MythTV: cavsdsp_mmx.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 optimised DSP functions, based on H.264 optimisations 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 "dsputil.h"
00026 #include "common.h"
00027 
00028 DECLARE_ALIGNED_8(static const uint64_t,ff_pw_4 ) = 0x0004000400040004ULL;
00029 DECLARE_ALIGNED_8(static const uint64_t,ff_pw_5 ) = 0x0005000500050005ULL;
00030 DECLARE_ALIGNED_8(static const uint64_t,ff_pw_7 ) = 0x0007000700070007ULL;
00031 DECLARE_ALIGNED_8(static const uint64_t,ff_pw_42) = 0x002A002A002A002AULL;
00032 DECLARE_ALIGNED_8(static const uint64_t,ff_pw_64) = 0x0040004000400040ULL;
00033 DECLARE_ALIGNED_8(static const uint64_t,ff_pw_96) = 0x0060006000600060ULL;
00034 
00035 /*****************************************************************************
00036  *
00037  * inverse transform
00038  *
00039  ****************************************************************************/
00040 
00041 #define SUMSUB_BA( a, b ) \
00042     "paddw "#b", "#a" \n\t"\
00043     "paddw "#b", "#b" \n\t"\
00044     "psubw "#a", "#b" \n\t"
00045 
00046 #define SBUTTERFLY(a,b,t,n)\
00047     "movq " #a ", " #t "              \n\t" /* abcd */\
00048     "punpckl" #n " " #b ", " #a "     \n\t" /* aebf */\
00049     "punpckh" #n " " #b ", " #t "     \n\t" /* cgdh */
00050 
00051 #define TRANSPOSE4(a,b,c,d,t)\
00052     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
00053     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
00054     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
00055     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
00056 
00057 static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)
00058 {
00059     asm volatile(
00060         "movq 112(%0), %%mm4  \n\t" /* mm4 = src7 */
00061         "movq  16(%0), %%mm5  \n\t" /* mm5 = src1 */
00062         "movq  80(%0), %%mm2  \n\t" /* mm2 = src5 */
00063         "movq  48(%0), %%mm7  \n\t" /* mm7 = src3 */
00064         "movq   %%mm4, %%mm0  \n\t"
00065         "movq   %%mm5, %%mm3  \n\t"
00066         "movq   %%mm2, %%mm6  \n\t"
00067         "movq   %%mm7, %%mm1  \n\t"
00068 
00069         "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */
00070         "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */
00071         "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */
00072         "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */
00073         "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */
00074         "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */
00075         "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */
00076         "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */
00077         "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */
00078         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */
00079         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */
00080         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */
00081 
00082         "movq   %%mm5, %%mm4  \n\t"
00083         "movq   %%mm7, %%mm6  \n\t"
00084         "movq   %%mm3, %%mm0  \n\t"
00085         "movq   %%mm1, %%mm2  \n\t"
00086         SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */
00087         "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */
00088         "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */
00089         "paddw  %%mm7, %%mm7  \n\t"
00090         "paddw  %%mm5, %%mm5  \n\t"
00091         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */
00092         "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */
00093 
00094         SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */
00095         "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */
00096         "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */
00097         "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */
00098         "paddw  %%mm1, %%mm1  \n\t"
00099         "paddw  %%mm3, %%mm3  \n\t"
00100         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */
00101         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */
00102 
00103         "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */
00104         "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */
00105         "movq   %%mm2, %%mm4  \n\t"
00106         "movq   %%mm6, %%mm0  \n\t"
00107         "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */
00108         "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */
