postprocess_template.c 161 KB
Newer Older
1
/*
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
 * Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20

Michael Niedermayer's avatar
Michael Niedermayer committed
21
/**
22
 * @file
Michael Niedermayer's avatar
Michael Niedermayer committed
23 24 25
 * mmx/mmx2/3dnow postprocess code.
 */

26
#include "libavutil/x86/asm.h"
Michael Niedermayer's avatar
Michael Niedermayer committed
27

28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
/* A single TEMPLATE_PP_* should be defined (to 1) when this template is
 * included. The following macros will define its dependencies to 1 as well
 * (like MMX2 depending on MMX), and will define to 0 all the others. Every
 * TEMPLATE_PP_* need to be undef at the end. */

#ifdef TEMPLATE_PP_C
#   define RENAME(a) a ## _C
#else
#   define TEMPLATE_PP_C 0
#endif

#ifdef TEMPLATE_PP_ALTIVEC
#   define RENAME(a) a ## _altivec
#else
#   define TEMPLATE_PP_ALTIVEC 0
#endif

#ifdef TEMPLATE_PP_MMX
#   define RENAME(a) a ## _MMX
#else
#   define TEMPLATE_PP_MMX 0
#endif

#ifdef TEMPLATE_PP_MMXEXT
#   undef  TEMPLATE_PP_MMX
#   define TEMPLATE_PP_MMX 1
#   define RENAME(a) a ## _MMX2
#else
#   define TEMPLATE_PP_MMXEXT 0
#endif

#ifdef TEMPLATE_PP_3DNOW
#   undef  TEMPLATE_PP_MMX
#   define TEMPLATE_PP_MMX 1
#   define RENAME(a) a ## _3DNow
#else
#   define TEMPLATE_PP_3DNOW 0
#endif

67 68 69 70 71 72 73 74 75 76
#ifdef TEMPLATE_PP_SSE2
#   undef  TEMPLATE_PP_MMX
#   define TEMPLATE_PP_MMX 1
#   undef  TEMPLATE_PP_MMXEXT
#   define TEMPLATE_PP_MMXEXT 1
#   define RENAME(a) a ## _SSE2
#else
#   define TEMPLATE_PP_SSE2 0
#endif

77
#undef REAL_PAVGB
78 79 80
#undef PAVGB
#undef PMINUB
#undef PMAXUB
Arpi's avatar
Arpi committed
81

82
#if   TEMPLATE_PP_MMXEXT
83
#define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
84
#elif TEMPLATE_PP_3DNOW
D Richard Felker III's avatar
D Richard Felker III committed
85
#define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
Arpi's avatar
Arpi committed
86
#endif
87
#define PAVGB(a,b)  REAL_PAVGB(a,b)
88

89
#if   TEMPLATE_PP_MMXEXT
Michael Niedermayer's avatar
Michael Niedermayer committed
90
#define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
91
#elif TEMPLATE_PP_MMX
Michael Niedermayer's avatar
Michael Niedermayer committed
92
#define PMINUB(b,a,t) \
93 94 95
    "movq " #a ", " #t " \n\t"\
    "psubusb " #b ", " #t " \n\t"\
    "psubb " #t ", " #a " \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
96 97
#endif

98
#if   TEMPLATE_PP_MMXEXT
Michael Niedermayer's avatar
Michael Niedermayer committed
99
#define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
100
#elif TEMPLATE_PP_MMX
Michael Niedermayer's avatar
Michael Niedermayer committed
101
#define PMAXUB(a,b) \
102 103
    "psubusb " #a ", " #b " \n\t"\
    "paddb " #a ", " #b " \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
104 105
#endif

Diego Biurrun's avatar
Diego Biurrun committed
106
//FIXME? |255-0| = 1 (should not be a problem ...)
107
#if TEMPLATE_PP_MMX
108
/**
109
 * Check if the middle 8x8 Block in the given 8x16 block is flat
110
 */
111
static inline int RENAME(vertClassify)(const uint8_t src[], int stride, PPContext *c){
112 113
    int numEq= 0, dcOk;
    src+= stride*4; // src points to begin of the 8x8 Block
114
    __asm__ volatile(
115 116 117 118 119
        "movq %0, %%mm7                         \n\t"
        "movq %1, %%mm6                         \n\t"
        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
        );

120
    __asm__ volatile(
121
        "lea (%2, %3), %%"REG_a"                \n\t"
122 123 124
//      0       1       2       3       4       5       6       7       8       9
//      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2

125 126 127 128 129 130
        "movq (%2), %%mm0                       \n\t"
        "movq (%%"REG_a"), %%mm1                \n\t"
        "movq %%mm0, %%mm3                      \n\t"
        "movq %%mm0, %%mm4                      \n\t"
        PMAXUB(%%mm1, %%mm4)
        PMINUB(%%mm1, %%mm3, %%mm5)
131
        "psubb %%mm1, %%mm0                     \n\t" // mm0 = difference
132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
        "paddb %%mm7, %%mm0                     \n\t"
        "pcmpgtb %%mm6, %%mm0                   \n\t"

        "movq (%%"REG_a",%3), %%mm2             \n\t"
        PMAXUB(%%mm2, %%mm4)
        PMINUB(%%mm2, %%mm3, %%mm5)
        "psubb %%mm2, %%mm1                     \n\t"
        "paddb %%mm7, %%mm1                     \n\t"
        "pcmpgtb %%mm6, %%mm1                   \n\t"
        "paddb %%mm1, %%mm0                     \n\t"

        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
        PMAXUB(%%mm1, %%mm4)
        PMINUB(%%mm1, %%mm3, %%mm5)
        "psubb %%mm1, %%mm2                     \n\t"
        "paddb %%mm7, %%mm2                     \n\t"
        "pcmpgtb %%mm6, %%mm2                   \n\t"
        "paddb %%mm2, %%mm0                     \n\t"

        "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"

        "movq (%2, %3, 4), %%mm2                \n\t"
        PMAXUB(%%mm2, %%mm4)
        PMINUB(%%mm2, %%mm3, %%mm5)
        "psubb %%mm2, %%mm1                     \n\t"
        "paddb %%mm7, %%mm1                     \n\t"
        "pcmpgtb %%mm6, %%mm1                   \n\t"
        "paddb %%mm1, %%mm0                     \n\t"

        "movq (%%"REG_a"), %%mm1                \n\t"
        PMAXUB(%%mm1, %%mm4)
        PMINUB(%%mm1, %%mm3, %%mm5)
        "psubb %%mm1, %%mm2                     \n\t"
        "paddb %%mm7, %%mm2                     \n\t"
        "pcmpgtb %%mm6, %%mm2                   \n\t"
        "paddb %%mm2, %%mm0                     \n\t"

        "movq (%%"REG_a", %3), %%mm2            \n\t"
        PMAXUB(%%mm2, %%mm4)
        PMINUB(%%mm2, %%mm3, %%mm5)
        "psubb %%mm2, %%mm1                     \n\t"
        "paddb %%mm7, %%mm1                     \n\t"
        "pcmpgtb %%mm6, %%mm1                   \n\t"
        "paddb %%mm1, %%mm0                     \n\t"

        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
        PMAXUB(%%mm1, %%mm4)
        PMINUB(%%mm1, %%mm3, %%mm5)
        "psubb %%mm1, %%mm2                     \n\t"
        "paddb %%mm7, %%mm2                     \n\t"
        "pcmpgtb %%mm6, %%mm2                   \n\t"
        "paddb %%mm2, %%mm0                     \n\t"
        "psubusb %%mm3, %%mm4                   \n\t"

        "                                       \n\t"
187
#if TEMPLATE_PP_MMXEXT
188 189
        "pxor %%mm7, %%mm7                      \n\t"
        "psadbw %%mm7, %%mm0                    \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
190
#else
191 192 193 194 195 196 197 198 199
        "movq %%mm0, %%mm1                      \n\t"
        "psrlw $8, %%mm0                        \n\t"
        "paddb %%mm1, %%mm0                     \n\t"
        "movq %%mm0, %%mm1                      \n\t"
        "psrlq $16, %%mm0                       \n\t"
        "paddb %%mm1, %%mm0                     \n\t"
        "movq %%mm0, %%mm1                      \n\t"
        "psrlq $32, %%mm0                       \n\t"
        "paddb %%mm1, %%mm0                     \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
200
#endif
201 202 203 204 205 206 207 208
        "movq %4, %%mm7                         \n\t" // QP,..., QP
        "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
        "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
        "packssdw %%mm4, %%mm4                  \n\t"
        "movd %%mm0, %0                         \n\t"
        "movd %%mm4, %1                         \n\t"

        : "=r" (numEq), "=r" (dcOk)
209
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
210 211 212 213 214 215 216 217 218 219
        : "%"REG_a
        );

    numEq= (-numEq) &0xFF;
    if(numEq > c->ppMode.flatnessThreshold){
        if(dcOk) return 0;
        else     return 1;
    }else{
        return 2;
    }
220
}
221
#endif //TEMPLATE_PP_MMX
222 223

/**
224
 * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
Michael Niedermayer's avatar
Michael Niedermayer committed
225
 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
226
 */
227
#if !TEMPLATE_PP_ALTIVEC
228
static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
229
{
230
#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
231
    src+= stride*3;
232
    __asm__ volatile(        //"movv %0 %1 %2\n\t"
233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
        "movq %2, %%mm0                         \n\t"  // QP,..., QP
        "pxor %%mm4, %%mm4                      \n\t"

        "movq (%0), %%mm6                       \n\t"
        "movq (%0, %1), %%mm5                   \n\t"
        "movq %%mm5, %%mm1                      \n\t"
        "movq %%mm6, %%mm2                      \n\t"
        "psubusb %%mm6, %%mm5                   \n\t"
        "psubusb %%mm1, %%mm2                   \n\t"
        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF

        "pand %%mm2, %%mm6                      \n\t"
        "pandn %%mm1, %%mm2                     \n\t"
        "por %%mm2, %%mm6                       \n\t"// First Line to Filter

        "movq (%0, %1, 8), %%mm5                \n\t"
        "lea (%0, %1, 4), %%"REG_a"             \n\t"
        "lea (%0, %1, 8), %%"REG_c"             \n\t"
        "sub %1, %%"REG_c"                      \n\t"
        "add %1, %0                             \n\t" // %0 points to line 1 not 0
        "movq (%0, %1, 8), %%mm7                \n\t"
        "movq %%mm5, %%mm1                      \n\t"
        "movq %%mm7, %%mm2                      \n\t"
        "psubusb %%mm7, %%mm5                   \n\t"
        "psubusb %%mm1, %%mm2                   \n\t"
        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF

        "pand %%mm2, %%mm7                      \n\t"
        "pandn %%mm1, %%mm2                     \n\t"
        "por %%mm2, %%mm7                       \n\t" // First Line to Filter


        //      1       2       3       4       5       6       7       8
        //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
        // 6 4 2 2 1 1
        // 6 4 4 2
        // 6 8 2

