  384.          ASMALIGN(3)

  385.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  399.          "jnz 1b                        \n\t"

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

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

  402.          : "%"REG_a, "memory"

  403.         );

  404. }

  405. 

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

  407. {

  408.     __asm __volatile(

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

  410.          ASMALIGN(3)

  411.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  425.          "jnz 1b                        \n\t"

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

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

  428.          : "%"REG_a, "memory"

  429.         );

  430. }

  431. 

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

  433. {

  434.     __asm __volatile(

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

  436.          ASMALIGN(3)

  437.          "1:                            \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  459.          "jnz 1b                        \n\t"

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

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

  462.          : "%"REG_a, "memory"

  463.         );

  464. }

  465. 

  466. static void clear_blocks_mmx(DCTELEM *blocks)

  467. {

  468.     __asm __volatile(

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

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

  471.                 "1:                             \n\t"

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

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

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

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

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

  477.                 " js 1b                         \n\t"

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

  479.                 : "%"REG_a

  480.         );

  481. }

  482. 

  483. #ifdef CONFIG_ENCODERS

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

  485.     const int h=16;

  486.     int sum;

  487.     long index= -line_size*h;

  488. 

  489.     __asm __volatile(

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

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

  492.                 "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

  506.                 " js 1b                         \n\t"

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

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

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

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

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

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

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

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

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

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

  517.         );

  518. 

  519.         return sum;

  520. }

  521. #endif //CONFIG_ENCODERS

  522. 

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

  524.     long i=0;

  525.     asm volatile(

  526.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

  537.         " jb 1b                         \n\t"

  538.         : "+r" (i)

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

  540.     );

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

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

  543. }

  544. 

  545. #define H263_LOOP_FILTER \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  615. 

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

  617.     const int strength= ff_h263_loop_filter_strength[qscale];

  618. 

  619.     asm volatile(

  620. 

  621.         H263_LOOP_FILTER

  622. 

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

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

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

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

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

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

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

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

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

  632.     );

  633. }

  634. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  652. 

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

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

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

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

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

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

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

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

  661.     );

  662. }

  663. 

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

  665.     const int strength= ff_h263_loop_filter_strength[qscale];

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

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

  668. 

  669.     src -= 2;

  670. 

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

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

  673.     asm volatile(

  674.         H263_LOOP_FILTER // 5 3 4 6

  675. 

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

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

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

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

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

  681.     );

  682. 

  683.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  708.         :: "r" (src),

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

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

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

  712.     );

  713. }

  714. 

  715. #ifdef CONFIG_ENCODERS

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

  717.     int tmp;

  718.   asm volatile (

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

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

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

  722.       "1:\n"

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

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

  725. 

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

  727. 

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

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

  730. 

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

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

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

  734. 

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

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

  737. 

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

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

  740. 

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

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

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

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

  745. 

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

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

  748.       "dec %%ecx\n"

  749.       "jnz 1b\n"

  750. 

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

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

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

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

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

  756.     return tmp;

  757. }

  758. 

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

  760.     int tmp;

  761.   asm volatile (

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

  763.       "shr $1,%%ecx\n"

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

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

  766.       "1:\n"

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

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

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

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

  771. 

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

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

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

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

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

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

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

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

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

  781. 

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

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

  784. 

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

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

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

  788. 

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

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

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

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

  793. 

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

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

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

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

  798. 

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

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

  801. 

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

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

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

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

  806. 

  807.       "decl %%ecx\n"

  808.       "jnz 1b\n"

  809. 

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

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

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

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

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

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

  816.       : "%ecx");

  817.     return tmp;

  818. }

  819. 

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

  821.     int tmp;

  822.   asm volatile (

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

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

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

  826.       "1:\n"

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

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

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

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

  831. 

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

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

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

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

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

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

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

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

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

  841. 

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

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

  844. 

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

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

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

  848. 

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

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

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

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

  853. 

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

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

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

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

  858. 

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

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

  861. 

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

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

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

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

  866. 

  867.       "decl %%ecx\n"

  868.       "jnz 1b\n"

  869. 

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

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

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

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

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

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

  876.       : "%ecx");

  877.     return tmp;

  878. }

  879. 

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

  881.     int tmp;

  882.   asm volatile (

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

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

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

  886.       "1:\n"

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

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

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

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

  891. 

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

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

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

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

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

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

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

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

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

  901. 

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

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

  904. 

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

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

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

  908. 

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

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

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

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

  913. 

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

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

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

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

  918. 

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

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

  921. 

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

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

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

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

  926. 

  927.       "decl %2\n"

  928.       "jnz 1b\n"

  929. 

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

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

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

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

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

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

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

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

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

  939.     return tmp;

  940. }

  941. 

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

  943.     int tmp;

  944.   asm volatile (

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

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

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

  948. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  962. 

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

  964. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  990. 

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

  992.       "1:\n"

  993. 

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

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

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

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

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

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

  1000.       "movq %%mm1, %%mm3\n"

  1001.       "punpcklbw %%mm7,%%mm0\n"

  1002.       "punpcklbw %%mm7,%%mm1\n"

  1003.       "punpckhbw %%mm7,%%mm2\n"

  1004.       "punpckhbw %%mm7,%%mm3\n"

  1005.       "psubw %%mm1, %%mm0\n"

  1006.       "psubw %%mm3, %%mm2\n"

  1007.       "psubw %%mm0, %%mm4\n"

  1008.       "psubw %%mm2, %%mm5\n"

  1009.       "pxor %%mm3, %%mm3\n"

  1010.       "pxor %%mm1, %%mm1\n"

  1011.       "pcmpgtw %%mm4, %%mm3\n\t"

  1012.       "pcmpgtw %%mm5, %%mm1\n\t"

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

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

  1015.       "psubw %%mm3, %%mm4\n"

  1016.       "psubw %%mm1, %%mm5\n"

  1017.       "paddw %%mm4, %%mm5\n"

  1018.       "paddw %%mm5, %%mm6\n"

  1019. 

  1020.       "add %2,%0\n"

  1021. 

  1022.       "movq (%0),%%mm4\n"

  1023.       "movq %%mm4, %%mm1\n"

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

  1025.       "psrlq $8, %%mm1\n"

  1026.       "psrlq $8, %%mm4\n"

  1027.       "movq %%mm4, %%mm5\n"

  1028.       "movq %%mm1, %%mm3\n"

  1029.       "punpcklbw %%mm7,%%mm4\n"

  1030.       "punpcklbw %%mm7,%%mm1\n"

  1031.       "punpckhbw %%mm7,%%mm5\n"

  1032.       "punpckhbw %%mm7,%%mm3\n"

  1033.       "psubw %%mm1, %%mm4\n"

  1034.       "psubw %%mm3, %%mm5\n"

  1035.       "psubw %%mm4, %%mm0\n"

  1036.       "psubw %%mm5, %%mm2\n"

  1037.       "pxor %%mm3, %%mm3\n"

  1038.       "pxor %%mm1, %%mm1\n"

  1039.       "pcmpgtw %%mm0, %%mm3\n\t"

  1040.       "pcmpgtw %%mm2, %%mm1\n\t"

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

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

  1043.       "psubw %%mm3, %%mm0\n"

  1044.       "psubw %%mm1, %%mm2\n"

  1045.       "paddw %%mm0, %%mm2\n"

  1046.       "paddw %%mm2, %%mm6\n"

  1047. 

  1048.       "add %2,%0\n"

  1049.       "subl $2, %%ecx\n"

  1050.       " jnz 1b\n"

  1051. 

  1052.       "movq %%mm6, %%mm0\n"

  1053.       "punpcklwd %%mm7,%%mm0\n"

  1054.       "punpckhwd %%mm7,%%mm6\n"

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

  1056. 

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

  1058.       "psrlq $32, %%mm6\n"

  1059.       "paddd %%mm6,%%mm0\n"

  1060.       "movd %%mm0,%1\n"

  1061.       : "+r" (pix1), "=r"(tmp)

  1062.       : "r" ((long)line_size) , "g" (h-2)

  1063.       : "%ecx");

  1064.       return tmp;

  1065. }

  1066. 

  1067. static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {

  1068.     int tmp;

  1069.     uint8_t * pix= pix1;

  1070.   asm volatile (

  1071.       "movl %3,%%ecx\n"

  1072.       "pxor %%mm7,%%mm7\n"

  1073.       "pxor %%mm6,%%mm6\n"

  1074. 

  1075.       "movq (%0),%%mm0\n"

  1076.       "movq 1(%0),%%mm1\n"

  1077.       "movq %%mm0, %%mm2\n"

  1078.       "movq %%mm1, %%mm3\n"

  1079.       "punpcklbw %%mm7,%%mm0\n"

  1080.       "punpcklbw %%mm7,%%mm1\n"

  1081.       "punpckhbw %%mm7,%%mm2\n"

  1082.       "punpckhbw %%mm7,%%mm3\n"

  1083.       "psubw %%mm1, %%mm0\n"

  1084.       "psubw %%mm3, %%mm2\n"

  1085. 

  1086.       "add %2,%0\n"

  1087. 

  1088.       "movq (%0),%%mm4\n"

  1089.       "movq 1(%0),%%mm1\n"

  1090.       "movq %%mm4, %%mm5\n"

  1091.       "movq %%mm1, %%mm3\n"

  1092.       "punpcklbw %%mm7,%%mm4\n"

  1093.       "punpcklbw %%mm7,%%mm1\n"

  1094.       "punpckhbw %%mm7,%%mm5\n"

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

  1096.       "psubw %%mm1, %%mm4\n"

  1097.       "psubw %%mm3, %%mm5\n"

  1098.       "psubw %%mm4, %%mm0\n"

  1099.       "psubw %%mm5, %%mm2\n"

  1100.       "pxor %%mm3, %%mm3\n"

  1101.       "pxor %%mm1, %%mm1\n"

  1102.       "pcmpgtw %%mm0, %%mm3\n\t"

  1103.       "pcmpgtw %%mm2, %%mm1\n\t"

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

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

  1106.       "psubw %%mm3, %%mm0\n"

  1107.       "psubw %%mm1, %%mm2\n"

  1108.       "paddw %%mm0, %%mm2\n"

  1109.       "paddw %%mm2, %%mm6\n"

  1110. 

  1111.       "add %2,%0\n"

  1112.       "1:\n"

  1113. 

  1114.       "movq (%0),%%mm0\n"

  1115.       "movq 1(%0),%%mm1\n"

  1116.       "movq %%mm0, %%mm2\n"

  1117.       "movq %%mm1, %%mm3\n"

  1118.       "punpcklbw %%mm7,%%mm0\n"

  1119.       "punpcklbw %%mm7,%%mm1\n"

  1120.       "punpckhbw %%mm7,%%mm2\n"

  1121.       "punpckhbw %%mm7,%%mm3\n"

  1122.       "psubw %%mm1, %%mm0\n"

  1123.       "psubw %%mm3, %%mm2\n"

  1124.       "psubw %%mm0, %%mm4\n"

  1125.       "psubw %%mm2, %%mm5\n"

  1126.       "pxor %%mm3, %%mm3\n"

  1127.       "pxor %%mm1, %%mm1\n"

  1128.       "pcmpgtw %%mm4, %%mm3\n\t"

  1129.       "pcmpgtw %%mm5, %%mm1\n\t"

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

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

  1132.       "psubw %%mm3, %%mm4\n"

  1133.       "psubw %%mm1, %%mm5\n"

  1134.       "paddw %%mm4, %%mm5\n"

  1135.       "paddw %%mm5, %%mm6\n"

  1136. 

  1137.       "add %2,%0\n"

  1138. 

  1139.       "movq (%0),%%mm4\n"

  1140.       "movq 1(%0),%%mm1\n"

  1141.       "movq %%mm4, %%mm5\n"

  1142.       "movq %%mm1, %%mm3\n"

  1143.       "punpcklbw %%mm7,%%mm4\n"

  1144.       "punpcklbw %%mm7,%%mm1\n"

  1145.       "punpckhbw %%mm7,%%mm5\n"

  1146.       "punpckhbw %%mm7,%%mm3\n"

  1147.       "psubw %%mm1, %%mm4\n"

  1148.       "psubw %%mm3, %%mm5\n"

  1149.       "psubw %%mm4, %%mm0\n"

  1150.       "psubw %%mm5, %%mm2\n"

  1151.       "pxor %%mm3, %%mm3\n"

  1152.       "pxor %%mm1, %%mm1\n"

  1153.       "pcmpgtw %%mm0, %%mm3\n\t"

  1154.       "pcmpgtw %%mm2, %%mm1\n\t"

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

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

  1157.       "psubw %%mm3, %%mm0\n"

  1158.       "psubw %%mm1, %%mm2\n"

  1159.       "paddw %%mm0, %%mm2\n"

  1160.       "paddw %%mm2, %%mm6\n"

  1161. 

  1162.       "add %2,%0\n"

  1163.       "subl $2, %%ecx\n"

  1164.       " jnz 1b\n"

  1165. 

  1166.       "movq %%mm6, %%mm0\n"

  1167.       "punpcklwd %%mm7,%%mm0\n"

  1168.       "punpckhwd %%mm7,%%mm6\n"

  1169.       "paddd %%mm0, %%mm6\n"

  1170. 

  1171.       "movq %%mm6,%%mm0\n"

  1172.       "psrlq $32, %%mm6\n"

  1173.       "paddd %%mm6,%%mm0\n"

  1174.       "movd %%mm0,%1\n"

  1175.       : "+r" (pix1), "=r"(tmp)

  1176.       : "r" ((long)line_size) , "g" (h-2)

  1177.       : "%ecx");

  1178.       return tmp + hf_noise8_mmx(pix+8, line_size, h);

  1179. }

  1180. 

  1181. static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1182.     MpegEncContext *c = p;

  1183.     int score1, score2;

  1184. 

  1185.     if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);

  1186.     else  score1 = sse16_mmx(c, pix1, pix2, line_size, h);

  1187.     score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

  1188. 

