FFmpeg: libavcodec/x86/cavsdsp.c Source File

00001 /*
00002  * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
00003  * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>
00004  *
00005  * MMX-optimized DSP functions, based on H.264 optimizations by
00006  * Michael Niedermayer and Loren Merritt
00007  *
00008  * This file is part of FFmpeg.
00009  *
00010  * FFmpeg is free software; you can redistribute it and/or
00011  * modify it under the terms of the GNU Lesser General Public
00012  * License as published by the Free Software Foundation; either
00013  * version 2.1 of the License, or (at your option) any later version.
00014  *
00015  * FFmpeg is distributed in the hope that it will be useful,
00016  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00017  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00018  * Lesser General Public License for more details.
00019  *
00020  * You should have received a copy of the GNU Lesser General Public
00021  * License along with FFmpeg; if not, write to the Free Software
00022  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00023  */
00024 
00025 #include "libavutil/common.h"
00026 #include "libavutil/cpu.h"
00027 #include "libavutil/x86/asm.h"
00028 #include "libavutil/x86/cpu.h"
00029 #include "libavcodec/dsputil.h"
00030 #include "libavcodec/cavsdsp.h"
00031 #include "dsputil_mmx.h"
00032 #include "config.h"
00033 
00034 #if (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE)
00035 
00036 /* in/out: mma=mma+mmb, mmb=mmb-mma */
00037 #define SUMSUB_BA( a, b ) \
00038     "paddw "#b", "#a" \n\t"\
00039     "paddw "#b", "#b" \n\t"\
00040     "psubw "#a", "#b" \n\t"
00041 
00042 /*****************************************************************************
00043  *
00044  * inverse transform
00045  *
00046  ****************************************************************************/
00047 
00048 static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)
00049 {
00050     __asm__ volatile(
00051         "movq 112(%0), %%mm4  \n\t" /* mm4 = src7 */
00052         "movq  16(%0), %%mm5  \n\t" /* mm5 = src1 */
00053         "movq  80(%0), %%mm2  \n\t" /* mm2 = src5 */
00054         "movq  48(%0), %%mm7  \n\t" /* mm7 = src3 */
00055         "movq   %%mm4, %%mm0  \n\t"
00056         "movq   %%mm5, %%mm3  \n\t"
00057         "movq   %%mm2, %%mm6  \n\t"
00058         "movq   %%mm7, %%mm1  \n\t"
00059 
00060         "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */
00061         "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */
00062         "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */
00063         "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */
00064         "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */
00065         "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */
00066         "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */
00067         "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */
00068         "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */
00069         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */
00070         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */
00071         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */
00072 
00073         "movq   %%mm5, %%mm4  \n\t"
00074         "movq   %%mm7, %%mm6  \n\t"
00075         "movq   %%mm3, %%mm0  \n\t"
00076         "movq   %%mm1, %%mm2  \n\t"
00077         SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */
00078         "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */
00079         "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */
00080         "paddw  %%mm7, %%mm7  \n\t"
00081         "paddw  %%mm5, %%mm5  \n\t"
00082         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */
00083         "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */
00084 
00085         SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */
00086         "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */
00087         "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */
00088         "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */
00089         "paddw  %%mm1, %%mm1  \n\t"
00090         "paddw  %%mm3, %%mm3  \n\t"
00091         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */
00092         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */
00093 
00094         "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */
00095         "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */
00096         "movq   %%mm2, %%mm4  \n\t"
00097         "movq   %%mm6, %%mm0  \n\t"
00098         "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */
00099         "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */
00100         "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */
00101         "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */
00102         "paddw  %%mm2, %%mm2  \n\t"
00103         "paddw  %%mm0, %%mm0  \n\t"
00104         "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */
00105         "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */
00106 
00107         "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */
00108         "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */
00109         SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */
00110         "psllw  $3,    %%mm0  \n\t"
00111         "psllw  $3,    %%mm2  \n\t"
00112         "paddw  %1,    %%mm0  \n\t" /* add rounding bias */
00113         "paddw  %1,    %%mm2  \n\t" /* add rounding bias */
00114 
00115         SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */
00116         SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */
00117         SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */
00118         SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */
00119         SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */
00120         SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */
00121         :: "r"(block), "m"(bias)
00122     );
00123 }
00124 
00125 static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
00126 {
00127     int i;
00128     DECLARE_ALIGNED(8, int16_t, b2)[64];
00129 
00130     for(i=0; i<2; i++){
00131         DECLARE_ALIGNED(8, uint64_t, tmp);
00132 
00133         cavs_idct8_1d(block+4*i, ff_pw_4.a);
00134 
00135         __asm__ volatile(
00136             "psraw     $3, %%mm7  \n\t"
00137             "psraw     $3, %%mm6  \n\t"
00138             "psraw     $3, %%mm5  \n\t"
00139             "psraw     $3, %%mm4  \n\t"
00140             "psraw     $3, %%mm3  \n\t"
00141             "psraw     $3, %%mm2  \n\t"
00142             "psraw     $3, %%mm1  \n\t"
00143             "psraw     $3, %%mm0  \n\t"
00144             "movq   %%mm7,    %0   \n\t"
00145             TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
00146             "movq   %%mm0,  8(%1)  \n\t"
00147             "movq   %%mm6, 24(%1)  \n\t"
00148             "movq   %%mm7, 40(%1)  \n\t"
00149             "movq   %%mm4, 56(%1)  \n\t"
00150             "movq    %0,    %%mm7  \n\t"
00151             TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
00152             "movq   %%mm7,   (%1)  \n\t"
00153             "movq   %%mm1, 16(%1)  \n\t"
00154             "movq   %%mm0, 32(%1)  \n\t"
00155             "movq   %%mm3, 48(%1)  \n\t"
00156             : "=m"(tmp)
00157             : "r"(b2+32*i)
00158             : "memory"
00159         );
00160     }
00161 
00162     for(i=0; i<2; i++){
00163         cavs_idct8_1d(b2+4*i, ff_pw_64.a);
00164 
00165         __asm__ volatile(
00166             "psraw     $7, %%mm7  \n\t"
00167             "psraw     $7, %%mm6  \n\t"
00168             "psraw     $7, %%mm5  \n\t"
00169             "psraw     $7, %%mm4  \n\t"
00170             "psraw     $7, %%mm3  \n\t"
00171             "psraw     $7, %%mm2  \n\t"
00172             "psraw     $7, %%mm1  \n\t"
00173             "psraw     $7, %%mm0  \n\t"
00174             "movq   %%mm7,    (%0)  \n\t"
00175             "movq   %%mm5,  16(%0)  \n\t"
00176             "movq   %%mm3,  32(%0)  \n\t"
00177             "movq   %%mm1,  48(%0)  \n\t"
00178             "movq   %%mm0,  64(%0)  \n\t"
00179             "movq   %%mm2,  80(%0)  \n\t"
00180             "movq   %%mm4,  96(%0)  \n\t"
00181             "movq   %%mm6, 112(%0)  \n\t"
00182             :: "r"(b2+4*i)
00183             : "memory"
00184         );
00185     }
00186 
00187     ff_add_pixels_clamped_mmx(b2, dst, stride);
00188 }
00189 
00190 /*****************************************************************************
00191  *
00192  * motion compensation
00193  *
00194  ****************************************************************************/
00195 
00196 /* vertical filter [-1 -2 96 42 -7  0]  */
00197 #define QPEL_CAVSV1(A,B,C,D,E,F,OP,MUL2) \
00198         "movd (%0), "#F"            \n\t"\
