FFmpeg: libpostproc/postprocess_template.c Source File

00001 /*
00002  * Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg is free software; you can redistribute it and/or modify
00007  * it under the terms of the GNU General Public License as published by
00008  * the Free Software Foundation; either version 2 of the License, or
00009  * (at your option) any later version.
00010  *
00011  * FFmpeg is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
00014  * GNU General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU General Public License
00017  * along with FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00026 #include "libavutil/x86_cpu.h"
00027 
00028 #undef REAL_PAVGB
00029 #undef PAVGB
00030 #undef PMINUB
00031 #undef PMAXUB
00032 
00033 #if   HAVE_MMX2
00034 #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
00035 #elif HAVE_AMD3DNOW
00036 #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
00037 #endif
00038 #define PAVGB(a,b)  REAL_PAVGB(a,b)
00039 
00040 #if   HAVE_MMX2
00041 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
00042 #elif HAVE_MMX
00043 #define PMINUB(b,a,t) \
00044     "movq " #a ", " #t " \n\t"\
00045     "psubusb " #b ", " #t " \n\t"\
00046     "psubb " #t ", " #a " \n\t"
00047 #endif
00048 
00049 #if   HAVE_MMX2
00050 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
00051 #elif HAVE_MMX
00052 #define PMAXUB(a,b) \
00053     "psubusb " #a ", " #b " \n\t"\
00054     "paddb " #a ", " #b " \n\t"
00055 #endif
00056 
00057 //FIXME? |255-0| = 1 (should not be a problem ...)
00058 #if HAVE_MMX
00059 
00062 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
00063     int numEq= 0, dcOk;
00064     src+= stride*4; // src points to begin of the 8x8 Block
00065     __asm__ volatile(
00066         "movq %0, %%mm7                         \n\t"
00067         "movq %1, %%mm6                         \n\t"
00068         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
00069         );
00070 
00071     __asm__ volatile(
00072         "lea (%2, %3), %%"REG_a"                \n\t"
00073 //      0       1       2       3       4       5       6       7       8       9
00074 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
00075 
00076         "movq (%2), %%mm0                       \n\t"
00077         "movq (%%"REG_a"), %%mm1                \n\t"
00078         "movq %%mm0, %%mm3                      \n\t"
00079         "movq %%mm0, %%mm4                      \n\t"
00080         PMAXUB(%%mm1, %%mm4)
00081         PMINUB(%%mm1, %%mm3, %%mm5)
00082         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
00083         "paddb %%mm7, %%mm0                     \n\t"
00084         "pcmpgtb %%mm6, %%mm0                   \n\t"
00085 
00086         "movq (%%"REG_a",%3), %%mm2             \n\t"
00087         PMAXUB(%%mm2, %%mm4)
00088         PMINUB(%%mm2, %%mm3, %%mm5)
00089         "psubb %%mm2, %%mm1                     \n\t"
00090         "paddb %%mm7, %%mm1                     \n\t"
00091         "pcmpgtb %%mm6, %%mm1                   \n\t"
00092         "paddb %%mm1, %%mm0                     \n\t"
00093 
00094         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
00095         PMAXUB(%%mm1, %%mm4)
00096         PMINUB(%%mm1, %%mm3, %%mm5)
00097         "psubb %%mm1, %%mm2                     \n\t"
00098         "paddb %%mm7, %%mm2                     \n\t"
00099         "pcmpgtb %%mm6, %%mm2                   \n\t"
00100         "paddb %%mm2, %%mm0                     \n\t"
00101 
00102         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
00103 
00104         "movq (%2, %3, 4), %%mm2                \n\t"
00105         PMAXUB(%%mm2, %%mm4)
00106         PMINUB(%%mm2, %%mm3, %%mm5)
00107         "psubb %%mm2, %%mm1                     \n\t"
00108         "paddb %%mm7, %%mm1                     \n\t"
00109         "pcmpgtb %%mm6, %%mm1                   \n\t"
00110         "paddb %%mm1, %%mm0                     \n\t"
00111 
00112         "movq (%%"REG_a"), %%mm1                \n\t"
00113         PMAXUB(%%mm1, %%mm4)
00114         PMINUB(%%mm1, %%mm3, %%mm5)
00115         "psubb %%mm1, %%mm2                     \n\t"
00116         "paddb %%mm7, %%mm2                     \n\t"
00117         "pcmpgtb %%mm6, %%mm2                   \n\t"
00118         "paddb %%mm2, %%mm0                     \n\t"
00119 
00120         "movq (%%"REG_a", %3), %%mm2            \n\t"
00121         PMAXUB(%%mm2, %%mm4)
00122         PMINUB(%%mm2, %%mm3, %%mm5)
00123         "psubb %%mm2, %%mm1                     \n\t"
00124         "paddb %%mm7, %%mm1                     \n\t"
00125         "pcmpgtb %%mm6, %%mm1                   \n\t"
00126         "paddb %%mm1, %%mm0                     \n\t"
00127 
00128         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
00129         PMAXUB(%%mm1, %%mm4)
00130         PMINUB(%%mm1, %%mm3, %%mm5)
00131         "psubb %%mm1, %%mm2                     \n\t"
00132         "paddb %%mm7, %%mm2                     \n\t"
00133         "pcmpgtb %%mm6, %%mm2                   \n\t"
00134         "paddb %%mm2, %%mm0                     \n\t"
00135         "psubusb %%mm3, %%mm4                   \n\t"
00136 
00137         "                                       \n\t"
00138 #if HAVE_MMX2
00139         "pxor %%mm7, %%mm7                      \n\t"
00140         "psadbw %%mm7, %%mm0                    \n\t"
00141 #else
00142         "movq %%mm0, %%mm1                      \n\t"
00143         "psrlw $8, %%mm0                        \n\t"
00144         "paddb %%mm1, %%mm0                     \n\t"
00145         "movq %%mm0, %%mm1                      \n\t"
00146         "psrlq $16, %%mm0                       \n\t"
00147         "paddb %%mm1, %%mm0                     \n\t"
00148         "movq %%mm0, %%mm1                      \n\t"
00149         "psrlq $32, %%mm0                       \n\t"
00150         "paddb %%mm1, %%mm0                     \n\t"
00151 #endif
00152         "movq %4, %%mm7                         \n\t" // QP,..., QP
00153         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
00154         "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
00155         "packssdw %%mm4, %%mm4                  \n\t"
00156         "movd %%mm0, %0                         \n\t"
00157         "movd %%mm4, %1                         \n\t"
00158 
00159         : "=r" (numEq), "=r" (dcOk)
00160         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00161         : "%"REG_a
00162         );
00163 
00164     numEq= (-numEq) &0xFF;
00165     if(numEq > c->ppMode.flatnessThreshold){
00166         if(dcOk) return 0;
00167         else     return 1;
00168     }else{
00169         return 2;
00170     }
00171 }
00172 #endif //HAVE_MMX
00173 
00178 #if !HAVE_ALTIVEC
00179 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
00180 {
00181 #if HAVE_MMX2 || HAVE_AMD3DNOW
00182     src+= stride*3;
00183     __asm__ volatile(        //"movv %0 %1 %2\n\t"
00184         "movq %2, %%mm0                         \n\t"  // QP,..., QP
00185         "pxor %%mm4, %%mm4                      \n\t"
00186 
00187         "movq (%0), %%mm6                       \n\t"
00188         "movq (%0, %1), %%mm5                   \n\t"
00189         "movq %%mm5, %%mm1                      \n\t"
00190         "movq %%mm6, %%mm2                      \n\t"
00191         "psubusb %%mm6, %%mm5                   \n\t"
00192         "psubusb %%mm1, %%mm2                   \n\t"
00193         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
00194         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
00195         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
00196 
00197         "pand %%mm2, %%mm6                      \n\t"
00198         "pandn %%mm1, %%mm2                     \n\t"
00199         "por %%mm2, %%mm6                       \n\t"// First Line to Filter
00200 
00201         "movq (%0, %1, 8), %%mm5                \n\t"
00202         "lea (%0, %1, 4), %%"REG_a"             \n\t"
00203         "lea (%0, %1, 8), %%"REG_c"             \n\t"
00204         "sub %1, %%"REG_c"                      \n\t"
00205         "add %1, %0                             \n\t" // %0 points to line 1 not 0
00206         "movq (%0, %1, 8), %%mm7                \n\t"
00207         "movq %%mm5, %%mm1                      \n\t"
00208         "movq %%mm7, %%mm2                      \n\t"
00209         "psubusb %%mm7, %%mm5                   \n\t"
00210         "psubusb %%mm1, %%mm2                   \n\t"
00211         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
00212         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
00213         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
00214 
00215         "pand %%mm2, %%mm7                      \n\t"
00216         "pandn %%mm1, %%mm2                     \n\t"
00217         "por %%mm2, %%mm7                       \n\t" // First Line to Filter
00218 
00219 
00220         //      1       2       3       4       5       6       7       8
00221         //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
00222         // 6 4 2 2 1 1
00223         // 6 4 4 2
00224         // 6 8 2
00225 
00226         "movq (%0, %1), %%mm0                   \n\t" //  1
00227         "movq %%mm0, %%mm1                      \n\t" //  1
00228         PAVGB(%%mm6, %%mm0)                           //1 1        /2
00229         PAVGB(%%mm6, %%mm0)                           //3 1        /4
00230 
00231         "movq (%0, %1, 4), %%mm2                \n\t" //     1
00232         "movq %%mm2, %%mm5                      \n\t" //     1
00233         PAVGB((%%REGa), %%mm2)                        //    11        /2
00234         PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
00235         "movq %%mm2, %%mm3                      \n\t" //   211        /4
00236         "movq (%0), %%mm4                       \n\t" // 1
00237         PAVGB(%%mm4, %%mm3)                           // 4 211        /8
00238         PAVGB(%%mm0, %%mm3)                           //642211        /16
00239         "movq %%mm3, (%0)                       \n\t" // X
00240         // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
00241         "movq %%mm1, %%mm0                      \n\t" //  1
00242         PAVGB(%%mm6, %%mm0)                           //1 1        /2
00243         "movq %%mm4, %%mm3                      \n\t" // 1
00244         PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
00245         PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
00246         PAVGB((%%REGa), %%mm5)                        //    211 /4
00247         PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
00248         PAVGB(%%mm0, %%mm3)                           //4242211 /16
00249         "movq %%mm3, (%0,%1)                    \n\t" //  X
00250         // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
00251         PAVGB(%%mm4, %%mm6)                                   //11        /2
00252         "movq (%%"REG_c"), %%mm0                \n\t" //       1
00253         PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
00254         "movq %%mm0, %%mm3                      \n\t" //      11/2
00255         PAVGB(%%mm1, %%mm0)                           //  2   11/4
00256         PAVGB(%%mm6, %%mm0)                           //222   11/8
00257         PAVGB(%%mm2, %%mm0)                           //22242211/16
00258         "movq (%0, %1, 2), %%mm2                \n\t" //   1
00259         "movq %%mm0, (%0, %1, 2)                \n\t" //   X
00260         // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
00261         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
00262         PAVGB((%%REGc), %%mm0)                        //       11        /2
00263         PAVGB(%%mm0, %%mm6)                           //11     11        /4
00264         PAVGB(%%mm1, %%mm4)                           // 11                /2
00265         PAVGB(%%mm2, %%mm1)                           //  11                /2
00266         PAVGB(%%mm1, %%mm6)                           //1122   11        /8
00267         PAVGB(%%mm5, %%mm6)                           //112242211        /16
00268         "movq (%%"REG_a"), %%mm5                \n\t" //    1
00269         "movq %%mm6, (%%"REG_a")                \n\t" //    X
00270         // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
00271         "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
00272         PAVGB(%%mm7, %%mm6)                           //        11        /2
00273         PAVGB(%%mm4, %%mm6)                           // 11     11        /4
00274         PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
00275         PAVGB(%%mm5, %%mm2)                           //   11                /2
00276         "movq (%0, %1, 4), %%mm4                \n\t" //     1
00277         PAVGB(%%mm4, %%mm2)                           //   112                /4
00278         PAVGB(%%mm2, %%mm6)                           // 112242211        /16
00279         "movq %%mm6, (%0, %1, 4)                \n\t" //     X
00280         // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
00281         PAVGB(%%mm7, %%mm1)                           //  11     2        /4
00282         PAVGB(%%mm4, %%mm5)                           //    11                /2
00283         PAVGB(%%mm5, %%mm0)                           //    11 11        /4
00284         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
00285         PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
00286         PAVGB(%%mm0, %%mm1)                           //  11224222        /16
00287         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
00288         // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
00289         PAVGB((%%REGc), %%mm2)                        //   112 4        /8
00290         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
00291         PAVGB(%%mm0, %%mm6)                           //      1 1        /2
00292         PAVGB(%%mm7, %%mm6)                           //      1 12        /4
00293         PAVGB(%%mm2, %%mm6)                           //   1122424        /4
00294         "movq %%mm6, (%%"REG_c")                \n\t" //       X
00295         // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
00296         PAVGB(%%mm7, %%mm5)                           //    11   2        /4
00297         PAVGB(%%mm7, %%mm5)                           //    11   6        /8
00298 
00299         PAVGB(%%mm3, %%mm0)                           //      112        /4
00300         PAVGB(%%mm0, %%mm5)                           //    112246        /16
00301         "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
00302         "sub %1, %0                             \n\t"
00303 
00304         :
00305         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00306         : "%"REG_a, "%"REG_c
00307     );
00308 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00309     const int l1= stride;
00310     const int l2= stride + l1;
00311     const int l3= stride + l2;
00312     const int l4= stride + l3;
00313     const int l5= stride + l4;
00314     const int l6= stride + l5;
00315     const int l7= stride + l6;
00316     const int l8= stride + l7;
00317     const int l9= stride + l8;
00318     int x;
00319     src+= stride*3;
00320     for(x=0; x<BLOCK_SIZE; x++){
00321         const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
00322         const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
00323 
00324         int sums[10];
00325         sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
00326         sums[1] = sums[0] - first  + src[l4];
00327         sums[2] = sums[1] - first  + src[l5];
00328         sums[3] = sums[2] - first  + src[l6];
00329         sums[4] = sums[3] - first  + src[l7];
00330         sums[5] = sums[4] - src[l1] + src[l8];
00331         sums[6] = sums[5] - src[l2] + last;
00332         sums[7] = sums[6] - src[l3] + last;
00333         sums[8] = sums[7] - src[l4] + last;
00334         sums[9] = sums[8] - src[l5] + last;
00335 
00336         src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
00337         src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
00338         src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
00339         src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
00340         src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
00341         src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
00342         src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
00343         src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
00344 
00345         src++;
00346     }
00347 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
00348 }
00349 #endif //HAVE_ALTIVEC
00350 
00358 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
00359 {
00360 #if HAVE_MMX2 || HAVE_AMD3DNOW
00361     src+= stride*3;
00362 
00363     __asm__ volatile(
00364         "pxor %%mm7, %%mm7                      \n\t" // 0
00365         "lea (%0, %1), %%"REG_a"                \n\t"
00366         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00367 //      0       1       2       3       4       5       6       7       8       9
00368 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
00369         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
00370         "movq (%0, %1, 4), %%mm1                \n\t" // line 4
00371         "movq %%mm1, %%mm2                      \n\t" // line 4
00372         "psubusb %%mm0, %%mm1                   \n\t"
00373         "psubusb %%mm2, %%mm0                   \n\t"
00374         "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
00375         "movq (%%"REG_c"), %%mm3                \n\t" // line 5
00376         "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
00377         "movq %%mm3, %%mm5                      \n\t" // line 5
00378         "psubusb %%mm4, %%mm3                   \n\t"
00379         "psubusb %%mm5, %%mm4                   \n\t"
00380         "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
00381         PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
00382         "movq %%mm2, %%mm1                      \n\t" // line 4
00383         "psubusb %%mm5, %%mm2                   \n\t"
00384         "movq %%mm2, %%mm4                      \n\t"
00385         "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
00386         "psubusb %%mm1, %%mm5                   \n\t"
00387         "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
00388         "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
00389         "movq %%mm4, %%mm3                      \n\t" // d
00390         "movq %2, %%mm0                         \n\t"
00391         "paddusb %%mm0, %%mm0                   \n\t"
00392         "psubusb %%mm0, %%mm4                   \n\t"
00393         "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
00394         "psubusb "MANGLE(b01)", %%mm3           \n\t"
00395         "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
00396 
00397         PAVGB(%%mm7, %%mm3)                           // d/2
00398         "movq %%mm3, %%mm1                      \n\t" // d/2
00399         PAVGB(%%mm7, %%mm3)                           // d/4
00400         PAVGB(%%mm1, %%mm3)                           // 3*d/8
00401 
00402         "movq (%0, %1, 4), %%mm0                \n\t" // line 4
00403         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
00404         "psubusb %%mm3, %%mm0                   \n\t"
00405         "pxor %%mm2, %%mm0                      \n\t"
00406         "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
00407 
00408         "movq (%%"REG_c"), %%mm0                \n\t" // line 5
00409         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
00410         "paddusb %%mm3, %%mm0                   \n\t"
00411         "pxor %%mm2, %%mm0                      \n\t"
00412         "movq %%mm0, (%%"REG_c")                \n\t" // line 5
00413 
00414         PAVGB(%%mm7, %%mm1)                           // d/4
00415 
00416         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
00417         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
00418         "psubusb %%mm1, %%mm0                   \n\t"
00419         "pxor %%mm2, %%mm0                      \n\t"
00420         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
00421 
00422         "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
00423         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
00424         "paddusb %%mm1, %%mm0                   \n\t"
00425         "pxor %%mm2, %%mm0                      \n\t"
00426         "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
00427 
00428         PAVGB(%%mm7, %%mm1)                           // d/8
00429 
00430         "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
00431         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
00432         "psubusb %%mm1, %%mm0                   \n\t"
00433         "pxor %%mm2, %%mm0                      \n\t"
00434         "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
00435 
00436         "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
00437         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
00438         "paddusb %%mm1, %%mm0                   \n\t"
00439         "pxor %%mm2, %%mm0                      \n\t"
00440         "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
00441 
00442         :
00443         : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
00444         : "%"REG_a, "%"REG_c