  1189.     if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;

  1190.     else  return score1 + FFABS(score2)*8;

  1191. }

  1192. 

  1193. static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1194.     MpegEncContext *c = p;

  1195.     int score1= sse8_mmx(c, pix1, pix2, line_size, h);

  1196.     int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

  1197. 

  1198.     if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;

  1199.     else  return score1 + FFABS(score2)*8;

  1200. }

  1201. 

  1202. static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1203.     int tmp;

  1204. 

  1205.     assert( (((int)pix) & 7) == 0);

  1206.     assert((line_size &7) ==0);

  1207. 

  1208. #define SUM(in0, in1, out0, out1) \

  1209.       "movq (%0), %%mm2\n"\

  1210.       "movq 8(%0), %%mm3\n"\

  1211.       "add %2,%0\n"\

  1212.       "movq %%mm2, " #out0 "\n"\

  1213.       "movq %%mm3, " #out1 "\n"\

  1214.       "psubusb " #in0 ", %%mm2\n"\

  1215.       "psubusb " #in1 ", %%mm3\n"\

  1216.       "psubusb " #out0 ", " #in0 "\n"\

  1217.       "psubusb " #out1 ", " #in1 "\n"\

  1218.       "por %%mm2, " #in0 "\n"\

  1219.       "por %%mm3, " #in1 "\n"\

  1220.       "movq " #in0 ", %%mm2\n"\

  1221.       "movq " #in1 ", %%mm3\n"\

  1222.       "punpcklbw %%mm7, " #in0 "\n"\

  1223.       "punpcklbw %%mm7, " #in1 "\n"\

  1224.       "punpckhbw %%mm7, %%mm2\n"\

  1225.       "punpckhbw %%mm7, %%mm3\n"\

  1226.       "paddw " #in1 ", " #in0 "\n"\

  1227.       "paddw %%mm3, %%mm2\n"\

  1228.       "paddw %%mm2, " #in0 "\n"\

  1229.       "paddw " #in0 ", %%mm6\n"

  1230. 

  1231. 

  1232.   asm volatile (

  1233.       "movl %3,%%ecx\n"

  1234.       "pxor %%mm6,%%mm6\n"

  1235.       "pxor %%mm7,%%mm7\n"

  1236.       "movq (%0),%%mm0\n"

  1237.       "movq 8(%0),%%mm1\n"

  1238.       "add %2,%0\n"

  1239.       "subl $2, %%ecx\n"

  1240.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1241.       "1:\n"

  1242. 

  1243.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1244. 

  1245.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1246. 

  1247.       "subl $2, %%ecx\n"

  1248.       "jnz 1b\n"

  1249. 

  1250.       "movq %%mm6,%%mm0\n"

  1251.       "psrlq $32, %%mm6\n"

  1252.       "paddw %%mm6,%%mm0\n"

  1253.       "movq %%mm0,%%mm6\n"

  1254.       "psrlq $16, %%mm0\n"

  1255.       "paddw %%mm6,%%mm0\n"

  1256.       "movd %%mm0,%1\n"

  1257.       : "+r" (pix), "=r"(tmp)

  1258.       : "r" ((long)line_size) , "m" (h)

  1259.       : "%ecx");

  1260.     return tmp & 0xFFFF;

  1261. }

  1262. #undef SUM

  1263. 

  1264. static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  1265.     int tmp;

  1266. 

  1267.     assert( (((int)pix) & 7) == 0);

  1268.     assert((line_size &7) ==0);

  1269. 

  1270. #define SUM(in0, in1, out0, out1) \

  1271.       "movq (%0), " #out0 "\n"\

  1272.       "movq 8(%0), " #out1 "\n"\

  1273.       "add %2,%0\n"\

  1274.       "psadbw " #out0 ", " #in0 "\n"\

  1275.       "psadbw " #out1 ", " #in1 "\n"\

  1276.       "paddw " #in1 ", " #in0 "\n"\

  1277.       "paddw " #in0 ", %%mm6\n"

  1278. 

  1279.   asm volatile (

  1280.       "movl %3,%%ecx\n"

  1281.       "pxor %%mm6,%%mm6\n"

  1282.       "pxor %%mm7,%%mm7\n"

  1283.       "movq (%0),%%mm0\n"

  1284.       "movq 8(%0),%%mm1\n"

  1285.       "add %2,%0\n"

  1286.       "subl $2, %%ecx\n"

  1287.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1288.       "1:\n"

  1289. 

  1290.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1291. 

  1292.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1293. 

  1294.       "subl $2, %%ecx\n"

  1295.       "jnz 1b\n"

  1296. 

  1297.       "movd %%mm6,%1\n"

  1298.       : "+r" (pix), "=r"(tmp)

  1299.       : "r" ((long)line_size) , "m" (h)

  1300.       : "%ecx");

  1301.     return tmp;

  1302. }

  1303. #undef SUM

  1304. 

  1305. static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1306.     int tmp;

  1307. 

  1308.     assert( (((int)pix1) & 7) == 0);

  1309.     assert( (((int)pix2) & 7) == 0);

  1310.     assert((line_size &7) ==0);

  1311. 

  1312. #define SUM(in0, in1, out0, out1) \

  1313.       "movq (%0),%%mm2\n"\

  1314.       "movq (%1)," #out0 "\n"\

  1315.       "movq 8(%0),%%mm3\n"\

  1316.       "movq 8(%1)," #out1 "\n"\

  1317.       "add %3,%0\n"\

  1318.       "add %3,%1\n"\

  1319.       "psubb " #out0 ", %%mm2\n"\

  1320.       "psubb " #out1 ", %%mm3\n"\

  1321.       "pxor %%mm7, %%mm2\n"\

  1322.       "pxor %%mm7, %%mm3\n"\

  1323.       "movq %%mm2, " #out0 "\n"\

  1324.       "movq %%mm3, " #out1 "\n"\

  1325.       "psubusb " #in0 ", %%mm2\n"\

  1326.       "psubusb " #in1 ", %%mm3\n"\

  1327.       "psubusb " #out0 ", " #in0 "\n"\

  1328.       "psubusb " #out1 ", " #in1 "\n"\

  1329.       "por %%mm2, " #in0 "\n"\

  1330.       "por %%mm3, " #in1 "\n"\

  1331.       "movq " #in0 ", %%mm2\n"\

  1332.       "movq " #in1 ", %%mm3\n"\

  1333.       "punpcklbw %%mm7, " #in0 "\n"\

  1334.       "punpcklbw %%mm7, " #in1 "\n"\

  1335.       "punpckhbw %%mm7, %%mm2\n"\

  1336.       "punpckhbw %%mm7, %%mm3\n"\

  1337.       "paddw " #in1 ", " #in0 "\n"\

  1338.       "paddw %%mm3, %%mm2\n"\

  1339.       "paddw %%mm2, " #in0 "\n"\

  1340.       "paddw " #in0 ", %%mm6\n"

  1341. 

  1342. 

  1343.   asm volatile (

  1344.       "movl %4,%%ecx\n"

  1345.       "pxor %%mm6,%%mm6\n"

  1346.       "pcmpeqw %%mm7,%%mm7\n"

  1347.       "psllw $15, %%mm7\n"

  1348.       "packsswb %%mm7, %%mm7\n"

  1349.       "movq (%0),%%mm0\n"

  1350.       "movq (%1),%%mm2\n"

  1351.       "movq 8(%0),%%mm1\n"

  1352.       "movq 8(%1),%%mm3\n"

  1353.       "add %3,%0\n"

  1354.       "add %3,%1\n"

  1355.       "subl $2, %%ecx\n"

  1356.       "psubb %%mm2, %%mm0\n"

  1357.       "psubb %%mm3, %%mm1\n"

  1358.       "pxor %%mm7, %%mm0\n"

  1359.       "pxor %%mm7, %%mm1\n"

  1360.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1361.       "1:\n"

  1362. 

  1363.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1364. 

  1365.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1366. 

  1367.       "subl $2, %%ecx\n"

  1368.       "jnz 1b\n"

  1369. 

  1370.       "movq %%mm6,%%mm0\n"

  1371.       "psrlq $32, %%mm6\n"

  1372.       "paddw %%mm6,%%mm0\n"

  1373.       "movq %%mm0,%%mm6\n"

  1374.       "psrlq $16, %%mm0\n"

  1375.       "paddw %%mm6,%%mm0\n"

  1376.       "movd %%mm0,%2\n"

  1377.       : "+r" (pix1), "+r" (pix2), "=r"(tmp)

  1378.       : "r" ((long)line_size) , "m" (h)

  1379.       : "%ecx");

  1380.     return tmp & 0x7FFF;

  1381. }

  1382. #undef SUM

  1383. 

  1384. static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  1385.     int tmp;

  1386. 

  1387.     assert( (((int)pix1) & 7) == 0);

  1388.     assert( (((int)pix2) & 7) == 0);

  1389.     assert((line_size &7) ==0);

  1390. 

  1391. #define SUM(in0, in1, out0, out1) \

  1392.       "movq (%0)," #out0 "\n"\

  1393.       "movq (%1),%%mm2\n"\

  1394.       "movq 8(%0)," #out1 "\n"\

  1395.       "movq 8(%1),%%mm3\n"\

  1396.       "add %3,%0\n"\

  1397.       "add %3,%1\n"\

  1398.       "psubb %%mm2, " #out0 "\n"\

  1399.       "psubb %%mm3, " #out1 "\n"\

  1400.       "pxor %%mm7, " #out0 "\n"\

  1401.       "pxor %%mm7, " #out1 "\n"\

  1402.       "psadbw " #out0 ", " #in0 "\n"\

  1403.       "psadbw " #out1 ", " #in1 "\n"\

  1404.       "paddw " #in1 ", " #in0 "\n"\

  1405.       "paddw " #in0 ", %%mm6\n"

  1406. 

  1407.   asm volatile (

  1408.       "movl %4,%%ecx\n"

  1409.       "pxor %%mm6,%%mm6\n"

  1410.       "pcmpeqw %%mm7,%%mm7\n"

  1411.       "psllw $15, %%mm7\n"

  1412.       "packsswb %%mm7, %%mm7\n"

  1413.       "movq (%0),%%mm0\n"

  1414.       "movq (%1),%%mm2\n"

  1415.       "movq 8(%0),%%mm1\n"

  1416.       "movq 8(%1),%%mm3\n"

  1417.       "add %3,%0\n"

  1418.       "add %3,%1\n"

  1419.       "subl $2, %%ecx\n"

  1420.       "psubb %%mm2, %%mm0\n"

  1421.       "psubb %%mm3, %%mm1\n"

  1422.       "pxor %%mm7, %%mm0\n"

  1423.       "pxor %%mm7, %%mm1\n"

  1424.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1425.       "1:\n"

  1426. 

  1427.       SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  1428. 

  1429.       SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  1430. 

  1431.       "subl $2, %%ecx\n"

  1432.       "jnz 1b\n"

  1433. 

  1434.       "movd %%mm6,%2\n"

  1435.       : "+r" (pix1), "+r" (pix2), "=r"(tmp)

  1436.       : "r" ((long)line_size) , "m" (h)

  1437.       : "%ecx");

  1438.     return tmp;

  1439. }

  1440. #undef SUM

  1441. 

  1442. static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){

  1443.     long i=0;

  1444.     asm volatile(

  1445.         "1:                             \n\t"

  1446.         "movq  (%2, %0), %%mm0          \n\t"

  1447.         "movq  (%1, %0), %%mm1          \n\t"

  1448.         "psubb %%mm0, %%mm1             \n\t"

  1449.         "movq %%mm1, (%3, %0)           \n\t"

  1450.         "movq 8(%2, %0), %%mm0          \n\t"

  1451.         "movq 8(%1, %0), %%mm1          \n\t"

  1452.         "psubb %%mm0, %%mm1             \n\t"

  1453.         "movq %%mm1, 8(%3, %0)          \n\t"

  1454.         "add $16, %0                    \n\t"

  1455.         "cmp %4, %0                     \n\t"

  1456.         " jb 1b                         \n\t"

  1457.         : "+r" (i)

  1458.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15)

  1459.     );

  1460.     for(; i<w; i++)

  1461.         dst[i+0] = src1[i+0]-src2[i+0];

  1462. }

  1463. 

  1464. static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){

  1465.     long i=0;

  1466.     uint8_t l, lt;

  1467. 

  1468.     asm volatile(

  1469.         "1:                             \n\t"

  1470.         "movq  -1(%1, %0), %%mm0        \n\t" // LT

  1471.         "movq  (%1, %0), %%mm1          \n\t" // T

  1472.         "movq  -1(%2, %0), %%mm2        \n\t" // L

  1473.         "movq  (%2, %0), %%mm3          \n\t" // X

  1474.         "movq %%mm2, %%mm4              \n\t" // L

  1475.         "psubb %%mm0, %%mm2             \n\t"

  1476.         "paddb %%mm1, %%mm2             \n\t" // L + T - LT

  1477.         "movq %%mm4, %%mm5              \n\t" // L

  1478.         "pmaxub %%mm1, %%mm4            \n\t" // max(T, L)

  1479.         "pminub %%mm5, %%mm1            \n\t" // min(T, L)

  1480.         "pminub %%mm2, %%mm4            \n\t"

  1481.         "pmaxub %%mm1, %%mm4            \n\t"

  1482.         "psubb %%mm4, %%mm3             \n\t" // dst - pred

  1483.         "movq %%mm3, (%3, %0)           \n\t"

  1484.         "add $8, %0                     \n\t"

  1485.         "cmp %4, %0                     \n\t"

  1486.         " jb 1b                         \n\t"

  1487.         : "+r" (i)

  1488.         : "r"(src1), "r"(src2), "r"(dst), "r"((long)w)

  1489.     );

  1490. 

  1491.     l= *left;

  1492.     lt= *left_top;

  1493. 

  1494.     dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);

  1495. 

  1496.     *left_top= src1[w-1];

  1497.     *left    = src2[w-1];

  1498. }

  1499. 