00199         "movq "#C", %%mm6           \n\t"\
00200         "pmullw %5, %%mm6           \n\t"\
00201         "movq "#D", %%mm7           \n\t"\
00202         "pmullw "MANGLE(MUL2)", %%mm7\n\t"\
00203         "psllw $3, "#E"             \n\t"\
00204         "psubw "#E", %%mm6          \n\t"\
00205         "psraw $3, "#E"             \n\t"\
00206         "paddw %%mm7, %%mm6         \n\t"\
00207         "paddw "#E", %%mm6          \n\t"\
00208         "paddw "#B", "#B"           \n\t"\
00209         "pxor %%mm7, %%mm7          \n\t"\
00210         "add %2, %0                 \n\t"\
00211         "punpcklbw %%mm7, "#F"      \n\t"\
00212         "psubw "#B", %%mm6          \n\t"\
00213         "psraw $1, "#B"             \n\t"\
00214         "psubw "#A", %%mm6          \n\t"\
00215         "paddw %4, %%mm6            \n\t"\
00216         "psraw $7, %%mm6            \n\t"\
00217         "packuswb %%mm6, %%mm6      \n\t"\
00218         OP(%%mm6, (%1), A, d)            \
00219         "add %3, %1                 \n\t"
00220 
00221 /* vertical filter [ 0 -1  5  5 -1  0]  */
00222 #define QPEL_CAVSV2(A,B,C,D,E,F,OP,MUL2) \
00223         "movd (%0), "#F"            \n\t"\
00224         "movq "#C", %%mm6           \n\t"\
00225         "paddw "#D", %%mm6          \n\t"\
00226         "pmullw %5, %%mm6           \n\t"\
00227         "add %2, %0                 \n\t"\
00228         "punpcklbw %%mm7, "#F"      \n\t"\
00229         "psubw "#B", %%mm6          \n\t"\
00230         "psubw "#E", %%mm6          \n\t"\
00231         "paddw %4, %%mm6            \n\t"\
00232         "psraw $3, %%mm6            \n\t"\
00233         "packuswb %%mm6, %%mm6      \n\t"\
00234         OP(%%mm6, (%1), A, d)            \
00235         "add %3, %1                 \n\t"
00236 
00237 /* vertical filter [ 0 -7 42 96 -2 -1]  */
00238 #define QPEL_CAVSV3(A,B,C,D,E,F,OP,MUL2) \
00239         "movd (%0), "#F"            \n\t"\
00240         "movq "#C", %%mm6           \n\t"\
00241         "pmullw "MANGLE(MUL2)", %%mm6\n\t"\
00242         "movq "#D", %%mm7           \n\t"\
00243         "pmullw %5, %%mm7           \n\t"\
00244         "psllw $3, "#B"             \n\t"\
00245         "psubw "#B", %%mm6          \n\t"\
00246         "psraw $3, "#B"             \n\t"\
00247         "paddw %%mm7, %%mm6         \n\t"\
00248         "paddw "#B", %%mm6          \n\t"\
00249         "paddw "#E", "#E"           \n\t"\
00250         "pxor %%mm7, %%mm7          \n\t"\
00251         "add %2, %0                 \n\t"\
00252         "punpcklbw %%mm7, "#F"      \n\t"\
00253         "psubw "#E", %%mm6          \n\t"\
00254         "psraw $1, "#E"             \n\t"\
00255         "psubw "#F", %%mm6          \n\t"\
00256         "paddw %4, %%mm6            \n\t"\
00257         "psraw $7, %%mm6            \n\t"\
00258         "packuswb %%mm6, %%mm6      \n\t"\
00259         OP(%%mm6, (%1), A, d)            \
00260         "add %3, %1                 \n\t"
00261 
00262 
00263 #define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
00264     int w= 2;\
00265     src -= 2*srcStride;\
00266     \
00267     while(w--){\
00268       __asm__ volatile(\
00269         "pxor %%mm7, %%mm7          \n\t"\
00270         "movd (%0), %%mm0           \n\t"\
00271         "add %2, %0                 \n\t"\
00272         "movd (%0), %%mm1           \n\t"\
00273         "add %2, %0                 \n\t"\
00274         "movd (%0), %%mm2           \n\t"\
00275         "add %2, %0                 \n\t"\
00276         "movd (%0), %%mm3           \n\t"\
00277         "add %2, %0                 \n\t"\
00278         "movd (%0), %%mm4           \n\t"\
00279         "add %2, %0                 \n\t"\
00280         "punpcklbw %%mm7, %%mm0     \n\t"\
00281         "punpcklbw %%mm7, %%mm1     \n\t"\
00282         "punpcklbw %%mm7, %%mm2     \n\t"\
00283         "punpcklbw %%mm7, %%mm3     \n\t"\
00284         "punpcklbw %%mm7, %%mm4     \n\t"\
00285         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00286         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00287         VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00288         VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00289         VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
00290         VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
00291         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00292         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00293         \
