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 "libavcodec/dsputil.h"
00029 #include "libavcodec/cavsdsp.h"
00030 #include "dsputil_mmx.h"
00031 
00032 #if HAVE_INLINE_ASM
00033 
00034 /* in/out: mma=mma+mmb, mmb=mmb-mma */
00035 #define SUMSUB_BA( a, b ) \
00036     "paddw "#b", "#a" \n\t"\
00037     "paddw "#b", "#b" \n\t"\
00038     "psubw "#a", "#b" \n\t"
00039 
00040 /*****************************************************************************
00041  *
00042  * inverse transform
00043  *
00044  ****************************************************************************/
00045 
00046 static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)
00047 {
00048     __asm__ volatile(
00049         "movq 112(%0), %%mm4  \n\t" /* mm4 = src7 */
00050         "movq  16(%0), %%mm5  \n\t" /* mm5 = src1 */
00051         "movq  80(%0), %%mm2  \n\t" /* mm2 = src5 */
00052         "movq  48(%0), %%mm7  \n\t" /* mm7 = src3 */
00053         "movq   %%mm4, %%mm0  \n\t"
00054         "movq   %%mm5, %%mm3  \n\t"
00055         "movq   %%mm2, %%mm6  \n\t"
00056         "movq   %%mm7, %%mm1  \n\t"
00057 
00058         "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */
00059         "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */
00060         "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */
00061         "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */
00062         "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */
00063         "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */
00064         "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */
00065         "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */
00066         "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */
00067         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */
00068         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */
00069         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */
00070 
00071         "movq   %%mm5, %%mm4  \n\t"
00072         "movq   %%mm7, %%mm6  \n\t"
00073         "movq   %%mm3, %%mm0  \n\t"
00074         "movq   %%mm1, %%mm2  \n\t"
00075         SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */
00076         "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */
00077         "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */
00078         "paddw  %%mm7, %%mm7  \n\t"
00079         "paddw  %%mm5, %%mm5  \n\t"
00080         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */
00081         "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */
00082 
00083         SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */
00084         "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */
00085         "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */
00086         "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */
00087         "paddw  %%mm1, %%mm1  \n\t"
00088         "paddw  %%mm3, %%mm3  \n\t"
00089         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */
00090         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */
00091 
00092         "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */
00093         "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */
00094         "movq   %%mm2, %%mm4  \n\t"
00095         "movq   %%mm6, %%mm0  \n\t"
00096         "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */
00097         "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */
00098         "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */
00099         "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */
00100         "paddw  %%mm2, %%mm2  \n\t"
00101         "paddw  %%mm0, %%mm0  \n\t"
00102         "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */
00103         "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */
00104 
00105         "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */
00106         "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */
00107         SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */
00108         "psllw  $3,    %%mm0  \n\t"
00109         "psllw  $3,    %%mm2  \n\t"
00110         "paddw  %1,    %%mm0  \n\t" /* add rounding bias */
00111         "paddw  %1,    %%mm2  \n\t" /* add rounding bias */
00112 
00113         SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */
00114         SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */
00115         SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */
00116         SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */
00117         SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */
00118         SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */
00119         :: "r"(block), "m"(bias)
00120     );
00121 }
00122 
00123 static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
00124 {
00125     int i;
00126     DECLARE_ALIGNED(8, int16_t, b2)[64];
00127 
00128     for(i=0; i<2; i++){
00129         DECLARE_ALIGNED(8, uint64_t, tmp);
00130 
00131         cavs_idct8_1d(block+4*i, ff_pw_4.a);
00132 
00133         __asm__ volatile(