  1500. #define LBUTTERFLY2(a1,b1,a2,b2)\

  1501.     "paddw " #b1 ", " #a1 "           \n\t"\

  1502.     "paddw " #b2 ", " #a2 "           \n\t"\

  1503.     "paddw " #b1 ", " #b1 "           \n\t"\

  1504.     "paddw " #b2 ", " #b2 "           \n\t"\

  1505.     "psubw " #a1 ", " #b1 "           \n\t"\

  1506.     "psubw " #a2 ", " #b2 "           \n\t"

  1507. 

  1508. #define HADAMARD48\

  1509.         LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\

  1510.         LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\

  1511.         LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\

  1512.         LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\

  1513.         LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\

  1514.         LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\

  1515. 

  1516. #define MMABS(a,z)\

  1517.     "pxor " #z ", " #z "              \n\t"\

  1518.     "pcmpgtw " #a ", " #z "           \n\t"\

  1519.     "pxor " #z ", " #a "              \n\t"\

  1520.     "psubw " #z ", " #a "             \n\t"

  1521. 

  1522. #define MMABS_SUM(a,z, sum)\

  1523.     "pxor " #z ", " #z "              \n\t"\

  1524.     "pcmpgtw " #a ", " #z "           \n\t"\

  1525.     "pxor " #z ", " #a "              \n\t"\

  1526.     "psubw " #z ", " #a "             \n\t"\

  1527.     "paddusw " #a ", " #sum "         \n\t"

  1528. 

  1529. #define MMABS_MMX2(a,z)\

  1530.     "pxor " #z ", " #z "              \n\t"\

  1531.     "psubw " #a ", " #z "             \n\t"\

  1532.     "pmaxsw " #z ", " #a "            \n\t"

  1533. 

  1534. #define MMABS_SUM_MMX2(a,z, sum)\

  1535.     "pxor " #z ", " #z "              \n\t"\

  1536.     "psubw " #a ", " #z "             \n\t"\

  1537.     "pmaxsw " #z ", " #a "            \n\t"\

  1538.     "paddusw " #a ", " #sum "         \n\t"

  1539. 

  1540. #define TRANSPOSE4(a,b,c,d,t)\

  1541.     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\

  1542.     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\

  1543.     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\

  1544.     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

  1545. 

  1546. #define LOAD4(o, a, b, c, d)\

  1547.         "movq "#o"(%1), " #a "        \n\t"\

  1548.         "movq "#o"+16(%1), " #b "     \n\t"\

  1549.         "movq "#o"+32(%1), " #c "     \n\t"\

  1550.         "movq "#o"+48(%1), " #d "     \n\t"

  1551. 

  1552. #define STORE4(o, a, b, c, d)\

  1553.         "movq "#a", "#o"(%1)          \n\t"\

  1554.         "movq "#b", "#o"+16(%1)       \n\t"\

  1555.         "movq "#c", "#o"+32(%1)       \n\t"\

  1556.         "movq "#d", "#o"+48(%1)       \n\t"\

  1557. 

  1558. static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1559.     DECLARE_ALIGNED_8(uint64_t, temp[16]);

  1560.     int sum=0;

  1561. 

  1562.     assert(h==8);

  1563. 

  1564.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1565. 

  1566.     asm volatile(

  1567.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1568.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1569. 

  1570.         HADAMARD48

  1571. 

  1572.         "movq %%mm7, 112(%1)            \n\t"

  1573. 

  1574.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1575.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1576. 

  1577.         "movq 112(%1), %%mm7            \n\t"

  1578.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1579.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1580. 

  1581.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1582.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1583. 

  1584.         HADAMARD48

  1585. 

  1586.         "movq %%mm7, 120(%1)            \n\t"

  1587. 

  1588.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1589.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1590. 

  1591.         "movq 120(%1), %%mm7            \n\t"

  1592.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1593.         "movq %%mm7, %%mm5              \n\t"//FIXME remove

  1594.         "movq %%mm6, %%mm7              \n\t"

  1595.         "movq %%mm0, %%mm6              \n\t"

  1596. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1597. 

  1598.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1599. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1600. 

  1601.         HADAMARD48

  1602.         "movq %%mm7, 64(%1)             \n\t"

  1603.         MMABS(%%mm0, %%mm7)

  1604.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1605.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1606.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1607.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1608.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1609.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1610.         "movq 64(%1), %%mm1             \n\t"

  1611.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1612.         "movq %%mm0, 64(%1)             \n\t"

  1613. 

  1614.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1615.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1616. 

  1617.         HADAMARD48

  1618.         "movq %%mm7, (%1)               \n\t"

  1619.         MMABS(%%mm0, %%mm7)

  1620.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1621.         MMABS_SUM(%%mm2, %%mm7, %%mm0)

  1622.         MMABS_SUM(%%mm3, %%mm7, %%mm0)

  1623.         MMABS_SUM(%%mm4, %%mm7, %%mm0)

  1624.         MMABS_SUM(%%mm5, %%mm7, %%mm0)

  1625.         MMABS_SUM(%%mm6, %%mm7, %%mm0)

  1626.         "movq (%1), %%mm1               \n\t"

  1627.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1628.         "movq 64(%1), %%mm1             \n\t"

  1629.         MMABS_SUM(%%mm1, %%mm7, %%mm0)

  1630. 

  1631.         "movq %%mm0, %%mm1              \n\t"

  1632.         "psrlq $32, %%mm0               \n\t"

  1633.         "paddusw %%mm1, %%mm0           \n\t"

  1634.         "movq %%mm0, %%mm1              \n\t"

  1635.         "psrlq $16, %%mm0               \n\t"

  1636.         "paddusw %%mm1, %%mm0           \n\t"

  1637.         "movd %%mm0, %0                 \n\t"

  1638. 

  1639.         : "=r" (sum)

  1640.         : "r"(temp)

  1641.     );

  1642.     return sum&0xFFFF;

  1643. }

  1644. 

  1645. static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){

  1646.     DECLARE_ALIGNED_8(uint64_t, temp[16]);

  1647.     int sum=0;

  1648. 

  1649.     assert(h==8);

  1650. 

  1651.     diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

  1652. 

  1653.     asm volatile(

  1654.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1655.         LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

  1656. 

  1657.         HADAMARD48

  1658. 

  1659.         "movq %%mm7, 112(%1)            \n\t"

  1660. 

  1661.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1662.         STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

  1663. 

  1664.         "movq 112(%1), %%mm7            \n\t"

  1665.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1666.         STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

  1667. 

  1668.         LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)

  1669.         LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1670. 

  1671.         HADAMARD48

  1672. 

  1673.         "movq %%mm7, 120(%1)            \n\t"

  1674. 

  1675.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)

  1676.         STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

  1677. 

  1678.         "movq 120(%1), %%mm7            \n\t"

  1679.         TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)

  1680.         "movq %%mm7, %%mm5              \n\t"//FIXME remove

  1681.         "movq %%mm6, %%mm7              \n\t"

  1682.         "movq %%mm0, %%mm6              \n\t"

  1683. //        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

  1684. 

  1685.         LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)

  1686. //        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

  1687. 

  1688.         HADAMARD48

  1689.         "movq %%mm7, 64(%1)             \n\t"

  1690.         MMABS_MMX2(%%mm0, %%mm7)

  1691.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1692.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1693.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1694.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1695.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1696.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1697.         "movq 64(%1), %%mm1             \n\t"

  1698.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1699.         "movq %%mm0, 64(%1)             \n\t"

  1700. 

  1701.         LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)

  1702.         LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

  1703. 

  1704.         HADAMARD48

  1705.         "movq %%mm7, (%1)               \n\t"

  1706.         MMABS_MMX2(%%mm0, %%mm7)

  1707.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1708.         MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)

  1709.         MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)

  1710.         MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)

  1711.         MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)

  1712.         MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)

  1713.         "movq (%1), %%mm1               \n\t"

  1714.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1715.         "movq 64(%1), %%mm1             \n\t"

  1716.         MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

  1717. 

  1718.         "pshufw $0x0E, %%mm0, %%mm1     \n\t"

  1719.         "paddusw %%mm1, %%mm0           \n\t"

  1720.         "pshufw $0x01, %%mm0, %%mm1     \n\t"

  1721.         "paddusw %%mm1, %%mm0           \n\t"

  1722.         "movd %%mm0, %0                 \n\t"

  1723. 

  1724.         : "=r" (sum)

  1725.         : "r"(temp)

  1726.     );

  1727.     return sum&0xFFFF;

  1728. }

  1729. 

  1730. 

  1731. WARPER8_16_SQ(hadamard8_diff_mmx, hadamard8_diff16_mmx)

  1732. WARPER8_16_SQ(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)

  1733. #endif //CONFIG_ENCODERS

  1734. 

  1735. #define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)

  1736. #define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

  1737. 

  1738. #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\

  1739.         "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\

  1740.         "movq "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */\

  1741.         "pmullw " #m3 ", %%mm4            \n\t" /* 20x1 */\

  1742.         "movq "#in7", " #m3 "             \n\t" /* d */\

  1743.         "movq "#in0", %%mm5               \n\t" /* D */\

  1744.         "paddw " #m3 ", %%mm5             \n\t" /* x4 */\

  1745.         "psubw %%mm5, %%mm4               \n\t" /* 20x1 - x4 */\

  1746.         "movq "#in1", %%mm5               \n\t" /* C */\

  1747.         "movq "#in2", %%mm6               \n\t" /* B */\

  1748.         "paddw " #m6 ", %%mm5             \n\t" /* x3 */\

  1749.         "paddw " #m5 ", %%mm6             \n\t" /* x2 */\

  1750.         "paddw %%mm6, %%mm6               \n\t" /* 2x2 */\

  1751.         "psubw %%mm6, %%mm5               \n\t" /* -2x2 + x3 */\

  1752.         "pmullw "MANGLE(ff_pw_3)", %%mm5  \n\t" /* -6x2 + 3x3 */\

  1753.         "paddw " #rnd ", %%mm4            \n\t" /* x2 */\

  1754.         "paddw %%mm4, %%mm5               \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\

  1755.         "psraw $5, %%mm5                  \n\t"\

  1756.         "packuswb %%mm5, %%mm5            \n\t"\

  1757.         OP(%%mm5, out, %%mm7, d)

  1758. 

  1759. #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\

  1760. static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1761.     uint64_t temp;\

  1762. \

  1763.     asm volatile(\

  1764.         "pxor %%mm7, %%mm7                \n\t"\

  1765.         "1:                               \n\t"\

  1766.         "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\

  1767.         "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\

  1768.         "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\

  1769.         "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\

  1770.         "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\

  1771.         "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\

  1772.         "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\

  1773.         "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\

  1774.         "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\

  1775.         "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\

  1776.         "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\

  1777.         "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\

  1778.         "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\

  1779.         "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\

  1780.         "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\

  1781.         "paddw %%mm3, %%mm5               \n\t" /* b */\

  1782.         "paddw %%mm2, %%mm6               \n\t" /* c */\

  1783.         "paddw %%mm5, %%mm5               \n\t" /* 2b */\

  1784.         "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\

  1785.         "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\

  1786.         "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\

  1787.         "paddw %%mm4, %%mm0               \n\t" /* a */\

  1788.         "paddw %%mm1, %%mm5               \n\t" /* d */\

  1789.         "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\

  1790.         "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\

  1791.         "paddw %6, %%mm6                  \n\t"\

  1792.         "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\

  1793.         "psraw $5, %%mm0                  \n\t"\

  1794.         "movq %%mm0, %5                   \n\t"\

  1795.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1796.         \

  1797.         "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\

  1798.         "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\

  1799.         "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\

  1800.         "psrlq $8, %%mm0                  \n\t" /* GHIJKLM0 */\

  1801.         "psrlq $16, %%mm5                 \n\t" /* HIJKLM00 */\

  1802.         "punpcklbw %%mm7, %%mm0           \n\t" /* 0G0H0I0J */\

  1803.         "punpcklbw %%mm7, %%mm5           \n\t" /* 0H0I0J0K */\

  1804.         "paddw %%mm0, %%mm2               \n\t" /* b */\

  1805.         "paddw %%mm5, %%mm3               \n\t" /* c */\

  1806.         "paddw %%mm2, %%mm2               \n\t" /* 2b */\

  1807.         "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\

  1808.         "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\

  1809.         "psrlq $24, %%mm6                 \n\t" /* IJKLM000 */\

  1810.         "punpcklbw %%mm7, %%mm2           \n\t" /* 0F0G0H0I */\

  1811.         "punpcklbw %%mm7, %%mm6           \n\t" /* 0I0J0K0L */\

  1812.         "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\

  1813.         "paddw %%mm2, %%mm1               \n\t" /* a */\

  1814.         "paddw %%mm6, %%mm4               \n\t" /* d */\

  1815.         "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\

  1816.         "psubw %%mm4, %%mm3               \n\t" /* - 6b +3c - d */\

  1817.         "paddw %6, %%mm1                  \n\t"\

  1818.         "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b +3c - d */\