00445     );
00446 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00447 
00448     const int l1= stride;
00449     const int l2= stride + l1;
00450     const int l3= stride + l2;
00451     const int l4= stride + l3;
00452     const int l5= stride + l4;
00453     const int l6= stride + l5;
00454     const int l7= stride + l6;
00455 //    const int l8= stride + l7;
00456 //    const int l9= stride + l8;
00457     int x;
00458 
00459     src+= stride*3;
00460     for(x=0; x<BLOCK_SIZE; x++){
00461         int a= src[l3] - src[l4];
00462         int b= src[l4] - src[l5];
00463         int c= src[l5] - src[l6];
00464 
00465         int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
00466         d= FFMAX(d, 0);
00467 
00468         if(d < co->QP*2){
00469             int v = d * FFSIGN(-b);
00470 
00471             src[l2] +=v>>3;
00472             src[l3] +=v>>2;
00473             src[l4] +=(3*v)>>3;
00474             src[l5] -=(3*v)>>3;
00475             src[l6] -=v>>2;
00476             src[l7] -=v>>3;
00477         }
00478         src++;
00479     }
00480 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
00481 }
00482 
00483 #if !HAVE_ALTIVEC
00484 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
00485 {
00486 #if HAVE_MMX2 || HAVE_AMD3DNOW
00487 /*
00488     uint8_t tmp[16];
00489     const int l1= stride;
00490     const int l2= stride + l1;
00491     const int l3= stride + l2;
00492     const int l4= (int)tmp - (int)src - stride*3;
00493     const int l5= (int)tmp - (int)src - stride*3 + 8;
00494     const int l6= stride*3 + l3;
00495     const int l7= stride + l6;
00496     const int l8= stride + l7;
00497 
00498     memcpy(tmp, src+stride*7, 8);
00499     memcpy(tmp+8, src+stride*8, 8);
00500 */
00501     src+= stride*4;
00502     __asm__ volatile(
00503 
00504 #if 0 //slightly more accurate and slightly slower
00505         "pxor %%mm7, %%mm7                      \n\t" // 0
00506         "lea (%0, %1), %%"REG_a"                \n\t"
00507         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00508 //      0       1       2       3       4       5       6       7
00509 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
00510 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
00511 
00512 
00513         "movq (%0, %1, 2), %%mm0                \n\t" // l2
00514         "movq (%0), %%mm1                       \n\t" // l0
00515         "movq %%mm0, %%mm2                      \n\t" // l2
00516         PAVGB(%%mm7, %%mm0)                           // ~l2/2
00517         PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
00518         PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
00519 
00520         "movq (%%"REG_a"), %%mm1                \n\t" // l1
00521         "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
00522         "movq %%mm1, %%mm4                      \n\t" // l1
00523         PAVGB(%%mm7, %%mm1)                           // ~l1/2
00524         PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
00525         PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
00526 
00527         "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
00528         "psubusb %%mm1, %%mm0                   \n\t"
00529         "psubusb %%mm4, %%mm1                   \n\t"
00530         "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
00531 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
00532 
00533         "movq (%0, %1, 4), %%mm0                \n\t" // l4
00534         "movq %%mm0, %%mm4                      \n\t" // l4
00535         PAVGB(%%mm7, %%mm0)                           // ~l4/2
00536         PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
00537         PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
00538 
00539         "movq (%%"REG_c"), %%mm2                \n\t" // l5
00540         "movq %%mm3, %%mm5                      \n\t" // l3
00541         PAVGB(%%mm7, %%mm3)                           // ~l3/2
00542         PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
00543         PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
00544 
00545         "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
00546         "psubusb %%mm3, %%mm0                   \n\t"
00547         "psubusb %%mm6, %%mm3                   \n\t"
00548         "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
00549         "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
00550 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
00551 
00552         "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
00553         "movq %%mm6, %%mm5                      \n\t" // l6
00554         PAVGB(%%mm7, %%mm6)                           // ~l6/2
00555         PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
00556         PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
00557 
00558         "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
00559         "movq %%mm2, %%mm4                      \n\t" // l5
00560         PAVGB(%%mm7, %%mm2)                           // ~l5/2
00561         PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
00562         PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
00563 
00564         "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
00565         "psubusb %%mm2, %%mm6                   \n\t"
00566         "psubusb %%mm4, %%mm2                   \n\t"
00567         "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
00568 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
00569 
00570 
00571         PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
00572         "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
00573         "paddusb "MANGLE(b01)", %%mm4           \n\t"
00574         "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
00575         "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
00576         "pand %%mm4, %%mm3                      \n\t"
00577 
00578         "movq %%mm3, %%mm1                      \n\t"
00579 //        "psubusb "MANGLE(b01)", %%mm3           \n\t"
00580         PAVGB(%%mm7, %%mm3)
00581         PAVGB(%%mm7, %%mm3)
00582         "paddusb %%mm1, %%mm3                   \n\t"
00583 //        "paddusb "MANGLE(b01)", %%mm3           \n\t"
00584 
00585         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
00586         "movq (%0, %1, 4), %%mm5                \n\t" //l4
00587         "movq (%0, %1, 4), %%mm4                \n\t" //l4
00588         "psubusb %%mm6, %%mm5                   \n\t"
00589         "psubusb %%mm4, %%mm6                   \n\t"
00590         "por %%mm6, %%mm5                       \n\t" // |l3-l4|
00591         "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
00592         "pxor %%mm6, %%mm0                      \n\t"
00593         "pand %%mm0, %%mm3                      \n\t"
00594         PMINUB(%%mm5, %%mm3, %%mm0)
00595 
00596         "psubusb "MANGLE(b01)", %%mm3           \n\t"
00597         PAVGB(%%mm7, %%mm3)
00598 
00599         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00600         "movq (%0, %1, 4), %%mm2                \n\t"
00601         "pxor %%mm6, %%mm0                      \n\t"
00602         "pxor %%mm6, %%mm2                      \n\t"
00603         "psubb %%mm3, %%mm0                     \n\t"
00604         "paddb %%mm3, %%mm2                     \n\t"
00605         "pxor %%mm6, %%mm0                      \n\t"
00606         "pxor %%mm6, %%mm2                      \n\t"
00607         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
00608         "movq %%mm2, (%0, %1, 4)                \n\t"
00609 #endif //0
00610 
00611         "lea (%0, %1), %%"REG_a"                \n\t"
00612         "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
00613 //      0       1       2       3       4       5       6       7
00614 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
00615 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
00616 
00617 
00618         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
00619         "movq (%0, %1, 4), %%mm0                \n\t" // l4
00620         "pxor %%mm6, %%mm1                      \n\t" // -l3-1
00621         PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
00622 // mm1=-l3-1, mm0=128-q
00623 
00624         "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
00625         "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
00626         "pxor %%mm6, %%mm2                      \n\t" // -l5-1
00627         "movq %%mm2, %%mm5                      \n\t" // -l5-1
00628         "movq "MANGLE(b80)", %%mm4              \n\t" // 128
00629         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00630         PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
00631         PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
00632         PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
00633         PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
00634 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
00635 
00636         "movq (%%"REG_a"), %%mm2                \n\t" // l1
00637         "pxor %%mm6, %%mm2                      \n\t" // -l1-1
00638         PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
00639         PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
00640         "movq "MANGLE(b80)", %%mm3              \n\t" // 128
00641         PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
00642         PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
00643         PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
00644 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
00645 
00646         PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
00647         "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
00648         "pxor %%mm6, %%mm1                      \n\t" // -l7-1
00649         PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
00650         "movq "MANGLE(b80)", %%mm2              \n\t" // 128
00651         PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
00652         PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
00653         PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
00654 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
00655 
00656         "movq "MANGLE(b00)", %%mm1              \n\t" // 0
00657         "movq "MANGLE(b00)", %%mm5              \n\t" // 0
00658         "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
00659         "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
00660         PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
00661         PMAXUB(%%mm5, %%mm3)                          // 128 + |lenergy/16|
00662         PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
00663 
00664 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
00665 
00666         "movq "MANGLE(b00)", %%mm7              \n\t" // 0
00667         "movq %2, %%mm2                         \n\t" // QP
00668         PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
00669         "psubb %%mm6, %%mm2                     \n\t"
00670 
00671         "movq %%mm4, %%mm1                      \n\t"
00672         "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
00673         "pxor %%mm1, %%mm4                      \n\t"
00674         "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
00675         "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
00676         "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
00677 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
00678 
00679         "movq %%mm4, %%mm3                      \n\t" // d
00680         "psubusb "MANGLE(b01)", %%mm4           \n\t"
00681         PAVGB(%%mm7, %%mm4)                           // d/32
00682         PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
00683         "paddb %%mm3, %%mm4                     \n\t" // 5d/64
00684         "pand %%mm2, %%mm4                      \n\t"
00685 
00686         "movq "MANGLE(b80)", %%mm5              \n\t" // 128
00687         "psubb %%mm0, %%mm5                     \n\t" // q
00688         "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
00689         "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
00690         "pxor %%mm7, %%mm5                      \n\t"
00691 
00692         PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
00693         "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
00694 
00695         "pand %%mm7, %%mm4                      \n\t"
00696         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00697         "movq (%0, %1, 4), %%mm2                \n\t"
00698         "pxor %%mm1, %%mm0                      \n\t"
00699         "pxor %%mm1, %%mm2                      \n\t"
00700         "paddb %%mm4, %%mm0                     \n\t"
00701         "psubb %%mm4, %%mm2                     \n\t"
00702         "pxor %%mm1, %%mm0                      \n\t"
00703         "pxor %%mm1, %%mm2                      \n\t"
00704         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
00705         "movq %%mm2, (%0, %1, 4)                \n\t"
00706 
00707         :
00708         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00709         : "%"REG_a, "%"REG_c
00710     );
00711 
00712 /*
00713     {
00714     int x;
00715     src-= stride;
00716     for(x=0; x<BLOCK_SIZE; x++){
00717         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
00718         if(FFABS(middleEnergy)< 8*QP){
00719             const int q=(src[l4] - src[l5])/2;
00720             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
00721             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
00722 
00723             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
00724             d= FFMAX(d, 0);
00725 
00726             d= (5*d + 32) >> 6;
00727             d*= FFSIGN(-middleEnergy);
00728 
00729             if(q>0){
00730                 d= d<0 ? 0 : d;
00731                 d= d>q ? q : d;
00732             }else{
00733                 d= d>0 ? 0 : d;
00734                 d= d<q ? q : d;
00735             }
00736 
00737             src[l4]-= d;
00738             src[l5]+= d;
00739         }
00740         src++;
00741     }
00742     src-=8;
00743     for(x=0; x<8; x++){
00744         int y;
00745         for(y=4; y<6; y++){
00746             int d= src[x+y*stride] - tmp[x+(y-4)*8];
00747             int ad= FFABS(d);
00748             static int max=0;
00749             static int sum=0;
00750             static int num=0;
00751             static int bias=0;
00752 
00753             if(max<ad) max=ad;
00754             sum+= ad>3 ? 1 : 0;
00755             if(ad>3){
00756                 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
00757             }
00758             if(y==4) bias+=d;
00759             num++;
00760             if(num%1000000 == 0){
00761                 av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
00762             }
00763         }
00764     }
00765 }
00766 */
00767 #elif HAVE_MMX
00768     DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars
00769     src+= stride*4;
00770     __asm__ volatile(
00771         "pxor %%mm7, %%mm7                      \n\t"
00772 //      0       1       2       3       4       5       6       7
00773 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
00774 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1
00775 
00776         "movq (%0), %%mm0                       \n\t"
00777         "movq %%mm0, %%mm1                      \n\t"
00778         "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
00779         "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
00780 
00781         "movq (%0, %1), %%mm2                   \n\t"
00782         "lea (%0, %1, 2), %%"REG_a"             \n\t"
00783         "movq %%mm2, %%mm3                      \n\t"
00784         "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
00785         "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
00786 
00787         "movq (%%"REG_a"), %%mm4                \n\t"
00788         "movq %%mm4, %%mm5                      \n\t"
00789         "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
00790         "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
00791 
00792         "paddw %%mm0, %%mm0                     \n\t" // 2L0
00793         "paddw %%mm1, %%mm1                     \n\t" // 2H0
00794         "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
00795         "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
00796         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
00797         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
00798 
00799         "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
00800         "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
00801         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
00802         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
00803 
00804         "movq (%%"REG_a", %1), %%mm2            \n\t"
00805         "movq %%mm2, %%mm3                      \n\t"
00806         "punpcklbw %%mm7, %%mm2                 \n\t" // L3
00807         "punpckhbw %%mm7, %%mm3                 \n\t" // H3
00808 
00809         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
00810         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
00811         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00812         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00813         "movq %%mm0, (%3)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00814         "movq %%mm1, 8(%3)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00815 
00816         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00817         "movq %%mm0, %%mm1                      \n\t"
00818         "punpcklbw %%mm7, %%mm0                 \n\t" // L4
00819         "punpckhbw %%mm7, %%mm1                 \n\t" // H4
00820 
00821         "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
00822         "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
00823         "movq %%mm2, 16(%3)                     \n\t" // L3 - L4
00824         "movq %%mm3, 24(%3)                     \n\t" // H3 - H4
00825         "paddw %%mm4, %%mm4                     \n\t" // 2L2
00826         "paddw %%mm5, %%mm5                     \n\t" // 2H2
00827         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
00828         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
00829 
00830         "lea (%%"REG_a", %1), %0                \n\t"
00831         "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
00832         "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
00833         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
00834         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
00835 //50 opcodes so far
00836         "movq (%0, %1, 2), %%mm2                \n\t"
00837         "movq %%mm2, %%mm3                      \n\t"
00838         "punpcklbw %%mm7, %%mm2                 \n\t" // L5
00839         "punpckhbw %%mm7, %%mm3                 \n\t" // H5
00840         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
00841         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
00842         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
00843         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
00844 
00845         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
00846         "punpcklbw %%mm7, %%mm6                 \n\t" // L6
00847         "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
00848         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
00849         "punpckhbw %%mm7, %%mm6                 \n\t" // H6
00850         "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
00851 
00852         "paddw %%mm0, %%mm0                     \n\t" // 2L4
00853         "paddw %%mm1, %%mm1                     \n\t" // 2H4
00854         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
00855         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
00856 
00857         "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
00858         "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
00859         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
00860         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
00861 
00862         "movq (%0, %1, 4), %%mm2                \n\t"
00863         "movq %%mm2, %%mm3                      \n\t"
00864         "punpcklbw %%mm7, %%mm2                 \n\t" // L7
00865         "punpckhbw %%mm7, %%mm3                 \n\t" // H7
00866 
00867         "paddw %%mm2, %%mm2                     \n\t" // 2L7
00868         "paddw %%mm3, %%mm3                     \n\t" // 2H7
00869         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
00870         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
00871 
00872         "movq (%3), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00873         "movq 8(%3), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00874 
00875 #if HAVE_MMX2
00876         "movq %%mm7, %%mm6                      \n\t" // 0
00877         "psubw %%mm0, %%mm6                     \n\t"
00878         "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
00879         "movq %%mm7, %%mm6                      \n\t" // 0
00880         "psubw %%mm1, %%mm6                     \n\t"
00881         "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
00882         "movq %%mm7, %%mm6                      \n\t" // 0
00883         "psubw %%mm2, %%mm6                     \n\t"
