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.
33469 Commits
32 Branches
1.0GB
Tree: 4b4a6ebd68
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '4b4a6ebd68'
${ noResults }
FFmpeg/libavcodec/x86/cavsdsp_mmx.c

481 lines
19KB
Raw Blame History 

  1. /*

  2.  * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.

  3.  * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>

  4.  *

  5.  * MMX-optimized DSP functions, based on H.264 optimizations by

  6.  * Michael Niedermayer and Loren Merritt

  7.  *

  8.  * This file is part of FFmpeg.

  9.  *

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

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

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

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

  14.  *

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

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

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

  18.  * Lesser General Public License for more details.

  19.  *

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

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

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

  23.  */

  24. 

  25. #include "libavutil/common.h"

  26. #include "libavutil/cpu.h"

  27. #include "libavutil/x86_cpu.h"

  28. #include "libavcodec/dsputil.h"

  29. #include "libavcodec/cavsdsp.h"

  30. #include "dsputil_mmx.h"

  31. 

  32. /*****************************************************************************

  33.  *

  34.  * inverse transform

  35.  *

  36.  ****************************************************************************/

  37. 

  38. static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)

  39. {

  40.     __asm__ volatile(

  41.         "movq 112(%0), %%mm4  \n\t" /* mm4 = src7 */

  42.         "movq  16(%0), %%mm5  \n\t" /* mm5 = src1 */

  43.         "movq  80(%0), %%mm2  \n\t" /* mm2 = src5 */

  44.         "movq  48(%0), %%mm7  \n\t" /* mm7 = src3 */

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

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

  47.         "movq   %%mm2, %%mm6  \n\t"

  48.         "movq   %%mm7, %%mm1  \n\t"

  49. 

  50.         "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */

  51.         "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */

  52.         "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */

  53.         "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */

  54.         "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */

  55.         "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */

  56.         "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */

  57.         "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */

  58.         "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */

  59.         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */

  60.         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */

  61.         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */

  62. 

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

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

  65.         "movq   %%mm3, %%mm0  \n\t"

  66.         "movq   %%mm1, %%mm2  \n\t"

  67.         SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */

  68.         "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */

  69.         "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */

  70.         "paddw  %%mm7, %%mm7  \n\t"

  71.         "paddw  %%mm5, %%mm5  \n\t"

  72.         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */

  73.         "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */

  74. 

  75.         SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */

  76.         "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */

  77.         "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */

  78.         "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */

  79.         "paddw  %%mm1, %%mm1  \n\t"

  80.         "paddw  %%mm3, %%mm3  \n\t"

  81.         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */

  82.         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */

  83. 

  84.         "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */

  85.         "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */

  86.         "movq   %%mm2, %%mm4  \n\t"

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

  88.         "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */

  89.         "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */

  90.         "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */

  91.         "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */

  92.         "paddw  %%mm2, %%mm2  \n\t"

  93.         "paddw  %%mm0, %%mm0  \n\t"

  94.         "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */

  95.         "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */

  96. 

  97.         "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */

  98.         "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */

  99.         SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */

  100.         "psllw  $3,    %%mm0  \n\t"

  101.         "psllw  $3,    %%mm2  \n\t"

  102.         "paddw  %1,    %%mm0  \n\t" /* add rounding bias */

  103.         "paddw  %1,    %%mm2  \n\t" /* add rounding bias */

  104. 

  105.         SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */

  106.         SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */

  107.         SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */

  108.         SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */

  109.         SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */

  110.         SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */

  111.         :: "r"(block), "m"(bias)

  112.     );

  113. }

  114. 

  115. static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)

  116. {

  117.     int i;

  118.     DECLARE_ALIGNED(8, int16_t, b2)[64];

  119. 

  120.     for(i=0; i<2; i++){

  121.         DECLARE_ALIGNED(8, uint64_t, tmp);

  122. 

  123.         cavs_idct8_1d(block+4*i, ff_pw_4.a);

  124. 

