FFmpeg: libavcodec/x86/cavsdsp

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/x86_cpu.h"
00027 #include "libavcodec/dsputil.h"
00028 #include "dsputil_mmx.h"
00029 
00030 /*****************************************************************************
00031  *
00032  * inverse transform
00033  *
00034  ****************************************************************************/
00035 
00036 static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)
00037 {
00038     __asm__ volatile(
00039         "movq 112(%0), %%mm4  \n\t" /* mm4 = src7 */
00040         "movq  16(%0), %%mm5  \n\t" /* mm5 = src1 */
00041         "movq  80(%0), %%mm2  \n\t" /* mm2 = src5 */
00042         "movq  48(%0), %%mm7  \n\t" /* mm7 = src3 */
00043         "movq   %%mm4, %%mm0  \n\t"
00044         "movq   %%mm5, %%mm3  \n\t"
00045         "movq   %%mm2, %%mm6  \n\t"
00046         "movq   %%mm7, %%mm1  \n\t"
00047 
00048         "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */
00049         "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */
00050         "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */
00051         "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */
00052         "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */
00053         "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */
00054         "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */
00055         "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */
00056         "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */
00057         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */
00058         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */
00059         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */
00060 
00061         "movq   %%mm5, %%mm4  \n\t"
00062         "movq   %%mm7, %%mm6  \n\t"
00063         "movq   %%mm3, %%mm0  \n\t"
00064         "movq   %%mm1, %%mm2  \n\t"
00065         SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */
00066         "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */
00067         "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */
00068         "paddw  %%mm7, %%mm7  \n\t"
00069         "paddw  %%mm5, %%mm5  \n\t"
00070         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */
00071         "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */
00072 
00073         SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */
00074         "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */
00075         "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */
00076         "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */
00077         "paddw  %%mm1, %%mm1  \n\t"
00078         "paddw  %%mm3, %%mm3  \n\t"
00079         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */
00080         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */
00081 
00082         "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */
00083         "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */
00084         "movq   %%mm2, %%mm4  \n\t"
00085         "movq   %%mm6, %%mm0  \n\t"
00086         "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */
00087         "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */
00088         "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */
00089         "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */
00090         "paddw  %%mm2, %%mm2  \n\t"
00091         "paddw  %%mm0, %%mm0  \n\t"
00092         "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */
00093         "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */
00094 
00095         "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */
00096         "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */
00097         SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */
00098         "psllw  $3,    %%mm0  \n\t"
00099         "psllw  $3,    %%mm2  \n\t"
00100         "paddw  %1,    %%mm0  \n\t" /* add rounding bias */
00101         "paddw  %1,    %%mm2  \n\t" /* add rounding bias */
00102 
00103         SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */
00104         SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */
00105         SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */
00106         SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */
00107         SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */
00108         SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */
00109         :: "r"(block), "m"(bias)
00110     );
00111 }
00112 
00113 static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
00114 {
00115     int i;
00116     DECLARE_ALIGNED(8, int16_t, b2)[64];
00117 
00118     for(i=0; i<2; i++){
00119         DECLARE_ALIGNED(8, uint64_t, tmp);
00120 
00121         cavs_idct8_1d(block+4*i, ff_pw_4);
00122 
00123         __asm__ volatile(
00124             "psraw     $3, %%mm7  \n\t"
00125             "psraw     $3, %%mm6  \n\t"
