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 #define ALIGN_MASK "$-8"
00029 
00030 #undef PAVGB
00031 #undef PMINUB
00032 #undef PMAXUB
00033 
00034 #if   HAVE_MMX2
00035 #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
00036 #elif HAVE_AMD3DNOW
00037 #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
00038 #endif
00039 #define PAVGB(a,b)  REAL_PAVGB(a,b)
00040 
00041 #if   HAVE_MMX2
00042 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
00043 #elif HAVE_MMX
00044 #define PMINUB(b,a,t) \
00045     "movq " #a ", " #t " \n\t"\
00046     "psubusb " #b ", " #t " \n\t"\
00047     "psubb " #t ", " #a " \n\t"
00048 #endif
00049 
00050 #if   HAVE_MMX2
00051 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
00052 #elif HAVE_MMX
00053 #define PMAXUB(a,b) \
00054     "psubusb " #a ", " #b " \n\t"\
00055     "paddb " #a ", " #b " \n\t"
00056 #endif
00057 
00058 //FIXME? |255-0| = 1 (should not be a problem ...)
00059 #if HAVE_MMX
00060 
00063 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
00064     int numEq= 0, dcOk;
00065     src+= stride*4; // src points to begin of the 8x8 Block
00066     __asm__ volatile(
00067         "movq %0, %%mm7                         \n\t"
00068         "movq %1, %%mm6                         \n\t"
00069         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
00070         );
00071 
00072     __asm__ volatile(
00073         "lea (%2, %3), %%"REG_a"                \n\t"
00074 //      0       1       2       3       4       5       6       7       8       9
00075 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
00076 
00077         "movq (%2), %%mm0                       \n\t"
00078         "movq (%%"REG_a"), %%mm1                \n\t"
00079         "movq %%mm0, %%mm3                      \n\t"
00080         "movq %%mm0, %%mm4                      \n\t"
00081         PMAXUB(%%mm1, %%mm4)
00082         PMINUB(%%mm1, %%mm3, %%mm5)
00083         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
00084         "paddb %%mm7, %%mm0                     \n\t"
00085         "pcmpgtb %%mm6, %%mm0                   \n\t"
00086 
00087         "movq (%%"REG_a",%3), %%mm2             \n\t"
00088         PMAXUB(%%mm2, %%mm4)
00089         PMINUB(%%mm2, %%mm3, %%mm5)
00090         "psubb %%mm2, %%mm1                     \n\t"
00091         "paddb %%mm7, %%mm1                     \n\t"
00092         "pcmpgtb %%mm6, %%mm1                   \n\t"
00093         "paddb %%mm1, %%mm0                     \n\t"
00094 
00095         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
00096         PMAXUB(%%mm1, %%mm4)
00097         PMINUB(%%mm1, %%mm3, %%mm5)
00098         "psubb %%mm1, %%mm2                     \n\t"
00099         "paddb %%mm7, %%mm2                     \n\t"
00100         "pcmpgtb %%mm6, %%mm2                   \n\t"
00101         "paddb %%mm2, %%mm0                     \n\t"
00102 
00103         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
00104 
00105         "movq (%2, %3, 4), %%mm2                \n\t"
00106         PMAXUB(%%mm2, %%mm4)
00107         PMINUB(%%mm2, %%mm3, %%mm5)
00108         "psubb %%mm2, %%mm1                     \n\t"
00109         "paddb %%mm7, %%mm1                     \n\t"
00110         "pcmpgtb %%mm6, %%mm1                   \n\t"
00111         "paddb %%mm1, %%mm0                     \n\t"
00112 
00113         "movq (%%"REG_a"), %%mm1                \n\t"
00114         PMAXUB(%%mm1, %%mm4)
00115         PMINUB(%%mm1, %%mm3, %%mm5)
00116         "psubb %%mm1, %%mm2                     \n\t"
00117         "paddb %%mm7, %%mm2                     \n\t"
00118         "pcmpgtb %%mm6, %%mm2                   \n\t"
00119         "paddb %%mm2, %%mm0                     \n\t"
00120 
00121         "movq (%%"REG_a", %3), %%mm2            \n\t"
00122         PMAXUB(%%mm2, %%mm4)
00123         PMINUB(%%mm2, %%mm3, %%mm5)
00124         "psubb %%mm2, %%mm1                     \n\t"
00125         "paddb %%mm7, %%mm1                     \n\t"
00126         "pcmpgtb %%mm6, %%mm1                   \n\t"
00127         "paddb %%mm1, %%mm0                     \n\t"
00128 
00129         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
00130         PMAXUB(%%mm1, %%mm4)
00131         PMINUB(%%mm1, %%mm3, %%mm5)
00132         "psubb %%mm1, %%mm2                     \n\t"
00133         "paddb %%mm7, %%mm2                     \n\t"
00134         "pcmpgtb %%mm6, %%mm2                   \n\t"
00135         "paddb %%mm2, %%mm0                     \n\t"
00136         "psubusb %%mm3, %%mm4                   \n\t"
00137 
00138         "                                       \n\t"
00139 #if HAVE_MMX2
00140         "pxor %%mm7, %%mm7                      \n\t"
00141         "psadbw %%mm7, %%mm0                    \n\t"
00142 #else
00143         "movq %%mm0, %%mm1                      \n\t"
00144         "psrlw $8, %%mm0                        \n\t"
00145         "paddb %%mm1, %%mm0                     \n\t"
00146         "movq %%mm0, %%mm1                      \n\t"
00147         "psrlq $16, %%mm0                       \n\t"
00148         "paddb %%mm1, %%mm0                     \n\t"
00149         "movq %%mm0, %%mm1                      \n\t"
00150         "psrlq $32, %%mm0                       \n\t"
00151         "paddb %%mm1, %%mm0                     \n\t"
00152 #endif
00153         "movq %4, %%mm7                         \n\t" // QP,..., QP
00154         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
00155         "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
00156         "packssdw %%mm4, %%mm4                  \n\t"
00157         "movd %%mm0, %0                         \n\t"
00158         "movd %%mm4, %1                         \n\t"
00159 
00160         : "=r" (numEq), "=r" (dcOk)
00161         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00162         : "%"REG_a
00163         );
00164 
00165     numEq= (-numEq) &0xFF;
00166     if(numEq > c->ppMode.flatnessThreshold){
00167         if(dcOk) return 0;
00168         else     return 1;
00169     }else{
00170         return 2;
00171     }
00172 }
00173 #endif //HAVE_MMX
00174 
00179 #if !HAVE_ALTIVEC
00180 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
00181 {
00182 #if HAVE_MMX2 || HAVE_AMD3DNOW
00183     src+= stride*3;
00184     __asm__ volatile(        //"movv %0 %1 %2\n\t"
00185         "movq %2, %%mm0                         \n\t"  // QP,..., QP
00186         "pxor %%mm4, %%mm4                      \n\t"
00187 
00188         "movq (%0), %%mm6                       \n\t"
00189         "movq (%0, %1), %%mm5                   \n\t"
00190         "movq %%mm5, %%mm1                      \n\t"
00191         "movq %%mm6, %%mm2                      \n\t"
00192         "psubusb %%mm6, %%mm5                   \n\t"
00193         "psubusb %%mm1, %%mm2                   \n\t"
00194         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
00195         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
00196         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
00197 
00198         "pand %%mm2, %%mm6                      \n\t"
00199         "pandn %%mm1, %%mm2                     \n\t"
00200         "por %%mm2, %%mm6                       \n\t"// First Line to Filter
00201 
00202         "movq (%0, %1, 8), %%mm5                \n\t"
00203         "lea (%0, %1, 4), %%"REG_a"             \n\t"
00204         "lea (%0, %1, 8), %%"REG_c"             \n\t"
00205         "sub %1, %%"REG_c"                      \n\t"
00206         "add %1, %0                             \n\t" // %0 points to line 1 not 0
00207         "movq (%0, %1, 8), %%mm7                \n\t"
00208         "movq %%mm5, %%mm1                      \n\t"
00209         "movq %%mm7, %%mm2                      \n\t"
00210         "psubusb %%mm7, %%mm5                   \n\t"
00211         "psubusb %%mm1, %%mm2                   \n\t"
00212         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
00213         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
00214         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
00215 
00216         "pand %%mm2, %%mm7                      \n\t"
00217         "pandn %%mm1, %%mm2                     \n\t"
00218         "por %%mm2, %%mm7                       \n\t" // First Line to Filter
00219 
00220 
00221         //      1       2       3       4       5       6       7       8
00222         //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
00223         // 6 4 2 2 1 1
00224         // 6 4 4 2
00225         // 6 8 2
00226 
00227         "movq (%0, %1), %%mm0                   \n\t" //  1
00228         "movq %%mm0, %%mm1                      \n\t" //  1
00229         PAVGB(%%mm6, %%mm0)                           //1 1        /2
00230         PAVGB(%%mm6, %%mm0)                           //3 1        /4
00231 
00232         "movq (%0, %1, 4), %%mm2                \n\t" //     1
00233         "movq %%mm2, %%mm5                      \n\t" //     1
00234         PAVGB((%%REGa), %%mm2)                        //    11        /2
00235         PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
00236         "movq %%mm2, %%mm3                      \n\t" //   211        /4
00237         "movq (%0), %%mm4                       \n\t" // 1
00238         PAVGB(%%mm4, %%mm3)                           // 4 211        /8
00239         PAVGB(%%mm0, %%mm3)                           //642211        /16
00240         "movq %%mm3, (%0)                       \n\t" // X
00241         // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
00242         "movq %%mm1, %%mm0                      \n\t" //  1
00243         PAVGB(%%mm6, %%mm0)                           //1 1        /2
00244         "movq %%mm4, %%mm3                      \n\t" // 1
00245         PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
00246         PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
00247         PAVGB((%%REGa), %%mm5)                        //    211 /4
00248         PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
00249         PAVGB(%%mm0, %%mm3)                           //4242211 /16
00250         "movq %%mm3, (%0,%1)                    \n\t" //  X
00251         // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
00252         PAVGB(%%mm4, %%mm6)                                   //11        /2
00253         "movq (%%"REG_c"), %%mm0                \n\t" //       1
00254         PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
00255         "movq %%mm0, %%mm3                      \n\t" //      11/2
00256         PAVGB(%%mm1, %%mm0)                           //  2   11/4
00257         PAVGB(%%mm6, %%mm0)                           //222   11/8
00258         PAVGB(%%mm2, %%mm0)                           //22242211/16
00259         "movq (%0, %1, 2), %%mm2                \n\t" //   1
00260         "movq %%mm0, (%0, %1, 2)                \n\t" //   X
00261         // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
00262         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
00263         PAVGB((%%REGc), %%mm0)                        //       11        /2
00264         PAVGB(%%mm0, %%mm6)                           //11     11        /4
00265         PAVGB(%%mm1, %%mm4)                           // 11                /2
00266         PAVGB(%%mm2, %%mm1)                           //  11                /2
00267         PAVGB(%%mm1, %%mm6)                           //1122   11        /8
00268         PAVGB(%%mm5, %%mm6)                           //112242211        /16
00269         "movq (%%"REG_a"), %%mm5                \n\t" //    1
00270         "movq %%mm6, (%%"REG_a")                \n\t" //    X
00271         // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
00272         "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
00273         PAVGB(%%mm7, %%mm6)                           //        11        /2
00274         PAVGB(%%mm4, %%mm6)                           // 11     11        /4
00275         PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
00276         PAVGB(%%mm5, %%mm2)                           //   11                /2
00277         "movq (%0, %1, 4), %%mm4                \n\t" //     1
00278         PAVGB(%%mm4, %%mm2)                           //   112                /4
00279         PAVGB(%%mm2, %%mm6)                           // 112242211        /16
00280         "movq %%mm6, (%0, %1, 4)                \n\t" //     X
00281         // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
00282         PAVGB(%%mm7, %%mm1)                           //  11     2        /4
00283         PAVGB(%%mm4, %%mm5)                           //    11                /2
00284         PAVGB(%%mm5, %%mm0)                           //    11 11        /4
00285         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
00286         PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
00287         PAVGB(%%mm0, %%mm1)                           //  11224222        /16
00288         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
00289         // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
00290         PAVGB((%%REGc), %%mm2)                        //   112 4        /8
00291         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
00292         PAVGB(%%mm0, %%mm6)                           //      1 1        /2
00293         PAVGB(%%mm7, %%mm6)                           //      1 12        /4
00294         PAVGB(%%mm2, %%mm6)                           //   1122424        /4
00295         "movq %%mm6, (%%"REG_c")                \n\t" //       X
00296         // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
00297         PAVGB(%%mm7, %%mm5)                           //    11   2        /4
00298         PAVGB(%%mm7, %%mm5)                           //    11   6        /8
00299 
00300         PAVGB(%%mm3, %%mm0)                           //      112        /4
00301         PAVGB(%%mm0, %%mm5)                           //    112246        /16
00302         "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
00303         "sub %1, %0                             \n\t"
00304 
00305         :
00306         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00307         : "%"REG_a, "%"REG_c
00308     );
00309 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00310     const int l1= stride;
00311     const int l2= stride + l1;
00312     const int l3= stride + l2;
00313     const int l4= stride + l3;
00314     const int l5= stride + l4;
00315     const int l6= stride + l5;
00316     const int l7= stride + l6;
00317     const int l8= stride + l7;
00318     const int l9= stride + l8;
00319     int x;
00320     src+= stride*3;
00321     for(x=0; x<BLOCK_SIZE; x++){
00322         const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
00323         const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
00324 
00325         int sums[10];
00326         sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
00327         sums[1] = sums[0] - first  + src[l4];
00328         sums[2] = sums[1] - first  + src[l5];
00329         sums[3] = sums[2] - first  + src[l6];
00330         sums[4] = sums[3] - first  + src[l7];
00331         sums[5] = sums[4] - src[l1] + src[l8];
00332         sums[6] = sums[5] - src[l2] + last;
00333         sums[7] = sums[6] - src[l3] + last;
00334         sums[8] = sums[7] - src[l4] + last;
00335         sums[9] = sums[8] - src[l5] + last;
00336 
00337         src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
00338         src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
00339         src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
00340         src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
00341         src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
00342         src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
00343         src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
00344         src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
00345 
00346         src++;
00347     }
00348 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
00349 }
00350 #endif //HAVE_ALTIVEC
00351 
00352 #if 0
00353 
00365 static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
00366 {
00367 #if HAVE_MMX2 || HAVE_AMD3DNOW
00368     src+= stride*3;
00369 // FIXME rounding
00370     __asm__ volatile(
00371         "pxor %%mm7, %%mm7                      \n\t" // 0
00372         "movq "MANGLE(b80)", %%mm6              \n\t" // MIN_SIGNED_BYTE
00373         "leal (%0, %1), %%"REG_a"               \n\t"
00374         "leal (%%"REG_a", %1, 4), %%"REG_c"     \n\t"
00375 //      0       1       2       3       4       5       6       7       8       9
00376 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
00377         "movq "MANGLE(pQPb)", %%mm0             \n\t" // QP,..., QP
00378         "movq %%mm0, %%mm1                      \n\t" // QP,..., QP
00379         "paddusb "MANGLE(b02)", %%mm0           \n\t"
00380         "psrlw $2, %%mm0                        \n\t"
00381         "pand "MANGLE(b3F)", %%mm0              \n\t" // QP/4,..., QP/4
00382         "paddusb %%mm1, %%mm0                   \n\t" // QP*1.25 ...