00134             "psraw     $3, %%mm7  \n\t"
00135             "psraw     $3, %%mm6  \n\t"
00136             "psraw     $3, %%mm5  \n\t"
00137             "psraw     $3, %%mm4  \n\t"
00138             "psraw     $3, %%mm3  \n\t"
00139             "psraw     $3, %%mm2  \n\t"
00140             "psraw     $3, %%mm1  \n\t"
00141             "psraw     $3, %%mm0  \n\t"
00142             "movq   %%mm7,    %0   \n\t"
00143             TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
00144             "movq   %%mm0,  8(%1)  \n\t"
00145             "movq   %%mm6, 24(%1)  \n\t"
00146             "movq   %%mm7, 40(%1)  \n\t"
00147             "movq   %%mm4, 56(%1)  \n\t"
00148             "movq    %0,    %%mm7  \n\t"
00149             TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
00150             "movq   %%mm7,   (%1)  \n\t"
00151             "movq   %%mm1, 16(%1)  \n\t"
00152             "movq   %%mm0, 32(%1)  \n\t"
00153             "movq   %%mm3, 48(%1)  \n\t"
00154             : "=m"(tmp)
00155             : "r"(b2+32*i)
00156             : "memory"
00157         );
00158     }
00159 
00160     for(i=0; i<2; i++){
00161         cavs_idct8_1d(b2+4*i, ff_pw_64.a);
00162 
00163         __asm__ volatile(
00164             "psraw     $7, %%mm7  \n\t"
00165             "psraw     $7, %%mm6  \n\t"
00166             "psraw     $7, %%mm5  \n\t"
00167             "psraw     $7, %%mm4  \n\t"
00168             "psraw     $7, %%mm3  \n\t"
00169             "psraw     $7, %%mm2  \n\t"
00170             "psraw     $7, %%mm1  \n\t"
00171             "psraw     $7, %%mm0  \n\t"
00172             "movq   %%mm7,    (%0)  \n\t"
00173             "movq   %%mm5,  16(%0)  \n\t"
00174             "movq   %%mm3,  32(%0)  \n\t"
00175             "movq   %%mm1,  48(%0)  \n\t"
00176             "movq   %%mm0,  64(%0)  \n\t"
00177             "movq   %%mm2,  80(%0)  \n\t"
00178             "movq   %%mm4,  96(%0)  \n\t"
00179             "movq   %%mm6, 112(%0)  \n\t"
00180             :: "r"(b2+4*i)
00181             : "memory"
00182         );
00183     }
00184 
00185     ff_add_pixels_clamped_mmx(b2, dst, stride);
00186 }
00187 
00188 /*****************************************************************************
00189  *
00190  * motion compensation
00191  *
00192  ****************************************************************************/
00193 
00194 /* vertical filter [-1 -2 96 42 -7  0]  */
00195 #define QPEL_CAVSV1(A,B,C,D,E,F,OP,MUL2) \
00196         "movd (%0), "#F"            \n\t"\
00197         "movq "#C", %%mm6           \n\t"\
00198         "pmullw %5, %%mm6           \n\t"\
00199         "movq "#D", %%mm7           \n\t"\
00200         "pmullw "MANGLE(MUL2)", %%mm7\n\t"\
00201         "psllw $3, "#E"             \n\t"\
00202         "psubw "#E", %%mm6          \n\t"\
00203         "psraw $3, "#E"             \n\t"\
00204         "paddw %%mm7, %%mm6         \n\t"\
00205         "paddw "#E", %%mm6          \n\t"\
00206         "paddw "#B", "#B"           \n\t"\
00207         "pxor %%mm7, %%mm7          \n\t"\
00208         "add %2, %0                 \n\t"\
00209         "punpcklbw %%mm7, "#F"      \n\t"\
00210         "psubw "#B", %%mm6          \n\t"\
00211         "psraw $1, "#B"             \n\t"\
00212         "psubw "#A", %%mm6          \n\t"\
00213         "paddw %4, %%mm6            \n\t"\
00214         "psraw $7, %%mm6            \n\t"\
00215         "packuswb %%mm6, %%mm6      \n\t"\
00216         OP(%%mm6, (%1), A, d)            \
00217         "add %3, %1                 \n\t"
00218 
00219 /* vertical filter [ 0 -1  5  5 -1  0]  */
00220 #define QPEL_CAVSV2(A,B,C,D,E,F,OP,MUL2) \
00221         "movd (%0), "#F"            \n\t"\
00222         "movq "#C", %%mm6           \n\t"\
00223         "paddw "#D", %%mm6          \n\t"\
00224         "pmullw %5, %%mm6           \n\t"\
00225         "add %2, %0                 \n\t"\
00226         "punpcklbw %%mm7, "#F"      \n\t"\
00227         "psubw "#B", %%mm6          \n\t"\
00228         "psubw "#E", %%mm6          \n\t"\
00229         "paddw %4, %%mm6            \n\t"\
00230         "psraw $3, %%mm6            \n\t"\
00231         "packuswb %%mm6, %%mm6      \n\t"\
00232         OP(%%mm6, (%1), A, d)            \
00233         "add %3, %1                 \n\t"
00234 
00235 /* vertical filter [ 0 -7 42 96 -2 -1]  */
00236 #define QPEL_CAVSV3(A,B,C,D,E,F,OP,MUL2) \
00237         "movd (%0), "#F"            \n\t"\
00238         "movq "#C", %%mm6           \n\t"\
00239         "pmullw "MANGLE(MUL2)", %%mm6\n\t"\
00240         "movq "#D", %%mm7           \n\t"\
00241         "pmullw %5, %%mm7           \n\t"\
00242         "psllw $3, "#B"             \n\t"\
00243         "psubw "#B", %%mm6          \n\t"\
00244         "psraw $3, "#B"             \n\t"\
00245         "paddw %%mm7, %%mm6         \n\t"\
00246         "paddw "#B", %%mm6          \n\t"\
00247         "paddw "#E", "#E"           \n\t"\
00248         "pxor %%mm7, %%mm7          \n\t"\
00249         "add %2, %0                 \n\t"\
00250         "punpcklbw %%mm7, "#F"      \n\t"\