00884         "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
00885         "movq %%mm7, %%mm6                      \n\t" // 0
00886         "psubw %%mm3, %%mm6                     \n\t"
00887         "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
00888 #else
00889         "movq %%mm7, %%mm6                      \n\t" // 0
00890         "pcmpgtw %%mm0, %%mm6                   \n\t"
00891         "pxor %%mm6, %%mm0                      \n\t"
00892         "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
00893         "movq %%mm7, %%mm6                      \n\t" // 0
00894         "pcmpgtw %%mm1, %%mm6                   \n\t"
00895         "pxor %%mm6, %%mm1                      \n\t"
00896         "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
00897         "movq %%mm7, %%mm6                      \n\t" // 0
00898         "pcmpgtw %%mm2, %%mm6                   \n\t"
00899         "pxor %%mm6, %%mm2                      \n\t"
00900         "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
00901         "movq %%mm7, %%mm6                      \n\t" // 0
00902         "pcmpgtw %%mm3, %%mm6                   \n\t"
00903         "pxor %%mm6, %%mm3                      \n\t"
00904         "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
00905 #endif
00906 
00907 #if HAVE_MMX2
00908         "pminsw %%mm2, %%mm0                    \n\t"
00909         "pminsw %%mm3, %%mm1                    \n\t"
00910 #else
00911         "movq %%mm0, %%mm6                      \n\t"
00912         "psubusw %%mm2, %%mm6                   \n\t"
00913         "psubw %%mm6, %%mm0                     \n\t"
00914         "movq %%mm1, %%mm6                      \n\t"
00915         "psubusw %%mm3, %%mm6                   \n\t"
00916         "psubw %%mm6, %%mm1                     \n\t"
00917 #endif
00918 
00919         "movd %2, %%mm2                         \n\t" // QP
00920         "punpcklbw %%mm7, %%mm2                 \n\t"
00921 
00922         "movq %%mm7, %%mm6                      \n\t" // 0
00923         "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
00924         "pxor %%mm6, %%mm4                      \n\t"
00925         "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
00926         "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
00927         "pxor %%mm7, %%mm5                      \n\t"
00928         "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
00929 // 100 opcodes
00930         "psllw $3, %%mm2                        \n\t" // 8QP
00931         "movq %%mm2, %%mm3                      \n\t" // 8QP
00932         "pcmpgtw %%mm4, %%mm2                   \n\t"
00933         "pcmpgtw %%mm5, %%mm3                   \n\t"
00934         "pand %%mm2, %%mm4                      \n\t"
00935         "pand %%mm3, %%mm5                      \n\t"
00936 
00937 
00938         "psubusw %%mm0, %%mm4                   \n\t" // hd
00939         "psubusw %%mm1, %%mm5                   \n\t" // ld
00940 
00941 
00942         "movq "MANGLE(w05)", %%mm2              \n\t" // 5
00943         "pmullw %%mm2, %%mm4                    \n\t"
00944         "pmullw %%mm2, %%mm5                    \n\t"
00945         "movq "MANGLE(w20)", %%mm2              \n\t" // 32
00946         "paddw %%mm2, %%mm4                     \n\t"
00947         "paddw %%mm2, %%mm5                     \n\t"
00948         "psrlw $6, %%mm4                        \n\t"
00949         "psrlw $6, %%mm5                        \n\t"
00950 
00951         "movq 16(%3), %%mm0                     \n\t" // L3 - L4
00952         "movq 24(%3), %%mm1                     \n\t" // H3 - H4
00953 
00954         "pxor %%mm2, %%mm2                      \n\t"
00955         "pxor %%mm3, %%mm3                      \n\t"
00956 
00957         "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
00958         "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
00959         "pxor %%mm2, %%mm0                      \n\t"
00960         "pxor %%mm3, %%mm1                      \n\t"
00961         "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
00962         "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
00963         "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
00964         "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
00965 
00966         "pxor %%mm6, %%mm2                      \n\t"
00967         "pxor %%mm7, %%mm3                      \n\t"
00968         "pand %%mm2, %%mm4                      \n\t"
00969         "pand %%mm3, %%mm5                      \n\t"
00970 
00971 #if HAVE_MMX2
00972         "pminsw %%mm0, %%mm4                    \n\t"
00973         "pminsw %%mm1, %%mm5                    \n\t"
00974 #else
00975         "movq %%mm4, %%mm2                      \n\t"
00976         "psubusw %%mm0, %%mm2                   \n\t"
00977         "psubw %%mm2, %%mm4                     \n\t"
00978         "movq %%mm5, %%mm2                      \n\t"
00979         "psubusw %%mm1, %%mm2                   \n\t"
00980         "psubw %%mm2, %%mm5                     \n\t"
00981 #endif
00982         "pxor %%mm6, %%mm4                      \n\t"
00983         "pxor %%mm7, %%mm5                      \n\t"
00984         "psubw %%mm6, %%mm4                     \n\t"
00985         "psubw %%mm7, %%mm5                     \n\t"
00986         "packsswb %%mm5, %%mm4                  \n\t"
00987         "movq (%0), %%mm0                       \n\t"
00988         "paddb   %%mm4, %%mm0                   \n\t"
00989         "movq %%mm0, (%0)                       \n\t"
00990         "movq (%0, %1), %%mm0                   \n\t"
00991         "psubb %%mm4, %%mm0                     \n\t"
00992         "movq %%mm0, (%0, %1)                   \n\t"
00993 
00994         : "+r" (src)
00995         : "r" ((x86_reg)stride), "m" (c->pQPb), "r"(tmp)
00996         : "%"REG_a
00997     );
00998 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00999     const int l1= stride;
01000     const int l2= stride + l1;
01001     const int l3= stride + l2;
01002     const int l4= stride + l3;
01003     const int l5= stride + l4;
01004     const int l6= stride + l5;
01005     const int l7= stride + l6;
01006     const int l8= stride + l7;
01007 //    const int l9= stride + l8;
01008     int x;
01009     src+= stride*3;
01010     for(x=0; x<BLOCK_SIZE; x++){
01011         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
01012         if(FFABS(middleEnergy) < 8*c->QP){
01013             const int q=(src[l4] - src[l5])/2;
01014             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
01015             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
01016 
01017             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
01018             d= FFMAX(d, 0);
01019 
01020             d= (5*d + 32) >> 6;
01021             d*= FFSIGN(-middleEnergy);
01022 
01023             if(q>0){
01024                 d= d<0 ? 0 : d;
01025                 d= d>q ? q : d;
01026             }else{
01027                 d= d>0 ? 0 : d;
01028                 d= d<q ? q : d;
01029             }
01030 
01031             src[l4]-= d;
01032             src[l5]+= d;
01033         }
01034         src++;
01035     }
01036 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01037 }
01038 #endif //HAVE_ALTIVEC
01039 
01040 #if !HAVE_ALTIVEC
01041 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
01042 {
01043 #if HAVE_MMX2 || HAVE_AMD3DNOW
01044     DECLARE_ALIGNED(8, uint64_t, tmp)[3];
01045     __asm__ volatile(
01046         "pxor %%mm6, %%mm6                      \n\t"
01047         "pcmpeqb %%mm7, %%mm7                   \n\t"
01048         "movq %2, %%mm0                         \n\t"
01049         "punpcklbw %%mm6, %%mm0                 \n\t"
01050         "psrlw $1, %%mm0                        \n\t"
01051         "psubw %%mm7, %%mm0                     \n\t"
01052         "packuswb %%mm0, %%mm0                  \n\t"
01053         "movq %%mm0, %3                         \n\t"
01054 
01055         "lea (%0, %1), %%"REG_a"                \n\t"
01056         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01057 
01058 //        0        1        2        3        4        5        6        7        8        9
01059 //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1
01060 
01061 #undef REAL_FIND_MIN_MAX
01062 #undef FIND_MIN_MAX
01063 #if HAVE_MMX2
01064 #define REAL_FIND_MIN_MAX(addr)\
01065         "movq " #addr ", %%mm0                  \n\t"\
01066         "pminub %%mm0, %%mm7                    \n\t"\
01067         "pmaxub %%mm0, %%mm6                    \n\t"
01068 #else
01069 #define REAL_FIND_MIN_MAX(addr)\
01070         "movq " #addr ", %%mm0                  \n\t"\
01071         "movq %%mm7, %%mm1                      \n\t"\
01072         "psubusb %%mm0, %%mm6                   \n\t"\
01073         "paddb %%mm0, %%mm6                     \n\t"\
01074         "psubusb %%mm0, %%mm1                   \n\t"\
01075         "psubb %%mm1, %%mm7                     \n\t"
01076 #endif
01077 #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
01078 
01079 FIND_MIN_MAX((%%REGa))
01080 FIND_MIN_MAX((%%REGa, %1))
01081 FIND_MIN_MAX((%%REGa, %1, 2))
01082 FIND_MIN_MAX((%0, %1, 4))
01083 FIND_MIN_MAX((%%REGd))
01084 FIND_MIN_MAX((%%REGd, %1))
01085 FIND_MIN_MAX((%%REGd, %1, 2))
01086 FIND_MIN_MAX((%0, %1, 8))
01087 
01088         "movq %%mm7, %%mm4                      \n\t"
01089         "psrlq $8, %%mm7                        \n\t"
01090 #if HAVE_MMX2
01091         "pminub %%mm4, %%mm7                    \n\t" // min of pixels
01092         "pshufw $0xF9, %%mm7, %%mm4             \n\t"
01093         "pminub %%mm4, %%mm7                    \n\t" // min of pixels
01094         "pshufw $0xFE, %%mm7, %%mm4             \n\t"
01095         "pminub %%mm4, %%mm7                    \n\t"
01096 #else
01097         "movq %%mm7, %%mm1                      \n\t"
01098         "psubusb %%mm4, %%mm1                   \n\t"
01099         "psubb %%mm1, %%mm7                     \n\t"
01100         "movq %%mm7, %%mm4                      \n\t"
01101         "psrlq $16, %%mm7                       \n\t"
01102         "movq %%mm7, %%mm1                      \n\t"
01103         "psubusb %%mm4, %%mm1                   \n\t"
01104         "psubb %%mm1, %%mm7                     \n\t"
01105         "movq %%mm7, %%mm4                      \n\t"
01106         "psrlq $32, %%mm7                       \n\t"
01107         "movq %%mm7, %%mm1                      \n\t"
01108         "psubusb %%mm4, %%mm1                   \n\t"
01109         "psubb %%mm1, %%mm7                     \n\t"
01110 #endif
01111 
01112 
01113         "movq %%mm6, %%mm4                      \n\t"
01114         "psrlq $8, %%mm6                        \n\t"
01115 #if HAVE_MMX2
01116         "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
01117         "pshufw $0xF9, %%mm6, %%mm4             \n\t"
01118         "pmaxub %%mm4, %%mm6                    \n\t"
01119         "pshufw $0xFE, %%mm6, %%mm4             \n\t"
01120         "pmaxub %%mm4, %%mm6                    \n\t"
01121 #else
01122         "psubusb %%mm4, %%mm6                   \n\t"
01123         "paddb %%mm4, %%mm6                     \n\t"
01124         "movq %%mm6, %%mm4                      \n\t"
01125         "psrlq $16, %%mm6                       \n\t"
01126         "psubusb %%mm4, %%mm6                   \n\t"
01127         "paddb %%mm4, %%mm6                     \n\t"
01128         "movq %%mm6, %%mm4                      \n\t"
01129         "psrlq $32, %%mm6                       \n\t"
01130         "psubusb %%mm4, %%mm6                   \n\t"
01131         "paddb %%mm4, %%mm6                     \n\t"
01132 #endif
01133         "movq %%mm6, %%mm0                      \n\t" // max
01134         "psubb %%mm7, %%mm6                     \n\t" // max - min
01135         "push %4                              \n\t"
01136         "movd %%mm6, %k4                        \n\t"
01137         "cmpb "MANGLE(deringThreshold)", %b4    \n\t"
01138         "pop %4                               \n\t"
01139         " jb 1f                                 \n\t"
01140         PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
01141         "punpcklbw %%mm7, %%mm7                 \n\t"
01142         "punpcklbw %%mm7, %%mm7                 \n\t"
01143         "punpcklbw %%mm7, %%mm7                 \n\t"
01144         "movq %%mm7, (%4)                       \n\t"
01145 
01146         "movq (%0), %%mm0                       \n\t" // L10
01147         "movq %%mm0, %%mm1                      \n\t" // L10
01148         "movq %%mm0, %%mm2                      \n\t" // L10
01149         "psllq $8, %%mm1                        \n\t"
01150         "psrlq $8, %%mm2                        \n\t"
01151         "movd -4(%0), %%mm3                     \n\t"
01152         "movd 8(%0), %%mm4                      \n\t"
01153         "psrlq $24, %%mm3                       \n\t"
01154         "psllq $56, %%mm4                       \n\t"
01155         "por %%mm3, %%mm1                       \n\t" // L00
01156         "por %%mm4, %%mm2                       \n\t" // L20
01157         "movq %%mm1, %%mm3                      \n\t" // L00
01158         PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
01159         PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
01160         "psubusb %%mm7, %%mm0                   \n\t"
01161         "psubusb %%mm7, %%mm2                   \n\t"
01162         "psubusb %%mm7, %%mm3                   \n\t"
01163         "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
01164         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
01165         "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
01166         "paddb %%mm2, %%mm0                     \n\t"
01167         "paddb %%mm3, %%mm0                     \n\t"
01168 
01169         "movq (%%"REG_a"), %%mm2                \n\t" // L11
01170         "movq %%mm2, %%mm3                      \n\t" // L11
01171         "movq %%mm2, %%mm4                      \n\t" // L11
01172         "psllq $8, %%mm3                        \n\t"
01173         "psrlq $8, %%mm4                        \n\t"
01174         "movd -4(%%"REG_a"), %%mm5              \n\t"
01175         "movd 8(%%"REG_a"), %%mm6               \n\t"
01176         "psrlq $24, %%mm5                       \n\t"
01177         "psllq $56, %%mm6                       \n\t"
01178         "por %%mm5, %%mm3                       \n\t" // L01
01179         "por %%mm6, %%mm4                       \n\t" // L21
01180         "movq %%mm3, %%mm5                      \n\t" // L01
01181         PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
01182         PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
01183         "psubusb %%mm7, %%mm2                   \n\t"
01184         "psubusb %%mm7, %%mm4                   \n\t"
01185         "psubusb %%mm7, %%mm5                   \n\t"
01186         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
01187         "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
01188         "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
01189         "paddb %%mm4, %%mm2                     \n\t"
01190         "paddb %%mm5, %%mm2                     \n\t"
01191 // 0, 2, 3, 1
01192 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
01193         "movq " #src ", " #sx "                 \n\t" /* src[0] */\
01194         "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
01195         "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
01196         "psllq $8, " #lx "                      \n\t"\
01197         "psrlq $8, " #t0 "                      \n\t"\
01198         "movd -4" #src ", " #t1 "               \n\t"\
01199         "psrlq $24, " #t1 "                     \n\t"\
01200         "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
01201         "movd 8" #src ", " #t1 "                \n\t"\
01202         "psllq $56, " #t1 "                     \n\t"\
01203         "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
01204         "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
01205         PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
01206         PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
01207         PAVGB(lx, pplx)                                     \
01208         "movq " #lx ", 8(%4)                    \n\t"\
01209         "movq (%4), " #lx "                     \n\t"\
01210         "psubusb " #lx ", " #t1 "               \n\t"\
01211         "psubusb " #lx ", " #t0 "               \n\t"\
01212         "psubusb " #lx ", " #sx "               \n\t"\
01213         "movq "MANGLE(b00)", " #lx "            \n\t"\
01214         "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
01215         "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
01216         "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
01217         "paddb " #t1 ", " #t0 "                 \n\t"\
01218         "paddb " #t0 ", " #sx "                 \n\t"\
01219 \
01220         PAVGB(plx, pplx)                              /* filtered */\
01221         "movq " #dst ", " #t0 "                 \n\t" /* dst */\
01222         "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
01223         "psubusb %3, " #t0 "                    \n\t"\
01224         "paddusb %3, " #t1 "                    \n\t"\
01225         PMAXUB(t0, pplx)\
01226         PMINUB(t1, pplx, t0)\
01227         "paddb " #sx ", " #ppsx "               \n\t"\
01228         "paddb " #psx ", " #ppsx "              \n\t"\
01229         "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
01230         "pand "MANGLE(b08)", " #ppsx "          \n\t"\
01231         "pcmpeqb " #lx ", " #ppsx "             \n\t"\
01232         "pand " #ppsx ", " #pplx "              \n\t"\
01233         "pandn " #dst ", " #ppsx "              \n\t"\
01234         "por " #pplx ", " #ppsx "               \n\t"\
01235         "movq " #ppsx ", " #dst "               \n\t"\
01236         "movq 8(%4), " #lx "                    \n\t"
01237 
01238 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
01239    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
01240 /*
01241 0000000
01242 1111111
01243 
01244 1111110
01245 1111101
01246 1111100
01247 1111011
01248 1111010
01249 1111001
01250 
01251 1111000
01252 1110111
01253 
01254 */
01255 //DERING_CORE(dst          ,src            ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)
01256 DERING_CORE((%%REGa)       ,(%%REGa, %1)   ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01257 DERING_CORE((%%REGa, %1)   ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01258 DERING_CORE((%%REGa, %1, 2),(%0, %1, 4)    ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
01259 DERING_CORE((%0, %1, 4)    ,(%%REGd)       ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01260 DERING_CORE((%%REGd)       ,(%%REGd, %1)   ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01261 DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
01262 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01263 DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01264 
01265         "1:                        \n\t"
01266         : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2), "q"(tmp)
01267         : "%"REG_a, "%"REG_d
01268     );
01269 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01270     int y;
01271     int min=255;
01272     int max=0;
01273     int avg;
01274     uint8_t *p;
01275     int s[10];
01276     const int QP2= c->QP/2 + 1;
01277 
01278     for(y=1; y<9; y++){
01279         int x;
01280         p= src + stride*y;
01281         for(x=1; x<9; x++){
01282             p++;
01283             if(*p > max) max= *p;
01284             if(*p < min) min= *p;
01285         }
01286     }
01287     avg= (min + max + 1)>>1;
01288 
01289     if(max - min <deringThreshold) return;
01290 
01291     for(y=0; y<10; y++){
01292         int t = 0;
01293 
01294         if(src[stride*y + 0] > avg) t+= 1;
01295         if(src[stride*y + 1] > avg) t+= 2;
01296         if(src[stride*y + 2] > avg) t+= 4;
01297         if(src[stride*y + 3] > avg) t+= 8;
01298         if(src[stride*y + 4] > avg) t+= 16;
01299         if(src[stride*y + 5] > avg) t+= 32;
01300         if(src[stride*y + 6] > avg) t+= 64;
01301         if(src[stride*y + 7] > avg) t+= 128;
01302         if(src[stride*y + 8] > avg) t+= 256;
01303         if(src[stride*y + 9] > avg) t+= 512;
01304 
01305         t |= (~t)<<16;
01306         t &= (t<<1) & (t>>1);
01307         s[y] = t;
01308     }
01309 
01310     for(y=1; y<9; y++){
01311         int t = s[y-1] & s[y] & s[y+1];
01312         t|= t>>16;
01313         s[y-1]= t;
01314     }
01315 
01316     for(y=1; y<9; y++){
01317         int x;
01318         int t = s[y-1];
01319 
01320         p= src + stride*y;
01321         for(x=1; x<9; x++){
01322             p++;
01323             if(t & (1<<x)){
01324                 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
01325                       +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
01326                       +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
01327                 f= (f + 8)>>4;
01328 
01329 #ifdef DEBUG_DERING_THRESHOLD
01330                     __asm__ volatile("emms\n\t":);
01331                     {
01332                     static long long numPixels=0;
01333                     if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
01334 //                    if((max-min)<20 || (max-min)*QP<200)
01335 //                    if((max-min)*QP < 500)
