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 REAL_PAVGB
00031 #undef PAVGB
00032 #undef PMINUB
00033 #undef PMAXUB
00034 
00035 #if   HAVE_MMX2
00036 #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
00037 #elif HAVE_AMD3DNOW
00038 #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
00039 #endif
00040 #define PAVGB(a,b)  REAL_PAVGB(a,b)
00041 
00042 #if   HAVE_MMX2
00043 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
00044 #elif HAVE_MMX
00045 #define PMINUB(b,a,t) \
00046     "movq " #a ", " #t " \n\t"\
00047     "psubusb " #b ", " #t " \n\t"\
00048     "psubb " #t ", " #a " \n\t"
00049 #endif
00050 
00051 #if   HAVE_MMX2
00052 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
00053 #elif HAVE_MMX
00054 #define PMAXUB(a,b) \
00055     "psubusb " #a ", " #b " \n\t"\
00056     "paddb " #a ", " #b " \n\t"
00057 #endif
00058 
00059 //FIXME? |255-0| = 1 (should not be a problem ...)
00060 #if HAVE_MMX
00061 
00064 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
00065     int numEq= 0, dcOk;
00066     src+= stride*4; // src points to begin of the 8x8 Block
00067     __asm__ volatile(
00068         "movq %0, %%mm7                         \n\t"
00069         "movq %1, %%mm6                         \n\t"
00070         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
00071         );
00072 
00073     __asm__ volatile(
00074         "lea (%2, %3), %%"REG_a"                \n\t"
00075 //      0       1       2       3       4       5       6       7       8       9
00076 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
00077 
00078         "movq (%2), %%mm0                       \n\t"
00079         "movq (%%"REG_a"), %%mm1                \n\t"
00080         "movq %%mm0, %%mm3                      \n\t"
00081         "movq %%mm0, %%mm4                      \n\t"
00082         PMAXUB(%%mm1, %%mm4)
00083         PMINUB(%%mm1, %%mm3, %%mm5)
00084         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
00085         "paddb %%mm7, %%mm0                     \n\t"
00086         "pcmpgtb %%mm6, %%mm0                   \n\t"
00087 
00088         "movq (%%"REG_a",%3), %%mm2             \n\t"
00089         PMAXUB(%%mm2, %%mm4)
00090         PMINUB(%%mm2, %%mm3, %%mm5)
00091         "psubb %%mm2, %%mm1                     \n\t"
00092         "paddb %%mm7, %%mm1                     \n\t"
00093         "pcmpgtb %%mm6, %%mm1                   \n\t"
00094         "paddb %%mm1, %%mm0                     \n\t"
00095 
00096         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
00097         PMAXUB(%%mm1, %%mm4)
00098         PMINUB(%%mm1, %%mm3, %%mm5)
00099         "psubb %%mm1, %%mm2                     \n\t"
00100         "paddb %%mm7, %%mm2                     \n\t"
00101         "pcmpgtb %%mm6, %%mm2                   \n\t"
00102         "paddb %%mm2, %%mm0                     \n\t"
00103 
00104         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
00105 
00106         "movq (%2, %3, 4), %%mm2                \n\t"
00107         PMAXUB(%%mm2, %%mm4)
00108         PMINUB(%%mm2, %%mm3, %%mm5)
00109         "psubb %%mm2, %%mm1                     \n\t"
00110         "paddb %%mm7, %%mm1                     \n\t"
00111         "pcmpgtb %%mm6, %%mm1                   \n\t"
00112         "paddb %%mm1, %%mm0                     \n\t"
00113 
00114         "movq (%%"REG_a"), %%mm1                \n\t"
00115         PMAXUB(%%mm1, %%mm4)
00116         PMINUB(%%mm1, %%mm3, %%mm5)
00117         "psubb %%mm1, %%mm2                     \n\t"
00118         "paddb %%mm7, %%mm2                     \n\t"
00119         "pcmpgtb %%mm6, %%mm2                   \n\t"
00120         "paddb %%mm2, %%mm0                     \n\t"
00121 
00122         "movq (%%"REG_a", %3), %%mm2            \n\t"
00123         PMAXUB(%%mm2, %%mm4)
00124         PMINUB(%%mm2, %%mm3, %%mm5)
00125         "psubb %%mm2, %%mm1                     \n\t"
00126         "paddb %%mm7, %%mm1                     \n\t"
00127         "pcmpgtb %%mm6, %%mm1                   \n\t"
00128         "paddb %%mm1, %%mm0                     \n\t"
00129 
00130         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
00131         PMAXUB(%%mm1, %%mm4)
00132         PMINUB(%%mm1, %%mm3, %%mm5)
00133         "psubb %%mm1, %%mm2                     \n\t"
00134         "paddb %%mm7, %%mm2                     \n\t"
00135         "pcmpgtb %%mm6, %%mm2                   \n\t"
00136         "paddb %%mm2, %%mm0                     \n\t"
00137         "psubusb %%mm3, %%mm4                   \n\t"
00138 
00139         "                                       \n\t"
00140 #if HAVE_MMX2
00141         "pxor %%mm7, %%mm7                      \n\t"
00142         "psadbw %%mm7, %%mm0                    \n\t"
00143 #else
00144         "movq %%mm0, %%mm1                      \n\t"
00145         "psrlw $8, %%mm0                        \n\t"
00146         "paddb %%mm1, %%mm0                     \n\t"
00147         "movq %%mm0, %%mm1                      \n\t"
00148         "psrlq $16, %%mm0                       \n\t"
00149         "paddb %%mm1, %%mm0                     \n\t"
00150         "movq %%mm0, %%mm1                      \n\t"
00151         "psrlq $32, %%mm0                       \n\t"
00152         "paddb %%mm1, %%mm0                     \n\t"
00153 #endif
00154         "movq %4, %%mm7                         \n\t" // QP,..., QP
00155         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
00156         "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
00157         "packssdw %%mm4, %%mm4                  \n\t"
00158         "movd %%mm0, %0                         \n\t"
00159         "movd %%mm4, %1                         \n\t"
00160 
00161         : "=r" (numEq), "=r" (dcOk)
00162         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00163         : "%"REG_a
00164         );
00165 
00166     numEq= (-numEq) &0xFF;
00167     if(numEq > c->ppMode.flatnessThreshold){
00168         if(dcOk) return 0;
00169         else     return 1;
00170     }else{
00171         return 2;
00172     }
00173 }
00174 #endif //HAVE_MMX
00175 
00180 #if !HAVE_ALTIVEC
00181 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
00182 {
00183 #if HAVE_MMX2 || HAVE_AMD3DNOW
00184     src+= stride*3;
00185     __asm__ volatile(        //"movv %0 %1 %2\n\t"
00186         "movq %2, %%mm0                         \n\t"  // QP,..., QP
00187         "pxor %%mm4, %%mm4                      \n\t"
00188 
00189         "movq (%0), %%mm6                       \n\t"
00190         "movq (%0, %1), %%mm5                   \n\t"
00191         "movq %%mm5, %%mm1                      \n\t"
00192         "movq %%mm6, %%mm2                      \n\t"
00193         "psubusb %%mm6, %%mm5                   \n\t"
00194         "psubusb %%mm1, %%mm2                   \n\t"
00195         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
00196         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
00197         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
00198 
00199         "pand %%mm2, %%mm6                      \n\t"
00200         "pandn %%mm1, %%mm2                     \n\t"
00201         "por %%mm2, %%mm6                       \n\t"// First Line to Filter
00202 
00203         "movq (%0, %1, 8), %%mm5                \n\t"
00204         "lea (%0, %1, 4), %%"REG_a"             \n\t"
00205         "lea (%0, %1, 8), %%"REG_c"             \n\t"
00206         "sub %1, %%"REG_c"                      \n\t"
00207         "add %1, %0                             \n\t" // %0 points to line 1 not 0
00208         "movq (%0, %1, 8), %%mm7                \n\t"
00209         "movq %%mm5, %%mm1                      \n\t"
00210         "movq %%mm7, %%mm2                      \n\t"
00211         "psubusb %%mm7, %%mm5                   \n\t"
00212         "psubusb %%mm1, %%mm2                   \n\t"
00213         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
00214         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
00215         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
00216 
00217         "pand %%mm2, %%mm7                      \n\t"
00218         "pandn %%mm1, %%mm2                     \n\t"
00219         "por %%mm2, %%mm7                       \n\t" // First Line to Filter
00220 
00221 
00222         //      1       2       3       4       5       6       7       8
00223         //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
00224         // 6 4 2 2 1 1
00225         // 6 4 4 2
00226         // 6 8 2
00227 
00228         "movq (%0, %1), %%mm0                   \n\t" //  1
00229         "movq %%mm0, %%mm1                      \n\t" //  1
00230         PAVGB(%%mm6, %%mm0)                           //1 1        /2
00231         PAVGB(%%mm6, %%mm0)                           //3 1        /4
00232 
00233         "movq (%0, %1, 4), %%mm2                \n\t" //     1
00234         "movq %%mm2, %%mm5                      \n\t" //     1
00235         PAVGB((%%REGa), %%mm2)                        //    11        /2
00236         PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
00237         "movq %%mm2, %%mm3                      \n\t" //   211        /4
00238         "movq (%0), %%mm4                       \n\t" // 1
00239         PAVGB(%%mm4, %%mm3)                           // 4 211        /8
00240         PAVGB(%%mm0, %%mm3)                           //642211        /16
00241         "movq %%mm3, (%0)                       \n\t" // X
00242         // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
00243         "movq %%mm1, %%mm0                      \n\t" //  1
00244         PAVGB(%%mm6, %%mm0)                           //1 1        /2
00245         "movq %%mm4, %%mm3                      \n\t" // 1
00246         PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
00247         PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
00248         PAVGB((%%REGa), %%mm5)                        //    211 /4
00249         PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
00250         PAVGB(%%mm0, %%mm3)                           //4242211 /16
00251         "movq %%mm3, (%0,%1)                    \n\t" //  X
00252         // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
00253         PAVGB(%%mm4, %%mm6)                                   //11        /2
00254         "movq (%%"REG_c"), %%mm0                \n\t" //       1
00255         PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
00256         "movq %%mm0, %%mm3                      \n\t" //      11/2
00257         PAVGB(%%mm1, %%mm0)                           //  2   11/4
00258         PAVGB(%%mm6, %%mm0)                           //222   11/8
00259         PAVGB(%%mm2, %%mm0)                           //22242211/16
00260         "movq (%0, %1, 2), %%mm2                \n\t" //   1
00261         "movq %%mm0, (%0, %1, 2)                \n\t" //   X
00262         // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
00263         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
00264         PAVGB((%%REGc), %%mm0)                        //       11        /2
00265         PAVGB(%%mm0, %%mm6)                           //11     11        /4
00266         PAVGB(%%mm1, %%mm4)                           // 11                /2
00267         PAVGB(%%mm2, %%mm1)                           //  11                /2
00268         PAVGB(%%mm1, %%mm6)                           //1122   11        /8
00269         PAVGB(%%mm5, %%mm6)                           //112242211        /16
00270         "movq (%%"REG_a"), %%mm5                \n\t" //    1
00271         "movq %%mm6, (%%"REG_a")                \n\t" //    X
00272         // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
00273         "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
00274         PAVGB(%%mm7, %%mm6)                           //        11        /2
00275         PAVGB(%%mm4, %%mm6)                           // 11     11        /4
00276         PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
00277         PAVGB(%%mm5, %%mm2)                           //   11                /2
00278         "movq (%0, %1, 4), %%mm4                \n\t" //     1
00279         PAVGB(%%mm4, %%mm2)                           //   112                /4
00280         PAVGB(%%mm2, %%mm6)                           // 112242211        /16
00281         "movq %%mm6, (%0, %1, 4)                \n\t" //     X
00282         // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
00283         PAVGB(%%mm7, %%mm1)                           //  11     2        /4
00284         PAVGB(%%mm4, %%mm5)                           //    11                /2
00285         PAVGB(%%mm5, %%mm0)                           //    11 11        /4
00286         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
00287         PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
00288         PAVGB(%%mm0, %%mm1)                           //  11224222        /16
00289         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
00290         // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
00291         PAVGB((%%REGc), %%mm2)                        //   112 4        /8
00292         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
00293         PAVGB(%%mm0, %%mm6)                           //      1 1        /2
00294         PAVGB(%%mm7, %%mm6)                           //      1 12        /4
00295         PAVGB(%%mm2, %%mm6)                           //   1122424        /4
00296         "movq %%mm6, (%%"REG_c")                \n\t" //       X
00297         // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
00298         PAVGB(%%mm7, %%mm5)                           //    11   2        /4
00299         PAVGB(%%mm7, %%mm5)                           //    11   6        /8
00300 
00301         PAVGB(%%mm3, %%mm0)                           //      112        /4
00302         PAVGB(%%mm0, %%mm5)                           //    112246        /16
00303         "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
00304         "sub %1, %0                             \n\t"
00305 
00306         :
00307         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00308         : "%"REG_a, "%"REG_c
00309     );
00310 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00311     const int l1= stride;
00312     const int l2= stride + l1;
00313     const int l3= stride + l2;
00314     const int l4= stride + l3;
00315     const int l5= stride + l4;
00316     const int l6= stride + l5;
00317     const int l7= stride + l6;
00318     const int l8= stride + l7;
00319     const int l9= stride + l8;
00320     int x;
00321     src+= stride*3;
00322     for(x=0; x<BLOCK_SIZE; x++){
00323         const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
00324         const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
00325 
00326         int sums[10];
00327         sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
00328         sums[1] = sums[0] - first  + src[l4];
00329         sums[2] = sums[1] - first  + src[l5];
00330         sums[3] = sums[2] - first  + src[l6];
00331         sums[4] = sums[3] - first  + src[l7];
00332         sums[5] = sums[4] - src[l1] + src[l8];
00333         sums[6] = sums[5] - src[l2] + last;
00334         sums[7] = sums[6] - src[l3] + last;
00335         sums[8] = sums[7] - src[l4] + last;
00336         sums[9] = sums[8] - src[l5] + last;
00337 
00338         src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
00339         src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
00340         src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
00341         src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
00342         src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
00343         src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
00344         src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
00345         src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
00346 
00347         src++;
00348     }
00349 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
00350 }
00351 #endif //HAVE_ALTIVEC
00352 
00360 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
00361 {
00362 #if HAVE_MMX2 || HAVE_AMD3DNOW
00363     src+= stride*3;
00364 
00365     __asm__ volatile(
00366         "pxor %%mm7, %%mm7                      \n\t" // 0
00367         "lea (%0, %1), %%"REG_a"                \n\t"
00368         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00369 //      0       1       2       3       4       5       6       7       8       9
00370 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
00371         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
00372         "movq (%0, %1, 4), %%mm1                \n\t" // line 4
00373         "movq %%mm1, %%mm2                      \n\t" // line 4
00374         "psubusb %%mm0, %%mm1                   \n\t"
00375         "psubusb %%mm2, %%mm0                   \n\t"
00376         "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
00377         "movq (%%"REG_c"), %%mm3                \n\t" // line 5
00378         "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
00379         "movq %%mm3, %%mm5                      \n\t" // line 5
00380         "psubusb %%mm4, %%mm3                   \n\t"
00381         "psubusb %%mm5, %%mm4                   \n\t"
00382         "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
00383         PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
00384         "movq %%mm2, %%mm1                      \n\t" // line 4
00385         "psubusb %%mm5, %%mm2                   \n\t"
00386         "movq %%mm2, %%mm4                      \n\t"
00387         "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
00388         "psubusb %%mm1, %%mm5                   \n\t"
00389         "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
00390         "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
00391         "movq %%mm4, %%mm3                      \n\t" // d
00392         "movq %2, %%mm0                         \n\t"
00393         "paddusb %%mm0, %%mm0                   \n\t"
00394         "psubusb %%mm0, %%mm4                   \n\t"
00395         "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
00396         "psubusb "MANGLE(b01)", %%mm3           \n\t"
00397         "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
00398 
00399         PAVGB(%%mm7, %%mm3)                           // d/2
00400         "movq %%mm3, %%mm1                      \n\t" // d/2
00401         PAVGB(%%mm7, %%mm3)                           // d/4
00402         PAVGB(%%mm1, %%mm3)                           // 3*d/8
00403 
00404         "movq (%0, %1, 4), %%mm0                \n\t" // line 4
00405         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
00406         "psubusb %%mm3, %%mm0                   \n\t"
00407         "pxor %%mm2, %%mm0                      \n\t"
00408         "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
00409 
00410         "movq (%%"REG_c"), %%mm0                \n\t" // line 5
00411         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
00412         "paddusb %%mm3, %%mm0                   \n\t"
00413         "pxor %%mm2, %%mm0                      \n\t"
00414         "movq %%mm0, (%%"REG_c")                \n\t" // line 5
00415 
00416         PAVGB(%%mm7, %%mm1)                           // d/4
00417 
00418         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
00419         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
00420         "psubusb %%mm1, %%mm0                   \n\t"
00421         "pxor %%mm2, %%mm0                      \n\t"
00422         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
00423 
00424         "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
00425         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
00426         "paddusb %%mm1, %%mm0                   \n\t"
00427         "pxor %%mm2, %%mm0                      \n\t"
00428         "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
00429 
00430         PAVGB(%%mm7, %%mm1)                           // d/8
00431 
00432         "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
00433         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
00434         "psubusb %%mm1, %%mm0                   \n\t"
00435         "pxor %%mm2, %%mm0                      \n\t"
00436         "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
00437 
00438         "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
00439         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
00440         "paddusb %%mm1, %%mm0                   \n\t"
00441         "pxor %%mm2, %%mm0                      \n\t"
00442         "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
00443 
00444         :
00445         : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
00446         : "%"REG_a, "%"REG_c
00447     );
00448 #else //HAVE_MMX2 || HAVE_AMD3DNOW
00449 
00450     const int l1= stride;
00451     const int l2= stride + l1;
00452     const int l3= stride + l2;
00453     const int l4= stride + l3;
00454     const int l5= stride + l4;
00455     const int l6= stride + l5;
00456     const int l7= stride + l6;
00457 //    const int l8= stride + l7;
00458 //    const int l9= stride + l8;
00459     int x;
00460 
00461     src+= stride*3;
00462     for(x=0; x<BLOCK_SIZE; x++){