00383         "movq (%0, %1, 4), %%mm2                \n\t" // line 4
00384         "movq (%%"REG_c"), %%mm3                \n\t" // line 5
00385         "movq %%mm2, %%mm4                      \n\t" // line 4
00386         "pcmpeqb %%mm5, %%mm5                   \n\t" // -1
00387         "pxor %%mm2, %%mm5                      \n\t" // -line 4 - 1
00388         PAVGB(%%mm3, %%mm5)
00389         "paddb %%mm6, %%mm5                     \n\t" // (l5-l4)/2
00390         "psubusb %%mm3, %%mm4                   \n\t"
00391         "psubusb %%mm2, %%mm3                   \n\t"
00392         "por %%mm3, %%mm4                       \n\t" // |l4 - l5|
00393         "psubusb %%mm0, %%mm4                   \n\t"
00394         "pcmpeqb %%mm7, %%mm4                   \n\t"
00395         "pand %%mm4, %%mm5                      \n\t" // d/2
00396 
00397 //        "paddb %%mm6, %%mm2                     \n\t" // line 4 + 0x80
00398         "paddb %%mm5, %%mm2                     \n\t"
00399 //        "psubb %%mm6, %%mm2                     \n\t"
00400         "movq %%mm2, (%0,%1, 4)                 \n\t"
00401 
00402         "movq (%%"REG_c"), %%mm2                \n\t"
00403 //        "paddb %%mm6, %%mm2                     \n\t" // line 5 + 0x80
00404         "psubb %%mm5, %%mm2                     \n\t"
00405 //        "psubb %%mm6, %%mm2                     \n\t"
00406         "movq %%mm2, (%%"REG_c")                \n\t"
00407 
00408         "paddb %%mm6, %%mm5                     \n\t"
00409         "psrlw $2, %%mm5                        \n\t"
00410         "pand "MANGLE(b3F)", %%mm5              \n\t"
00411         "psubb "MANGLE(b20)", %%mm5             \n\t" // (l5-l4)/8
00412 
00413         "movq (%%"REG_a", %1, 2), %%mm2         \n\t"
00414         "paddb %%mm6, %%mm2                     \n\t" // line 3 + 0x80
00415         "paddsb %%mm5, %%mm2                    \n\t"
00416         "psubb %%mm6, %%mm2                     \n\t"
00417         "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
00418 
00419         "movq (%%"REG_c", %1), %%mm2            \n\t"
00420         "paddb %%mm6, %%mm2                     \n\t" // line 6 + 0x80
00421         "psubsb %%mm5, %%mm2                    \n\t"
00422         "psubb %%mm6, %%mm2                     \n\t"
00423         "movq %%mm2, (%%"REG_c", %1)            \n\t"
00424 
00425         :
00426         : "r" (src), "r" ((x86_reg)stride)
00427         : "%"REG_a, "%"REG_c
00428     );
00429 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00430     const int l1= stride;
00431     const int l2= stride + l1;
00432     const int l3= stride + l2;
00433     const int l4= stride + l3;
00434     const int l5= stride + l4;
00435     const int l6= stride + l5;
00436 //    const int l7= stride + l6;
00437 //    const int l8= stride + l7;
00438 //    const int l9= stride + l8;
00439     int x;
00440     const int QP15= QP + (QP>>2);
00441     src+= stride*3;
00442     for(x=0; x<BLOCK_SIZE; x++){
00443         const int v = (src[x+l5] - src[x+l4]);
00444         if(FFABS(v) < QP15){
00445             src[x+l3] +=v>>3;
00446             src[x+l4] +=v>>1;
00447             src[x+l5] -=v>>1;
00448             src[x+l6] -=v>>3;
00449         }
00450     }
00451 
00452 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
00453 }
00454 #endif //0
00455 
00463 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
00464 {
00465 #if HAVE_MMX2 || HAVE_AMD3DNOW
00466     src+= stride*3;
00467 
00468     __asm__ volatile(
00469         "pxor %%mm7, %%mm7                      \n\t" // 0
00470         "lea (%0, %1), %%"REG_a"                \n\t"
00471         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00472 //      0       1       2       3       4       5       6       7       8       9
00473 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
00474         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
00475         "movq (%0, %1, 4), %%mm1                \n\t" // line 4
00476         "movq %%mm1, %%mm2                      \n\t" // line 4
00477         "psubusb %%mm0, %%mm1                   \n\t"
00478         "psubusb %%mm2, %%mm0                   \n\t"
00479         "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
00480         "movq (%%"REG_c"), %%mm3                \n\t" // line 5
00481         "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
00482         "movq %%mm3, %%mm5                      \n\t" // line 5
00483         "psubusb %%mm4, %%mm3                   \n\t"
00484         "psubusb %%mm5, %%mm4                   \n\t"
00485         "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
00486         PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
00487         "movq %%mm2, %%mm1                      \n\t" // line 4
00488         "psubusb %%mm5, %%mm2                   \n\t"
00489         "movq %%mm2, %%mm4                      \n\t"
00490         "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
00491         "psubusb %%mm1, %%mm5                   \n\t"
00492         "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
00493         "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
00494         "movq %%mm4, %%mm3                      \n\t" // d
00495         "movq %2, %%mm0                         \n\t"
00496         "paddusb %%mm0, %%mm0                   \n\t"
00497         "psubusb %%mm0, %%mm4                   \n\t"
00498         "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
00499         "psubusb "MANGLE(b01)", %%mm3           \n\t"
00500         "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
00501 
00502         PAVGB(%%mm7, %%mm3)                           // d/2
00503         "movq %%mm3, %%mm1                      \n\t" // d/2
00504         PAVGB(%%mm7, %%mm3)                           // d/4
00505         PAVGB(%%mm1, %%mm3)                           // 3*d/8
00506 
00507         "movq (%0, %1, 4), %%mm0                \n\t" // line 4
00508         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
00509         "psubusb %%mm3, %%mm0                   \n\t"
00510         "pxor %%mm2, %%mm0                      \n\t"
00511         "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
00512 
00513         "movq (%%"REG_c"), %%mm0                \n\t" // line 5
00514         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
00515         "paddusb %%mm3, %%mm0                   \n\t"
00516         "pxor %%mm2, %%mm0                      \n\t"
00517         "movq %%mm0, (%%"REG_c")                \n\t" // line 5
00518 
00519         PAVGB(%%mm7, %%mm1)                           // d/4
00520 
00521         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
00522         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
00523         "psubusb %%mm1, %%mm0                   \n\t"
00524         "pxor %%mm2, %%mm0                      \n\t"
00525         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
00526 
00527         "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
00528         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
00529         "paddusb %%mm1, %%mm0                   \n\t"
00530         "pxor %%mm2, %%mm0                      \n\t"
00531         "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
00532 
00533         PAVGB(%%mm7, %%mm1)                           // d/8
00534 
00535         "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
00536         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
00537         "psubusb %%mm1, %%mm0                   \n\t"
00538         "pxor %%mm2, %%mm0                      \n\t"
00539         "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
00540 
00541         "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
00542         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
00543         "paddusb %%mm1, %%mm0                   \n\t"
00544         "pxor %%mm2, %%mm0                      \n\t"
00545         "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
00546 
00547         :
00548         : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
00549         : "%"REG_a, "%"REG_c
00550     );
00551 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00552 
00553     const int l1= stride;
00554     const int l2= stride + l1;
00555     const int l3= stride + l2;
00556     const int l4= stride + l3;
00557     const int l5= stride + l4;
00558     const int l6= stride + l5;
00559     const int l7= stride + l6;
00560 //    const int l8= stride + l7;
00561 //    const int l9= stride + l8;
00562     int x;
00563 
00564     src+= stride*3;
00565     for(x=0; x<BLOCK_SIZE; x++){
00566         int a= src[l3] - src[l4];
00567         int b= src[l4] - src[l5];
00568         int c= src[l5] - src[l6];
00569 
00570         int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
00571         d= FFMAX(d, 0);
00572 
00573         if(d < co->QP*2){
00574             int v = d * FFSIGN(-b);
00575 
00576             src[l2] +=v>>3;
00577             src[l3] +=v>>2;
00578             src[l4] +=(3*v)>>3;
00579             src[l5] -=(3*v)>>3;
00580             src[l6] -=v>>2;
00581             src[l7] -=v>>3;
00582         }
00583         src++;
00584     }
00585 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
00586 }
00587 
00588 #if !HAVE_ALTIVEC
00589 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
00590 {
00591 #if HAVE_MMX2 || HAVE_AMD3DNOW
00592 /*
00593     uint8_t tmp[16];
00594     const int l1= stride;
00595     const int l2= stride + l1;
00596     const int l3= stride + l2;
00597     const int l4= (int)tmp - (int)src - stride*3;
00598     const int l5= (int)tmp - (int)src - stride*3 + 8;
00599     const int l6= stride*3 + l3;
00600     const int l7= stride + l6;
00601     const int l8= stride + l7;
00602 
00603     memcpy(tmp, src+stride*7, 8);
00604     memcpy(tmp+8, src+stride*8, 8);
00605 */
00606     src+= stride*4;
00607     __asm__ volatile(
00608 
00609 #if 0 //slightly more accurate and slightly slower
00610         "pxor %%mm7, %%mm7                      \n\t" // 0
00611         "lea (%0, %1), %%"REG_a"                \n\t"
00612         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
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 (%0, %1, 2), %%mm0                \n\t" // l2
00619         "movq (%0), %%mm1                       \n\t" // l0
00620         "movq %%mm0, %%mm2                      \n\t" // l2
00621         PAVGB(%%mm7, %%mm0)                           // ~l2/2
00622         PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
00623         PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
00624 
00625         "movq (%%"REG_a"), %%mm1                \n\t" // l1
00626         "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
00627         "movq %%mm1, %%mm4                      \n\t" // l1
00628         PAVGB(%%mm7, %%mm1)                           // ~l1/2
00629         PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
00630         PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
00631 
00632         "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
00633         "psubusb %%mm1, %%mm0                   \n\t"
00634         "psubusb %%mm4, %%mm1                   \n\t"
00635         "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
00636 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
00637 
00638         "movq (%0, %1, 4), %%mm0                \n\t" // l4
00639         "movq %%mm0, %%mm4                      \n\t" // l4
00640         PAVGB(%%mm7, %%mm0)                           // ~l4/2
00641         PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
00642         PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
00643 
00644         "movq (%%"REG_c"), %%mm2                \n\t" // l5
00645         "movq %%mm3, %%mm5                      \n\t" // l3
00646         PAVGB(%%mm7, %%mm3)                           // ~l3/2
00647         PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
00648         PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
00649 
00650         "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
00651         "psubusb %%mm3, %%mm0                   \n\t"
00652         "psubusb %%mm6, %%mm3                   \n\t"
00653         "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
00654         "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
00655 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
00656 
00657         "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
00658         "movq %%mm6, %%mm5                      \n\t" // l6
00659         PAVGB(%%mm7, %%mm6)                           // ~l6/2
00660         PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
00661         PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
00662 
00663         "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
00664         "movq %%mm2, %%mm4                      \n\t" // l5
00665         PAVGB(%%mm7, %%mm2)                           // ~l5/2
00666         PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
00667         PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
00668 
00669         "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
00670         "psubusb %%mm2, %%mm6                   \n\t"
00671         "psubusb %%mm4, %%mm2                   \n\t"
00672         "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
00673 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
00674 
00675 
00676         PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
00677         "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
00678         "paddusb "MANGLE(b01)", %%mm4           \n\t"
00679         "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
00680         "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
00681         "pand %%mm4, %%mm3                      \n\t"
00682 
00683         "movq %%mm3, %%mm1                      \n\t"
00684 //        "psubusb "MANGLE(b01)", %%mm3           \n\t"
00685         PAVGB(%%mm7, %%mm3)
00686         PAVGB(%%mm7, %%mm3)
00687         "paddusb %%mm1, %%mm3                   \n\t"
00688 //        "paddusb "MANGLE(b01)", %%mm3           \n\t"
00689 
00690         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
00691         "movq (%0, %1, 4), %%mm5                \n\t" //l4
00692         "movq (%0, %1, 4), %%mm4                \n\t" //l4
00693         "psubusb %%mm6, %%mm5                   \n\t"
00694         "psubusb %%mm4, %%mm6                   \n\t"
00695         "por %%mm6, %%mm5                       \n\t" // |l3-l4|
00696         "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
00697         "pxor %%mm6, %%mm0                      \n\t"
00698         "pand %%mm0, %%mm3                      \n\t"
00699         PMINUB(%%mm5, %%mm3, %%mm0)
00700 
00701         "psubusb "MANGLE(b01)", %%mm3           \n\t"
00702         PAVGB(%%mm7, %%mm3)
00703 
00704         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00705         "movq (%0, %1, 4), %%mm2                \n\t"
00706         "pxor %%mm6, %%mm0                      \n\t"
00707         "pxor %%mm6, %%mm2                      \n\t"
00708         "psubb %%mm3, %%mm0                     \n\t"
00709         "paddb %%mm3, %%mm2                     \n\t"
00710         "pxor %%mm6, %%mm0                      \n\t"
00711         "pxor %%mm6, %%mm2                      \n\t"
00712         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
00713         "movq %%mm2, (%0, %1, 4)                \n\t"
00714 #endif //0
00715 
00716         "lea (%0, %1), %%"REG_a"                \n\t"
00717         "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
00718 //      0       1       2       3       4       5       6       7
00719 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
00720 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
00721 
00722 
00723         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
00724         "movq (%0, %1, 4), %%mm0                \n\t" // l4
00725         "pxor %%mm6, %%mm1                      \n\t" // -l3-1
00726         PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
00727 // mm1=-l3-1, mm0=128-q
00728 
00729         "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
00730         "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
00731         "pxor %%mm6, %%mm2                      \n\t" // -l5-1
00732         "movq %%mm2, %%mm5                      \n\t" // -l5-1
00733         "movq "MANGLE(b80)", %%mm4              \n\t" // 128
00734         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00735         PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
00736         PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
00737         PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
00738         PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
00739 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
00740 
00741         "movq (%%"REG_a"), %%mm2                \n\t" // l1
00742         "pxor %%mm6, %%mm2                      \n\t" // -l1-1
00743         PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
00744         PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
00745         "movq "MANGLE(b80)", %%mm3              \n\t" // 128
00746         PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
00747         PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
00748         PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
00749 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
00750 
00751         PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
00752         "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
00753         "pxor %%mm6, %%mm1                      \n\t" // -l7-1
00754         PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
00755         "movq "MANGLE(b80)", %%mm2              \n\t" // 128
00756         PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
00757         PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
00758         PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
00759 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
00760 
00761         "movq "MANGLE(b00)", %%mm1              \n\t" // 0
00762         "movq "MANGLE(b00)", %%mm5              \n\t" // 0
00763         "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
00764         "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
00765         PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
00766         PMAXUB(%%mm5, %%mm3)                          // 128 + |lenergy/16|
00767         PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
00768 
00769 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
00770 
00771         "movq "MANGLE(b00)", %%mm7              \n\t" // 0
00772         "movq %2, %%mm2                         \n\t" // QP
00773         PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
00774         "psubb %%mm6, %%mm2                     \n\t"
00775 
00776         "movq %%mm4, %%mm1                      \n\t"
00777         "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
00778         "pxor %%mm1, %%mm4                      \n\t"
00779         "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
00780         "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
00781         "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
00782 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
00783 
00784         "movq %%mm4, %%mm3                      \n\t" // d
00785         "psubusb "MANGLE(b01)", %%mm4           \n\t"
00786         PAVGB(%%mm7, %%mm4)                           // d/32
00787         PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
00788         "paddb %%mm3, %%mm4                     \n\t" // 5d/64
00789         "pand %%mm2, %%mm4                      \n\t"
00790 
00791         "movq "MANGLE(b80)", %%mm5              \n\t" // 128
00792         "psubb %%mm0, %%mm5                     \n\t" // q
00793         "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
00794         "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
00795         "pxor %%mm7, %%mm5                      \n\t"
00796 
00797         PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
00798         "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
00799 
00800         "pand %%mm7, %%mm4                      \n\t"
00801         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00802         "movq (%0, %1, 4), %%mm2                \n\t"
00803         "pxor %%mm1, %%mm0                      \n\t"
00804         "pxor %%mm1, %%mm2                      \n\t"
00805         "paddb %%mm4, %%mm0                     \n\t"
00806         "psubb %%mm4, %%mm2                     \n\t"
00807         "pxor %%mm1, %%mm0                      \n\t"
00808         "pxor %%mm1, %%mm2                      \n\t"
00809         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
00810         "movq %%mm2, (%0, %1, 4)                \n\t"
00811 
00812         :
00813         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00814         : "%"REG_a, "%"REG_c
00815     );
00816 
00817 /*
00818     {
00819     int x;
00820     src-= stride;
00821     for(x=0; x<BLOCK_SIZE; x++){
00822         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
00823         if(FFABS(middleEnergy)< 8*QP){
00824             const int q=(src[l4] - src[l5])/2;