01336 //                    if(max-min<QP/2)
01337                     if(max-min < 20){
01338                         static int numSkipped=0;
01339                         static int errorSum=0;
01340                         static int worstQP=0;
01341                         static int worstRange=0;
01342                         static int worstDiff=0;
01343                         int diff= (f - *p);
01344                         int absDiff= FFABS(diff);
01345                         int error= diff*diff;
01346 
01347                         if(x==1 || x==8 || y==1 || y==8) continue;
01348 
01349                         numSkipped++;
01350                         if(absDiff > worstDiff){
01351                             worstDiff= absDiff;
01352                             worstQP= QP;
01353                             worstRange= max-min;
01354                         }
01355                         errorSum+= error;
01356 
01357                         if(1024LL*1024LL*1024LL % numSkipped == 0){
01358                             av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
01359                                    "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
01360                                    (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
01361                                    worstDiff, (float)numSkipped/numPixels);
01362                         }
01363                     }
01364                     }
01365 #endif
01366                     if     (*p + QP2 < f) *p= *p + QP2;
01367                     else if(*p - QP2 > f) *p= *p - QP2;
01368                     else *p=f;
01369             }
01370         }
01371     }
01372 #ifdef DEBUG_DERING_THRESHOLD
01373     if(max-min < 20){
01374         for(y=1; y<9; y++){
01375             int x;
01376             int t = 0;
01377             p= src + stride*y;
01378             for(x=1; x<9; x++){
01379                 p++;
01380                 *p = FFMIN(*p + 20, 255);
01381             }
01382         }
01383 //        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
01384     }
01385 #endif
01386 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01387 }
01388 #endif //HAVE_ALTIVEC
01389 
01396 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
01397 {
01398 #if HAVE_MMX2 || HAVE_AMD3DNOW
01399     src+= 4*stride;
01400     __asm__ volatile(
01401         "lea (%0, %1), %%"REG_a"                \n\t"
01402         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
01403 //      0       1       2       3       4       5       6       7       8       9
01404 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
01405 
01406         "movq (%0), %%mm0                       \n\t"
01407         "movq (%%"REG_a", %1), %%mm1            \n\t"
01408         PAVGB(%%mm1, %%mm0)
01409         "movq %%mm0, (%%"REG_a")                \n\t"
01410         "movq (%0, %1, 4), %%mm0                \n\t"
01411         PAVGB(%%mm0, %%mm1)
01412         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
01413         "movq (%%"REG_c", %1), %%mm1            \n\t"
01414         PAVGB(%%mm1, %%mm0)
01415         "movq %%mm0, (%%"REG_c")                \n\t"
01416         "movq (%0, %1, 8), %%mm0                \n\t"
01417         PAVGB(%%mm0, %%mm1)
01418         "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
01419 
01420         : : "r" (src), "r" ((x86_reg)stride)
01421         : "%"REG_a, "%"REG_c
01422     );
01423 #else
01424     int a, b, x;
01425     src+= 4*stride;
01426 
01427     for(x=0; x<2; x++){
01428         a= *(uint32_t*)&src[stride*0];
01429         b= *(uint32_t*)&src[stride*2];
01430         *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01431         a= *(uint32_t*)&src[stride*4];
01432         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01433         b= *(uint32_t*)&src[stride*6];
01434         *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01435         a= *(uint32_t*)&src[stride*8];
01436         *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01437         src += 4;
01438     }
01439 #endif
01440 }
01441 
01449 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
01450 {
01451 #if HAVE_MMX2 || HAVE_AMD3DNOW
01452     src+= stride*3;
01453     __asm__ volatile(
01454         "lea (%0, %1), %%"REG_a"                \n\t"
01455         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01456         "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
01457         "add %1, %%"REG_c"                      \n\t"
01458         "pxor %%mm7, %%mm7                      \n\t"
01459 //      0       1       2       3       4       5       6       7       8       9       10
01460 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01461 
01462 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
01463         "movq " #a ", %%mm0                     \n\t"\
01464         "movq " #b ", %%mm1                     \n\t"\
01465         "movq " #d ", %%mm2                     \n\t"\
01466         "movq " #e ", %%mm3                     \n\t"\
01467         PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
01468         PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
01469         "movq %%mm0, %%mm2                      \n\t"\
01470         "punpcklbw %%mm7, %%mm0                 \n\t"\
01471         "punpckhbw %%mm7, %%mm2                 \n\t"\
01472         "movq %%mm1, %%mm3                      \n\t"\
01473         "punpcklbw %%mm7, %%mm1                 \n\t"\
01474         "punpckhbw %%mm7, %%mm3                 \n\t"\
01475         "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
01476         "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
01477         "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
01478         "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
01479         "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
01480         "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
01481         "packuswb %%mm3, %%mm1                  \n\t"\
01482         "movq %%mm1, " #c "                     \n\t"
01483 #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)
01484 
01485 DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
01486 DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))
01487 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
01488 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
01489 
01490         : : "r" (src), "r" ((x86_reg)stride)
01491         : "%"REG_a, "%"REG_d, "%"REG_c
01492     );
01493 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01494     int x;
01495     src+= stride*3;
01496     for(x=0; x<8; x++){
01497         src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
01498         src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
01499         src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
01500         src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
01501         src++;
01502     }
01503 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01504 }
01505 
01513 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
01514 {
01515 #if HAVE_MMX2 || HAVE_AMD3DNOW
01516     src+= stride*4;
01517     __asm__ volatile(
01518         "lea (%0, %1), %%"REG_a"                \n\t"
01519         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01520         "pxor %%mm7, %%mm7                      \n\t"
01521         "movq (%2), %%mm0                       \n\t"
01522 //      0       1       2       3       4       5       6       7       8       9       10
01523 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01524 
01525 #define REAL_DEINT_FF(a,b,c,d)\
01526         "movq " #a ", %%mm1                     \n\t"\
01527         "movq " #b ", %%mm2                     \n\t"\
01528         "movq " #c ", %%mm3                     \n\t"\
01529         "movq " #d ", %%mm4                     \n\t"\
01530         PAVGB(%%mm3, %%mm1)                          \
01531         PAVGB(%%mm4, %%mm0)                          \
01532         "movq %%mm0, %%mm3                      \n\t"\
01533         "punpcklbw %%mm7, %%mm0                 \n\t"\
01534         "punpckhbw %%mm7, %%mm3                 \n\t"\
01535         "movq %%mm1, %%mm4                      \n\t"\
01536         "punpcklbw %%mm7, %%mm1                 \n\t"\
01537         "punpckhbw %%mm7, %%mm4                 \n\t"\
01538         "psllw $2, %%mm1                        \n\t"\
01539         "psllw $2, %%mm4                        \n\t"\
01540         "psubw %%mm0, %%mm1                     \n\t"\
01541         "psubw %%mm3, %%mm4                     \n\t"\
01542         "movq %%mm2, %%mm5                      \n\t"\
01543         "movq %%mm2, %%mm0                      \n\t"\
01544         "punpcklbw %%mm7, %%mm2                 \n\t"\
01545         "punpckhbw %%mm7, %%mm5                 \n\t"\
01546         "paddw %%mm2, %%mm1                     \n\t"\
01547         "paddw %%mm5, %%mm4                     \n\t"\
01548         "psraw $2, %%mm1                        \n\t"\
01549         "psraw $2, %%mm4                        \n\t"\
01550         "packuswb %%mm4, %%mm1                  \n\t"\
01551         "movq %%mm1, " #b "                     \n\t"\
01552 
01553 #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)
01554 
01555 DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))
01556 DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
01557 DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
01558 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
01559 
01560         "movq %%mm0, (%2)                       \n\t"
01561         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)
01562         : "%"REG_a, "%"REG_d
01563     );
01564 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01565     int x;
01566     src+= stride*4;
01567     for(x=0; x<8; x++){
01568         int t1= tmp[x];
01569         int t2= src[stride*1];
01570 
01571         src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
01572         t1= src[stride*4];
01573         src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
01574         t2= src[stride*6];
01575         src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
01576         t1= src[stride*8];
01577         src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
01578         tmp[x]= t1;
01579 
01580         src++;
01581     }
01582 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01583 }
01584 
01592 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
01593 {
01594 #if HAVE_MMX2 || HAVE_AMD3DNOW
01595     src+= stride*4;
01596     __asm__ volatile(
01597         "lea (%0, %1), %%"REG_a"                \n\t"
01598         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01599         "pxor %%mm7, %%mm7                      \n\t"
01600         "movq (%2), %%mm0                       \n\t"
01601         "movq (%3), %%mm1                       \n\t"
01602 //      0       1       2       3       4       5       6       7       8       9       10
01603 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01604 
01605 #define REAL_DEINT_L5(t1,t2,a,b,c)\
01606         "movq " #a ", %%mm2                     \n\t"\
01607         "movq " #b ", %%mm3                     \n\t"\
01608         "movq " #c ", %%mm4                     \n\t"\
01609         PAVGB(t2, %%mm3)                             \
01610         PAVGB(t1, %%mm4)                             \
01611         "movq %%mm2, %%mm5                      \n\t"\
01612         "movq %%mm2, " #t1 "                    \n\t"\
01613         "punpcklbw %%mm7, %%mm2                 \n\t"\
01614         "punpckhbw %%mm7, %%mm5                 \n\t"\
01615         "movq %%mm2, %%mm6                      \n\t"\
01616         "paddw %%mm2, %%mm2                     \n\t"\
01617         "paddw %%mm6, %%mm2                     \n\t"\
01618         "movq %%mm5, %%mm6                      \n\t"\
01619         "paddw %%mm5, %%mm5                     \n\t"\
01620         "paddw %%mm6, %%mm5                     \n\t"\
01621         "movq %%mm3, %%mm6                      \n\t"\
01622         "punpcklbw %%mm7, %%mm3                 \n\t"\
01623         "punpckhbw %%mm7, %%mm6                 \n\t"\
01624         "paddw %%mm3, %%mm3                     \n\t"\
01625         "paddw %%mm6, %%mm6                     \n\t"\
01626         "paddw %%mm3, %%mm2                     \n\t"\
01627         "paddw %%mm6, %%mm5                     \n\t"\
01628         "movq %%mm4, %%mm6                      \n\t"\
01629         "punpcklbw %%mm7, %%mm4                 \n\t"\
01630         "punpckhbw %%mm7, %%mm6                 \n\t"\
01631         "psubw %%mm4, %%mm2                     \n\t"\
01632         "psubw %%mm6, %%mm5                     \n\t"\
01633         "psraw $2, %%mm2                        \n\t"\
01634         "psraw $2, %%mm5                        \n\t"\
01635         "packuswb %%mm5, %%mm2                  \n\t"\
01636         "movq %%mm2, " #a "                     \n\t"\
01637 
01638 #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)
01639 
01640 DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )
01641 DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))
01642 DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )
01643 DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )
01644 DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )
01645 DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
01646 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
01647 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
01648 
01649         "movq %%mm0, (%2)                       \n\t"
01650         "movq %%mm1, (%3)                       \n\t"
01651         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)
01652         : "%"REG_a, "%"REG_d
01653     );
01654 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01655     int x;
01656     src+= stride*4;
01657     for(x=0; x<8; x++){
01658         int t1= tmp[x];
01659         int t2= tmp2[x];
01660         int t3= src[0];
01661 
01662         src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
01663         t1= src[stride*1];
01664         src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
01665         t2= src[stride*2];
01666         src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
01667         t3= src[stride*3];
01668         src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
01669         t1= src[stride*4];
01670         src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
01671         t2= src[stride*5];
01672         src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
01673         t3= src[stride*6];
01674         src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
01675         t1= src[stride*7];
01676         src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
01677 
01678         tmp[x]= t3;
01679         tmp2[x]= t1;
01680 
01681         src++;
01682     }
01683 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01684 }
01685 
01693 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
01694 {
01695 #if HAVE_MMX2 || HAVE_AMD3DNOW
01696     src+= 4*stride;
01697     __asm__ volatile(
01698         "lea (%0, %1), %%"REG_a"                \n\t"
01699         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01700 //      0       1       2       3       4       5       6       7       8       9
01701 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01702 
01703         "movq (%2), %%mm0                       \n\t" // L0
01704         "movq (%%"REG_a"), %%mm1                \n\t" // L2
01705         PAVGB(%%mm1, %%mm0)                           // L0+L2
01706         "movq (%0), %%mm2                       \n\t" // L1
01707         PAVGB(%%mm2, %%mm0)
01708         "movq %%mm0, (%0)                       \n\t"
01709         "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
01710         PAVGB(%%mm0, %%mm2)                           // L1+L3
01711         PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
01712         "movq %%mm2, (%%"REG_a")                \n\t"
01713         "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
01714         PAVGB(%%mm2, %%mm1)                           // L2+L4
01715         PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
01716         "movq %%mm1, (%%"REG_a", %1)            \n\t"
01717         "movq (%0, %1, 4), %%mm1                \n\t" // L5
01718         PAVGB(%%mm1, %%mm0)                           // L3+L5
01719         PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
01720         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
01721         "movq (%%"REG_d"), %%mm0                \n\t" // L6
01722         PAVGB(%%mm0, %%mm2)                           // L4+L6
01723         PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
01724         "movq %%mm2, (%0, %1, 4)                \n\t"
01725         "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
01726         PAVGB(%%mm2, %%mm1)                           // L5+L7
01727         PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
01728         "movq %%mm1, (%%"REG_d")                \n\t"
01729         "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
01730         PAVGB(%%mm1, %%mm0)                           // L6+L8
01731         PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
01732         "movq %%mm0, (%%"REG_d", %1)            \n\t"
01733         "movq (%0, %1, 8), %%mm0                \n\t" // L9
01734         PAVGB(%%mm0, %%mm2)                           // L7+L9
01735         PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
01736         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
01737         "movq %%mm1, (%2)                       \n\t"
01738 
01739         : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
01740         : "%"REG_a, "%"REG_d
01741     );
01742 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01743     int a, b, c, x;
01744     src+= 4*stride;
01745 
01746     for(x=0; x<2; x++){
01747         a= *(uint32_t*)&tmp[stride*0];
01748         b= *(uint32_t*)&src[stride*0];
01749         c= *(uint32_t*)&src[stride*1];
01750         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01751         *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01752 
01753         a= *(uint32_t*)&src[stride*2];
01754         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01755         *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01756 
01757         b= *(uint32_t*)&src[stride*3];
01758         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
01759         *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
01760 
01761         c= *(uint32_t*)&src[stride*4];
01762         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01763         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01764 
01765         a= *(uint32_t*)&src[stride*5];
01766         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01767         *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01768 
01769         b= *(uint32_t*)&src[stride*6];
01770         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
01771         *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
01772 
01773         c= *(uint32_t*)&src[stride*7];
01774         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01775         *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01776 
01777         a= *(uint32_t*)&src[stride*8];
01778         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01779         *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01780 
01781         *(uint32_t*)&tmp[stride*0]= c;
01782         src += 4;
01783         tmp += 4;
01784     }
01785 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01786 }
01787 
01794 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
01795 {
01796 #if HAVE_MMX
01797     src+= 4*stride;
01798 #if HAVE_MMX2
01799     __asm__ volatile(
01800         "lea (%0, %1), %%"REG_a"                \n\t"
01801         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01802 //      0       1       2       3       4       5       6       7       8       9
01803 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01804 
01805         "movq (%0), %%mm0                       \n\t" //
01806         "movq (%%"REG_a", %1), %%mm2            \n\t" //
01807         "movq (%%"REG_a"), %%mm1                \n\t" //
01808         "movq %%mm0, %%mm3                      \n\t"
01809         "pmaxub %%mm1, %%mm0                    \n\t" //
01810         "pminub %%mm3, %%mm1                    \n\t" //
01811         "pmaxub %%mm2, %%mm1                    \n\t" //
01812         "pminub %%mm1, %%mm0                    \n\t"
01813         "movq %%mm0, (%%"REG_a")                \n\t"
01814 
01815         "movq (%0, %1, 4), %%mm0                \n\t" //
01816         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
01817         "movq %%mm2, %%mm3                      \n\t"
01818         "pmaxub %%mm1, %%mm2                    \n\t" //
01819         "pminub %%mm3, %%mm1                    \n\t" //
01820         "pmaxub %%mm0, %%mm1                    \n\t" //
01821         "pminub %%mm1, %%mm2                    \n\t"
01822         "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
01823 
01824         "movq (%%"REG_d"), %%mm2                \n\t" //
01825         "movq (%%"REG_d", %1), %%mm1            \n\t" //
01826         "movq %%mm2, %%mm3                      \n\t"
01827         "pmaxub %%mm0, %%mm2                    \n\t" //