00463         int a= src[l3] - src[l4];
00464         int b= src[l4] - src[l5];
00465         int c= src[l5] - src[l6];
00466 
00467         int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
00468         d= FFMAX(d, 0);
00469 
00470         if(d < co->QP*2){
00471             int v = d * FFSIGN(-b);
00472 
00473             src[l2] +=v>>3;
00474             src[l3] +=v>>2;
00475             src[l4] +=(3*v)>>3;
00476             src[l5] -=(3*v)>>3;
00477             src[l6] -=v>>2;
00478             src[l7] -=v>>3;
00479         }
00480         src++;
00481     }
00482 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
00483 }
00484 
00485 #if !HAVE_ALTIVEC
00486 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
00487 {
00488 #if HAVE_MMX2 || HAVE_AMD3DNOW
00489 /*
00490     uint8_t tmp[16];
00491     const int l1= stride;
00492     const int l2= stride + l1;
00493     const int l3= stride + l2;
00494     const int l4= (int)tmp - (int)src - stride*3;
00495     const int l5= (int)tmp - (int)src - stride*3 + 8;
00496     const int l6= stride*3 + l3;
00497     const int l7= stride + l6;
00498     const int l8= stride + l7;
00499 
00500     memcpy(tmp, src+stride*7, 8);
00501     memcpy(tmp+8, src+stride*8, 8);
00502 */
00503     src+= stride*4;
00504     __asm__ volatile(
00505 
00506 #if 0 //slightly more accurate and slightly slower
00507         "pxor %%mm7, %%mm7                      \n\t" // 0
00508         "lea (%0, %1), %%"REG_a"                \n\t"
00509         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00510 //      0       1       2       3       4       5       6       7
00511 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
00512 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
00513 
00514 
00515         "movq (%0, %1, 2), %%mm0                \n\t" // l2
00516         "movq (%0), %%mm1                       \n\t" // l0
00517         "movq %%mm0, %%mm2                      \n\t" // l2
00518         PAVGB(%%mm7, %%mm0)                           // ~l2/2
00519         PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
00520         PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
00521 
00522         "movq (%%"REG_a"), %%mm1                \n\t" // l1
00523         "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
00524         "movq %%mm1, %%mm4                      \n\t" // l1
00525         PAVGB(%%mm7, %%mm1)                           // ~l1/2
00526         PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
00527         PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
00528 
00529         "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
00530         "psubusb %%mm1, %%mm0                   \n\t"
00531         "psubusb %%mm4, %%mm1                   \n\t"
00532         "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
00533 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
00534 
00535         "movq (%0, %1, 4), %%mm0                \n\t" // l4
00536         "movq %%mm0, %%mm4                      \n\t" // l4
00537         PAVGB(%%mm7, %%mm0)                           // ~l4/2
00538         PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
00539         PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
00540 
00541         "movq (%%"REG_c"), %%mm2                \n\t" // l5
00542         "movq %%mm3, %%mm5                      \n\t" // l3
00543         PAVGB(%%mm7, %%mm3)                           // ~l3/2
00544         PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
00545         PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
00546 
00547         "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
00548         "psubusb %%mm3, %%mm0                   \n\t"
00549         "psubusb %%mm6, %%mm3                   \n\t"
00550         "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
00551         "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
00552 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
00553 
00554         "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
00555         "movq %%mm6, %%mm5                      \n\t" // l6
00556         PAVGB(%%mm7, %%mm6)                           // ~l6/2
00557         PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
00558         PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
00559 
00560         "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
00561         "movq %%mm2, %%mm4                      \n\t" // l5
00562         PAVGB(%%mm7, %%mm2)                           // ~l5/2
00563         PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
00564         PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
00565 
00566         "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
00567         "psubusb %%mm2, %%mm6                   \n\t"
00568         "psubusb %%mm4, %%mm2                   \n\t"
00569         "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
00570 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
00571 
00572 
00573         PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
00574         "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
00575         "paddusb "MANGLE(b01)", %%mm4           \n\t"
00576         "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
00577         "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
00578         "pand %%mm4, %%mm3                      \n\t"
00579 
00580         "movq %%mm3, %%mm1                      \n\t"
00581 //        "psubusb "MANGLE(b01)", %%mm3           \n\t"
00582         PAVGB(%%mm7, %%mm3)
00583         PAVGB(%%mm7, %%mm3)
00584         "paddusb %%mm1, %%mm3                   \n\t"
00585 //        "paddusb "MANGLE(b01)", %%mm3           \n\t"
00586 
00587         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
00588         "movq (%0, %1, 4), %%mm5                \n\t" //l4
00589         "movq (%0, %1, 4), %%mm4                \n\t" //l4
00590         "psubusb %%mm6, %%mm5                   \n\t"
00591         "psubusb %%mm4, %%mm6                   \n\t"
00592         "por %%mm6, %%mm5                       \n\t" // |l3-l4|
00593         "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
00594         "pxor %%mm6, %%mm0                      \n\t"
00595         "pand %%mm0, %%mm3                      \n\t"
00596         PMINUB(%%mm5, %%mm3, %%mm0)
00597 
00598         "psubusb "MANGLE(b01)", %%mm3           \n\t"
00599         PAVGB(%%mm7, %%mm3)
00600 
00601         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00602         "movq (%0, %1, 4), %%mm2                \n\t"
00603         "pxor %%mm6, %%mm0                      \n\t"
00604         "pxor %%mm6, %%mm2                      \n\t"
00605         "psubb %%mm3, %%mm0                     \n\t"
00606         "paddb %%mm3, %%mm2                     \n\t"
00607         "pxor %%mm6, %%mm0                      \n\t"
00608         "pxor %%mm6, %%mm2                      \n\t"
00609         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
00610         "movq %%mm2, (%0, %1, 4)                \n\t"
00611 #endif //0
00612 
00613         "lea (%0, %1), %%"REG_a"                \n\t"
00614         "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
00615 //      0       1       2       3       4       5       6       7
00616 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
00617 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
00618 
00619 
00620         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
00621         "movq (%0, %1, 4), %%mm0                \n\t" // l4
00622         "pxor %%mm6, %%mm1                      \n\t" // -l3-1
00623         PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
00624 // mm1=-l3-1, mm0=128-q
00625 
00626         "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
00627         "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
00628         "pxor %%mm6, %%mm2                      \n\t" // -l5-1
00629         "movq %%mm2, %%mm5                      \n\t" // -l5-1
00630         "movq "MANGLE(b80)", %%mm4              \n\t" // 128
00631         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
00632         PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
00633         PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
00634         PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
00635         PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
00636 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
00637 
00638         "movq (%%"REG_a"), %%mm2                \n\t" // l1
00639         "pxor %%mm6, %%mm2                      \n\t" // -l1-1
00640         PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
00641         PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
00642         "movq "MANGLE(b80)", %%mm3              \n\t" // 128
00643         PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
00644         PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
00645         PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
00646 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
00647 
00648         PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
00649         "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
00650         "pxor %%mm6, %%mm1                      \n\t" // -l7-1
00651         PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
00652         "movq "MANGLE(b80)", %%mm2              \n\t" // 128
00653         PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
00654         PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
00655         PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
00656 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
00657 
00658         "movq "MANGLE(b00)", %%mm1              \n\t" // 0
00659         "movq "MANGLE(b00)", %%mm5              \n\t" // 0
00660         "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
00661         "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
00662         PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
00663         PMAXUB(%%mm5, %%mm3)                          // 128 + |lenergy/16|
00664         PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
00665 
00666 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
00667 
00668         "movq "MANGLE(b00)", %%mm7              \n\t" // 0
00669         "movq %2, %%mm2                         \n\t" // QP
00670         PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
00671         "psubb %%mm6, %%mm2                     \n\t"
00672 
00673         "movq %%mm4, %%mm1                      \n\t"
00674         "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
00675         "pxor %%mm1, %%mm4                      \n\t"
00676         "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
00677         "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
00678         "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
00679 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
00680 
00681         "movq %%mm4, %%mm3                      \n\t" // d
00682         "psubusb "MANGLE(b01)", %%mm4           \n\t"
00683         PAVGB(%%mm7, %%mm4)                           // d/32
00684         PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
00685         "paddb %%mm3, %%mm4                     \n\t" // 5d/64
00686         "pand %%mm2, %%mm4                      \n\t"
00687 
00688         "movq "MANGLE(b80)", %%mm5              \n\t" // 128
00689         "psubb %%mm0, %%mm5                     \n\t" // q
00690         "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
00691         "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
00692         "pxor %%mm7, %%mm5                      \n\t"
00693 
00694         PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
00695         "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
00696 
00697         "pand %%mm7, %%mm4                      \n\t"
00698         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00699         "movq (%0, %1, 4), %%mm2                \n\t"
00700         "pxor %%mm1, %%mm0                      \n\t"
00701         "pxor %%mm1, %%mm2                      \n\t"
00702         "paddb %%mm4, %%mm0                     \n\t"
00703         "psubb %%mm4, %%mm2                     \n\t"
00704         "pxor %%mm1, %%mm0                      \n\t"
00705         "pxor %%mm1, %%mm2                      \n\t"
00706         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
00707         "movq %%mm2, (%0, %1, 4)                \n\t"
00708 
00709         :
00710         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
00711         : "%"REG_a, "%"REG_c
00712     );
00713 
00714 /*
00715     {
00716     int x;
00717     src-= stride;
00718     for(x=0; x<BLOCK_SIZE; x++){
00719         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
00720         if(FFABS(middleEnergy)< 8*QP){
00721             const int q=(src[l4] - src[l5])/2;
00722             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
00723             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
00724 
00725             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
00726             d= FFMAX(d, 0);
00727 
00728             d= (5*d + 32) >> 6;
00729             d*= FFSIGN(-middleEnergy);
00730 
00731             if(q>0){
00732                 d= d<0 ? 0 : d;
00733                 d= d>q ? q : d;
00734             }else{
00735                 d= d>0 ? 0 : d;
00736                 d= d<q ? q : d;
00737             }
00738 
00739             src[l4]-= d;
00740             src[l5]+= d;
00741         }
00742         src++;
00743     }
00744     src-=8;
00745     for(x=0; x<8; x++){
00746         int y;
00747         for(y=4; y<6; y++){
00748             int d= src[x+y*stride] - tmp[x+(y-4)*8];
00749             int ad= FFABS(d);
00750             static int max=0;
00751             static int sum=0;
00752             static int num=0;
00753             static int bias=0;
00754 
00755             if(max<ad) max=ad;
00756             sum+= ad>3 ? 1 : 0;
00757             if(ad>3){
00758                 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
00759             }
00760             if(y==4) bias+=d;
00761             num++;
00762             if(num%1000000 == 0){
00763                 av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
00764             }
00765         }
00766     }
00767 }
00768 */
00769 #elif HAVE_MMX
00770     src+= stride*4;
00771     __asm__ volatile(
00772         "pxor %%mm7, %%mm7                      \n\t"
00773         "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
00774         "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
00775 //      0       1       2       3       4       5       6       7
00776 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
00777 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1
00778 
00779         "movq (%0), %%mm0                       \n\t"
00780         "movq %%mm0, %%mm1                      \n\t"
00781         "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
00782         "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
00783 
00784         "movq (%0, %1), %%mm2                   \n\t"
00785         "lea (%0, %1, 2), %%"REG_a"             \n\t"
00786         "movq %%mm2, %%mm3                      \n\t"
00787         "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
00788         "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
00789 
00790         "movq (%%"REG_a"), %%mm4                \n\t"
00791         "movq %%mm4, %%mm5                      \n\t"
00792         "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
00793         "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
00794 
00795         "paddw %%mm0, %%mm0                     \n\t" // 2L0
00796         "paddw %%mm1, %%mm1                     \n\t" // 2H0
00797         "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
00798         "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
00799         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
00800         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
00801 
00802         "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
00803         "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
00804         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
00805         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
00806 
00807         "movq (%%"REG_a", %1), %%mm2            \n\t"
00808         "movq %%mm2, %%mm3                      \n\t"
00809         "punpcklbw %%mm7, %%mm2                 \n\t" // L3
00810         "punpckhbw %%mm7, %%mm3                 \n\t" // H3
00811 
00812         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
00813         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
00814         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00815         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00816         "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00817         "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00818 
00819         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
00820         "movq %%mm0, %%mm1                      \n\t"
00821         "punpcklbw %%mm7, %%mm0                 \n\t" // L4
00822         "punpckhbw %%mm7, %%mm1                 \n\t" // H4
00823 
00824         "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
00825         "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
00826         "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
00827         "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
00828         "paddw %%mm4, %%mm4                     \n\t" // 2L2
00829         "paddw %%mm5, %%mm5                     \n\t" // 2H2
00830         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
00831         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
00832 
00833         "lea (%%"REG_a", %1), %0                \n\t"
00834         "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
00835         "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
00836         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
00837         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
00838 //50 opcodes so far
00839         "movq (%0, %1, 2), %%mm2                \n\t"
00840         "movq %%mm2, %%mm3                      \n\t"
00841         "punpcklbw %%mm7, %%mm2                 \n\t" // L5
00842         "punpckhbw %%mm7, %%mm3                 \n\t" // H5
00843         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
00844         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
00845         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
00846         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
00847 
00848         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
00849         "punpcklbw %%mm7, %%mm6                 \n\t" // L6
00850         "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
00851         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
00852         "punpckhbw %%mm7, %%mm6                 \n\t" // H6
00853         "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
00854 
00855         "paddw %%mm0, %%mm0                     \n\t" // 2L4
00856         "paddw %%mm1, %%mm1                     \n\t" // 2H4
00857         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
00858         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
00859 
00860         "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
00861         "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
00862         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
00863         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
00864 
00865         "movq (%0, %1, 4), %%mm2                \n\t"
00866         "movq %%mm2, %%mm3                      \n\t"
00867         "punpcklbw %%mm7, %%mm2                 \n\t" // L7
00868         "punpckhbw %%mm7, %%mm3                 \n\t" // H7
00869 
00870         "paddw %%mm2, %%mm2                     \n\t" // 2L7
00871         "paddw %%mm3, %%mm3                     \n\t" // 2H7
00872         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
00873         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
00874 
00875         "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
00876         "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
00877 
00878 #if HAVE_MMX2
00879         "movq %%mm7, %%mm6                      \n\t" // 0
00880         "psubw %%mm0, %%mm6                     \n\t"
00881         "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
00882         "movq %%mm7, %%mm6                      \n\t" // 0
00883         "psubw %%mm1, %%mm6                     \n\t"
00884         "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
00885         "movq %%mm7, %%mm6                      \n\t" // 0
00886         "psubw %%mm2, %%mm6                     \n\t"
00887         "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
00888         "movq %%mm7, %%mm6                      \n\t" // 0
00889         "psubw %%mm3, %%mm6                     \n\t"
00890         "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
00891 #else
00892         "movq %%mm7, %%mm6                      \n\t" // 0
00893         "pcmpgtw %%mm0, %%mm6                   \n\t"