00294         : "+a"(src), "+c"(dst)\
00295         : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD), "m"(MUL1)\
00296         : "memory"\
00297      );\
00298      if(h==16){\
00299         __asm__ volatile(\
00300             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00301             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00302             VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
00303             VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
00304             VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00305             VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00306             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00307             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00308             \
00309            : "+a"(src), "+c"(dst)\
00310            : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD),  "m"(MUL1)\
00311            : "memory"\
00312         );\
00313      }\
00314      src += 4-(h+5)*srcStride;\
00315      dst += 4-h*dstStride;\
00316    }
00317 
00318 #define QPEL_CAVS(OPNAME, OP, MMX)\
00319 static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00320     int h=8;\
00321     __asm__ volatile(\
00322         "pxor %%mm7, %%mm7          \n\t"\
00323         "movq %5, %%mm6             \n\t"\
00324         "1:                         \n\t"\
00325         "movq    (%0), %%mm0        \n\t"\
00326         "movq   1(%0), %%mm2        \n\t"\
00327         "movq %%mm0, %%mm1          \n\t"\
00328         "movq %%mm2, %%mm3          \n\t"\
00329         "punpcklbw %%mm7, %%mm0     \n\t"\
00330         "punpckhbw %%mm7, %%mm1     \n\t"\
00331         "punpcklbw %%mm7, %%mm2     \n\t"\
00332         "punpckhbw %%mm7, %%mm3     \n\t"\
00333         "paddw %%mm2, %%mm0         \n\t"\
00334         "paddw %%mm3, %%mm1         \n\t"\
00335         "pmullw %%mm6, %%mm0        \n\t"\
00336         "pmullw %%mm6, %%mm1        \n\t"\
00337         "movq   -1(%0), %%mm2       \n\t"\
00338         "movq    2(%0), %%mm4       \n\t"\
00339         "movq %%mm2, %%mm3          \n\t"\
00340         "movq %%mm4, %%mm5          \n\t"\
00341         "punpcklbw %%mm7, %%mm2     \n\t"\
00342         "punpckhbw %%mm7, %%mm3     \n\t"\
00343         "punpcklbw %%mm7, %%mm4     \n\t"\
00344         "punpckhbw %%mm7, %%mm5     \n\t"\
00345         "paddw %%mm4, %%mm2         \n\t"\
00346         "paddw %%mm3, %%mm5         \n\t"\
00347         "psubw %%mm2, %%mm0         \n\t"\
00348         "psubw %%mm5, %%mm1         \n\t"\
00349         "movq %6, %%mm5             \n\t"\
00350         "paddw %%mm5, %%mm0         \n\t"\
00351         "paddw %%mm5, %%mm1         \n\t"\
00352         "psraw $3, %%mm0            \n\t"\
00353         "psraw $3, %%mm1            \n\t"\
00354         "packuswb %%mm1, %%mm0      \n\t"\
00355         OP(%%mm0, (%1),%%mm5, q)         \
00356         "add %3, %0                 \n\t"\
00357         "add %4, %1                 \n\t"\
00358         "decl %2                    \n\t"\
00359         " jnz 1b                    \n\t"\
00360         : "+a"(src), "+c"(dst), "+m"(h)\
00361         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\
00362         : "memory"\
00363     );\
00364 }\
00365 \
00366 static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00367   QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00368 }\
00369 \
00370 static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00371   QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \
00372 }\
00373 \
00374 static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00375   QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00376 }\
00377 \
00378 static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00379     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00380 }\
00381 static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00382     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00383     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00384 }\
00385 \
00386 static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00387     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00388 }\
00389 static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00390     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00391     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00392 }\
00393 \