  125.         __asm__ volatile(

  126.             "psraw     $3, %%mm7  \n\t"

  127.             "psraw     $3, %%mm6  \n\t"

  128.             "psraw     $3, %%mm5  \n\t"

  129.             "psraw     $3, %%mm4  \n\t"

  130.             "psraw     $3, %%mm3  \n\t"

  131.             "psraw     $3, %%mm2  \n\t"

  132.             "psraw     $3, %%mm1  \n\t"

  133.             "psraw     $3, %%mm0  \n\t"

  134.             "movq   %%mm7,    %0   \n\t"

  135.             TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )

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

  137.             "movq   %%mm6, 24(%1)  \n\t"

  138.             "movq   %%mm7, 40(%1)  \n\t"

  139.             "movq   %%mm4, 56(%1)  \n\t"

  140.             "movq    %0,    %%mm7  \n\t"

  141.             TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )

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

  143.             "movq   %%mm1, 16(%1)  \n\t"

  144.             "movq   %%mm0, 32(%1)  \n\t"

  145.             "movq   %%mm3, 48(%1)  \n\t"

  146.             : "=m"(tmp)

  147.             : "r"(b2+32*i)

  148.             : "memory"

  149.         );

  150.     }

  151. 

  152.     for(i=0; i<2; i++){

  153.         cavs_idct8_1d(b2+4*i, ff_pw_64.a);

  154. 

  155.         __asm__ volatile(

  156.             "psraw     $7, %%mm7  \n\t"

  157.             "psraw     $7, %%mm6  \n\t"

  158.             "psraw     $7, %%mm5  \n\t"

  159.             "psraw     $7, %%mm4  \n\t"

  160.             "psraw     $7, %%mm3  \n\t"

  161.             "psraw     $7, %%mm2  \n\t"

  162.             "psraw     $7, %%mm1  \n\t"

  163.             "psraw     $7, %%mm0  \n\t"

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

  165.             "movq   %%mm5,  16(%0)  \n\t"

  166.             "movq   %%mm3,  32(%0)  \n\t"

  167.             "movq   %%mm1,  48(%0)  \n\t"

  168.             "movq   %%mm0,  64(%0)  \n\t"

  169.             "movq   %%mm2,  80(%0)  \n\t"

  170.             "movq   %%mm4,  96(%0)  \n\t"

  171.             "movq   %%mm6, 112(%0)  \n\t"

  172.             :: "r"(b2+4*i)

  173.             : "memory"

  174.         );

  175.     }

  176. 

  177.     ff_add_pixels_clamped_mmx(b2, dst, stride);

  178. }

  179. 

  180. /*****************************************************************************

  181.  *

  182.  * motion compensation

  183.  *

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

  185. 

  186. /* vertical filter [-1 -2 96 42 -7  0]  */

  187. #define QPEL_CAVSV1(A,B,C,D,E,F,OP,MUL2) \

  188.         "movd (%0), "#F"            \n\t"\

  189.         "movq "#C", %%mm6           \n\t"\

  190.         "pmullw %5, %%mm6           \n\t"\

  191.         "movq "#D", %%mm7           \n\t"\

  192.         "pmullw "MANGLE(MUL2)", %%mm7\n\t"\

  193.         "psllw $3, "#E"             \n\t"\

  194.         "psubw "#E", %%mm6          \n\t"\

  195.         "psraw $3, "#E"             \n\t"\

  196.         "paddw %%mm7, %%mm6         \n\t"\

  197.         "paddw "#E", %%mm6          \n\t"\

  198.         "paddw "#B", "#B"           \n\t"\

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

  200.         "add %2, %0                 \n\t"\

  201.         "punpcklbw %%mm7, "#F"      \n\t"\

  202.         "psubw "#B", %%mm6          \n\t"\

  203.         "psraw $1, "#B"             \n\t"\

  204.         "psubw "#A", %%mm6          \n\t"\

  205.         "paddw %4, %%mm6            \n\t"\

  206.         "psraw $7, %%mm6            \n\t"\

  207.         "packuswb %%mm6, %%mm6      \n\t"\

  208.         OP(%%mm6, (%1), A, d)            \

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

  210. 

  211. /* vertical filter [ 0 -1  5  5 -1  0]  */

  212. #define QPEL_CAVSV2(A,B,C,D,E,F,OP,MUL2) \

  213.         "movd (%0), "#F"            \n\t"\

  214.         "movq "#C", %%mm6           \n\t"\

  215.         "paddw "#D", %%mm6          \n\t"\

  216.         "pmullw %5, %%mm6           \n\t"\

  217.         "add %2, %0                 \n\t"\

  218.         "punpcklbw %%mm7, "#F"      \n\t"\

  219.         "psubw "#B", %%mm6          \n\t"\

  220.         "psubw "#E", %%mm6          \n\t"\

  221.         "paddw %4, %%mm6            \n\t"\

  222.         "psraw $3, %%mm6            \n\t"\

  223.         "packuswb %%mm6, %%mm6      \n\t"\

  224.         OP(%%mm6, (%1), A, d)            \

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

  226. 