        "movq (%0, %1), %%mm0                   \n\t" //  1
        "movq %%mm0, %%mm1                      \n\t" //  1
        PAVGB(%%mm6, %%mm0)                           //1 1        /2
        PAVGB(%%mm6, %%mm0)                           //3 1        /4

        "movq (%0, %1, 4), %%mm2                \n\t" //     1
        "movq %%mm2, %%mm5                      \n\t" //     1
        PAVGB((%%REGa), %%mm2)                        //    11        /2
        PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
        "movq %%mm2, %%mm3                      \n\t" //   211        /4
        "movq (%0), %%mm4                       \n\t" // 1
        PAVGB(%%mm4, %%mm3)                           // 4 211        /8
        PAVGB(%%mm0, %%mm3)                           //642211        /16
        "movq %%mm3, (%0)                       \n\t" // X
        // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
        "movq %%mm1, %%mm0                      \n\t" //  1
        PAVGB(%%mm6, %%mm0)                           //1 1        /2
        "movq %%mm4, %%mm3                      \n\t" // 1
        PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
        PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
        PAVGB((%%REGa), %%mm5)                        //    211 /4
        PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
        PAVGB(%%mm0, %%mm3)                           //4242211 /16
        "movq %%mm3, (%0,%1)                    \n\t" //  X
        // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
        PAVGB(%%mm4, %%mm6)                                   //11        /2
        "movq (%%"REG_c"), %%mm0                \n\t" //       1
        PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
        "movq %%mm0, %%mm3                      \n\t" //      11/2
        PAVGB(%%mm1, %%mm0)                           //  2   11/4
        PAVGB(%%mm6, %%mm0)                           //222   11/8
        PAVGB(%%mm2, %%mm0)                           //22242211/16
        "movq (%0, %1, 2), %%mm2                \n\t" //   1
        "movq %%mm0, (%0, %1, 2)                \n\t" //   X
        // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
        PAVGB((%%REGc), %%mm0)                        //       11        /2
        PAVGB(%%mm0, %%mm6)                           //11     11        /4
        PAVGB(%%mm1, %%mm4)                           // 11                /2
        PAVGB(%%mm2, %%mm1)                           //  11                /2
        PAVGB(%%mm1, %%mm6)                           //1122   11        /8
        PAVGB(%%mm5, %%mm6)                           //112242211        /16
        "movq (%%"REG_a"), %%mm5                \n\t" //    1
        "movq %%mm6, (%%"REG_a")                \n\t" //    X
        // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
        "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
        PAVGB(%%mm7, %%mm6)                           //        11        /2
        PAVGB(%%mm4, %%mm6)                           // 11     11        /4
        PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
        PAVGB(%%mm5, %%mm2)                           //   11                /2
        "movq (%0, %1, 4), %%mm4                \n\t" //     1
        PAVGB(%%mm4, %%mm2)                           //   112                /4
        PAVGB(%%mm2, %%mm6)                           // 112242211        /16
        "movq %%mm6, (%0, %1, 4)                \n\t" //     X
        // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
        PAVGB(%%mm7, %%mm1)                           //  11     2        /4
        PAVGB(%%mm4, %%mm5)                           //    11                /2
        PAVGB(%%mm5, %%mm0)                           //    11 11        /4
        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
        PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
        PAVGB(%%mm0, %%mm1)                           //  11224222        /16
        "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
        // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
        PAVGB((%%REGc), %%mm2)                        //   112 4        /8
        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
        PAVGB(%%mm0, %%mm6)                           //      1 1        /2
        PAVGB(%%mm7, %%mm6)                           //      1 12        /4
        PAVGB(%%mm2, %%mm6)                           //   1122424        /4
        "movq %%mm6, (%%"REG_c")                \n\t" //       X
        // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
        PAVGB(%%mm7, %%mm5)                           //    11   2        /4
        PAVGB(%%mm7, %%mm5)                           //    11   6        /8

        PAVGB(%%mm3, %%mm0)                           //      112        /4
        PAVGB(%%mm0, %%mm5)                           //    112246        /16
        "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
        "sub %1, %0                             \n\t"

        :
354
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
355 356
        : "%"REG_a, "%"REG_c
    );
357
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
    const int l1= stride;
    const int l2= stride + l1;
    const int l3= stride + l2;
    const int l4= stride + l3;
    const int l5= stride + l4;
    const int l6= stride + l5;
    const int l7= stride + l6;
    const int l8= stride + l7;
    const int l9= stride + l8;
    int x;
    src+= stride*3;
    for(x=0; x<BLOCK_SIZE; x++){
        const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
        const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];

        int sums[10];
        sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
        sums[1] = sums[0] - first  + src[l4];
        sums[2] = sums[1] - first  + src[l5];
        sums[3] = sums[2] - first  + src[l6];
        sums[4] = sums[3] - first  + src[l7];
        sums[5] = sums[4] - src[l1] + src[l8];
        sums[6] = sums[5] - src[l2] + last;
        sums[7] = sums[6] - src[l3] + last;
        sums[8] = sums[7] - src[l4] + last;
        sums[9] = sums[8] - src[l5] + last;

        src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
        src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
        src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
        src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
        src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
        src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
        src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
        src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;

        src++;
    }
396
#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
397
}
398
#endif //TEMPLATE_PP_ALTIVEC
399

400 401
/**
 * Experimental Filter 1
402
 * will not damage linear gradients
Diego Biurrun's avatar
Diego Biurrun committed
403
 * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
Diego Biurrun's avatar
Diego Biurrun committed
404 405
 * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
 * MMX2 version does correct clipping C version does not
406
 */
407
static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
408
{
409
#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
410
    src+= stride*3;
411

412
    __asm__ volatile(
413 414 415
        "pxor %%mm7, %%mm7                      \n\t" // 0
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
416 417
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491
        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
        "movq (%0, %1, 4), %%mm1                \n\t" // line 4
        "movq %%mm1, %%mm2                      \n\t" // line 4
        "psubusb %%mm0, %%mm1                   \n\t"
        "psubusb %%mm2, %%mm0                   \n\t"
        "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
        "movq (%%"REG_c"), %%mm3                \n\t" // line 5
        "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
        "movq %%mm3, %%mm5                      \n\t" // line 5
        "psubusb %%mm4, %%mm3                   \n\t"
        "psubusb %%mm5, %%mm4                   \n\t"
        "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
        PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
        "movq %%mm2, %%mm1                      \n\t" // line 4
        "psubusb %%mm5, %%mm2                   \n\t"
        "movq %%mm2, %%mm4                      \n\t"
        "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
        "psubusb %%mm1, %%mm5                   \n\t"
        "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
        "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
        "movq %%mm4, %%mm3                      \n\t" // d
        "movq %2, %%mm0                         \n\t"
        "paddusb %%mm0, %%mm0                   \n\t"
        "psubusb %%mm0, %%mm4                   \n\t"
        "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
        "psubusb "MANGLE(b01)", %%mm3           \n\t"
        "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0

        PAVGB(%%mm7, %%mm3)                           // d/2
        "movq %%mm3, %%mm1                      \n\t" // d/2
        PAVGB(%%mm7, %%mm3)                           // d/4
        PAVGB(%%mm1, %%mm3)                           // 3*d/8

        "movq (%0, %1, 4), %%mm0                \n\t" // line 4
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
        "psubusb %%mm3, %%mm0                   \n\t"
        "pxor %%mm2, %%mm0                      \n\t"
        "movq %%mm0, (%0, %1, 4)                \n\t" // line 4

        "movq (%%"REG_c"), %%mm0                \n\t" // line 5
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
        "paddusb %%mm3, %%mm0                   \n\t"
        "pxor %%mm2, %%mm0                      \n\t"
        "movq %%mm0, (%%"REG_c")                \n\t" // line 5

        PAVGB(%%mm7, %%mm1)                           // d/4

        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
        "psubusb %%mm1, %%mm0                   \n\t"
        "pxor %%mm2, %%mm0                      \n\t"
        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3

        "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
        "paddusb %%mm1, %%mm0                   \n\t"
        "pxor %%mm2, %%mm0                      \n\t"
        "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6

        PAVGB(%%mm7, %%mm1)                           // d/8

        "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
        "psubusb %%mm1, %%mm0                   \n\t"
        "pxor %%mm2, %%mm0                      \n\t"
        "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2

        "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
        "paddusb %%mm1, %%mm0                   \n\t"
        "pxor %%mm2, %%mm0                      \n\t"
        "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7

        :
492
        : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
493
          NAMED_CONSTRAINTS_ADD(b01)
494 495
        : "%"REG_a, "%"REG_c
    );
496
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
497

498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526
    const int l1= stride;
    const int l2= stride + l1;
    const int l3= stride + l2;
    const int l4= stride + l3;
    const int l5= stride + l4;
    const int l6= stride + l5;
    const int l7= stride + l6;
//    const int l8= stride + l7;
//    const int l9= stride + l8;
    int x;

    src+= stride*3;
    for(x=0; x<BLOCK_SIZE; x++){
        int a= src[l3] - src[l4];
        int b= src[l4] - src[l5];
        int c= src[l5] - src[l6];

        int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
        d= FFMAX(d, 0);

        if(d < co->QP*2){
            int v = d * FFSIGN(-b);

            src[l2] +=v>>3;
            src[l3] +=v>>2;
            src[l4] +=(3*v)>>3;
            src[l5] -=(3*v)>>3;
            src[l6] -=v>>2;
            src[l7] -=v>>3;
527
        }
528 529
        src++;
    }
530
#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
531 532
}

533
#if !TEMPLATE_PP_ALTIVEC
534
static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
535
{
536
#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
537
/*
538 539 540 541 542 543 544 545 546 547 548 549
    uint8_t tmp[16];
    const int l1= stride;
    const int l2= stride + l1;
    const int l3= stride + l2;
    const int l4= (int)tmp - (int)src - stride*3;
    const int l5= (int)tmp - (int)src - stride*3 + 8;
    const int l6= stride*3 + l3;
    const int l7= stride + l6;
    const int l8= stride + l7;

    memcpy(tmp, src+stride*7, 8);
    memcpy(tmp+8, src+stride*8, 8);
550
*/
551
    src+= stride*4;
552
    __asm__ volatile(
553

Diego Biurrun's avatar
Diego Biurrun committed
554
#if 0 //slightly more accurate and slightly slower
555 556 557
        "pxor %%mm7, %%mm7                      \n\t" // 0
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
558 559 560 561 562
//      0       1       2       3       4       5       6       7
//      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1


563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
        "movq (%0, %1, 2), %%mm0                \n\t" // l2
        "movq (%0), %%mm1                       \n\t" // l0
        "movq %%mm0, %%mm2                      \n\t" // l2
        PAVGB(%%mm7, %%mm0)                           // ~l2/2
        PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
        PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8

        "movq (%%"REG_a"), %%mm1                \n\t" // l1
        "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
        "movq %%mm1, %%mm4                      \n\t" // l1
        PAVGB(%%mm7, %%mm1)                           // ~l1/2
        PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
        PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8

