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.
37302 Commits
32 Branches
325MB
Tree: 71469f3b63
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '71469f3b63'
${ noResults }
FFmpeg/libavcodec/x86/cavsdsp.c

548 lines
21KB
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 Libav.

  9.  *

  10.  * Libav 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.  * Libav 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 Libav; 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/attributes.h"

  26. #include "libavutil/common.h"

  27. #include "libavutil/cpu.h"

  28. #include "libavutil/x86/asm.h"

  29. #include "libavutil/x86/cpu.h"

  30. #include "libavcodec/cavsdsp.h"

  31. #include "constants.h"

  32. #include "dsputil_mmx.h"

  33. #include "config.h"

  34. 

  35. #if HAVE_MMX_INLINE

  36. 

  37. /* in/out: mma=mma+mmb, mmb=mmb-mma */

  38. #define SUMSUB_BA( a, b ) \

  39.     "paddw "#b", "#a" \n\t"\

  40.     "paddw "#b", "#b" \n\t"\

  41.     "psubw "#a", "#b" \n\t"

  42. 

  43. /*****************************************************************************

  44.  *

  45.  * inverse transform

  46.  *

  47.  ****************************************************************************/

  48. 

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

  50. {

  51.     __asm__ volatile(

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

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

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

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

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

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

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

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

  60. 

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

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

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

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

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

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

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

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

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

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

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

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

  73. 

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

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

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

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

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

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

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

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

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

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

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

  85. 

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

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

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

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

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

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

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

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

  94. 

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

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

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

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

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

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

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

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

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

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

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

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

  107. 

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

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

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

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

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

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

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

  115. 

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

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

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

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

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

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

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

  123.     );

  124. }

  125. 

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

  127. {

  128.     int i;

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

  130. 

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

  132.         DECLARE_ALIGNED(8, uint64_t, tmp);

  133. 

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

  135. 

  136.         __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  157.             : "=m"(tmp)

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

  159.             : "memory"

  160.         );

  161.     }

  162. 

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

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

  165. 

  166.         __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  184.             : "memory"

  185.         );

  186.     }

  187. 

  188.     ff_add_pixels_clamped_mmx(b2, dst, stride);

  189. }

  190. 

  191. #endif /* HAVE_MMX_INLINE */

  192. 

  193. #if (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE)

  194. 

  195. /*****************************************************************************

  196.  *

  197.  * motion compensation

  198.  *

  199.  ****************************************************************************/

  200. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  225. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  241. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  266. 

  267. 

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

  269.     int w= 2;\

  270.     src -= 2*srcStride;\

  271.     \

  272.     while(w--){\

  273.       __asm__ volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  297.         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 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.      if(h==16){\

  304.         __asm__ volatile(\

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

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

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

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

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

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

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

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

  313.             \

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

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

  316.            : "memory"\

  317.         );\

  318.      }\

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

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

  321.    }

  322. 

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

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

  325.     int h=8;\

  326.     __asm__ volatile(\

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

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

  329.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  367.         : "memory"\

  368.     );\

  369. }\

  370. \

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

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

  373. }\

  374. \

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

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

  377. }\

  378. \

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

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

  381. }\

  382. \

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

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

  385. }\

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

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

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

  389. }\

  390. \

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

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

  393. }\

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

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

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

  397. }\

  398. \

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

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

  401. }\

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

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

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

  405. }\

  406. \

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

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

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

  410.     src += 8*srcStride;\

  411.     dst += 8*dstStride;\

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

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

  414. }\

  415. 

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

  417. static void OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\

  418. {\

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

  420. }\

  421. \

  422. static void OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\

  423. {\

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

  425. }\

  426. \

  427. static void OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\

  428. {\

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

  430. }\

  431. \

  432. static void OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\

  433. {\

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

  435. }\

  436. 

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

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

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

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

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

  442. #define AVG_MMXEXT_OP(a, b, temp, size) \

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

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

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

  446. 

  447. #endif /* (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE) */

  448. 

  449. #if HAVE_MMX_INLINE

  450. static void put_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src,

  451.                                     ptrdiff_t stride)

  452. {

  453.     ff_put_pixels8_mmx(dst, src, stride, 8);

  454. }

  455. 

  456. static void avg_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src,

  457.                                     ptrdiff_t stride)

  458. {

  459.     ff_avg_pixels8_mmx(dst, src, stride, 8);

  460. }

  461. 

  462. static void put_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src,

  463.                                      ptrdiff_t stride)

  464. {

  465.     ff_put_pixels16_mmx(dst, src, stride, 16);

  466. }

  467. 