00825             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
00826             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
00827 
00828             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
00829             d= FFMAX(d, 0);
00830 
00831             d= (5*d + 32) >> 6;
00832             d*= FFSIGN(-middleEnergy);
00833 
00834             if(q>0){
00835                 d= d<0 ? 0 : d;
00836                 d= d>q ? q : d;
00837             }else{
00838                 d= d>0 ? 0 : d;
00839                 d= d<q ? q : d;
00840             }
00841 
00842             src[l4]-= d;
00843             src[l5]+= d;
00844         }
00845         src++;
00846     }
00847     src-=8;
00848     for(x=0; x<8; x++){
00849         int y;
00850         for(y=4; y<6; y++){
00851             int d= src[x+y*stride] - tmp[x+(y-4)*8];
00852             int ad= FFABS(d);
00853             static int max=0;
00854             static int sum=0;
00855             static int num=0;
00856             static int bias=0;
00857 
00858             if(max<ad) max=ad;
00859             sum+= ad>3 ? 1 : 0;
00860             if(ad>3){
00861                 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
00862             }
00863             if(y==4) bias+=d;
00864             num++;
00865             if(num%1000000 == 0){
00866                 av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
00867             }
00868         }
00869     }
00870 }
00871 */
00872 #elif HAVE_MMX
00873     src+= stride*4;
00874     __asm__ volatile(
00875         "pxor %%mm7, %%mm7                      \n\t"
00876         "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
00877         "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
00878 //      0       1       2       3       4       5       6       7
00879 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
00880 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1
00881 
00882         "movq (%0), %%mm0                       \n\t"
00883         "movq %%mm0, %%mm1                      \n\t"
00884         "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
00885         "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
00886 
00887         "movq (%0, %1), %%mm2                   \n\t"
00888         "lea (%0, %1, 2), %%"REG_a"             \n\t"
00889         "movq %%mm2, %%mm3                      \n\t"
00890         "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
00891         "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
00892 
00893         "movq (%%"REG_a"), %%mm4                \n\t"
00894         "movq %%mm4, %%mm5                      \n\t"
00895         "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
00896         "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
00897 
00898         "paddw %%mm0, %%mm0                     \n\t" // 2L0
00899         "paddw %%mm1, %%mm1                     \n\t" // 2H0
00900         "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
00901         "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
00902         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
00903         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
00904 
00905         "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
00906         "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
00907         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
00908         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
00909 
00910         "movq (%%"REG_a", %1), %%mm2            \n\t"
00911         "movq %%mm2, %%mm3                      \n\t"
00912         "punpcklbw %%mm7, %%mm2                 \n\t" // L3
00913         "punpckhbw %%mm7, %%mm3                 \n\t" // H3
00914 
00915         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
00916         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
00917         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00918         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00919         "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00920         "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00921 
00922         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00923         "movq %%mm0, %%mm1                      \n\t"
00924         "punpcklbw %%mm7, %%mm0                 \n\t" // L4
00925         "punpckhbw %%mm7, %%mm1                 \n\t" // H4
00926 
00927         "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
00928         "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
00929         "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
00930         "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
00931         "paddw %%mm4, %%mm4                     \n\t" // 2L2
00932         "paddw %%mm5, %%mm5                     \n\t" // 2H2
00933         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
00934         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
00935 
00936         "lea (%%"REG_a", %1), %0                \n\t"
00937         "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
00938         "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
00939         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
00940         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
00941 //50 opcodes so far
00942         "movq (%0, %1, 2), %%mm2                \n\t"
00943         "movq %%mm2, %%mm3                      \n\t"
00944         "punpcklbw %%mm7, %%mm2                 \n\t" // L5
00945         "punpckhbw %%mm7, %%mm3                 \n\t" // H5
00946         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
00947         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
00948         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
00949         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
00950 
00951         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
00952         "punpcklbw %%mm7, %%mm6                 \n\t" // L6
00953         "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
00954         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
00955         "punpckhbw %%mm7, %%mm6                 \n\t" // H6
00956         "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
00957 
00958         "paddw %%mm0, %%mm0                     \n\t" // 2L4
00959         "paddw %%mm1, %%mm1                     \n\t" // 2H4
00960         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
00961         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
00962 
00963         "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
00964         "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
00965         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
00966         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
00967 
00968         "movq (%0, %1, 4), %%mm2                \n\t"
00969         "movq %%mm2, %%mm3                      \n\t"
00970         "punpcklbw %%mm7, %%mm2                 \n\t" // L7
00971         "punpckhbw %%mm7, %%mm3                 \n\t" // H7
00972 
00973         "paddw %%mm2, %%mm2                     \n\t" // 2L7
00974         "paddw %%mm3, %%mm3                     \n\t" // 2H7
00975         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
00976         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
00977 
00978         "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00979         "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00980 
00981 #if HAVE_MMX2
00982         "movq %%mm7, %%mm6                      \n\t" // 0
00983         "psubw %%mm0, %%mm6                     \n\t"
00984         "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
00985         "movq %%mm7, %%mm6                      \n\t" // 0
00986         "psubw %%mm1, %%mm6                     \n\t"
00987         "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
00988         "movq %%mm7, %%mm6                      \n\t" // 0
00989         "psubw %%mm2, %%mm6                     \n\t"
00990         "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
00991         "movq %%mm7, %%mm6                      \n\t" // 0
00992         "psubw %%mm3, %%mm6                     \n\t"
00993         "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
00994 #else
00995         "movq %%mm7, %%mm6                      \n\t" // 0
00996         "pcmpgtw %%mm0, %%mm6                   \n\t"
00997         "pxor %%mm6, %%mm0                      \n\t"
00998         "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
00999         "movq %%mm7, %%mm6                      \n\t" // 0
01000         "pcmpgtw %%mm1, %%mm6                   \n\t"
01001         "pxor %%mm6, %%mm1                      \n\t"
01002         "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
01003         "movq %%mm7, %%mm6                      \n\t" // 0
01004         "pcmpgtw %%mm2, %%mm6                   \n\t"
01005         "pxor %%mm6, %%mm2                      \n\t"
01006         "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
01007         "movq %%mm7, %%mm6                      \n\t" // 0
01008         "pcmpgtw %%mm3, %%mm6                   \n\t"
01009         "pxor %%mm6, %%mm3                      \n\t"
01010         "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
01011 #endif
01012 
01013 #if HAVE_MMX2
01014         "pminsw %%mm2, %%mm0                    \n\t"
01015         "pminsw %%mm3, %%mm1                    \n\t"
01016 #else
01017         "movq %%mm0, %%mm6                      \n\t"
01018         "psubusw %%mm2, %%mm6                   \n\t"
01019         "psubw %%mm6, %%mm0                     \n\t"
01020         "movq %%mm1, %%mm6                      \n\t"
01021         "psubusw %%mm3, %%mm6                   \n\t"
01022         "psubw %%mm6, %%mm1                     \n\t"
01023 #endif
01024 
01025         "movd %2, %%mm2                         \n\t" // QP
01026         "punpcklbw %%mm7, %%mm2                 \n\t"
01027 
01028         "movq %%mm7, %%mm6                      \n\t" // 0
01029         "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
01030         "pxor %%mm6, %%mm4                      \n\t"
01031         "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
01032         "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
01033         "pxor %%mm7, %%mm5                      \n\t"
01034         "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
01035 // 100 opcodes
01036         "psllw $3, %%mm2                        \n\t" // 8QP
01037         "movq %%mm2, %%mm3                      \n\t" // 8QP
01038         "pcmpgtw %%mm4, %%mm2                   \n\t"
01039         "pcmpgtw %%mm5, %%mm3                   \n\t"
01040         "pand %%mm2, %%mm4                      \n\t"
01041         "pand %%mm3, %%mm5                      \n\t"
01042 
01043 
01044         "psubusw %%mm0, %%mm4                   \n\t" // hd
01045         "psubusw %%mm1, %%mm5                   \n\t" // ld
01046 
01047 
01048         "movq "MANGLE(w05)", %%mm2              \n\t" // 5
01049         "pmullw %%mm2, %%mm4                    \n\t"
01050         "pmullw %%mm2, %%mm5                    \n\t"
01051         "movq "MANGLE(w20)", %%mm2              \n\t" // 32
01052         "paddw %%mm2, %%mm4                     \n\t"
01053         "paddw %%mm2, %%mm5                     \n\t"
01054         "psrlw $6, %%mm4                        \n\t"
01055         "psrlw $6, %%mm5                        \n\t"
01056 
01057         "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
01058         "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
01059 
01060         "pxor %%mm2, %%mm2                      \n\t"
01061         "pxor %%mm3, %%mm3                      \n\t"
01062 
01063         "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
01064         "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
01065         "pxor %%mm2, %%mm0                      \n\t"
01066         "pxor %%mm3, %%mm1                      \n\t"
01067         "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
01068         "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
01069         "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
01070         "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
01071 
01072         "pxor %%mm6, %%mm2                      \n\t"
01073         "pxor %%mm7, %%mm3                      \n\t"
01074         "pand %%mm2, %%mm4                      \n\t"
01075         "pand %%mm3, %%mm5                      \n\t"
01076 
01077 #if HAVE_MMX2
01078         "pminsw %%mm0, %%mm4                    \n\t"
01079         "pminsw %%mm1, %%mm5                    \n\t"
01080 #else
01081         "movq %%mm4, %%mm2                      \n\t"
01082         "psubusw %%mm0, %%mm2                   \n\t"
01083         "psubw %%mm2, %%mm4                     \n\t"
01084         "movq %%mm5, %%mm2                      \n\t"
01085         "psubusw %%mm1, %%mm2                   \n\t"
01086         "psubw %%mm2, %%mm5                     \n\t"
01087 #endif
01088         "pxor %%mm6, %%mm4                      \n\t"
01089         "pxor %%mm7, %%mm5                      \n\t"
01090         "psubw %%mm6, %%mm4                     \n\t"
01091         "psubw %%mm7, %%mm5                     \n\t"
01092         "packsswb %%mm5, %%mm4                  \n\t"
01093         "movq (%0), %%mm0                       \n\t"
01094         "paddb   %%mm4, %%mm0                   \n\t"
01095         "movq %%mm0, (%0)                       \n\t"
01096         "movq (%0, %1), %%mm0                   \n\t"
01097         "psubb %%mm4, %%mm0                     \n\t"
01098         "movq %%mm0, (%0, %1)                   \n\t"
01099 
01100         : "+r" (src)
01101         : "r" ((x86_reg)stride), "m" (c->pQPb)
01102         : "%"REG_a, "%"REG_c
01103     );
01104 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01105     const int l1= stride;
01106     const int l2= stride + l1;
01107     const int l3= stride + l2;
01108     const int l4= stride + l3;
01109     const int l5= stride + l4;
01110     const int l6= stride + l5;
01111     const int l7= stride + l6;
01112     const int l8= stride + l7;
01113 //    const int l9= stride + l8;
01114     int x;
01115     src+= stride*3;
01116     for(x=0; x<BLOCK_SIZE; x++){
01117         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
01118         if(FFABS(middleEnergy) < 8*c->QP){
01119             const int q=(src[l4] - src[l5])/2;
01120             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
01121             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
01122 
01123             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
01124             d= FFMAX(d, 0);
01125 
01126             d= (5*d + 32) >> 6;
01127             d*= FFSIGN(-middleEnergy);
01128 
01129             if(q>0){
01130                 d= d<0 ? 0 : d;
01131                 d= d>q ? q : d;
01132             }else{
01133                 d= d>0 ? 0 : d;
01134                 d= d<q ? q : d;
01135             }
01136 
01137             src[l4]-= d;
01138             src[l5]+= d;
01139         }
01140         src++;
01141     }
01142 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01143 }
01144 #endif //HAVE_ALTIVEC
01145 
01146 #if !HAVE_ALTIVEC
01147 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
01148 {
01149 #if HAVE_MMX2 || HAVE_AMD3DNOW
01150     __asm__ volatile(
01151         "pxor %%mm6, %%mm6                      \n\t"
01152         "pcmpeqb %%mm7, %%mm7                   \n\t"
01153         "movq %2, %%mm0                         \n\t"
01154         "punpcklbw %%mm6, %%mm0                 \n\t"
01155         "psrlw $1, %%mm0                        \n\t"
01156         "psubw %%mm7, %%mm0                     \n\t"
01157         "packuswb %%mm0, %%mm0                  \n\t"
01158         "movq %%mm0, %3                         \n\t"
01159 
01160         "lea (%0, %1), %%"REG_a"                \n\t"
01161         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01162 
01163 //        0        1        2        3        4        5        6        7        8        9
01164 //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1
01165 
01166 #undef FIND_MIN_MAX
01167 #if HAVE_MMX2
01168 #define REAL_FIND_MIN_MAX(addr)\
01169         "movq " #addr ", %%mm0                  \n\t"\
01170         "pminub %%mm0, %%mm7                    \n\t"\
01171         "pmaxub %%mm0, %%mm6                    \n\t"
01172 #else
01173 #define REAL_FIND_MIN_MAX(addr)\
01174         "movq " #addr ", %%mm0                  \n\t"\
01175         "movq %%mm7, %%mm1                      \n\t"\
01176         "psubusb %%mm0, %%mm6                   \n\t"\
01177         "paddb %%mm0, %%mm6                     \n\t"\
01178         "psubusb %%mm0, %%mm1                   \n\t"\
01179         "psubb %%mm1, %%mm7                     \n\t"
01180 #endif
01181 #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
01182 
01183 FIND_MIN_MAX((%%REGa))
01184 FIND_MIN_MAX((%%REGa, %1))
01185 FIND_MIN_MAX((%%REGa, %1, 2))
01186 FIND_MIN_MAX((%0, %1, 4))
01187 FIND_MIN_MAX((%%REGd))
01188 FIND_MIN_MAX((%%REGd, %1))
01189 FIND_MIN_MAX((%%REGd, %1, 2))
01190 FIND_MIN_MAX((%0, %1, 8))
01191 
01192         "movq %%mm7, %%mm4                      \n\t"
01193         "psrlq $8, %%mm7                        \n\t"
01194 #if HAVE_MMX2
01195         "pminub %%mm4, %%mm7                    \n\t" // min of pixels
01196         "pshufw $0xF9, %%mm7, %%mm4             \n\t"
01197         "pminub %%mm4, %%mm7                    \n\t" // min of pixels
01198         "pshufw $0xFE, %%mm7, %%mm4             \n\t"
01199         "pminub %%mm4, %%mm7                    \n\t"
01200 #else
01201         "movq %%mm7, %%mm1                      \n\t"
01202         "psubusb %%mm4, %%mm1                   \n\t"
01203         "psubb %%mm1, %%mm7                     \n\t"
01204         "movq %%mm7, %%mm4                      \n\t"
01205         "psrlq $16, %%mm7                       \n\t"
01206         "movq %%mm7, %%mm1                      \n\t"
01207         "psubusb %%mm4, %%mm1                   \n\t"
01208         "psubb %%mm1, %%mm7                     \n\t"
01209         "movq %%mm7, %%mm4                      \n\t"
01210         "psrlq $32, %%mm7                       \n\t"
01211         "movq %%mm7, %%mm1                      \n\t"
01212         "psubusb %%mm4, %%mm1                   \n\t"
01213         "psubb %%mm1, %%mm7                     \n\t"
01214 #endif
01215 
01216 
01217         "movq %%mm6, %%mm4                      \n\t"
01218         "psrlq $8, %%mm6                        \n\t"
01219 #if HAVE_MMX2
01220         "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
01221         "pshufw $0xF9, %%mm6, %%mm4             \n\t"
01222         "pmaxub %%mm4, %%mm6                    \n\t"
01223         "pshufw $0xFE, %%mm6, %%mm4             \n\t"
01224         "pmaxub %%mm4, %%mm6                    \n\t"
01225 #else
01226         "psubusb %%mm4, %%mm6                   \n\t"
01227         "paddb %%mm4, %%mm6                     \n\t"
01228         "movq %%mm6, %%mm4                      \n\t"
01229         "psrlq $16, %%mm6                       \n\t"
01230         "psubusb %%mm4, %%mm6                   \n\t"
01231         "paddb %%mm4, %%mm6                     \n\t"
01232         "movq %%mm6, %%mm4                      \n\t"
01233         "psrlq $32, %%mm6                       \n\t"
01234         "psubusb %%mm4, %%mm6                   \n\t"
01235         "paddb %%mm4, %%mm6                     \n\t"
01236 #endif
01237         "movq %%mm6, %%mm0                      \n\t" // max
01238         "psubb %%mm7, %%mm6                     \n\t" // max - min
01239         "movd %%mm6, %%ecx                      \n\t"
01240         "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
01241         " jb 1f                                 \n\t"
01242         "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
01243         "and "ALIGN_MASK", %%"REG_c"            \n\t"
01244         PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
01245         "punpcklbw %%mm7, %%mm7                 \n\t"
01246         "punpcklbw %%mm7, %%mm7                 \n\t"
01247         "punpcklbw %%mm7, %%mm7                 \n\t"
01248         "movq %%mm7, (%%"REG_c")                \n\t"
01249 
01250         "movq (%0), %%mm0                       \n\t" // L10
01251         "movq %%mm0, %%mm1                      \n\t" // L10
01252         "movq %%mm0, %%mm2                      \n\t" // L10
01253         "psllq $8, %%mm1                        \n\t"
01254         "psrlq $8, %%mm2                        \n\t"
01255         "movd -4(%0), %%mm3                     \n\t"
01256         "movd 8(%0), %%mm4                      \n\t"
01257         "psrlq $24, %%mm3                       \n\t"
01258         "psllq $56, %%mm4                       \n\t"
01259         "por %%mm3, %%mm1                       \n\t" // L00
01260         "por %%mm4, %%mm2                       \n\t" // L20
01261         "movq %%mm1, %%mm3                      \n\t" // L00
01262         PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
01263         PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
01264         "psubusb %%mm7, %%mm0                   \n\t"