        "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
        "psubusb %%mm1, %%mm0                   \n\t"
        "psubusb %%mm4, %%mm1                   \n\t"
        "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
581 582
// mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0

583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599
        "movq (%0, %1, 4), %%mm0                \n\t" // l4
        "movq %%mm0, %%mm4                      \n\t" // l4
        PAVGB(%%mm7, %%mm0)                           // ~l4/2
        PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
        PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8

        "movq (%%"REG_c"), %%mm2                \n\t" // l5
        "movq %%mm3, %%mm5                      \n\t" // l3
        PAVGB(%%mm7, %%mm3)                           // ~l3/2
        PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
        PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8

        "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
        "psubusb %%mm3, %%mm0                   \n\t"
        "psubusb %%mm6, %%mm3                   \n\t"
        "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
        "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
600 601
// mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0

602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617
        "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
        "movq %%mm6, %%mm5                      \n\t" // l6
        PAVGB(%%mm7, %%mm6)                           // ~l6/2
        PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
        PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8

        "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
        "movq %%mm2, %%mm4                      \n\t" // l5
        PAVGB(%%mm7, %%mm2)                           // ~l5/2
        PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
        PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8

        "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
        "psubusb %%mm2, %%mm6                   \n\t"
        "psubusb %%mm4, %%mm2                   \n\t"
        "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
618 619 620
// mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0


621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658
        PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
        "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
        "paddusb "MANGLE(b01)", %%mm4           \n\t"
        "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
        "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
        "pand %%mm4, %%mm3                      \n\t"

        "movq %%mm3, %%mm1                      \n\t"
//        "psubusb "MANGLE(b01)", %%mm3           \n\t"
        PAVGB(%%mm7, %%mm3)
        PAVGB(%%mm7, %%mm3)
        "paddusb %%mm1, %%mm3                   \n\t"
//        "paddusb "MANGLE(b01)", %%mm3           \n\t"

        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
        "movq (%0, %1, 4), %%mm5                \n\t" //l4
        "movq (%0, %1, 4), %%mm4                \n\t" //l4
        "psubusb %%mm6, %%mm5                   \n\t"
        "psubusb %%mm4, %%mm6                   \n\t"
        "por %%mm6, %%mm5                       \n\t" // |l3-l4|
        "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
        "pxor %%mm6, %%mm0                      \n\t"
        "pand %%mm0, %%mm3                      \n\t"
        PMINUB(%%mm5, %%mm3, %%mm0)

        "psubusb "MANGLE(b01)", %%mm3           \n\t"
        PAVGB(%%mm7, %%mm3)

        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
        "movq (%0, %1, 4), %%mm2                \n\t"
        "pxor %%mm6, %%mm0                      \n\t"
        "pxor %%mm6, %%mm2                      \n\t"
        "psubb %%mm3, %%mm0                     \n\t"
        "paddb %%mm3, %%mm2                     \n\t"
        "pxor %%mm6, %%mm0                      \n\t"
        "pxor %%mm6, %%mm2                      \n\t"
        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
        "movq %%mm2, (%0, %1, 4)                \n\t"
659
#endif //0
660

661 662
        "lea (%0, %1), %%"REG_a"                \n\t"
        "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
663 664 665
//      0       1       2       3       4       5       6       7
//      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
666 667


668 669 670 671
        "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
        "movq (%0, %1, 4), %%mm0                \n\t" // l4
        "pxor %%mm6, %%mm1                      \n\t" // -l3-1
        PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
672 673
// mm1=-l3-1, mm0=128-q

674 675 676 677 678 679 680 681 682 683
        "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
        "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
        "pxor %%mm6, %%mm2                      \n\t" // -l5-1
        "movq %%mm2, %%mm5                      \n\t" // -l5-1
        "movq "MANGLE(b80)", %%mm4              \n\t" // 128
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
        PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
        PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
        PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
        PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
684 685
// mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1

686 687 688 689 690 691 692 693
        "movq (%%"REG_a"), %%mm2                \n\t" // l1
        "pxor %%mm6, %%mm2                      \n\t" // -l1-1
        PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
        PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
        "movq "MANGLE(b80)", %%mm3              \n\t" // 128
        PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
        PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
        PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
694 695
// mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1

696 697 698 699 700 701 702 703
        PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
        "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
        "pxor %%mm6, %%mm1                      \n\t" // -l7-1
        PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
        "movq "MANGLE(b80)", %%mm2              \n\t" // 128
        PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
        PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
        PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
704 705
// mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128

706 707 708 709 710 711 712
        "movq "MANGLE(b00)", %%mm1              \n\t" // 0
        "movq "MANGLE(b00)", %%mm5              \n\t" // 0
        "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
        "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
        PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
        PMAXUB(%%mm5, %%mm3)                          // 128 + |lenergy/16|
        PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
713 714 715

// mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128

716 717 718 719 720 721 722 723 724 725 726
        "movq "MANGLE(b00)", %%mm7              \n\t" // 0
        "movq %2, %%mm2                         \n\t" // QP
        PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
        "psubb %%mm6, %%mm2                     \n\t"

        "movq %%mm4, %%mm1                      \n\t"
        "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
        "pxor %%mm1, %%mm4                      \n\t"
        "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
        "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
        "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
727 728
// mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16

729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757
        "movq %%mm4, %%mm3                      \n\t" // d
        "psubusb "MANGLE(b01)", %%mm4           \n\t"
        PAVGB(%%mm7, %%mm4)                           // d/32
        PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
        "paddb %%mm3, %%mm4                     \n\t" // 5d/64
        "pand %%mm2, %%mm4                      \n\t"

        "movq "MANGLE(b80)", %%mm5              \n\t" // 128
        "psubb %%mm0, %%mm5                     \n\t" // q
        "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
        "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
        "pxor %%mm7, %%mm5                      \n\t"

        PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
        "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)

        "pand %%mm7, %%mm4                      \n\t"
        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
        "movq (%0, %1, 4), %%mm2                \n\t"
        "pxor %%mm1, %%mm0                      \n\t"
        "pxor %%mm1, %%mm2                      \n\t"
        "paddb %%mm4, %%mm0                     \n\t"
        "psubb %%mm4, %%mm2                     \n\t"
        "pxor %%mm1, %%mm0                      \n\t"
        "pxor %%mm1, %%mm2                      \n\t"
        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
        "movq %%mm2, (%0, %1, 4)                \n\t"

        :
758
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
759
          NAMED_CONSTRAINTS_ADD(b80,b00,b01)
760 761
        : "%"REG_a, "%"REG_c
    );
762 763

/*
764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789
    {
    int x;
    src-= stride;
    for(x=0; x<BLOCK_SIZE; x++){
        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
        if(FFABS(middleEnergy)< 8*QP){
            const int q=(src[l4] - src[l5])/2;
            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
            d= FFMAX(d, 0);

            d= (5*d + 32) >> 6;
            d*= FFSIGN(-middleEnergy);

            if(q>0){
                d= d<0 ? 0 : d;
                d= d>q ? q : d;
            }else{
                d= d>0 ? 0 : d;
                d= d<q ? q : d;
            }

            src[l4]-= d;
            src[l5]+= d;
790
        }
791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813
        src++;
    }
    src-=8;
    for(x=0; x<8; x++){
        int y;
        for(y=4; y<6; y++){
            int d= src[x+y*stride] - tmp[x+(y-4)*8];
            int ad= FFABS(d);
            static int max=0;
            static int sum=0;
            static int num=0;
            static int bias=0;

            if(max<ad) max=ad;
            sum+= ad>3 ? 1 : 0;
            if(ad>3){
                src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
            }
            if(y==4) bias+=d;
            num++;
            if(num%1000000 == 0){
                av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
            }
814
        }
815
    }
816 817
}
*/
818
#elif TEMPLATE_PP_MMX
819
    DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars
820
    src+= stride*4;
821
    __asm__ volatile(
822
        "pxor %%mm7, %%mm7                      \n\t"
823 824 825 826
//      0       1       2       3       4       5       6       7
//      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1

827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863
        "movq (%0), %%mm0                       \n\t"
        "movq %%mm0, %%mm1                      \n\t"
        "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
        "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0

        "movq (%0, %1), %%mm2                   \n\t"
        "lea (%0, %1, 2), %%"REG_a"             \n\t"
        "movq %%mm2, %%mm3                      \n\t"
        "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
        "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1

        "movq (%%"REG_a"), %%mm4                \n\t"
        "movq %%mm4, %%mm5                      \n\t"
        "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
        "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2

        "paddw %%mm0, %%mm0                     \n\t" // 2L0
        "paddw %%mm1, %%mm1                     \n\t" // 2H0
        "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
        "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2

        "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
        "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2

        "movq (%%"REG_a", %1), %%mm2            \n\t"
        "movq %%mm2, %%mm3                      \n\t"
        "punpcklbw %%mm7, %%mm2                 \n\t" // L3
        "punpckhbw %%mm7, %%mm3                 \n\t" // H3

        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
864 865
        "movq %%mm0, (%3)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
        "movq %%mm1, 8(%3)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
866 867 868 869 870 871 872 873

        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
        "movq %%mm0, %%mm1                      \n\t"
        "punpcklbw %%mm7, %%mm0                 \n\t" // L4
        "punpckhbw %%mm7, %%mm1                 \n\t" // H4

        "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
        "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
874 875
        "movq %%mm2, 16(%3)                     \n\t" // L3 - L4
        "movq %%mm3, 24(%3)                     \n\t" // H3 - H4
876 877 878 879 880 881 882 883 884 885
        "paddw %%mm4, %%mm4                     \n\t" // 2L2
        "paddw %%mm5, %%mm5                     \n\t" // 2H2
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4

        "lea (%%"REG_a", %1), %0                \n\t"
        "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
        "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
886
//50 opcodes so far
887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922
        "movq (%0, %1, 2), %%mm2                \n\t"
        "movq %%mm2, %%mm3                      \n\t"
        "punpcklbw %%mm7, %%mm2                 \n\t" // L5
        "punpckhbw %%mm7, %%mm3                 \n\t" // H5
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5

        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
        "punpcklbw %%mm7, %%mm6                 \n\t" // L6
        "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
        "punpckhbw %%mm7, %%mm6                 \n\t" // H6
        "psubw %%mm6, %%mm3                     \n\t" // H5 - H6

        "paddw %%mm0, %%mm0                     \n\t" // 2L4
        "paddw %%mm1, %%mm1                     \n\t" // 2H4
        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6

        "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
        "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6

        "movq (%0, %1, 4), %%mm2                \n\t"
        "movq %%mm2, %%mm3                      \n\t"
        "punpcklbw %%mm7, %%mm2                 \n\t" // L7
        "punpckhbw %%mm7, %%mm3                 \n\t" // H7

        "paddw %%mm2, %%mm2                     \n\t" // 2L7
        "paddw %%mm3, %%mm3                     \n\t" // 2H7
        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7