  1819.         "psraw $5, %%mm3                  \n\t"\

  1820.         "movq %5, %%mm1                   \n\t"\

  1821.         "packuswb %%mm3, %%mm1            \n\t"\

  1822.         OP_MMX2(%%mm1, (%1),%%mm4, q)\

  1823.         /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\

  1824.         \

  1825.         "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\

  1826.         "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\

  1827.         "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\

  1828.         "psrlq $8, %%mm1                  \n\t" /* KLMNOPQ0 */\

  1829.         "psrlq $16, %%mm4                 \n\t" /* LMNOPQ00 */\

  1830.         "punpcklbw %%mm7, %%mm1           \n\t" /* 0K0L0M0N */\

  1831.         "punpcklbw %%mm7, %%mm4           \n\t" /* 0L0M0N0O */\

  1832.         "paddw %%mm1, %%mm5               \n\t" /* b */\

  1833.         "paddw %%mm4, %%mm0               \n\t" /* c */\

  1834.         "paddw %%mm5, %%mm5               \n\t" /* 2b */\

  1835.         "psubw %%mm5, %%mm0               \n\t" /* c - 2b */\

  1836.         "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\

  1837.         "psrlq $24, %%mm3                 \n\t" /* MNOPQ000 */\

  1838.         "pmullw "MANGLE(ff_pw_3)", %%mm0  \n\t" /* 3c - 6b */\

  1839.         "punpcklbw %%mm7, %%mm3           \n\t" /* 0M0N0O0P */\

  1840.         "paddw %%mm3, %%mm2               \n\t" /* d */\

  1841.         "psubw %%mm2, %%mm0               \n\t" /* -6b + 3c - d */\

  1842.         "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\

  1843.         "punpcklbw %%mm7, %%mm2           \n\t" /* 0J0K0L0M */\

  1844.         "punpckhbw %%mm7, %%mm5           \n\t" /* 0N0O0P0Q */\

  1845.         "paddw %%mm2, %%mm6               \n\t" /* a */\

  1846.         "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\

  1847.         "paddw %6, %%mm0                  \n\t"\

  1848.         "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\

  1849.         "psraw $5, %%mm0                  \n\t"\

  1850.         /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\

  1851.         \

  1852.         "paddw %%mm5, %%mm3               \n\t" /* a */\

  1853.         "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */\

  1854.         "paddw %%mm4, %%mm6               \n\t" /* b */\

  1855.         "pshufw $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */\

  1856.         "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */\

  1857.         "paddw %%mm1, %%mm4               \n\t" /* c */\

  1858.         "paddw %%mm2, %%mm5               \n\t" /* d */\

  1859.         "paddw %%mm6, %%mm6               \n\t" /* 2b */\

  1860.         "psubw %%mm6, %%mm4               \n\t" /* c - 2b */\

  1861.         "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\

  1862.         "pmullw "MANGLE(ff_pw_3)", %%mm4  \n\t" /* 3c - 6b */\

  1863.         "psubw %%mm5, %%mm3               \n\t" /* -6b + 3c - d */\

  1864.         "paddw %6, %%mm4                  \n\t"\

  1865.         "paddw %%mm3, %%mm4               \n\t" /* 20a - 6b + 3c - d */\

  1866.         "psraw $5, %%mm4                  \n\t"\

  1867.         "packuswb %%mm4, %%mm0            \n\t"\

  1868.         OP_MMX2(%%mm0, 8(%1), %%mm4, q)\

  1869.         \

  1870.         "add %3, %0                       \n\t"\

  1871.         "add %4, %1                       \n\t"\

  1872.         "decl %2                          \n\t"\

  1873.         " jnz 1b                          \n\t"\

  1874.         : "+a"(src), "+c"(dst), "+m"(h)\

  1875.         : "d"((long)srcStride), "S"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1876.         : "memory"\

  1877.     );\

  1878. }\

  1879. \

  1880. static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1881.     int i;\

  1882.     int16_t temp[16];\

  1883.     /* quick HACK, XXX FIXME MUST be optimized */\

  1884.     for(i=0; i<h; i++)\

  1885.     {\

  1886.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  1887.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  1888.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  1889.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  1890.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  1891.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\

  1892.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\

  1893.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\

  1894.         temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\

  1895.         temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\

  1896.         temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\

  1897.         temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\

  1898.         temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\

  1899.         temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\

  1900.         temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\

  1901.         temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\

  1902.         asm volatile(\

  1903.             "movq (%0), %%mm0               \n\t"\

  1904.             "movq 8(%0), %%mm1              \n\t"\

  1905.             "paddw %2, %%mm0                \n\t"\

  1906.             "paddw %2, %%mm1                \n\t"\

  1907.             "psraw $5, %%mm0                \n\t"\

  1908.             "psraw $5, %%mm1                \n\t"\

  1909.             "packuswb %%mm1, %%mm0          \n\t"\

  1910.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  1911.             "movq 16(%0), %%mm0             \n\t"\

  1912.             "movq 24(%0), %%mm1             \n\t"\

  1913.             "paddw %2, %%mm0                \n\t"\

  1914.             "paddw %2, %%mm1                \n\t"\

  1915.             "psraw $5, %%mm0                \n\t"\

  1916.             "psraw $5, %%mm1                \n\t"\

  1917.             "packuswb %%mm1, %%mm0          \n\t"\

  1918.             OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\

  1919.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  1920.             : "memory"\

  1921.         );\

  1922.         dst+=dstStride;\

  1923.         src+=srcStride;\

  1924.     }\

  1925. }\

  1926. \

  1927. static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1928.     uint64_t temp;\

  1929. \

  1930.     asm volatile(\

  1931.         "pxor %%mm7, %%mm7                \n\t"\

  1932.         "1:                               \n\t"\

  1933.         "movq  (%0), %%mm0                \n\t" /* ABCDEFGH */\

  1934.         "movq %%mm0, %%mm1                \n\t" /* ABCDEFGH */\

  1935.         "movq %%mm0, %%mm2                \n\t" /* ABCDEFGH */\

  1936.         "punpcklbw %%mm7, %%mm0           \n\t" /* 0A0B0C0D */\

  1937.         "punpckhbw %%mm7, %%mm1           \n\t" /* 0E0F0G0H */\

  1938.         "pshufw $0x90, %%mm0, %%mm5       \n\t" /* 0A0A0B0C */\

  1939.         "pshufw $0x41, %%mm0, %%mm6       \n\t" /* 0B0A0A0B */\

  1940.         "movq %%mm2, %%mm3                \n\t" /* ABCDEFGH */\

  1941.         "movq %%mm2, %%mm4                \n\t" /* ABCDEFGH */\

  1942.         "psllq $8, %%mm2                  \n\t" /* 0ABCDEFG */\

  1943.         "psllq $16, %%mm3                 \n\t" /* 00ABCDEF */\

  1944.         "psllq $24, %%mm4                 \n\t" /* 000ABCDE */\

  1945.         "punpckhbw %%mm7, %%mm2           \n\t" /* 0D0E0F0G */\

  1946.         "punpckhbw %%mm7, %%mm3           \n\t" /* 0C0D0E0F */\

  1947.         "punpckhbw %%mm7, %%mm4           \n\t" /* 0B0C0D0E */\

  1948.         "paddw %%mm3, %%mm5               \n\t" /* b */\

  1949.         "paddw %%mm2, %%mm6               \n\t" /* c */\

  1950.         "paddw %%mm5, %%mm5               \n\t" /* 2b */\

  1951.         "psubw %%mm5, %%mm6               \n\t" /* c - 2b */\

  1952.         "pshufw $0x06, %%mm0, %%mm5       \n\t" /* 0C0B0A0A */\

  1953.         "pmullw "MANGLE(ff_pw_3)", %%mm6  \n\t" /* 3c - 6b */\

  1954.         "paddw %%mm4, %%mm0               \n\t" /* a */\

  1955.         "paddw %%mm1, %%mm5               \n\t" /* d */\

  1956.         "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\

  1957.         "psubw %%mm5, %%mm0               \n\t" /* 20a - d */\

  1958.         "paddw %6, %%mm6                  \n\t"\

  1959.         "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\

  1960.         "psraw $5, %%mm0                  \n\t"\

  1961.         /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\

  1962.         \

  1963.         "movd 5(%0), %%mm5                \n\t" /* FGHI */\

  1964.         "punpcklbw %%mm7, %%mm5           \n\t" /* 0F0G0H0I */\

  1965.         "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0G0H0I0I */\

  1966.         "paddw %%mm5, %%mm1               \n\t" /* a */\

  1967.         "paddw %%mm6, %%mm2               \n\t" /* b */\

  1968.         "pshufw $0xBE, %%mm5, %%mm6       \n\t" /* 0H0I0I0H */\

  1969.         "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0I0I0H0G */\

  1970.         "paddw %%mm6, %%mm3               \n\t" /* c */\

  1971.         "paddw %%mm5, %%mm4               \n\t" /* d */\

  1972.         "paddw %%mm2, %%mm2               \n\t" /* 2b */\

  1973.         "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\

  1974.         "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\

  1975.         "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\

  1976.         "psubw %%mm4, %%mm3               \n\t" /* -6b + 3c - d */\

  1977.         "paddw %6, %%mm1                  \n\t"\

  1978.         "paddw %%mm1, %%mm3               \n\t" /* 20a - 6b + 3c - d */\

  1979.         "psraw $5, %%mm3                  \n\t"\

  1980.         "packuswb %%mm3, %%mm0            \n\t"\

  1981.         OP_MMX2(%%mm0, (%1), %%mm4, q)\

  1982.         \

  1983.         "add %3, %0                       \n\t"\

  1984.         "add %4, %1                       \n\t"\

  1985.         "decl %2                          \n\t"\

  1986.         " jnz 1b                          \n\t"\

  1987.         : "+a"(src), "+c"(dst), "+m"(h)\

  1988.         : "S"((long)srcStride), "D"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\

  1989.         : "memory"\

  1990.     );\

  1991. }\

  1992. \

  1993. static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  1994.     int i;\

  1995.     int16_t temp[8];\

  1996.     /* quick HACK, XXX FIXME MUST be optimized */\

  1997.     for(i=0; i<h; i++)\

  1998.     {\

  1999.         temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\

  2000.         temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\

  2001.         temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\

  2002.         temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\

  2003.         temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\

  2004.         temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\

  2005.         temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\

  2006.         temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\

  2007.         asm volatile(\

  2008.             "movq (%0), %%mm0           \n\t"\

  2009.             "movq 8(%0), %%mm1          \n\t"\

  2010.             "paddw %2, %%mm0            \n\t"\

  2011.             "paddw %2, %%mm1            \n\t"\

  2012.             "psraw $5, %%mm0            \n\t"\

  2013.             "psraw $5, %%mm1            \n\t"\

  2014.             "packuswb %%mm1, %%mm0      \n\t"\

  2015.             OP_3DNOW(%%mm0, (%1), %%mm1, q)\

  2016.             :: "r"(temp), "r"(dst), "m"(ROUNDER)\

  2017.             :"memory"\

  2018.         );\

  2019.         dst+=dstStride;\

  2020.         src+=srcStride;\

  2021.     }\

  2022. }

  2023. 

  2024. #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\

  2025. \

  2026. static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2027.     uint64_t temp[17*4];\

  2028.     uint64_t *temp_ptr= temp;\

  2029.     int count= 17;\

  2030. \

  2031.     /*FIXME unroll */\

  2032.     asm volatile(\

  2033.         "pxor %%mm7, %%mm7              \n\t"\

  2034.         "1:                             \n\t"\

  2035.         "movq (%0), %%mm0               \n\t"\

  2036.         "movq (%0), %%mm1               \n\t"\

  2037.         "movq 8(%0), %%mm2              \n\t"\

  2038.         "movq 8(%0), %%mm3              \n\t"\

  2039.         "punpcklbw %%mm7, %%mm0         \n\t"\

  2040.         "punpckhbw %%mm7, %%mm1         \n\t"\

  2041.         "punpcklbw %%mm7, %%mm2         \n\t"\

  2042.         "punpckhbw %%mm7, %%mm3         \n\t"\

  2043.         "movq %%mm0, (%1)               \n\t"\

  2044.         "movq %%mm1, 17*8(%1)           \n\t"\

  2045.         "movq %%mm2, 2*17*8(%1)         \n\t"\

  2046.         "movq %%mm3, 3*17*8(%1)         \n\t"\

  2047.         "add $8, %1                     \n\t"\

  2048.         "add %3, %0                     \n\t"\

  2049.         "decl %2                        \n\t"\

  2050.         " jnz 1b                        \n\t"\

  2051.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  2052.         : "r" ((long)srcStride)\

  2053.         : "memory"\

  2054.     );\

  2055.     \

  2056.     temp_ptr= temp;\

  2057.     count=4;\

  2058.     \

  2059. /*FIXME reorder for speed */\

  2060.     asm volatile(\

  2061.         /*"pxor %%mm7, %%mm7              \n\t"*/\

  2062.         "1:                             \n\t"\

  2063.         "movq (%0), %%mm0               \n\t"\

  2064.         "movq 8(%0), %%mm1              \n\t"\

  2065.         "movq 16(%0), %%mm2             \n\t"\

  2066.         "movq 24(%0), %%mm3             \n\t"\

  2067.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  2068.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  2069.         "add %4, %1                     \n\t"\

  2070.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  2071.         \

  2072.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  2073.         "add %4, %1                     \n\t"\

  2074.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  2075.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\

  2076.         "add %4, %1                     \n\t"\

  2077.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\

  2078.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\

  2079.         "add %4, %1                     \n\t"\

  2080.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\

  2081.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\

  2082.         "add %4, %1                     \n\t"\

  2083.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\

  2084.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\

  2085.         "add %4, %1                     \n\t"\

  2086.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\

  2087.         \

  2088.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\

  2089.         "add %4, %1                     \n\t"  \

  2090.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\

  2091.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\

  2092.         \

  2093.         "add $136, %0                   \n\t"\

  2094.         "add %6, %1                     \n\t"\

  2095.         "decl %2                        \n\t"\

  2096.         " jnz 1b                        \n\t"\

  2097.         \

  2098.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2099.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(long)dstStride)\

  2100.         :"memory"\

  2101.     );\

  2102. }\

  2103. \

  2104. static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  2105.     uint64_t temp[9*2];\