00126             "psraw     $3, %%mm5  \n\t"
00127             "psraw     $3, %%mm4  \n\t"
00128             "psraw     $3, %%mm3  \n\t"
00129             "psraw     $3, %%mm2  \n\t"
00130             "psraw     $3, %%mm1  \n\t"
00131             "psraw     $3, %%mm0  \n\t"
00132             "movq   %%mm7,    %0   \n\t"
00133             TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
00134             "movq   %%mm0,  8(%1)  \n\t"
00135             "movq   %%mm6, 24(%1)  \n\t"
00136             "movq   %%mm7, 40(%1)  \n\t"
00137             "movq   %%mm4, 56(%1)  \n\t"
00138             "movq    %0,    %%mm7  \n\t"
00139             TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
00140             "movq   %%mm7,   (%1)  \n\t"
00141             "movq   %%mm1, 16(%1)  \n\t"
00142             "movq   %%mm0, 32(%1)  \n\t"
00143             "movq   %%mm3, 48(%1)  \n\t"
00144             : "=m"(tmp)
00145             : "r"(b2+32*i)
00146             : "memory"
00147         );
00148     }
00149 
00150     for(i=0; i<2; i++){
00151         cavs_idct8_1d(b2+4*i, ff_pw_64.a);
00152 
00153         __asm__ volatile(
00154             "psraw     $7, %%mm7  \n\t"
00155             "psraw     $7, %%mm6  \n\t"
00156             "psraw     $7, %%mm5  \n\t"
00157             "psraw     $7, %%mm4  \n\t"
00158             "psraw     $7, %%mm3  \n\t"
00159             "psraw     $7, %%mm2  \n\t"
00160             "psraw     $7, %%mm1  \n\t"
00161             "psraw     $7, %%mm0  \n\t"
00162             "movq   %%mm7,    (%0)  \n\t"
00163             "movq   %%mm5,  16(%0)  \n\t"
00164             "movq   %%mm3,  32(%0)  \n\t"
00165             "movq   %%mm1,  48(%0)  \n\t"
00166             "movq   %%mm0,  64(%0)  \n\t"
00167             "movq   %%mm2,  80(%0)  \n\t"
00168             "movq   %%mm4,  96(%0)  \n\t"
00169             "movq   %%mm6, 112(%0)  \n\t"
00170             :: "r"(b2+4*i)
00171             : "memory"
00172         );
00173     }
00174 
00175     add_pixels_clamped_mmx(b2, dst, stride);
00176 }
00177 
00178 /*****************************************************************************
00179  *
00180  * motion compensation
00181  *
00182  ****************************************************************************/
00183 
00184 /* vertical filter [-1 -2 96 42 -7  0]  */
00185 #define QPEL_CAVSV1(A,B,C,D,E,F,OP,MUL2) \
00186         "movd (%0), "#F"            \n\t"\
00187         "movq "#C", %%mm6           \n\t"\
00188         "pmullw %5, %%mm6           \n\t"\
00189         "movq "#D", %%mm7           \n\t"\
00190         "pmullw "MANGLE(MUL2)", %%mm7\n\t"\
00191         "psllw $3, "#E"             \n\t"\
00192         "psubw "#E", %%mm6          \n\t"\
00193         "psraw $3, "#E"             \n\t"\
00194         "paddw %%mm7, %%mm6         \n\t"\
00195         "paddw "#E", %%mm6          \n\t"\
00196         "paddw "#B", "#B"           \n\t"\
00197         "pxor %%mm7, %%mm7          \n\t"\
00198         "add %2, %0                 \n\t"\
00199         "punpcklbw %%mm7, "#F"      \n\t"\
00200         "psubw "#B", %%mm6          \n\t"\
00201         "psraw $1, "#B"             \n\t"\
00202         "psubw "#A", %%mm6          \n\t"\
00203         "paddw %4, %%mm6            \n\t"\
00204         "psraw $7, %%mm6            \n\t"\
00205         "packuswb %%mm6, %%mm6      \n\t"\
00206         OP(%%mm6, (%1), A, d)            \
00207         "add %3, %1                 \n\t"
00208 
00209 /* vertical filter [ 0 -1  5  5 -1  0]  */
00210 #define QPEL_CAVSV2(A,B,C,D,E,F,OP,MUL2) \
00211         "movd (%0), "#F"            \n\t"\
00212         "movq "#C", %%mm6           \n\t"\
00213         "paddw "#D", %%mm6          \n\t"\
00214         "pmullw %5, %%mm6           \n\t"\
00215         "add %2, %0                 \n\t"\
00216         "punpcklbw %%mm7, "#F"      \n\t"\
00217         "psubw "#B", %%mm6          \n\t"\
00218         "psubw "#E", %%mm6          \n\t"\
00219         "paddw %4, %%mm6            \n\t"\
00220         "psraw $3, %%mm6            \n\t"\
00221         "packuswb %%mm6, %%mm6      \n\t"\
00222         OP(%%mm6, (%1), A, d)            \
00223         "add %3, %1                 \n\t"
00224 
00225 /* vertical filter [ 0 -7 42 96 -2 -1]  */
00226 #define QPEL_CAVSV3(A,B,C,D,E,F,OP,MUL2) \
00227         "movd (%0), "#F"            \n\t"\
00228         "movq "#C", %%mm6           \n\t"\
00229         "pmullw "MANGLE(MUL2)", %%mm6\n\t"\
00230         "movq "#D", %%mm7           \n\t"\
00231         "pmullw %5, %%mm7           \n\t"\
00232         "psllw $3, "#B"             \n\t"\
00233         "psubw "#B", %%mm6          \n\t"\
00234         "psraw $3, "#B"             \n\t"\
00235         "paddw %%mm7, %%mm6         \n\t"\
00236         "paddw "#B", %%mm6          \n\t"\
00237         "paddw "#E", "#E"           \n\t"\
00238         "pxor %%mm7, %%mm7          \n\t"\
00239         "add %2, %0                 \n\t"\
00240         "punpcklbw %%mm7, "#F"      \n\t"\