923 924
        "movq (%3), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
        "movq 8(%3), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
Michael Niedermayer's avatar
Michael Niedermayer committed
925

926
#if TEMPLATE_PP_MMXEXT
927 928 929 930 931 932 933 934 935 936 937 938
        "movq %%mm7, %%mm6                      \n\t" // 0
        "psubw %%mm0, %%mm6                     \n\t"
        "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
        "movq %%mm7, %%mm6                      \n\t" // 0
        "psubw %%mm1, %%mm6                     \n\t"
        "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
        "movq %%mm7, %%mm6                      \n\t" // 0
        "psubw %%mm2, %%mm6                     \n\t"
        "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
        "movq %%mm7, %%mm6                      \n\t" // 0
        "psubw %%mm3, %%mm6                     \n\t"
        "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
Michael Niedermayer's avatar
Michael Niedermayer committed
939
#else
940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955
        "movq %%mm7, %%mm6                      \n\t" // 0
        "pcmpgtw %%mm0, %%mm6                   \n\t"
        "pxor %%mm6, %%mm0                      \n\t"
        "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
        "movq %%mm7, %%mm6                      \n\t" // 0
        "pcmpgtw %%mm1, %%mm6                   \n\t"
        "pxor %%mm6, %%mm1                      \n\t"
        "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
        "movq %%mm7, %%mm6                      \n\t" // 0
        "pcmpgtw %%mm2, %%mm6                   \n\t"
        "pxor %%mm6, %%mm2                      \n\t"
        "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
        "movq %%mm7, %%mm6                      \n\t" // 0
        "pcmpgtw %%mm3, %%mm6                   \n\t"
        "pxor %%mm6, %%mm3                      \n\t"
        "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
Michael Niedermayer's avatar
Michael Niedermayer committed
956
#endif
957

958
#if TEMPLATE_PP_MMXEXT
959 960
        "pminsw %%mm2, %%mm0                    \n\t"
        "pminsw %%mm3, %%mm1                    \n\t"
961
#else
962 963 964 965 966 967
        "movq %%mm0, %%mm6                      \n\t"
        "psubusw %%mm2, %%mm6                   \n\t"
        "psubw %%mm6, %%mm0                     \n\t"
        "movq %%mm1, %%mm6                      \n\t"
        "psubusw %%mm3, %%mm6                   \n\t"
        "psubw %%mm6, %%mm1                     \n\t"
968 969
#endif

970 971
        "movd %2, %%mm2                         \n\t" // QP
        "punpcklbw %%mm7, %%mm2                 \n\t"
972

973 974 975 976 977 978 979
        "movq %%mm7, %%mm6                      \n\t" // 0
        "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
        "pxor %%mm6, %%mm4                      \n\t"
        "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
        "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
        "pxor %%mm7, %%mm5                      \n\t"
        "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
980
// 100 opcodes
981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001
        "psllw $3, %%mm2                        \n\t" // 8QP
        "movq %%mm2, %%mm3                      \n\t" // 8QP
        "pcmpgtw %%mm4, %%mm2                   \n\t"
        "pcmpgtw %%mm5, %%mm3                   \n\t"
        "pand %%mm2, %%mm4                      \n\t"
        "pand %%mm3, %%mm5                      \n\t"


        "psubusw %%mm0, %%mm4                   \n\t" // hd
        "psubusw %%mm1, %%mm5                   \n\t" // ld


        "movq "MANGLE(w05)", %%mm2              \n\t" // 5
        "pmullw %%mm2, %%mm4                    \n\t"
        "pmullw %%mm2, %%mm5                    \n\t"
        "movq "MANGLE(w20)", %%mm2              \n\t" // 32
        "paddw %%mm2, %%mm4                     \n\t"
        "paddw %%mm2, %%mm5                     \n\t"
        "psrlw $6, %%mm4                        \n\t"
        "psrlw $6, %%mm5                        \n\t"

1002 1003
        "movq 16(%3), %%mm0                     \n\t" // L3 - L4
        "movq 24(%3), %%mm1                     \n\t" // H3 - H4
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020

        "pxor %%mm2, %%mm2                      \n\t"
        "pxor %%mm3, %%mm3                      \n\t"

        "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
        "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
        "pxor %%mm2, %%mm0                      \n\t"
        "pxor %%mm3, %%mm1                      \n\t"
        "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
        "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
        "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
        "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2

        "pxor %%mm6, %%mm2                      \n\t"
        "pxor %%mm7, %%mm3                      \n\t"
        "pand %%mm2, %%mm4                      \n\t"
        "pand %%mm3, %%mm5                      \n\t"
1021

1022
#if TEMPLATE_PP_MMXEXT
1023 1024
        "pminsw %%mm0, %%mm4                    \n\t"
        "pminsw %%mm1, %%mm5                    \n\t"
1025
#else
1026 1027 1028 1029 1030 1031
        "movq %%mm4, %%mm2                      \n\t"
        "psubusw %%mm0, %%mm2                   \n\t"
        "psubw %%mm2, %%mm4                     \n\t"
        "movq %%mm5, %%mm2                      \n\t"
        "psubusw %%mm1, %%mm2                   \n\t"
        "psubw %%mm2, %%mm5                     \n\t"
1032
#endif
1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045
        "pxor %%mm6, %%mm4                      \n\t"
        "pxor %%mm7, %%mm5                      \n\t"
        "psubw %%mm6, %%mm4                     \n\t"
        "psubw %%mm7, %%mm5                     \n\t"
        "packsswb %%mm5, %%mm4                  \n\t"
        "movq (%0), %%mm0                       \n\t"
        "paddb   %%mm4, %%mm0                   \n\t"
        "movq %%mm0, (%0)                       \n\t"
        "movq (%0, %1), %%mm0                   \n\t"
        "psubb %%mm4, %%mm0                     \n\t"
        "movq %%mm0, (%0, %1)                   \n\t"

        : "+r" (src)
1046
        : "r" ((x86_reg)stride), "m" (c->pQPb), "r"(tmp)
1047
          NAMED_CONSTRAINTS_ADD(w05,w20)
1048
        : "%"REG_a
1049
    );
1050
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
    const int l1= stride;
    const int l2= stride + l1;
    const int l3= stride + l2;
    const int l4= stride + l3;
    const int l5= stride + l4;
    const int l6= stride + l5;
    const int l7= stride + l6;
    const int l8= stride + l7;
//    const int l9= stride + l8;
    int x;
    src+= stride*3;
    for(x=0; x<BLOCK_SIZE; x++){
        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
        if(FFABS(middleEnergy) < 8*c->QP){
            const int q=(src[l4] - src[l5])/2;
            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
            d= FFMAX(d, 0);

            d= (5*d + 32) >> 6;
            d*= FFSIGN(-middleEnergy);

            if(q>0){
1076 1077
                d = FFMAX(d, 0);
                d = FFMIN(d, q);
1078
            }else{
1079 1080
                d = FFMIN(d, 0);
                d = FFMAX(d, q);
1081 1082 1083 1084
            }

            src[l4]-= d;
            src[l5]+= d;
1085
        }
1086 1087
        src++;
    }
1088
#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1089
}
1090
#endif //TEMPLATE_PP_ALTIVEC
1091

1092
#if !TEMPLATE_PP_ALTIVEC
1093
static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1094
{
1095
#if HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW)
1096
    DECLARE_ALIGNED(8, uint64_t, tmp)[3];
1097
    __asm__ volatile(
1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
        "pxor %%mm6, %%mm6                      \n\t"
        "pcmpeqb %%mm7, %%mm7                   \n\t"
        "movq %2, %%mm0                         \n\t"
        "punpcklbw %%mm6, %%mm0                 \n\t"
        "psrlw $1, %%mm0                        \n\t"
        "psubw %%mm7, %%mm0                     \n\t"
        "packuswb %%mm0, %%mm0                  \n\t"
        "movq %%mm0, %3                         \n\t"

        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1109 1110 1111

//        0        1        2        3        4        5        6        7        8        9
//        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1
1112

1113
#undef REAL_FIND_MIN_MAX
1114
#undef FIND_MIN_MAX
1115
#if TEMPLATE_PP_MMXEXT
1116
#define REAL_FIND_MIN_MAX(addr)\
1117 1118 1119
        "movq " #addr ", %%mm0                  \n\t"\
        "pminub %%mm0, %%mm7                    \n\t"\
        "pmaxub %%mm0, %%mm6                    \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1120
#else
1121
#define REAL_FIND_MIN_MAX(addr)\
1122 1123 1124 1125 1126 1127
        "movq " #addr ", %%mm0                  \n\t"\
        "movq %%mm7, %%mm1                      \n\t"\
        "psubusb %%mm0, %%mm6                   \n\t"\
        "paddb %%mm0, %%mm6                     \n\t"\
        "psubusb %%mm0, %%mm1                   \n\t"\
        "psubb %%mm1, %%mm7                     \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1128
#endif
1129
#define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
1130

1131 1132 1133
FIND_MIN_MAX((%%REGa))
FIND_MIN_MAX((%%REGa, %1))
FIND_MIN_MAX((%%REGa, %1, 2))
Michael Niedermayer's avatar
Michael Niedermayer committed
1134
FIND_MIN_MAX((%0, %1, 4))
1135 1136 1137
FIND_MIN_MAX((%%REGd))
FIND_MIN_MAX((%%REGd, %1))
FIND_MIN_MAX((%%REGd, %1, 2))
Michael Niedermayer's avatar
Michael Niedermayer committed
1138
FIND_MIN_MAX((%0, %1, 8))
1139

1140 1141
        "movq %%mm7, %%mm4                      \n\t"
        "psrlq $8, %%mm7                        \n\t"
1142
#if TEMPLATE_PP_MMXEXT
1143 1144 1145 1146 1147
        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
        "pshufw $0xF9, %%mm7, %%mm4             \n\t"
        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
        "pshufw $0xFE, %%mm7, %%mm4             \n\t"
        "pminub %%mm4, %%mm7                    \n\t"
1148
#else
1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161
        "movq %%mm7, %%mm1                      \n\t"
        "psubusb %%mm4, %%mm1                   \n\t"
        "psubb %%mm1, %%mm7                     \n\t"
        "movq %%mm7, %%mm4                      \n\t"
        "psrlq $16, %%mm7                       \n\t"
        "movq %%mm7, %%mm1                      \n\t"
        "psubusb %%mm4, %%mm1                   \n\t"
        "psubb %%mm1, %%mm7                     \n\t"
        "movq %%mm7, %%mm4                      \n\t"
        "psrlq $32, %%mm7                       \n\t"
        "movq %%mm7, %%mm1                      \n\t"
        "psubusb %%mm4, %%mm1                   \n\t"
        "psubb %%mm1, %%mm7                     \n\t"
1162
#endif
Michael Niedermayer's avatar
Michael Niedermayer committed
1163 1164