  2106.     uint64_t *temp_ptr= temp;\

  2107.     int count= 9;\

  2108. \

  2109.     /*FIXME unroll */\

  2110.     asm volatile(\

  2111.         "pxor %%mm7, %%mm7              \n\t"\

  2112.         "1:                             \n\t"\

  2113.         "movq (%0), %%mm0               \n\t"\

  2114.         "movq (%0), %%mm1               \n\t"\

  2115.         "punpcklbw %%mm7, %%mm0         \n\t"\

  2116.         "punpckhbw %%mm7, %%mm1         \n\t"\

  2117.         "movq %%mm0, (%1)               \n\t"\

  2118.         "movq %%mm1, 9*8(%1)            \n\t"\

  2119.         "add $8, %1                     \n\t"\

  2120.         "add %3, %0                     \n\t"\

  2121.         "decl %2                        \n\t"\

  2122.         " jnz 1b                        \n\t"\

  2123.         : "+r" (src), "+r" (temp_ptr), "+r"(count)\

  2124.         : "r" ((long)srcStride)\

  2125.         : "memory"\

  2126.     );\

  2127.     \

  2128.     temp_ptr= temp;\

  2129.     count=2;\

  2130.     \

  2131. /*FIXME reorder for speed */\

  2132.     asm volatile(\

  2133.         /*"pxor %%mm7, %%mm7              \n\t"*/\

  2134.         "1:                             \n\t"\

  2135.         "movq (%0), %%mm0               \n\t"\

  2136.         "movq 8(%0), %%mm1              \n\t"\

  2137.         "movq 16(%0), %%mm2             \n\t"\

  2138.         "movq 24(%0), %%mm3             \n\t"\

  2139.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\

  2140.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\

  2141.         "add %4, %1                     \n\t"\

  2142.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\

  2143.         \

  2144.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\

  2145.         "add %4, %1                     \n\t"\

  2146.         QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\

  2147.         \

  2148.         QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\

  2149.         "add %4, %1                     \n\t"\

  2150.         QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\

  2151.         QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\

  2152.                 \

  2153.         "add $72, %0                    \n\t"\

  2154.         "add %6, %1                     \n\t"\

  2155.         "decl %2                        \n\t"\

  2156.         " jnz 1b                        \n\t"\

  2157.          \

  2158.         : "+r"(temp_ptr), "+r"(dst), "+g"(count)\

  2159.         : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(long)dstStride)\

  2160.         : "memory"\

  2161.    );\

  2162. }\

  2163. \

  2164. static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2165.     OPNAME ## pixels8_mmx(dst, src, stride, 8);\

  2166. }\

  2167. \

  2168. static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2169.     uint64_t temp[8];\

  2170.     uint8_t * const half= (uint8_t*)temp;\

  2171.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2172.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2173. }\

  2174. \

  2175. static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2176.     OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\

  2177. }\

  2178. \

  2179. static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2180.     uint64_t temp[8];\

  2181.     uint8_t * const half= (uint8_t*)temp;\

  2182.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\

  2183.     OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\

  2184. }\

  2185. \

  2186. static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2187.     uint64_t temp[8];\

  2188.     uint8_t * const half= (uint8_t*)temp;\

  2189.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2190.     OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\

  2191. }\

  2192. \

  2193. static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2194.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2195. }\

  2196. \

  2197. static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2198.     uint64_t temp[8];\

  2199.     uint8_t * const half= (uint8_t*)temp;\

  2200.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\

  2201.     OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\

  2202. }\

  2203. static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2204.     uint64_t half[8 + 9];\

  2205.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2206.     uint8_t * const halfHV= ((uint8_t*)half);\

  2207.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2208.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2209.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2210.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2211. }\

  2212. static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2213.     uint64_t half[8 + 9];\

  2214.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2215.     uint8_t * const halfHV= ((uint8_t*)half);\

  2216.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2217.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2218.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2219.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2220. }\

  2221. static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2222.     uint64_t half[8 + 9];\

  2223.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2224.     uint8_t * const halfHV= ((uint8_t*)half);\

  2225.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2226.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2227.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2228.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2229. }\

  2230. static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2231.     uint64_t half[8 + 9];\

  2232.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2233.     uint8_t * const halfHV= ((uint8_t*)half);\

  2234.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2235.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2236.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2237.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2238. }\

  2239. static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2240.     uint64_t half[8 + 9];\

  2241.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2242.     uint8_t * const halfHV= ((uint8_t*)half);\

  2243.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2244.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2245.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\

  2246. }\

  2247. static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2248.     uint64_t half[8 + 9];\

  2249.     uint8_t * const halfH= ((uint8_t*)half) + 64;\

  2250.     uint8_t * const halfHV= ((uint8_t*)half);\

  2251.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2252.     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\

  2253.     OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\

  2254. }\

  2255. static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2256.     uint64_t half[8 + 9];\

  2257.     uint8_t * const halfH= ((uint8_t*)half);\

  2258.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2259.     put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\

  2260.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2261. }\

  2262. static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2263.     uint64_t half[8 + 9];\

  2264.     uint8_t * const halfH= ((uint8_t*)half);\

  2265.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2266.     put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\

  2267.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2268. }\

  2269. static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2270.     uint64_t half[9];\

  2271.     uint8_t * const halfH= ((uint8_t*)half);\

  2272.     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\

  2273.     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\

  2274. }\

  2275. static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  2276.     OPNAME ## pixels16_mmx(dst, src, stride, 16);\

  2277. }\

  2278. \

  2279. static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2280.     uint64_t temp[32];\

  2281.     uint8_t * const half= (uint8_t*)temp;\

  2282.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2283.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2284. }\

  2285. \

  2286. static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2287.     OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\

  2288. }\

  2289. \

  2290. static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2291.     uint64_t temp[32];\

  2292.     uint8_t * const half= (uint8_t*)temp;\

  2293.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\

  2294.     OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\

  2295. }\

  2296. \

  2297. static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2298.     uint64_t temp[32];\

  2299.     uint8_t * const half= (uint8_t*)temp;\

  2300.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2301.     OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\

  2302. }\

  2303. \

  2304. static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2305.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\

  2306. }\

  2307. \

  2308. static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2309.     uint64_t temp[32];\

  2310.     uint8_t * const half= (uint8_t*)temp;\

  2311.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\

  2312.     OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\

  2313. }\

  2314. static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2315.     uint64_t half[16*2 + 17*2];\

  2316.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2317.     uint8_t * const halfHV= ((uint8_t*)half);\

  2318.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2319.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2320.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2321.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2322. }\

  2323. static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2324.     uint64_t half[16*2 + 17*2];\

  2325.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2326.     uint8_t * const halfHV= ((uint8_t*)half);\

  2327.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2328.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2329.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2330.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2331. }\

  2332. static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2333.     uint64_t half[16*2 + 17*2];\

  2334.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2335.     uint8_t * const halfHV= ((uint8_t*)half);\

  2336.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2337.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2338.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2339.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2340. }\

  2341. static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2342.     uint64_t half[16*2 + 17*2];\

  2343.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2344.     uint8_t * const halfHV= ((uint8_t*)half);\

  2345.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2346.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2347.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2348.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2349. }\

  2350. static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2351.     uint64_t half[16*2 + 17*2];\

  2352.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2353.     uint8_t * const halfHV= ((uint8_t*)half);\

  2354.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2355.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2356.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\

  2357. }\

  2358. static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2359.     uint64_t half[16*2 + 17*2];\

  2360.     uint8_t * const halfH= ((uint8_t*)half) + 256;\

  2361.     uint8_t * const halfHV= ((uint8_t*)half);\

  2362.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2363.     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\

  2364.     OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\

  2365. }\

  2366. static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2367.     uint64_t half[17*2];\

  2368.     uint8_t * const halfH= ((uint8_t*)half);\

  2369.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2370.     put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\

  2371.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2372. }\

  2373. static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2374.     uint64_t half[17*2];\

  2375.     uint8_t * const halfH= ((uint8_t*)half);\

  2376.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2377.     put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\

  2378.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2379. }\

  2380. static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2381.     uint64_t half[17*2];\

  2382.     uint8_t * const halfH= ((uint8_t*)half);\

  2383.     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\

  2384.     OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\

  2385. }

  2386. 

  2387. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "        \n\t"

  2388. #define AVG_3DNOW_OP(a,b,temp, size) \

  2389. "mov" #size " " #b ", " #temp "   \n\t"\

  2390. "pavgusb " #temp ", " #a "        \n\t"\

  2391. "mov" #size " " #a ", " #b "      \n\t"

  2392. #define AVG_MMX2_OP(a,b,temp, size) \

  2393. "mov" #size " " #b ", " #temp "   \n\t"\

  2394. "pavgb " #temp ", " #a "          \n\t"\

  2395. "mov" #size " " #a ", " #b "      \n\t"

  2396. 

  2397. QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)

  2398. QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)

  2399. QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)

  2400. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)

  2401. QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)

  2402. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)

  2403. QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)

  2404. QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)

  2405. QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)

  2406. 

  2407. /***********************************/

  2408. /* bilinear qpel: not compliant to any spec, only for -lavdopts fast */

  2409. 

  2410. #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\

  2411. static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2412.     OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\

  2413. }

  2414. #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\

  2415. static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2416.     OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\

  2417. }

  2418. 

  2419. #define QPEL_2TAP(OPNAME, SIZE, MMX)\

  2420. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\

  2421. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\

  2422. QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\

  2423. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\

  2424.                           OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\

  2425. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\

  2426.                           OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\

  2427. static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\

  2428.                           OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\

  2429. static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2430.     OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\

  2431. }\

  2432. static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  2433.     OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\

  2434. }\

  2435. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0,         1,       0)\

  2436. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1,        -1,       0)\

  2437. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0,         stride,  0)\

  2438. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride,   -stride,  0)\

  2439. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0,         stride,  1)\

  2440. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1,         stride, -1)\

  2441. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride,   -stride,  1)\

  2442. QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\

  2443. 

  2444. QPEL_2TAP(put_, 16, mmx2)

  2445. QPEL_2TAP(avg_, 16, mmx2)

  2446. QPEL_2TAP(put_,  8, mmx2)

  2447. QPEL_2TAP(avg_,  8, mmx2)

  2448. QPEL_2TAP(put_, 16, 3dnow)

  2449. QPEL_2TAP(avg_, 16, 3dnow)

  2450. QPEL_2TAP(put_,  8, 3dnow)

  2451. QPEL_2TAP(avg_,  8, 3dnow)

  2452. 

  2453. 

  2454. #if 0

  2455. static void just_return() { return; }

  2456. #endif

  2457. 

  2458. #define SET_QPEL_FUNC(postfix1, postfix2) \

  2459.     c->put_ ## postfix1 = put_ ## postfix2;\

  2460.     c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\

  2461.     c->avg_ ## postfix1 = avg_ ## postfix2;

  2462. 

  2463. static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  2464.                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){

  2465.     const int w = 8;

  2466.     const int ix = ox>>(16+shift);

  2467.     const int iy = oy>>(16+shift);

  2468.     const int oxs = ox>>4;

  2469.     const int oys = oy>>4;

  2470.     const int dxxs = dxx>>4;

  2471.     const int dxys = dxy>>4;

  2472.     const int dyxs = dyx>>4;

  2473.     const int dyys = dyy>>4;

  2474.     const uint16_t r4[4] = {r,r,r,r};

  2475.     const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};

  2476.     const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};

  2477.     const uint64_t shift2 = 2*shift;

  2478.     uint8_t edge_buf[(h+1)*stride];

  2479.     int x, y;

  2480. 

  2481.     const int dxw = (dxx-(1<<(16+shift)))*(w-1);

  2482.     const int dyh = (dyy-(1<<(16+shift)))*(h-1);

  2483.     const int dxh = dxy*(h-1);

  2484.     const int dyw = dyx*(w-1);

  2485.     if( // non-constant fullpel offset (3% of blocks)

  2486.         (ox^(ox+dxw) | ox^(ox+dxh) | ox^(ox+dxw+dxh) |

  2487.          oy^(oy+dyw) | oy^(oy+dyh) | oy^(oy+dyw+dyh)) >> (16+shift)

  2488.         // uses more than 16 bits of subpel mv (only at huge resolution)

  2489.         || (dxx|dxy|dyx|dyy)&15 )

  2490.     {

  2491.         //FIXME could still use mmx for some of the rows

  2492.         ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);

  2493.         return;

  2494.     }

  2495. 

  2496.     src += ix + iy*stride;

  2497.     if( (unsigned)ix >= width-w ||

  2498.         (unsigned)iy >= height-h )

  2499.     {

  2500.         ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);

  2501.         src = edge_buf;

  2502.     }

  2503. 

  2504.     asm volatile(

  2505.         "movd         %0, %%mm6 \n\t"

  2506.         "pxor      %%mm7, %%mm7 \n\t"

  2507.         "punpcklwd %%mm6, %%mm6 \n\t"

  2508.         "punpcklwd %%mm6, %%mm6 \n\t"

  2509.         :: "r"(1<<shift)

  2510.     );

  2511. 

  2512.     for(x=0; x<w; x+=4){

  2513.         uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),

  2514.                             oxs - dxys + dxxs*(x+1),

  2515.                             oxs - dxys + dxxs*(x+2),

  2516.                             oxs - dxys + dxxs*(x+3) };

  2517.         uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),

  2518.                             oys - dyys + dyxs*(x+1),

  2519.                             oys - dyys + dyxs*(x+2),

  2520.                             oys - dyys + dyxs*(x+3) };

  2521. 

  2522.         for(y=0; y<h; y++){

  2523.             asm volatile(

  2524.                 "movq   %0,  %%mm4 \n\t"

  2525.                 "movq   %1,  %%mm5 \n\t"

  2526.                 "paddw  %2,  %%mm4 \n\t"

  2527.                 "paddw  %3,  %%mm5 \n\t"

  2528.                 "movq   %%mm4, %0  \n\t"

  2529.                 "movq   %%mm5, %1  \n\t"

  2530.                 "psrlw  $12, %%mm4 \n\t"

  2531.                 "psrlw  $12, %%mm5 \n\t"

  2532.                 : "+m"(*dx4), "+m"(*dy4)

  2533.                 : "m"(*dxy4), "m"(*dyy4)

  2534.             );

  2535. 