01828         "pminub %%mm3, %%mm0                    \n\t" //
01829         "pmaxub %%mm1, %%mm0                    \n\t" //
01830         "pminub %%mm0, %%mm2                    \n\t"
01831         "movq %%mm2, (%%"REG_d")                \n\t"
01832 
01833         "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
01834         "movq (%0, %1, 8), %%mm0                \n\t" //
01835         "movq %%mm2, %%mm3                      \n\t"
01836         "pmaxub %%mm0, %%mm2                    \n\t" //
01837         "pminub %%mm3, %%mm0                    \n\t" //
01838         "pmaxub %%mm1, %%mm0                    \n\t" //
01839         "pminub %%mm0, %%mm2                    \n\t"
01840         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
01841 
01842 
01843         : : "r" (src), "r" ((x86_reg)stride)
01844         : "%"REG_a, "%"REG_d
01845     );
01846 
01847 #else // MMX without MMX2
01848     __asm__ volatile(
01849         "lea (%0, %1), %%"REG_a"                \n\t"
01850         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01851 //      0       1       2       3       4       5       6       7       8       9
01852 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01853         "pxor %%mm7, %%mm7                      \n\t"
01854 
01855 #define REAL_MEDIAN(a,b,c)\
01856         "movq " #a ", %%mm0                     \n\t"\
01857         "movq " #b ", %%mm2                     \n\t"\
01858         "movq " #c ", %%mm1                     \n\t"\
01859         "movq %%mm0, %%mm3                      \n\t"\
01860         "movq %%mm1, %%mm4                      \n\t"\
01861         "movq %%mm2, %%mm5                      \n\t"\
01862         "psubusb %%mm1, %%mm3                   \n\t"\
01863         "psubusb %%mm2, %%mm4                   \n\t"\
01864         "psubusb %%mm0, %%mm5                   \n\t"\
01865         "pcmpeqb %%mm7, %%mm3                   \n\t"\
01866         "pcmpeqb %%mm7, %%mm4                   \n\t"\
01867         "pcmpeqb %%mm7, %%mm5                   \n\t"\
01868         "movq %%mm3, %%mm6                      \n\t"\
01869         "pxor %%mm4, %%mm3                      \n\t"\
01870         "pxor %%mm5, %%mm4                      \n\t"\
01871         "pxor %%mm6, %%mm5                      \n\t"\
01872         "por %%mm3, %%mm1                       \n\t"\
01873         "por %%mm4, %%mm2                       \n\t"\
01874         "por %%mm5, %%mm0                       \n\t"\
01875         "pand %%mm2, %%mm0                      \n\t"\
01876         "pand %%mm1, %%mm0                      \n\t"\
01877         "movq %%mm0, " #b "                     \n\t"
01878 #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
01879 
01880 MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))
01881 MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
01882 MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))
01883 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
01884 
01885         : : "r" (src), "r" ((x86_reg)stride)
01886         : "%"REG_a, "%"REG_d
01887     );
01888 #endif //HAVE_MMX2
01889 #else //HAVE_MMX
01890     int x, y;
01891     src+= 4*stride;
01892     // FIXME - there should be a way to do a few columns in parallel like w/mmx
01893     for(x=0; x<8; x++){
01894         uint8_t *colsrc = src;
01895         for (y=0; y<4; y++){
01896             int a, b, c, d, e, f;
01897             a = colsrc[0       ];
01898             b = colsrc[stride  ];
01899             c = colsrc[stride*2];
01900             d = (a-b)>>31;
01901             e = (b-c)>>31;
01902             f = (c-a)>>31;
01903             colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
01904             colsrc += stride*2;
01905         }
01906         src++;
01907     }
01908 #endif //HAVE_MMX
01909 }
01910 
01911 #if HAVE_MMX
01912 
01915 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
01916 {
01917     __asm__(
01918         "lea (%0, %1), %%"REG_a"                \n\t"
01919 //      0       1       2       3       4       5       6       7       8       9
01920 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01921         "movq (%0), %%mm0                       \n\t" // 12345678
01922         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
01923         "movq %%mm0, %%mm2                      \n\t" // 12345678
01924         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
01925         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
01926 
01927         "movq (%%"REG_a", %1), %%mm1            \n\t"
01928         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
01929         "movq %%mm1, %%mm4                      \n\t"
01930         "punpcklbw %%mm3, %%mm1                 \n\t"
01931         "punpckhbw %%mm3, %%mm4                 \n\t"
01932 
01933         "movq %%mm0, %%mm3                      \n\t"
01934         "punpcklwd %%mm1, %%mm0                 \n\t"
01935         "punpckhwd %%mm1, %%mm3                 \n\t"
01936         "movq %%mm2, %%mm1                      \n\t"
01937         "punpcklwd %%mm4, %%mm2                 \n\t"
01938         "punpckhwd %%mm4, %%mm1                 \n\t"
01939 
01940         "movd %%mm0, 128(%2)                    \n\t"
01941         "psrlq $32, %%mm0                       \n\t"
01942         "movd %%mm0, 144(%2)                    \n\t"
01943         "movd %%mm3, 160(%2)                    \n\t"
01944         "psrlq $32, %%mm3                       \n\t"
01945         "movd %%mm3, 176(%2)                    \n\t"
01946         "movd %%mm3, 48(%3)                     \n\t"
01947         "movd %%mm2, 192(%2)                    \n\t"
01948         "movd %%mm2, 64(%3)                     \n\t"
01949         "psrlq $32, %%mm2                       \n\t"
01950         "movd %%mm2, 80(%3)                     \n\t"
01951         "movd %%mm1, 96(%3)                     \n\t"
01952         "psrlq $32, %%mm1                       \n\t"
01953         "movd %%mm1, 112(%3)                    \n\t"
01954 
01955         "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
01956 
01957         "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
01958         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
01959         "movq %%mm0, %%mm2                      \n\t" // 12345678
01960         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
01961         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
01962 
01963         "movq (%%"REG_a", %1), %%mm1            \n\t"
01964         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
01965         "movq %%mm1, %%mm4                      \n\t"
01966         "punpcklbw %%mm3, %%mm1                 \n\t"
01967         "punpckhbw %%mm3, %%mm4                 \n\t"
01968 
01969         "movq %%mm0, %%mm3                      \n\t"
01970         "punpcklwd %%mm1, %%mm0                 \n\t"
01971         "punpckhwd %%mm1, %%mm3                 \n\t"
01972         "movq %%mm2, %%mm1                      \n\t"
01973         "punpcklwd %%mm4, %%mm2                 \n\t"
01974         "punpckhwd %%mm4, %%mm1                 \n\t"
01975 
01976         "movd %%mm0, 132(%2)                    \n\t"
01977         "psrlq $32, %%mm0                       \n\t"
01978         "movd %%mm0, 148(%2)                    \n\t"
01979         "movd %%mm3, 164(%2)                    \n\t"
01980         "psrlq $32, %%mm3                       \n\t"
01981         "movd %%mm3, 180(%2)                    \n\t"
01982         "movd %%mm3, 52(%3)                     \n\t"
01983         "movd %%mm2, 196(%2)                    \n\t"
01984         "movd %%mm2, 68(%3)                     \n\t"
01985         "psrlq $32, %%mm2                       \n\t"
01986         "movd %%mm2, 84(%3)                     \n\t"
01987         "movd %%mm1, 100(%3)                    \n\t"
01988         "psrlq $32, %%mm1                       \n\t"
01989         "movd %%mm1, 116(%3)                    \n\t"
01990 
01991 
01992         :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)
01993         : "%"REG_a
01994     );
01995 }
01996 
02000 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
02001 {
02002     __asm__(
02003         "lea (%0, %1), %%"REG_a"                \n\t"
02004         "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
02005 //      0       1       2       3       4       5       6       7       8       9
02006 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
02007         "movq (%2), %%mm0                       \n\t" // 12345678
02008         "movq 16(%2), %%mm1                     \n\t" // abcdefgh
02009         "movq %%mm0, %%mm2                      \n\t" // 12345678
02010         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02011         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02012 
02013         "movq 32(%2), %%mm1                     \n\t"
02014         "movq 48(%2), %%mm3                     \n\t"
02015         "movq %%mm1, %%mm4                      \n\t"
02016         "punpcklbw %%mm3, %%mm1                 \n\t"
02017         "punpckhbw %%mm3, %%mm4                 \n\t"
02018 
02019         "movq %%mm0, %%mm3                      \n\t"
02020         "punpcklwd %%mm1, %%mm0                 \n\t"
02021         "punpckhwd %%mm1, %%mm3                 \n\t"
02022         "movq %%mm2, %%mm1                      \n\t"
02023         "punpcklwd %%mm4, %%mm2                 \n\t"
02024         "punpckhwd %%mm4, %%mm1                 \n\t"
02025 
02026         "movd %%mm0, (%0)                       \n\t"
02027         "psrlq $32, %%mm0                       \n\t"
02028         "movd %%mm0, (%%"REG_a")                \n\t"
02029         "movd %%mm3, (%%"REG_a", %1)            \n\t"
02030         "psrlq $32, %%mm3                       \n\t"
02031         "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
02032         "movd %%mm2, (%0, %1, 4)                \n\t"
02033         "psrlq $32, %%mm2                       \n\t"
02034         "movd %%mm2, (%%"REG_d")                \n\t"
02035         "movd %%mm1, (%%"REG_d", %1)            \n\t"
02036         "psrlq $32, %%mm1                       \n\t"
02037         "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
02038 
02039 
02040         "movq 64(%2), %%mm0                     \n\t" // 12345678
02041         "movq 80(%2), %%mm1                     \n\t" // abcdefgh
02042         "movq %%mm0, %%mm2                      \n\t" // 12345678
02043         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02044         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02045 
02046         "movq 96(%2), %%mm1                     \n\t"
02047         "movq 112(%2), %%mm3                    \n\t"
02048         "movq %%mm1, %%mm4                      \n\t"
02049         "punpcklbw %%mm3, %%mm1                 \n\t"
02050         "punpckhbw %%mm3, %%mm4                 \n\t"
02051 
02052         "movq %%mm0, %%mm3                      \n\t"
02053         "punpcklwd %%mm1, %%mm0                 \n\t"
02054         "punpckhwd %%mm1, %%mm3                 \n\t"
02055         "movq %%mm2, %%mm1                      \n\t"
02056         "punpcklwd %%mm4, %%mm2                 \n\t"
02057         "punpckhwd %%mm4, %%mm1                 \n\t"
02058 
02059         "movd %%mm0, 4(%0)                      \n\t"
02060         "psrlq $32, %%mm0                       \n\t"
02061         "movd %%mm0, 4(%%"REG_a")               \n\t"
02062         "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
02063         "psrlq $32, %%mm3                       \n\t"
02064         "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
02065         "movd %%mm2, 4(%0, %1, 4)               \n\t"
02066         "psrlq $32, %%mm2                       \n\t"
02067         "movd %%mm2, 4(%%"REG_d")               \n\t"
02068         "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
02069         "psrlq $32, %%mm1                       \n\t"
02070         "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
02071 
02072         :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)
02073         : "%"REG_a, "%"REG_d
02074     );
02075 }
02076 #endif //HAVE_MMX
02077 //static long test=0;
02078 
02079 #if !HAVE_ALTIVEC
02080 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
02081                                     uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)
02082 {
02083     // to save a register (FIXME do this outside of the loops)
02084     tempBlurredPast[127]= maxNoise[0];
02085     tempBlurredPast[128]= maxNoise[1];
02086     tempBlurredPast[129]= maxNoise[2];
02087 
02088 #define FAST_L2_DIFF
02089 //#define L1_DIFF //u should change the thresholds too if u try that one
02090 #if HAVE_MMX2 || HAVE_AMD3DNOW
02091     __asm__ volatile(
02092         "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
02093         "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
02094         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02095 //      0       1       2       3       4       5       6       7       8       9
02096 //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
02097 //FIXME reorder?
02098 #ifdef L1_DIFF //needs mmx2
02099         "movq (%0), %%mm0                       \n\t" // L0
02100         "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
02101         "movq (%0, %2), %%mm1                   \n\t" // L1
02102         "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
02103         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02104         "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
02105         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02106         "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
02107 
02108         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02109         "paddw %%mm1, %%mm0                     \n\t"
02110         "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
02111         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02112         "paddw %%mm2, %%mm0                     \n\t"
02113         "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
02114         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02115         "paddw %%mm3, %%mm0                     \n\t"
02116         "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
02117         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02118         "paddw %%mm4, %%mm0                     \n\t"
02119         "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
02120         "paddw %%mm5, %%mm6                     \n\t"
02121         "paddw %%mm7, %%mm6                     \n\t"
02122         "paddw %%mm6, %%mm0                     \n\t"
02123 #else //L1_DIFF
02124 #if defined (FAST_L2_DIFF)
02125         "pcmpeqb %%mm7, %%mm7                   \n\t"
02126         "movq "MANGLE(b80)", %%mm6              \n\t"
02127         "pxor %%mm0, %%mm0                      \n\t"
02128 #define REAL_L2_DIFF_CORE(a, b)\
02129         "movq " #a ", %%mm5                     \n\t"\
02130         "movq " #b ", %%mm2                     \n\t"\
02131         "pxor %%mm7, %%mm2                      \n\t"\
02132         PAVGB(%%mm2, %%mm5)\
02133         "paddb %%mm6, %%mm5                     \n\t"\
02134         "movq %%mm5, %%mm2                      \n\t"\
02135         "psllw $8, %%mm5                        \n\t"\
02136         "pmaddwd %%mm5, %%mm5                   \n\t"\
02137         "pmaddwd %%mm2, %%mm2                   \n\t"\
02138         "paddd %%mm2, %%mm5                     \n\t"\
02139         "psrld $14, %%mm5                       \n\t"\
02140         "paddd %%mm5, %%mm0                     \n\t"
02141 
02142 #else //defined (FAST_L2_DIFF)
02143         "pxor %%mm7, %%mm7                      \n\t"
02144         "pxor %%mm0, %%mm0                      \n\t"
02145 #define REAL_L2_DIFF_CORE(a, b)\
02146         "movq " #a ", %%mm5                     \n\t"\
02147         "movq " #b ", %%mm2                     \n\t"\
02148         "movq %%mm5, %%mm1                      \n\t"\
02149         "movq %%mm2, %%mm3                      \n\t"\
02150         "punpcklbw %%mm7, %%mm5                 \n\t"\
02151         "punpckhbw %%mm7, %%mm1                 \n\t"\
02152         "punpcklbw %%mm7, %%mm2                 \n\t"\
02153         "punpckhbw %%mm7, %%mm3                 \n\t"\
02154         "psubw %%mm2, %%mm5                     \n\t"\
02155         "psubw %%mm3, %%mm1                     \n\t"\
02156         "pmaddwd %%mm5, %%mm5                   \n\t"\
02157         "pmaddwd %%mm1, %%mm1                   \n\t"\
02158         "paddd %%mm1, %%mm5                     \n\t"\
02159         "paddd %%mm5, %%mm0                     \n\t"
02160 
02161 #endif //defined (FAST_L2_DIFF)
02162 
02163 #define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)
02164 
02165 L2_DIFF_CORE((%0)          , (%1))
02166 L2_DIFF_CORE((%0, %2)      , (%1, %2))
02167 L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))
02168 L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))
02169 L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))
02170 L2_DIFF_CORE((%0, %%REGd)  , (%1, %%REGd))
02171 L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
02172 L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))
02173 
02174 #endif //L1_DIFF
02175 
02176         "movq %%mm0, %%mm4                      \n\t"
02177         "psrlq $32, %%mm0                       \n\t"
02178         "paddd %%mm0, %%mm4                     \n\t"
02179         "movd %%mm4, %%ecx                      \n\t"
02180         "shll $2, %%ecx                         \n\t"
02181         "mov %3, %%"REG_d"                      \n\t"
02182         "addl -4(%%"REG_d"), %%ecx              \n\t"
02183         "addl 4(%%"REG_d"), %%ecx               \n\t"
02184         "addl -1024(%%"REG_d"), %%ecx           \n\t"
02185         "addl $4, %%ecx                         \n\t"
02186         "addl 1024(%%"REG_d"), %%ecx            \n\t"
02187         "shrl $3, %%ecx                         \n\t"
02188         "movl %%ecx, (%%"REG_d")                \n\t"
02189 
02190 //        "mov %3, %%"REG_c"                      \n\t"
02191 //        "mov %%"REG_c", test                    \n\t"
02192 //        "jmp 4f                                 \n\t"
02193         "cmpl 512(%%"REG_d"), %%ecx             \n\t"
02194         " jb 2f                                 \n\t"
02195         "cmpl 516(%%"REG_d"), %%ecx             \n\t"
02196         " jb 1f                                 \n\t"
02197 
02198         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02199         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02200         "movq (%0), %%mm0                       \n\t" // L0
02201         "movq (%0, %2), %%mm1                   \n\t" // L1
02202         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02203         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02204         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02205         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02206         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02207         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02208         "movq %%mm0, (%1)                       \n\t" // L0
02209         "movq %%mm1, (%1, %2)                   \n\t" // L1
02210         "movq %%mm2, (%1, %2, 2)                \n\t" // L2
02211         "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
02212         "movq %%mm4, (%1, %2, 4)                \n\t" // L4
02213         "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
02214         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
02215         "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
02216         "jmp 4f                                 \n\t"
02217 
02218         "1:                                     \n\t"
02219         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02220         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02221         "movq (%0), %%mm0                       \n\t" // L0
02222         PAVGB((%1), %%mm0)                            // L0
02223         "movq (%0, %2), %%mm1                   \n\t" // L1
02224         PAVGB((%1, %2), %%mm1)                        // L1
02225         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02226         PAVGB((%1, %2, 2), %%mm2)                     // L2
02227         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02228         PAVGB((%1, %%REGa), %%mm3)                    // L3
02229         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02230         PAVGB((%1, %2, 4), %%mm4)                     // L4
02231         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02232         PAVGB((%1, %%REGd), %%mm5)                    // L5
02233         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02234         PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
02235         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02236         PAVGB((%1, %%REGc), %%mm7)                    // L7
02237         "movq %%mm0, (%1)                       \n\t" // R0
02238         "movq %%mm1, (%1, %2)                   \n\t" // R1
02239         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02240         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02241         "movq %%mm4, (%1, %2, 4)                \n\t" // R4
02242         "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
02243         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