1165 1166
        "movq %%mm6, %%mm4                      \n\t"
        "psrlq $8, %%mm6                        \n\t"
1167
#if TEMPLATE_PP_MMXEXT
1168 1169 1170 1171 1172
        "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
        "pshufw $0xF9, %%mm6, %%mm4             \n\t"
        "pmaxub %%mm4, %%mm6                    \n\t"
        "pshufw $0xFE, %%mm6, %%mm4             \n\t"
        "pmaxub %%mm4, %%mm6                    \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1173
#else
1174 1175 1176 1177 1178 1179 1180 1181 1182 1183
        "psubusb %%mm4, %%mm6                   \n\t"
        "paddb %%mm4, %%mm6                     \n\t"
        "movq %%mm6, %%mm4                      \n\t"
        "psrlq $16, %%mm6                       \n\t"
        "psubusb %%mm4, %%mm6                   \n\t"
        "paddb %%mm4, %%mm6                     \n\t"
        "movq %%mm6, %%mm4                      \n\t"
        "psrlq $32, %%mm6                       \n\t"
        "psubusb %%mm4, %%mm6                   \n\t"
        "paddb %%mm4, %%mm6                     \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1184
#endif
1185 1186
        "movq %%mm6, %%mm0                      \n\t" // max
        "psubb %%mm7, %%mm6                     \n\t" // max - min
1187
        "push %4                              \n\t"
1188 1189
        "movd %%mm6, %k4                        \n\t"
        "cmpb "MANGLE(deringThreshold)", %b4    \n\t"
1190
        "pop %4                               \n\t"
1191 1192 1193 1194 1195
        " jb 1f                                 \n\t"
        PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "punpcklbw %%mm7, %%mm7                 \n\t"
1196
        "movq %%mm7, (%4)                       \n\t"
1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242

        "movq (%0), %%mm0                       \n\t" // L10
        "movq %%mm0, %%mm1                      \n\t" // L10
        "movq %%mm0, %%mm2                      \n\t" // L10
        "psllq $8, %%mm1                        \n\t"
        "psrlq $8, %%mm2                        \n\t"
        "movd -4(%0), %%mm3                     \n\t"
        "movd 8(%0), %%mm4                      \n\t"
        "psrlq $24, %%mm3                       \n\t"
        "psllq $56, %%mm4                       \n\t"
        "por %%mm3, %%mm1                       \n\t" // L00
        "por %%mm4, %%mm2                       \n\t" // L20
        "movq %%mm1, %%mm3                      \n\t" // L00
        PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
        PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
        "psubusb %%mm7, %%mm0                   \n\t"
        "psubusb %%mm7, %%mm2                   \n\t"
        "psubusb %%mm7, %%mm3                   \n\t"
        "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
        "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
        "paddb %%mm2, %%mm0                     \n\t"
        "paddb %%mm3, %%mm0                     \n\t"

        "movq (%%"REG_a"), %%mm2                \n\t" // L11
        "movq %%mm2, %%mm3                      \n\t" // L11
        "movq %%mm2, %%mm4                      \n\t" // L11
        "psllq $8, %%mm3                        \n\t"
        "psrlq $8, %%mm4                        \n\t"
        "movd -4(%%"REG_a"), %%mm5              \n\t"
        "movd 8(%%"REG_a"), %%mm6               \n\t"
        "psrlq $24, %%mm5                       \n\t"
        "psllq $56, %%mm6                       \n\t"
        "por %%mm5, %%mm3                       \n\t" // L01
        "por %%mm6, %%mm4                       \n\t" // L21
        "movq %%mm3, %%mm5                      \n\t" // L01
        PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
        PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
        "psubusb %%mm7, %%mm2                   \n\t"
        "psubusb %%mm7, %%mm4                   \n\t"
        "psubusb %%mm7, %%mm5                   \n\t"
        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
        "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
        "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
        "paddb %%mm4, %%mm2                     \n\t"
        "paddb %%mm5, %%mm2                     \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1243
// 0, 2, 3, 1
1244
#define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259
        "movq " #src ", " #sx "                 \n\t" /* src[0] */\
        "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
        "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
        "psllq $8, " #lx "                      \n\t"\
        "psrlq $8, " #t0 "                      \n\t"\
        "movd -4" #src ", " #t1 "               \n\t"\
        "psrlq $24, " #t1 "                     \n\t"\
        "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
        "movd 8" #src ", " #t1 "                \n\t"\
        "psllq $56, " #t1 "                     \n\t"\
        "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
        "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
        PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
        PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
        PAVGB(lx, pplx)                                     \
1260 1261
        "movq " #lx ", 8(%4)                    \n\t"\
        "movq (%4), " #lx "                     \n\t"\
1262 1263 1264 1265 1266 1267 1268 1269 1270
        "psubusb " #lx ", " #t1 "               \n\t"\
        "psubusb " #lx ", " #t0 "               \n\t"\
        "psubusb " #lx ", " #sx "               \n\t"\
        "movq "MANGLE(b00)", " #lx "            \n\t"\
        "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
        "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
        "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
        "paddb " #t1 ", " #t0 "                 \n\t"\
        "paddb " #t0 ", " #sx "                 \n\t"\
Michael Niedermayer's avatar
Michael Niedermayer committed
1271
\
1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287
        PAVGB(plx, pplx)                              /* filtered */\
        "movq " #dst ", " #t0 "                 \n\t" /* dst */\
        "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
        "psubusb %3, " #t0 "                    \n\t"\
        "paddusb %3, " #t1 "                    \n\t"\
        PMAXUB(t0, pplx)\
        PMINUB(t1, pplx, t0)\
        "paddb " #sx ", " #ppsx "               \n\t"\
        "paddb " #psx ", " #ppsx "              \n\t"\
        "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
        "pand "MANGLE(b08)", " #ppsx "          \n\t"\
        "pcmpeqb " #lx ", " #ppsx "             \n\t"\
        "pand " #ppsx ", " #pplx "              \n\t"\
        "pandn " #dst ", " #ppsx "              \n\t"\
        "por " #pplx ", " #ppsx "               \n\t"\
        "movq " #ppsx ", " #dst "               \n\t"\
1288
        "movq 8(%4), " #lx "                    \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1289

1290 1291
#define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
   REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
Michael Niedermayer's avatar
Michael Niedermayer committed
1292 1293 1294
/*
0000000
1111111
1295

Michael Niedermayer's avatar
Michael Niedermayer committed
1296 1297 1298 1299 1300 1301
1111110
1111101
1111100
1111011
1111010
1111001
1302

Michael Niedermayer's avatar
Michael Niedermayer committed
1303 1304
1111000
1110111
1305

Michael Niedermayer's avatar
Michael Niedermayer committed
1306
*/
1307 1308 1309 1310 1311 1312 1313 1314 1315 1316
//DERING_CORE(dst          ,src            ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)
DERING_CORE((%%REGa)       ,(%%REGa, %1)   ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
DERING_CORE((%%REGa, %1)   ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
DERING_CORE((%%REGa, %1, 2),(%0, %1, 4)    ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
DERING_CORE((%0, %1, 4)    ,(%%REGd)       ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
DERING_CORE((%%REGd)       ,(%%REGd, %1)   ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

1317
        "1:                        \n\t"
1318
        : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2), "q"(tmp)
1319
          NAMED_CONSTRAINTS_ADD(deringThreshold,b00,b02,b08)
1320
        : "%"REG_a, "%"REG_d, "%"REG_SP
1321
    );
1322
#else // HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW)
1323 1324 1325 1326 1327 1328 1329 1330
    int y;
    int min=255;
    int max=0;
    int avg;
    uint8_t *p;
    int s[10];
    const int QP2= c->QP/2 + 1;

1331
    src --;
1332 1333 1334 1335 1336 1337 1338
    for(y=1; y<9; y++){
        int x;
        p= src + stride*y;
        for(x=1; x<9; x++){
            p++;
            if(*p > max) max= *p;
            if(*p < min) min= *p;
1339
        }
1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372
    }
    avg= (min + max + 1)>>1;

    if(max - min <deringThreshold) return;

    for(y=0; y<10; y++){
        int t = 0;

        if(src[stride*y + 0] > avg) t+= 1;
        if(src[stride*y + 1] > avg) t+= 2;
        if(src[stride*y + 2] > avg) t+= 4;
        if(src[stride*y + 3] > avg) t+= 8;
        if(src[stride*y + 4] > avg) t+= 16;
        if(src[stride*y + 5] > avg) t+= 32;
        if(src[stride*y + 6] > avg) t+= 64;
        if(src[stride*y + 7] > avg) t+= 128;
        if(src[stride*y + 8] > avg) t+= 256;
        if(src[stride*y + 9] > avg) t+= 512;

        t |= (~t)<<16;
        t &= (t<<1) & (t>>1);
        s[y] = t;
    }

    for(y=1; y<9; y++){
        int t = s[y-1] & s[y] & s[y+1];
        t|= t>>16;
        s[y-1]= t;
    }

    for(y=1; y<9; y++){
        int x;
        int t = s[y-1];
1373

1374 1375 1376 1377 1378 1379 1380 1381
        p= src + stride*y;
        for(x=1; x<9; x++){
            p++;
            if(t & (1<<x)){
                int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
                      +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
                      +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
                f= (f + 8)>>4;
Michael Niedermayer's avatar
Michael Niedermayer committed
1382

Michael Niedermayer's avatar
Michael Niedermayer committed
1383
#ifdef DEBUG_DERING_THRESHOLD
1384
                    __asm__ volatile("emms\n\t":);
1385
                    {
1386
                    static uint64_t numPixels=0;
1387 1388 1389 1390 1391
                    if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
//                    if((max-min)<20 || (max-min)*QP<200)
//                    if((max-min)*QP < 500)
//                    if(max-min<QP/2)
                    if(max-min < 20){
1392
                        static int numSkipped=0;
1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
                        static int errorSum=0;
                        static int worstQP=0;
                        static int worstRange=0;
                        static int worstDiff=0;
                        int diff= (f - *p);
                        int absDiff= FFABS(diff);
                        int error= diff*diff;

                        if(x==1 || x==8 || y==1 || y==8) continue;

1403
                        numSkipped++;
1404 1405 1406 1407
                        if(absDiff > worstDiff){
                            worstDiff= absDiff;
                            worstQP= QP;
                            worstRange= max-min;
1408
                        }
1409 1410
                        errorSum+= error;

1411
                        if(1024LL*1024LL*1024LL % numSkipped == 0){
1412 1413
                            av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
                                   "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1414 1415
                                   (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
                                   worstDiff, (float)numSkipped/numPixels);
1416 1417 1418 1419 1420 1421 1422 1423
                        }
                    }
                    }
#endif
                    if     (*p + QP2 < f) *p= *p + QP2;
                    else if(*p - QP2 > f) *p= *p - QP2;
                    else *p=f;
            }
1424
        }
1425
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
1426
#ifdef DEBUG_DERING_THRESHOLD
1427 1428 1429 1430 1431 1432 1433 1434 1435
    if(max-min < 20){
        for(y=1; y<9; y++){
            int x;
            int t = 0;
            p= src + stride*y;
            for(x=1; x<9; x++){
                p++;
                *p = FFMIN(*p + 20, 255);
            }
1436
        }
1437 1438
//        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
1439
#endif
1440
#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1441
}
1442
#endif //TEMPLATE_PP_ALTIVEC
1443