  2536.             asm volatile(

  2537.                 "movq   %%mm6, %%mm2 \n\t"

  2538.                 "movq   %%mm6, %%mm1 \n\t"

  2539.                 "psubw  %%mm4, %%mm2 \n\t"

  2540.                 "psubw  %%mm5, %%mm1 \n\t"

  2541.                 "movq   %%mm2, %%mm0 \n\t"

  2542.                 "movq   %%mm4, %%mm3 \n\t"

  2543.                 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)

  2544.                 "pmullw %%mm5, %%mm3 \n\t" // dx*dy

  2545.                 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy

  2546.                 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)

  2547. 

  2548.                 "movd   %4,    %%mm5 \n\t"

  2549.                 "movd   %3,    %%mm4 \n\t"

  2550.                 "punpcklbw %%mm7, %%mm5 \n\t"

  2551.                 "punpcklbw %%mm7, %%mm4 \n\t"

  2552.                 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy

  2553.                 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy

  2554. 

  2555.                 "movd   %2,    %%mm5 \n\t"

  2556.                 "movd   %1,    %%mm4 \n\t"

  2557.                 "punpcklbw %%mm7, %%mm5 \n\t"

  2558.                 "punpcklbw %%mm7, %%mm4 \n\t"

  2559.                 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)

  2560.                 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)

  2561.                 "paddw  %5,    %%mm1 \n\t"

  2562.                 "paddw  %%mm3, %%mm2 \n\t"

  2563.                 "paddw  %%mm1, %%mm0 \n\t"

  2564.                 "paddw  %%mm2, %%mm0 \n\t"

  2565. 

  2566.                 "psrlw    %6,    %%mm0 \n\t"

  2567.                 "packuswb %%mm0, %%mm0 \n\t"

  2568.                 "movd     %%mm0, %0    \n\t"

  2569. 

  2570.                 : "=m"(dst[x+y*stride])

  2571.                 : "m"(src[0]), "m"(src[1]),

  2572.                   "m"(src[stride]), "m"(src[stride+1]),

  2573.                   "m"(*r4), "m"(shift2)

  2574.             );

  2575.             src += stride;

  2576.         }

  2577.         src += 4-h*stride;

  2578.     }

  2579. }

  2580. 

  2581. #ifdef CONFIG_ENCODERS

  2582. static int try_8x8basis_mmx(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale){

  2583.     long i=0;

  2584. 

  2585.     assert(FFABS(scale) < 256);

  2586.     scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2587. 

  2588.     asm volatile(

  2589.         "pcmpeqw %%mm6, %%mm6           \n\t" // -1w

  2590.         "psrlw $15, %%mm6               \n\t" //  1w

  2591.         "pxor %%mm7, %%mm7              \n\t"

  2592.         "movd  %4, %%mm5                \n\t"

  2593.         "punpcklwd %%mm5, %%mm5         \n\t"

  2594.         "punpcklwd %%mm5, %%mm5         \n\t"

  2595.         "1:                             \n\t"

  2596.         "movq  (%1, %0), %%mm0          \n\t"

  2597.         "movq  8(%1, %0), %%mm1         \n\t"

  2598.         "pmulhw %%mm5, %%mm0            \n\t"

  2599.         "pmulhw %%mm5, %%mm1            \n\t"

  2600.         "paddw %%mm6, %%mm0             \n\t"

  2601.         "paddw %%mm6, %%mm1             \n\t"

  2602.         "psraw $1, %%mm0                \n\t"

  2603.         "psraw $1, %%mm1                \n\t"

  2604.         "paddw (%2, %0), %%mm0          \n\t"

  2605.         "paddw 8(%2, %0), %%mm1         \n\t"

  2606.         "psraw $6, %%mm0                \n\t"

  2607.         "psraw $6, %%mm1                \n\t"

  2608.         "pmullw (%3, %0), %%mm0         \n\t"

  2609.         "pmullw 8(%3, %0), %%mm1        \n\t"

  2610.         "pmaddwd %%mm0, %%mm0           \n\t"

  2611.         "pmaddwd %%mm1, %%mm1           \n\t"

  2612.         "paddd %%mm1, %%mm0             \n\t"

  2613.         "psrld $4, %%mm0                \n\t"

  2614.         "paddd %%mm0, %%mm7             \n\t"

  2615.         "add $16, %0                    \n\t"

  2616.         "cmp $128, %0                   \n\t" //FIXME optimize & bench

  2617.         " jb 1b                         \n\t"

  2618.         "movq %%mm7, %%mm6              \n\t"

  2619.         "psrlq $32, %%mm7               \n\t"

  2620.         "paddd %%mm6, %%mm7             \n\t"

  2621.         "psrld $2, %%mm7                \n\t"

  2622.         "movd %%mm7, %0                 \n\t"

  2623. 

  2624.         : "+r" (i)

  2625.         : "r"(basis), "r"(rem), "r"(weight), "g"(scale)

  2626.     );

  2627.     return i;

  2628. }

  2629. 

  2630. static void add_8x8basis_mmx(int16_t rem[64], int16_t basis[64], int scale){

  2631.     long i=0;

  2632. 

  2633.     if(FFABS(scale) < 256){

  2634.         scale<<= 16 + 1 - BASIS_SHIFT + RECON_SHIFT;

  2635.         asm volatile(

  2636.                 "pcmpeqw %%mm6, %%mm6   \n\t" // -1w

  2637.                 "psrlw $15, %%mm6       \n\t" //  1w

  2638.                 "movd  %3, %%mm5        \n\t"

  2639.                 "punpcklwd %%mm5, %%mm5 \n\t"

  2640.                 "punpcklwd %%mm5, %%mm5 \n\t"

  2641.                 "1:                     \n\t"

  2642.                 "movq  (%1, %0), %%mm0  \n\t"

  2643.                 "movq  8(%1, %0), %%mm1 \n\t"

  2644.                 "pmulhw %%mm5, %%mm0    \n\t"

  2645.                 "pmulhw %%mm5, %%mm1    \n\t"

  2646.                 "paddw %%mm6, %%mm0     \n\t"

  2647.                 "paddw %%mm6, %%mm1     \n\t"

  2648.                 "psraw $1, %%mm0        \n\t"

  2649.                 "psraw $1, %%mm1        \n\t"

  2650.                 "paddw (%2, %0), %%mm0  \n\t"

  2651.                 "paddw 8(%2, %0), %%mm1 \n\t"

  2652.                 "movq %%mm0, (%2, %0)   \n\t"

  2653.                 "movq %%mm1, 8(%2, %0)  \n\t"

  2654.                 "add $16, %0            \n\t"

  2655.                 "cmp $128, %0           \n\t" //FIXME optimize & bench

  2656.                 " jb 1b                 \n\t"

  2657. 

  2658.                 : "+r" (i)

  2659.                 : "r"(basis), "r"(rem), "g"(scale)

  2660.         );

  2661.     }else{

  2662.         for(i=0; i<8*8; i++){

  2663.             rem[i] += (basis[i]*scale + (1<<(BASIS_SHIFT - RECON_SHIFT-1)))>>(BASIS_SHIFT - RECON_SHIFT);

  2664.         }

  2665.     }

  2666. }

  2667. #endif /* CONFIG_ENCODERS */

  2668. 

  2669. #define PREFETCH(name, op) \

  2670. static void name(void *mem, int stride, int h){\

  2671.     const uint8_t *p= mem;\

  2672.     do{\

  2673.         asm volatile(#op" %0" :: "m"(*p));\

  2674.         p+= stride;\

  2675.     }while(--h);\

  2676. }

  2677. PREFETCH(prefetch_mmx2,  prefetcht0)

  2678. PREFETCH(prefetch_3dnow, prefetch)

  2679. #undef PREFETCH

  2680. 

  2681. #include "h264dsp_mmx.c"

  2682. 

  2683. /* AVS specific */

  2684. void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx);

  2685. 

  2686. void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2687.     put_pixels8_mmx(dst, src, stride, 8);

  2688. }

  2689. void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2690.     avg_pixels8_mmx(dst, src, stride, 8);

  2691. }

  2692. void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2693.     put_pixels16_mmx(dst, src, stride, 16);

  2694. }

  2695. void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {

  2696.     avg_pixels16_mmx(dst, src, stride, 16);

  2697. }

  2698. 

  2699. /* external functions, from idct_mmx.c */

  2700. void ff_mmx_idct(DCTELEM *block);

  2701. void ff_mmxext_idct(DCTELEM *block);

  2702. 

  2703. void ff_vp3_idct_sse2(int16_t *input_data);

  2704. void ff_vp3_idct_mmx(int16_t *data);

  2705. void ff_vp3_dsp_init_mmx(void);

  2706. 

  2707. /* XXX: those functions should be suppressed ASAP when all IDCTs are

  2708.    converted */

  2709. #ifdef CONFIG_GPL

  2710. static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2711. {

  2712.     ff_mmx_idct (block);

  2713.     put_pixels_clamped_mmx(block, dest, line_size);

  2714. }

  2715. static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2716. {

  2717.     ff_mmx_idct (block);

  2718.     add_pixels_clamped_mmx(block, dest, line_size);

  2719. }

  2720. static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)

  2721. {

  2722.     ff_mmxext_idct (block);

  2723.     put_pixels_clamped_mmx(block, dest, line_size);

  2724. }

  2725. static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)

  2726. {

  2727.     ff_mmxext_idct (block);

  2728.     add_pixels_clamped_mmx(block, dest, line_size);

  2729. }

  2730. #endif

  2731. static void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block)

  2732. {

  2733.     ff_vp3_idct_sse2(block);

  2734.     put_signed_pixels_clamped_mmx(block, dest, line_size);

  2735. }

  2736. static void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block)

  2737. {

  2738.     ff_vp3_idct_sse2(block);

  2739.     add_pixels_clamped_mmx(block, dest, line_size);

  2740. }

  2741. static void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)

  2742. {

  2743.     ff_vp3_idct_mmx(block);

  2744.     put_signed_pixels_clamped_mmx(block, dest, line_size);

  2745. }

  2746. static void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)

  2747. {

  2748.     ff_vp3_idct_mmx(block);

  2749.     add_pixels_clamped_mmx(block, dest, line_size);

  2750. }

  2751. static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)

  2752. {

  2753.     ff_idct_xvid_mmx (block);

  2754.     put_pixels_clamped_mmx(block, dest, line_size);

  2755. }

  2756. static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)

  2757. {

  2758.     ff_idct_xvid_mmx (block);

  2759.     add_pixels_clamped_mmx(block, dest, line_size);

  2760. }

  2761. static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)

  2762. {

  2763.     ff_idct_xvid_mmx2 (block);

  2764.     put_pixels_clamped_mmx(block, dest, line_size);

  2765. }

  2766. static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)

  2767. {

  2768.     ff_idct_xvid_mmx2 (block);

  2769.     add_pixels_clamped_mmx(block, dest, line_size);

  2770. }

  2771. 

  2772. static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)

  2773. {

  2774.     int i;

  2775.     asm volatile("pxor %%mm7, %%mm7":);

  2776.     for(i=0; i<blocksize; i+=2) {

  2777.         asm volatile(

  2778.             "movq    %0,    %%mm0 \n\t"

  2779.             "movq    %1,    %%mm1 \n\t"

  2780.             "movq    %%mm0, %%mm2 \n\t"

  2781.             "movq    %%mm1, %%mm3 \n\t"

  2782.             "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0

  2783.             "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0

  2784.             "pslld   $31,   %%mm2 \n\t" // keep only the sign bit

  2785.             "pxor    %%mm2, %%mm1 \n\t"

  2786.             "movq    %%mm3, %%mm4 \n\t"

  2787.             "pand    %%mm1, %%mm3 \n\t"

  2788.             "pandn   %%mm1, %%mm4 \n\t"

  2789.             "pfadd   %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))

  2790.             "pfsub   %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))

  2791.             "movq    %%mm3, %1    \n\t"

  2792.             "movq    %%mm0, %0    \n\t"

  2793.             :"+m"(mag[i]), "+m"(ang[i])

  2794.             ::"memory"

  2795.         );

  2796.     }

  2797.     asm volatile("femms");

  2798. }

  2799. static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)

  2800. {

  2801.     int i;

  2802. 

  2803.     asm volatile(

  2804.             "movaps  %0,     %%xmm5 \n\t"

  2805.         ::"m"(ff_pdw_80000000[0])

  2806.     );

  2807.     for(i=0; i<blocksize; i+=4) {

  2808.         asm volatile(

  2809.             "movaps  %0,     %%xmm0 \n\t"

  2810.             "movaps  %1,     %%xmm1 \n\t"

  2811.             "xorps   %%xmm2, %%xmm2 \n\t"

  2812.             "xorps   %%xmm3, %%xmm3 \n\t"

  2813.             "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0

  2814.             "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0

  2815.             "andps   %%xmm5, %%xmm2 \n\t" // keep only the sign bit

  2816.             "xorps   %%xmm2, %%xmm1 \n\t"

  2817.             "movaps  %%xmm3, %%xmm4 \n\t"

  2818.             "andps   %%xmm1, %%xmm3 \n\t"

  2819.             "andnps  %%xmm1, %%xmm4 \n\t"

  2820.             "addps   %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))

  2821.             "subps   %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))

  2822.             "movaps  %%xmm3, %1     \n\t"

  2823.             "movaps  %%xmm0, %0     \n\t"

  2824.             :"+m"(mag[i]), "+m"(ang[i])

  2825.             ::"memory"

  2826.         );

  2827.     }

  2828. }

  2829. 