00894         "pxor %%mm6, %%mm0                      \n\t"
00895         "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
00896         "movq %%mm7, %%mm6                      \n\t" // 0
00897         "pcmpgtw %%mm1, %%mm6                   \n\t"
00898         "pxor %%mm6, %%mm1                      \n\t"
00899         "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
00900         "movq %%mm7, %%mm6                      \n\t" // 0
00901         "pcmpgtw %%mm2, %%mm6                   \n\t"
00902         "pxor %%mm6, %%mm2                      \n\t"
00903         "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
00904         "movq %%mm7, %%mm6                      \n\t" // 0
00905         "pcmpgtw %%mm3, %%mm6                   \n\t"
00906         "pxor %%mm6, %%mm3                      \n\t"
00907         "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
00908 #endif
00909 
00910 #if HAVE_MMX2
00911         "pminsw %%mm2, %%mm0                    \n\t"
00912         "pminsw %%mm3, %%mm1                    \n\t"
00913 #else
00914         "movq %%mm0, %%mm6                      \n\t"
00915         "psubusw %%mm2, %%mm6                   \n\t"
00916         "psubw %%mm6, %%mm0                     \n\t"
00917         "movq %%mm1, %%mm6                      \n\t"
00918         "psubusw %%mm3, %%mm6                   \n\t"
00919         "psubw %%mm6, %%mm1                     \n\t"
00920 #endif
00921 
00922         "movd %2, %%mm2                         \n\t" // QP
00923         "punpcklbw %%mm7, %%mm2                 \n\t"
00924 
00925         "movq %%mm7, %%mm6                      \n\t" // 0
00926         "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
00927         "pxor %%mm6, %%mm4                      \n\t"
00928         "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
00929         "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
00930         "pxor %%mm7, %%mm5                      \n\t"
00931         "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
00932 // 100 opcodes
00933         "psllw $3, %%mm2                        \n\t" // 8QP
00934         "movq %%mm2, %%mm3                      \n\t" // 8QP
00935         "pcmpgtw %%mm4, %%mm2                   \n\t"
00936         "pcmpgtw %%mm5, %%mm3                   \n\t"
00937         "pand %%mm2, %%mm4                      \n\t"
00938         "pand %%mm3, %%mm5                      \n\t"
00939 
00940 
00941         "psubusw %%mm0, %%mm4                   \n\t" // hd
00942         "psubusw %%mm1, %%mm5                   \n\t" // ld
00943 
00944 
00945         "movq "MANGLE(w05)", %%mm2              \n\t" // 5
00946         "pmullw %%mm2, %%mm4                    \n\t"
00947         "pmullw %%mm2, %%mm5                    \n\t"
00948         "movq "MANGLE(w20)", %%mm2              \n\t" // 32
00949         "paddw %%mm2, %%mm4                     \n\t"
00950         "paddw %%mm2, %%mm5                     \n\t"
00951         "psrlw $6, %%mm4                        \n\t"
00952         "psrlw $6, %%mm5                        \n\t"
00953 
00954         "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
00955         "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
00956 
00957         "pxor %%mm2, %%mm2                      \n\t"
00958         "pxor %%mm3, %%mm3                      \n\t"
00959 
00960         "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
00961         "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
00962         "pxor %%mm2, %%mm0                      \n\t"
00963         "pxor %%mm3, %%mm1                      \n\t"
00964         "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
00965         "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
00966         "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
00967         "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
00968 
00969         "pxor %%mm6, %%mm2                      \n\t"
00970         "pxor %%mm7, %%mm3                      \n\t"
00971         "pand %%mm2, %%mm4                      \n\t"
00972         "pand %%mm3, %%mm5                      \n\t"
00973 
00974 #if HAVE_MMX2
00975         "pminsw %%mm0, %%mm4                    \n\t"
00976         "pminsw %%mm1, %%mm5                    \n\t"
00977 #else
00978         "movq %%mm4, %%mm2                      \n\t"
00979         "psubusw %%mm0, %%mm2                   \n\t"
00980         "psubw %%mm2, %%mm4                     \n\t"
00981         "movq %%mm5, %%mm2                      \n\t"
00982         "psubusw %%mm1, %%mm2                   \n\t"
00983         "psubw %%mm2, %%mm5                     \n\t"
00984 #endif
00985         "pxor %%mm6, %%mm4                      \n\t"
00986         "pxor %%mm7, %%mm5                      \n\t"
00987         "psubw %%mm6, %%mm4                     \n\t"
00988         "psubw %%mm7, %%mm5                     \n\t"
00989         "packsswb %%mm5, %%mm4                  \n\t"
00990         "movq (%0), %%mm0                       \n\t"
00991         "paddb   %%mm4, %%mm0                   \n\t"
00992         "movq %%mm0, (%0)                       \n\t"
00993         "movq (%0, %1), %%mm0                   \n\t"
00994         "psubb %%mm4, %%mm0                     \n\t"
00995         "movq %%mm0, (%0, %1)                   \n\t"
00996 
00997         : "+r" (src)
00998         : "r" ((x86_reg)stride), "m" (c->pQPb)
00999         : "%"REG_a, "%"REG_c
01000     );
01001 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01002     const int l1= stride;
01003     const int l2= stride + l1;
01004     const int l3= stride + l2;
01005     const int l4= stride + l3;
01006     const int l5= stride + l4;
01007     const int l6= stride + l5;
01008     const int l7= stride + l6;
01009     const int l8= stride + l7;
01010 //    const int l9= stride + l8;
01011     int x;
01012     src+= stride*3;
01013     for(x=0; x<BLOCK_SIZE; x++){
01014         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
01015         if(FFABS(middleEnergy) < 8*c->QP){
01016             const int q=(src[l4] - src[l5])/2;
01017             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
01018             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
01019 
01020             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
01021             d= FFMAX(d, 0);
01022 
01023             d= (5*d + 32) >> 6;
01024             d*= FFSIGN(-middleEnergy);
01025 
01026             if(q>0){
01027                 d= d<0 ? 0 : d;
01028                 d= d>q ? q : d;
01029             }else{
01030                 d= d>0 ? 0 : d;
01031                 d= d<q ? q : d;
01032             }
01033 
01034             src[l4]-= d;
01035             src[l5]+= d;
01036         }
01037         src++;
01038     }
01039 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01040 }
01041 #endif //HAVE_ALTIVEC
01042 
01043 #if !HAVE_ALTIVEC
01044 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
01045 {
01046 #if HAVE_MMX2 || HAVE_AMD3DNOW
01047     __asm__ volatile(
01048         "pxor %%mm6, %%mm6                      \n\t"
01049         "pcmpeqb %%mm7, %%mm7                   \n\t"
01050         "movq %2, %%mm0                         \n\t"
01051         "punpcklbw %%mm6, %%mm0                 \n\t"
01052         "psrlw $1, %%mm0                        \n\t"
01053         "psubw %%mm7, %%mm0                     \n\t"
01054         "packuswb %%mm0, %%mm0                  \n\t"
01055         "movq %%mm0, %3                         \n\t"
01056 
01057         "lea (%0, %1), %%"REG_a"                \n\t"
01058         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01059 
01060 //        0        1        2        3        4        5        6        7        8        9
01061 //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1
01062 
01063 #undef REAL_FIND_MIN_MAX
01064 #undef FIND_MIN_MAX
01065 #if HAVE_MMX2
01066 #define REAL_FIND_MIN_MAX(addr)\
01067         "movq " #addr ", %%mm0                  \n\t"\
01068         "pminub %%mm0, %%mm7                    \n\t"\
01069         "pmaxub %%mm0, %%mm6                    \n\t"
01070 #else
01071 #define REAL_FIND_MIN_MAX(addr)\
01072         "movq " #addr ", %%mm0                  \n\t"\
01073         "movq %%mm7, %%mm1                      \n\t"\
01074         "psubusb %%mm0, %%mm6                   \n\t"\
01075         "paddb %%mm0, %%mm6                     \n\t"\
01076         "psubusb %%mm0, %%mm1                   \n\t"\
01077         "psubb %%mm1, %%mm7                     \n\t"
01078 #endif
01079 #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
01080 
01081 FIND_MIN_MAX((%%REGa))
01082 FIND_MIN_MAX((%%REGa, %1))
01083 FIND_MIN_MAX((%%REGa, %1, 2))
01084 FIND_MIN_MAX((%0, %1, 4))
01085 FIND_MIN_MAX((%%REGd))
01086 FIND_MIN_MAX((%%REGd, %1))
01087 FIND_MIN_MAX((%%REGd, %1, 2))
01088 FIND_MIN_MAX((%0, %1, 8))
01089 
01090         "movq %%mm7, %%mm4                      \n\t"
01091         "psrlq $8, %%mm7                        \n\t"
01092 #if HAVE_MMX2
01093         "pminub %%mm4, %%mm7                    \n\t" // min of pixels
01094         "pshufw $0xF9, %%mm7, %%mm4             \n\t"
01095         "pminub %%mm4, %%mm7                    \n\t" // min of pixels
01096         "pshufw $0xFE, %%mm7, %%mm4             \n\t"
01097         "pminub %%mm4, %%mm7                    \n\t"
01098 #else
01099         "movq %%mm7, %%mm1                      \n\t"
01100         "psubusb %%mm4, %%mm1                   \n\t"
01101         "psubb %%mm1, %%mm7                     \n\t"
01102         "movq %%mm7, %%mm4                      \n\t"
01103         "psrlq $16, %%mm7                       \n\t"
01104         "movq %%mm7, %%mm1                      \n\t"
01105         "psubusb %%mm4, %%mm1                   \n\t"
01106         "psubb %%mm1, %%mm7                     \n\t"
01107         "movq %%mm7, %%mm4                      \n\t"
01108         "psrlq $32, %%mm7                       \n\t"
01109         "movq %%mm7, %%mm1                      \n\t"
01110         "psubusb %%mm4, %%mm1                   \n\t"
01111         "psubb %%mm1, %%mm7                     \n\t"
01112 #endif
01113 
01114 
01115         "movq %%mm6, %%mm4                      \n\t"
01116         "psrlq $8, %%mm6                        \n\t"
01117 #if HAVE_MMX2
01118         "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
01119         "pshufw $0xF9, %%mm6, %%mm4             \n\t"
01120         "pmaxub %%mm4, %%mm6                    \n\t"
01121         "pshufw $0xFE, %%mm6, %%mm4             \n\t"
01122         "pmaxub %%mm4, %%mm6                    \n\t"
01123 #else
01124         "psubusb %%mm4, %%mm6                   \n\t"
01125         "paddb %%mm4, %%mm6                     \n\t"
01126         "movq %%mm6, %%mm4                      \n\t"
01127         "psrlq $16, %%mm6                       \n\t"
01128         "psubusb %%mm4, %%mm6                   \n\t"
01129         "paddb %%mm4, %%mm6                     \n\t"
01130         "movq %%mm6, %%mm4                      \n\t"
01131         "psrlq $32, %%mm6                       \n\t"
01132         "psubusb %%mm4, %%mm6                   \n\t"
01133         "paddb %%mm4, %%mm6                     \n\t"
01134 #endif
01135         "movq %%mm6, %%mm0                      \n\t" // max
01136         "psubb %%mm7, %%mm6                     \n\t" // max - min
01137         "movd %%mm6, %%ecx                      \n\t"
01138         "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
01139         " jb 1f                                 \n\t"
01140         "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
01141         "and "ALIGN_MASK", %%"REG_c"            \n\t"
01142         PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
01143         "punpcklbw %%mm7, %%mm7                 \n\t"
01144         "punpcklbw %%mm7, %%mm7                 \n\t"
01145         "punpcklbw %%mm7, %%mm7                 \n\t"
01146         "movq %%mm7, (%%"REG_c")                \n\t"
01147 
01148         "movq (%0), %%mm0                       \n\t" // L10
01149         "movq %%mm0, %%mm1                      \n\t" // L10
01150         "movq %%mm0, %%mm2                      \n\t" // L10
01151         "psllq $8, %%mm1                        \n\t"
01152         "psrlq $8, %%mm2                        \n\t"
01153         "movd -4(%0), %%mm3                     \n\t"
01154         "movd 8(%0), %%mm4                      \n\t"
01155         "psrlq $24, %%mm3                       \n\t"
01156         "psllq $56, %%mm4                       \n\t"
01157         "por %%mm3, %%mm1                       \n\t" // L00
01158         "por %%mm4, %%mm2                       \n\t" // L20
01159         "movq %%mm1, %%mm3                      \n\t" // L00
01160         PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
01161         PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
01162         "psubusb %%mm7, %%mm0                   \n\t"
01163         "psubusb %%mm7, %%mm2                   \n\t"
01164         "psubusb %%mm7, %%mm3                   \n\t"
01165         "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
01166         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
01167         "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
01168         "paddb %%mm2, %%mm0                     \n\t"
01169         "paddb %%mm3, %%mm0                     \n\t"
01170 
01171         "movq (%%"REG_a"), %%mm2                \n\t" // L11
01172         "movq %%mm2, %%mm3                      \n\t" // L11
01173         "movq %%mm2, %%mm4                      \n\t" // L11
01174         "psllq $8, %%mm3                        \n\t"
01175         "psrlq $8, %%mm4                        \n\t"
01176         "movd -4(%%"REG_a"), %%mm5              \n\t"
01177         "movd 8(%%"REG_a"), %%mm6               \n\t"
01178         "psrlq $24, %%mm5                       \n\t"
01179         "psllq $56, %%mm6                       \n\t"
01180         "por %%mm5, %%mm3                       \n\t" // L01
01181         "por %%mm6, %%mm4                       \n\t" // L21
01182         "movq %%mm3, %%mm5                      \n\t" // L01
01183         PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
01184         PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
01185         "psubusb %%mm7, %%mm2                   \n\t"
01186         "psubusb %%mm7, %%mm4                   \n\t"
01187         "psubusb %%mm7, %%mm5                   \n\t"
01188         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
01189         "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
01190         "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
01191         "paddb %%mm4, %%mm2                     \n\t"
01192         "paddb %%mm5, %%mm2                     \n\t"
01193 // 0, 2, 3, 1
01194 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
01195         "movq " #src ", " #sx "                 \n\t" /* src[0] */\
01196         "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
01197         "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
01198         "psllq $8, " #lx "                      \n\t"\
01199         "psrlq $8, " #t0 "                      \n\t"\
01200         "movd -4" #src ", " #t1 "               \n\t"\
01201         "psrlq $24, " #t1 "                     \n\t"\
01202         "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
01203         "movd 8" #src ", " #t1 "                \n\t"\
01204         "psllq $56, " #t1 "                     \n\t"\
01205         "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
01206         "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
01207         PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
01208         PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
01209         PAVGB(lx, pplx)                                     \
01210         "movq " #lx ", 8(%%"REG_c")             \n\t"\
01211         "movq (%%"REG_c"), " #lx "              \n\t"\
01212         "psubusb " #lx ", " #t1 "               \n\t"\
01213         "psubusb " #lx ", " #t0 "               \n\t"\
01214         "psubusb " #lx ", " #sx "               \n\t"\
01215         "movq "MANGLE(b00)", " #lx "            \n\t"\
01216         "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
01217         "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
01218         "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
01219         "paddb " #t1 ", " #t0 "                 \n\t"\
01220         "paddb " #t0 ", " #sx "                 \n\t"\
01221 \
01222         PAVGB(plx, pplx)                              /* filtered */\
01223         "movq " #dst ", " #t0 "                 \n\t" /* dst */\
01224         "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
01225         "psubusb %3, " #t0 "                    \n\t"\
01226         "paddusb %3, " #t1 "                    \n\t"\
01227         PMAXUB(t0, pplx)\
01228         PMINUB(t1, pplx, t0)\
01229         "paddb " #sx ", " #ppsx "               \n\t"\
01230         "paddb " #psx ", " #ppsx "              \n\t"\
01231         "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
01232         "pand "MANGLE(b08)", " #ppsx "          \n\t"\
01233         "pcmpeqb " #lx ", " #ppsx "             \n\t"\
01234         "pand " #ppsx ", " #pplx "              \n\t"\
01235         "pandn " #dst ", " #ppsx "              \n\t"\
01236         "por " #pplx ", " #ppsx "               \n\t"\
01237         "movq " #ppsx ", " #dst "               \n\t"\
01238         "movq 8(%%"REG_c"), " #lx "             \n\t"
01239 
01240 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
01241    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
01242 /*
01243 0000000
01244 1111111
01245 
01246 1111110
01247 1111101
01248 1111100
01249 1111011
01250 1111010
01251 1111001
01252 
01253 1111000
01254 1110111
01255 
01256 */
01257 //DERING_CORE(dst          ,src            ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)
01258 DERING_CORE((%%REGa)       ,(%%REGa, %1)   ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01259 DERING_CORE((%%REGa, %1)   ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01260 DERING_CORE((%%REGa, %1, 2),(%0, %1, 4)    ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
01261 DERING_CORE((%0, %1, 4)    ,(%%REGd)       ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01262 DERING_CORE((%%REGd)       ,(%%REGd, %1)   ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01263 DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
01264 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
01265 DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
01266 
01267         "1:                        \n\t"
01268         : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2)
01269         : "%"REG_a, "%"REG_d, "%"REG_c
01270     );
01271 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01272     int y;
01273     int min=255;
01274     int max=0;
01275     int avg;
01276     uint8_t *p;
01277     int s[10];
01278     const int QP2= c->QP/2 + 1;
01279 
01280     for(y=1; y<9; y++){
01281         int x;
01282         p= src + stride*y;
01283         for(x=1; x<9; x++){
01284             p++;
01285             if(*p > max) max= *p;
01286             if(*p < min) min= *p;
01287         }
01288     }
01289     avg= (min + max + 1)>>1;
01290 
01291     if(max - min <deringThreshold) return;
01292 
01293     for(y=0; y<10; y++){
01294         int t = 0;
01295 
01296         if(src[stride*y + 0] > avg) t+= 1;
01297         if(src[stride*y + 1] > avg) t+= 2;
01298         if(src[stride*y + 2] > avg) t+= 4;
01299         if(src[stride*y + 3] > avg) t+= 8;
01300         if(src[stride*y + 4] > avg) t+= 16;
01301         if(src[stride*y + 5] > avg) t+= 32;
01302         if(src[stride*y + 6] > avg) t+= 64;
01303         if(src[stride*y + 7] > avg) t+= 128;
01304         if(src[stride*y + 8] > avg) t+= 256;
01305         if(src[stride*y + 9] > avg) t+= 512;
01306 
01307         t |= (~t)<<16;
01308         t &= (t<<1) & (t>>1);
01309         s[y] = t;
01310     }
01311 
01312     for(y=1; y<9; y++){
01313         int t = s[y-1] & s[y] & s[y+1];
01314         t|= t>>16;
01315         s[y-1]= t;
01316     }
01317 
01318     for(y=1; y<9; y++){
01319         int x;
01320         int t = s[y-1];
01321 
01322         p= src + stride*y;
01323         for(x=1; x<9; x++){
01324             p++;
01325             if(t & (1<<x)){
01326                 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
01327                       +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
01328                       +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
01329                 f= (f + 8)>>4;
01330 
01331 #ifdef DEBUG_DERING_THRESHOLD
01332                     __asm__ volatile("emms\n\t":);
01333                     {
01334                     static long long numPixels=0;
01335                     if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
01336 //                    if((max-min)<20 || (max-min)*QP<200)
01337 //                    if((max-min)*QP < 500)
01338 //                    if(max-min<QP/2)
01339                     if(max-min < 20){
01340                         static int numSkipped=0;
01341                         static int errorSum=0;
01342                         static int worstQP=0;
01343                         static int worstRange=0;