1444
/**
1445
 * Deinterlace the given block by linearly interpolating every second line.
Michael Niedermayer's avatar
Michael Niedermayer committed
1446
 * will be called for every 8x8 block and can read & write from line 4-15
Diego Biurrun's avatar
Diego Biurrun committed
1447
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
Michael Niedermayer's avatar
Michael Niedermayer committed
1448
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1449
 */
1450
static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
1451
{
1452
#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1453
    src+= 4*stride;
1454
    __asm__ volatile(
1455 1456
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
1457 1458 1459
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1

1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473
        "movq (%0), %%mm0                       \n\t"
        "movq (%%"REG_a", %1), %%mm1            \n\t"
        PAVGB(%%mm1, %%mm0)
        "movq %%mm0, (%%"REG_a")                \n\t"
        "movq (%0, %1, 4), %%mm0                \n\t"
        PAVGB(%%mm0, %%mm1)
        "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
        "movq (%%"REG_c", %1), %%mm1            \n\t"
        PAVGB(%%mm1, %%mm0)
        "movq %%mm0, (%%"REG_c")                \n\t"
        "movq (%0, %1, 8), %%mm0                \n\t"
        PAVGB(%%mm0, %%mm1)
        "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"

1474
        : : "r" (src), "r" ((x86_reg)stride)
1475 1476
        : "%"REG_a, "%"REG_c
    );
1477
#else
1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492
    int a, b, x;
    src+= 4*stride;

    for(x=0; x<2; x++){
        a= *(uint32_t*)&src[stride*0];
        b= *(uint32_t*)&src[stride*2];
        *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
        a= *(uint32_t*)&src[stride*4];
        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
        b= *(uint32_t*)&src[stride*6];
        *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
        a= *(uint32_t*)&src[stride*8];
        *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
        src += 4;
    }
1493 1494 1495 1496
#endif
}

/**
1497
 * Deinterlace the given block by cubic interpolating every second line.
Michael Niedermayer's avatar
Michael Niedermayer committed
1498
 * will be called for every 8x8 block and can read & write from line 4-15
Diego Biurrun's avatar
Diego Biurrun committed
1499
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
Michael Niedermayer's avatar
Michael Niedermayer committed
1500 1501
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
 * this filter will read lines 3-15 and write 7-13
1502
 */
1503
static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
1504
{
1505
#if TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1506
    src+= stride*3;
1507
    __asm__ volatile(
1508 1509 1510 1511
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
        "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
        "add %1, %%"REG_c"                      \n\t"
1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528
#if TEMPLATE_PP_SSE2
        "pxor %%xmm7, %%xmm7                    \n\t"
#define REAL_DEINT_CUBIC(a,b,c,d,e)\
        "movq " #a ", %%xmm0                    \n\t"\
        "movq " #b ", %%xmm1                    \n\t"\
        "movq " #d ", %%xmm2                    \n\t"\
        "movq " #e ", %%xmm3                    \n\t"\
        "pavgb %%xmm2, %%xmm1                   \n\t"\
        "pavgb %%xmm3, %%xmm0                   \n\t"\
        "punpcklbw %%xmm7, %%xmm0               \n\t"\
        "punpcklbw %%xmm7, %%xmm1               \n\t"\
        "psubw %%xmm1, %%xmm0                   \n\t"\
        "psraw $3, %%xmm0                       \n\t"\
        "psubw %%xmm0, %%xmm1                   \n\t"\
        "packuswb %%xmm1, %%xmm1                \n\t"\
        "movlps %%xmm1, " #c "                  \n\t"
#else //TEMPLATE_PP_SSE2
1529
        "pxor %%mm7, %%mm7                      \n\t"
1530 1531
//      0       1       2       3       4       5       6       7       8       9       10
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1532

1533
#define REAL_DEINT_CUBIC(a,b,c,d,e)\
1534 1535 1536 1537 1538
        "movq " #a ", %%mm0                     \n\t"\
        "movq " #b ", %%mm1                     \n\t"\
        "movq " #d ", %%mm2                     \n\t"\
        "movq " #e ", %%mm3                     \n\t"\
        PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
1539
        PAVGB(%%mm3, %%mm0)                             /* (a+e) /2 */\
1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553
        "movq %%mm0, %%mm2                      \n\t"\
        "punpcklbw %%mm7, %%mm0                 \n\t"\
        "punpckhbw %%mm7, %%mm2                 \n\t"\
        "movq %%mm1, %%mm3                      \n\t"\
        "punpcklbw %%mm7, %%mm1                 \n\t"\
        "punpckhbw %%mm7, %%mm3                 \n\t"\
        "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
        "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
        "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
        "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
        "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
        "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
        "packuswb %%mm3, %%mm1                  \n\t"\
        "movq %%mm1, " #c "                     \n\t"
1554
#endif //TEMPLATE_PP_SSE2
1555
#define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)
1556

1557 1558 1559 1560
DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))
DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
1561

1562
        : : "r" (src), "r" ((x86_reg)stride)
1563 1564 1565 1566 1567
        :
#if TEMPLATE_PP_SSE2
        XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm7",)
#endif
        "%"REG_a, "%"REG_d, "%"REG_c
1568
    );
1569 1570
#undef REAL_DEINT_CUBIC
#else //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1571 1572 1573
    int x;
    src+= stride*3;
    for(x=0; x<8; x++){
1574 1575 1576 1577
        src[stride*3] = av_clip_uint8((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
        src[stride*5] = av_clip_uint8((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
        src[stride*7] = av_clip_uint8((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
        src[stride*9] = av_clip_uint8((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1578 1579
        src++;
    }
1580
#endif //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1581 1582
}

1583
/**
1584
 * Deinterlace the given block by filtering every second line with a (-1 4 2 4 -1) filter.
1585
 * will be called for every 8x8 block and can read & write from line 4-15
Diego Biurrun's avatar
Diego Biurrun committed
1586
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1587 1588 1589 1590 1591
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
 * this filter will read lines 4-13 and write 5-11
 */
static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
{
1592
#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1593
    src+= stride*4;
1594
    __asm__ volatile(
1595 1596 1597 1598
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
        "pxor %%mm7, %%mm7                      \n\t"
        "movq (%2), %%mm0                       \n\t"
1599 1600
//      0       1       2       3       4       5       6       7       8       9       10
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1601

1602
#define REAL_DEINT_FF(a,b,c,d)\
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628
        "movq " #a ", %%mm1                     \n\t"\
        "movq " #b ", %%mm2                     \n\t"\
        "movq " #c ", %%mm3                     \n\t"\
        "movq " #d ", %%mm4                     \n\t"\
        PAVGB(%%mm3, %%mm1)                          \
        PAVGB(%%mm4, %%mm0)                          \
        "movq %%mm0, %%mm3                      \n\t"\
        "punpcklbw %%mm7, %%mm0                 \n\t"\
        "punpckhbw %%mm7, %%mm3                 \n\t"\
        "movq %%mm1, %%mm4                      \n\t"\
        "punpcklbw %%mm7, %%mm1                 \n\t"\
        "punpckhbw %%mm7, %%mm4                 \n\t"\
        "psllw $2, %%mm1                        \n\t"\
        "psllw $2, %%mm4                        \n\t"\
        "psubw %%mm0, %%mm1                     \n\t"\
        "psubw %%mm3, %%mm4                     \n\t"\
        "movq %%mm2, %%mm5                      \n\t"\
        "movq %%mm2, %%mm0                      \n\t"\
        "punpcklbw %%mm7, %%mm2                 \n\t"\
        "punpckhbw %%mm7, %%mm5                 \n\t"\
        "paddw %%mm2, %%mm1                     \n\t"\
        "paddw %%mm5, %%mm4                     \n\t"\
        "psraw $2, %%mm1                        \n\t"\
        "psraw $2, %%mm4                        \n\t"\
        "packuswb %%mm4, %%mm1                  \n\t"\
        "movq %%mm1, " #b "                     \n\t"\
1629

1630 1631
#define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)

1632 1633 1634 1635
DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))
DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1636

1637
        "movq %%mm0, (%2)                       \n\t"
1638
        : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)
1639 1640
        : "%"REG_a, "%"REG_d
    );
1641
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1642 1643 1644 1645 1646 1647
    int x;
    src+= stride*4;
    for(x=0; x<8; x++){
        int t1= tmp[x];
        int t2= src[stride*1];

1648
        src[stride*1]= av_clip_uint8((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1649
        t1= src[stride*4];
1650
        src[stride*3]= av_clip_uint8((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1651
        t2= src[stride*6];
1652
        src[stride*5]= av_clip_uint8((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1653
        t1= src[stride*8];
1654
        src[stride*7]= av_clip_uint8((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1655 1656 1657 1658
        tmp[x]= t1;

        src++;
    }
1659
#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1660 1661
}

1662
/**
1663
 * Deinterlace the given block by filtering every line with a (-1 2 6 2 -1) filter.
1664
 * will be called for every 8x8 block and can read & write from line 4-15
Diego Biurrun's avatar
Diego Biurrun committed
1665
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1666 1667 1668 1669 1670
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
 * this filter will read lines 4-13 and write 4-11
 */
static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
{
1671
#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
1672
    src+= stride*4;
1673
    __asm__ volatile(
1674 1675 1676 1677 1678
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
        "pxor %%mm7, %%mm7                      \n\t"
        "movq (%2), %%mm0                       \n\t"
        "movq (%3), %%mm1                       \n\t"
1679 1680
//      0       1       2       3       4       5       6       7       8       9       10
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1681

1682
#define REAL_DEINT_L5(t1,t2,a,b,c)\
1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713
        "movq " #a ", %%mm2                     \n\t"\
        "movq " #b ", %%mm3                     \n\t"\
        "movq " #c ", %%mm4                     \n\t"\
        PAVGB(t2, %%mm3)                             \
        PAVGB(t1, %%mm4)                             \
        "movq %%mm2, %%mm5                      \n\t"\
        "movq %%mm2, " #t1 "                    \n\t"\
        "punpcklbw %%mm7, %%mm2                 \n\t"\
        "punpckhbw %%mm7, %%mm5                 \n\t"\
        "movq %%mm2, %%mm6                      \n\t"\
        "paddw %%mm2, %%mm2                     \n\t"\
        "paddw %%mm6, %%mm2                     \n\t"\
        "movq %%mm5, %%mm6                      \n\t"\
        "paddw %%mm5, %%mm5                     \n\t"\
        "paddw %%mm6, %%mm5                     \n\t"\
        "movq %%mm3, %%mm6                      \n\t"\
        "punpcklbw %%mm7, %%mm3                 \n\t"\
        "punpckhbw %%mm7, %%mm6                 \n\t"\
        "paddw %%mm3, %%mm3                     \n\t"\
        "paddw %%mm6, %%mm6                     \n\t"\
        "paddw %%mm3, %%mm2                     \n\t"\
        "paddw %%mm6, %%mm5                     \n\t"\
        "movq %%mm4, %%mm6                      \n\t"\
        "punpcklbw %%mm7, %%mm4                 \n\t"\
        "punpckhbw %%mm7, %%mm6                 \n\t"\
        "psubw %%mm4, %%mm2                     \n\t"\
        "psubw %%mm6, %%mm5                     \n\t"\
        "psraw $2, %%mm2                        \n\t"\
        "psraw $2, %%mm5                        \n\t"\
        "packuswb %%mm5, %%mm2                  \n\t"\
        "movq %%mm2, " #a "                     \n\t"\
1714