02244         "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
02245         "movq %%mm0, (%0)                       \n\t" // L0
02246         "movq %%mm1, (%0, %2)                   \n\t" // L1
02247         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02248         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02249         "movq %%mm4, (%0, %2, 4)                \n\t" // L4
02250         "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
02251         "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
02252         "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
02253         "jmp 4f                                 \n\t"
02254 
02255         "2:                                     \n\t"
02256         "cmpl 508(%%"REG_d"), %%ecx             \n\t"
02257         " jb 3f                                 \n\t"
02258 
02259         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02260         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02261         "movq (%0), %%mm0                       \n\t" // L0
02262         "movq (%0, %2), %%mm1                   \n\t" // L1
02263         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02264         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02265         "movq (%1), %%mm4                       \n\t" // R0
02266         "movq (%1, %2), %%mm5                   \n\t" // R1
02267         "movq (%1, %2, 2), %%mm6                \n\t" // R2
02268         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
02269         PAVGB(%%mm4, %%mm0)
02270         PAVGB(%%mm5, %%mm1)
02271         PAVGB(%%mm6, %%mm2)
02272         PAVGB(%%mm7, %%mm3)
02273         PAVGB(%%mm4, %%mm0)
02274         PAVGB(%%mm5, %%mm1)
02275         PAVGB(%%mm6, %%mm2)
02276         PAVGB(%%mm7, %%mm3)
02277         "movq %%mm0, (%1)                       \n\t" // R0
02278         "movq %%mm1, (%1, %2)                   \n\t" // R1
02279         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02280         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02281         "movq %%mm0, (%0)                       \n\t" // L0
02282         "movq %%mm1, (%0, %2)                   \n\t" // L1
02283         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02284         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02285 
02286         "movq (%0, %2, 4), %%mm0                \n\t" // L4
02287         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
02288         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
02289         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
02290         "movq (%1, %2, 4), %%mm4                \n\t" // R4
02291         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
02292         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
02293         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
02294         PAVGB(%%mm4, %%mm0)
02295         PAVGB(%%mm5, %%mm1)
02296         PAVGB(%%mm6, %%mm2)
02297         PAVGB(%%mm7, %%mm3)
02298         PAVGB(%%mm4, %%mm0)
02299         PAVGB(%%mm5, %%mm1)
02300         PAVGB(%%mm6, %%mm2)
02301         PAVGB(%%mm7, %%mm3)
02302         "movq %%mm0, (%1, %2, 4)                \n\t" // R4
02303         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
02304         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
02305         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
02306         "movq %%mm0, (%0, %2, 4)                \n\t" // L4
02307         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
02308         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
02309         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
02310         "jmp 4f                                 \n\t"
02311 
02312         "3:                                     \n\t"
02313         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02314         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02315         "movq (%0), %%mm0                       \n\t" // L0
02316         "movq (%0, %2), %%mm1                   \n\t" // L1
02317         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02318         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02319         "movq (%1), %%mm4                       \n\t" // R0
02320         "movq (%1, %2), %%mm5                   \n\t" // R1
02321         "movq (%1, %2, 2), %%mm6                \n\t" // R2
02322         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
02323         PAVGB(%%mm4, %%mm0)
02324         PAVGB(%%mm5, %%mm1)
02325         PAVGB(%%mm6, %%mm2)
02326         PAVGB(%%mm7, %%mm3)
02327         PAVGB(%%mm4, %%mm0)
02328         PAVGB(%%mm5, %%mm1)
02329         PAVGB(%%mm6, %%mm2)
02330         PAVGB(%%mm7, %%mm3)
02331         PAVGB(%%mm4, %%mm0)
02332         PAVGB(%%mm5, %%mm1)
02333         PAVGB(%%mm6, %%mm2)
02334         PAVGB(%%mm7, %%mm3)
02335         "movq %%mm0, (%1)                       \n\t" // R0
02336         "movq %%mm1, (%1, %2)                   \n\t" // R1
02337         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02338         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02339         "movq %%mm0, (%0)                       \n\t" // L0
02340         "movq %%mm1, (%0, %2)                   \n\t" // L1
02341         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02342         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02343 
02344         "movq (%0, %2, 4), %%mm0                \n\t" // L4
02345         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
02346         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
02347         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
02348         "movq (%1, %2, 4), %%mm4                \n\t" // R4
02349         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
02350         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
02351         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
02352         PAVGB(%%mm4, %%mm0)
02353         PAVGB(%%mm5, %%mm1)
02354         PAVGB(%%mm6, %%mm2)
02355         PAVGB(%%mm7, %%mm3)
02356         PAVGB(%%mm4, %%mm0)
02357         PAVGB(%%mm5, %%mm1)
02358         PAVGB(%%mm6, %%mm2)
02359         PAVGB(%%mm7, %%mm3)
02360         PAVGB(%%mm4, %%mm0)
02361         PAVGB(%%mm5, %%mm1)
02362         PAVGB(%%mm6, %%mm2)
02363         PAVGB(%%mm7, %%mm3)
02364         "movq %%mm0, (%1, %2, 4)                \n\t" // R4
02365         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
02366         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
02367         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
02368         "movq %%mm0, (%0, %2, 4)                \n\t" // L4
02369         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
02370         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
02371         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
02372 
02373         "4:                                     \n\t"
02374 
02375         :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)
02376         : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
02377     );
02378 #else //HAVE_MMX2 || HAVE_AMD3DNOW
02379 {
02380     int y;
02381     int d=0;
02382 //    int sysd=0;
02383     int i;
02384 
02385     for(y=0; y<8; y++){
02386         int x;
02387         for(x=0; x<8; x++){
02388             int ref= tempBlurred[ x + y*stride ];
02389             int cur= src[ x + y*stride ];
02390             int d1=ref - cur;
02391 //            if(x==0 || x==7) d1+= d1>>1;
02392 //            if(y==0 || y==7) d1+= d1>>1;
02393 //            d+= FFABS(d1);
02394             d+= d1*d1;
02395 //            sysd+= d1;
02396         }
02397     }
02398     i=d;
02399     d=  (
02400         4*d
02401         +(*(tempBlurredPast-256))
02402         +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
02403         +(*(tempBlurredPast+256))
02404         +4)>>3;
02405     *tempBlurredPast=i;
02406 //    ((*tempBlurredPast)*3 + d + 2)>>2;
02407 
02408 /*
02409 Switch between
02410  1  0  0  0  0  0  0  (0)
02411 64 32 16  8  4  2  1  (1)
02412 64 48 36 27 20 15 11 (33) (approx)
02413 64 56 49 43 37 33 29 (200) (approx)
02414 */
02415     if(d > maxNoise[1]){
02416         if(d < maxNoise[2]){
02417             for(y=0; y<8; y++){
02418                 int x;
02419                 for(x=0; x<8; x++){
02420                     int ref= tempBlurred[ x + y*stride ];
02421                     int cur= src[ x + y*stride ];
02422                     tempBlurred[ x + y*stride ]=
02423                     src[ x + y*stride ]=
02424                         (ref + cur + 1)>>1;
02425                 }
02426             }
02427         }else{
02428             for(y=0; y<8; y++){
02429                 int x;
02430                 for(x=0; x<8; x++){
02431                     tempBlurred[ x + y*stride ]= src[ x + y*stride ];
02432                 }
02433             }
02434         }
02435     }else{
02436         if(d < maxNoise[0]){
02437             for(y=0; y<8; y++){
02438                 int x;
02439                 for(x=0; x<8; x++){
02440                     int ref= tempBlurred[ x + y*stride ];
02441                     int cur= src[ x + y*stride ];
02442                     tempBlurred[ x + y*stride ]=
02443                     src[ x + y*stride ]=
02444                         (ref*7 + cur + 4)>>3;
02445                 }
02446             }
02447         }else{
02448             for(y=0; y<8; y++){
02449                 int x;
02450                 for(x=0; x<8; x++){
02451                     int ref= tempBlurred[ x + y*stride ];
02452                     int cur= src[ x + y*stride ];
02453                     tempBlurred[ x + y*stride ]=
02454                     src[ x + y*stride ]=
02455                         (ref*3 + cur + 2)>>2;
02456                 }
02457             }
02458         }
02459     }
02460 }
02461 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
02462 }
02463 #endif //HAVE_ALTIVEC
02464 
02465 #if HAVE_MMX
02466 
02469 static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
02470     int64_t dc_mask, eq_mask, both_masks;
02471     int64_t sums[10*8*2];
02472     src+= step*3; // src points to begin of the 8x8 Block
02473     //{ START_TIMER
02474     __asm__ volatile(
02475         "movq %0, %%mm7                         \n\t"
02476         "movq %1, %%mm6                         \n\t"
02477         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
02478         );
02479 
02480     __asm__ volatile(
02481         "lea (%2, %3), %%"REG_a"                \n\t"
02482 //      0       1       2       3       4       5       6       7       8       9
02483 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
02484 
02485         "movq (%2), %%mm0                       \n\t"
02486         "movq (%%"REG_a"), %%mm1                \n\t"
02487         "movq %%mm1, %%mm3                      \n\t"
02488         "movq %%mm1, %%mm4                      \n\t"
02489         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
02490         "paddb %%mm7, %%mm0                     \n\t"
02491         "pcmpgtb %%mm6, %%mm0                   \n\t"
02492 
02493         "movq (%%"REG_a",%3), %%mm2             \n\t"
02494         PMAXUB(%%mm2, %%mm4)
02495         PMINUB(%%mm2, %%mm3, %%mm5)
02496         "psubb %%mm2, %%mm1                     \n\t"
02497         "paddb %%mm7, %%mm1                     \n\t"
02498         "pcmpgtb %%mm6, %%mm1                   \n\t"
02499         "paddb %%mm1, %%mm0                     \n\t"
02500 
02501         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
02502         PMAXUB(%%mm1, %%mm4)
02503         PMINUB(%%mm1, %%mm3, %%mm5)
02504         "psubb %%mm1, %%mm2                     \n\t"
02505         "paddb %%mm7, %%mm2                     \n\t"
02506         "pcmpgtb %%mm6, %%mm2                   \n\t"
02507         "paddb %%mm2, %%mm0                     \n\t"
02508 
02509         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
02510 
02511         "movq (%2, %3, 4), %%mm2                \n\t"
02512         PMAXUB(%%mm2, %%mm4)
02513         PMINUB(%%mm2, %%mm3, %%mm5)
02514         "psubb %%mm2, %%mm1                     \n\t"
02515         "paddb %%mm7, %%mm1                     \n\t"
02516         "pcmpgtb %%mm6, %%mm1                   \n\t"
02517         "paddb %%mm1, %%mm0                     \n\t"
02518 
02519         "movq (%%"REG_a"), %%mm1                \n\t"
02520         PMAXUB(%%mm1, %%mm4)
02521         PMINUB(%%mm1, %%mm3, %%mm5)
02522         "psubb %%mm1, %%mm2                     \n\t"
02523         "paddb %%mm7, %%mm2                     \n\t"
02524         "pcmpgtb %%mm6, %%mm2                   \n\t"
02525         "paddb %%mm2, %%mm0                     \n\t"
02526 
02527         "movq (%%"REG_a", %3), %%mm2            \n\t"
02528         PMAXUB(%%mm2, %%mm4)
02529         PMINUB(%%mm2, %%mm3, %%mm5)
02530         "psubb %%mm2, %%mm1                     \n\t"
02531         "paddb %%mm7, %%mm1                     \n\t"
02532         "pcmpgtb %%mm6, %%mm1                   \n\t"
02533         "paddb %%mm1, %%mm0                     \n\t"
02534 
02535         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
02536         PMAXUB(%%mm1, %%mm4)
02537         PMINUB(%%mm1, %%mm3, %%mm5)
02538         "psubb %%mm1, %%mm2                     \n\t"
02539         "paddb %%mm7, %%mm2                     \n\t"
02540         "pcmpgtb %%mm6, %%mm2                   \n\t"
02541         "paddb %%mm2, %%mm0                     \n\t"
02542 
02543         "movq (%2, %3, 8), %%mm2                \n\t"
02544         PMAXUB(%%mm2, %%mm4)
02545         PMINUB(%%mm2, %%mm3, %%mm5)
02546         "psubb %%mm2, %%mm1                     \n\t"
02547         "paddb %%mm7, %%mm1                     \n\t"
02548         "pcmpgtb %%mm6, %%mm1                   \n\t"
02549         "paddb %%mm1, %%mm0                     \n\t"
02550 
02551         "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
02552         "psubb %%mm1, %%mm2                     \n\t"
02553         "paddb %%mm7, %%mm2                     \n\t"
02554         "pcmpgtb %%mm6, %%mm2                   \n\t"
02555         "paddb %%mm2, %%mm0                     \n\t"
02556         "psubusb %%mm3, %%mm4                   \n\t"
02557 
02558         "pxor %%mm6, %%mm6                      \n\t"
02559         "movq %4, %%mm7                         \n\t" // QP,..., QP
02560         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
02561         "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
02562         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
02563         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
02564         "movq %%mm7, %1                         \n\t"
02565 
02566         "movq %5, %%mm7                         \n\t"
02567         "punpcklbw %%mm7, %%mm7                 \n\t"
02568         "punpcklbw %%mm7, %%mm7                 \n\t"
02569         "punpcklbw %%mm7, %%mm7                 \n\t"
02570         "psubb %%mm0, %%mm6                     \n\t"
02571         "pcmpgtb %%mm7, %%mm6                   \n\t"
02572         "movq %%mm6, %0                         \n\t"
02573 
02574         : "=m" (eq_mask), "=m" (dc_mask)
02575         : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
02576         : "%"REG_a
02577     );
02578 
02579     both_masks = dc_mask & eq_mask;
02580 
02581     if(both_masks){
02582         x86_reg offset= -8*step;
02583         int64_t *temp_sums= sums;
02584 
02585         __asm__ volatile(
02586             "movq %2, %%mm0                         \n\t"  // QP,..., QP
02587             "pxor %%mm4, %%mm4                      \n\t"
02588 
02589             "movq (%0), %%mm6                       \n\t"
02590             "movq (%0, %1), %%mm5                   \n\t"
02591             "movq %%mm5, %%mm1                      \n\t"
02592             "movq %%mm6, %%mm2                      \n\t"
02593             "psubusb %%mm6, %%mm5                   \n\t"
02594             "psubusb %%mm1, %%mm2                   \n\t"
02595             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
02596             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
02597             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
02598 
02599             "pxor %%mm6, %%mm1                      \n\t"
02600             "pand %%mm0, %%mm1                      \n\t"
02601             "pxor %%mm1, %%mm6                      \n\t"
02602             // 0:QP  6:First
02603 
02604             "movq (%0, %1, 8), %%mm5                \n\t"
02605             "add %1, %0                             \n\t" // %0 points to line 1 not 0
02606             "movq (%0, %1, 8), %%mm7                \n\t"
02607             "movq %%mm5, %%mm1                      \n\t"
02608             "movq %%mm7, %%mm2                      \n\t"
02609             "psubusb %%mm7, %%mm5                   \n\t"
02610             "psubusb %%mm1, %%mm2                   \n\t"
02611             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
02612             "movq %2, %%mm0                         \n\t"  // QP,..., QP
02613             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
02614             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
02615 
02616             "pxor %%mm7, %%mm1                      \n\t"
02617             "pand %%mm0, %%mm1                      \n\t"
02618             "pxor %%mm1, %%mm7                      \n\t"
02619 
02620             "movq %%mm6, %%mm5                      \n\t"
02621             "punpckhbw %%mm4, %%mm6                 \n\t"
02622             "punpcklbw %%mm4, %%mm5                 \n\t"
02623             // 4:0 5/6:First 7:Last
02624 
02625             "movq %%mm5, %%mm0                      \n\t"
02626             "movq %%mm6, %%mm1                      \n\t"
02627             "psllw $2, %%mm0                        \n\t"
02628             "psllw $2, %%mm1                        \n\t"
02629             "paddw "MANGLE(w04)", %%mm0             \n\t"
02630             "paddw "MANGLE(w04)", %%mm1             \n\t"
02631 
02632 #define NEXT\
02633             "movq (%0), %%mm2                       \n\t"\
02634             "movq (%0), %%mm3                       \n\t"\
02635             "add %1, %0                             \n\t"\
02636             "punpcklbw %%mm4, %%mm2                 \n\t"\
02637             "punpckhbw %%mm4, %%mm3                 \n\t"\
02638             "paddw %%mm2, %%mm0                     \n\t"\
02639             "paddw %%mm3, %%mm1                     \n\t"
02640 
02641 #define PREV\
02642             "movq (%0), %%mm2                       \n\t"\
02643             "movq (%0), %%mm3                       \n\t"\
02644             "add %1, %0                             \n\t"\
02645             "punpcklbw %%mm4, %%mm2                 \n\t"\
02646             "punpckhbw %%mm4, %%mm3                 \n\t"\
02647             "psubw %%mm2, %%mm0                     \n\t"\
02648             "psubw %%mm3, %%mm1                     \n\t"
02649 
02650 
02651             NEXT //0
02652             NEXT //1
02653             NEXT //2
02654             "movq %%mm0, (%3)                       \n\t"
02655             "movq %%mm1, 8(%3)                      \n\t"
02656 
02657             NEXT //3
02658             "psubw %%mm5, %%mm0                     \n\t"
02659             "psubw %%mm6, %%mm1                     \n\t"
02660             "movq %%mm0, 16(%3)                     \n\t"
02661             "movq %%mm1, 24(%3)                     \n\t"
02662 
02663             NEXT //4
02664             "psubw %%mm5, %%mm0                     \n\t"
02665             "psubw %%mm6, %%mm1                     \n\t"
02666             "movq %%mm0, 32(%3)                     \n\t"
02667             "movq %%mm1, 40(%3)                     \n\t"
02668 
02669             NEXT //5
02670             "psubw %%mm5, %%mm0                     \n\t"
02671             "psubw %%mm6, %%mm1                     \n\t"
02672             "movq %%mm0, 48(%3)                     \n\t"
02673             "movq %%mm1, 56(%3)                     \n\t"
02674 
02675             NEXT //6
02676             "psubw %%mm5, %%mm0                     \n\t"
02677             "psubw %%mm6, %%mm1                     \n\t"
02678             "movq %%mm0, 64(%3)                     \n\t"
02679             "movq %%mm1, 72(%3)                     \n\t"
02680 
02681             "movq %%mm7, %%mm6                      \n\t"
02682             "punpckhbw %%mm4, %%mm7                 \n\t"
02683             "punpcklbw %%mm4, %%mm6                 \n\t"
02684 
02685             NEXT //7
02686             "mov %4, %0                             \n\t"
02687             "add %1, %0                             \n\t"
02688             PREV //0
02689             "movq %%mm0, 80(%3)                     \n\t"
02690             "movq %%mm1, 88(%3)                     \n\t"
02691 
02692             PREV //1
02693             "paddw %%mm6, %%mm0                     \n\t"
02694             "paddw %%mm7, %%mm1                     \n\t"
02695             "movq %%mm0, 96(%3)                     \n\t"
02696             "movq %%mm1, 104(%3)                    \n\t"
02697 
02698             PREV //2
02699             "paddw %%mm6, %%mm0                     \n\t"
02700             "paddw %%mm7, %%mm1                     \n\t"
02701             "movq %%mm0, 112(%3)                    \n\t"
02702             "movq %%mm1, 120(%3)                    \n\t"