  227. /* vertical filter [ 0 -7 42 96 -2 -1]  */

  228. #define QPEL_CAVSV3(A,B,C,D,E,F,OP,MUL2) \

  229.         "movd (%0), "#F"            \n\t"\

  230.         "movq "#C", %%mm6           \n\t"\

  231.         "pmullw "MANGLE(MUL2)", %%mm6\n\t"\

  232.         "movq "#D", %%mm7           \n\t"\

  233.         "pmullw %5, %%mm7           \n\t"\

  234.         "psllw $3, "#B"             \n\t"\

  235.         "psubw "#B", %%mm6          \n\t"\

  236.         "psraw $3, "#B"             \n\t"\

  237.         "paddw %%mm7, %%mm6         \n\t"\

  238.         "paddw "#B", %%mm6          \n\t"\

  239.         "paddw "#E", "#E"           \n\t"\

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

  241.         "add %2, %0                 \n\t"\

  242.         "punpcklbw %%mm7, "#F"      \n\t"\

  243.         "psubw "#E", %%mm6          \n\t"\

  244.         "psraw $1, "#E"             \n\t"\

  245.         "psubw "#F", %%mm6          \n\t"\

  246.         "paddw %4, %%mm6            \n\t"\

  247.         "psraw $7, %%mm6            \n\t"\

  248.         "packuswb %%mm6, %%mm6      \n\t"\

  249.         OP(%%mm6, (%1), A, d)            \

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

  251. 

  252. 

  253. #define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\

  254.     int w= 2;\

  255.     src -= 2*srcStride;\

  256.     \

  257.     while(w--){\

  258.       __asm__ volatile(\

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

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

  261.         "add %2, %0                 \n\t"\

  262.         "movd (%0), %%mm1           \n\t"\

  263.         "add %2, %0                 \n\t"\

  264.         "movd (%0), %%mm2           \n\t"\

  265.         "add %2, %0                 \n\t"\

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

  267.         "add %2, %0                 \n\t"\

  268.         "movd (%0), %%mm4           \n\t"\

  269.         "add %2, %0                 \n\t"\

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

  271.         "punpcklbw %%mm7, %%mm1     \n\t"\

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

  273.         "punpcklbw %%mm7, %%mm3     \n\t"\

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

  275.         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\

  276.         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\

  277.         VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\

  278.         VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\

  279.         VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\

  280.         VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\

  281.         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\

  282.         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\

  283.         \

  284.         : "+a"(src), "+c"(dst)\

  285.         : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD), "m"(MUL1)\

  286.         : "memory"\

  287.      );\

  288.      if(h==16){\

  289.         __asm__ volatile(\

  290.             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\

  291.             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\

  292.             VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\

  293.             VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\

  294.             VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\

  295.             VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\

  296.             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\

  297.             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\

  298.             \

  299.            : "+a"(src), "+c"(dst)\

  300.            : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD),  "m"(MUL1)\

  301.            : "memory"\

  302.         );\

  303.      }\

  304.      src += 4-(h+5)*srcStride;\

  305.      dst += 4-h*dstStride;\

  306.    }

  307. 

  308. #define QPEL_CAVS(OPNAME, OP, MMX)\

  309. static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  310.     int h=8;\

  311.     __asm__ volatile(\

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

  313.         "movq %5, %%mm6             \n\t"\

  314.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

  323.         "paddw %%mm2, %%mm0         \n\t"\

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

  325.         "pmullw %%mm6, %%mm0        \n\t"\

  326.         "pmullw %%mm6, %%mm1        \n\t"\

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

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

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

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

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

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

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

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

  335.         "paddw %%mm4, %%mm2         \n\t"\

  336.         "paddw %%mm3, %%mm5         \n\t"\

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

  338.         "psubw %%mm5, %%mm1         \n\t"\

  339.         "movq %6, %%mm5             \n\t"\

  340.         "paddw %%mm5, %%mm0         \n\t"\

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

  342.         "psraw $3, %%mm0            \n\t"\

  343.         "psraw $3, %%mm1            \n\t"\

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

  345.         OP(%%mm0, (%1),%%mm5, q)         \

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

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

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

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

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

  351.         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\

  352.         : "memory"\

  353.     );\

  354. }\

  355. \

  356. static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  357.   QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \

  358. }\

  359. \

  360. static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  361.   QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \

  362. }\

  363. \

  364. static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  365.   QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \

  366. }\

  367. \

  368. static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  369.     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  370. }\

  371. static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  372.     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  373.     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  374. }\

  375. \

  376. static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  377.     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  378. }\

  379. static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  380.     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  381.     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  382. }\

  383. \

  384. static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  385.     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  386. }\

  387. static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  388.     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  389.     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  390. }\

  391. \

  392. static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  393.     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\

  394.     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  395.     src += 8*srcStride;\

  396.     dst += 8*dstStride;\

  397.     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\

  398.     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  399. }\

  400. 

  401. #define CAVS_MC(OPNAME, SIZE, MMX) \

  402. static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  403.     OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\

  404. }\

  405. \

  406. static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  407.     OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\

  408. }\

  409. \

  410. static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  411.     OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\

  412. }\

  413. \

  414. static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  415.     OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\

  416. }\

  417. 

  418. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"

  419. #define AVG_3DNOW_OP(a,b,temp, size) \

  420. "mov" #size " " #b ", " #temp "   \n\t"\

  421. "pavgusb " #temp ", " #a "        \n\t"\

  422. "mov" #size " " #a ", " #b "      \n\t"

  423. #define AVG_MMX2_OP(a,b,temp, size) \

  424. "mov" #size " " #b ", " #temp "   \n\t"\

  425. "pavgb " #temp ", " #a "          \n\t"\

  426. "mov" #size " " #a ", " #b "      \n\t"

  427. 

  428. QPEL_CAVS(put_,       PUT_OP, 3dnow)

  429. QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)

  430. QPEL_CAVS(put_,       PUT_OP, mmx2)

  431. QPEL_CAVS(avg_,  AVG_MMX2_OP, mmx2)

  432. 

  433. CAVS_MC(put_, 8, 3dnow)

  434. CAVS_MC(put_, 16,3dnow)

  435. CAVS_MC(avg_, 8, 3dnow)

  436. CAVS_MC(avg_, 16,3dnow)

  437. CAVS_MC(put_, 8, mmx2)

  438. CAVS_MC(put_, 16,mmx2)

  439. CAVS_MC(avg_, 8, mmx2)

  440. CAVS_MC(avg_, 16,mmx2)

  441. 

  442. static void ff_cavsdsp_init_mmx2(CAVSDSPContext* c, AVCodecContext *avctx) {

  443. #define dspfunc(PFX, IDX, NUM) \

  444.     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \

  445.     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \

  446.     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \

  447.     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \

  448.     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \

  449. 

  450.     dspfunc(put_cavs_qpel, 0, 16);

  451.     dspfunc(put_cavs_qpel, 1, 8);

  452.     dspfunc(avg_cavs_qpel, 0, 16);

  453.     dspfunc(avg_cavs_qpel, 1, 8);

  454. #undef dspfunc

  455.     c->cavs_idct8_add = cavs_idct8_add_mmx;

  456. }

  457. 

  458. static void ff_cavsdsp_init_3dnow(CAVSDSPContext* c, AVCodecContext *avctx) {

  459. #define dspfunc(PFX, IDX, NUM) \

  460.     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \

  461.     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \

  462.     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \

  463.     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \

  464.     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \

  465. 

  466.     dspfunc(put_cavs_qpel, 0, 16);

  467.     dspfunc(put_cavs_qpel, 1, 8);

  468.     dspfunc(avg_cavs_qpel, 0, 16);

  469.     dspfunc(avg_cavs_qpel, 1, 8);

  470. #undef dspfunc

  471.     c->cavs_idct8_add = cavs_idct8_add_mmx;

  472. }

  473. 

  474. void ff_cavsdsp_init_mmx(CAVSDSPContext *c, AVCodecContext *avctx)

  475. {

  476.     int mm_flags = av_get_cpu_flags();

  477. 

  478.     if (mm_flags & AV_CPU_FLAG_MMX2)  ff_cavsdsp_init_mmx2 (c, avctx);

  479.     if (mm_flags & AV_CPU_FLAG_3DNOW) ff_cavsdsp_init_3dnow(c, avctx);

  480. }