  2830. static void vector_fmul_3dnow(float *dst, const float *src, int len){

  2831.     long i = (len-4)*4;

  2832.     asm volatile(

  2833.         "1: \n\t"

  2834.         "movq    (%1,%0), %%mm0 \n\t"

  2835.         "movq   8(%1,%0), %%mm1 \n\t"

  2836.         "pfmul   (%2,%0), %%mm0 \n\t"

  2837.         "pfmul  8(%2,%0), %%mm1 \n\t"

  2838.         "movq   %%mm0,  (%1,%0) \n\t"

  2839.         "movq   %%mm1, 8(%1,%0) \n\t"

  2840.         "sub  $16, %0 \n\t"

  2841.         "jge 1b \n\t"

  2842.         "femms  \n\t"

  2843.         :"+r"(i)

  2844.         :"r"(dst), "r"(src)

  2845.         :"memory"

  2846.     );

  2847. }

  2848. static void vector_fmul_sse(float *dst, const float *src, int len){

  2849.     long i = (len-8)*4;

  2850.     asm volatile(

  2851.         "1: \n\t"

  2852.         "movaps    (%1,%0), %%xmm0 \n\t"

  2853.         "movaps  16(%1,%0), %%xmm1 \n\t"

  2854.         "mulps     (%2,%0), %%xmm0 \n\t"

  2855.         "mulps   16(%2,%0), %%xmm1 \n\t"

  2856.         "movaps  %%xmm0,   (%1,%0) \n\t"

  2857.         "movaps  %%xmm1, 16(%1,%0) \n\t"

  2858.         "sub  $32, %0 \n\t"

  2859.         "jge 1b \n\t"

  2860.         :"+r"(i)

  2861.         :"r"(dst), "r"(src)

  2862.         :"memory"

  2863.     );

  2864. }

  2865. 

  2866. static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){

  2867.     long i = len*4-16;

  2868.     asm volatile(

  2869.         "1: \n\t"

  2870.         "pswapd   8(%1), %%mm0 \n\t"

  2871.         "pswapd    (%1), %%mm1 \n\t"

  2872.         "pfmul  (%3,%0), %%mm0 \n\t"

  2873.         "pfmul 8(%3,%0), %%mm1 \n\t"

  2874.         "movq  %%mm0,  (%2,%0) \n\t"

  2875.         "movq  %%mm1, 8(%2,%0) \n\t"

  2876.         "add   $16, %1 \n\t"

  2877.         "sub   $16, %0 \n\t"

  2878.         "jge   1b \n\t"

  2879.         :"+r"(i), "+r"(src1)

  2880.         :"r"(dst), "r"(src0)

  2881.     );

  2882.     asm volatile("femms");

  2883. }

  2884. static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){

  2885.     long i = len*4-32;

  2886.     asm volatile(

  2887.         "1: \n\t"

  2888.         "movaps        16(%1), %%xmm0 \n\t"

  2889.         "movaps          (%1), %%xmm1 \n\t"

  2890.         "shufps $0x1b, %%xmm0, %%xmm0 \n\t"

  2891.         "shufps $0x1b, %%xmm1, %%xmm1 \n\t"

  2892.         "mulps        (%3,%0), %%xmm0 \n\t"

  2893.         "mulps      16(%3,%0), %%xmm1 \n\t"

  2894.         "movaps     %%xmm0,   (%2,%0) \n\t"

  2895.         "movaps     %%xmm1, 16(%2,%0) \n\t"

  2896.         "add    $32, %1 \n\t"

  2897.         "sub    $32, %0 \n\t"

  2898.         "jge    1b \n\t"

  2899.         :"+r"(i), "+r"(src1)

  2900.         :"r"(dst), "r"(src0)

  2901.     );

  2902. }

  2903. 

  2904. static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1,

  2905.                                       const float *src2, int src3, int len, int step){

  2906.     long i = (len-4)*4;

  2907.     if(step == 2 && src3 == 0){

  2908.         dst += (len-4)*2;

  2909.         asm volatile(

  2910.             "1: \n\t"

  2911.             "movq   (%2,%0),  %%mm0 \n\t"

  2912.             "movq  8(%2,%0),  %%mm1 \n\t"

  2913.             "pfmul  (%3,%0),  %%mm0 \n\t"

  2914.             "pfmul 8(%3,%0),  %%mm1 \n\t"

  2915.             "pfadd  (%4,%0),  %%mm0 \n\t"

  2916.             "pfadd 8(%4,%0),  %%mm1 \n\t"

  2917.             "movd     %%mm0,   (%1) \n\t"

  2918.             "movd     %%mm1, 16(%1) \n\t"

  2919.             "psrlq      $32,  %%mm0 \n\t"

  2920.             "psrlq      $32,  %%mm1 \n\t"

  2921.             "movd     %%mm0,  8(%1) \n\t"

  2922.             "movd     %%mm1, 24(%1) \n\t"

  2923.             "sub  $32, %1 \n\t"

  2924.             "sub  $16, %0 \n\t"

  2925.             "jge  1b \n\t"

  2926.             :"+r"(i), "+r"(dst)

  2927.             :"r"(src0), "r"(src1), "r"(src2)

  2928.             :"memory"

  2929.         );

  2930.     }

  2931.     else if(step == 1 && src3 == 0){

  2932.         asm volatile(

  2933.             "1: \n\t"

  2934.             "movq    (%2,%0), %%mm0 \n\t"

  2935.             "movq   8(%2,%0), %%mm1 \n\t"

  2936.             "pfmul   (%3,%0), %%mm0 \n\t"

  2937.             "pfmul  8(%3,%0), %%mm1 \n\t"

  2938.             "pfadd   (%4,%0), %%mm0 \n\t"

  2939.             "pfadd  8(%4,%0), %%mm1 \n\t"

  2940.             "movq  %%mm0,   (%1,%0) \n\t"

  2941.             "movq  %%mm1,  8(%1,%0) \n\t"

  2942.             "sub  $16, %0 \n\t"

  2943.             "jge  1b \n\t"

  2944.             :"+r"(i)

  2945.             :"r"(dst), "r"(src0), "r"(src1), "r"(src2)

  2946.             :"memory"

  2947.         );

  2948.     }

  2949.     else

  2950.         ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);

  2951.     asm volatile("femms");

  2952. }

  2953. static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *src1,

  2954.                                     const float *src2, int src3, int len, int step){

  2955.     long i = (len-8)*4;

  2956.     if(step == 2 && src3 == 0){

  2957.         dst += (len-8)*2;

  2958.         asm volatile(

  2959.             "1: \n\t"

  2960.             "movaps   (%2,%0), %%xmm0 \n\t"

  2961.             "movaps 16(%2,%0), %%xmm1 \n\t"

  2962.             "mulps    (%3,%0), %%xmm0 \n\t"

  2963.             "mulps  16(%3,%0), %%xmm1 \n\t"

  2964.             "addps    (%4,%0), %%xmm0 \n\t"

  2965.             "addps  16(%4,%0), %%xmm1 \n\t"

  2966.             "movss     %%xmm0,   (%1) \n\t"

  2967.             "movss     %%xmm1, 32(%1) \n\t"

  2968.             "movhlps   %%xmm0, %%xmm2 \n\t"

  2969.             "movhlps   %%xmm1, %%xmm3 \n\t"

  2970.             "movss     %%xmm2, 16(%1) \n\t"

  2971.             "movss     %%xmm3, 48(%1) \n\t"

  2972.             "shufps $0xb1, %%xmm0, %%xmm0 \n\t"

  2973.             "shufps $0xb1, %%xmm1, %%xmm1 \n\t"

  2974.             "movss     %%xmm0,  8(%1) \n\t"

  2975.             "movss     %%xmm1, 40(%1) \n\t"

  2976.             "movhlps   %%xmm0, %%xmm2 \n\t"

  2977.             "movhlps   %%xmm1, %%xmm3 \n\t"

  2978.             "movss     %%xmm2, 24(%1) \n\t"

  2979.             "movss     %%xmm3, 56(%1) \n\t"

  2980.             "sub  $64, %1 \n\t"

  2981.             "sub  $32, %0 \n\t"

  2982.             "jge  1b \n\t"

  2983.             :"+r"(i), "+r"(dst)

  2984.             :"r"(src0), "r"(src1), "r"(src2)

  2985.             :"memory"

  2986.         );

  2987.     }

  2988.     else if(step == 1 && src3 == 0){

  2989.         asm volatile(

  2990.             "1: \n\t"

  2991.             "movaps   (%2,%0), %%xmm0 \n\t"

  2992.             "movaps 16(%2,%0), %%xmm1 \n\t"

  2993.             "mulps    (%3,%0), %%xmm0 \n\t"

  2994.             "mulps  16(%3,%0), %%xmm1 \n\t"

  2995.             "addps    (%4,%0), %%xmm0 \n\t"

  2996.             "addps  16(%4,%0), %%xmm1 \n\t"

  2997.             "movaps %%xmm0,   (%1,%0) \n\t"

  2998.             "movaps %%xmm1, 16(%1,%0) \n\t"

  2999.             "sub  $32, %0 \n\t"

  3000.             "jge  1b \n\t"

  3001.             :"+r"(i)

  3002.             :"r"(dst), "r"(src0), "r"(src1), "r"(src2)

  3003.             :"memory"

  3004.         );

  3005.     }

  3006.     else

  3007.         ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);

  3008. }

  3009. 

  3010. static void float_to_int16_3dnow(int16_t *dst, const float *src, int len){

  3011.     // not bit-exact: pf2id uses different rounding than C and SSE

  3012.     int i;

  3013.     for(i=0; i<len; i+=4) {

  3014.         asm volatile(

  3015.             "pf2id       %1, %%mm0 \n\t"

  3016.             "pf2id       %2, %%mm1 \n\t"

  3017.             "packssdw %%mm1, %%mm0 \n\t"

  3018.             "movq     %%mm0, %0    \n\t"

  3019.             :"=m"(dst[i])

  3020.             :"m"(src[i]), "m"(src[i+2])

  3021.         );

  3022.     }

  3023.     asm volatile("femms");

  3024. }

  3025. static void float_to_int16_sse(int16_t *dst, const float *src, int len){

  3026.     int i;

  3027.     for(i=0; i<len; i+=4) {

  3028.         asm volatile(

  3029.             "cvtps2pi    %1, %%mm0 \n\t"

  3030.             "cvtps2pi    %2, %%mm1 \n\t"

  3031.             "packssdw %%mm1, %%mm0 \n\t"

  3032.             "movq     %%mm0, %0    \n\t"

  3033.             :"=m"(dst[i])

  3034.             :"m"(src[i]), "m"(src[i+2])

  3035.         );

  3036.     }

  3037.     asm volatile("emms");

  3038. }

  3039. 

  3040. #ifdef CONFIG_SNOW_DECODER

  3041. extern void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width);

  3042. extern void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width);

  3043. extern void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);

  3044. extern void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);

  3045. extern void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,

  3046.                            int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);

  3047. extern void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,

  3048.                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);

  3049. #endif

  3050. 

  3051. void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)

  3052. {

  3053.     mm_flags = mm_support();

  3054. 

  3055.     if (avctx->dsp_mask) {

  3056.         if (avctx->dsp_mask & FF_MM_FORCE)

  3057.             mm_flags |= (avctx->dsp_mask & 0xffff);

  3058.         else

  3059.             mm_flags &= ~(avctx->dsp_mask & 0xffff);

  3060.     }

  3061. 

  3062. #if 0

  3063.     av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");

  3064.     if (mm_flags & MM_MMX)

  3065.         av_log(avctx, AV_LOG_INFO, " mmx");

  3066.     if (mm_flags & MM_MMXEXT)

  3067.         av_log(avctx, AV_LOG_INFO, " mmxext");

  3068.     if (mm_flags & MM_3DNOW)

  3069.         av_log(avctx, AV_LOG_INFO, " 3dnow");

  3070.     if (mm_flags & MM_SSE)

  3071.         av_log(avctx, AV_LOG_INFO, " sse");

  3072.     if (mm_flags & MM_SSE2)

  3073.         av_log(avctx, AV_LOG_INFO, " sse2");

  3074.     av_log(avctx, AV_LOG_INFO, "\n");

  3075. #endif

  3076. 

  3077.     if (mm_flags & MM_MMX) {

  3078.         const int idct_algo= avctx->idct_algo;

  3079. 

  3080. #ifdef CONFIG_ENCODERS

  3081.         const int dct_algo = avctx->dct_algo;

  3082.         if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){

  3083.             if(mm_flags & MM_SSE2){

  3084.                 c->fdct = ff_fdct_sse2;

  3085.             }else if(mm_flags & MM_MMXEXT){

  3086.                 c->fdct = ff_fdct_mmx2;

  3087.             }else{

  3088.                 c->fdct = ff_fdct_mmx;

  3089.             }

  3090.         }

  3091. #endif //CONFIG_ENCODERS

  3092.         if(avctx->lowres==0){

  3093.             if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){

  3094.                 c->idct_put= ff_simple_idct_put_mmx;

  3095.                 c->idct_add= ff_simple_idct_add_mmx;

  3096.                 c->idct    = ff_simple_idct_mmx;

  3097.                 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;

  3098. #ifdef CONFIG_GPL

  3099.             }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){

  3100.                 if(mm_flags & MM_MMXEXT){

  3101.                     c->idct_put= ff_libmpeg2mmx2_idct_put;

  3102.                     c->idct_add= ff_libmpeg2mmx2_idct_add;

  3103.                     c->idct    = ff_mmxext_idct;

  3104.                 }else{

  3105.                     c->idct_put= ff_libmpeg2mmx_idct_put;

  3106.                     c->idct_add= ff_libmpeg2mmx_idct_add;

  3107.                     c->idct    = ff_mmx_idct;

  3108.                 }

  3109.                 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;

  3110. #endif

  3111.             }else if(idct_algo==FF_IDCT_VP3 &&

  3112.                      avctx->codec->id!=CODEC_ID_THEORA &&

  3113.                      !(avctx->flags & CODEC_FLAG_BITEXACT)){

  3114.                 if(mm_flags & MM_SSE2){

  3115.                     c->idct_put= ff_vp3_idct_put_sse2;

  3116.                     c->idct_add= ff_vp3_idct_add_sse2;

  3117.                     c->idct    = ff_vp3_idct_sse2;

  3118.                     c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;

  3119.                 }else{

  3120.                     ff_vp3_dsp_init_mmx();

  3121.                     c->idct_put= ff_vp3_idct_put_mmx;

  3122.                     c->idct_add= ff_vp3_idct_add_mmx;

  3123.                     c->idct    = ff_vp3_idct_mmx;

  3124.                     c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;

  3125.                 }

  3126.             }else if(idct_algo==FF_IDCT_CAVS){

  3127.                     c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;

  3128.             }else if(idct_algo==FF_IDCT_XVIDMMX){

  3129.                 if(mm_flags & MM_MMXEXT){

  3130.                     c->idct_put= ff_idct_xvid_mmx2_put;

  3131.                     c->idct_add= ff_idct_xvid_mmx2_add;

  3132.                     c->idct    = ff_idct_xvid_mmx2;

  3133.                 }else{

  3134.                     c->idct_put= ff_idct_xvid_mmx_put;

  3135.                     c->idct_add= ff_idct_xvid_mmx_add;

  3136.                     c->idct    = ff_idct_xvid_mmx;

  3137.                 }

  3138.             }

  3139.         }

  3140. 