00109         "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */
00110         "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */
00111         "paddw  %%mm2, %%mm2  \n\t"
00112         "paddw  %%mm0, %%mm0  \n\t"
00113         "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */
00114         "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */
00115 
00116         "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */
00117         "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */
00118         SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */
00119         "psllw  $3,    %%mm0  \n\t"
00120         "psllw  $3,    %%mm2  \n\t"
00121         "paddw  %1,    %%mm0  \n\t" /* add rounding bias */
00122         "paddw  %1,    %%mm2  \n\t" /* add rounding bias */
00123 
00124         SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */
00125         SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */
00126         SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */
00127         SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */
00128         SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */
00129         SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */
00130         :: "r"(block), "m"(bias)
00131     );
00132 }
00133 
00134 static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
00135 {
00136     int i;
00137     DECLARE_ALIGNED_8(int16_t, b2[64]);
00138 
00139     for(i=0; i<2; i++){
00140         DECLARE_ALIGNED_8(uint64_t, tmp);
00141 
00142         cavs_idct8_1d(block+4*i, ff_pw_4);
00143 
00144         asm volatile(
00145             "psraw     $3, %%mm7  \n\t"
00146             "psraw     $3, %%mm6  \n\t"
00147             "psraw     $3, %%mm5  \n\t"
00148             "psraw     $3, %%mm4  \n\t"
00149             "psraw     $3, %%mm3  \n\t"
00150             "psraw     $3, %%mm2  \n\t"
00151             "psraw     $3, %%mm1  \n\t"
00152             "psraw     $3, %%mm0  \n\t"
00153             "movq   %%mm7,    %0   \n\t"
00154             TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
00155             "movq   %%mm0,  8(%1)  \n\t"
00156             "movq   %%mm6, 24(%1)  \n\t"
00157             "movq   %%mm7, 40(%1)  \n\t"
00158             "movq   %%mm4, 56(%1)  \n\t"
00159             "movq    %0,    %%mm7  \n\t"
00160             TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
00161             "movq   %%mm7,   (%1)  \n\t"
00162             "movq   %%mm1, 16(%1)  \n\t"
00163             "movq   %%mm0, 32(%1)  \n\t"
00164             "movq   %%mm3, 48(%1)  \n\t"
00165             : "=m"(tmp)
00166             : "r"(b2+32*i)
00167             : "memory"
00168         );
00169     }
00170 
00171     for(i=0; i<2; i++){
00172         cavs_idct8_1d(b2+4*i, ff_pw_64);
00173 
00174         asm volatile(
00175             "psraw     $7, %%mm7  \n\t"
00176             "psraw     $7, %%mm6  \n\t"
00177             "psraw     $7, %%mm5  \n\t"
00178             "psraw     $7, %%mm4  \n\t"
00179             "psraw     $7, %%mm3  \n\t"
00180             "psraw     $7, %%mm2  \n\t"
00181             "psraw     $7, %%mm1  \n\t"
00182             "psraw     $7, %%mm0  \n\t"
00183             "movq   %%mm7,    (%0)  \n\t"
00184             "movq   %%mm5,  16(%0)  \n\t"
00185             "movq   %%mm3,  32(%0)  \n\t"
00186             "movq   %%mm1,  48(%0)  \n\t"
00187             "movq   %%mm0,  64(%0)  \n\t"
00188             "movq   %%mm2,  80(%0)  \n\t"
00189             "movq   %%mm4,  96(%0)  \n\t"
00190             "movq   %%mm6, 112(%0)  \n\t"
00191             :: "r"(b2+4*i)
00192             : "memory"
00193         );
00194     }
00195 
00196     add_pixels_clamped_mmx(b2, dst, stride);
00197 
00198     /* clear block */
00199     asm volatile(
00200             "pxor %%mm7, %%mm7   \n\t"
00201             "movq %%mm7, (%0)    \n\t"
00202             "movq %%mm7, 8(%0)   \n\t"
00203             "movq %%mm7, 16(%0)  \n\t"
00204             "movq %%mm7, 24(%0)  \n\t"
00205             "movq %%mm7, 32(%0)  \n\t"
00206             "movq %%mm7, 40(%0)  \n\t"
00207             "movq %%mm7, 48(%0)  \n\t"
00208             "movq %%mm7, 56(%0)  \n\t"
00209             "movq %%mm7, 64(%0)  \n\t"
00210             "movq %%mm7, 72(%0)  \n\t"
00211             "movq %%mm7, 80(%0)  \n\t"
00212             "movq %%mm7, 88(%0)  \n\t"
00213             "movq %%mm7, 96(%0)  \n\t"