  468. static void avg_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src,

  469.                                      ptrdiff_t stride)

  470. {

  471.     ff_avg_pixels16_mmx(dst, src, stride, 16);

  472. }

  473. 

  474. static av_cold void cavsdsp_init_mmx(CAVSDSPContext *c,

  475.                                      AVCodecContext *avctx)

  476. {

  477.     c->put_cavs_qpel_pixels_tab[0][0] = put_cavs_qpel16_mc00_mmx;

  478.     c->put_cavs_qpel_pixels_tab[1][0] = put_cavs_qpel8_mc00_mmx;

  479.     c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_mmx;

  480.     c->avg_cavs_qpel_pixels_tab[1][0] = avg_cavs_qpel8_mc00_mmx;

  481. 

  482.     c->cavs_idct8_add = cavs_idct8_add_mmx;

  483.     c->idct_perm      = FF_TRANSPOSE_IDCT_PERM;

  484. }

  485. #endif /* HAVE_MMX_INLINE */

  486. 

  487. #define DSPFUNC(PFX, IDX, NUM, EXT)                                                       \

  488.     c->PFX ## _cavs_qpel_pixels_tab[IDX][ 2] = PFX ## _cavs_qpel ## NUM ## _mc20_ ## EXT; \

  489.     c->PFX ## _cavs_qpel_pixels_tab[IDX][ 4] = PFX ## _cavs_qpel ## NUM ## _mc01_ ## EXT; \

  490.     c->PFX ## _cavs_qpel_pixels_tab[IDX][ 8] = PFX ## _cavs_qpel ## NUM ## _mc02_ ## EXT; \

  491.     c->PFX ## _cavs_qpel_pixels_tab[IDX][12] = PFX ## _cavs_qpel ## NUM ## _mc03_ ## EXT; \

  492. 

  493. #if HAVE_MMXEXT_INLINE

  494. QPEL_CAVS(put_,        PUT_OP, mmxext)

  495. QPEL_CAVS(avg_, AVG_MMXEXT_OP, mmxext)

  496. 

  497. CAVS_MC(put_,  8, mmxext)

  498. CAVS_MC(put_, 16, mmxext)

  499. CAVS_MC(avg_,  8, mmxext)

  500. CAVS_MC(avg_, 16, mmxext)

  501. 

  502. static av_cold void cavsdsp_init_mmxext(CAVSDSPContext *c,

  503.                                         AVCodecContext *avctx)

  504. {

  505.     DSPFUNC(put, 0, 16, mmxext);

  506.     DSPFUNC(put, 1,  8, mmxext);

  507.     DSPFUNC(avg, 0, 16, mmxext);

  508.     DSPFUNC(avg, 1,  8, mmxext);

  509. }

  510. #endif /* HAVE_MMXEXT_INLINE */

  511. 

  512. #if HAVE_AMD3DNOW_INLINE

  513. QPEL_CAVS(put_,       PUT_OP, 3dnow)

  514. QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)

  515. 

  516. CAVS_MC(put_, 8, 3dnow)

  517. CAVS_MC(put_, 16,3dnow)

  518. CAVS_MC(avg_, 8, 3dnow)

  519. CAVS_MC(avg_, 16,3dnow)

  520. 

  521. static av_cold void cavsdsp_init_3dnow(CAVSDSPContext *c,

  522.                                        AVCodecContext *avctx)

  523. {

  524.     DSPFUNC(put, 0, 16, 3dnow);

  525.     DSPFUNC(put, 1,  8, 3dnow);

  526.     DSPFUNC(avg, 0, 16, 3dnow);

  527.     DSPFUNC(avg, 1,  8, 3dnow);

  528. }

  529. #endif /* HAVE_AMD3DNOW_INLINE */

  530. 

  531. av_cold void ff_cavsdsp_init_x86(CAVSDSPContext *c, AVCodecContext *avctx)

  532. {

  533. #if HAVE_MMX_INLINE

  534.     int mm_flags = av_get_cpu_flags();

  535. 

  536.     if (mm_flags & AV_CPU_FLAG_MMX)

  537.         cavsdsp_init_mmx(c, avctx);

  538. #endif /* HAVE_MMX_INLINE */

  539. #if HAVE_MMXEXT_INLINE

  540.     if (mm_flags & AV_CPU_FLAG_MMXEXT)

  541.         cavsdsp_init_mmxext(c, avctx);

  542. #endif /* HAVE_MMXEXT_INLINE */

  543. #if HAVE_AMD3DNOW_INLINE

  544.     if (mm_flags & AV_CPU_FLAG_3DNOW)

  545.         cavsdsp_init_3dnow(c, avctx);

  546. #endif /* HAVE_AMD3DNOW_INLINE */

  547. }