00394 static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00395     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00396 }\
00397 static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00398     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00399     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00400 }\
00401 \
00402 static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00403     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00404     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00405     src += 8*srcStride;\
00406     dst += 8*dstStride;\
00407     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00408     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00409 }\
00410 
00411 #define CAVS_MC(OPNAME, SIZE, MMX) \
00412 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00413     OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
00414 }\
00415 \
00416 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00417     OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
00418 }\
00419 \
00420 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00421     OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
00422 }\
00423 \
00424 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00425     OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
00426 }\
00427 
00428 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"
00429 #define AVG_3DNOW_OP(a,b,temp, size) \
00430 "mov" #size " " #b ", " #temp "   \n\t"\
00431 "pavgusb " #temp ", " #a "        \n\t"\
00432 "mov" #size " " #a ", " #b "      \n\t"
00433 #define AVG_MMXEXT_OP(a, b, temp, size) \
00434 "mov" #size " " #b ", " #temp "   \n\t"\
00435 "pavgb " #temp ", " #a "          \n\t"\
00436 "mov" #size " " #a ", " #b "      \n\t"
00437 
00438 #endif /* (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE) */
00439 
00440 #if HAVE_MMXEXT_INLINE
00441 QPEL_CAVS(put_,        PUT_OP, mmxext)
00442 QPEL_CAVS(avg_, AVG_MMXEXT_OP, mmxext)
00443 
00444 CAVS_MC(put_,  8, mmxext)
00445 CAVS_MC(put_, 16, mmxext)
00446 CAVS_MC(avg_,  8, mmxext)
00447 CAVS_MC(avg_, 16, mmxext)
00448 
00449 static void ff_cavsdsp_init_mmxext(CAVSDSPContext *c, AVCodecContext *avctx)
00450 {
00451 #define dspfunc(PFX, IDX, NUM) \
00452     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmxext; \
00453     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmxext; \
00454     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmxext; \
00455     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmxext; \
00456     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmxext; \
00457 
00458     dspfunc(put_cavs_qpel, 0, 16);
00459     dspfunc(put_cavs_qpel, 1, 8);
00460     dspfunc(avg_cavs_qpel, 0, 16);
00461     dspfunc(avg_cavs_qpel, 1, 8);
00462 #undef dspfunc
00463     c->cavs_idct8_add = cavs_idct8_add_mmx;
00464     c->idct_perm = FF_TRANSPOSE_IDCT_PERM;
00465 }
00466 #endif /* HAVE_MMXEXT_INLINE */
00467 
00468 #if HAVE_AMD3DNOW_INLINE
00469 QPEL_CAVS(put_,       PUT_OP, 3dnow)
00470 QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
00471 
00472 CAVS_MC(put_, 8, 3dnow)
00473 CAVS_MC(put_, 16,3dnow)
00474 CAVS_MC(avg_, 8, 3dnow)
00475 CAVS_MC(avg_, 16,3dnow)
00476 
00477 static void ff_cavsdsp_init_3dnow(CAVSDSPContext* c, AVCodecContext *avctx) {
00478 #define dspfunc(PFX, IDX, NUM) \
00479     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmxext; \
00480     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
00481     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
00482     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
00483     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \
00484 
00485     dspfunc(put_cavs_qpel, 0, 16);
00486     dspfunc(put_cavs_qpel, 1, 8);
00487     dspfunc(avg_cavs_qpel, 0, 16);
00488     dspfunc(avg_cavs_qpel, 1, 8);
00489 #undef dspfunc
00490     c->cavs_idct8_add = cavs_idct8_add_mmx;
00491     c->idct_perm = FF_TRANSPOSE_IDCT_PERM;
00492 }
00493 #endif /* HAVE_AMD3DNOW_INLINE */
00494 
00495 av_cold void ff_cavsdsp_init_x86(CAVSDSPContext *c, AVCodecContext *avctx)
00496 {
00497     int mm_flags = av_get_cpu_flags();
00498 
00499 #if HAVE_MMXEXT_INLINE
00500     if (mm_flags & AV_CPU_FLAG_MMXEXT) ff_cavsdsp_init_mmxext(c, avctx);
00501 #endif /* HAVE_MMXEXT_INLINE */
00502 #if HAVE_AMD3DNOW_INLINE
00503     if (mm_flags & AV_CPU_FLAG_3DNOW) ff_cavsdsp_init_3dnow(c, avctx);
00504 #endif /* HAVE_AMD3DNOW_INLINE */
00505 }