02703 
02704             PREV //3
02705             "paddw %%mm6, %%mm0                     \n\t"
02706             "paddw %%mm7, %%mm1                     \n\t"
02707             "movq %%mm0, 128(%3)                    \n\t"
02708             "movq %%mm1, 136(%3)                    \n\t"
02709 
02710             PREV //4
02711             "paddw %%mm6, %%mm0                     \n\t"
02712             "paddw %%mm7, %%mm1                     \n\t"
02713             "movq %%mm0, 144(%3)                    \n\t"
02714             "movq %%mm1, 152(%3)                    \n\t"
02715 
02716             "mov %4, %0                             \n\t" //FIXME
02717 
02718             : "+&r"(src)
02719             : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)
02720         );
02721 
02722         src+= step; // src points to begin of the 8x8 Block
02723 
02724         __asm__ volatile(
02725             "movq %4, %%mm6                         \n\t"
02726             "pcmpeqb %%mm5, %%mm5                   \n\t"
02727             "pxor %%mm6, %%mm5                      \n\t"
02728             "pxor %%mm7, %%mm7                      \n\t"
02729 
02730             "1:                                     \n\t"
02731             "movq (%1), %%mm0                       \n\t"
02732             "movq 8(%1), %%mm1                      \n\t"
02733             "paddw 32(%1), %%mm0                    \n\t"
02734             "paddw 40(%1), %%mm1                    \n\t"
02735             "movq (%0, %3), %%mm2                   \n\t"
02736             "movq %%mm2, %%mm3                      \n\t"
02737             "movq %%mm2, %%mm4                      \n\t"
02738             "punpcklbw %%mm7, %%mm2                 \n\t"
02739             "punpckhbw %%mm7, %%mm3                 \n\t"
02740             "paddw %%mm2, %%mm0                     \n\t"
02741             "paddw %%mm3, %%mm1                     \n\t"
02742             "paddw %%mm2, %%mm0                     \n\t"
02743             "paddw %%mm3, %%mm1                     \n\t"
02744             "psrlw $4, %%mm0                        \n\t"
02745             "psrlw $4, %%mm1                        \n\t"
02746             "packuswb %%mm1, %%mm0                  \n\t"
02747             "pand %%mm6, %%mm0                      \n\t"
02748             "pand %%mm5, %%mm4                      \n\t"
02749             "por %%mm4, %%mm0                       \n\t"
02750             "movq %%mm0, (%0, %3)                   \n\t"
02751             "add $16, %1                            \n\t"
02752             "add %2, %0                             \n\t"
02753             " js 1b                                 \n\t"
02754 
02755             : "+r"(offset), "+r"(temp_sums)
02756             : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)
02757         );
02758     }else
02759         src+= step; // src points to begin of the 8x8 Block
02760 
02761     if(eq_mask != -1LL){
02762         uint8_t *temp_src= src;
02763         DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars
02764         __asm__ volatile(
02765             "pxor %%mm7, %%mm7                      \n\t"
02766 //      0       1       2       3       4       5       6       7       8       9
02767 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1
02768 
02769             "movq (%0), %%mm0                       \n\t"
02770             "movq %%mm0, %%mm1                      \n\t"
02771             "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
02772             "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
02773 
02774             "movq (%0, %1), %%mm2                   \n\t"
02775             "lea (%0, %1, 2), %%"REG_a"             \n\t"
02776             "movq %%mm2, %%mm3                      \n\t"
02777             "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
02778             "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
02779 
02780             "movq (%%"REG_a"), %%mm4                \n\t"
02781             "movq %%mm4, %%mm5                      \n\t"
02782             "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
02783             "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
02784 
02785             "paddw %%mm0, %%mm0                     \n\t" // 2L0
02786             "paddw %%mm1, %%mm1                     \n\t" // 2H0
02787             "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
02788             "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
02789             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
02790             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
02791 
02792             "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
02793             "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
02794             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
02795             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
02796 
02797             "movq (%%"REG_a", %1), %%mm2            \n\t"
02798             "movq %%mm2, %%mm3                      \n\t"
02799             "punpcklbw %%mm7, %%mm2                 \n\t" // L3
02800             "punpckhbw %%mm7, %%mm3                 \n\t" // H3
02801 
02802             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
02803             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
02804             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02805             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02806             "movq %%mm0, (%4)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02807             "movq %%mm1, 8(%4)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02808 
02809             "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
02810             "movq %%mm0, %%mm1                      \n\t"
02811             "punpcklbw %%mm7, %%mm0                 \n\t" // L4
02812             "punpckhbw %%mm7, %%mm1                 \n\t" // H4
02813 
02814             "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
02815             "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
02816             "movq %%mm2, 16(%4)                     \n\t" // L3 - L4
02817             "movq %%mm3, 24(%4)                     \n\t" // H3 - H4
02818             "paddw %%mm4, %%mm4                     \n\t" // 2L2
02819             "paddw %%mm5, %%mm5                     \n\t" // 2H2
02820             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
02821             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
02822 
02823             "lea (%%"REG_a", %1), %0                \n\t"
02824             "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
02825             "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
02826             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
02827             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
02828 //50 opcodes so far
02829             "movq (%0, %1, 2), %%mm2                \n\t"
02830             "movq %%mm2, %%mm3                      \n\t"
02831             "punpcklbw %%mm7, %%mm2                 \n\t" // L5
02832             "punpckhbw %%mm7, %%mm3                 \n\t" // H5
02833             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
02834             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
02835             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
02836             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
02837 
02838             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
02839             "punpcklbw %%mm7, %%mm6                 \n\t" // L6
02840             "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
02841             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
02842             "punpckhbw %%mm7, %%mm6                 \n\t" // H6
02843             "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
02844 
02845             "paddw %%mm0, %%mm0                     \n\t" // 2L4
02846             "paddw %%mm1, %%mm1                     \n\t" // 2H4
02847             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
02848             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
02849 
02850             "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
02851             "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
02852             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
02853             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
02854 
02855             "movq (%0, %1, 4), %%mm2                \n\t"
02856             "movq %%mm2, %%mm3                      \n\t"
02857             "punpcklbw %%mm7, %%mm2                 \n\t" // L7
02858             "punpckhbw %%mm7, %%mm3                 \n\t" // H7
02859 
02860             "paddw %%mm2, %%mm2                     \n\t" // 2L7
02861             "paddw %%mm3, %%mm3                     \n\t" // 2H7
02862             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
02863             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
02864 
02865             "movq (%4), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02866             "movq 8(%4), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02867 
02868 #if HAVE_MMX2
02869             "movq %%mm7, %%mm6                      \n\t" // 0
02870             "psubw %%mm0, %%mm6                     \n\t"
02871             "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
02872             "movq %%mm7, %%mm6                      \n\t" // 0
02873             "psubw %%mm1, %%mm6                     \n\t"
02874             "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
02875             "movq %%mm7, %%mm6                      \n\t" // 0
02876             "psubw %%mm2, %%mm6                     \n\t"
02877             "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
02878             "movq %%mm7, %%mm6                      \n\t" // 0
02879             "psubw %%mm3, %%mm6                     \n\t"
02880             "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
02881 #else
02882             "movq %%mm7, %%mm6                      \n\t" // 0
02883             "pcmpgtw %%mm0, %%mm6                   \n\t"
02884             "pxor %%mm6, %%mm0                      \n\t"
02885             "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
02886             "movq %%mm7, %%mm6                      \n\t" // 0
02887             "pcmpgtw %%mm1, %%mm6                   \n\t"
02888             "pxor %%mm6, %%mm1                      \n\t"
02889             "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
02890             "movq %%mm7, %%mm6                      \n\t" // 0
02891             "pcmpgtw %%mm2, %%mm6                   \n\t"
02892             "pxor %%mm6, %%mm2                      \n\t"
02893             "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
02894             "movq %%mm7, %%mm6                      \n\t" // 0
02895             "pcmpgtw %%mm3, %%mm6                   \n\t"
02896             "pxor %%mm6, %%mm3                      \n\t"
02897             "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
02898 #endif
02899 
02900 #if HAVE_MMX2
02901             "pminsw %%mm2, %%mm0                    \n\t"
02902             "pminsw %%mm3, %%mm1                    \n\t"
02903 #else
02904             "movq %%mm0, %%mm6                      \n\t"
02905             "psubusw %%mm2, %%mm6                   \n\t"
02906             "psubw %%mm6, %%mm0                     \n\t"
02907             "movq %%mm1, %%mm6                      \n\t"
02908             "psubusw %%mm3, %%mm6                   \n\t"
02909             "psubw %%mm6, %%mm1                     \n\t"
02910 #endif
02911 
02912             "movd %2, %%mm2                         \n\t" // QP
02913             "punpcklbw %%mm7, %%mm2                 \n\t"
02914 
02915             "movq %%mm7, %%mm6                      \n\t" // 0
02916             "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
02917             "pxor %%mm6, %%mm4                      \n\t"
02918             "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
02919             "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
02920             "pxor %%mm7, %%mm5                      \n\t"
02921             "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
02922 // 100 opcodes
02923             "psllw $3, %%mm2                        \n\t" // 8QP
02924             "movq %%mm2, %%mm3                      \n\t" // 8QP
02925             "pcmpgtw %%mm4, %%mm2                   \n\t"
02926             "pcmpgtw %%mm5, %%mm3                   \n\t"
02927             "pand %%mm2, %%mm4                      \n\t"
02928             "pand %%mm3, %%mm5                      \n\t"
02929 
02930 
02931             "psubusw %%mm0, %%mm4                   \n\t" // hd
02932             "psubusw %%mm1, %%mm5                   \n\t" // ld
02933 
02934 
02935             "movq "MANGLE(w05)", %%mm2              \n\t" // 5
02936             "pmullw %%mm2, %%mm4                    \n\t"
02937             "pmullw %%mm2, %%mm5                    \n\t"
02938             "movq "MANGLE(w20)", %%mm2              \n\t" // 32
02939             "paddw %%mm2, %%mm4                     \n\t"
02940             "paddw %%mm2, %%mm5                     \n\t"
02941             "psrlw $6, %%mm4                        \n\t"
02942             "psrlw $6, %%mm5                        \n\t"
02943 
02944             "movq 16(%4), %%mm0                     \n\t" // L3 - L4
02945             "movq 24(%4), %%mm1                     \n\t" // H3 - H4
02946 
02947             "pxor %%mm2, %%mm2                      \n\t"
02948             "pxor %%mm3, %%mm3                      \n\t"
02949 
02950             "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
02951             "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
02952             "pxor %%mm2, %%mm0                      \n\t"
02953             "pxor %%mm3, %%mm1                      \n\t"
02954             "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
02955             "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
02956             "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
02957             "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
02958 
02959             "pxor %%mm6, %%mm2                      \n\t"
02960             "pxor %%mm7, %%mm3                      \n\t"
02961             "pand %%mm2, %%mm4                      \n\t"
02962             "pand %%mm3, %%mm5                      \n\t"
02963 
02964 #if HAVE_MMX2
02965             "pminsw %%mm0, %%mm4                    \n\t"
02966             "pminsw %%mm1, %%mm5                    \n\t"
02967 #else
02968             "movq %%mm4, %%mm2                      \n\t"
02969             "psubusw %%mm0, %%mm2                   \n\t"
02970             "psubw %%mm2, %%mm4                     \n\t"
02971             "movq %%mm5, %%mm2                      \n\t"
02972             "psubusw %%mm1, %%mm2                   \n\t"
02973             "psubw %%mm2, %%mm5                     \n\t"
02974 #endif
02975             "pxor %%mm6, %%mm4                      \n\t"
02976             "pxor %%mm7, %%mm5                      \n\t"
02977             "psubw %%mm6, %%mm4                     \n\t"
02978             "psubw %%mm7, %%mm5                     \n\t"
02979             "packsswb %%mm5, %%mm4                  \n\t"
02980             "movq %3, %%mm1                         \n\t"
02981             "pandn %%mm4, %%mm1                     \n\t"
02982             "movq (%0), %%mm0                       \n\t"
02983             "paddb   %%mm1, %%mm0                   \n\t"
02984             "movq %%mm0, (%0)                       \n\t"
02985             "movq (%0, %1), %%mm0                   \n\t"
02986             "psubb %%mm1, %%mm0                     \n\t"
02987             "movq %%mm0, (%0, %1)                   \n\t"
02988 
02989             : "+r" (temp_src)
02990             : "r" ((x86_reg)step), "m" (c->pQPb), "m"(eq_mask), "r"(tmp)
02991             : "%"REG_a
02992         );
02993     }
02994 /*if(step==16){
02995     STOP_TIMER("step16")
02996 }else{
02997     STOP_TIMER("stepX")
02998 }
02999     } */
03000 }
03001 #endif //HAVE_MMX
03002 
03003 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
03004                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
03005 
03010 #undef REAL_SCALED_CPY
03011 #undef SCALED_CPY
03012 
03013 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
03014                                      int levelFix, int64_t *packedOffsetAndScale)
03015 {
03016 #if !HAVE_MMX
03017     int i;
03018 #endif
03019     if(levelFix){
03020 #if HAVE_MMX
03021     __asm__ volatile(
03022         "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
03023         "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
03024         "lea (%2,%4), %%"REG_a"         \n\t"
03025         "lea (%3,%5), %%"REG_d"         \n\t"
03026         "pxor %%mm4, %%mm4              \n\t"
03027 #if HAVE_MMX2
03028 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \
03029         "movq " #src1 ", %%mm0          \n\t"\
03030         "movq " #src1 ", %%mm5          \n\t"\
03031         "movq " #src2 ", %%mm1          \n\t"\
03032         "movq " #src2 ", %%mm6          \n\t"\
03033         "punpcklbw %%mm0, %%mm0         \n\t"\
03034         "punpckhbw %%mm5, %%mm5         \n\t"\
03035         "punpcklbw %%mm1, %%mm1         \n\t"\
03036         "punpckhbw %%mm6, %%mm6         \n\t"\
03037         "pmulhuw %%mm3, %%mm0           \n\t"\
03038         "pmulhuw %%mm3, %%mm5           \n\t"\
03039         "pmulhuw %%mm3, %%mm1           \n\t"\
03040         "pmulhuw %%mm3, %%mm6           \n\t"\
03041         "psubw %%mm2, %%mm0             \n\t"\
03042         "psubw %%mm2, %%mm5             \n\t"\
03043         "psubw %%mm2, %%mm1             \n\t"\
03044         "psubw %%mm2, %%mm6             \n\t"\
03045         "packuswb %%mm5, %%mm0          \n\t"\
03046         "packuswb %%mm6, %%mm1          \n\t"\
03047         "movq %%mm0, " #dst1 "          \n\t"\
03048         "movq %%mm1, " #dst2 "          \n\t"\
03049 
03050 #else //HAVE_MMX2
03051 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \
03052         "movq " #src1 ", %%mm0          \n\t"\
03053         "movq " #src1 ", %%mm5          \n\t"\
03054         "punpcklbw %%mm4, %%mm0         \n\t"\
03055         "punpckhbw %%mm4, %%mm5         \n\t"\
03056         "psubw %%mm2, %%mm0             \n\t"\
03057         "psubw %%mm2, %%mm5             \n\t"\
03058         "movq " #src2 ", %%mm1          \n\t"\
03059         "psllw $6, %%mm0                \n\t"\
03060         "psllw $6, %%mm5                \n\t"\
03061         "pmulhw %%mm3, %%mm0            \n\t"\
03062         "movq " #src2 ", %%mm6          \n\t"\
03063         "pmulhw %%mm3, %%mm5            \n\t"\
03064         "punpcklbw %%mm4, %%mm1         \n\t"\
03065         "punpckhbw %%mm4, %%mm6         \n\t"\
03066         "psubw %%mm2, %%mm1             \n\t"\
03067         "psubw %%mm2, %%mm6             \n\t"\
03068         "psllw $6, %%mm1                \n\t"\
03069         "psllw $6, %%mm6                \n\t"\
03070         "pmulhw %%mm3, %%mm1            \n\t"\
03071         "pmulhw %%mm3, %%mm6            \n\t"\
03072         "packuswb %%mm5, %%mm0          \n\t"\
03073         "packuswb %%mm6, %%mm1          \n\t"\
03074         "movq %%mm0, " #dst1 "          \n\t"\
03075         "movq %%mm1, " #dst2 "          \n\t"\
03076 
03077 #endif //HAVE_MMX2
03078 #define SCALED_CPY(src1, src2, dst1, dst2)\
03079    REAL_SCALED_CPY(src1, src2, dst1, dst2)
03080 
03081 SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
03082 SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
03083 SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
03084         "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
03085         "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
03086 SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
03087 
03088 
03089         : "=&a" (packedOffsetAndScale)
03090         : "0" (packedOffsetAndScale),
03091         "r"(src),
03092         "r"(dst),
03093         "r" ((x86_reg)srcStride),
03094         "r" ((x86_reg)dstStride)
03095         : "%"REG_d
03096     );
03097 #else //HAVE_MMX
03098     for(i=0; i<8; i++)
03099         memcpy( &(dst[dstStride*i]),
03100                 &(src[srcStride*i]), BLOCK_SIZE);
03101 #endif //HAVE_MMX
03102     }else{
03103 #if HAVE_MMX
03104     __asm__ volatile(
03105         "lea (%0,%2), %%"REG_a"                 \n\t"
03106         "lea (%1,%3), %%"REG_d"                 \n\t"
03107 
03108 #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \
03109         "movq " #src1 ", %%mm0          \n\t"\
03110         "movq " #src2 ", %%mm1          \n\t"\
03111         "movq %%mm0, " #dst1 "          \n\t"\
03112         "movq %%mm1, " #dst2 "          \n\t"\
03113 
03114 #define SIMPLE_CPY(src1, src2, dst1, dst2)\
03115    REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
03116 
03117 SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))
03118 SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
03119 SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
03120         "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
03121         "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