1715 1716 1717 1718 1719 1720
#define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)

DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )
DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))
DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )
DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )
1721
DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )
1722 1723 1724
DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
1725

1726 1727
        "movq %%mm0, (%2)                       \n\t"
        "movq %%mm1, (%3)                       \n\t"
1728
        : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)
1729 1730
        : "%"REG_a, "%"REG_d
    );
1731
#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
1732 1733 1734 1735 1736 1737 1738
    int x;
    src+= stride*4;
    for(x=0; x<8; x++){
        int t1= tmp[x];
        int t2= tmp2[x];
        int t3= src[0];

1739
        src[stride*0]= av_clip_uint8((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1740
        t1= src[stride*1];
1741
        src[stride*1]= av_clip_uint8((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1742
        t2= src[stride*2];
1743
        src[stride*2]= av_clip_uint8((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1744
        t3= src[stride*3];
1745
        src[stride*3]= av_clip_uint8((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1746
        t1= src[stride*4];
1747
        src[stride*4]= av_clip_uint8((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1748
        t2= src[stride*5];
1749
        src[stride*5]= av_clip_uint8((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1750
        t3= src[stride*6];
1751
        src[stride*6]= av_clip_uint8((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1752
        t1= src[stride*7];
1753
        src[stride*7]= av_clip_uint8((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1754 1755 1756 1757 1758 1759

        tmp[x]= t3;
        tmp2[x]= t1;

        src++;
    }
1760
#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
1761 1762
}

1763
/**
1764
 * Deinterlace the given block by filtering all lines with a (1 2 1) filter.
Michael Niedermayer's avatar
Michael Niedermayer committed
1765
 * will be called for every 8x8 block and can read & write from line 4-15
Diego Biurrun's avatar
Diego Biurrun committed
1766
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
Michael Niedermayer's avatar
Michael Niedermayer committed
1767 1768
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
 * this filter will read lines 4-13 and write 4-11
1769
 */
1770
static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
1771
{
1772
#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1773
    src+= 4*stride;
1774
    __asm__ volatile(
1775 1776
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1777 1778 1779
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815
        "movq (%2), %%mm0                       \n\t" // L0
        "movq (%%"REG_a"), %%mm1                \n\t" // L2
        PAVGB(%%mm1, %%mm0)                           // L0+L2
        "movq (%0), %%mm2                       \n\t" // L1
        PAVGB(%%mm2, %%mm0)
        "movq %%mm0, (%0)                       \n\t"
        "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
        PAVGB(%%mm0, %%mm2)                           // L1+L3
        PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
        "movq %%mm2, (%%"REG_a")                \n\t"
        "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
        PAVGB(%%mm2, %%mm1)                           // L2+L4
        PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
        "movq %%mm1, (%%"REG_a", %1)            \n\t"
        "movq (%0, %1, 4), %%mm1                \n\t" // L5
        PAVGB(%%mm1, %%mm0)                           // L3+L5
        PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
        "movq (%%"REG_d"), %%mm0                \n\t" // L6
        PAVGB(%%mm0, %%mm2)                           // L4+L6
        PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
        "movq %%mm2, (%0, %1, 4)                \n\t"
        "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
        PAVGB(%%mm2, %%mm1)                           // L5+L7
        PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
        "movq %%mm1, (%%"REG_d")                \n\t"
        "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
        PAVGB(%%mm1, %%mm0)                           // L6+L8
        PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
        "movq %%mm0, (%%"REG_d", %1)            \n\t"
        "movq (%0, %1, 8), %%mm0                \n\t" // L9
        PAVGB(%%mm0, %%mm2)                           // L7+L9
        PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
        "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
        "movq %%mm1, (%2)                       \n\t"

1816
        : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
1817 1818
        : "%"REG_a, "%"REG_d
    );
1819
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861
    int a, b, c, x;
    src+= 4*stride;

    for(x=0; x<2; x++){
        a= *(uint32_t*)&tmp[stride*0];
        b= *(uint32_t*)&src[stride*0];
        c= *(uint32_t*)&src[stride*1];
        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

        a= *(uint32_t*)&src[stride*2];
        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

        b= *(uint32_t*)&src[stride*3];
        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

        c= *(uint32_t*)&src[stride*4];
        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

        a= *(uint32_t*)&src[stride*5];
        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

        b= *(uint32_t*)&src[stride*6];
        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

        c= *(uint32_t*)&src[stride*7];
        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

        a= *(uint32_t*)&src[stride*8];
        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
        *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

        *(uint32_t*)&tmp[stride*0]= c;
        src += 4;
        tmp += 4;
    }
1862
#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1863 1864 1865
}

/**
1866
 * Deinterlace the given block by applying a median filter to every second line.
Michael Niedermayer's avatar
Michael Niedermayer committed
1867
 * will be called for every 8x8 block and can read & write from line 4-15,
Diego Biurrun's avatar
Diego Biurrun committed
1868
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
Michael Niedermayer's avatar
Michael Niedermayer committed
1869
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1870
 */
1871
static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
1872
{
1873
#if TEMPLATE_PP_MMX
1874
    src+= 4*stride;
1875
#if TEMPLATE_PP_MMXEXT
1876
    __asm__ volatile(
1877 1878
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1879 1880 1881
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

1882 1883 1884
        "movq (%0), %%mm0                       \n\t"
        "movq (%%"REG_a", %1), %%mm2            \n\t"
        "movq (%%"REG_a"), %%mm1                \n\t"
1885
        "movq %%mm0, %%mm3                      \n\t"
1886 1887 1888
        "pmaxub %%mm1, %%mm0                    \n\t"
        "pminub %%mm3, %%mm1                    \n\t"
        "pmaxub %%mm2, %%mm1                    \n\t"
1889 1890 1891
        "pminub %%mm1, %%mm0                    \n\t"
        "movq %%mm0, (%%"REG_a")                \n\t"

1892 1893
        "movq (%0, %1, 4), %%mm0                \n\t"
        "movq (%%"REG_a", %1, 2), %%mm1         \n\t"
1894
        "movq %%mm2, %%mm3                      \n\t"
1895 1896 1897
        "pmaxub %%mm1, %%mm2                    \n\t"
        "pminub %%mm3, %%mm1                    \n\t"
        "pmaxub %%mm0, %%mm1                    \n\t"
1898 1899 1900
        "pminub %%mm1, %%mm2                    \n\t"
        "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"

1901 1902
        "movq (%%"REG_d"), %%mm2                \n\t"
        "movq (%%"REG_d", %1), %%mm1            \n\t"
1903
        "movq %%mm2, %%mm3                      \n\t"
1904 1905 1906
        "pmaxub %%mm0, %%mm2                    \n\t"
        "pminub %%mm3, %%mm0                    \n\t"
        "pmaxub %%mm1, %%mm0                    \n\t"
1907 1908 1909
        "pminub %%mm0, %%mm2                    \n\t"
        "movq %%mm2, (%%"REG_d")                \n\t"

1910 1911
        "movq (%%"REG_d", %1, 2), %%mm2         \n\t"
        "movq (%0, %1, 8), %%mm0                \n\t"
1912
        "movq %%mm2, %%mm3                      \n\t"
1913 1914 1915
        "pmaxub %%mm0, %%mm2                    \n\t"
        "pminub %%mm3, %%mm0                    \n\t"
        "pmaxub %%mm1, %%mm0                    \n\t"
1916 1917 1918 1919
        "pminub %%mm0, %%mm2                    \n\t"
        "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"


1920
        : : "r" (src), "r" ((x86_reg)stride)
1921 1922
        : "%"REG_a, "%"REG_d
    );
Michael Niedermayer's avatar
Michael Niedermayer committed
1923 1924

#else // MMX without MMX2
1925
    __asm__ volatile(
1926 1927
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1928 1929
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1930
        "pxor %%mm7, %%mm7                      \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1931

1932
#define REAL_MEDIAN(a,b,c)\
1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954
        "movq " #a ", %%mm0                     \n\t"\
        "movq " #b ", %%mm2                     \n\t"\
        "movq " #c ", %%mm1                     \n\t"\
        "movq %%mm0, %%mm3                      \n\t"\
        "movq %%mm1, %%mm4                      \n\t"\
        "movq %%mm2, %%mm5                      \n\t"\
        "psubusb %%mm1, %%mm3                   \n\t"\
        "psubusb %%mm2, %%mm4                   \n\t"\
        "psubusb %%mm0, %%mm5                   \n\t"\
        "pcmpeqb %%mm7, %%mm3                   \n\t"\
        "pcmpeqb %%mm7, %%mm4                   \n\t"\
        "pcmpeqb %%mm7, %%mm5                   \n\t"\
        "movq %%mm3, %%mm6                      \n\t"\
        "pxor %%mm4, %%mm3                      \n\t"\
        "pxor %%mm5, %%mm4                      \n\t"\
        "pxor %%mm6, %%mm5                      \n\t"\
        "por %%mm3, %%mm1                       \n\t"\
        "por %%mm4, %%mm2                       \n\t"\
        "por %%mm5, %%mm0                       \n\t"\
        "pand %%mm2, %%mm0                      \n\t"\
        "pand %%mm1, %%mm0                      \n\t"\
        "movq %%mm0, " #b "                     \n\t"
1955
#define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
Michael Niedermayer's avatar
Michael Niedermayer committed
1956

1957
MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))
1958
MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
1959
MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))
1960
MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
Michael Niedermayer's avatar
Michael Niedermayer committed
1961

1962
        : : "r" (src), "r" ((x86_reg)stride)
1963 1964
        : "%"REG_a, "%"REG_d
    );
1965 1966
#endif //TEMPLATE_PP_MMXEXT
#else //TEMPLATE_PP_MMX
1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981
    int x, y;
    src+= 4*stride;
    // FIXME - there should be a way to do a few columns in parallel like w/mmx
    for(x=0; x<8; x++){
        uint8_t *colsrc = src;
        for (y=0; y<4; y++){
            int a, b, c, d, e, f;
            a = colsrc[0       ];
            b = colsrc[stride  ];
            c = colsrc[stride*2];
            d = (a-b)>>31;
            e = (b-c)>>31;
            f = (c-a)>>31;
            colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
            colsrc += stride*2;
1982
        }
1983 1984
        src++;
    }
1985
#endif //TEMPLATE_PP_MMX
1986 1987
}