00251         "psubw "#E", %%mm6          \n\t"\
00252         "psraw $1, "#E"             \n\t"\
00253         "psubw "#F", %%mm6          \n\t"\
00254         "paddw %4, %%mm6            \n\t"\
00255         "psraw $7, %%mm6            \n\t"\
00256         "packuswb %%mm6, %%mm6      \n\t"\
00257         OP(%%mm6, (%1), A, d)            \
00258         "add %3, %1                 \n\t"
00259 
00260 
00261 #define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
00262     int w= 2;\
00263     src -= 2*srcStride;\
00264     \
00265     while(w--){\
00266       __asm__ volatile(\
00267         "pxor %%mm7, %%mm7          \n\t"\
00268         "movd (%0), %%mm0           \n\t"\
00269         "add %2, %0                 \n\t"\
00270         "movd (%0), %%mm1           \n\t"\
00271         "add %2, %0                 \n\t"\
00272         "movd (%0), %%mm2           \n\t"\
00273         "add %2, %0                 \n\t"\
00274         "movd (%0), %%mm3           \n\t"\
00275         "add %2, %0                 \n\t"\
00276         "movd (%0), %%mm4           \n\t"\
00277         "add %2, %0                 \n\t"\
00278         "punpcklbw %%mm7, %%mm0     \n\t"\
00279         "punpcklbw %%mm7, %%mm1     \n\t"\
00280         "punpcklbw %%mm7, %%mm2     \n\t"\
00281         "punpcklbw %%mm7, %%mm3     \n\t"\
00282         "punpcklbw %%mm7, %%mm4     \n\t"\
00283         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00284         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00285         VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00286         VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00287         VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
00288         VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
00289         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00290         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00291         \
00292         : "+a"(src), "+c"(dst)\
00293         : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD), "m"(MUL1)\
00294         : "memory"\
00295      );\
00296      if(h==16){\
00297         __asm__ volatile(\
00298             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00299             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00300             VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\
00301             VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\
00302             VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\
00303             VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\
00304             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\
00305             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\
00306             \
00307            : "+a"(src), "+c"(dst)\
00308            : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD),  "m"(MUL1)\
00309            : "memory"\
00310         );\
00311      }\
00312      src += 4-(h+5)*srcStride;\
00313      dst += 4-h*dstStride;\
00314    }
00315 
00316 #define QPEL_CAVS(OPNAME, OP, MMX)\
00317 static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00318     int h=8;\
00319     __asm__ volatile(\
00320         "pxor %%mm7, %%mm7          \n\t"\
00321         "movq %5, %%mm6             \n\t"\
00322         "1:                         \n\t"\
00323         "movq    (%0), %%mm0        \n\t"\
00324         "movq   1(%0), %%mm2        \n\t"\
00325         "movq %%mm0, %%mm1          \n\t"\
00326         "movq %%mm2, %%mm3          \n\t"\
00327         "punpcklbw %%mm7, %%mm0     \n\t"\
00328         "punpckhbw %%mm7, %%mm1     \n\t"\
00329         "punpcklbw %%mm7, %%mm2     \n\t"\
00330         "punpckhbw %%mm7, %%mm3     \n\t"\
00331         "paddw %%mm2, %%mm0         \n\t"\
00332         "paddw %%mm3, %%mm1         \n\t"\
00333         "pmullw %%mm6, %%mm0        \n\t"\
00334         "pmullw %%mm6, %%mm1        \n\t"\
00335         "movq   -1(%0), %%mm2       \n\t"\
00336         "movq    2(%0), %%mm4       \n\t"\
00337         "movq %%mm2, %%mm3          \n\t"\
00338         "movq %%mm4, %%mm5          \n\t"\
00339         "punpcklbw %%mm7, %%mm2     \n\t"\
00340         "punpckhbw %%mm7, %%mm3     \n\t"\
00341         "punpcklbw %%mm7, %%mm4     \n\t"\
00342         "punpckhbw %%mm7, %%mm5     \n\t"\
00343         "paddw %%mm4, %%mm2         \n\t"\
00344         "paddw %%mm3, %%mm5         \n\t"\
00345         "psubw %%mm2, %%mm0         \n\t"\
00346         "psubw %%mm5, %%mm1         \n\t"\
00347         "movq %6, %%mm5             \n\t"\
00348         "paddw %%mm5, %%mm0         \n\t"\
00349         "paddw %%mm5, %%mm1         \n\t"\
00350         "psraw $3, %%mm0            \n\t"\
00351         "psraw $3, %%mm1            \n\t"\
00352         "packuswb %%mm1, %%mm0      \n\t"\
00353         OP(%%mm0, (%1),%%mm5, q)         \
00354         "add %3, %0                 \n\t"\
00355         "add %4, %1                 \n\t"\
00356         "decl %2                    \n\t"\
00357         " jnz 1b                    \n\t"\
00358         : "+a"(src), "+c"(dst), "+m"(h)\