03122 SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
03123 
03124         : : "r" (src),
03125         "r" (dst),
03126         "r" ((x86_reg)srcStride),
03127         "r" ((x86_reg)dstStride)
03128         : "%"REG_a, "%"REG_d
03129     );
03130 #else //HAVE_MMX
03131     for(i=0; i<8; i++)
03132         memcpy( &(dst[dstStride*i]),
03133                 &(src[srcStride*i]), BLOCK_SIZE);
03134 #endif //HAVE_MMX
03135     }
03136 }
03137 
03141 static inline void RENAME(duplicate)(uint8_t src[], int stride)
03142 {
03143 #if HAVE_MMX
03144     __asm__ volatile(
03145         "movq (%0), %%mm0               \n\t"
03146         "add %1, %0                     \n\t"
03147         "movq %%mm0, (%0)               \n\t"
03148         "movq %%mm0, (%0, %1)           \n\t"
03149         "movq %%mm0, (%0, %1, 2)        \n\t"
03150         : "+r" (src)
03151         : "r" ((x86_reg)-stride)
03152     );
03153 #else
03154     int i;
03155     uint8_t *p=src;
03156     for(i=0; i<3; i++){
03157         p-= stride;
03158         memcpy(p, src, 8);
03159     }
03160 #endif
03161 }
03162 
03166 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
03167                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
03168 {
03169     DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
03170     int x,y;
03171 #ifdef COMPILE_TIME_MODE
03172     const int mode= COMPILE_TIME_MODE;
03173 #else
03174     const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
03175 #endif
03176     int black=0, white=255; // blackest black and whitest white in the picture
03177     int QPCorrecture= 256*256;
03178 
03179     int copyAhead;
03180 #if HAVE_MMX
03181     int i;
03182 #endif
03183 
03184     const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
03185     const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
03186 
03187     //FIXME remove
03188     uint64_t * const yHistogram= c.yHistogram;
03189     uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
03190     uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
03191     //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
03192 
03193 #if HAVE_MMX
03194     for(i=0; i<57; i++){
03195         int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
03196         int threshold= offset*2 + 1;
03197         c.mmxDcOffset[i]= 0x7F - offset;
03198         c.mmxDcThreshold[i]= 0x7F - threshold;
03199         c.mmxDcOffset[i]*= 0x0101010101010101LL;
03200         c.mmxDcThreshold[i]*= 0x0101010101010101LL;
03201     }
03202 #endif
03203 
03204     if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
03205     else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
03206             || (mode & FFMPEG_DEINT_FILTER)
03207             || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
03208     else if(   (mode & V_DEBLOCK)
03209             || (mode & LINEAR_IPOL_DEINT_FILTER)
03210             || (mode & MEDIAN_DEINT_FILTER)
03211             || (mode & V_A_DEBLOCK)) copyAhead=13;
03212     else if(mode & V_X1_FILTER) copyAhead=11;
03213 //    else if(mode & V_RK1_FILTER) copyAhead=10;
03214     else if(mode & DERING) copyAhead=9;
03215     else copyAhead=8;
03216 
03217     copyAhead-= 8;
03218 
03219     if(!isColor){
03220         uint64_t sum= 0;
03221         int i;
03222         uint64_t maxClipped;
03223         uint64_t clipped;
03224         double scale;
03225 
03226         c.frameNum++;
03227         // first frame is fscked so we ignore it
03228         if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
03229 
03230         for(i=0; i<256; i++){
03231             sum+= yHistogram[i];
03232         }
03233 
03234         /* We always get a completely black picture first. */
03235         maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
03236 
03237         clipped= sum;
03238         for(black=255; black>0; black--){
03239             if(clipped < maxClipped) break;
03240             clipped-= yHistogram[black];
03241         }
03242 
03243         clipped= sum;
03244         for(white=0; white<256; white++){
03245             if(clipped < maxClipped) break;
03246             clipped-= yHistogram[white];
03247         }
03248 
03249         scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
03250 
03251 #if HAVE_MMX2
03252         c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
03253         c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
03254 #else
03255         c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
03256         c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
03257 #endif
03258 
03259         c.packedYOffset|= c.packedYOffset<<32;
03260         c.packedYOffset|= c.packedYOffset<<16;
03261 
03262         c.packedYScale|= c.packedYScale<<32;
03263         c.packedYScale|= c.packedYScale<<16;
03264 
03265         if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);
03266         else                        QPCorrecture= 256*256;
03267     }else{
03268         c.packedYScale= 0x0100010001000100LL;
03269         c.packedYOffset= 0;
03270         QPCorrecture= 256*256;
03271     }
03272 
03273     /* copy & deinterlace first row of blocks */
03274     y=-BLOCK_SIZE;
03275     {
03276         const uint8_t *srcBlock= &(src[y*srcStride]);
03277         uint8_t *dstBlock= tempDst + dstStride;
03278 
03279         // From this point on it is guaranteed that we can read and write 16 lines downward
03280         // finish 1 block before the next otherwise we might have a problem
03281         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
03282         for(x=0; x<width; x+=BLOCK_SIZE){
03283 
03284 #if HAVE_MMX2
03285 /*
03286             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
03287             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
03288             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
03289             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
03290 */
03291 
03292             __asm__(
03293                 "mov %4, %%"REG_a"              \n\t"
03294                 "shr $2, %%"REG_a"              \n\t"
03295                 "and $6, %%"REG_a"              \n\t"
03296                 "add %5, %%"REG_a"              \n\t"
03297                 "mov %%"REG_a", %%"REG_d"       \n\t"
03298                 "imul %1, %%"REG_a"             \n\t"
03299                 "imul %3, %%"REG_d"             \n\t"
03300                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03301                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03302                 "add %1, %%"REG_a"              \n\t"
03303                 "add %3, %%"REG_d"              \n\t"
03304                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03305                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03306                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
03307                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
03308                 : "%"REG_a, "%"REG_d
03309             );
03310 
03311 #elif HAVE_AMD3DNOW
03312 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
03313 /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
03314             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
03315             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
03316             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
03317 */
03318 #endif
03319 
03320             RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
03321                               srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
03322 
03323             RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
03324 
03325             if(mode & LINEAR_IPOL_DEINT_FILTER)
03326                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
03327             else if(mode & LINEAR_BLEND_DEINT_FILTER)
03328                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
03329             else if(mode & MEDIAN_DEINT_FILTER)
03330                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
03331             else if(mode & CUBIC_IPOL_DEINT_FILTER)
03332                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
03333             else if(mode & FFMPEG_DEINT_FILTER)
03334                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
03335             else if(mode & LOWPASS5_DEINT_FILTER)
03336                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
03337 /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
03338                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
03339 */
03340             dstBlock+=8;
03341             srcBlock+=8;
03342         }
03343         if(width==FFABS(dstStride))
03344             linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
03345         else{
03346             int i;
03347             for(i=0; i<copyAhead; i++){
03348                 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
03349             }
03350         }
03351     }
03352 
03353     for(y=0; y<height; y+=BLOCK_SIZE){
03354         //1% speedup if these are here instead of the inner loop
03355         const uint8_t *srcBlock= &(src[y*srcStride]);
03356         uint8_t *dstBlock= &(dst[y*dstStride]);
03357 #if HAVE_MMX
03358         uint8_t *tempBlock1= c.tempBlocks;
03359         uint8_t *tempBlock2= c.tempBlocks + 8;
03360 #endif
03361         const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
03362         int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
03363         int QP=0;
03364         /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
03365            if not than use a temporary buffer */
03366         if(y+15 >= height){
03367             int i;
03368             /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
03369                blockcopy to dst later */
03370             linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
03371                     FFMAX(height-y-copyAhead, 0), srcStride);
03372 
03373             /* duplicate last line of src to fill the void up to line (copyAhead+7) */
03374             for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
03375                     memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
03376 
03377             /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
03378             linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
03379 
03380             /* duplicate last line of dst to fill the void up to line (copyAhead) */
03381             for(i=height-y+1; i<=copyAhead; i++)
03382                     memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
03383 
03384             dstBlock= tempDst + dstStride;
03385             srcBlock= tempSrc;
03386         }
03387 
03388         // From this point on it is guaranteed that we can read and write 16 lines downward
03389         // finish 1 block before the next otherwise we might have a problem
03390         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
03391         for(x=0; x<width; x+=BLOCK_SIZE){
03392             const int stride= dstStride;
03393 #if HAVE_MMX
03394             uint8_t *tmpXchg;
03395 #endif
03396             if(isColor){
03397                 QP= QPptr[x>>qpHShift];
03398                 c.nonBQP= nonBQPptr[x>>qpHShift];
03399             }else{
03400                 QP= QPptr[x>>4];
03401                 QP= (QP* QPCorrecture + 256*128)>>16;
03402                 c.nonBQP= nonBQPptr[x>>4];
03403                 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
03404                 yHistogram[ srcBlock[srcStride*12 + 4] ]++;
03405             }
03406             c.QP= QP;
03407 #if HAVE_MMX
03408             __asm__ volatile(
03409                 "movd %1, %%mm7         \n\t"
03410                 "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
03411                 "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
03412                 "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP
03413                 "movq %%mm7, %0         \n\t"
03414                 : "=m" (c.pQPb)
03415                 : "r" (QP)
03416             );
03417 #endif
03418 
03419 
03420 #if HAVE_MMX2
03421 /*
03422             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
03423             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
03424             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
03425             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
03426 */
03427 
03428             __asm__(
03429                 "mov %4, %%"REG_a"              \n\t"
03430                 "shr $2, %%"REG_a"              \n\t"
03431                 "and $6, %%"REG_a"              \n\t"
03432                 "add %5, %%"REG_a"              \n\t"
03433                 "mov %%"REG_a", %%"REG_d"       \n\t"
03434                 "imul %1, %%"REG_a"             \n\t"
03435                 "imul %3, %%"REG_d"             \n\t"
03436                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03437                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03438                 "add %1, %%"REG_a"              \n\t"
03439                 "add %3, %%"REG_d"              \n\t"
03440                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03441                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03442                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
03443                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
03444                 : "%"REG_a, "%"REG_d
03445             );
03446 
03447 #elif HAVE_AMD3DNOW
03448 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
03449 /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
03450             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
03451             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
03452             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
03453 */
03454 #endif
03455 
03456             RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
03457                               srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
03458 
03459             if(mode & LINEAR_IPOL_DEINT_FILTER)
03460                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
03461             else if(mode & LINEAR_BLEND_DEINT_FILTER)
03462                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
03463             else if(mode & MEDIAN_DEINT_FILTER)
03464                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
03465             else if(mode & CUBIC_IPOL_DEINT_FILTER)
03466                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
03467             else if(mode & FFMPEG_DEINT_FILTER)
03468                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
03469             else if(mode & LOWPASS5_DEINT_FILTER)
03470                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
03471 /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
03472                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
03473 */
03474 
03475             /* only deblock if we have 2 blocks */
03476             if(y + 8 < height){
03477                 if(mode & V_X1_FILTER)
03478                     RENAME(vertX1Filter)(dstBlock, stride, &c);
03479                 else if(mode & V_DEBLOCK){
03480                     const int t= RENAME(vertClassify)(dstBlock, stride, &c);
03481 
03482                     if(t==1)
03483                         RENAME(doVertLowPass)(dstBlock, stride, &c);
03484                     else if(t==2)
03485                         RENAME(doVertDefFilter)(dstBlock, stride, &c);
03486                 }else if(mode & V_A_DEBLOCK){
03487                     RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
03488                 }
03489             }
03490 
03491 #if HAVE_MMX
03492             RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
03493 #endif
03494             /* check if we have a previous block to deblock it with dstBlock */
03495             if(x - 8 >= 0){
03496 #if HAVE_MMX
03497                 if(mode & H_X1_FILTER)
03498                         RENAME(vertX1Filter)(tempBlock1, 16, &c);
03499                 else if(mode & H_DEBLOCK){
03500 //START_TIMER
03501                     const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
03502 //STOP_TIMER("dc & minmax")
03503                     if(t==1)
03504                         RENAME(doVertLowPass)(tempBlock1, 16, &c);
03505                     else if(t==2)
03506                         RENAME(doVertDefFilter)(tempBlock1, 16, &c);
03507                 }else if(mode & H_A_DEBLOCK){
03508                         RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
03509                 }
03510 
03511                 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
03512 
03513 #else
03514                 if(mode & H_X1_FILTER)
03515                     horizX1Filter(dstBlock-4, stride, QP);
03516                 else if(mode & H_DEBLOCK){
03517 #if HAVE_ALTIVEC
03518                     DECLARE_ALIGNED(16, unsigned char, tempBlock)[272];
03519                     int t;
03520                     transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
03521 
03522                     t = vertClassify_altivec(tempBlock-48, 16, &c);
03523                     if(t==1) {
03524                         doVertLowPass_altivec(tempBlock-48, 16, &c);
03525                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
03526                     }
03527                     else if(t==2) {
03528                         doVertDefFilter_altivec(tempBlock-48, 16, &c);
03529                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
03530                     }
03531 #else
03532                     const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
03533 
03534                     if(t==1)
03535                         RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
03536                     else if(t==2)
03537                         RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
03538 #endif
03539                 }else if(mode & H_A_DEBLOCK){
03540                     RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
03541                 }
03542 #endif //HAVE_MMX
03543                 if(mode & DERING){
03544                 //FIXME filter first line
03545                     if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
03546                 }
03547 
03548                 if(mode & TEMP_NOISE_FILTER)
03549                 {
03550                     RENAME(tempNoiseReducer)(dstBlock-8, stride,
03551                             c.tempBlurred[isColor] + y*dstStride + x,
03552                             c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256,
03553                             c.ppMode.maxTmpNoise);
03554                 }
03555             }
03556 
03557             dstBlock+=8;
03558             srcBlock+=8;
03559 
03560 #if HAVE_MMX
03561             tmpXchg= tempBlock1;
03562             tempBlock1= tempBlock2;
03563             tempBlock2 = tmpXchg;
03564 #endif
03565         }
03566 
03567         if(mode & DERING){
03568             if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
03569         }
03570 
03571         if((mode & TEMP_NOISE_FILTER)){
03572             RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
03573                     c.tempBlurred[isColor] + y*dstStride + x,
03574                     c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256,
03575                     c.ppMode.maxTmpNoise);
03576         }
03577 
03578         /* did we use a tmp buffer for the last lines*/
03579         if(y+15 >= height){
03580             uint8_t *dstBlock= &(dst[y*dstStride]);
03581             if(width==FFABS(dstStride))
03582                 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
03583             else{
03584                 int i;
03585                 for(i=0; i<height-y; i++){
03586                     memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
03587                 }
03588             }
03589         }
03590 /*
03591         for(x=0; x<width; x+=32){
03592             volatile int i;
03593             i+=   dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
03594                 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
03595                 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
03596                 + dstBlock[x +13*dstStride]
03597                 + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
03598         }*/
03599     }
03600 #if   HAVE_AMD3DNOW
03601     __asm__ volatile("femms");
03602 #elif HAVE_MMX
03603     __asm__ volatile("emms");
03604 #endif
03605 
03606 #ifdef DEBUG_BRIGHTNESS
03607     if(!isColor){
03608         int max=1;
03609         int i;
03610         for(i=0; i<256; i++)
03611             if(yHistogram[i] > max) max=yHistogram[i];
03612 
03613         for(i=1; i<256; i++){
03614             int x;
03615             int start=yHistogram[i-1]/(max/256+1);
03616             int end=yHistogram[i]/(max/256+1);
03617             int inc= end > start ? 1 : -1;
03618             for(x=start; x!=end+inc; x+=inc)
03619                 dst[ i*dstStride + x]+=128;
03620         }
03621 
03622         for(i=0; i<100; i+=2){
03623             dst[ (white)*dstStride + i]+=128;
03624             dst[ (black)*dstStride + i]+=128;
03625         }
03626     }
03627 #endif
03628 
03629     *c2= c; //copy local context back
03630 
03631 }