01265         "psubusb %%mm7, %%mm2                   \n\t"
01266         "psubusb %%mm7, %%mm3                   \n\t"
01267         "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
01268         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
01269         "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
01270         "paddb %%mm2, %%mm0                     \n\t"
01271         "paddb %%mm3, %%mm0                     \n\t"
01272 
01273         "movq (%%"REG_a"), %%mm2                \n\t" // L11
01274         "movq %%mm2, %%mm3                      \n\t" // L11
01275         "movq %%mm2, %%mm4                      \n\t" // L11
01276         "psllq $8, %%mm3                        \n\t"
01277         "psrlq $8, %%mm4                        \n\t"
01278         "movd -4(%%"REG_a"), %%mm5              \n\t"
01279         "movd 8(%%"REG_a"), %%mm6               \n\t"
01280         "psrlq $24, %%mm5                       \n\t"
01281         "psllq $56, %%mm6                       \n\t"
01282         "por %%mm5, %%mm3                       \n\t" // L01
01283         "por %%mm6, %%mm4                       \n\t" // L21
01284         "movq %%mm3, %%mm5                      \n\t" // L01
01285         PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
01286         PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
01287         "psubusb %%mm7, %%mm2                   \n\t"
01288         "psubusb %%mm7, %%mm4                   \n\t"
01289         "psubusb %%mm7, %%mm5                   \n\t"
01290         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
01291         "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
01292         "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
01293         "paddb %%mm4, %%mm2                     \n\t"
01294         "paddb %%mm5, %%mm2                     \n\t"
01295 // 0, 2, 3, 1
01296 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
01297         "movq " #src ", " #sx "                 \n\t" /* src[0] */\
01298         "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
01299         "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
01300         "psllq $8, " #lx "                      \n\t"\
01301         "psrlq $8, " #t0 "                      \n\t"\
01302         "movd -4" #src ", " #t1 "               \n\t"\
01303         "psrlq $24, " #t1 "                     \n\t"\
01304         "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
01305         "movd 8" #src ", " #t1 "                \n\t"\
01306         "psllq $56, " #t1 "                     \n\t"\
01307         "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
01308         "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
01309         PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
01310         PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
01311         PAVGB(lx, pplx)                                     \
01312         "movq " #lx ", 8(%%"REG_c")             \n\t"\
01313         "movq (%%"REG_c"), " #lx "              \n\t"\
01314         "psubusb " #lx ", " #t1 "               \n\t"\
01315         "psubusb " #lx ", " #t0 "               \n\t"\
01316         "psubusb " #lx ", " #sx "               \n\t"\
01317         "movq "MANGLE(b00)", " #lx "            \n\t"\
01318         "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
01319         "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
01320         "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
01321         "paddb " #t1 ", " #t0 "                 \n\t"\
01322         "paddb " #t0 ", " #sx "                 \n\t"\
01323 \
01324         PAVGB(plx, pplx)                              /* filtered */\
01325         "movq " #dst ", " #t0 "                 \n\t" /* dst */\
01326         "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
01327         "psubusb %3, " #t0 "                    \n\t"\
01328         "paddusb %3, " #t1 "                    \n\t"\
01329         PMAXUB(t0, pplx)\
01330         PMINUB(t1, pplx, t0)\
01331         "paddb " #sx ", " #ppsx "               \n\t"\
01332         "paddb " #psx ", " #ppsx "              \n\t"\
01333         "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
01334         "pand "MANGLE(b08)", " #ppsx "          \n\t"\
01335         "pcmpeqb " #lx ", " #ppsx "             \n\t"\
01336         "pand " #ppsx ", " #pplx "              \n\t"\
01337         "pandn " #dst ", " #ppsx "              \n\t"\
01338         "por " #pplx ", " #ppsx "               \n\t"\
01339         "movq " #ppsx ", " #dst "               \n\t"\
01340         "movq 8(%%"REG_c"), " #lx "             \n\t"
01341 
01342 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
01343    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
01344 /*
01345 0000000
01346 1111111
01347 
01348 1111110
01349 1111101
01350 1111100
01351 1111011
01352 1111010
01353 1111001
01354 
01355 1111000
01356 1110111
01357 
01358 */
01359 //DERING_CORE(dst          ,src            ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)
01360 DERING_CORE((%%REGa)       ,(%%REGa, %1)   ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01361 DERING_CORE((%%REGa, %1)   ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01362 DERING_CORE((%%REGa, %1, 2),(%0, %1, 4)    ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
01363 DERING_CORE((%0, %1, 4)    ,(%%REGd)       ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01364 DERING_CORE((%%REGd)       ,(%%REGd, %1)   ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01365 DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
01366 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01367 DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01368 
01369         "1:                        \n\t"
01370         : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2)
01371         : "%"REG_a, "%"REG_d, "%"REG_c
01372     );
01373 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01374     int y;
01375     int min=255;
01376     int max=0;
01377     int avg;
01378     uint8_t *p;
01379     int s[10];
01380     const int QP2= c->QP/2 + 1;
01381 
01382     for(y=1; y<9; y++){
01383         int x;
01384         p= src + stride*y;
01385         for(x=1; x<9; x++){
01386             p++;
01387             if(*p > max) max= *p;
01388             if(*p < min) min= *p;
01389         }
01390     }
01391     avg= (min + max + 1)>>1;
01392 
01393     if(max - min <deringThreshold) return;
01394 
01395     for(y=0; y<10; y++){
01396         int t = 0;
01397 
01398         if(src[stride*y + 0] > avg) t+= 1;
01399         if(src[stride*y + 1] > avg) t+= 2;
01400         if(src[stride*y + 2] > avg) t+= 4;
01401         if(src[stride*y + 3] > avg) t+= 8;
01402         if(src[stride*y + 4] > avg) t+= 16;
01403         if(src[stride*y + 5] > avg) t+= 32;
01404         if(src[stride*y + 6] > avg) t+= 64;
01405         if(src[stride*y + 7] > avg) t+= 128;
01406         if(src[stride*y + 8] > avg) t+= 256;
01407         if(src[stride*y + 9] > avg) t+= 512;
01408 
01409         t |= (~t)<<16;
01410         t &= (t<<1) & (t>>1);
01411         s[y] = t;
01412     }
01413 
01414     for(y=1; y<9; y++){
01415         int t = s[y-1] & s[y] & s[y+1];
01416         t|= t>>16;
01417         s[y-1]= t;
01418     }
01419 
01420     for(y=1; y<9; y++){
01421         int x;
01422         int t = s[y-1];
01423 
01424         p= src + stride*y;
01425         for(x=1; x<9; x++){
01426             p++;
01427             if(t & (1<<x)){
01428                 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
01429                       +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
01430                       +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
01431                 f= (f + 8)>>4;
01432 
01433 #ifdef DEBUG_DERING_THRESHOLD
01434                     __asm__ volatile("emms\n\t":);
01435                     {
01436                     static long long numPixels=0;
01437                     if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
01438 //                    if((max-min)<20 || (max-min)*QP<200)
01439 //                    if((max-min)*QP < 500)
01440 //                    if(max-min<QP/2)
01441                     if(max-min < 20){
01442                         static int numSkipped=0;
01443                         static int errorSum=0;
01444                         static int worstQP=0;
01445                         static int worstRange=0;
01446                         static int worstDiff=0;
01447                         int diff= (f - *p);
01448                         int absDiff= FFABS(diff);
01449                         int error= diff*diff;
01450 
01451                         if(x==1 || x==8 || y==1 || y==8) continue;
01452 
01453                         numSkipped++;
01454                         if(absDiff > worstDiff){
01455                             worstDiff= absDiff;
01456                             worstQP= QP;
01457                             worstRange= max-min;
01458                         }
01459                         errorSum+= error;
01460 
01461                         if(1024LL*1024LL*1024LL % numSkipped == 0){
01462                             av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
01463                                    "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
01464                                    (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
01465                                    worstDiff, (float)numSkipped/numPixels);
01466                         }
01467                     }
01468                     }
01469 #endif
01470                     if     (*p + QP2 < f) *p= *p + QP2;
01471                     else if(*p - QP2 > f) *p= *p - QP2;
01472                     else *p=f;
01473             }
01474         }
01475     }
01476 #ifdef DEBUG_DERING_THRESHOLD
01477     if(max-min < 20){
01478         for(y=1; y<9; y++){
01479             int x;
01480             int t = 0;
01481             p= src + stride*y;
01482             for(x=1; x<9; x++){
01483                 p++;
01484                 *p = FFMIN(*p + 20, 255);
01485             }
01486         }
01487 //        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
01488     }
01489 #endif
01490 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01491 }
01492 #endif //HAVE_ALTIVEC
01493 
01500 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
01501 {
01502 #if HAVE_MMX2 || HAVE_AMD3DNOW
01503     src+= 4*stride;
01504     __asm__ volatile(
01505         "lea (%0, %1), %%"REG_a"                \n\t"
01506         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
01507 //      0       1       2       3       4       5       6       7       8       9
01508 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
01509 
01510         "movq (%0), %%mm0                       \n\t"
01511         "movq (%%"REG_a", %1), %%mm1            \n\t"
01512         PAVGB(%%mm1, %%mm0)
01513         "movq %%mm0, (%%"REG_a")                \n\t"
01514         "movq (%0, %1, 4), %%mm0                \n\t"
01515         PAVGB(%%mm0, %%mm1)
01516         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
01517         "movq (%%"REG_c", %1), %%mm1            \n\t"
01518         PAVGB(%%mm1, %%mm0)
01519         "movq %%mm0, (%%"REG_c")                \n\t"
01520         "movq (%0, %1, 8), %%mm0                \n\t"
01521         PAVGB(%%mm0, %%mm1)
01522         "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
01523 
01524         : : "r" (src), "r" ((x86_reg)stride)
01525         : "%"REG_a, "%"REG_c
01526     );
01527 #else
01528     int a, b, x;
01529     src+= 4*stride;
01530 
01531     for(x=0; x<2; x++){
01532         a= *(uint32_t*)&src[stride*0];
01533         b= *(uint32_t*)&src[stride*2];
01534         *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01535         a= *(uint32_t*)&src[stride*4];
01536         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01537         b= *(uint32_t*)&src[stride*6];
01538         *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01539         a= *(uint32_t*)&src[stride*8];
01540         *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01541         src += 4;
01542     }
01543 #endif
01544 }
01545 
01553 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
01554 {
01555 #if HAVE_MMX2 || HAVE_AMD3DNOW
01556     src+= stride*3;
01557     __asm__ volatile(
01558         "lea (%0, %1), %%"REG_a"                \n\t"
01559         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01560         "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
01561         "add %1, %%"REG_c"                      \n\t"
01562         "pxor %%mm7, %%mm7                      \n\t"
01563 //      0       1       2       3       4       5       6       7       8       9       10
01564 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01565 
01566 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
01567         "movq " #a ", %%mm0                     \n\t"\
01568         "movq " #b ", %%mm1                     \n\t"\
01569         "movq " #d ", %%mm2                     \n\t"\
01570         "movq " #e ", %%mm3                     \n\t"\
01571         PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
01572         PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
01573         "movq %%mm0, %%mm2                      \n\t"\
01574         "punpcklbw %%mm7, %%mm0                 \n\t"\
01575         "punpckhbw %%mm7, %%mm2                 \n\t"\
01576         "movq %%mm1, %%mm3                      \n\t"\
01577         "punpcklbw %%mm7, %%mm1                 \n\t"\
01578         "punpckhbw %%mm7, %%mm3                 \n\t"\
01579         "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
01580         "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
01581         "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
01582         "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
01583         "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
01584         "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
01585         "packuswb %%mm3, %%mm1                  \n\t"\
01586         "movq %%mm1, " #c "                     \n\t"
01587 #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)
01588 
01589 DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
01590 DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))
01591 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
01592 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
01593 
01594         : : "r" (src), "r" ((x86_reg)stride)
01595         : "%"REG_a, "%"REG_d, "%"REG_c
01596     );
01597 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01598     int x;
01599     src+= stride*3;
01600     for(x=0; x<8; x++){
01601         src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
01602         src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
01603         src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
01604         src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
01605         src++;
01606     }
01607 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01608 }
01609 
01617 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
01618 {
01619 #if HAVE_MMX2 || HAVE_AMD3DNOW
01620     src+= stride*4;
01621     __asm__ volatile(
01622         "lea (%0, %1), %%"REG_a"                \n\t"
01623         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01624         "pxor %%mm7, %%mm7                      \n\t"
01625         "movq (%2), %%mm0                       \n\t"
01626 //      0       1       2       3       4       5       6       7       8       9       10
01627 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01628 
01629 #define REAL_DEINT_FF(a,b,c,d)\
01630         "movq " #a ", %%mm1                     \n\t"\
01631         "movq " #b ", %%mm2                     \n\t"\
01632         "movq " #c ", %%mm3                     \n\t"\
01633         "movq " #d ", %%mm4                     \n\t"\
01634         PAVGB(%%mm3, %%mm1)                          \
01635         PAVGB(%%mm4, %%mm0)                          \
01636         "movq %%mm0, %%mm3                      \n\t"\
01637         "punpcklbw %%mm7, %%mm0                 \n\t"\
01638         "punpckhbw %%mm7, %%mm3                 \n\t"\
01639         "movq %%mm1, %%mm4                      \n\t"\
01640         "punpcklbw %%mm7, %%mm1                 \n\t"\
01641         "punpckhbw %%mm7, %%mm4                 \n\t"\
01642         "psllw $2, %%mm1                        \n\t"\
01643         "psllw $2, %%mm4                        \n\t"\
01644         "psubw %%mm0, %%mm1                     \n\t"\
01645         "psubw %%mm3, %%mm4                     \n\t"\
01646         "movq %%mm2, %%mm5                      \n\t"\
01647         "movq %%mm2, %%mm0                      \n\t"\
01648         "punpcklbw %%mm7, %%mm2                 \n\t"\
01649         "punpckhbw %%mm7, %%mm5                 \n\t"\
01650         "paddw %%mm2, %%mm1                     \n\t"\
01651         "paddw %%mm5, %%mm4                     \n\t"\
01652         "psraw $2, %%mm1                        \n\t"\
01653         "psraw $2, %%mm4                        \n\t"\
01654         "packuswb %%mm4, %%mm1                  \n\t"\
01655         "movq %%mm1, " #b "                     \n\t"\
01656 
01657 #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)
01658 
01659 DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))
01660 DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
01661 DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
01662 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
01663 
01664         "movq %%mm0, (%2)                       \n\t"
01665         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)
01666         : "%"REG_a, "%"REG_d
01667     );
01668 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01669     int x;
01670     src+= stride*4;
01671     for(x=0; x<8; x++){
01672         int t1= tmp[x];
01673         int t2= src[stride*1];
01674 
01675         src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
01676         t1= src[stride*4];
01677         src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
01678         t2= src[stride*6];
01679         src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
01680         t1= src[stride*8];
01681         src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
01682         tmp[x]= t1;
01683 
01684         src++;
01685     }
01686 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01687 }
01688 
01696 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
01697 {
01698 #if HAVE_MMX2 || HAVE_AMD3DNOW
01699     src+= stride*4;
01700     __asm__ volatile(
01701         "lea (%0, %1), %%"REG_a"                \n\t"
01702         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01703         "pxor %%mm7, %%mm7                      \n\t"
01704         "movq (%2), %%mm0                       \n\t"
01705         "movq (%3), %%mm1                       \n\t"
01706 //      0       1       2       3       4       5       6       7       8       9       10
01707 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01708 
01709 #define REAL_DEINT_L5(t1,t2,a,b,c)\
01710         "movq " #a ", %%mm2                     \n\t"\
01711         "movq " #b ", %%mm3                     \n\t"\
01712         "movq " #c ", %%mm4                     \n\t"\
01713         PAVGB(t2, %%mm3)                             \
01714         PAVGB(t1, %%mm4)                             \
01715         "movq %%mm2, %%mm5                      \n\t"\
01716         "movq %%mm2, " #t1 "                    \n\t"\
01717         "punpcklbw %%mm7, %%mm2                 \n\t"\
01718         "punpckhbw %%mm7, %%mm5                 \n\t"\
01719         "movq %%mm2, %%mm6                      \n\t"\
01720         "paddw %%mm2, %%mm2                     \n\t"\
01721         "paddw %%mm6, %%mm2                     \n\t"\
01722         "movq %%mm5, %%mm6                      \n\t"\
01723         "paddw %%mm5, %%mm5                     \n\t"\
01724         "paddw %%mm6, %%mm5                     \n\t"\
01725         "movq %%mm3, %%mm6                      \n\t"\
01726         "punpcklbw %%mm7, %%mm3                 \n\t"\
01727         "punpckhbw %%mm7, %%mm6                 \n\t"\
01728         "paddw %%mm3, %%mm3                     \n\t"\
01729         "paddw %%mm6, %%mm6                     \n\t"\
01730         "paddw %%mm3, %%mm2                     \n\t"\
01731         "paddw %%mm6, %%mm5                     \n\t"\
01732         "movq %%mm4, %%mm6                      \n\t"\
01733         "punpcklbw %%mm7, %%mm4                 \n\t"\
01734         "punpckhbw %%mm7, %%mm6                 \n\t"\
01735         "psubw %%mm4, %%mm2                     \n\t"\
01736         "psubw %%mm6, %%mm5                     \n\t"\
01737         "psraw $2, %%mm2                        \n\t"\
01738         "psraw $2, %%mm5                        \n\t"\
01739         "packuswb %%mm5, %%mm2                  \n\t"\
01740         "movq %%mm2, " #a "                     \n\t"\
01741 
01742 #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)
01743 
01744 DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )
01745 DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))
01746 DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )
01747 DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )
01748 DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )
01749 DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
01750 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
01751 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
01752 
01753         "movq %%mm0, (%2)                       \n\t"
01754         "movq %%mm1, (%3)                       \n\t"
01755         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)
01756         : "%"REG_a, "%"REG_d
01757     );
01758 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01759     int x;
01760     src+= stride*4;
01761     for(x=0; x<8; x++){
01762         int t1= tmp[x];
01763         int t2= tmp2[x];
01764         int t3= src[0];
01765 
01766         src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
01767         t1= src[stride*1];
01768         src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
01769         t2= src[stride*2];
01770         src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
01771         t3= src[stride*3];
01772         src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
01773         t1= src[stride*4];
01774         src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
01775         t2= src[stride*5];
01776         src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
01777         t3= src[stride*6];
01778         src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
01779         t1= src[stride*7];
01780         src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
01781 
01782         tmp[x]= t3;
01783         tmp2[x]= t1;
01784 
01785         src++;
01786     }
01787 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01788 }
01789 
01797 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
01798 {
01799 #if HAVE_MMX2 || HAVE_AMD3DNOW
01800     src+= 4*stride;
01801     __asm__ volatile(
01802         "lea (%0, %1), %%"REG_a"                \n\t"
01803         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01804 //      0       1       2       3       4       5       6       7       8       9
01805 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01806 
01807         "movq (%2), %%mm0                       \n\t" // L0
01808         "movq (%%"REG_a"), %%mm1                \n\t" // L2
01809         PAVGB(%%mm1, %%mm0)                           // L0+L2
01810         "movq (%0), %%mm2                       \n\t" // L1
01811         PAVGB(%%mm2, %%mm0)
01812         "movq %%mm0, (%0)                       \n\t"
01813         "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
01814         PAVGB(%%mm0, %%mm2)                           // L1+L3
01815         PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
01816         "movq %%mm2, (%%"REG_a")                \n\t"
01817         "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
01818         PAVGB(%%mm2, %%mm1)                           // L2+L4
01819         PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
01820         "movq %%mm1, (%%"REG_a", %1)            \n\t"
01821         "movq (%0, %1, 4), %%mm1                \n\t" // L5
01822         PAVGB(%%mm1, %%mm0)                           // L3+L5
01823         PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
01824         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
01825         "movq (%%"REG_d"), %%mm0                \n\t" // L6
01826         PAVGB(%%mm0, %%mm2)                           // L4+L6
01827         PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
01828         "movq %%mm2, (%0, %1, 4)                \n\t"
01829         "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
01830         PAVGB(%%mm2, %%mm1)                           // L5+L7
01831         PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
01832         "movq %%mm1, (%%"REG_d")                \n\t"
01833         "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
01834         PAVGB(%%mm1, %%mm0)                           // L6+L8
01835         PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
01836         "movq %%mm0, (%%"REG_d", %1)            \n\t"
01837         "movq (%0, %1, 8), %%mm0                \n\t" // L9
01838         PAVGB(%%mm0, %%mm2)                           // L7+L9
01839         PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
01840         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
01841         "movq %%mm1, (%2)                       \n\t"
01842 
01843         : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
01844         : "%"REG_a, "%"REG_d
01845     );
01846 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01847     int a, b, c, x;
01848     src+= 4*stride;
01849 
01850     for(x=0; x<2; x++){
01851         a= *(uint32_t*)&tmp[stride*0];
01852         b= *(uint32_t*)&src[stride*0];
01853         c= *(uint32_t*)&src[stride*1];
01854         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01855         *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01856 
01857         a= *(uint32_t*)&src[stride*2];
01858         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01859         *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01860 
01861         b= *(uint32_t*)&src[stride*3];
01862         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
01863         *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
01864 
01865         c= *(uint32_t*)&src[stride*4];
01866         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01867         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01868 
01869         a= *(uint32_t*)&src[stride*5];
01870         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01871         *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01872 
01873         b= *(uint32_t*)&src[stride*6];
01874         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
01875         *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
01876 
01877         c= *(uint32_t*)&src[stride*7];
01878         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01879         *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01880 
01881         a= *(uint32_t*)&src[stride*8];
01882         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01883         *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01884 
01885         *(uint32_t*)&tmp[stride*0]= c;
01886         src += 4;
01887         tmp += 4;
01888     }
01889 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01890 }
01891 
01898 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
01899 {
01900 #if HAVE_MMX
01901     src+= 4*stride;
01902 #if HAVE_MMX2
01903     __asm__ volatile(
01904         "lea (%0, %1), %%"REG_a"                \n\t"
01905         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01906 //      0       1       2       3       4       5       6       7       8       9
01907 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01908 
01909         "movq (%0), %%mm0                       \n\t" //
01910         "movq (%%"REG_a", %1), %%mm2            \n\t" //
01911         "movq (%%"REG_a"), %%mm1                \n\t" //
01912         "movq %%mm0, %%mm3                      \n\t"
01913         "pmaxub %%mm1, %%mm0                    \n\t" //
01914         "pminub %%mm3, %%mm1                    \n\t" //
01915         "pmaxub %%mm2, %%mm1                    \n\t" //
01916         "pminub %%mm1, %%mm0                    \n\t"
01917         "movq %%mm0, (%%"REG_a")                \n\t"
01918 
01919         "movq (%0, %1, 4), %%mm0                \n\t" //
01920         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
01921         "movq %%mm2, %%mm3                      \n\t"
01922         "pmaxub %%mm1, %%mm2                    \n\t" //
01923         "pminub %%mm3, %%mm1                    \n\t" //
01924         "pmaxub %%mm0, %%mm1                    \n\t" //
01925         "pminub %%mm1, %%mm2                    \n\t"
01926         "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
01927 
01928         "movq (%%"REG_d"), %%mm2                \n\t" //
01929         "movq (%%"REG_d", %1), %%mm1            \n\t" //
01930         "movq %%mm2, %%mm3                      \n\t"
01931         "pmaxub %%mm0, %%mm2                    \n\t" //
01932         "pminub %%mm3, %%mm0                    \n\t" //
01933         "pmaxub %%mm1, %%mm0                    \n\t" //
01934         "pminub %%mm0, %%mm2                    \n\t"
01935         "movq %%mm2, (%%"REG_d")                \n\t"
01936 
01937         "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
01938         "movq (%0, %1, 8), %%mm0                \n\t" //
01939         "movq %%mm2, %%mm3                      \n\t"
01940         "pmaxub %%mm0, %%mm2                    \n\t" //
01941         "pminub %%mm3, %%mm0                    \n\t" //
01942         "pmaxub %%mm1, %%mm0                    \n\t" //
01943         "pminub %%mm0, %%mm2                    \n\t"
01944         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
01945 
01946 
01947         : : "r" (src), "r" ((x86_reg)stride)
01948         : "%"REG_a, "%"REG_d
01949     );
01950 
01951 #else // MMX without MMX2
01952     __asm__ volatile(
01953         "lea (%0, %1), %%"REG_a"                \n\t"
01954         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01955 //      0       1       2       3       4       5       6       7       8       9
01956 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01957         "pxor %%mm7, %%mm7                      \n\t"
01958 
01959 #define REAL_MEDIAN(a,b,c)\
01960         "movq " #a ", %%mm0                     \n\t"\
01961         "movq " #b ", %%mm2                     \n\t"\
01962         "movq " #c ", %%mm1                     \n\t"\
01963         "movq %%mm0, %%mm3                      \n\t"\
01964         "movq %%mm1, %%mm4                      \n\t"\
01965         "movq %%mm2, %%mm5                      \n\t"\
01966         "psubusb %%mm1, %%mm3                   \n\t"\
01967         "psubusb %%mm2, %%mm4                   \n\t"\
01968         "psubusb %%mm0, %%mm5                   \n\t"\
01969         "pcmpeqb %%mm7, %%mm3                   \n\t"\
01970         "pcmpeqb %%mm7, %%mm4                   \n\t"\
01971         "pcmpeqb %%mm7, %%mm5                   \n\t"\
01972         "movq %%mm3, %%mm6                      \n\t"\
01973         "pxor %%mm4, %%mm3                      \n\t"\
01974         "pxor %%mm5, %%mm4                      \n\t"\
01975         "pxor %%mm6, %%mm5                      \n\t"\
01976         "por %%mm3, %%mm1                       \n\t"\
01977         "por %%mm4, %%mm2                       \n\t"\
01978         "por %%mm5, %%mm0                       \n\t"\
01979         "pand %%mm2, %%mm0                      \n\t"\
01980         "pand %%mm1, %%mm0                      \n\t"\
01981         "movq %%mm0, " #b "                     \n\t"
01982 #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
01983 
01984 MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))
01985 MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
01986 MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))
01987 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
01988 
01989         : : "r" (src), "r" ((x86_reg)stride)
01990         : "%"REG_a, "%"REG_d
01991     );
01992 #endif //HAVE_MMX2
01993 #else //HAVE_MMX
01994     int x, y;
01995     src+= 4*stride;
01996     // FIXME - there should be a way to do a few columns in parallel like w/mmx
01997     for(x=0; x<8; x++){
01998         uint8_t *colsrc = src;
01999         for (y=0; y<4; y++){
02000             int a, b, c, d, e, f;
02001             a = colsrc[0       ];
02002             b = colsrc[stride  ];
02003             c = colsrc[stride*2];
02004             d = (a-b)>>31;
02005             e = (b-c)>>31;
02006             f = (c-a)>>31;
02007             colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
02008             colsrc += stride*2;
02009         }
02010         src++;
02011     }
02012 #endif //HAVE_MMX
02013 }
02014 
02015 #if HAVE_MMX
02016 
02019 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
02020 {
02021     __asm__(
02022         "lea (%0, %1), %%"REG_a"                \n\t"
02023 //      0       1       2       3       4       5       6       7       8       9
02024 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
02025         "movq (%0), %%mm0                       \n\t" // 12345678
02026         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
02027         "movq %%mm0, %%mm2                      \n\t" // 12345678
02028         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02029         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02030 
02031         "movq (%%"REG_a", %1), %%mm1            \n\t"
02032         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
02033         "movq %%mm1, %%mm4                      \n\t"
02034         "punpcklbw %%mm3, %%mm1                 \n\t"
02035         "punpckhbw %%mm3, %%mm4                 \n\t"
02036 
02037         "movq %%mm0, %%mm3                      \n\t"
02038         "punpcklwd %%mm1, %%mm0                 \n\t"
02039         "punpckhwd %%mm1, %%mm3                 \n\t"
02040         "movq %%mm2, %%mm1                      \n\t"
02041         "punpcklwd %%mm4, %%mm2                 \n\t"
02042         "punpckhwd %%mm4, %%mm1                 \n\t"
02043 
02044         "movd %%mm0, 128(%2)                    \n\t"
02045         "psrlq $32, %%mm0                       \n\t"
02046         "movd %%mm0, 144(%2)                    \n\t"
02047         "movd %%mm3, 160(%2)                    \n\t"
02048         "psrlq $32, %%mm3                       \n\t"
02049         "movd %%mm3, 176(%2)                    \n\t"
02050         "movd %%mm3, 48(%3)                     \n\t"
02051         "movd %%mm2, 192(%2)                    \n\t"
02052         "movd %%mm2, 64(%3)                     \n\t"
02053         "psrlq $32, %%mm2                       \n\t"
02054         "movd %%mm2, 80(%3)                     \n\t"
02055         "movd %%mm1, 96(%3)                     \n\t"
02056         "psrlq $32, %%mm1                       \n\t"
02057         "movd %%mm1, 112(%3)                    \n\t"
02058 
02059         "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
02060 
02061         "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
02062         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
02063         "movq %%mm0, %%mm2                      \n\t" // 12345678
02064         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02065         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02066 
02067         "movq (%%"REG_a", %1), %%mm1            \n\t"
02068         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
02069         "movq %%mm1, %%mm4                      \n\t"
02070         "punpcklbw %%mm3, %%mm1                 \n\t"
02071         "punpckhbw %%mm3, %%mm4                 \n\t"
02072 
02073         "movq %%mm0, %%mm3                      \n\t"
02074         "punpcklwd %%mm1, %%mm0                 \n\t"
02075         "punpckhwd %%mm1, %%mm3                 \n\t"
02076         "movq %%mm2, %%mm1                      \n\t"
02077         "punpcklwd %%mm4, %%mm2                 \n\t"
02078         "punpckhwd %%mm4, %%mm1                 \n\t"
02079 
02080         "movd %%mm0, 132(%2)                    \n\t"
02081         "psrlq $32, %%mm0                       \n\t"
02082         "movd %%mm0, 148(%2)                    \n\t"
02083         "movd %%mm3, 164(%2)                    \n\t"
02084         "psrlq $32, %%mm3                       \n\t"
02085         "movd %%mm3, 180(%2)                    \n\t"
02086         "movd %%mm3, 52(%3)                     \n\t"
02087         "movd %%mm2, 196(%2)                    \n\t"
02088         "movd %%mm2, 68(%3)                     \n\t"
02089         "psrlq $32, %%mm2                       \n\t"
02090         "movd %%mm2, 84(%3)                     \n\t"
02091         "movd %%mm1, 100(%3)                    \n\t"
02092         "psrlq $32, %%mm1                       \n\t"
02093         "movd %%mm1, 116(%3)                    \n\t"
02094 
02095 
02096         :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)
02097         : "%"REG_a
02098     );
02099 }
02100 
02104 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
02105 {
02106     __asm__(
02107         "lea (%0, %1), %%"REG_a"                \n\t"
02108         "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
02109 //      0       1       2       3       4       5       6       7       8       9
02110 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
02111         "movq (%2), %%mm0                       \n\t" // 12345678
02112         "movq 16(%2), %%mm1                     \n\t" // abcdefgh
02113         "movq %%mm0, %%mm2                      \n\t" // 12345678
02114         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02115         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02116 
02117         "movq 32(%2), %%mm1                     \n\t"
02118         "movq 48(%2), %%mm3                     \n\t"
02119         "movq %%mm1, %%mm4                      \n\t"
02120         "punpcklbw %%mm3, %%mm1                 \n\t"
02121         "punpckhbw %%mm3, %%mm4                 \n\t"
02122 
02123         "movq %%mm0, %%mm3                      \n\t"
02124         "punpcklwd %%mm1, %%mm0                 \n\t"
02125         "punpckhwd %%mm1, %%mm3                 \n\t"
02126         "movq %%mm2, %%mm1                      \n\t"
02127         "punpcklwd %%mm4, %%mm2                 \n\t"
02128         "punpckhwd %%mm4, %%mm1                 \n\t"
02129 
02130         "movd %%mm0, (%0)                       \n\t"
02131         "psrlq $32, %%mm0                       \n\t"
02132         "movd %%mm0, (%%"REG_a")                \n\t"
02133         "movd %%mm3, (%%"REG_a", %1)            \n\t"
02134         "psrlq $32, %%mm3                       \n\t"
02135         "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
02136         "movd %%mm2, (%0, %1, 4)                \n\t"
02137         "psrlq $32, %%mm2                       \n\t"
02138         "movd %%mm2, (%%"REG_d")                \n\t"
02139         "movd %%mm1, (%%"REG_d", %1)            \n\t"
02140         "psrlq $32, %%mm1                       \n\t"
02141         "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
02142 
02143 
02144         "movq 64(%2), %%mm0                     \n\t" // 12345678
02145         "movq 80(%2), %%mm1                     \n\t" // abcdefgh
02146         "movq %%mm0, %%mm2                      \n\t" // 12345678
02147         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02148         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02149 
02150         "movq 96(%2), %%mm1                     \n\t"
02151         "movq 112(%2), %%mm3                    \n\t"
02152         "movq %%mm1, %%mm4                      \n\t"
02153         "punpcklbw %%mm3, %%mm1                 \n\t"
02154         "punpckhbw %%mm3, %%mm4                 \n\t"
02155 
02156         "movq %%mm0, %%mm3                      \n\t"
02157         "punpcklwd %%mm1, %%mm0                 \n\t"
02158         "punpckhwd %%mm1, %%mm3                 \n\t"
02159         "movq %%mm2, %%mm1                      \n\t"
02160         "punpcklwd %%mm4, %%mm2                 \n\t"
02161         "punpckhwd %%mm4, %%mm1                 \n\t"
02162 
02163         "movd %%mm0, 4(%0)                      \n\t"
02164         "psrlq $32, %%mm0                       \n\t"
02165         "movd %%mm0, 4(%%"REG_a")               \n\t"
02166         "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
02167         "psrlq $32, %%mm3                       \n\t"
02168         "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
02169         "movd %%mm2, 4(%0, %1, 4)               \n\t"
02170         "psrlq $32, %%mm2                       \n\t"
02171         "movd %%mm2, 4(%%"REG_d")               \n\t"
02172         "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
02173         "psrlq $32, %%mm1                       \n\t"
02174         "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
02175 
02176         :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)
02177         : "%"REG_a, "%"REG_d
02178     );
02179 }
02180 #endif //HAVE_MMX
02181 //static long test=0;
02182 
02183 #if !HAVE_ALTIVEC
02184 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
02185                                     uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)
02186 {
02187     // to save a register (FIXME do this outside of the loops)
02188     tempBlurredPast[127]= maxNoise[0];
02189     tempBlurredPast[128]= maxNoise[1];
02190     tempBlurredPast[129]= maxNoise[2];
02191 
02192 #define FAST_L2_DIFF
02193 //#define L1_DIFF //u should change the thresholds too if u try that one
02194 #if HAVE_MMX2 || HAVE_AMD3DNOW
02195     __asm__ volatile(
02196         "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
02197         "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
02198         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02199 //      0       1       2       3       4       5       6       7       8       9
02200 //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
02201 //FIXME reorder?