00214             "movq %%mm7, 104(%0) \n\t"
00215             "movq %%mm7, 112(%0) \n\t"
00216             "movq %%mm7, 120(%0) \n\t"
00217             :: "r" (block)
00218     );
00219 }
00220 
00221 /*****************************************************************************
00222  *
00223  * motion compensation
00224  *
00225  ****************************************************************************/
00226 
00227 /* vertical filter [-1 -2 96 42 -7  0]  */
00228 #define QPEL_CAVSV1(A,B,C,D,E,F,OP)      \
00229         "movd (%0), "#F"            \n\t"\
00230         "movq "#C", %%mm6           \n\t"\
00231         "pmullw %5, %%mm6           \n\t"\
00232         "movq "#D", %%mm7           \n\t"\
00233         "pmullw %6, %%mm7           \n\t"\
00234         "psllw $3, "#E"             \n\t"\
00235         "psubw "#E", %%mm6          \n\t"\
00236         "psraw $3, "#E"             \n\t"\
00237         "paddw %%mm7, %%mm6         \n\t"\
00238         "paddw "#E", %%mm6          \n\t"\
00239         "paddw "#B", "#B"           \n\t"\
00240         "pxor %%mm7, %%mm7          \n\t"\
00241         "add %2, %0                 \n\t"\
00242         "punpcklbw %%mm7, "#F"      \n\t"\
00243         "psubw "#B", %%mm6          \n\t"\
00244         "psraw $1, "#B"             \n\t"\
00245         "psubw "#A", %%mm6          \n\t"\
00246         "paddw %4, %%mm6            \n\t"\
00247         "psraw $7, %%mm6            \n\t"\
00248         "packuswb %%mm6, %%mm6      \n\t"\
00249         OP(%%mm6, (%1), A, d)            \
00250         "add %3, %1                 \n\t"
00251 
00252 /* vertical filter [ 0 -1  5  5 -1  0]  */
00253 #define QPEL_CAVSV2(A,B,C,D,E,F,OP)      \
00254         "movd (%0), "#F"            \n\t"\
00255         "movq "#C", %%mm6           \n\t"\
00256         "paddw "#D", %%mm6          \n\t"\
00257         "pmullw %5, %%mm6           \n\t"\
00258         "add %2, %0                 \n\t"\
00259         "punpcklbw %%mm7, "#F"      \n\t"\
00260         "psubw "#B", %%mm6          \n\t"\
00261         "psubw "#E", %%mm6          \n\t"\
00262         "paddw %4, %%mm6            \n\t"\
00263         "psraw $3, %%mm6            \n\t"\
00264         "packuswb %%mm6, %%mm6      \n\t"\
00265         OP(%%mm6, (%1), A, d)            \
00266         "add %3, %1                 \n\t"
00267 
00268 /* vertical filter [ 0 -7 42 96 -2 -1]  */
00269 #define QPEL_CAVSV3(A,B,C,D,E,F,OP)      \
00270         "movd (%0), "#F"            \n\t"\
00271         "movq "#C", %%mm6           \n\t"\
00272         "pmullw %6, %%mm6           \n\t"\
00273         "movq "#D", %%mm7           \n\t"\
00274         "pmullw %5, %%mm7           \n\t"\
00275         "psllw $3, "#B"             \n\t"\
00276         "psubw "#B", %%mm6          \n\t"\
00277         "psraw $3, "#B"             \n\t"\
00278         "paddw %%mm7, %%mm6         \n\t"\
00279         "paddw "#B", %%mm6          \n\t"\
00280         "paddw "#E", "#E"           \n\t"\
00281         "pxor %%mm7, %%mm7          \n\t"\
00282         "add %2, %0                 \n\t"\
00283         "punpcklbw %%mm7, "#F"      \n\t"\
00284         "psubw "#E", %%mm6          \n\t"\
00285         "psraw $1, "#E"             \n\t"\
00286         "psubw "#F", %%mm6          \n\t"\
00287         "paddw %4, %%mm6            \n\t"\
00288         "psraw $7, %%mm6            \n\t"\
00289         "packuswb %%mm6, %%mm6      \n\t"\
00290         OP(%%mm6, (%1), A, d)            \
00291         "add %3, %1                 \n\t"
00292 
00293 
00294 #define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
00295     int w= 2;\
00296     src -= 2*srcStride;\
00297     \
00298     while(w--){\
00299       asm volatile(\
00300         "pxor %%mm7, %%mm7          \n\t"\
00301         "movd (%0), %%mm0           \n\t"\
00302         "add %2, %0                 \n\t"\
00303         "movd (%0), %%mm1           \n\t"\
00304         "add %2, %0                 \n\t"\
00305         "movd (%0), %%mm2           \n\t"\
00306         "add %2, %0                 \n\t"\
00307         "movd (%0), %%mm3           \n\t"\
00308         "add %2, %0                 \n\t"\
00309         "movd (%0), %%mm4           \n\t"\
00310         "add %2, %0                 \n\t"\
00311         "punpcklbw %%mm7, %%mm0     \n\t"\
00312         "punpcklbw %%mm7, %%mm1     \n\t"\
00313         "punpcklbw %%mm7, %%mm2     \n\t"\
00314         "punpcklbw %%mm7, %%mm3     \n\t"\