01344                         static int worstDiff=0;
01345                         int diff= (f - *p);
01346                         int absDiff= FFABS(diff);
01347                         int error= diff*diff;
01348 
01349                         if(x==1 || x==8 || y==1 || y==8) continue;
01350 
01351                         numSkipped++;
01352                         if(absDiff > worstDiff){
01353                             worstDiff= absDiff;
01354                             worstQP= QP;
01355                             worstRange= max-min;
01356                         }
01357                         errorSum+= error;
01358 
01359                         if(1024LL*1024LL*1024LL % numSkipped == 0){
01360                             av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
01361                                    "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
01362                                    (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
01363                                    worstDiff, (float)numSkipped/numPixels);
01364                         }
01365                     }
01366                     }
01367 #endif
01368                     if     (*p + QP2 < f) *p= *p + QP2;
01369                     else if(*p - QP2 > f) *p= *p - QP2;
01370                     else *p=f;
01371             }
01372         }
01373     }
01374 #ifdef DEBUG_DERING_THRESHOLD
01375     if(max-min < 20){
01376         for(y=1; y<9; y++){
01377             int x;
01378             int t = 0;
01379             p= src + stride*y;
01380             for(x=1; x<9; x++){
01381                 p++;
01382                 *p = FFMIN(*p + 20, 255);
01383             }
01384         }
01385 //        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
01386     }
01387 #endif
01388 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01389 }
01390 #endif //HAVE_ALTIVEC
01391 
01398 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
01399 {
01400 #if HAVE_MMX2 || HAVE_AMD3DNOW
01401     src+= 4*stride;
01402     __asm__ volatile(
01403         "lea (%0, %1), %%"REG_a"                \n\t"
01404         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
01405 //      0       1       2       3       4       5       6       7       8       9
01406 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
01407 
01408         "movq (%0), %%mm0                       \n\t"
01409         "movq (%%"REG_a", %1), %%mm1            \n\t"
01410         PAVGB(%%mm1, %%mm0)
01411         "movq %%mm0, (%%"REG_a")                \n\t"
01412         "movq (%0, %1, 4), %%mm0                \n\t"
01413         PAVGB(%%mm0, %%mm1)
01414         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
01415         "movq (%%"REG_c", %1), %%mm1            \n\t"
01416         PAVGB(%%mm1, %%mm0)
01417         "movq %%mm0, (%%"REG_c")                \n\t"
01418         "movq (%0, %1, 8), %%mm0                \n\t"
01419         PAVGB(%%mm0, %%mm1)
01420         "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
01421 
01422         : : "r" (src), "r" ((x86_reg)stride)
01423         : "%"REG_a, "%"REG_c
01424     );
01425 #else
01426     int a, b, x;
01427     src+= 4*stride;
01428 
01429     for(x=0; x<2; x++){
01430         a= *(uint32_t*)&src[stride*0];
01431         b= *(uint32_t*)&src[stride*2];
01432         *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01433         a= *(uint32_t*)&src[stride*4];
01434         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01435         b= *(uint32_t*)&src[stride*6];
01436         *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01437         a= *(uint32_t*)&src[stride*8];
01438         *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01439         src += 4;
01440     }
01441 #endif
01442 }
01443 
01451 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
01452 {
01453 #if HAVE_MMX2 || HAVE_AMD3DNOW
01454     src+= stride*3;
01455     __asm__ volatile(
01456         "lea (%0, %1), %%"REG_a"                \n\t"
01457         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01458         "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
01459         "add %1, %%"REG_c"                      \n\t"
01460         "pxor %%mm7, %%mm7                      \n\t"
01461 //      0       1       2       3       4       5       6       7       8       9       10
01462 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01463 
01464 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
01465         "movq " #a ", %%mm0                     \n\t"\
01466         "movq " #b ", %%mm1                     \n\t"\
01467         "movq " #d ", %%mm2                     \n\t"\
01468         "movq " #e ", %%mm3                     \n\t"\
01469         PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
01470         PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
01471         "movq %%mm0, %%mm2                      \n\t"\
01472         "punpcklbw %%mm7, %%mm0                 \n\t"\
01473         "punpckhbw %%mm7, %%mm2                 \n\t"\
01474         "movq %%mm1, %%mm3                      \n\t"\
01475         "punpcklbw %%mm7, %%mm1                 \n\t"\
01476         "punpckhbw %%mm7, %%mm3                 \n\t"\
01477         "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
01478         "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
01479         "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
01480         "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
01481         "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
01482         "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
01483         "packuswb %%mm3, %%mm1                  \n\t"\
01484         "movq %%mm1, " #c "                     \n\t"
01485 #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)
01486 
01487 DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
01488 DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))
01489 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
01490 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
01491 
01492         : : "r" (src), "r" ((x86_reg)stride)
01493         : "%"REG_a, "%"REG_d, "%"REG_c
01494     );
01495 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01496     int x;
01497     src+= stride*3;
01498     for(x=0; x<8; x++){
01499         src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
01500         src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
01501         src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
01502         src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
01503         src++;
01504     }
01505 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01506 }
01507 
01515 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
01516 {
01517 #if HAVE_MMX2 || HAVE_AMD3DNOW
01518     src+= stride*4;
01519     __asm__ volatile(
01520         "lea (%0, %1), %%"REG_a"                \n\t"
01521         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01522         "pxor %%mm7, %%mm7                      \n\t"
01523         "movq (%2), %%mm0                       \n\t"
01524 //      0       1       2       3       4       5       6       7       8       9       10
01525 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01526 
01527 #define REAL_DEINT_FF(a,b,c,d)\
01528         "movq " #a ", %%mm1                     \n\t"\
01529         "movq " #b ", %%mm2                     \n\t"\
01530         "movq " #c ", %%mm3                     \n\t"\
01531         "movq " #d ", %%mm4                     \n\t"\
01532         PAVGB(%%mm3, %%mm1)                          \
01533         PAVGB(%%mm4, %%mm0)                          \
01534         "movq %%mm0, %%mm3                      \n\t"\
01535         "punpcklbw %%mm7, %%mm0                 \n\t"\
01536         "punpckhbw %%mm7, %%mm3                 \n\t"\
01537         "movq %%mm1, %%mm4                      \n\t"\
01538         "punpcklbw %%mm7, %%mm1                 \n\t"\
01539         "punpckhbw %%mm7, %%mm4                 \n\t"\
01540         "psllw $2, %%mm1                        \n\t"\
01541         "psllw $2, %%mm4                        \n\t"\
01542         "psubw %%mm0, %%mm1                     \n\t"\
01543         "psubw %%mm3, %%mm4                     \n\t"\
01544         "movq %%mm2, %%mm5                      \n\t"\
01545         "movq %%mm2, %%mm0                      \n\t"\
01546         "punpcklbw %%mm7, %%mm2                 \n\t"\
01547         "punpckhbw %%mm7, %%mm5                 \n\t"\
01548         "paddw %%mm2, %%mm1                     \n\t"\
01549         "paddw %%mm5, %%mm4                     \n\t"\
01550         "psraw $2, %%mm1                        \n\t"\
01551         "psraw $2, %%mm4                        \n\t"\
01552         "packuswb %%mm4, %%mm1                  \n\t"\
01553         "movq %%mm1, " #b "                     \n\t"\
01554 
01555 #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)
01556 
01557 DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))
01558 DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
01559 DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
01560 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
01561 
01562         "movq %%mm0, (%2)                       \n\t"
01563         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)
01564         : "%"REG_a, "%"REG_d
01565     );
01566 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01567     int x;
01568     src+= stride*4;
01569     for(x=0; x<8; x++){
01570         int t1= tmp[x];
01571         int t2= src[stride*1];
01572 
01573         src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
01574         t1= src[stride*4];
01575         src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
01576         t2= src[stride*6];
01577         src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
01578         t1= src[stride*8];
01579         src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
01580         tmp[x]= t1;
01581 
01582         src++;
01583     }
01584 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01585 }
01586 
01594 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
01595 {
01596 #if HAVE_MMX2 || HAVE_AMD3DNOW
01597     src+= stride*4;
01598     __asm__ volatile(
01599         "lea (%0, %1), %%"REG_a"                \n\t"
01600         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01601         "pxor %%mm7, %%mm7                      \n\t"
01602         "movq (%2), %%mm0                       \n\t"
01603         "movq (%3), %%mm1                       \n\t"
01604 //      0       1       2       3       4       5       6       7       8       9       10
01605 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
01606 
01607 #define REAL_DEINT_L5(t1,t2,a,b,c)\
01608         "movq " #a ", %%mm2                     \n\t"\
01609         "movq " #b ", %%mm3                     \n\t"\
01610         "movq " #c ", %%mm4                     \n\t"\
01611         PAVGB(t2, %%mm3)                             \
01612         PAVGB(t1, %%mm4)                             \
01613         "movq %%mm2, %%mm5                      \n\t"\
01614         "movq %%mm2, " #t1 "                    \n\t"\
01615         "punpcklbw %%mm7, %%mm2                 \n\t"\
01616         "punpckhbw %%mm7, %%mm5                 \n\t"\
01617         "movq %%mm2, %%mm6                      \n\t"\
01618         "paddw %%mm2, %%mm2                     \n\t"\
01619         "paddw %%mm6, %%mm2                     \n\t"\
01620         "movq %%mm5, %%mm6                      \n\t"\
01621         "paddw %%mm5, %%mm5                     \n\t"\
01622         "paddw %%mm6, %%mm5                     \n\t"\
01623         "movq %%mm3, %%mm6                      \n\t"\
01624         "punpcklbw %%mm7, %%mm3                 \n\t"\
01625         "punpckhbw %%mm7, %%mm6                 \n\t"\
01626         "paddw %%mm3, %%mm3                     \n\t"\
01627         "paddw %%mm6, %%mm6                     \n\t"\
01628         "paddw %%mm3, %%mm2                     \n\t"\
01629         "paddw %%mm6, %%mm5                     \n\t"\
01630         "movq %%mm4, %%mm6                      \n\t"\
01631         "punpcklbw %%mm7, %%mm4                 \n\t"\
01632         "punpckhbw %%mm7, %%mm6                 \n\t"\
01633         "psubw %%mm4, %%mm2                     \n\t"\
01634         "psubw %%mm6, %%mm5                     \n\t"\
01635         "psraw $2, %%mm2                        \n\t"\
01636         "psraw $2, %%mm5                        \n\t"\
01637         "packuswb %%mm5, %%mm2                  \n\t"\
01638         "movq %%mm2, " #a "                     \n\t"\
01639 
01640 #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)
01641 
01642 DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )
01643 DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))
01644 DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )
01645 DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )
01646 DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )
01647 DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
01648 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
01649 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
01650 
01651         "movq %%mm0, (%2)                       \n\t"
01652         "movq %%mm1, (%3)                       \n\t"
01653         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)
01654         : "%"REG_a, "%"REG_d
01655     );
01656 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01657     int x;
01658     src+= stride*4;
01659     for(x=0; x<8; x++){
01660         int t1= tmp[x];
01661         int t2= tmp2[x];
01662         int t3= src[0];
01663 
01664         src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
01665         t1= src[stride*1];
01666         src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
01667         t2= src[stride*2];
01668         src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
01669         t3= src[stride*3];
01670         src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
01671         t1= src[stride*4];
01672         src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
01673         t2= src[stride*5];
01674         src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
01675         t3= src[stride*6];
01676         src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
01677         t1= src[stride*7];
01678         src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
01679 
01680         tmp[x]= t3;
01681         tmp2[x]= t1;
01682 
01683         src++;
01684     }
01685 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01686 }
01687 
01695 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
01696 {
01697 #if HAVE_MMX2 || HAVE_AMD3DNOW
01698     src+= 4*stride;
01699     __asm__ volatile(
01700         "lea (%0, %1), %%"REG_a"                \n\t"
01701         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01702 //      0       1       2       3       4       5       6       7       8       9
01703 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01704 
01705         "movq (%2), %%mm0                       \n\t" // L0
01706         "movq (%%"REG_a"), %%mm1                \n\t" // L2
01707         PAVGB(%%mm1, %%mm0)                           // L0+L2
01708         "movq (%0), %%mm2                       \n\t" // L1
01709         PAVGB(%%mm2, %%mm0)
01710         "movq %%mm0, (%0)                       \n\t"
01711         "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
01712         PAVGB(%%mm0, %%mm2)                           // L1+L3
01713         PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
01714         "movq %%mm2, (%%"REG_a")                \n\t"
01715         "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
01716         PAVGB(%%mm2, %%mm1)                           // L2+L4
01717         PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
01718         "movq %%mm1, (%%"REG_a", %1)            \n\t"
01719         "movq (%0, %1, 4), %%mm1                \n\t" // L5
01720         PAVGB(%%mm1, %%mm0)                           // L3+L5
01721         PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
01722         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
01723         "movq (%%"REG_d"), %%mm0                \n\t" // L6
01724         PAVGB(%%mm0, %%mm2)                           // L4+L6
01725         PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
01726         "movq %%mm2, (%0, %1, 4)                \n\t"
01727         "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
01728         PAVGB(%%mm2, %%mm1)                           // L5+L7
01729         PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
01730         "movq %%mm1, (%%"REG_d")                \n\t"
01731         "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
01732         PAVGB(%%mm1, %%mm0)                           // L6+L8
01733         PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
01734         "movq %%mm0, (%%"REG_d", %1)            \n\t"
01735         "movq (%0, %1, 8), %%mm0                \n\t" // L9
01736         PAVGB(%%mm0, %%mm2)                           // L7+L9
01737         PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
01738         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
01739         "movq %%mm1, (%2)                       \n\t"
01740 
01741         : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
01742         : "%"REG_a, "%"REG_d
01743     );
01744 #else //HAVE_MMX2 || HAVE_AMD3DNOW
01745     int a, b, c, x;
01746     src+= 4*stride;
01747 
01748     for(x=0; x<2; x++){
01749         a= *(uint32_t*)&tmp[stride*0];
01750         b= *(uint32_t*)&src[stride*0];
01751         c= *(uint32_t*)&src[stride*1];
01752         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01753         *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01754 
01755         a= *(uint32_t*)&src[stride*2];
01756         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01757         *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01758 
01759         b= *(uint32_t*)&src[stride*3];
01760         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
01761         *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
01762 
01763         c= *(uint32_t*)&src[stride*4];
01764         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01765         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01766 
01767         a= *(uint32_t*)&src[stride*5];
01768         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01769         *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01770 
01771         b= *(uint32_t*)&src[stride*6];
01772         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
01773         *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
01774 
01775         c= *(uint32_t*)&src[stride*7];
01776         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
01777         *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
01778 
01779         a= *(uint32_t*)&src[stride*8];
01780         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
01781         *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
01782 
01783         *(uint32_t*)&tmp[stride*0]= c;
01784         src += 4;
01785         tmp += 4;
01786     }
01787 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
01788 }
01789 
01796 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
01797 {
01798 #if HAVE_MMX
01799     src+= 4*stride;
01800 #if HAVE_MMX2
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 (%0), %%mm0                       \n\t" //
01808         "movq (%%"REG_a", %1), %%mm2            \n\t" //
01809         "movq (%%"REG_a"), %%mm1                \n\t" //
01810         "movq %%mm0, %%mm3                      \n\t"
01811         "pmaxub %%mm1, %%mm0                    \n\t" //
01812         "pminub %%mm3, %%mm1                    \n\t" //
01813         "pmaxub %%mm2, %%mm1                    \n\t" //
01814         "pminub %%mm1, %%mm0                    \n\t"
01815         "movq %%mm0, (%%"REG_a")                \n\t"
01816 
01817         "movq (%0, %1, 4), %%mm0                \n\t" //
01818         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
01819         "movq %%mm2, %%mm3                      \n\t"
01820         "pmaxub %%mm1, %%mm2                    \n\t" //
01821         "pminub %%mm3, %%mm1                    \n\t" //
01822         "pmaxub %%mm0, %%mm1                    \n\t" //
01823         "pminub %%mm1, %%mm2                    \n\t"
01824         "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
01825 
01826         "movq (%%"REG_d"), %%mm2                \n\t" //
01827         "movq (%%"REG_d", %1), %%mm1            \n\t" //
01828         "movq %%mm2, %%mm3                      \n\t"
01829         "pmaxub %%mm0, %%mm2                    \n\t" //
01830         "pminub %%mm3, %%mm0                    \n\t" //
01831         "pmaxub %%mm1, %%mm0                    \n\t" //
01832         "pminub %%mm0, %%mm2                    \n\t"
01833         "movq %%mm2, (%%"REG_d")                \n\t"
01834 
01835         "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
01836         "movq (%0, %1, 8), %%mm0                \n\t" //
01837         "movq %%mm2, %%mm3                      \n\t"