  3141. #ifdef CONFIG_ENCODERS

  3142.         c->get_pixels = get_pixels_mmx;

  3143.         c->diff_pixels = diff_pixels_mmx;

  3144. #endif //CONFIG_ENCODERS

  3145.         c->put_pixels_clamped = put_pixels_clamped_mmx;

  3146.         c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;

  3147.         c->add_pixels_clamped = add_pixels_clamped_mmx;

  3148.         c->clear_blocks = clear_blocks_mmx;

  3149. #ifdef CONFIG_ENCODERS

  3150.         c->pix_sum = pix_sum16_mmx;

  3151. #endif //CONFIG_ENCODERS

  3152. 

  3153.         c->put_pixels_tab[0][0] = put_pixels16_mmx;

  3154.         c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;

  3155.         c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;

  3156.         c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;

  3157. 

  3158.         c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;

  3159.         c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;

  3160.         c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;

  3161.         c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;

  3162. 

  3163.         c->avg_pixels_tab[0][0] = avg_pixels16_mmx;

  3164.         c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;

  3165.         c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;

  3166.         c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;

  3167. 

  3168.         c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;

  3169.         c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;

  3170.         c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;

  3171.         c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;

  3172. 

  3173.         c->put_pixels_tab[1][0] = put_pixels8_mmx;

  3174.         c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;

  3175.         c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;

  3176.         c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;

  3177. 

  3178.         c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;

  3179.         c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;

  3180.         c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;

  3181.         c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;

  3182. 

  3183.         c->avg_pixels_tab[1][0] = avg_pixels8_mmx;

  3184.         c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;

  3185.         c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;

  3186.         c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;

  3187. 

  3188.         c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;

  3189.         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;

  3190.         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;

  3191.         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;

  3192. 

  3193.         c->gmc= gmc_mmx;

  3194. 

  3195.         c->add_bytes= add_bytes_mmx;

  3196. #ifdef CONFIG_ENCODERS

  3197.         c->diff_bytes= diff_bytes_mmx;

  3198. 

  3199.         c->hadamard8_diff[0]= hadamard8_diff16_mmx;

  3200.         c->hadamard8_diff[1]= hadamard8_diff_mmx;

  3201. 

  3202.         c->pix_norm1 = pix_norm1_mmx;

  3203.         c->sse[0] = (mm_flags & MM_SSE2) ? sse16_sse2 : sse16_mmx;

  3204.           c->sse[1] = sse8_mmx;

  3205.         c->vsad[4]= vsad_intra16_mmx;

  3206. 

  3207.         c->nsse[0] = nsse16_mmx;

  3208.         c->nsse[1] = nsse8_mmx;

  3209.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3210.             c->vsad[0] = vsad16_mmx;

  3211.         }

  3212. 

  3213.         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3214.             c->try_8x8basis= try_8x8basis_mmx;

  3215.         }

  3216.         c->add_8x8basis= add_8x8basis_mmx;

  3217. 

  3218. #endif //CONFIG_ENCODERS

  3219. 

  3220.         c->h263_v_loop_filter= h263_v_loop_filter_mmx;

  3221.         c->h263_h_loop_filter= h263_h_loop_filter_mmx;

  3222.         c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx;

  3223.         c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx;

  3224. 

  3225.         c->h264_idct_dc_add=

  3226.         c->h264_idct_add= ff_h264_idct_add_mmx;

  3227.         c->h264_idct8_dc_add=

  3228.         c->h264_idct8_add= ff_h264_idct8_add_mmx;

  3229. 

  3230.         if (mm_flags & MM_MMXEXT) {

  3231.             c->prefetch = prefetch_mmx2;

  3232. 

  3233.             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;

  3234.             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;

  3235. 

  3236.             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;

  3237.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;

  3238.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;

  3239. 

  3240.             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;

  3241.             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;

  3242. 

  3243.             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;

  3244.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;

  3245.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;

  3246. 

  3247. #ifdef CONFIG_ENCODERS

  3248.             c->hadamard8_diff[0]= hadamard8_diff16_mmx2;

  3249.             c->hadamard8_diff[1]= hadamard8_diff_mmx2;

  3250.             c->vsad[4]= vsad_intra16_mmx2;

  3251. #endif //CONFIG_ENCODERS

  3252. 

  3253.             c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;

  3254.             c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;

  3255. 

  3256.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3257.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;

  3258.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;

  3259.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;

  3260.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;

  3261.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;

  3262.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;

  3263. #ifdef CONFIG_ENCODERS

  3264.                 c->vsad[0] = vsad16_mmx2;

  3265. #endif //CONFIG_ENCODERS

  3266.             }

  3267. 

  3268. #if 1

  3269.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)

  3270.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)

  3271.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)

  3272.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)

  3273.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)

  3274.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)

  3275.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)

  3276.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)

  3277.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)

  3278.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)

  3279.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)

  3280.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)

  3281.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)

  3282.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)

  3283.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)

  3284.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)

  3285.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)

  3286.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)

  3287.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)

  3288.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)

  3289.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)

  3290.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)

  3291.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)

  3292.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)

  3293.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)

  3294.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)

  3295.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)

  3296.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)

  3297.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)

  3298.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)

  3299.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)

  3300.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)

  3301. #endif

  3302. 

  3303. //FIXME 3dnow too

  3304. #define dspfunc(PFX, IDX, NUM) \

  3305.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_mmx2; \

  3306.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_mmx2; \

  3307.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_mmx2; \

  3308.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_mmx2; \

  3309.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_mmx2; \

  3310.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_mmx2; \

  3311.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_mmx2; \

  3312.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_mmx2; \

  3313.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_mmx2; \

  3314.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_mmx2; \

  3315.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_mmx2; \

  3316.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_mmx2; \

  3317.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_mmx2; \

  3318.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_mmx2; \

  3319.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_mmx2; \

  3320.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_mmx2

  3321. 

  3322.             dspfunc(put_h264_qpel, 0, 16);

  3323.             dspfunc(put_h264_qpel, 1, 8);

  3324.             dspfunc(put_h264_qpel, 2, 4);

  3325.             dspfunc(avg_h264_qpel, 0, 16);

  3326.             dspfunc(avg_h264_qpel, 1, 8);

  3327.             dspfunc(avg_h264_qpel, 2, 4);

  3328. 

  3329.             dspfunc(put_2tap_qpel, 0, 16);

  3330.             dspfunc(put_2tap_qpel, 1, 8);

  3331.             dspfunc(avg_2tap_qpel, 0, 16);

  3332.             dspfunc(avg_2tap_qpel, 1, 8);

  3333. #undef dspfunc

  3334. 

  3335.             c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2;

  3336.             c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2;

  3337.             c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2;

  3338.             c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2;

  3339.             c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;

  3340.             c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;

  3341.             c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;

  3342.             c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;

  3343.             c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;

  3344.             c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;

  3345.             c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;

  3346. 

  3347.             c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;

  3348.             c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;

  3349.             c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;

  3350.             c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;

  3351.             c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;

  3352.             c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;

  3353.             c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;

  3354.             c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;

  3355. 

  3356.             c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;

  3357.             c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;

  3358.             c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;

  3359.             c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;

  3360.             c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;

  3361.             c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;

  3362.             c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;

  3363.             c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;

  3364. 

  3365. #ifdef CONFIG_CAVS_DECODER

  3366.             ff_cavsdsp_init_mmx2(c, avctx);

  3367. #endif

  3368. 

  3369. #ifdef CONFIG_ENCODERS

  3370.             c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;

  3371. #endif //CONFIG_ENCODERS

  3372.         } else if (mm_flags & MM_3DNOW) {

  3373.             c->prefetch = prefetch_3dnow;

  3374. 

  3375.             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;

  3376.             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;

  3377. 

  3378.             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;

  3379.             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;

  3380.             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;

  3381. 

  3382.             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;

  3383.             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;

  3384. 

  3385.             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;

  3386.             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;

  3387.             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;

  3388. 

  3389.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  3390.                 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;

  3391.                 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;

  3392.                 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;

  3393.                 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;

  3394.                 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;

  3395.                 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;

  3396.             }

  3397. 

  3398.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)

  3399.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)

  3400.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)

  3401.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)

  3402.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)

  3403.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)

  3404.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)

  3405.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)

  3406.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)

  3407.             SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)

  3408.             SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)

  3409.             SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)

  3410.             SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)

  3411.             SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)

  3412.             SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)

  3413.             SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)

  3414.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)

  3415.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)

  3416.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)

  3417.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)

  3418.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)

  3419.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)

  3420.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)

  3421.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)

  3422.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)

  3423.             SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)

  3424.             SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)

  3425.             SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)

  3426.             SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)

  3427.             SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)

  3428.             SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)

  3429.             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)

  3430. 

  3431. #define dspfunc(PFX, IDX, NUM) \

  3432.     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_3dnow; \

  3433.     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_3dnow; \

  3434.     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_3dnow; \

  3435.     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_3dnow; \

  3436.     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_3dnow; \

  3437.     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_3dnow; \

  3438.     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_3dnow; \

  3439.     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_3dnow; \

  3440.     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_3dnow; \

  3441.     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_3dnow; \

  3442.     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_3dnow; \

  3443.     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_3dnow; \

  3444.     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_3dnow; \

  3445.     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_3dnow; \

  3446.     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_3dnow; \

  3447.     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_3dnow

  3448. 

  3449.             dspfunc(put_h264_qpel, 0, 16);

  3450.             dspfunc(put_h264_qpel, 1, 8);

  3451.             dspfunc(put_h264_qpel, 2, 4);

  3452.             dspfunc(avg_h264_qpel, 0, 16);

  3453.             dspfunc(avg_h264_qpel, 1, 8);

  3454.             dspfunc(avg_h264_qpel, 2, 4);

  3455. 

  3456.             dspfunc(put_2tap_qpel, 0, 16);

  3457.             dspfunc(put_2tap_qpel, 1, 8);

  3458.             dspfunc(avg_2tap_qpel, 0, 16);

  3459.             dspfunc(avg_2tap_qpel, 1, 8);

  3460. 

  3461.             c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow;

  3462.             c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow;

  3463.         }

  3464. 

  3465. #ifdef CONFIG_SNOW_DECODER

  3466.         if(mm_flags & MM_SSE2){

  3467.             c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;

  3468.             c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;

  3469.             c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;

  3470.         }

  3471.         else{

  3472.             c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;

  3473.             c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;

  3474.             c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;

  3475.         }

  3476. #endif

  3477. 

  3478.         if(mm_flags & MM_3DNOW){

  3479.             c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;

  3480.             c->vector_fmul = vector_fmul_3dnow;

  3481.             if(!(avctx->flags & CODEC_FLAG_BITEXACT))

  3482.                 c->float_to_int16 = float_to_int16_3dnow;

  3483.         }

  3484.         if(mm_flags & MM_3DNOWEXT)

  3485.             c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;

  3486.         if(mm_flags & MM_SSE){

  3487.             c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;

  3488.             c->vector_fmul = vector_fmul_sse;

  3489.             c->float_to_int16 = float_to_int16_sse;

  3490.             c->vector_fmul_reverse = vector_fmul_reverse_sse;

  3491.             c->vector_fmul_add_add = vector_fmul_add_add_sse;

  3492.         }

  3493.         if(mm_flags & MM_3DNOW)

  3494.             c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse

  3495.     }

  3496. 

  3497. #ifdef CONFIG_ENCODERS

  3498.     dsputil_init_pix_mmx(c, avctx);

  3499. #endif //CONFIG_ENCODERS

  3500. #if 0

  3501.     // for speed testing

  3502.     get_pixels = just_return;

  3503.     put_pixels_clamped = just_return;

  3504.     add_pixels_clamped = just_return;

  3505. 

  3506.     pix_abs16x16 = just_return;

  3507.     pix_abs16x16_x2 = just_return;

  3508.     pix_abs16x16_y2 = just_return;

  3509.     pix_abs16x16_xy2 = just_return;

  3510. 

  3511.     put_pixels_tab[0] = just_return;

  3512.     put_pixels_tab[1] = just_return;

  3513.     put_pixels_tab[2] = just_return;

  3514.     put_pixels_tab[3] = just_return;

  3515. 

  3516.     put_no_rnd_pixels_tab[0] = just_return;

  3517.     put_no_rnd_pixels_tab[1] = just_return;

  3518.     put_no_rnd_pixels_tab[2] = just_return;

  3519.     put_no_rnd_pixels_tab[3] = just_return;

  3520. 

  3521.     avg_pixels_tab[0] = just_return;

  3522.     avg_pixels_tab[1] = just_return;

  3523.     avg_pixels_tab[2] = just_return;

  3524.     avg_pixels_tab[3] = just_return;

  3525. 

  3526.     avg_no_rnd_pixels_tab[0] = just_return;

  3527.     avg_no_rnd_pixels_tab[1] = just_return;

  3528.     avg_no_rnd_pixels_tab[2] = just_return;

  3529.     avg_no_rnd_pixels_tab[3] = just_return;

  3530. 

  3531.     //av_fdct = just_return;

  3532.     //ff_idct = just_return;

  3533. #endif

  3534. }