00359         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\
00360         : "memory"\
00361     );\
00362 }\
00363 \
00364 static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00365   QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00366 }\
00367 \
00368 static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00369   QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \
00370 }\
00371 \
00372 static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00373   QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00374 }\
00375 \
00376 static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00377     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00378 }\
00379 static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00380     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00381     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00382 }\
00383 \
00384 static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00385     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00386 }\
00387 static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00388     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00389     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00390 }\
00391 \
00392 static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00393     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00394 }\
00395 static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00396     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00397     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00398 }\
00399 \
00400 static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00401     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00402     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00403     src += 8*srcStride;\
00404     dst += 8*dstStride;\
00405     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00406     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00407 }\
00408 
00409 #define CAVS_MC(OPNAME, SIZE, MMX) \
00410 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00411     OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
00412 }\
00413 \
00414 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00415     OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
00416 }\
00417 \
00418 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00419     OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
00420 }\
00421 \
00422 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00423     OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
00424 }\
00425 
00426 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"
00427 #define AVG_3DNOW_OP(a,b,temp, size) \
00428 "mov" #size " " #b ", " #temp "   \n\t"\
00429 "pavgusb " #temp ", " #a "        \n\t"\
00430 "mov" #size " " #a ", " #b "      \n\t"
00431 #define AVG_MMX2_OP(a,b,temp, size) \
00432 "mov" #size " " #b ", " #temp "   \n\t"\
00433 "pavgb " #temp ", " #a "          \n\t"\
00434 "mov" #size " " #a ", " #b "      \n\t"
00435 
00436 QPEL_CAVS(put_,       PUT_OP, 3dnow)
00437 QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
00438 QPEL_CAVS(put_,       PUT_OP, mmx2)
00439 QPEL_CAVS(avg_,  AVG_MMX2_OP, mmx2)
00440 
00441 CAVS_MC(put_, 8, 3dnow)
00442 CAVS_MC(put_, 16,3dnow)
00443 CAVS_MC(avg_, 8, 3dnow)
00444 CAVS_MC(avg_, 16,3dnow)
00445 CAVS_MC(put_, 8, mmx2)
00446 CAVS_MC(put_, 16,mmx2)
00447 CAVS_MC(avg_, 8, mmx2)
00448 CAVS_MC(avg_, 16,mmx2)
00449 
00450 static void ff_cavsdsp_init_mmx2(CAVSDSPContext* c, AVCodecContext *avctx) {
00451 #define dspfunc(PFX, IDX, NUM) \
00452     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
00453     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \
00454     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \
00455     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \
00456     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \
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 
00467 static void ff_cavsdsp_init_3dnow(CAVSDSPContext* c, AVCodecContext *avctx) {
00468 #define dspfunc(PFX, IDX, NUM) \
00469     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
00470     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
00471     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
00472     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
00473     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \
00474 
00475     dspfunc(put_cavs_qpel, 0, 16);
00476     dspfunc(put_cavs_qpel, 1, 8);
00477     dspfunc(avg_cavs_qpel, 0, 16);
00478     dspfunc(avg_cavs_qpel, 1, 8);
00479 #undef dspfunc
00480     c->cavs_idct8_add = cavs_idct8_add_mmx;
00481     c->idct_perm = FF_TRANSPOSE_IDCT_PERM;
00482 }
00483 
00484 #endif /* HAVE_INLINE_ASM */
00485 
00486 void ff_cavsdsp_init_mmx(CAVSDSPContext *c, AVCodecContext *avctx)
00487 {
00488     int mm_flags = av_get_cpu_flags();
00489 
00490 #if HAVE_INLINE_ASM
00491     if (mm_flags & AV_CPU_FLAG_MMXEXT) ff_cavsdsp_init_mmx2(c, avctx);
00492     if (mm_flags & AV_CPU_FLAG_3DNOW) ff_cavsdsp_init_3dnow(c, avctx);
00493 #endif /* HAVE_INLINE_ASM */
00494 }