01838         "pmaxub %%mm0, %%mm2                    \n\t" //
01839         "pminub %%mm3, %%mm0                    \n\t" //
01840         "pmaxub %%mm1, %%mm0                    \n\t" //
01841         "pminub %%mm0, %%mm2                    \n\t"
01842         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
01843 
01844 
01845         : : "r" (src), "r" ((x86_reg)stride)
01846         : "%"REG_a, "%"REG_d
01847     );
01848 
01849 #else // MMX without MMX2
01850     __asm__ volatile(
01851         "lea (%0, %1), %%"REG_a"                \n\t"
01852         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
01853 //      0       1       2       3       4       5       6       7       8       9
01854 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01855         "pxor %%mm7, %%mm7                      \n\t"
01856 
01857 #define REAL_MEDIAN(a,b,c)\
01858         "movq " #a ", %%mm0                     \n\t"\
01859         "movq " #b ", %%mm2                     \n\t"\
01860         "movq " #c ", %%mm1                     \n\t"\
01861         "movq %%mm0, %%mm3                      \n\t"\
01862         "movq %%mm1, %%mm4                      \n\t"\
01863         "movq %%mm2, %%mm5                      \n\t"\
01864         "psubusb %%mm1, %%mm3                   \n\t"\
01865         "psubusb %%mm2, %%mm4                   \n\t"\
01866         "psubusb %%mm0, %%mm5                   \n\t"\
01867         "pcmpeqb %%mm7, %%mm3                   \n\t"\
01868         "pcmpeqb %%mm7, %%mm4                   \n\t"\
01869         "pcmpeqb %%mm7, %%mm5                   \n\t"\
01870         "movq %%mm3, %%mm6                      \n\t"\
01871         "pxor %%mm4, %%mm3                      \n\t"\
01872         "pxor %%mm5, %%mm4                      \n\t"\
01873         "pxor %%mm6, %%mm5                      \n\t"\
01874         "por %%mm3, %%mm1                       \n\t"\
01875         "por %%mm4, %%mm2                       \n\t"\
01876         "por %%mm5, %%mm0                       \n\t"\
01877         "pand %%mm2, %%mm0                      \n\t"\
01878         "pand %%mm1, %%mm0                      \n\t"\
01879         "movq %%mm0, " #b "                     \n\t"
01880 #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
01881 
01882 MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))
01883 MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
01884 MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))
01885 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
01886 
01887         : : "r" (src), "r" ((x86_reg)stride)
01888         : "%"REG_a, "%"REG_d
01889     );
01890 #endif //HAVE_MMX2
01891 #else //HAVE_MMX
01892     int x, y;
01893     src+= 4*stride;
01894     // FIXME - there should be a way to do a few columns in parallel like w/mmx
01895     for(x=0; x<8; x++){
01896         uint8_t *colsrc = src;
01897         for (y=0; y<4; y++){
01898             int a, b, c, d, e, f;
01899             a = colsrc[0       ];
01900             b = colsrc[stride  ];
01901             c = colsrc[stride*2];
01902             d = (a-b)>>31;
01903             e = (b-c)>>31;
01904             f = (c-a)>>31;
01905             colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
01906             colsrc += stride*2;
01907         }
01908         src++;
01909     }
01910 #endif //HAVE_MMX
01911 }
01912 
01913 #if HAVE_MMX
01914 
01917 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
01918 {
01919     __asm__(
01920         "lea (%0, %1), %%"REG_a"                \n\t"
01921 //      0       1       2       3       4       5       6       7       8       9
01922 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
01923         "movq (%0), %%mm0                       \n\t" // 12345678
01924         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
01925         "movq %%mm0, %%mm2                      \n\t" // 12345678
01926         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
01927         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
01928 
01929         "movq (%%"REG_a", %1), %%mm1            \n\t"
01930         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
01931         "movq %%mm1, %%mm4                      \n\t"
01932         "punpcklbw %%mm3, %%mm1                 \n\t"
01933         "punpckhbw %%mm3, %%mm4                 \n\t"
01934 
01935         "movq %%mm0, %%mm3                      \n\t"
01936         "punpcklwd %%mm1, %%mm0                 \n\t"
01937         "punpckhwd %%mm1, %%mm3                 \n\t"
01938         "movq %%mm2, %%mm1                      \n\t"
01939         "punpcklwd %%mm4, %%mm2                 \n\t"
01940         "punpckhwd %%mm4, %%mm1                 \n\t"
01941 
01942         "movd %%mm0, 128(%2)                    \n\t"
01943         "psrlq $32, %%mm0                       \n\t"
01944         "movd %%mm0, 144(%2)                    \n\t"
01945         "movd %%mm3, 160(%2)                    \n\t"
01946         "psrlq $32, %%mm3                       \n\t"
01947         "movd %%mm3, 176(%2)                    \n\t"
01948         "movd %%mm3, 48(%3)                     \n\t"
01949         "movd %%mm2, 192(%2)                    \n\t"
01950         "movd %%mm2, 64(%3)                     \n\t"
01951         "psrlq $32, %%mm2                       \n\t"
01952         "movd %%mm2, 80(%3)                     \n\t"
01953         "movd %%mm1, 96(%3)                     \n\t"
01954         "psrlq $32, %%mm1                       \n\t"
01955         "movd %%mm1, 112(%3)                    \n\t"
01956 
01957         "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
01958 
01959         "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
01960         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
01961         "movq %%mm0, %%mm2                      \n\t" // 12345678
01962         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
01963         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
01964 
01965         "movq (%%"REG_a", %1), %%mm1            \n\t"
01966         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
01967         "movq %%mm1, %%mm4                      \n\t"
01968         "punpcklbw %%mm3, %%mm1                 \n\t"
01969         "punpckhbw %%mm3, %%mm4                 \n\t"
01970 
01971         "movq %%mm0, %%mm3                      \n\t"
01972         "punpcklwd %%mm1, %%mm0                 \n\t"
01973         "punpckhwd %%mm1, %%mm3                 \n\t"
01974         "movq %%mm2, %%mm1                      \n\t"
01975         "punpcklwd %%mm4, %%mm2                 \n\t"
01976         "punpckhwd %%mm4, %%mm1                 \n\t"
01977 
01978         "movd %%mm0, 132(%2)                    \n\t"
01979         "psrlq $32, %%mm0                       \n\t"
01980         "movd %%mm0, 148(%2)                    \n\t"
01981         "movd %%mm3, 164(%2)                    \n\t"
01982         "psrlq $32, %%mm3                       \n\t"
01983         "movd %%mm3, 180(%2)                    \n\t"
01984         "movd %%mm3, 52(%3)                     \n\t"
01985         "movd %%mm2, 196(%2)                    \n\t"
01986         "movd %%mm2, 68(%3)                     \n\t"
01987         "psrlq $32, %%mm2                       \n\t"
01988         "movd %%mm2, 84(%3)                     \n\t"
01989         "movd %%mm1, 100(%3)                    \n\t"
01990         "psrlq $32, %%mm1                       \n\t"
01991         "movd %%mm1, 116(%3)                    \n\t"
01992 
01993 
01994         :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)
01995         : "%"REG_a
01996     );
01997 }
01998 
02002 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
02003 {
02004     __asm__(
02005         "lea (%0, %1), %%"REG_a"                \n\t"
02006         "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
02007 //      0       1       2       3       4       5       6       7       8       9
02008 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
02009         "movq (%2), %%mm0                       \n\t" // 12345678
02010         "movq 16(%2), %%mm1                     \n\t" // abcdefgh
02011         "movq %%mm0, %%mm2                      \n\t" // 12345678
02012         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02013         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02014 
02015         "movq 32(%2), %%mm1                     \n\t"
02016         "movq 48(%2), %%mm3                     \n\t"
02017         "movq %%mm1, %%mm4                      \n\t"
02018         "punpcklbw %%mm3, %%mm1                 \n\t"
02019         "punpckhbw %%mm3, %%mm4                 \n\t"
02020 
02021         "movq %%mm0, %%mm3                      \n\t"
02022         "punpcklwd %%mm1, %%mm0                 \n\t"
02023         "punpckhwd %%mm1, %%mm3                 \n\t"
02024         "movq %%mm2, %%mm1                      \n\t"
02025         "punpcklwd %%mm4, %%mm2                 \n\t"
02026         "punpckhwd %%mm4, %%mm1                 \n\t"
02027 
02028         "movd %%mm0, (%0)                       \n\t"
02029         "psrlq $32, %%mm0                       \n\t"
02030         "movd %%mm0, (%%"REG_a")                \n\t"
02031         "movd %%mm3, (%%"REG_a", %1)            \n\t"
02032         "psrlq $32, %%mm3                       \n\t"
02033         "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
02034         "movd %%mm2, (%0, %1, 4)                \n\t"
02035         "psrlq $32, %%mm2                       \n\t"
02036         "movd %%mm2, (%%"REG_d")                \n\t"
02037         "movd %%mm1, (%%"REG_d", %1)            \n\t"
02038         "psrlq $32, %%mm1                       \n\t"
02039         "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
02040 
02041 
02042         "movq 64(%2), %%mm0                     \n\t" // 12345678
02043         "movq 80(%2), %%mm1                     \n\t" // abcdefgh
02044         "movq %%mm0, %%mm2                      \n\t" // 12345678
02045         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
02046         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
02047 
02048         "movq 96(%2), %%mm1                     \n\t"
02049         "movq 112(%2), %%mm3                    \n\t"
02050         "movq %%mm1, %%mm4                      \n\t"
02051         "punpcklbw %%mm3, %%mm1                 \n\t"
02052         "punpckhbw %%mm3, %%mm4                 \n\t"
02053 
02054         "movq %%mm0, %%mm3                      \n\t"
02055         "punpcklwd %%mm1, %%mm0                 \n\t"
02056         "punpckhwd %%mm1, %%mm3                 \n\t"
02057         "movq %%mm2, %%mm1                      \n\t"
02058         "punpcklwd %%mm4, %%mm2                 \n\t"
02059         "punpckhwd %%mm4, %%mm1                 \n\t"
02060 
02061         "movd %%mm0, 4(%0)                      \n\t"
02062         "psrlq $32, %%mm0                       \n\t"
02063         "movd %%mm0, 4(%%"REG_a")               \n\t"
02064         "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
02065         "psrlq $32, %%mm3                       \n\t"
02066         "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
02067         "movd %%mm2, 4(%0, %1, 4)               \n\t"
02068         "psrlq $32, %%mm2                       \n\t"
02069         "movd %%mm2, 4(%%"REG_d")               \n\t"
02070         "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
02071         "psrlq $32, %%mm1                       \n\t"
02072         "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
02073 
02074         :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)
02075         : "%"REG_a, "%"REG_d
02076     );
02077 }
02078 #endif //HAVE_MMX
02079 //static long test=0;
02080 
02081 #if !HAVE_ALTIVEC
02082 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
02083                                     uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)
02084 {
02085     // to save a register (FIXME do this outside of the loops)
02086     tempBlurredPast[127]= maxNoise[0];
02087     tempBlurredPast[128]= maxNoise[1];
02088     tempBlurredPast[129]= maxNoise[2];
02089 
02090 #define FAST_L2_DIFF
02091 //#define L1_DIFF //u should change the thresholds too if u try that one
02092 #if HAVE_MMX2 || HAVE_AMD3DNOW
02093     __asm__ volatile(
02094         "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
02095         "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
02096         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02097 //      0       1       2       3       4       5       6       7       8       9
02098 //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
02099 //FIXME reorder?
02100 #ifdef L1_DIFF //needs mmx2
02101         "movq (%0), %%mm0                       \n\t" // L0
02102         "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
02103         "movq (%0, %2), %%mm1                   \n\t" // L1
02104         "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
02105         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02106         "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
02107         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02108         "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
02109 
02110         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02111         "paddw %%mm1, %%mm0                     \n\t"
02112         "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
02113         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02114         "paddw %%mm2, %%mm0                     \n\t"
02115         "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
02116         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02117         "paddw %%mm3, %%mm0                     \n\t"
02118         "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
02119         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02120         "paddw %%mm4, %%mm0                     \n\t"
02121         "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
02122         "paddw %%mm5, %%mm6                     \n\t"
02123         "paddw %%mm7, %%mm6                     \n\t"
02124         "paddw %%mm6, %%mm0                     \n\t"
02125 #else //L1_DIFF
02126 #if defined (FAST_L2_DIFF)
02127         "pcmpeqb %%mm7, %%mm7                   \n\t"
02128         "movq "MANGLE(b80)", %%mm6              \n\t"
02129         "pxor %%mm0, %%mm0                      \n\t"
02130 #define REAL_L2_DIFF_CORE(a, b)\
02131         "movq " #a ", %%mm5                     \n\t"\
02132         "movq " #b ", %%mm2                     \n\t"\
02133         "pxor %%mm7, %%mm2                      \n\t"\
02134         PAVGB(%%mm2, %%mm5)\
02135         "paddb %%mm6, %%mm5                     \n\t"\
02136         "movq %%mm5, %%mm2                      \n\t"\
02137         "psllw $8, %%mm5                        \n\t"\
02138         "pmaddwd %%mm5, %%mm5                   \n\t"\
02139         "pmaddwd %%mm2, %%mm2                   \n\t"\
02140         "paddd %%mm2, %%mm5                     \n\t"\
02141         "psrld $14, %%mm5                       \n\t"\
02142         "paddd %%mm5, %%mm0                     \n\t"
02143 
02144 #else //defined (FAST_L2_DIFF)
02145         "pxor %%mm7, %%mm7                      \n\t"
02146         "pxor %%mm0, %%mm0                      \n\t"
02147 #define REAL_L2_DIFF_CORE(a, b)\
02148         "movq " #a ", %%mm5                     \n\t"\
02149         "movq " #b ", %%mm2                     \n\t"\
02150         "movq %%mm5, %%mm1                      \n\t"\
02151         "movq %%mm2, %%mm3                      \n\t"\
02152         "punpcklbw %%mm7, %%mm5                 \n\t"\
02153         "punpckhbw %%mm7, %%mm1                 \n\t"\
02154         "punpcklbw %%mm7, %%mm2                 \n\t"\
02155         "punpckhbw %%mm7, %%mm3                 \n\t"\
02156         "psubw %%mm2, %%mm5                     \n\t"\
02157         "psubw %%mm3, %%mm1                     \n\t"\
02158         "pmaddwd %%mm5, %%mm5                   \n\t"\
02159         "pmaddwd %%mm1, %%mm1                   \n\t"\
02160         "paddd %%mm1, %%mm5                     \n\t"\
02161         "paddd %%mm5, %%mm0                     \n\t"
02162 
02163 #endif //defined (FAST_L2_DIFF)
02164 
02165 #define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)
02166 
02167 L2_DIFF_CORE((%0)          , (%1))
02168 L2_DIFF_CORE((%0, %2)      , (%1, %2))
02169 L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))
02170 L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))
02171 L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))
02172 L2_DIFF_CORE((%0, %%REGd)  , (%1, %%REGd))
02173 L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
02174 L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))
02175 
02176 #endif //L1_DIFF
02177 
02178         "movq %%mm0, %%mm4                      \n\t"
02179         "psrlq $32, %%mm0                       \n\t"
02180         "paddd %%mm0, %%mm4                     \n\t"
02181         "movd %%mm4, %%ecx                      \n\t"
02182         "shll $2, %%ecx                         \n\t"
02183         "mov %3, %%"REG_d"                      \n\t"
02184         "addl -4(%%"REG_d"), %%ecx              \n\t"
02185         "addl 4(%%"REG_d"), %%ecx               \n\t"
02186         "addl -1024(%%"REG_d"), %%ecx           \n\t"
02187         "addl $4, %%ecx                         \n\t"
02188         "addl 1024(%%"REG_d"), %%ecx            \n\t"
02189         "shrl $3, %%ecx                         \n\t"
02190         "movl %%ecx, (%%"REG_d")                \n\t"
02191 
02192 //        "mov %3, %%"REG_c"                      \n\t"
02193 //        "mov %%"REG_c", test                    \n\t"
02194 //        "jmp 4f                                 \n\t"
02195         "cmpl 512(%%"REG_d"), %%ecx             \n\t"
02196         " jb 2f                                 \n\t"
02197         "cmpl 516(%%"REG_d"), %%ecx             \n\t"
02198         " jb 1f                                 \n\t"
02199 
02200         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02201         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02202         "movq (%0), %%mm0                       \n\t" // L0
02203         "movq (%0, %2), %%mm1                   \n\t" // L1
02204         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02205         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02206         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02207         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02208         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02209         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02210         "movq %%mm0, (%1)                       \n\t" // L0
02211         "movq %%mm1, (%1, %2)                   \n\t" // L1
02212         "movq %%mm2, (%1, %2, 2)                \n\t" // L2
02213         "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
02214         "movq %%mm4, (%1, %2, 4)                \n\t" // L4
02215         "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
02216         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
02217         "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
02218         "jmp 4f                                 \n\t"
02219 
02220         "1:                                     \n\t"
02221         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02222         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02223         "movq (%0), %%mm0                       \n\t" // L0
02224         PAVGB((%1), %%mm0)                            // L0
02225         "movq (%0, %2), %%mm1                   \n\t" // L1
02226         PAVGB((%1, %2), %%mm1)                        // L1
02227         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02228         PAVGB((%1, %2, 2), %%mm2)                     // L2
02229         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02230         PAVGB((%1, %%REGa), %%mm3)                    // L3
02231         "movq (%0, %2, 4), %%mm4                \n\t" // L4
02232         PAVGB((%1, %2, 4), %%mm4)                     // L4
02233         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
02234         PAVGB((%1, %%REGd), %%mm5)                    // L5
02235         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
02236         PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
02237         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
02238         PAVGB((%1, %%REGc), %%mm7)                    // L7
02239         "movq %%mm0, (%1)                       \n\t" // R0
02240         "movq %%mm1, (%1, %2)                   \n\t" // R1
02241         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02242         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02243         "movq %%mm4, (%1, %2, 4)                \n\t" // R4
02244         "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
02245         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
02246         "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
02247         "movq %%mm0, (%0)                       \n\t" // L0
02248         "movq %%mm1, (%0, %2)                   \n\t" // L1
02249         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02250         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02251         "movq %%mm4, (%0, %2, 4)                \n\t" // L4