02202 #ifdef L1_DIFF //needs mmx2
02203         "movq (%0), %%mm0                       \n\t" // L0
02204         "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
02205         "movq (%0, %2), %%mm1                   \n\t" // L1
02206         "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
02207         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02208         "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
02209         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02210         "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
02211 
02212         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02213         "paddw %%mm1, %%mm0                     \n\t"
02214         "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
02215         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02216         "paddw %%mm2, %%mm0                     \n\t"
02217         "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
02218         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02219         "paddw %%mm3, %%mm0                     \n\t"
02220         "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
02221         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02222         "paddw %%mm4, %%mm0                     \n\t"
02223         "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
02224         "paddw %%mm5, %%mm6                     \n\t"
02225         "paddw %%mm7, %%mm6                     \n\t"
02226         "paddw %%mm6, %%mm0                     \n\t"
02227 #else //L1_DIFF
02228 #if defined (FAST_L2_DIFF)
02229         "pcmpeqb %%mm7, %%mm7                   \n\t"
02230         "movq "MANGLE(b80)", %%mm6              \n\t"
02231         "pxor %%mm0, %%mm0                      \n\t"
02232 #define REAL_L2_DIFF_CORE(a, b)\
02233         "movq " #a ", %%mm5                     \n\t"\
02234         "movq " #b ", %%mm2                     \n\t"\
02235         "pxor %%mm7, %%mm2                      \n\t"\
02236         PAVGB(%%mm2, %%mm5)\
02237         "paddb %%mm6, %%mm5                     \n\t"\
02238         "movq %%mm5, %%mm2                      \n\t"\
02239         "psllw $8, %%mm5                        \n\t"\
02240         "pmaddwd %%mm5, %%mm5                   \n\t"\
02241         "pmaddwd %%mm2, %%mm2                   \n\t"\
02242         "paddd %%mm2, %%mm5                     \n\t"\
02243         "psrld $14, %%mm5                       \n\t"\
02244         "paddd %%mm5, %%mm0                     \n\t"
02245 
02246 #else //defined (FAST_L2_DIFF)
02247         "pxor %%mm7, %%mm7                      \n\t"
02248         "pxor %%mm0, %%mm0                      \n\t"
02249 #define REAL_L2_DIFF_CORE(a, b)\
02250         "movq " #a ", %%mm5                     \n\t"\
02251         "movq " #b ", %%mm2                     \n\t"\
02252         "movq %%mm5, %%mm1                      \n\t"\
02253         "movq %%mm2, %%mm3                      \n\t"\
02254         "punpcklbw %%mm7, %%mm5                 \n\t"\
02255         "punpckhbw %%mm7, %%mm1                 \n\t"\
02256         "punpcklbw %%mm7, %%mm2                 \n\t"\
02257         "punpckhbw %%mm7, %%mm3                 \n\t"\
02258         "psubw %%mm2, %%mm5                     \n\t"\
02259         "psubw %%mm3, %%mm1                     \n\t"\
02260         "pmaddwd %%mm5, %%mm5                   \n\t"\
02261         "pmaddwd %%mm1, %%mm1                   \n\t"\
02262         "paddd %%mm1, %%mm5                     \n\t"\
02263         "paddd %%mm5, %%mm0                     \n\t"
02264 
02265 #endif //defined (FAST_L2_DIFF)
02266 
02267 #define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)
02268 
02269 L2_DIFF_CORE((%0)          , (%1))
02270 L2_DIFF_CORE((%0, %2)      , (%1, %2))
02271 L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))
02272 L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))
02273 L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))
02274 L2_DIFF_CORE((%0, %%REGd)  , (%1, %%REGd))
02275 L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
02276 L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))
02277 
02278 #endif //L1_DIFF
02279 
02280         "movq %%mm0, %%mm4                      \n\t"
02281         "psrlq $32, %%mm0                       \n\t"
02282         "paddd %%mm0, %%mm4                     \n\t"
02283         "movd %%mm4, %%ecx                      \n\t"
02284         "shll $2, %%ecx                         \n\t"
02285         "mov %3, %%"REG_d"                      \n\t"
02286         "addl -4(%%"REG_d"), %%ecx              \n\t"
02287         "addl 4(%%"REG_d"), %%ecx               \n\t"
02288         "addl -1024(%%"REG_d"), %%ecx           \n\t"
02289         "addl $4, %%ecx                         \n\t"
02290         "addl 1024(%%"REG_d"), %%ecx            \n\t"
02291         "shrl $3, %%ecx                         \n\t"
02292         "movl %%ecx, (%%"REG_d")                \n\t"
02293 
02294 //        "mov %3, %%"REG_c"                      \n\t"
02295 //        "mov %%"REG_c", test                    \n\t"
02296 //        "jmp 4f                                 \n\t"
02297         "cmpl 512(%%"REG_d"), %%ecx             \n\t"
02298         " jb 2f                                 \n\t"
02299         "cmpl 516(%%"REG_d"), %%ecx             \n\t"
02300         " jb 1f                                 \n\t"
02301 
02302         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02303         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02304         "movq (%0), %%mm0                       \n\t" // L0
02305         "movq (%0, %2), %%mm1                   \n\t" // L1
02306         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02307         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02308         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02309         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02310         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02311         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02312         "movq %%mm0, (%1)                       \n\t" // L0
02313         "movq %%mm1, (%1, %2)                   \n\t" // L1
02314         "movq %%mm2, (%1, %2, 2)                \n\t" // L2
02315         "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
02316         "movq %%mm4, (%1, %2, 4)                \n\t" // L4
02317         "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
02318         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
02319         "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
02320         "jmp 4f                                 \n\t"
02321 
02322         "1:                                     \n\t"
02323         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02324         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02325         "movq (%0), %%mm0                       \n\t" // L0
02326         PAVGB((%1), %%mm0)                            // L0
02327         "movq (%0, %2), %%mm1                   \n\t" // L1
02328         PAVGB((%1, %2), %%mm1)                        // L1
02329         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02330         PAVGB((%1, %2, 2), %%mm2)                     // L2
02331         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02332         PAVGB((%1, %%REGa), %%mm3)                    // L3
02333         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02334         PAVGB((%1, %2, 4), %%mm4)                     // L4
02335         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02336         PAVGB((%1, %%REGd), %%mm5)                    // L5
02337         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02338         PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
02339         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02340         PAVGB((%1, %%REGc), %%mm7)                    // L7
02341         "movq %%mm0, (%1)                       \n\t" // R0
02342         "movq %%mm1, (%1, %2)                   \n\t" // R1
02343         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02344         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02345         "movq %%mm4, (%1, %2, 4)                \n\t" // R4
02346         "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
02347         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
02348         "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
02349         "movq %%mm0, (%0)                       \n\t" // L0
02350         "movq %%mm1, (%0, %2)                   \n\t" // L1
02351         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02352         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02353         "movq %%mm4, (%0, %2, 4)                \n\t" // L4
02354         "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
02355         "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
02356         "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
02357         "jmp 4f                                 \n\t"
02358 
02359         "2:                                     \n\t"
02360         "cmpl 508(%%"REG_d"), %%ecx             \n\t"
02361         " jb 3f                                 \n\t"
02362 
02363         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02364         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02365         "movq (%0), %%mm0                       \n\t" // L0
02366         "movq (%0, %2), %%mm1                   \n\t" // L1
02367         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02368         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02369         "movq (%1), %%mm4                       \n\t" // R0
02370         "movq (%1, %2), %%mm5                   \n\t" // R1
02371         "movq (%1, %2, 2), %%mm6                \n\t" // R2
02372         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
02373         PAVGB(%%mm4, %%mm0)
02374         PAVGB(%%mm5, %%mm1)
02375         PAVGB(%%mm6, %%mm2)
02376         PAVGB(%%mm7, %%mm3)
02377         PAVGB(%%mm4, %%mm0)
02378         PAVGB(%%mm5, %%mm1)
02379         PAVGB(%%mm6, %%mm2)
02380         PAVGB(%%mm7, %%mm3)
02381         "movq %%mm0, (%1)                       \n\t" // R0
02382         "movq %%mm1, (%1, %2)                   \n\t" // R1
02383         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02384         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02385         "movq %%mm0, (%0)                       \n\t" // L0
02386         "movq %%mm1, (%0, %2)                   \n\t" // L1
02387         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02388         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02389 
02390         "movq (%0, %2, 4), %%mm0                \n\t" // L4
02391         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
02392         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
02393         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
02394         "movq (%1, %2, 4), %%mm4                \n\t" // R4
02395         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
02396         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
02397         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
02398         PAVGB(%%mm4, %%mm0)
02399         PAVGB(%%mm5, %%mm1)
02400         PAVGB(%%mm6, %%mm2)
02401         PAVGB(%%mm7, %%mm3)
02402         PAVGB(%%mm4, %%mm0)
02403         PAVGB(%%mm5, %%mm1)
02404         PAVGB(%%mm6, %%mm2)
02405         PAVGB(%%mm7, %%mm3)
02406         "movq %%mm0, (%1, %2, 4)                \n\t" // R4
02407         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
02408         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
02409         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
02410         "movq %%mm0, (%0, %2, 4)                \n\t" // L4
02411         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
02412         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
02413         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
02414         "jmp 4f                                 \n\t"
02415 
02416         "3:                                     \n\t"
02417         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02418         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02419         "movq (%0), %%mm0                       \n\t" // L0
02420         "movq (%0, %2), %%mm1                   \n\t" // L1
02421         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02422         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02423         "movq (%1), %%mm4                       \n\t" // R0
02424         "movq (%1, %2), %%mm5                   \n\t" // R1
02425         "movq (%1, %2, 2), %%mm6                \n\t" // R2
02426         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
02427         PAVGB(%%mm4, %%mm0)
02428         PAVGB(%%mm5, %%mm1)
02429         PAVGB(%%mm6, %%mm2)
02430         PAVGB(%%mm7, %%mm3)
02431         PAVGB(%%mm4, %%mm0)
02432         PAVGB(%%mm5, %%mm1)
02433         PAVGB(%%mm6, %%mm2)
02434         PAVGB(%%mm7, %%mm3)
02435         PAVGB(%%mm4, %%mm0)
02436         PAVGB(%%mm5, %%mm1)
02437         PAVGB(%%mm6, %%mm2)
02438         PAVGB(%%mm7, %%mm3)
02439         "movq %%mm0, (%1)                       \n\t" // R0
02440         "movq %%mm1, (%1, %2)                   \n\t" // R1
02441         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02442         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02443         "movq %%mm0, (%0)                       \n\t" // L0
02444         "movq %%mm1, (%0, %2)                   \n\t" // L1
02445         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02446         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02447 
02448         "movq (%0, %2, 4), %%mm0                \n\t" // L4
02449         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
02450         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
02451         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
02452         "movq (%1, %2, 4), %%mm4                \n\t" // R4
02453         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
02454         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
02455         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
02456         PAVGB(%%mm4, %%mm0)
02457         PAVGB(%%mm5, %%mm1)
02458         PAVGB(%%mm6, %%mm2)
02459         PAVGB(%%mm7, %%mm3)
02460         PAVGB(%%mm4, %%mm0)
02461         PAVGB(%%mm5, %%mm1)
02462         PAVGB(%%mm6, %%mm2)
02463         PAVGB(%%mm7, %%mm3)
02464         PAVGB(%%mm4, %%mm0)
02465         PAVGB(%%mm5, %%mm1)
02466         PAVGB(%%mm6, %%mm2)
02467         PAVGB(%%mm7, %%mm3)
02468         "movq %%mm0, (%1, %2, 4)                \n\t" // R4
02469         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
02470         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
02471         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
02472         "movq %%mm0, (%0, %2, 4)                \n\t" // L4
02473         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
02474         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
02475         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
02476 
02477         "4:                                     \n\t"
02478 
02479         :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)
02480         : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
02481     );
02482 #else //HAVE_MMX2 || HAVE_AMD3DNOW
02483 {
02484     int y;
02485     int d=0;
02486 //    int sysd=0;
02487     int i;
02488 
02489     for(y=0; y<8; y++){
02490         int x;
02491         for(x=0; x<8; x++){
02492             int ref= tempBlurred[ x + y*stride ];
02493             int cur= src[ x + y*stride ];
02494             int d1=ref - cur;
02495 //            if(x==0 || x==7) d1+= d1>>1;
02496 //            if(y==0 || y==7) d1+= d1>>1;
02497 //            d+= FFABS(d1);
02498             d+= d1*d1;
02499 //            sysd+= d1;
02500         }
02501     }
02502     i=d;
02503     d=  (
02504         4*d
02505         +(*(tempBlurredPast-256))
02506         +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
02507         +(*(tempBlurredPast+256))
02508         +4)>>3;
02509     *tempBlurredPast=i;
02510 //    ((*tempBlurredPast)*3 + d + 2)>>2;
02511 
02512 /*
02513 Switch between
02514  1  0  0  0  0  0  0  (0)
02515 64 32 16  8  4  2  1  (1)
02516 64 48 36 27 20 15 11 (33) (approx)
02517 64 56 49 43 37 33 29 (200) (approx)
02518 */
02519     if(d > maxNoise[1]){
02520         if(d < maxNoise[2]){
02521             for(y=0; y<8; y++){
02522                 int x;
02523                 for(x=0; x<8; x++){
02524                     int ref= tempBlurred[ x + y*stride ];
02525                     int cur= src[ x + y*stride ];
02526                     tempBlurred[ x + y*stride ]=
02527                     src[ x + y*stride ]=
02528                         (ref + cur + 1)>>1;
02529                 }
02530             }
02531         }else{
02532             for(y=0; y<8; y++){
02533                 int x;
02534                 for(x=0; x<8; x++){
02535                     tempBlurred[ x + y*stride ]= src[ x + y*stride ];
02536                 }
02537             }
02538         }
02539     }else{
02540         if(d < maxNoise[0]){
02541             for(y=0; y<8; y++){
02542                 int x;
02543                 for(x=0; x<8; x++){
02544                     int ref= tempBlurred[ x + y*stride ];
02545                     int cur= src[ x + y*stride ];
02546                     tempBlurred[ x + y*stride ]=
02547                     src[ x + y*stride ]=
02548                         (ref*7 + cur + 4)>>3;
02549                 }
02550             }
02551         }else{
02552             for(y=0; y<8; y++){
02553                 int x;
02554                 for(x=0; x<8; x++){
02555                     int ref= tempBlurred[ x + y*stride ];
02556                     int cur= src[ x + y*stride ];
02557                     tempBlurred[ x + y*stride ]=
02558                     src[ x + y*stride ]=
02559                         (ref*3 + cur + 2)>>2;
02560                 }
02561             }
02562         }
02563     }
02564 }
02565 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
02566 }
02567 #endif //HAVE_ALTIVEC
02568 
02569 #if HAVE_MMX
02570 
02573 static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
02574     int64_t dc_mask, eq_mask, both_masks;
02575     int64_t sums[10*8*2];
02576     src+= step*3; // src points to begin of the 8x8 Block
02577 //START_TIMER
02578     __asm__ volatile(
02579         "movq %0, %%mm7                         \n\t"
02580         "movq %1, %%mm6                         \n\t"
02581         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
02582         );
02583 
02584     __asm__ volatile(
02585         "lea (%2, %3), %%"REG_a"                \n\t"
02586 //      0       1       2       3       4       5       6       7       8       9
02587 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
02588 
02589         "movq (%2), %%mm0                       \n\t"
02590         "movq (%%"REG_a"), %%mm1                \n\t"
02591         "movq %%mm1, %%mm3                      \n\t"
02592         "movq %%mm1, %%mm4                      \n\t"
02593         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
02594         "paddb %%mm7, %%mm0                     \n\t"
02595         "pcmpgtb %%mm6, %%mm0                   \n\t"
02596 
02597         "movq (%%"REG_a",%3), %%mm2             \n\t"
02598         PMAXUB(%%mm2, %%mm4)
02599         PMINUB(%%mm2, %%mm3, %%mm5)
02600         "psubb %%mm2, %%mm1                     \n\t"
02601         "paddb %%mm7, %%mm1                     \n\t"
02602         "pcmpgtb %%mm6, %%mm1                   \n\t"
02603         "paddb %%mm1, %%mm0                     \n\t"
02604 
02605         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
02606         PMAXUB(%%mm1, %%mm4)
02607         PMINUB(%%mm1, %%mm3, %%mm5)
02608         "psubb %%mm1, %%mm2                     \n\t"
02609         "paddb %%mm7, %%mm2                     \n\t"
02610         "pcmpgtb %%mm6, %%mm2                   \n\t"
02611         "paddb %%mm2, %%mm0                     \n\t"
02612 
02613         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
02614 
02615         "movq (%2, %3, 4), %%mm2                \n\t"
02616         PMAXUB(%%mm2, %%mm4)
02617         PMINUB(%%mm2, %%mm3, %%mm5)
02618         "psubb %%mm2, %%mm1                     \n\t"
02619         "paddb %%mm7, %%mm1                     \n\t"
02620         "pcmpgtb %%mm6, %%mm1                   \n\t"
02621         "paddb %%mm1, %%mm0                     \n\t"
02622 
02623         "movq (%%"REG_a"), %%mm1                \n\t"
02624         PMAXUB(%%mm1, %%mm4)
02625         PMINUB(%%mm1, %%mm3, %%mm5)
02626         "psubb %%mm1, %%mm2                     \n\t"
02627         "paddb %%mm7, %%mm2                     \n\t"
02628         "pcmpgtb %%mm6, %%mm2                   \n\t"
02629         "paddb %%mm2, %%mm0                     \n\t"
02630 
02631         "movq (%%"REG_a", %3), %%mm2            \n\t"
02632         PMAXUB(%%mm2, %%mm4)
02633         PMINUB(%%mm2, %%mm3, %%mm5)
02634         "psubb %%mm2, %%mm1                     \n\t"
02635         "paddb %%mm7, %%mm1                     \n\t"
02636         "pcmpgtb %%mm6, %%mm1                   \n\t"
02637         "paddb %%mm1, %%mm0                     \n\t"
02638 
02639         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
02640         PMAXUB(%%mm1, %%mm4)
02641         PMINUB(%%mm1, %%mm3, %%mm5)
02642         "psubb %%mm1, %%mm2                     \n\t"
02643         "paddb %%mm7, %%mm2                     \n\t"
02644         "pcmpgtb %%mm6, %%mm2                   \n\t"
02645         "paddb %%mm2, %%mm0                     \n\t"
02646 
02647         "movq (%2, %3, 8), %%mm2                \n\t"
02648         PMAXUB(%%mm2, %%mm4)
02649         PMINUB(%%mm2, %%mm3, %%mm5)
02650         "psubb %%mm2, %%mm1                     \n\t"
02651         "paddb %%mm7, %%mm1                     \n\t"
02652         "pcmpgtb %%mm6, %%mm1                   \n\t"
02653         "paddb %%mm1, %%mm0                     \n\t"
02654 
02655         "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
02656         "psubb %%mm1, %%mm2                     \n\t"
02657         "paddb %%mm7, %%mm2                     \n\t"
02658         "pcmpgtb %%mm6, %%mm2                   \n\t"
02659         "paddb %%mm2, %%mm0                     \n\t"
02660         "psubusb %%mm3, %%mm4                   \n\t"
02661 
02662         "pxor %%mm6, %%mm6                      \n\t"
02663         "movq %4, %%mm7                         \n\t" // QP,..., QP
02664         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
02665         "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
02666         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
02667         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
02668         "movq %%mm7, %1                         \n\t"
02669 
02670         "movq %5, %%mm7                         \n\t"
02671         "punpcklbw %%mm7, %%mm7                 \n\t"
02672         "punpcklbw %%mm7, %%mm7                 \n\t"
02673         "punpcklbw %%mm7, %%mm7                 \n\t"
02674         "psubb %%mm0, %%mm6                     \n\t"
02675         "pcmpgtb %%mm7, %%mm6                   \n\t"
02676         "movq %%mm6, %0                         \n\t"
02677 
02678         : "=m" (eq_mask), "=m" (dc_mask)
02679         : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
02680         : "%"REG_a
02681     );
02682 
02683     both_masks = dc_mask & eq_mask;
02684 
02685     if(both_masks){
02686         x86_reg offset= -8*step;
02687         int64_t *temp_sums= sums;
02688 
02689         __asm__ volatile(
02690             "movq %2, %%mm0                         \n\t"  // QP,..., QP
02691             "pxor %%mm4, %%mm4                      \n\t"
02692 
02693             "movq (%0), %%mm6                       \n\t"
02694             "movq (%0, %1), %%mm5                   \n\t"
02695             "movq %%mm5, %%mm1                      \n\t"