00315         "punpcklbw %%mm7, %%mm4     \n\t"\
00316         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00317         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00318         VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00319         VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00320         VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00321         VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00322         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00323         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00324         \
00325         : "+a"(src), "+c"(dst)\
00326         : "S"((long)srcStride), "D"((long)dstStride), "m"(ADD), "m"(MUL1), "m"(MUL2)\
00327         : "memory"\
00328      );\
00329      if(h==16){\
00330         asm volatile(\
00331             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00332             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00333             VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
00334             VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
00335             VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
00336             VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
00337             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
00338             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
00339             \
00340            : "+a"(src), "+c"(dst)\
00341            : "S"((long)srcStride), "D"((long)dstStride), "m"(ADD),  "m"(MUL1), "m"(MUL2)\
00342            : "memory"\
00343         );\
00344      }\
00345      src += 4-(h+5)*srcStride;\
00346      dst += 4-h*dstStride;\
00347    }
00348 
00349 #define QPEL_CAVS(OPNAME, OP, MMX)\
00350 static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00351     int h=8;\
00352     asm volatile(\
00353         "pxor %%mm7, %%mm7          \n\t"\
00354         "movq %5, %%mm6             \n\t"\
00355         "1:                         \n\t"\
00356         "movq    (%0), %%mm0        \n\t"\
00357         "movq   1(%0), %%mm2        \n\t"\
00358         "movq %%mm0, %%mm1          \n\t"\
00359         "movq %%mm2, %%mm3          \n\t"\
00360         "punpcklbw %%mm7, %%mm0     \n\t"\
00361         "punpckhbw %%mm7, %%mm1     \n\t"\
00362         "punpcklbw %%mm7, %%mm2     \n\t"\
00363         "punpckhbw %%mm7, %%mm3     \n\t"\
00364         "paddw %%mm2, %%mm0         \n\t"\
00365         "paddw %%mm3, %%mm1         \n\t"\
00366         "pmullw %%mm6, %%mm0        \n\t"\
00367         "pmullw %%mm6, %%mm1        \n\t"\
00368         "movq   -1(%0), %%mm2       \n\t"\
00369         "movq    2(%0), %%mm4       \n\t"\
00370         "movq %%mm2, %%mm3          \n\t"\
00371         "movq %%mm4, %%mm5          \n\t"\
00372         "punpcklbw %%mm7, %%mm2     \n\t"\
00373         "punpckhbw %%mm7, %%mm3     \n\t"\
00374         "punpcklbw %%mm7, %%mm4     \n\t"\
00375         "punpckhbw %%mm7, %%mm5     \n\t"\
00376         "paddw %%mm4, %%mm2         \n\t"\
00377         "paddw %%mm3, %%mm5         \n\t"\
00378         "psubw %%mm2, %%mm0         \n\t"\
00379         "psubw %%mm5, %%mm1         \n\t"\
00380         "movq %6, %%mm5             \n\t"\
00381         "paddw %%mm5, %%mm0         \n\t"\
00382         "paddw %%mm5, %%mm1         \n\t"\
00383         "psraw $3, %%mm0            \n\t"\
00384         "psraw $3, %%mm1            \n\t"\
00385         "packuswb %%mm1, %%mm0      \n\t"\
00386         OP(%%mm0, (%1),%%mm5, q)         \
00387         "add %3, %0                 \n\t"\
00388         "add %4, %1                 \n\t"\
00389         "decl %2                    \n\t"\
00390         " jnz 1b                    \n\t"\
00391         : "+a"(src), "+c"(dst), "+m"(h)\
00392         : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\
00393         : "memory"\
00394     );\
00395 }\
00396 \
00397 static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00398   QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00399 }\
00400 \
00401 static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00402   QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \
00403 }\
00404 \
00405 static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
00406   QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
00407 }\
00408 \
00409 static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00410     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00411 }\