02252         "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
02253         "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
02254         "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
02255         "jmp 4f                                 \n\t"
02256 
02257         "2:                                     \n\t"
02258         "cmpl 508(%%"REG_d"), %%ecx             \n\t"
02259         " jb 3f                                 \n\t"
02260 
02261         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02262         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02263         "movq (%0), %%mm0                       \n\t" // L0
02264         "movq (%0, %2), %%mm1                   \n\t" // L1
02265         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02266         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02267         "movq (%1), %%mm4                       \n\t" // R0
02268         "movq (%1, %2), %%mm5                   \n\t" // R1
02269         "movq (%1, %2, 2), %%mm6                \n\t" // R2
02270         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
02271         PAVGB(%%mm4, %%mm0)
02272         PAVGB(%%mm5, %%mm1)
02273         PAVGB(%%mm6, %%mm2)
02274         PAVGB(%%mm7, %%mm3)
02275         PAVGB(%%mm4, %%mm0)
02276         PAVGB(%%mm5, %%mm1)
02277         PAVGB(%%mm6, %%mm2)
02278         PAVGB(%%mm7, %%mm3)
02279         "movq %%mm0, (%1)                       \n\t" // R0
02280         "movq %%mm1, (%1, %2)                   \n\t" // R1
02281         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02282         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02283         "movq %%mm0, (%0)                       \n\t" // L0
02284         "movq %%mm1, (%0, %2)                   \n\t" // L1
02285         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02286         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02287 
02288         "movq (%0, %2, 4), %%mm0                \n\t" // L4
02289         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
02290         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
02291         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
02292         "movq (%1, %2, 4), %%mm4                \n\t" // R4
02293         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
02294         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
02295         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
02296         PAVGB(%%mm4, %%mm0)
02297         PAVGB(%%mm5, %%mm1)
02298         PAVGB(%%mm6, %%mm2)
02299         PAVGB(%%mm7, %%mm3)
02300         PAVGB(%%mm4, %%mm0)
02301         PAVGB(%%mm5, %%mm1)
02302         PAVGB(%%mm6, %%mm2)
02303         PAVGB(%%mm7, %%mm3)
02304         "movq %%mm0, (%1, %2, 4)                \n\t" // R4
02305         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
02306         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
02307         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
02308         "movq %%mm0, (%0, %2, 4)                \n\t" // L4
02309         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
02310         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
02311         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
02312         "jmp 4f                                 \n\t"
02313 
02314         "3:                                     \n\t"
02315         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
02316         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
02317         "movq (%0), %%mm0                       \n\t" // L0
02318         "movq (%0, %2), %%mm1                   \n\t" // L1
02319         "movq (%0, %2, 2), %%mm2                \n\t" // L2
02320         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
02321         "movq (%1), %%mm4                       \n\t" // R0
02322         "movq (%1, %2), %%mm5                   \n\t" // R1
02323         "movq (%1, %2, 2), %%mm6                \n\t" // R2
02324         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
02325         PAVGB(%%mm4, %%mm0)
02326         PAVGB(%%mm5, %%mm1)
02327         PAVGB(%%mm6, %%mm2)
02328         PAVGB(%%mm7, %%mm3)
02329         PAVGB(%%mm4, %%mm0)
02330         PAVGB(%%mm5, %%mm1)
02331         PAVGB(%%mm6, %%mm2)
02332         PAVGB(%%mm7, %%mm3)
02333         PAVGB(%%mm4, %%mm0)
02334         PAVGB(%%mm5, %%mm1)
02335         PAVGB(%%mm6, %%mm2)
02336         PAVGB(%%mm7, %%mm3)
02337         "movq %%mm0, (%1)                       \n\t" // R0
02338         "movq %%mm1, (%1, %2)                   \n\t" // R1
02339         "movq %%mm2, (%1, %2, 2)                \n\t" // R2
02340         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
02341         "movq %%mm0, (%0)                       \n\t" // L0
02342         "movq %%mm1, (%0, %2)                   \n\t" // L1
02343         "movq %%mm2, (%0, %2, 2)                \n\t" // L2
02344         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
02345 
02346         "movq (%0, %2, 4), %%mm0                \n\t" // L4
02347         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
02348         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
02349         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
02350         "movq (%1, %2, 4), %%mm4                \n\t" // R4
02351         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
02352         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
02353         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
02354         PAVGB(%%mm4, %%mm0)
02355         PAVGB(%%mm5, %%mm1)
02356         PAVGB(%%mm6, %%mm2)
02357         PAVGB(%%mm7, %%mm3)
02358         PAVGB(%%mm4, %%mm0)
02359         PAVGB(%%mm5, %%mm1)
02360         PAVGB(%%mm6, %%mm2)
02361         PAVGB(%%mm7, %%mm3)
02362         PAVGB(%%mm4, %%mm0)
02363         PAVGB(%%mm5, %%mm1)
02364         PAVGB(%%mm6, %%mm2)
02365         PAVGB(%%mm7, %%mm3)
02366         "movq %%mm0, (%1, %2, 4)                \n\t" // R4
02367         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
02368         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
02369         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
02370         "movq %%mm0, (%0, %2, 4)                \n\t" // L4
02371         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
02372         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
02373         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
02374 
02375         "4:                                     \n\t"
02376 
02377         :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)
02378         : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
02379     );
02380 #else //HAVE_MMX2 || HAVE_AMD3DNOW
02381 {
02382     int y;
02383     int d=0;
02384 //    int sysd=0;
02385     int i;
02386 
02387     for(y=0; y<8; y++){
02388         int x;
02389         for(x=0; x<8; x++){
02390             int ref= tempBlurred[ x + y*stride ];
02391             int cur= src[ x + y*stride ];
02392             int d1=ref - cur;
02393 //            if(x==0 || x==7) d1+= d1>>1;
02394 //            if(y==0 || y==7) d1+= d1>>1;
02395 //            d+= FFABS(d1);
02396             d+= d1*d1;
02397 //            sysd+= d1;
02398         }
02399     }
02400     i=d;
02401     d=  (
02402         4*d
02403         +(*(tempBlurredPast-256))
02404         +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
02405         +(*(tempBlurredPast+256))
02406         +4)>>3;
02407     *tempBlurredPast=i;
02408 //    ((*tempBlurredPast)*3 + d + 2)>>2;
02409 
02410 /*
02411 Switch between
02412  1  0  0  0  0  0  0  (0)
02413 64 32 16  8  4  2  1  (1)
02414 64 48 36 27 20 15 11 (33) (approx)
02415 64 56 49 43 37 33 29 (200) (approx)
02416 */
02417     if(d > maxNoise[1]){
02418         if(d < maxNoise[2]){
02419             for(y=0; y<8; y++){
02420                 int x;
02421                 for(x=0; x<8; x++){
02422                     int ref= tempBlurred[ x + y*stride ];
02423                     int cur= src[ x + y*stride ];
02424                     tempBlurred[ x + y*stride ]=
02425                     src[ x + y*stride ]=
02426                         (ref + cur + 1)>>1;
02427                 }
02428             }
02429         }else{
02430             for(y=0; y<8; y++){
02431                 int x;
02432                 for(x=0; x<8; x++){
02433                     tempBlurred[ x + y*stride ]= src[ x + y*stride ];
02434                 }
02435             }
02436         }
02437     }else{
02438         if(d < maxNoise[0]){
02439             for(y=0; y<8; y++){
02440                 int x;
02441                 for(x=0; x<8; x++){
02442                     int ref= tempBlurred[ x + y*stride ];
02443                     int cur= src[ x + y*stride ];
02444                     tempBlurred[ x + y*stride ]=
02445                     src[ x + y*stride ]=
02446                         (ref*7 + cur + 4)>>3;
02447                 }
02448             }
02449         }else{
02450             for(y=0; y<8; y++){
02451                 int x;
02452                 for(x=0; x<8; x++){
02453                     int ref= tempBlurred[ x + y*stride ];
02454                     int cur= src[ x + y*stride ];
02455                     tempBlurred[ x + y*stride ]=
02456                     src[ x + y*stride ]=
02457                         (ref*3 + cur + 2)>>2;
02458                 }
02459             }
02460         }
02461     }
02462 }
02463 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
02464 }
02465 #endif //HAVE_ALTIVEC
02466 
02467 #if HAVE_MMX
02468 
02471 static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
02472     int64_t dc_mask, eq_mask, both_masks;
02473     int64_t sums[10*8*2];
02474     src+= step*3; // src points to begin of the 8x8 Block
02475 //START_TIMER
02476     __asm__ volatile(
02477         "movq %0, %%mm7                         \n\t"
02478         "movq %1, %%mm6                         \n\t"
02479         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
02480         );
02481 
02482     __asm__ volatile(
02483         "lea (%2, %3), %%"REG_a"                \n\t"
02484 //      0       1       2       3       4       5       6       7       8       9
02485 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
02486 
02487         "movq (%2), %%mm0                       \n\t"
02488         "movq (%%"REG_a"), %%mm1                \n\t"
02489         "movq %%mm1, %%mm3                      \n\t"
02490         "movq %%mm1, %%mm4                      \n\t"
02491         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
02492         "paddb %%mm7, %%mm0                     \n\t"
02493         "pcmpgtb %%mm6, %%mm0                   \n\t"
02494 
02495         "movq (%%"REG_a",%3), %%mm2             \n\t"
02496         PMAXUB(%%mm2, %%mm4)
02497         PMINUB(%%mm2, %%mm3, %%mm5)
02498         "psubb %%mm2, %%mm1                     \n\t"
02499         "paddb %%mm7, %%mm1                     \n\t"
02500         "pcmpgtb %%mm6, %%mm1                   \n\t"
02501         "paddb %%mm1, %%mm0                     \n\t"
02502 
02503         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
02504         PMAXUB(%%mm1, %%mm4)
02505         PMINUB(%%mm1, %%mm3, %%mm5)
02506         "psubb %%mm1, %%mm2                     \n\t"
02507         "paddb %%mm7, %%mm2                     \n\t"
02508         "pcmpgtb %%mm6, %%mm2                   \n\t"
02509         "paddb %%mm2, %%mm0                     \n\t"
02510 
02511         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
02512 
02513         "movq (%2, %3, 4), %%mm2                \n\t"
02514         PMAXUB(%%mm2, %%mm4)
02515         PMINUB(%%mm2, %%mm3, %%mm5)
02516         "psubb %%mm2, %%mm1                     \n\t"
02517         "paddb %%mm7, %%mm1                     \n\t"
02518         "pcmpgtb %%mm6, %%mm1                   \n\t"
02519         "paddb %%mm1, %%mm0                     \n\t"
02520 
02521         "movq (%%"REG_a"), %%mm1                \n\t"
02522         PMAXUB(%%mm1, %%mm4)
02523         PMINUB(%%mm1, %%mm3, %%mm5)
02524         "psubb %%mm1, %%mm2                     \n\t"
02525         "paddb %%mm7, %%mm2                     \n\t"
02526         "pcmpgtb %%mm6, %%mm2                   \n\t"
02527         "paddb %%mm2, %%mm0                     \n\t"
02528 
02529         "movq (%%"REG_a", %3), %%mm2            \n\t"
02530         PMAXUB(%%mm2, %%mm4)
02531         PMINUB(%%mm2, %%mm3, %%mm5)
02532         "psubb %%mm2, %%mm1                     \n\t"
02533         "paddb %%mm7, %%mm1                     \n\t"
02534         "pcmpgtb %%mm6, %%mm1                   \n\t"
02535         "paddb %%mm1, %%mm0                     \n\t"
02536 
02537         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
02538         PMAXUB(%%mm1, %%mm4)
02539         PMINUB(%%mm1, %%mm3, %%mm5)
02540         "psubb %%mm1, %%mm2                     \n\t"
02541         "paddb %%mm7, %%mm2                     \n\t"
02542         "pcmpgtb %%mm6, %%mm2                   \n\t"
02543         "paddb %%mm2, %%mm0                     \n\t"
02544 
02545         "movq (%2, %3, 8), %%mm2                \n\t"
02546         PMAXUB(%%mm2, %%mm4)
02547         PMINUB(%%mm2, %%mm3, %%mm5)
02548         "psubb %%mm2, %%mm1                     \n\t"
02549         "paddb %%mm7, %%mm1                     \n\t"
02550         "pcmpgtb %%mm6, %%mm1                   \n\t"
02551         "paddb %%mm1, %%mm0                     \n\t"
02552 
02553         "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
02554         "psubb %%mm1, %%mm2                     \n\t"
02555         "paddb %%mm7, %%mm2                     \n\t"
02556         "pcmpgtb %%mm6, %%mm2                   \n\t"
02557         "paddb %%mm2, %%mm0                     \n\t"
02558         "psubusb %%mm3, %%mm4                   \n\t"
02559 
02560         "pxor %%mm6, %%mm6                      \n\t"
02561         "movq %4, %%mm7                         \n\t" // QP,..., QP
02562         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
02563         "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
02564         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
02565         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
02566         "movq %%mm7, %1                         \n\t"
02567 
02568         "movq %5, %%mm7                         \n\t"
02569         "punpcklbw %%mm7, %%mm7                 \n\t"
02570         "punpcklbw %%mm7, %%mm7                 \n\t"
02571         "punpcklbw %%mm7, %%mm7                 \n\t"
02572         "psubb %%mm0, %%mm6                     \n\t"
02573         "pcmpgtb %%mm7, %%mm6                   \n\t"
02574         "movq %%mm6, %0                         \n\t"
02575 
02576         : "=m" (eq_mask), "=m" (dc_mask)
02577         : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
02578         : "%"REG_a
02579     );
02580 
02581     both_masks = dc_mask & eq_mask;
02582 
02583     if(both_masks){
02584         x86_reg offset= -8*step;
02585         int64_t *temp_sums= sums;
02586 
02587         __asm__ volatile(
02588             "movq %2, %%mm0                         \n\t"  // QP,..., QP
02589             "pxor %%mm4, %%mm4                      \n\t"
02590 
02591             "movq (%0), %%mm6                       \n\t"
02592             "movq (%0, %1), %%mm5                   \n\t"
02593             "movq %%mm5, %%mm1                      \n\t"
02594             "movq %%mm6, %%mm2                      \n\t"
02595             "psubusb %%mm6, %%mm5                   \n\t"
02596             "psubusb %%mm1, %%mm2                   \n\t"
02597             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
02598             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
02599             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
02600 
02601             "pxor %%mm6, %%mm1                      \n\t"
02602             "pand %%mm0, %%mm1                      \n\t"
02603             "pxor %%mm1, %%mm6                      \n\t"
02604             // 0:QP  6:First
02605 
02606             "movq (%0, %1, 8), %%mm5                \n\t"
02607             "add %1, %0                             \n\t" // %0 points to line 1 not 0
02608             "movq (%0, %1, 8), %%mm7                \n\t"
02609             "movq %%mm5, %%mm1                      \n\t"
02610             "movq %%mm7, %%mm2                      \n\t"
02611             "psubusb %%mm7, %%mm5                   \n\t"
02612             "psubusb %%mm1, %%mm2                   \n\t"
02613             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
02614             "movq %2, %%mm0                         \n\t"  // QP,..., QP
02615             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
02616             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
02617 
02618             "pxor %%mm7, %%mm1                      \n\t"
02619             "pand %%mm0, %%mm1                      \n\t"
02620             "pxor %%mm1, %%mm7                      \n\t"
02621 
02622             "movq %%mm6, %%mm5                      \n\t"
02623             "punpckhbw %%mm4, %%mm6                 \n\t"
02624             "punpcklbw %%mm4, %%mm5                 \n\t"
02625             // 4:0 5/6:First 7:Last
02626 
02627             "movq %%mm5, %%mm0                      \n\t"
02628             "movq %%mm6, %%mm1                      \n\t"
02629             "psllw $2, %%mm0                        \n\t"
02630             "psllw $2, %%mm1                        \n\t"
02631             "paddw "MANGLE(w04)", %%mm0             \n\t"
02632             "paddw "MANGLE(w04)", %%mm1             \n\t"
02633 
02634 #define NEXT\
02635             "movq (%0), %%mm2                       \n\t"\
02636             "movq (%0), %%mm3                       \n\t"\
02637             "add %1, %0                             \n\t"\
02638             "punpcklbw %%mm4, %%mm2                 \n\t"\
02639             "punpckhbw %%mm4, %%mm3                 \n\t"\
02640             "paddw %%mm2, %%mm0                     \n\t"\
02641             "paddw %%mm3, %%mm1                     \n\t"
02642 
02643 #define PREV\
02644             "movq (%0), %%mm2                       \n\t"\
02645             "movq (%0), %%mm3                       \n\t"\
02646             "add %1, %0                             \n\t"\
02647             "punpcklbw %%mm4, %%mm2                 \n\t"\
02648             "punpckhbw %%mm4, %%mm3                 \n\t"\
02649             "psubw %%mm2, %%mm0                     \n\t"\
02650             "psubw %%mm3, %%mm1                     \n\t"
02651 
02652 
02653             NEXT //0
02654             NEXT //1
02655             NEXT //2
02656             "movq %%mm0, (%3)                       \n\t"
02657             "movq %%mm1, 8(%3)                      \n\t"
02658 
02659             NEXT //3
02660             "psubw %%mm5, %%mm0                     \n\t"
02661             "psubw %%mm6, %%mm1                     \n\t"
02662             "movq %%mm0, 16(%3)                     \n\t"
02663             "movq %%mm1, 24(%3)                     \n\t"
02664 
02665             NEXT //4
02666             "psubw %%mm5, %%mm0                     \n\t"
02667             "psubw %%mm6, %%mm1                     \n\t"
02668             "movq %%mm0, 32(%3)                     \n\t"
02669             "movq %%mm1, 40(%3)                     \n\t"
02670 
02671             NEXT //5
02672             "psubw %%mm5, %%mm0                     \n\t"
02673             "psubw %%mm6, %%mm1                     \n\t"
02674             "movq %%mm0, 48(%3)                     \n\t"
02675             "movq %%mm1, 56(%3)                     \n\t"
02676 
02677             NEXT //6
02678             "psubw %%mm5, %%mm0                     \n\t"
02679             "psubw %%mm6, %%mm1                     \n\t"
02680             "movq %%mm0, 64(%3)                     \n\t"
02681             "movq %%mm1, 72(%3)                     \n\t"
02682 
02683             "movq %%mm7, %%mm6                      \n\t"
02684             "punpckhbw %%mm4, %%mm7                 \n\t"
02685             "punpcklbw %%mm4, %%mm6                 \n\t"
02686 
02687             NEXT //7
02688             "mov %4, %0                             \n\t"
02689             "add %1, %0                             \n\t"
02690             PREV //0
02691             "movq %%mm0, 80(%3)                     \n\t"
02692             "movq %%mm1, 88(%3)                     \n\t"
02693 
02694             PREV //1
02695             "paddw %%mm6, %%mm0                     \n\t"
02696             "paddw %%mm7, %%mm1                     \n\t"
02697             "movq %%mm0, 96(%3)                     \n\t"
02698             "movq %%mm1, 104(%3)                    \n\t"
02699 
02700             PREV //2
02701             "paddw %%mm6, %%mm0                     \n\t"
02702             "paddw %%mm7, %%mm1                     \n\t"
02703             "movq %%mm0, 112(%3)                    \n\t"
02704             "movq %%mm1, 120(%3)                    \n\t"
02705 
02706             PREV //3
02707             "paddw %%mm6, %%mm0                     \n\t"
02708             "paddw %%mm7, %%mm1                     \n\t"