02696             "movq %%mm6, %%mm2                      \n\t"
02697             "psubusb %%mm6, %%mm5                   \n\t"
02698             "psubusb %%mm1, %%mm2                   \n\t"
02699             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
02700             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
02701             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
02702 
02703             "pxor %%mm6, %%mm1                      \n\t"
02704             "pand %%mm0, %%mm1                      \n\t"
02705             "pxor %%mm1, %%mm6                      \n\t"
02706             // 0:QP  6:First
02707 
02708             "movq (%0, %1, 8), %%mm5                \n\t"
02709             "add %1, %0                             \n\t" // %0 points to line 1 not 0
02710             "movq (%0, %1, 8), %%mm7                \n\t"
02711             "movq %%mm5, %%mm1                      \n\t"
02712             "movq %%mm7, %%mm2                      \n\t"
02713             "psubusb %%mm7, %%mm5                   \n\t"
02714             "psubusb %%mm1, %%mm2                   \n\t"
02715             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
02716             "movq %2, %%mm0                         \n\t"  // QP,..., QP
02717             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
02718             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
02719 
02720             "pxor %%mm7, %%mm1                      \n\t"
02721             "pand %%mm0, %%mm1                      \n\t"
02722             "pxor %%mm1, %%mm7                      \n\t"
02723 
02724             "movq %%mm6, %%mm5                      \n\t"
02725             "punpckhbw %%mm4, %%mm6                 \n\t"
02726             "punpcklbw %%mm4, %%mm5                 \n\t"
02727             // 4:0 5/6:First 7:Last
02728 
02729             "movq %%mm5, %%mm0                      \n\t"
02730             "movq %%mm6, %%mm1                      \n\t"
02731             "psllw $2, %%mm0                        \n\t"
02732             "psllw $2, %%mm1                        \n\t"
02733             "paddw "MANGLE(w04)", %%mm0             \n\t"
02734             "paddw "MANGLE(w04)", %%mm1             \n\t"
02735 
02736 #define NEXT\
02737             "movq (%0), %%mm2                       \n\t"\
02738             "movq (%0), %%mm3                       \n\t"\
02739             "add %1, %0                             \n\t"\
02740             "punpcklbw %%mm4, %%mm2                 \n\t"\
02741             "punpckhbw %%mm4, %%mm3                 \n\t"\
02742             "paddw %%mm2, %%mm0                     \n\t"\
02743             "paddw %%mm3, %%mm1                     \n\t"
02744 
02745 #define PREV\
02746             "movq (%0), %%mm2                       \n\t"\
02747             "movq (%0), %%mm3                       \n\t"\
02748             "add %1, %0                             \n\t"\
02749             "punpcklbw %%mm4, %%mm2                 \n\t"\
02750             "punpckhbw %%mm4, %%mm3                 \n\t"\
02751             "psubw %%mm2, %%mm0                     \n\t"\
02752             "psubw %%mm3, %%mm1                     \n\t"
02753 
02754 
02755             NEXT //0
02756             NEXT //1
02757             NEXT //2
02758             "movq %%mm0, (%3)                       \n\t"
02759             "movq %%mm1, 8(%3)                      \n\t"
02760 
02761             NEXT //3
02762             "psubw %%mm5, %%mm0                     \n\t"
02763             "psubw %%mm6, %%mm1                     \n\t"
02764             "movq %%mm0, 16(%3)                     \n\t"
02765             "movq %%mm1, 24(%3)                     \n\t"
02766 
02767             NEXT //4
02768             "psubw %%mm5, %%mm0                     \n\t"
02769             "psubw %%mm6, %%mm1                     \n\t"
02770             "movq %%mm0, 32(%3)                     \n\t"
02771             "movq %%mm1, 40(%3)                     \n\t"
02772 
02773             NEXT //5
02774             "psubw %%mm5, %%mm0                     \n\t"
02775             "psubw %%mm6, %%mm1                     \n\t"
02776             "movq %%mm0, 48(%3)                     \n\t"
02777             "movq %%mm1, 56(%3)                     \n\t"
02778 
02779             NEXT //6
02780             "psubw %%mm5, %%mm0                     \n\t"
02781             "psubw %%mm6, %%mm1                     \n\t"
02782             "movq %%mm0, 64(%3)                     \n\t"
02783             "movq %%mm1, 72(%3)                     \n\t"
02784 
02785             "movq %%mm7, %%mm6                      \n\t"
02786             "punpckhbw %%mm4, %%mm7                 \n\t"
02787             "punpcklbw %%mm4, %%mm6                 \n\t"
02788 
02789             NEXT //7
02790             "mov %4, %0                             \n\t"
02791             "add %1, %0                             \n\t"
02792             PREV //0
02793             "movq %%mm0, 80(%3)                     \n\t"
02794             "movq %%mm1, 88(%3)                     \n\t"
02795 
02796             PREV //1
02797             "paddw %%mm6, %%mm0                     \n\t"
02798             "paddw %%mm7, %%mm1                     \n\t"
02799             "movq %%mm0, 96(%3)                     \n\t"
02800             "movq %%mm1, 104(%3)                    \n\t"
02801 
02802             PREV //2
02803             "paddw %%mm6, %%mm0                     \n\t"
02804             "paddw %%mm7, %%mm1                     \n\t"
02805             "movq %%mm0, 112(%3)                    \n\t"
02806             "movq %%mm1, 120(%3)                    \n\t"
02807 
02808             PREV //3
02809             "paddw %%mm6, %%mm0                     \n\t"
02810             "paddw %%mm7, %%mm1                     \n\t"
02811             "movq %%mm0, 128(%3)                    \n\t"
02812             "movq %%mm1, 136(%3)                    \n\t"
02813 
02814             PREV //4
02815             "paddw %%mm6, %%mm0                     \n\t"
02816             "paddw %%mm7, %%mm1                     \n\t"
02817             "movq %%mm0, 144(%3)                    \n\t"
02818             "movq %%mm1, 152(%3)                    \n\t"
02819 
02820             "mov %4, %0                             \n\t" //FIXME
02821 
02822             : "+&r"(src)
02823             : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)
02824         );
02825 
02826         src+= step; // src points to begin of the 8x8 Block
02827 
02828         __asm__ volatile(
02829             "movq %4, %%mm6                         \n\t"
02830             "pcmpeqb %%mm5, %%mm5                   \n\t"
02831             "pxor %%mm6, %%mm5                      \n\t"
02832             "pxor %%mm7, %%mm7                      \n\t"
02833 
02834             "1:                                     \n\t"
02835             "movq (%1), %%mm0                       \n\t"
02836             "movq 8(%1), %%mm1                      \n\t"
02837             "paddw 32(%1), %%mm0                    \n\t"
02838             "paddw 40(%1), %%mm1                    \n\t"
02839             "movq (%0, %3), %%mm2                   \n\t"
02840             "movq %%mm2, %%mm3                      \n\t"
02841             "movq %%mm2, %%mm4                      \n\t"
02842             "punpcklbw %%mm7, %%mm2                 \n\t"
02843             "punpckhbw %%mm7, %%mm3                 \n\t"
02844             "paddw %%mm2, %%mm0                     \n\t"
02845             "paddw %%mm3, %%mm1                     \n\t"
02846             "paddw %%mm2, %%mm0                     \n\t"
02847             "paddw %%mm3, %%mm1                     \n\t"
02848             "psrlw $4, %%mm0                        \n\t"
02849             "psrlw $4, %%mm1                        \n\t"
02850             "packuswb %%mm1, %%mm0                  \n\t"
02851             "pand %%mm6, %%mm0                      \n\t"
02852             "pand %%mm5, %%mm4                      \n\t"
02853             "por %%mm4, %%mm0                       \n\t"
02854             "movq %%mm0, (%0, %3)                   \n\t"
02855             "add $16, %1                            \n\t"
02856             "add %2, %0                             \n\t"
02857             " js 1b                                 \n\t"
02858 
02859             : "+r"(offset), "+r"(temp_sums)
02860             : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)
02861         );
02862     }else
02863         src+= step; // src points to begin of the 8x8 Block
02864 
02865     if(eq_mask != -1LL){
02866         uint8_t *temp_src= src;
02867         __asm__ volatile(
02868             "pxor %%mm7, %%mm7                      \n\t"
02869             "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
02870             "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
02871 //      0       1       2       3       4       5       6       7       8       9
02872 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1
02873 
02874             "movq (%0), %%mm0                       \n\t"
02875             "movq %%mm0, %%mm1                      \n\t"
02876             "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
02877             "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
02878 
02879             "movq (%0, %1), %%mm2                   \n\t"
02880             "lea (%0, %1, 2), %%"REG_a"             \n\t"
02881             "movq %%mm2, %%mm3                      \n\t"
02882             "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
02883             "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
02884 
02885             "movq (%%"REG_a"), %%mm4                \n\t"
02886             "movq %%mm4, %%mm5                      \n\t"
02887             "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
02888             "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
02889 
02890             "paddw %%mm0, %%mm0                     \n\t" // 2L0
02891             "paddw %%mm1, %%mm1                     \n\t" // 2H0
02892             "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
02893             "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
02894             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
02895             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
02896 
02897             "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
02898             "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
02899             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
02900             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
02901 
02902             "movq (%%"REG_a", %1), %%mm2            \n\t"
02903             "movq %%mm2, %%mm3                      \n\t"
02904             "punpcklbw %%mm7, %%mm2                 \n\t" // L3
02905             "punpckhbw %%mm7, %%mm3                 \n\t" // H3
02906 
02907             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
02908             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
02909             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02910             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02911             "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02912             "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02913 
02914             "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
02915             "movq %%mm0, %%mm1                      \n\t"
02916             "punpcklbw %%mm7, %%mm0                 \n\t" // L4
02917             "punpckhbw %%mm7, %%mm1                 \n\t" // H4
02918 
02919             "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
02920             "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
02921             "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
02922             "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
02923             "paddw %%mm4, %%mm4                     \n\t" // 2L2
02924             "paddw %%mm5, %%mm5                     \n\t" // 2H2
02925             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
02926             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
02927 
02928             "lea (%%"REG_a", %1), %0                \n\t"
02929             "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
02930             "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
02931             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
02932             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
02933 //50 opcodes so far
02934             "movq (%0, %1, 2), %%mm2                \n\t"
02935             "movq %%mm2, %%mm3                      \n\t"
02936             "punpcklbw %%mm7, %%mm2                 \n\t" // L5
02937             "punpckhbw %%mm7, %%mm3                 \n\t" // H5
02938             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
02939             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
02940             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
02941             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
02942 
02943             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
02944             "punpcklbw %%mm7, %%mm6                 \n\t" // L6
02945             "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
02946             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
02947             "punpckhbw %%mm7, %%mm6                 \n\t" // H6
02948             "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
02949 
02950             "paddw %%mm0, %%mm0                     \n\t" // 2L4
02951             "paddw %%mm1, %%mm1                     \n\t" // 2H4
02952             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
02953             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
02954 
02955             "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
02956             "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
02957             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
02958             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
02959 
02960             "movq (%0, %1, 4), %%mm2                \n\t"
02961             "movq %%mm2, %%mm3                      \n\t"
02962             "punpcklbw %%mm7, %%mm2                 \n\t" // L7
02963             "punpckhbw %%mm7, %%mm3                 \n\t" // H7
02964 
02965             "paddw %%mm2, %%mm2                     \n\t" // 2L7
02966             "paddw %%mm3, %%mm3                     \n\t" // 2H7
02967             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
02968             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
02969 
02970             "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02971             "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02972 
02973 #if HAVE_MMX2
02974             "movq %%mm7, %%mm6                      \n\t" // 0
02975             "psubw %%mm0, %%mm6                     \n\t"
02976             "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
02977             "movq %%mm7, %%mm6                      \n\t" // 0
02978             "psubw %%mm1, %%mm6                     \n\t"
02979             "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
02980             "movq %%mm7, %%mm6                      \n\t" // 0
02981             "psubw %%mm2, %%mm6                     \n\t"
02982             "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
02983             "movq %%mm7, %%mm6                      \n\t" // 0
02984             "psubw %%mm3, %%mm6                     \n\t"
02985             "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
02986 #else
02987             "movq %%mm7, %%mm6                      \n\t" // 0
02988             "pcmpgtw %%mm0, %%mm6                   \n\t"
02989             "pxor %%mm6, %%mm0                      \n\t"
02990             "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
02991             "movq %%mm7, %%mm6                      \n\t" // 0
02992             "pcmpgtw %%mm1, %%mm6                   \n\t"
02993             "pxor %%mm6, %%mm1                      \n\t"
02994             "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
02995             "movq %%mm7, %%mm6                      \n\t" // 0
02996             "pcmpgtw %%mm2, %%mm6                   \n\t"
02997             "pxor %%mm6, %%mm2                      \n\t"
02998             "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
02999             "movq %%mm7, %%mm6                      \n\t" // 0
03000             "pcmpgtw %%mm3, %%mm6                   \n\t"
03001             "pxor %%mm6, %%mm3                      \n\t"
03002             "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
03003 #endif
03004 
03005 #if HAVE_MMX2
03006             "pminsw %%mm2, %%mm0                    \n\t"
03007             "pminsw %%mm3, %%mm1                    \n\t"
03008 #else
03009             "movq %%mm0, %%mm6                      \n\t"
03010             "psubusw %%mm2, %%mm6                   \n\t"
03011             "psubw %%mm6, %%mm0                     \n\t"
03012             "movq %%mm1, %%mm6                      \n\t"
03013             "psubusw %%mm3, %%mm6                   \n\t"
03014             "psubw %%mm6, %%mm1                     \n\t"
03015 #endif
03016 
03017             "movd %2, %%mm2                         \n\t" // QP
03018             "punpcklbw %%mm7, %%mm2                 \n\t"
03019 
03020             "movq %%mm7, %%mm6                      \n\t" // 0
03021             "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
03022             "pxor %%mm6, %%mm4                      \n\t"
03023             "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
03024             "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
03025             "pxor %%mm7, %%mm5                      \n\t"
03026             "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
03027 // 100 opcodes
03028             "psllw $3, %%mm2                        \n\t" // 8QP
03029             "movq %%mm2, %%mm3                      \n\t" // 8QP
03030             "pcmpgtw %%mm4, %%mm2                   \n\t"
03031             "pcmpgtw %%mm5, %%mm3                   \n\t"
03032             "pand %%mm2, %%mm4                      \n\t"
03033             "pand %%mm3, %%mm5                      \n\t"
03034 
03035 
03036             "psubusw %%mm0, %%mm4                   \n\t" // hd
03037             "psubusw %%mm1, %%mm5                   \n\t" // ld
03038 
03039 
03040             "movq "MANGLE(w05)", %%mm2              \n\t" // 5
03041             "pmullw %%mm2, %%mm4                    \n\t"
03042             "pmullw %%mm2, %%mm5                    \n\t"
03043             "movq "MANGLE(w20)", %%mm2              \n\t" // 32
03044             "paddw %%mm2, %%mm4                     \n\t"
03045             "paddw %%mm2, %%mm5                     \n\t"
03046             "psrlw $6, %%mm4                        \n\t"
03047             "psrlw $6, %%mm5                        \n\t"
03048 
03049             "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
03050             "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
03051 
03052             "pxor %%mm2, %%mm2                      \n\t"
03053             "pxor %%mm3, %%mm3                      \n\t"
03054 
03055             "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
03056             "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
03057             "pxor %%mm2, %%mm0                      \n\t"
03058             "pxor %%mm3, %%mm1                      \n\t"
03059             "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
03060             "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
03061             "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
03062             "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
03063 
03064             "pxor %%mm6, %%mm2                      \n\t"
03065             "pxor %%mm7, %%mm3                      \n\t"
03066             "pand %%mm2, %%mm4                      \n\t"
03067             "pand %%mm3, %%mm5                      \n\t"
03068 
03069 #if HAVE_MMX2
03070             "pminsw %%mm0, %%mm4                    \n\t"
03071             "pminsw %%mm1, %%mm5                    \n\t"
03072 #else
03073             "movq %%mm4, %%mm2                      \n\t"
03074             "psubusw %%mm0, %%mm2                   \n\t"
03075             "psubw %%mm2, %%mm4                     \n\t"
03076             "movq %%mm5, %%mm2                      \n\t"
03077             "psubusw %%mm1, %%mm2                   \n\t"
03078             "psubw %%mm2, %%mm5                     \n\t"
03079 #endif
03080             "pxor %%mm6, %%mm4                      \n\t"
03081             "pxor %%mm7, %%mm5                      \n\t"
03082             "psubw %%mm6, %%mm4                     \n\t"
03083             "psubw %%mm7, %%mm5                     \n\t"
03084             "packsswb %%mm5, %%mm4                  \n\t"
03085             "movq %3, %%mm1                         \n\t"
03086             "pandn %%mm4, %%mm1                     \n\t"
03087             "movq (%0), %%mm0                       \n\t"
03088             "paddb   %%mm1, %%mm0                   \n\t"
03089             "movq %%mm0, (%0)                       \n\t"
03090             "movq (%0, %1), %%mm0                   \n\t"
03091             "psubb %%mm1, %%mm0                     \n\t"
03092             "movq %%mm0, (%0, %1)                   \n\t"
03093 
03094             : "+r" (temp_src)
03095             : "r" ((x86_reg)step), "m" (c->pQPb), "m"(eq_mask)
03096             : "%"REG_a, "%"REG_c
03097         );
03098     }
03099 /*if(step==16){
03100     STOP_TIMER("step16")
03101 }else{
03102     STOP_TIMER("stepX")
03103 }*/
03104 }
03105 #endif //HAVE_MMX
03106 
03107 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
03108                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
03109 
03114 #undef SCALED_CPY
03115 
03116 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
03117                                      int levelFix, int64_t *packedOffsetAndScale)
03118 {
03119 #if !HAVE_MMX
03120     int i;
03121 #endif
03122     if(levelFix){
03123 #if HAVE_MMX
03124     __asm__ volatile(
03125         "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
03126         "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
03127         "lea (%2,%4), %%"REG_a"         \n\t"
03128         "lea (%3,%5), %%"REG_d"         \n\t"
03129         "pxor %%mm4, %%mm4              \n\t"
03130 #if HAVE_MMX2
03131 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \
03132         "movq " #src1 ", %%mm0          \n\t"\
03133         "movq " #src1 ", %%mm5          \n\t"\
03134         "movq " #src2 ", %%mm1          \n\t"\
03135         "movq " #src2 ", %%mm6          \n\t"\
03136         "punpcklbw %%mm0, %%mm0         \n\t"\
03137         "punpckhbw %%mm5, %%mm5         \n\t"\
03138         "punpcklbw %%mm1, %%mm1         \n\t"\
03139         "punpckhbw %%mm6, %%mm6         \n\t"\
03140         "pmulhuw %%mm3, %%mm0           \n\t"\
03141         "pmulhuw %%mm3, %%mm5           \n\t"\
03142         "pmulhuw %%mm3, %%mm1           \n\t"\
03143         "pmulhuw %%mm3, %%mm6           \n\t"\
03144         "psubw %%mm2, %%mm0             \n\t"\
03145         "psubw %%mm2, %%mm5             \n\t"\
03146         "psubw %%mm2, %%mm1             \n\t"\
03147         "psubw %%mm2, %%mm6             \n\t"\
03148         "packuswb %%mm5, %%mm0          \n\t"\
03149         "packuswb %%mm6, %%mm1          \n\t"\
03150         "movq %%mm0, " #dst1 "          \n\t"\
03151         "movq %%mm1, " #dst2 "          \n\t"\
03152 
03153 #else //HAVE_MMX2
03154 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \
03155         "movq " #src1 ", %%mm0          \n\t"\
03156         "movq " #src1 ", %%mm5          \n\t"\
03157         "punpcklbw %%mm4, %%mm0         \n\t"\
03158         "punpckhbw %%mm4, %%mm5         \n\t"\
03159         "psubw %%mm2, %%mm0             \n\t"\
03160         "psubw %%mm2, %%mm5             \n\t"\