00412 static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00413     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00414     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00415 }\
00416 \
00417 static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00418     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00419 }\
00420 static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00421     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00422     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00423 }\
00424 \
00425 static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00426     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
00427 }\
00428 static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00429     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
00430     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
00431 }\
00432 \
00433 static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
00434     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00435     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00436     src += 8*srcStride;\
00437     dst += 8*dstStride;\
00438     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
00439     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
00440 }\
00441 
00442 #define CAVS_MC(OPNAME, SIZE, MMX) \
00443 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00444     OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
00445 }\
00446 \
00447 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00448     OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
00449 }\
00450 \
00451 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00452     OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
00453 }\
00454 \
00455 static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
00456     OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
00457 }\
00458 
00459 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"
00460 #define AVG_3DNOW_OP(a,b,temp, size) \
00461 "mov" #size " " #b ", " #temp "   \n\t"\
00462 "pavgusb " #temp ", " #a "        \n\t"\
00463 "mov" #size " " #a ", " #b "      \n\t"
00464 #define AVG_MMX2_OP(a,b,temp, size) \
00465 "mov" #size " " #b ", " #temp "   \n\t"\
00466 "pavgb " #temp ", " #a "          \n\t"\
00467 "mov" #size " " #a ", " #b "      \n\t"
00468 
00469 QPEL_CAVS(put_,       PUT_OP, 3dnow)
00470 QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
00471 QPEL_CAVS(put_,       PUT_OP, mmx2)
00472 QPEL_CAVS(avg_,  AVG_MMX2_OP, mmx2)
00473 
00474 CAVS_MC(put_, 8, 3dnow)
00475 CAVS_MC(put_, 16,3dnow)
00476 CAVS_MC(avg_, 8, 3dnow)
00477 CAVS_MC(avg_, 16,3dnow)
00478 CAVS_MC(put_, 8, mmx2)
00479 CAVS_MC(put_, 16,mmx2)
00480 CAVS_MC(avg_, 8, mmx2)
00481 CAVS_MC(avg_, 16,mmx2)
00482 
00483 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00484 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00485 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00486 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00487 
00488 void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx) {
00489 #define dspfunc(PFX, IDX, NUM) \
00490     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
00491     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \
00492     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \
00493     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \
00494     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \
00495 
00496     dspfunc(put_cavs_qpel, 0, 16);
00497     dspfunc(put_cavs_qpel, 1, 8);
00498     dspfunc(avg_cavs_qpel, 0, 16);
00499     dspfunc(avg_cavs_qpel, 1, 8);
00500 #undef dspfunc
00501     c->cavs_idct8_add = cavs_idct8_add_mmx;
00502 }
00503 
00504 void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx) {
00505 #define dspfunc(PFX, IDX, NUM) \
00506     c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
00507     c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
00508     c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
00509     c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
00510     c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \
00511 
00512     dspfunc(put_cavs_qpel, 0, 16);
00513     dspfunc(put_cavs_qpel, 1, 8);
00514     dspfunc(avg_cavs_qpel, 0, 16);
00515     dspfunc(avg_cavs_qpel, 1, 8);
00516 #undef dspfunc
00517     c->cavs_idct8_add = cavs_idct8_add_mmx;
00518 }