02709             "movq %%mm0, 128(%3)                    \n\t"
02710             "movq %%mm1, 136(%3)                    \n\t"
02711 
02712             PREV //4
02713             "paddw %%mm6, %%mm0                     \n\t"
02714             "paddw %%mm7, %%mm1                     \n\t"
02715             "movq %%mm0, 144(%3)                    \n\t"
02716             "movq %%mm1, 152(%3)                    \n\t"
02717 
02718             "mov %4, %0                             \n\t" //FIXME
02719 
02720             : "+&r"(src)
02721             : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)
02722         );
02723 
02724         src+= step; // src points to begin of the 8x8 Block
02725 
02726         __asm__ volatile(
02727             "movq %4, %%mm6                         \n\t"
02728             "pcmpeqb %%mm5, %%mm5                   \n\t"
02729             "pxor %%mm6, %%mm5                      \n\t"
02730             "pxor %%mm7, %%mm7                      \n\t"
02731 
02732             "1:                                     \n\t"
02733             "movq (%1), %%mm0                       \n\t"
02734             "movq 8(%1), %%mm1                      \n\t"
02735             "paddw 32(%1), %%mm0                    \n\t"
02736             "paddw 40(%1), %%mm1                    \n\t"
02737             "movq (%0, %3), %%mm2                   \n\t"
02738             "movq %%mm2, %%mm3                      \n\t"
02739             "movq %%mm2, %%mm4                      \n\t"
02740             "punpcklbw %%mm7, %%mm2                 \n\t"
02741             "punpckhbw %%mm7, %%mm3                 \n\t"
02742             "paddw %%mm2, %%mm0                     \n\t"
02743             "paddw %%mm3, %%mm1                     \n\t"
02744             "paddw %%mm2, %%mm0                     \n\t"
02745             "paddw %%mm3, %%mm1                     \n\t"
02746             "psrlw $4, %%mm0                        \n\t"
02747             "psrlw $4, %%mm1                        \n\t"
02748             "packuswb %%mm1, %%mm0                  \n\t"
02749             "pand %%mm6, %%mm0                      \n\t"
02750             "pand %%mm5, %%mm4                      \n\t"
02751             "por %%mm4, %%mm0                       \n\t"
02752             "movq %%mm0, (%0, %3)                   \n\t"
02753             "add $16, %1                            \n\t"
02754             "add %2, %0                             \n\t"
02755             " js 1b                                 \n\t"
02756 
02757             : "+r"(offset), "+r"(temp_sums)
02758             : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)
02759         );
02760     }else
02761         src+= step; // src points to begin of the 8x8 Block
02762 
02763     if(eq_mask != -1LL){
02764         uint8_t *temp_src= src;
02765         __asm__ volatile(
02766             "pxor %%mm7, %%mm7                      \n\t"
02767             "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
02768             "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
02769 //      0       1       2       3       4       5       6       7       8       9
02770 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1
02771 
02772             "movq (%0), %%mm0                       \n\t"
02773             "movq %%mm0, %%mm1                      \n\t"
02774             "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
02775             "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
02776 
02777             "movq (%0, %1), %%mm2                   \n\t"
02778             "lea (%0, %1, 2), %%"REG_a"             \n\t"
02779             "movq %%mm2, %%mm3                      \n\t"
02780             "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
02781             "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
02782 
02783             "movq (%%"REG_a"), %%mm4                \n\t"
02784             "movq %%mm4, %%mm5                      \n\t"
02785             "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
02786             "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
02787 
02788             "paddw %%mm0, %%mm0                     \n\t" // 2L0
02789             "paddw %%mm1, %%mm1                     \n\t" // 2H0
02790             "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
02791             "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
02792             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
02793             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
02794 
02795             "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
02796             "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
02797             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
02798             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
02799 
02800             "movq (%%"REG_a", %1), %%mm2            \n\t"
02801             "movq %%mm2, %%mm3                      \n\t"
02802             "punpcklbw %%mm7, %%mm2                 \n\t" // L3
02803             "punpckhbw %%mm7, %%mm3                 \n\t" // H3
02804 
02805             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
02806             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
02807             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02808             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02809             "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02810             "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02811 
02812             "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
02813             "movq %%mm0, %%mm1                      \n\t"
02814             "punpcklbw %%mm7, %%mm0                 \n\t" // L4
02815             "punpckhbw %%mm7, %%mm1                 \n\t" // H4
02816 
02817             "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
02818             "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
02819             "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
02820             "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
02821             "paddw %%mm4, %%mm4                     \n\t" // 2L2
02822             "paddw %%mm5, %%mm5                     \n\t" // 2H2
02823             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
02824             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
02825 
02826             "lea (%%"REG_a", %1), %0                \n\t"
02827             "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
02828             "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
02829             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
02830             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
02831 //50 opcodes so far
02832             "movq (%0, %1, 2), %%mm2                \n\t"
02833             "movq %%mm2, %%mm3                      \n\t"
02834             "punpcklbw %%mm7, %%mm2                 \n\t" // L5
02835             "punpckhbw %%mm7, %%mm3                 \n\t" // H5
02836             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
02837             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
02838             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
02839             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
02840 
02841             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
02842             "punpcklbw %%mm7, %%mm6                 \n\t" // L6
02843             "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
02844             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
02845             "punpckhbw %%mm7, %%mm6                 \n\t" // H6
02846             "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
02847 
02848             "paddw %%mm0, %%mm0                     \n\t" // 2L4
02849             "paddw %%mm1, %%mm1                     \n\t" // 2H4
02850             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
02851             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
02852 
02853             "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
02854             "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
02855             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
02856             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
02857 
02858             "movq (%0, %1, 4), %%mm2                \n\t"
02859             "movq %%mm2, %%mm3                      \n\t"
02860             "punpcklbw %%mm7, %%mm2                 \n\t" // L7
02861             "punpckhbw %%mm7, %%mm3                 \n\t" // H7
02862 
02863             "paddw %%mm2, %%mm2                     \n\t" // 2L7
02864             "paddw %%mm3, %%mm3                     \n\t" // 2H7
02865             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
02866             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
02867 
02868             "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
02869             "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
02870 
02871 #if HAVE_MMX2
02872             "movq %%mm7, %%mm6                      \n\t" // 0
02873             "psubw %%mm0, %%mm6                     \n\t"
02874             "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
02875             "movq %%mm7, %%mm6                      \n\t" // 0
02876             "psubw %%mm1, %%mm6                     \n\t"
02877             "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
02878             "movq %%mm7, %%mm6                      \n\t" // 0
02879             "psubw %%mm2, %%mm6                     \n\t"
02880             "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
02881             "movq %%mm7, %%mm6                      \n\t" // 0
02882             "psubw %%mm3, %%mm6                     \n\t"
02883             "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
02884 #else
02885             "movq %%mm7, %%mm6                      \n\t" // 0
02886             "pcmpgtw %%mm0, %%mm6                   \n\t"
02887             "pxor %%mm6, %%mm0                      \n\t"
02888             "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
02889             "movq %%mm7, %%mm6                      \n\t" // 0
02890             "pcmpgtw %%mm1, %%mm6                   \n\t"
02891             "pxor %%mm6, %%mm1                      \n\t"
02892             "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
02893             "movq %%mm7, %%mm6                      \n\t" // 0
02894             "pcmpgtw %%mm2, %%mm6                   \n\t"
02895             "pxor %%mm6, %%mm2                      \n\t"
02896             "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
02897             "movq %%mm7, %%mm6                      \n\t" // 0
02898             "pcmpgtw %%mm3, %%mm6                   \n\t"
02899             "pxor %%mm6, %%mm3                      \n\t"
02900             "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
02901 #endif
02902 
02903 #if HAVE_MMX2
02904             "pminsw %%mm2, %%mm0                    \n\t"
02905             "pminsw %%mm3, %%mm1                    \n\t"
02906 #else
02907             "movq %%mm0, %%mm6                      \n\t"
02908             "psubusw %%mm2, %%mm6                   \n\t"
02909             "psubw %%mm6, %%mm0                     \n\t"
02910             "movq %%mm1, %%mm6                      \n\t"
02911             "psubusw %%mm3, %%mm6                   \n\t"
02912             "psubw %%mm6, %%mm1                     \n\t"
02913 #endif
02914 
02915             "movd %2, %%mm2                         \n\t" // QP
02916             "punpcklbw %%mm7, %%mm2                 \n\t"
02917 
02918             "movq %%mm7, %%mm6                      \n\t" // 0
02919             "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
02920             "pxor %%mm6, %%mm4                      \n\t"
02921             "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
02922             "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
02923             "pxor %%mm7, %%mm5                      \n\t"
02924             "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
02925 // 100 opcodes
02926             "psllw $3, %%mm2                        \n\t" // 8QP
02927             "movq %%mm2, %%mm3                      \n\t" // 8QP
02928             "pcmpgtw %%mm4, %%mm2                   \n\t"
02929             "pcmpgtw %%mm5, %%mm3                   \n\t"
02930             "pand %%mm2, %%mm4                      \n\t"
02931             "pand %%mm3, %%mm5                      \n\t"
02932 
02933 
02934             "psubusw %%mm0, %%mm4                   \n\t" // hd
02935             "psubusw %%mm1, %%mm5                   \n\t" // ld
02936 
02937 
02938             "movq "MANGLE(w05)", %%mm2              \n\t" // 5
02939             "pmullw %%mm2, %%mm4                    \n\t"
02940             "pmullw %%mm2, %%mm5                    \n\t"
02941             "movq "MANGLE(w20)", %%mm2              \n\t" // 32
02942             "paddw %%mm2, %%mm4                     \n\t"
02943             "paddw %%mm2, %%mm5                     \n\t"
02944             "psrlw $6, %%mm4                        \n\t"
02945             "psrlw $6, %%mm5                        \n\t"
02946 
02947             "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
02948             "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
02949 
02950             "pxor %%mm2, %%mm2                      \n\t"
02951             "pxor %%mm3, %%mm3                      \n\t"
02952 
02953             "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
02954             "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
02955             "pxor %%mm2, %%mm0                      \n\t"
02956             "pxor %%mm3, %%mm1                      \n\t"
02957             "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
02958             "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
02959             "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
02960             "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
02961 
02962             "pxor %%mm6, %%mm2                      \n\t"
02963             "pxor %%mm7, %%mm3                      \n\t"
02964             "pand %%mm2, %%mm4                      \n\t"
02965             "pand %%mm3, %%mm5                      \n\t"
02966 
02967 #if HAVE_MMX2
02968             "pminsw %%mm0, %%mm4                    \n\t"
02969             "pminsw %%mm1, %%mm5                    \n\t"
02970 #else
02971             "movq %%mm4, %%mm2                      \n\t"
02972             "psubusw %%mm0, %%mm2                   \n\t"
02973             "psubw %%mm2, %%mm4                     \n\t"
02974             "movq %%mm5, %%mm2                      \n\t"
02975             "psubusw %%mm1, %%mm2                   \n\t"
02976             "psubw %%mm2, %%mm5                     \n\t"
02977 #endif
02978             "pxor %%mm6, %%mm4                      \n\t"
02979             "pxor %%mm7, %%mm5                      \n\t"
02980             "psubw %%mm6, %%mm4                     \n\t"
02981             "psubw %%mm7, %%mm5                     \n\t"
02982             "packsswb %%mm5, %%mm4                  \n\t"
02983             "movq %3, %%mm1                         \n\t"
02984             "pandn %%mm4, %%mm1                     \n\t"
02985             "movq (%0), %%mm0                       \n\t"
02986             "paddb   %%mm1, %%mm0                   \n\t"
02987             "movq %%mm0, (%0)                       \n\t"
02988             "movq (%0, %1), %%mm0                   \n\t"
02989             "psubb %%mm1, %%mm0                     \n\t"
02990             "movq %%mm0, (%0, %1)                   \n\t"
02991 
02992             : "+r" (temp_src)
02993             : "r" ((x86_reg)step), "m" (c->pQPb), "m"(eq_mask)
02994             : "%"REG_a, "%"REG_c
02995         );
02996     }
02997 /*if(step==16){
02998     STOP_TIMER("step16")
02999 }else{
03000     STOP_TIMER("stepX")
03001 }*/
03002 }
03003 #endif //HAVE_MMX
03004 
03005 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
03006                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
03007 
03012 #undef REAL_SCALED_CPY
03013 #undef SCALED_CPY
03014 
03015 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
03016                                      int levelFix, int64_t *packedOffsetAndScale)
03017 {
03018 #if !HAVE_MMX
03019     int i;
03020 #endif
03021     if(levelFix){
03022 #if HAVE_MMX
03023     __asm__ volatile(
03024         "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
03025         "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
03026         "lea (%2,%4), %%"REG_a"         \n\t"
03027         "lea (%3,%5), %%"REG_d"         \n\t"
03028         "pxor %%mm4, %%mm4              \n\t"
03029 #if HAVE_MMX2
03030 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \
03031         "movq " #src1 ", %%mm0          \n\t"\
03032         "movq " #src1 ", %%mm5          \n\t"\
03033         "movq " #src2 ", %%mm1          \n\t"\
03034         "movq " #src2 ", %%mm6          \n\t"\
03035         "punpcklbw %%mm0, %%mm0         \n\t"\
03036         "punpckhbw %%mm5, %%mm5         \n\t"\
03037         "punpcklbw %%mm1, %%mm1         \n\t"\
03038         "punpckhbw %%mm6, %%mm6         \n\t"\
03039         "pmulhuw %%mm3, %%mm0           \n\t"\
03040         "pmulhuw %%mm3, %%mm5           \n\t"\
03041         "pmulhuw %%mm3, %%mm1           \n\t"\
03042         "pmulhuw %%mm3, %%mm6           \n\t"\
03043         "psubw %%mm2, %%mm0             \n\t"\
03044         "psubw %%mm2, %%mm5             \n\t"\
03045         "psubw %%mm2, %%mm1             \n\t"\
03046         "psubw %%mm2, %%mm6             \n\t"\
03047         "packuswb %%mm5, %%mm0          \n\t"\
03048         "packuswb %%mm6, %%mm1          \n\t"\
03049         "movq %%mm0, " #dst1 "          \n\t"\
03050         "movq %%mm1, " #dst2 "          \n\t"\
03051 
03052 #else //HAVE_MMX2
03053 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \
03054         "movq " #src1 ", %%mm0          \n\t"\
03055         "movq " #src1 ", %%mm5          \n\t"\
03056         "punpcklbw %%mm4, %%mm0         \n\t"\
03057         "punpckhbw %%mm4, %%mm5         \n\t"\
03058         "psubw %%mm2, %%mm0             \n\t"\
03059         "psubw %%mm2, %%mm5             \n\t"\
03060         "movq " #src2 ", %%mm1          \n\t"\
03061         "psllw $6, %%mm0                \n\t"\
03062         "psllw $6, %%mm5                \n\t"\
03063         "pmulhw %%mm3, %%mm0            \n\t"\
03064         "movq " #src2 ", %%mm6          \n\t"\
03065         "pmulhw %%mm3, %%mm5            \n\t"\
03066         "punpcklbw %%mm4, %%mm1         \n\t"\
03067         "punpckhbw %%mm4, %%mm6         \n\t"\
03068         "psubw %%mm2, %%mm1             \n\t"\
03069         "psubw %%mm2, %%mm6             \n\t"\
03070         "psllw $6, %%mm1                \n\t"\
03071         "psllw $6, %%mm6                \n\t"\
03072         "pmulhw %%mm3, %%mm1            \n\t"\
03073         "pmulhw %%mm3, %%mm6            \n\t"\
03074         "packuswb %%mm5, %%mm0          \n\t"\
03075         "packuswb %%mm6, %%mm1          \n\t"\
03076         "movq %%mm0, " #dst1 "          \n\t"\
03077         "movq %%mm1, " #dst2 "          \n\t"\
03078 
03079 #endif //HAVE_MMX2
03080 #define SCALED_CPY(src1, src2, dst1, dst2)\
03081    REAL_SCALED_CPY(src1, src2, dst1, dst2)
03082 
03083 SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
03084 SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
03085 SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
03086         "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
03087         "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
03088 SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
03089 
03090 
03091         : "=&a" (packedOffsetAndScale)
03092         : "0" (packedOffsetAndScale),
03093         "r"(src),
03094         "r"(dst),
03095         "r" ((x86_reg)srcStride),
03096         "r" ((x86_reg)dstStride)
03097         : "%"REG_d
03098     );
03099 #else //HAVE_MMX
03100     for(i=0; i<8; i++)
03101         memcpy( &(dst[dstStride*i]),
03102                 &(src[srcStride*i]), BLOCK_SIZE);
03103 #endif //HAVE_MMX
03104     }else{
03105 #if HAVE_MMX
03106     __asm__ volatile(
03107         "lea (%0,%2), %%"REG_a"                 \n\t"
03108         "lea (%1,%3), %%"REG_d"                 \n\t"
03109 
03110 #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \
03111         "movq " #src1 ", %%mm0          \n\t"\
03112         "movq " #src2 ", %%mm1          \n\t"\
03113         "movq %%mm0, " #dst1 "          \n\t"\
03114         "movq %%mm1, " #dst2 "          \n\t"\
03115 
03116 #define SIMPLE_CPY(src1, src2, dst1, dst2)\
03117    REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
03118 
03119 SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))
03120 SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
03121 SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
03122         "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
03123         "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
03124 SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