03161         "movq " #src2 ", %%mm1          \n\t"\
03162         "psllw $6, %%mm0                \n\t"\
03163         "psllw $6, %%mm5                \n\t"\
03164         "pmulhw %%mm3, %%mm0            \n\t"\
03165         "movq " #src2 ", %%mm6          \n\t"\
03166         "pmulhw %%mm3, %%mm5            \n\t"\
03167         "punpcklbw %%mm4, %%mm1         \n\t"\
03168         "punpckhbw %%mm4, %%mm6         \n\t"\
03169         "psubw %%mm2, %%mm1             \n\t"\
03170         "psubw %%mm2, %%mm6             \n\t"\
03171         "psllw $6, %%mm1                \n\t"\
03172         "psllw $6, %%mm6                \n\t"\
03173         "pmulhw %%mm3, %%mm1            \n\t"\
03174         "pmulhw %%mm3, %%mm6            \n\t"\
03175         "packuswb %%mm5, %%mm0          \n\t"\
03176         "packuswb %%mm6, %%mm1          \n\t"\
03177         "movq %%mm0, " #dst1 "          \n\t"\
03178         "movq %%mm1, " #dst2 "          \n\t"\
03179 
03180 #endif //HAVE_MMX2
03181 #define SCALED_CPY(src1, src2, dst1, dst2)\
03182    REAL_SCALED_CPY(src1, src2, dst1, dst2)
03183 
03184 SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
03185 SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
03186 SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
03187         "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
03188         "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
03189 SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
03190 
03191 
03192         : "=&a" (packedOffsetAndScale)
03193         : "0" (packedOffsetAndScale),
03194         "r"(src),
03195         "r"(dst),
03196         "r" ((x86_reg)srcStride),
03197         "r" ((x86_reg)dstStride)
03198         : "%"REG_d
03199     );
03200 #else //HAVE_MMX
03201     for(i=0; i<8; i++)
03202         memcpy( &(dst[dstStride*i]),
03203                 &(src[srcStride*i]), BLOCK_SIZE);
03204 #endif //HAVE_MMX
03205     }else{
03206 #if HAVE_MMX
03207     __asm__ volatile(
03208         "lea (%0,%2), %%"REG_a"                 \n\t"
03209         "lea (%1,%3), %%"REG_d"                 \n\t"
03210 
03211 #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \
03212         "movq " #src1 ", %%mm0          \n\t"\
03213         "movq " #src2 ", %%mm1          \n\t"\
03214         "movq %%mm0, " #dst1 "          \n\t"\
03215         "movq %%mm1, " #dst2 "          \n\t"\
03216 
03217 #define SIMPLE_CPY(src1, src2, dst1, dst2)\
03218    REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
03219 
03220 SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))
03221 SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
03222 SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
03223         "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
03224         "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
03225 SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
03226 
03227         : : "r" (src),
03228         "r" (dst),
03229         "r" ((x86_reg)srcStride),
03230         "r" ((x86_reg)dstStride)
03231         : "%"REG_a, "%"REG_d
03232     );
03233 #else //HAVE_MMX
03234     for(i=0; i<8; i++)
03235         memcpy( &(dst[dstStride*i]),
03236                 &(src[srcStride*i]), BLOCK_SIZE);
03237 #endif //HAVE_MMX
03238     }
03239 }
03240 
03244 static inline void RENAME(duplicate)(uint8_t src[], int stride)
03245 {
03246 #if HAVE_MMX
03247     __asm__ volatile(
03248         "movq (%0), %%mm0               \n\t"
03249         "add %1, %0                     \n\t"
03250         "movq %%mm0, (%0)               \n\t"
03251         "movq %%mm0, (%0, %1)           \n\t"
03252         "movq %%mm0, (%0, %1, 2)        \n\t"
03253         : "+r" (src)
03254         : "r" ((x86_reg)-stride)
03255     );
03256 #else
03257     int i;
03258     uint8_t *p=src;
03259     for(i=0; i<3; i++){
03260         p-= stride;
03261         memcpy(p, src, 8);
03262     }
03263 #endif
03264 }
03265 
03269 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
03270                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
03271 {
03272     DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
03273     int x,y;
03274 #ifdef COMPILE_TIME_MODE
03275     const int mode= COMPILE_TIME_MODE;
03276 #else
03277     const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
03278 #endif
03279     int black=0, white=255; // blackest black and whitest white in the picture
03280     int QPCorrecture= 256*256;
03281 
03282     int copyAhead;
03283 #if HAVE_MMX
03284     int i;
03285 #endif
03286 
03287     const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
03288     const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
03289 
03290     //FIXME remove
03291     uint64_t * const yHistogram= c.yHistogram;
03292     uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
03293     uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
03294     //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
03295 
03296 #if HAVE_MMX
03297     for(i=0; i<57; i++){
03298         int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
03299         int threshold= offset*2 + 1;
03300         c.mmxDcOffset[i]= 0x7F - offset;
03301         c.mmxDcThreshold[i]= 0x7F - threshold;
03302         c.mmxDcOffset[i]*= 0x0101010101010101LL;
03303         c.mmxDcThreshold[i]*= 0x0101010101010101LL;
03304     }
03305 #endif
03306 
03307     if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
03308     else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
03309             || (mode & FFMPEG_DEINT_FILTER)
03310             || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
03311     else if(   (mode & V_DEBLOCK)
03312             || (mode & LINEAR_IPOL_DEINT_FILTER)
03313             || (mode & MEDIAN_DEINT_FILTER)
03314             || (mode & V_A_DEBLOCK)) copyAhead=13;
03315     else if(mode & V_X1_FILTER) copyAhead=11;
03316 //    else if(mode & V_RK1_FILTER) copyAhead=10;
03317     else if(mode & DERING) copyAhead=9;
03318     else copyAhead=8;
03319 
03320     copyAhead-= 8;
03321 
03322     if(!isColor){
03323         uint64_t sum= 0;
03324         int i;
03325         uint64_t maxClipped;
03326         uint64_t clipped;
03327         double scale;
03328 
03329         c.frameNum++;
03330         // first frame is fscked so we ignore it
03331         if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
03332 
03333         for(i=0; i<256; i++){
03334             sum+= yHistogram[i];
03335         }
03336 
03337         /* We always get a completely black picture first. */
03338         maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
03339 
03340         clipped= sum;
03341         for(black=255; black>0; black--){
03342             if(clipped < maxClipped) break;
03343             clipped-= yHistogram[black];
03344         }
03345 
03346         clipped= sum;
03347         for(white=0; white<256; white++){
03348             if(clipped < maxClipped) break;
03349             clipped-= yHistogram[white];
03350         }
03351 
03352         scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
03353 
03354 #if HAVE_MMX2
03355         c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
03356         c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
03357 #else
03358         c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
03359         c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
03360 #endif
03361 
03362         c.packedYOffset|= c.packedYOffset<<32;
03363         c.packedYOffset|= c.packedYOffset<<16;
03364 
03365         c.packedYScale|= c.packedYScale<<32;
03366         c.packedYScale|= c.packedYScale<<16;
03367 
03368         if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);
03369         else                        QPCorrecture= 256*256;
03370     }else{
03371         c.packedYScale= 0x0100010001000100LL;
03372         c.packedYOffset= 0;
03373         QPCorrecture= 256*256;
03374     }
03375 
03376     /* copy & deinterlace first row of blocks */
03377     y=-BLOCK_SIZE;
03378     {
03379         const uint8_t *srcBlock= &(src[y*srcStride]);
03380         uint8_t *dstBlock= tempDst + dstStride;
03381 
03382         // From this point on it is guaranteed that we can read and write 16 lines downward
03383         // finish 1 block before the next otherwise we might have a problem
03384         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
03385         for(x=0; x<width; x+=BLOCK_SIZE){
03386 
03387 #if HAVE_MMX2
03388 /*
03389             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
03390             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
03391             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
03392             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
03393 */
03394 
03395             __asm__(
03396                 "mov %4, %%"REG_a"              \n\t"
03397                 "shr $2, %%"REG_a"              \n\t"
03398                 "and $6, %%"REG_a"              \n\t"
03399                 "add %5, %%"REG_a"              \n\t"
03400                 "mov %%"REG_a", %%"REG_d"       \n\t"
03401                 "imul %1, %%"REG_a"             \n\t"
03402                 "imul %3, %%"REG_d"             \n\t"
03403                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03404                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03405                 "add %1, %%"REG_a"              \n\t"
03406                 "add %3, %%"REG_d"              \n\t"
03407                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03408                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03409                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
03410                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
03411                 : "%"REG_a, "%"REG_d
03412             );
03413 
03414 #elif HAVE_AMD3DNOW
03415 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
03416 /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
03417             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
03418             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
03419             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
03420 */
03421 #endif
03422 
03423             RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
03424                               srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
03425 
03426             RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
03427 
03428             if(mode & LINEAR_IPOL_DEINT_FILTER)
03429                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
03430             else if(mode & LINEAR_BLEND_DEINT_FILTER)
03431                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
03432             else if(mode & MEDIAN_DEINT_FILTER)
03433                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
03434             else if(mode & CUBIC_IPOL_DEINT_FILTER)
03435                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
03436             else if(mode & FFMPEG_DEINT_FILTER)
03437                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
03438             else if(mode & LOWPASS5_DEINT_FILTER)
03439                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
03440 /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
03441                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
03442 */
03443             dstBlock+=8;
03444             srcBlock+=8;
03445         }
03446         if(width==FFABS(dstStride))
03447             linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
03448         else{
03449             int i;
03450             for(i=0; i<copyAhead; i++){
03451                 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
03452             }
03453         }
03454     }
03455 
03456     for(y=0; y<height; y+=BLOCK_SIZE){
03457         //1% speedup if these are here instead of the inner loop
03458         const uint8_t *srcBlock= &(src[y*srcStride]);
03459         uint8_t *dstBlock= &(dst[y*dstStride]);
03460 #if HAVE_MMX
03461         uint8_t *tempBlock1= c.tempBlocks;
03462         uint8_t *tempBlock2= c.tempBlocks + 8;
03463 #endif
03464         const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
03465         int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
03466         int QP=0;
03467         /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
03468            if not than use a temporary buffer */
03469         if(y+15 >= height){
03470             int i;
03471             /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
03472                blockcopy to dst later */
03473             linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
03474                     FFMAX(height-y-copyAhead, 0), srcStride);
03475 
03476             /* duplicate last line of src to fill the void upto line (copyAhead+7) */
03477             for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
03478                     memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
03479 
03480             /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
03481             linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
03482 
03483             /* duplicate last line of dst to fill the void upto line (copyAhead) */
03484             for(i=height-y+1; i<=copyAhead; i++)
03485                     memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
03486 
03487             dstBlock= tempDst + dstStride;
03488             srcBlock= tempSrc;
03489         }
03490 
03491         // From this point on it is guaranteed that we can read and write 16 lines downward
03492         // finish 1 block before the next otherwise we might have a problem
03493         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
03494         for(x=0; x<width; x+=BLOCK_SIZE){
03495             const int stride= dstStride;
03496 #if HAVE_MMX
03497             uint8_t *tmpXchg;
03498 #endif
03499             if(isColor){
03500                 QP= QPptr[x>>qpHShift];
03501                 c.nonBQP= nonBQPptr[x>>qpHShift];
03502             }else{
03503                 QP= QPptr[x>>4];
03504                 QP= (QP* QPCorrecture + 256*128)>>16;
03505                 c.nonBQP= nonBQPptr[x>>4];
03506                 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
03507                 yHistogram[ srcBlock[srcStride*12 + 4] ]++;
03508             }
03509             c.QP= QP;
03510 #if HAVE_MMX
03511             __asm__ volatile(
03512                 "movd %1, %%mm7         \n\t"
03513                 "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
03514                 "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
03515                 "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP
03516                 "movq %%mm7, %0         \n\t"
03517                 : "=m" (c.pQPb)
03518                 : "r" (QP)
03519             );
03520 #endif
03521 
03522 
03523 #if HAVE_MMX2
03524 /*
03525             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
03526             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
03527             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
03528             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
03529 */
03530 
03531             __asm__(
03532                 "mov %4, %%"REG_a"              \n\t"
03533                 "shr $2, %%"REG_a"              \n\t"
03534                 "and $6, %%"REG_a"              \n\t"
03535                 "add %5, %%"REG_a"              \n\t"
03536                 "mov %%"REG_a", %%"REG_d"       \n\t"
03537                 "imul %1, %%"REG_a"             \n\t"
03538                 "imul %3, %%"REG_d"             \n\t"
03539                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03540                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03541                 "add %1, %%"REG_a"              \n\t"
03542                 "add %3, %%"REG_d"              \n\t"
03543                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03544                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03545                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
03546                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
03547                 : "%"REG_a, "%"REG_d
03548             );
03549 
03550 #elif HAVE_AMD3DNOW
03551 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
03552 /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
03553             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
03554             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
03555             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
03556 */
03557 #endif
03558 
03559             RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
03560                               srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
03561 
03562             if(mode & LINEAR_IPOL_DEINT_FILTER)
03563                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
03564             else if(mode & LINEAR_BLEND_DEINT_FILTER)
03565                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
03566             else if(mode & MEDIAN_DEINT_FILTER)
03567                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
03568             else if(mode & CUBIC_IPOL_DEINT_FILTER)
03569                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
03570             else if(mode & FFMPEG_DEINT_FILTER)
03571                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
03572             else if(mode & LOWPASS5_DEINT_FILTER)
03573                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
03574 /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
03575                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
03576 */
03577 
03578             /* only deblock if we have 2 blocks */
03579             if(y + 8 < height){
03580                 if(mode & V_X1_FILTER)
03581                     RENAME(vertX1Filter)(dstBlock, stride, &c);
03582                 else if(mode & V_DEBLOCK){
03583                     const int t= RENAME(vertClassify)(dstBlock, stride, &c);
03584 
03585                     if(t==1)
03586                         RENAME(doVertLowPass)(dstBlock, stride, &c);
03587                     else if(t==2)
03588                         RENAME(doVertDefFilter)(dstBlock, stride, &c);
03589                 }else if(mode & V_A_DEBLOCK){
03590                     RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
03591                 }
03592             }
03593 
03594 #if HAVE_MMX
03595             RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
03596 #endif
03597             /* check if we have a previous block to deblock it with dstBlock */
03598             if(x - 8 >= 0){
03599 #if HAVE_MMX
03600                 if(mode & H_X1_FILTER)
03601                         RENAME(vertX1Filter)(tempBlock1, 16, &c);
03602                 else if(mode & H_DEBLOCK){
03603 //START_TIMER
03604                     const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
03605 //STOP_TIMER("dc & minmax")
03606                     if(t==1)
03607                         RENAME(doVertLowPass)(tempBlock1, 16, &c);
03608                     else if(t==2)
03609                         RENAME(doVertDefFilter)(tempBlock1, 16, &c);
03610                 }else if(mode & H_A_DEBLOCK){
03611                         RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
03612                 }
03613 
03614                 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
03615 
03616 #else
03617                 if(mode & H_X1_FILTER)
03618                     horizX1Filter(dstBlock-4, stride, QP);
03619                 else if(mode & H_DEBLOCK){
03620 #if HAVE_ALTIVEC
03621                     DECLARE_ALIGNED(16, unsigned char, tempBlock[272]);
03622                     transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
03623 
03624                     const int t=vertClassify_altivec(tempBlock-48, 16, &c);
03625                     if(t==1) {
03626                         doVertLowPass_altivec(tempBlock-48, 16, &c);
03627                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
03628                     }
03629                     else if(t==2) {
03630                         doVertDefFilter_altivec(tempBlock-48, 16, &c);
03631                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
03632                     }
03633 #else
03634                     const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
03635 
03636                     if(t==1)
03637                         RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
03638                     else if(t==2)
03639                         RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
03640 #endif
03641                 }else if(mode & H_A_DEBLOCK){
03642                     RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
03643                 }
03644 #endif //HAVE_MMX
03645                 if(mode & DERING){
03646                 //FIXME filter first line
03647                     if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
03648                 }
03649 
03650                 if(mode & TEMP_NOISE_FILTER)
03651                 {
03652                     RENAME(tempNoiseReducer)(dstBlock-8, stride,
03653                             c.tempBlurred[isColor] + y*dstStride + x,
03654                             c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
03655                             c.ppMode.maxTmpNoise);
03656                 }
03657             }
03658 
03659             dstBlock+=8;
03660             srcBlock+=8;
03661 
03662 #if HAVE_MMX
03663             tmpXchg= tempBlock1;
03664             tempBlock1= tempBlock2;
03665             tempBlock2 = tmpXchg;
03666 #endif
03667         }
03668 
03669         if(mode & DERING){
03670             if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
03671         }
03672 
03673         if((mode & TEMP_NOISE_FILTER)){
03674             RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
03675                     c.tempBlurred[isColor] + y*dstStride + x,
03676                     c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
03677                     c.ppMode.maxTmpNoise);
03678         }
03679 
03680         /* did we use a tmp buffer for the last lines*/
03681         if(y+15 >= height){
03682             uint8_t *dstBlock= &(dst[y*dstStride]);
03683             if(width==FFABS(dstStride))
03684                 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
03685             else{
03686                 int i;
03687                 for(i=0; i<height-y; i++){
03688                     memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
03689                 }
03690             }
03691         }
03692 /*
03693         for(x=0; x<width; x+=32){
03694             volatile int i;
03695             i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
03696                 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
03697                 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
03698                 + dstBlock[x +13*dstStride]
03699                 + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
03700         }*/
03701     }
03702 #if   HAVE_AMD3DNOW
03703     __asm__ volatile("femms");
03704 #elif HAVE_MMX
03705     __asm__ volatile("emms");
03706 #endif
03707 
03708 #ifdef DEBUG_BRIGHTNESS
03709     if(!isColor){
03710         int max=1;
03711         int i;
03712         for(i=0; i<256; i++)
03713             if(yHistogram[i] > max) max=yHistogram[i];
03714 
03715         for(i=1; i<256; i++){
03716             int x;
03717             int start=yHistogram[i-1]/(max/256+1);
03718             int end=yHistogram[i]/(max/256+1);
03719             int inc= end > start ? 1 : -1;
03720             for(x=start; x!=end+inc; x+=inc)
03721                 dst[ i*dstStride + x]+=128;
03722         }
03723 
03724         for(i=0; i<100; i+=2){
03725             dst[ (white)*dstStride + i]+=128;
03726             dst[ (black)*dstStride + i]+=128;
03727         }
03728     }
03729 #endif
03730 
03731     *c2= c; //copy local context back
03732 
03733 }