00241         "psubw "#E", %%mm6          \n\t"\
00242         "psraw $1, "#E"             \n\t"\
00243         "psubw "#F", %%mm6          \n\t"\
00244         "paddw %4, %%mm6            \n\t"\
00245         "psraw $7, %%mm6            \n\t"\
00246         "packuswb %%mm6, %%mm6      \n\t"\
00247         OP(%%mm6, (%1), A, d)            \
00248         "add %3, %1                 \n\t"
00249 
00250 
00251 #define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
00252     int w= 2;\
00253     src -= 2*srcStride;\
00254     \
00255     while(w--){\
00256       __asm__ volatile(\
00257         "pxor %%mm7, %%mm7          \n\t"\
00258         "movd (%0), %%mm0           \n\t"\
00259         "add %2, %0                 \n\t"\
00260         "movd (%0), %%mm1           \n\t"\
00261         "add %2, %0                 \n\t"\
00262         "movd (%0), %%mm2           \n\t"\
00263         "add %2, %0                 \n\t"\
00264         "movd (%0), %%mm3           \n\t"\
00265         "add %2, %0                 \n\t"\
00266         "movd (%0), %%mm4           \n\t"\
00267         "add %2, %0                 \n\t"\
00268         "punpcklbw %%mm7, %%mm0     \n\t"\
00269         "punpcklbw %%mm7, %%mm1     \n\t"\
00270         "punpcklbw %%mm7, %%mm2     \n\t"\
00271         "punpcklbw %%mm7, %%mm3     \n\t"\
00272         "punpcklbw %%mm7, %%mm4     \n\t"\
00273         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00274         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00275         VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00276         VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00277         VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
00278         VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
00279         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00280         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00281         \
00282         : "+a"(src), "+c"(dst)\
00283         : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD), "m"(MUL1)\
00284         : "memory"\
00285      );\
00286      if(h==16){\
00287         __asm__ volatile(\
00288             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00289             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00290             VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
00291             VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
00292             VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00293             VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00294             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00295             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00296             \
00297            : "+a"(src), "+c"(dst)\
00298            : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD),  "m"(MUL1)\
00299            : "memory"\
00300         );\
00301      }\
00302      src += 4-(h+5)*srcStride;\
00303      dst += 4-h*dstStride;\
00304    }
00305 
00306 #define QPEL_CAVS(OPNAME, OP, MMX)\
00307 static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00308     int h=8;\
00309     __asm__ volatile(\
00310         "pxor %%mm7, %%mm7          \n\t"\
00311         "movq %5, %%mm6             \n\t"\
00312         "1:                         \n\t"\
00313         "movq    (%0), %%mm0        \n\t"\
00314         "movq   1(%0), %%mm2        \n\t"\
00315         "movq %%mm0, %%mm1          \n\t"\
00316         "movq %%mm2, %%mm3          \n\t"\
00317         "punpcklbw %%mm7, %%mm0     \n\t"\
00318         "punpckhbw %%mm7, %%mm1     \n\t"\
00319         "punpcklbw %%mm7, %%mm2     \n\t"\
00320         "punpckhbw %%mm7, %%mm3     \n\t"\
00321         "paddw %%mm2, %%mm0         \n\t"\
00322         "paddw %%mm3, %%mm1         \n\t"\
00323         "pmullw %%mm6, %%mm0        \n\t"\
00324         "pmullw %%mm6, %%mm1        \n\t"\
00325         "movq   -1(%0), %%mm2       \n\t"\
00326         "movq    2(%0), %%mm4       \n\t"\
00327         "movq %%mm2, %%mm3          \n\t"\
00328         "movq %%mm4, %%mm5          \n\t"\
00329         "punpcklbw %%mm7, %%mm2     \n\t"\
00330         "punpckhbw %%mm7, %%mm3     \n\t"\
00331         "punpcklbw %%mm7, %%mm4     \n\t"\
00332         "punpckhbw %%mm7, %%mm5     \n\t"\
00333         "paddw %%mm4, %%mm2         \n\t"\
00334         "paddw %%mm3, %%mm5         \n\t"\
00335         "psubw %%mm2, %%mm0         \n\t"\
00336         "psubw %%mm5, %%mm1         \n\t"\
00337         "movq %6, %%mm5             \n\t"\
00338         "paddw %%mm5, %%mm0         \n\t"\
00339         "paddw %%mm5, %%mm1         \n\t"\
00340         "psraw $3, %%mm0            \n\t"\
00341         "psraw $3, %%mm1            \n\t"\
00342         "packuswb %%mm1, %%mm0      \n\t"\
00343         OP(%%mm0, (%1),%%mm5, q)         \
00344         "add %3, %0                 \n\t"\
00345         "add %4, %1                 \n\t"\
00346         "decl %2                    \n\t"\