03125 
03126         : : "r" (src),
03127         "r" (dst),
03128         "r" ((x86_reg)srcStride),
03129         "r" ((x86_reg)dstStride)
03130         : "%"REG_a, "%"REG_d
03131     );
03132 #else //HAVE_MMX
03133     for(i=0; i<8; i++)
03134         memcpy( &(dst[dstStride*i]),
03135                 &(src[srcStride*i]), BLOCK_SIZE);
03136 #endif //HAVE_MMX
03137     }
03138 }
03139 
03143 static inline void RENAME(duplicate)(uint8_t src[], int stride)
03144 {
03145 #if HAVE_MMX
03146     __asm__ volatile(
03147         "movq (%0), %%mm0               \n\t"
03148         "add %1, %0                     \n\t"
03149         "movq %%mm0, (%0)               \n\t"
03150         "movq %%mm0, (%0, %1)           \n\t"
03151         "movq %%mm0, (%0, %1, 2)        \n\t"
03152         : "+r" (src)
03153         : "r" ((x86_reg)-stride)
03154     );
03155 #else
03156     int i;
03157     uint8_t *p=src;
03158     for(i=0; i<3; i++){
03159         p-= stride;
03160         memcpy(p, src, 8);
03161     }
03162 #endif
03163 }
03164 
03168 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
03169                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
03170 {
03171     DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
03172     int x,y;
03173 #ifdef COMPILE_TIME_MODE
03174     const int mode= COMPILE_TIME_MODE;
03175 #else
03176     const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
03177 #endif
03178     int black=0, white=255; // blackest black and whitest white in the picture
03179     int QPCorrecture= 256*256;
03180 
03181     int copyAhead;
03182 #if HAVE_MMX
03183     int i;
03184 #endif
03185 
03186     const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
03187     const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
03188 
03189     //FIXME remove
03190     uint64_t * const yHistogram= c.yHistogram;
03191     uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
03192     uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
03193     //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
03194 
03195 #if HAVE_MMX
03196     for(i=0; i<57; i++){
03197         int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
03198         int threshold= offset*2 + 1;
03199         c.mmxDcOffset[i]= 0x7F - offset;
03200         c.mmxDcThreshold[i]= 0x7F - threshold;
03201         c.mmxDcOffset[i]*= 0x0101010101010101LL;
03202         c.mmxDcThreshold[i]*= 0x0101010101010101LL;
03203     }
03204 #endif
03205 
03206     if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
03207     else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
03208             || (mode & FFMPEG_DEINT_FILTER)
03209             || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
03210     else if(   (mode & V_DEBLOCK)
03211             || (mode & LINEAR_IPOL_DEINT_FILTER)
03212             || (mode & MEDIAN_DEINT_FILTER)
03213             || (mode & V_A_DEBLOCK)) copyAhead=13;
03214     else if(mode & V_X1_FILTER) copyAhead=11;
03215 //    else if(mode & V_RK1_FILTER) copyAhead=10;
03216     else if(mode & DERING) copyAhead=9;
03217     else copyAhead=8;
03218 
03219     copyAhead-= 8;
03220 
03221     if(!isColor){
03222         uint64_t sum= 0;
03223         int i;
03224         uint64_t maxClipped;
03225         uint64_t clipped;
03226         double scale;
03227 
03228         c.frameNum++;
03229         // first frame is fscked so we ignore it
03230         if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
03231 
03232         for(i=0; i<256; i++){
03233             sum+= yHistogram[i];
03234         }
03235 
03236         /* We always get a completely black picture first. */
03237         maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
03238 
03239         clipped= sum;
03240         for(black=255; black>0; black--){
03241             if(clipped < maxClipped) break;
03242             clipped-= yHistogram[black];
03243         }
03244 
03245         clipped= sum;
03246         for(white=0; white<256; white++){
03247             if(clipped < maxClipped) break;
03248             clipped-= yHistogram[white];
03249         }
03250 
03251         scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
03252 
03253 #if HAVE_MMX2
03254         c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
03255         c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
03256 #else
03257         c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
03258         c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
03259 #endif
03260 
03261         c.packedYOffset|= c.packedYOffset<<32;
03262         c.packedYOffset|= c.packedYOffset<<16;
03263 
03264         c.packedYScale|= c.packedYScale<<32;
03265         c.packedYScale|= c.packedYScale<<16;
03266 
03267         if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);
03268         else                        QPCorrecture= 256*256;
03269     }else{
03270         c.packedYScale= 0x0100010001000100LL;
03271         c.packedYOffset= 0;
03272         QPCorrecture= 256*256;
03273     }
03274 
03275     /* copy & deinterlace first row of blocks */
03276     y=-BLOCK_SIZE;
03277     {
03278         const uint8_t *srcBlock= &(src[y*srcStride]);
03279         uint8_t *dstBlock= tempDst + dstStride;
03280 
03281         // From this point on it is guaranteed that we can read and write 16 lines downward
03282         // finish 1 block before the next otherwise we might have a problem
03283         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
03284         for(x=0; x<width; x+=BLOCK_SIZE){
03285 
03286 #if HAVE_MMX2
03287 /*
03288             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
03289             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
03290             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
03291             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
03292 */
03293 
03294             __asm__(
03295                 "mov %4, %%"REG_a"              \n\t"
03296                 "shr $2, %%"REG_a"              \n\t"
03297                 "and $6, %%"REG_a"              \n\t"
03298                 "add %5, %%"REG_a"              \n\t"
03299                 "mov %%"REG_a", %%"REG_d"       \n\t"
03300                 "imul %1, %%"REG_a"             \n\t"
03301                 "imul %3, %%"REG_d"             \n\t"
03302                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03303                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03304                 "add %1, %%"REG_a"              \n\t"
03305                 "add %3, %%"REG_d"              \n\t"
03306                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03307                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03308                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
03309                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
03310                 : "%"REG_a, "%"REG_d
03311             );
03312 
03313 #elif HAVE_AMD3DNOW
03314 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
03315 /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
03316             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
03317             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
03318             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
03319 */
03320 #endif
03321 
03322             RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
03323                               srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
03324 
03325             RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
03326 
03327             if(mode & LINEAR_IPOL_DEINT_FILTER)
03328                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
03329             else if(mode & LINEAR_BLEND_DEINT_FILTER)
03330                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
03331             else if(mode & MEDIAN_DEINT_FILTER)
03332                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
03333             else if(mode & CUBIC_IPOL_DEINT_FILTER)
03334                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
03335             else if(mode & FFMPEG_DEINT_FILTER)
03336                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
03337             else if(mode & LOWPASS5_DEINT_FILTER)
03338                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
03339 /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
03340                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
03341 */
03342             dstBlock+=8;
03343             srcBlock+=8;
03344         }
03345         if(width==FFABS(dstStride))
03346             linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
03347         else{
03348             int i;
03349             for(i=0; i<copyAhead; i++){
03350                 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
03351             }
03352         }
03353     }
03354 
03355     for(y=0; y<height; y+=BLOCK_SIZE){
03356         //1% speedup if these are here instead of the inner loop
03357         const uint8_t *srcBlock= &(src[y*srcStride]);
03358         uint8_t *dstBlock= &(dst[y*dstStride]);
03359 #if HAVE_MMX
03360         uint8_t *tempBlock1= c.tempBlocks;
03361         uint8_t *tempBlock2= c.tempBlocks + 8;
03362 #endif
03363         const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
03364         int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
03365         int QP=0;
03366         /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
03367            if not than use a temporary buffer */
03368         if(y+15 >= height){
03369             int i;
03370             /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
03371                blockcopy to dst later */
03372             linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
03373                     FFMAX(height-y-copyAhead, 0), srcStride);
03374 
03375             /* duplicate last line of src to fill the void upto line (copyAhead+7) */
03376             for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
03377                     memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
03378 
03379             /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
03380             linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
03381 
03382             /* duplicate last line of dst to fill the void upto line (copyAhead) */
03383             for(i=height-y+1; i<=copyAhead; i++)
03384                     memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
03385 
03386             dstBlock= tempDst + dstStride;
03387             srcBlock= tempSrc;
03388         }
03389 
03390         // From this point on it is guaranteed that we can read and write 16 lines downward
03391         // finish 1 block before the next otherwise we might have a problem
03392         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
03393         for(x=0; x<width; x+=BLOCK_SIZE){
03394             const int stride= dstStride;
03395 #if HAVE_MMX
03396             uint8_t *tmpXchg;
03397 #endif
03398             if(isColor){
03399                 QP= QPptr[x>>qpHShift];
03400                 c.nonBQP= nonBQPptr[x>>qpHShift];
03401             }else{
03402                 QP= QPptr[x>>4];
03403                 QP= (QP* QPCorrecture + 256*128)>>16;
03404                 c.nonBQP= nonBQPptr[x>>4];
03405                 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
03406                 yHistogram[ srcBlock[srcStride*12 + 4] ]++;
03407             }
03408             c.QP= QP;
03409 #if HAVE_MMX
03410             __asm__ volatile(
03411                 "movd %1, %%mm7         \n\t"
03412                 "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
03413                 "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
03414                 "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP
03415                 "movq %%mm7, %0         \n\t"
03416                 : "=m" (c.pQPb)
03417                 : "r" (QP)
03418             );
03419 #endif
03420 
03421 
03422 #if HAVE_MMX2
03423 /*
03424             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
03425             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
03426             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
03427             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
03428 */
03429 
03430             __asm__(
03431                 "mov %4, %%"REG_a"              \n\t"
03432                 "shr $2, %%"REG_a"              \n\t"
03433                 "and $6, %%"REG_a"              \n\t"
03434                 "add %5, %%"REG_a"              \n\t"
03435                 "mov %%"REG_a", %%"REG_d"       \n\t"
03436                 "imul %1, %%"REG_a"             \n\t"
03437                 "imul %3, %%"REG_d"             \n\t"
03438                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03439                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03440                 "add %1, %%"REG_a"              \n\t"
03441                 "add %3, %%"REG_d"              \n\t"
03442                 "prefetchnta 32(%%"REG_a", %0)  \n\t"
03443                 "prefetcht0 32(%%"REG_d", %2)   \n\t"
03444                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
03445                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
03446                 : "%"REG_a, "%"REG_d
03447             );
03448 
03449 #elif HAVE_AMD3DNOW
03450 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
03451 /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
03452             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
03453             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
03454             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
03455 */
03456 #endif
03457 
03458             RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
03459                               srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
03460 
03461             if(mode & LINEAR_IPOL_DEINT_FILTER)
03462                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
03463             else if(mode & LINEAR_BLEND_DEINT_FILTER)
03464                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
03465             else if(mode & MEDIAN_DEINT_FILTER)
03466                 RENAME(deInterlaceMedian)(dstBlock, dstStride);
03467             else if(mode & CUBIC_IPOL_DEINT_FILTER)
03468                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
03469             else if(mode & FFMPEG_DEINT_FILTER)
03470                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
03471             else if(mode & LOWPASS5_DEINT_FILTER)
03472                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
03473 /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
03474                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
03475 */
03476 
03477             /* only deblock if we have 2 blocks */
03478             if(y + 8 < height){
03479                 if(mode & V_X1_FILTER)
03480                     RENAME(vertX1Filter)(dstBlock, stride, &c);
03481                 else if(mode & V_DEBLOCK){
03482                     const int t= RENAME(vertClassify)(dstBlock, stride, &c);
03483 
03484                     if(t==1)
03485                         RENAME(doVertLowPass)(dstBlock, stride, &c);
03486                     else if(t==2)
03487                         RENAME(doVertDefFilter)(dstBlock, stride, &c);
03488                 }else if(mode & V_A_DEBLOCK){
03489                     RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
03490                 }
03491             }
03492 
03493 #if HAVE_MMX
03494             RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
03495 #endif
03496             /* check if we have a previous block to deblock it with dstBlock */
03497             if(x - 8 >= 0){
03498 #if HAVE_MMX
03499                 if(mode & H_X1_FILTER)
03500                         RENAME(vertX1Filter)(tempBlock1, 16, &c);
03501                 else if(mode & H_DEBLOCK){
03502 //START_TIMER
03503                     const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
03504 //STOP_TIMER("dc & minmax")
03505                     if(t==1)
03506                         RENAME(doVertLowPass)(tempBlock1, 16, &c);
03507                     else if(t==2)
03508                         RENAME(doVertDefFilter)(tempBlock1, 16, &c);
03509                 }else if(mode & H_A_DEBLOCK){
03510                         RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
03511                 }
03512 
03513                 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
03514 
03515 #else
03516                 if(mode & H_X1_FILTER)
03517                     horizX1Filter(dstBlock-4, stride, QP);
03518                 else if(mode & H_DEBLOCK){
03519 #if HAVE_ALTIVEC
03520                     DECLARE_ALIGNED(16, unsigned char, tempBlock)[272];
03521                     transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
03522 
03523                     const int t=vertClassify_altivec(tempBlock-48, 16, &c);
03524                     if(t==1) {
03525                         doVertLowPass_altivec(tempBlock-48, 16, &c);
03526                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
03527                     }
03528                     else if(t==2) {
03529                         doVertDefFilter_altivec(tempBlock-48, 16, &c);
03530                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
03531                     }
03532 #else
03533                     const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
03534 
03535                     if(t==1)
03536                         RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
03537                     else if(t==2)
03538                         RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
03539 #endif
03540                 }else if(mode & H_A_DEBLOCK){
03541                     RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
03542                 }
03543 #endif //HAVE_MMX
03544                 if(mode & DERING){
03545                 //FIXME filter first line
03546                     if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
03547                 }
03548 
03549                 if(mode & TEMP_NOISE_FILTER)
03550                 {
03551                     RENAME(tempNoiseReducer)(dstBlock-8, stride,
03552                             c.tempBlurred[isColor] + y*dstStride + x,
03553                             c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
03554                             c.ppMode.maxTmpNoise);
03555                 }
03556             }
03557 
03558             dstBlock+=8;
03559             srcBlock+=8;
03560 
03561 #if HAVE_MMX
03562             tmpXchg= tempBlock1;
03563             tempBlock1= tempBlock2;
03564             tempBlock2 = tmpXchg;
03565 #endif
03566         }
03567 
03568         if(mode & DERING){
03569             if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
03570         }
03571 
03572         if((mode & TEMP_NOISE_FILTER)){
03573             RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
03574                     c.tempBlurred[isColor] + y*dstStride + x,
03575                     c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
03576                     c.ppMode.maxTmpNoise);
03577         }
03578 
03579         /* did we use a tmp buffer for the last lines*/
03580         if(y+15 >= height){
03581             uint8_t *dstBlock= &(dst[y*dstStride]);
03582             if(width==FFABS(dstStride))
03583                 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
03584             else{
03585                 int i;
03586                 for(i=0; i<height-y; i++){
03587                     memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
03588                 }
03589             }
03590         }
03591 /*
03592         for(x=0; x<width; x+=32){
03593             volatile int i;
03594             i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
03595                 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
03596                 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
03597                 + dstBlock[x +13*dstStride]
03598                 + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
03599         }*/
03600     }
03601 #if   HAVE_AMD3DNOW
03602     __asm__ volatile("femms");
03603 #elif HAVE_MMX
03604     __asm__ volatile("emms");
03605 #endif
03606 
03607 #ifdef DEBUG_BRIGHTNESS
03608     if(!isColor){
03609         int max=1;
03610         int i;
03611         for(i=0; i<256; i++)
03612             if(yHistogram[i] > max) max=yHistogram[i];
03613 
03614         for(i=1; i<256; i++){
03615             int x;
03616             int start=yHistogram[i-1]/(max/256+1);
03617             int end=yHistogram[i]/(max/256+1);
03618             int inc= end > start ? 1 : -1;
03619             for(x=start; x!=end+inc; x+=inc)
03620                 dst[ i*dstStride + x]+=128;
03621         }
03622 
03623         for(i=0; i<100; i+=2){
03624             dst[ (white)*dstStride + i]+=128;
03625             dst[ (black)*dstStride + i]+=128;
03626         }
03627     }
03628 #endif
03629 
03630     *c2= c; //copy local context back
03631 
03632 }