00347         " jnz 1b                    \n\t"\
00348         : "+a"(src), "+c"(dst), "+m"(h)\
00349         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\
00350         : "memory"\
00351     );\
00352 }\
00353 \
00354 static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00355   QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00356 }\
00357 \
00358 static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00359   QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \
00360 }\
00361 \
00362 static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00363   QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00364 }\
00365 \
00366 static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00367     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00368 }\
00369 static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00370     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00371     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00372 }\
00373 \
00374 static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00375     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00376 }\
00377 static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00378     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00379     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00380 }\
00381 \
00382 static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00383     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00384 }\
00385 static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00386     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00387     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00388 }\
00389 \
00390 static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00391     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00392     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00393     src += 8*srcStride;\
00394     dst += 8*dstStride;\
00395     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00396     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00397 }\
00398 
00399 #define CAVS_MC(OPNAME, SIZE, MMX) \
00400 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00401     OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
00402 }\
00403 \
00404 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00405     OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
00406 }\
00407 \
00408 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00409     OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
00410 }\
00411 \
00412 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00413     OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
00414 }\
00415 
00416 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"
00417 #define AVG_3DNOW_OP(a,b,temp, size) \
00418 "mov" #size " " #b ", " #temp "   \n\t"\
00419 "pavgusb " #temp ", " #a "        \n\t"\
00420 "mov" #size " " #a ", " #b "      \n\t"
00421 #define AVG_MMX2_OP(a,b,temp, size) \
00422 "mov" #size " " #b ", " #temp "   \n\t"\
00423 "pavgb " #temp ", " #a "          \n\t"\
00424 "mov" #size " " #a ", " #b "      \n\t"
00425 
00426 QPEL_CAVS(put_,       PUT_OP, 3dnow)
00427 QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
00428 QPEL_CAVS(put_,       PUT_OP, mmx2)
00429 QPEL_CAVS(avg_,  AVG_MMX2_OP, mmx2)
00430 
00431 CAVS_MC(put_, 8, 3dnow)
00432 CAVS_MC(put_, 16,3dnow)
00433 CAVS_MC(avg_, 8, 3dnow)
00434 CAVS_MC(avg_, 16,3dnow)
00435 CAVS_MC(put_, 8, mmx2)
00436 CAVS_MC(put_, 16,mmx2)
00437 CAVS_MC(avg_, 8, mmx2)
00438 CAVS_MC(avg_, 16,mmx2)
00439 
00440 void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx) {
00441 #define dspfunc(PFX, IDX, NUM) \
00442     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
00443     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \
00444     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \
00445     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \
00446     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \
00447 
00448     dspfunc(put_cavs_qpel, 0, 16);
00449     dspfunc(put_cavs_qpel, 1, 8);
00450     dspfunc(avg_cavs_qpel, 0, 16);
00451     dspfunc(avg_cavs_qpel, 1, 8);
00452 #undef dspfunc
00453     c->cavs_idct8_add = cavs_idct8_add_mmx;
00454 }
00455 
00456 void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx) {
00457 #define dspfunc(PFX, IDX, NUM) \
00458     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
00459     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
00460     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
00461     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
00462     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \
00463 
00464     dspfunc(put_cavs_qpel, 0, 16);
00465     dspfunc(put_cavs_qpel, 1, 8);
00466     dspfunc(avg_cavs_qpel, 0, 16);
00467     dspfunc(avg_cavs_qpel, 1, 8);
00468 #undef dspfunc
00469     c->cavs_idct8_add = cavs_idct8_add_mmx;
00470 }
libavcodec/x86/cavsdsp_mmx.c