1988
#if TEMPLATE_PP_MMX
1989
/**
1990
 * Transpose and shift the given 8x8 Block into dst1 and dst2.
1991
 */
1992
static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, const uint8_t *src, int srcStride)
1993
{
1994
    __asm__(
1995
        "lea (%0, %1), %%"REG_a"                \n\t"
1996 1997
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066
        "movq (%0), %%mm0                       \n\t" // 12345678
        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
        "movq %%mm0, %%mm2                      \n\t" // 12345678
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

        "movq (%%"REG_a", %1), %%mm1            \n\t"
        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
        "movq %%mm1, %%mm4                      \n\t"
        "punpcklbw %%mm3, %%mm1                 \n\t"
        "punpckhbw %%mm3, %%mm4                 \n\t"

        "movq %%mm0, %%mm3                      \n\t"
        "punpcklwd %%mm1, %%mm0                 \n\t"
        "punpckhwd %%mm1, %%mm3                 \n\t"
        "movq %%mm2, %%mm1                      \n\t"
        "punpcklwd %%mm4, %%mm2                 \n\t"
        "punpckhwd %%mm4, %%mm1                 \n\t"

        "movd %%mm0, 128(%2)                    \n\t"
        "psrlq $32, %%mm0                       \n\t"
        "movd %%mm0, 144(%2)                    \n\t"
        "movd %%mm3, 160(%2)                    \n\t"
        "psrlq $32, %%mm3                       \n\t"
        "movd %%mm3, 176(%2)                    \n\t"
        "movd %%mm3, 48(%3)                     \n\t"
        "movd %%mm2, 192(%2)                    \n\t"
        "movd %%mm2, 64(%3)                     \n\t"
        "psrlq $32, %%mm2                       \n\t"
        "movd %%mm2, 80(%3)                     \n\t"
        "movd %%mm1, 96(%3)                     \n\t"
        "psrlq $32, %%mm1                       \n\t"
        "movd %%mm1, 112(%3)                    \n\t"

        "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"

        "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
        "movq %%mm0, %%mm2                      \n\t" // 12345678
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

        "movq (%%"REG_a", %1), %%mm1            \n\t"
        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
        "movq %%mm1, %%mm4                      \n\t"
        "punpcklbw %%mm3, %%mm1                 \n\t"
        "punpckhbw %%mm3, %%mm4                 \n\t"

        "movq %%mm0, %%mm3                      \n\t"
        "punpcklwd %%mm1, %%mm0                 \n\t"
        "punpckhwd %%mm1, %%mm3                 \n\t"
        "movq %%mm2, %%mm1                      \n\t"
        "punpcklwd %%mm4, %%mm2                 \n\t"
        "punpckhwd %%mm4, %%mm1                 \n\t"

        "movd %%mm0, 132(%2)                    \n\t"
        "psrlq $32, %%mm0                       \n\t"
        "movd %%mm0, 148(%2)                    \n\t"
        "movd %%mm3, 164(%2)                    \n\t"
        "psrlq $32, %%mm3                       \n\t"
        "movd %%mm3, 180(%2)                    \n\t"
        "movd %%mm3, 52(%3)                     \n\t"
        "movd %%mm2, 196(%2)                    \n\t"
        "movd %%mm2, 68(%3)                     \n\t"
        "psrlq $32, %%mm2                       \n\t"
        "movd %%mm2, 84(%3)                     \n\t"
        "movd %%mm1, 100(%3)                    \n\t"
        "psrlq $32, %%mm1                       \n\t"
        "movd %%mm1, 116(%3)                    \n\t"
2067 2068


2069
        :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)
2070
        : "%"REG_a
2071
    );
2072 2073 2074
}

/**
2075
 * Transpose the given 8x8 block.
2076
 */
2077
static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, const uint8_t *src)
2078
{
2079
    __asm__(
2080 2081
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
2082 2083
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147
        "movq (%2), %%mm0                       \n\t" // 12345678
        "movq 16(%2), %%mm1                     \n\t" // abcdefgh
        "movq %%mm0, %%mm2                      \n\t" // 12345678
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

        "movq 32(%2), %%mm1                     \n\t"
        "movq 48(%2), %%mm3                     \n\t"
        "movq %%mm1, %%mm4                      \n\t"
        "punpcklbw %%mm3, %%mm1                 \n\t"
        "punpckhbw %%mm3, %%mm4                 \n\t"

        "movq %%mm0, %%mm3                      \n\t"
        "punpcklwd %%mm1, %%mm0                 \n\t"
        "punpckhwd %%mm1, %%mm3                 \n\t"
        "movq %%mm2, %%mm1                      \n\t"
        "punpcklwd %%mm4, %%mm2                 \n\t"
        "punpckhwd %%mm4, %%mm1                 \n\t"

        "movd %%mm0, (%0)                       \n\t"
        "psrlq $32, %%mm0                       \n\t"
        "movd %%mm0, (%%"REG_a")                \n\t"
        "movd %%mm3, (%%"REG_a", %1)            \n\t"
        "psrlq $32, %%mm3                       \n\t"
        "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
        "movd %%mm2, (%0, %1, 4)                \n\t"
        "psrlq $32, %%mm2                       \n\t"
        "movd %%mm2, (%%"REG_d")                \n\t"
        "movd %%mm1, (%%"REG_d", %1)            \n\t"
        "psrlq $32, %%mm1                       \n\t"
        "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"


        "movq 64(%2), %%mm0                     \n\t" // 12345678
        "movq 80(%2), %%mm1                     \n\t" // abcdefgh
        "movq %%mm0, %%mm2                      \n\t" // 12345678
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

        "movq 96(%2), %%mm1                     \n\t"
        "movq 112(%2), %%mm3                    \n\t"
        "movq %%mm1, %%mm4                      \n\t"
        "punpcklbw %%mm3, %%mm1                 \n\t"
        "punpckhbw %%mm3, %%mm4                 \n\t"

        "movq %%mm0, %%mm3                      \n\t"
        "punpcklwd %%mm1, %%mm0                 \n\t"
        "punpckhwd %%mm1, %%mm3                 \n\t"
        "movq %%mm2, %%mm1                      \n\t"
        "punpcklwd %%mm4, %%mm2                 \n\t"
        "punpckhwd %%mm4, %%mm1                 \n\t"

        "movd %%mm0, 4(%0)                      \n\t"
        "psrlq $32, %%mm0                       \n\t"
        "movd %%mm0, 4(%%"REG_a")               \n\t"
        "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
        "psrlq $32, %%mm3                       \n\t"
        "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
        "movd %%mm2, 4(%0, %1, 4)               \n\t"
        "psrlq $32, %%mm2                       \n\t"
        "movd %%mm2, 4(%%"REG_d")               \n\t"
        "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
        "psrlq $32, %%mm1                       \n\t"
        "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
2148

2149
        :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)
2150
        : "%"REG_a, "%"REG_d
2151
    );
2152
}
2153
#endif //TEMPLATE_PP_MMX
2154
//static long test=0;
2155

2156
#if !TEMPLATE_PP_ALTIVEC
Michael Niedermayer's avatar
Michael Niedermayer committed
2157
static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
2158
                                    uint8_t *tempBlurred, uint32_t *tempBlurredPast, const int *maxNoise)
2159
{
2160
    // to save a register (FIXME do this outside of the loops)
2161 2162 2163
    tempBlurredPast[127]= maxNoise[0];
    tempBlurredPast[128]= maxNoise[1];
    tempBlurredPast[129]= maxNoise[2];
2164

2165 2166
#define FAST_L2_DIFF
//#define L1_DIFF //u should change the thresholds too if u try that one
2167
#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
2168
    __asm__ volatile(
2169 2170 2171
        "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
        "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2172 2173
//      0       1       2       3       4       5       6       7       8       9
//      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
2174 2175
//FIXME reorder?
#ifdef L1_DIFF //needs mmx2
2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199
        "movq (%0), %%mm0                       \n\t" // L0
        "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
        "movq (%0, %2), %%mm1                   \n\t" // L1
        "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
        "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
        "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|

        "movq (%0, %2, 4), %%mm4                \n\t" // L4
        "paddw %%mm1, %%mm0                     \n\t"
        "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
        "paddw %%mm2, %%mm0                     \n\t"
        "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
        "paddw %%mm3, %%mm0                     \n\t"
        "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
        "paddw %%mm4, %%mm0                     \n\t"
        "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
        "paddw %%mm5, %%mm6                     \n\t"
        "paddw %%mm7, %%mm6                     \n\t"
        "paddw %%mm6, %%mm0                     \n\t"
2200
#else //L1_DIFF
2201
#if defined (FAST_L2_DIFF)
2202 2203 2204
        "pcmpeqb %%mm7, %%mm7                   \n\t"
        "movq "MANGLE(b80)", %%mm6              \n\t"
        "pxor %%mm0, %%mm0                      \n\t"
2205
#define REAL_L2_DIFF_CORE(a, b)\
2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217
        "movq " #a ", %%mm5                     \n\t"\
        "movq " #b ", %%mm2                     \n\t"\
        "pxor %%mm7, %%mm2                      \n\t"\
        PAVGB(%%mm2, %%mm5)\
        "paddb %%mm6, %%mm5                     \n\t"\
        "movq %%mm5, %%mm2                      \n\t"\
        "psllw $8, %%mm5                        \n\t"\
        "pmaddwd %%mm5, %%mm5                   \n\t"\
        "pmaddwd %%mm2, %%mm2                   \n\t"\
        "paddd %%mm2, %%mm5                     \n\t"\
        "psrld $14, %%mm5                       \n\t"\
        "paddd %%mm5, %%mm0                     \n\t"
2218

2219
#else //defined (FAST_L2_DIFF)
2220 2221
        "pxor %%mm7, %%mm7                      \n\t"
        "pxor %%mm0, %%mm0                      \n\t"
2222
#define REAL_L2_DIFF_CORE(a, b)\
2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236
        "movq " #a ", %%mm5                     \n\t"\
        "movq " #b ", %%mm2                     \n\t"\
        "movq %%mm5, %%mm1                      \n\t"\
        "movq %%mm2, %%mm3                      \n\t"\
        "punpcklbw %%mm7, %%mm5                 \n\t"\
        "punpckhbw %%mm7, %%mm1                 \n\t"\
        "punpcklbw %%mm7, %%mm2                 \n\t"\
        "punpckhbw %%mm7, %%mm3                 \n\t"\
        "psubw %%mm2, %%mm5                     \n\t"\
        "psubw %%mm3, %%mm1                     \n\t"\
        "pmaddwd %%mm5, %%mm5                   \n\t"\
        "pmaddwd %%mm1, %%mm1                   \n\t"\
        "paddd %%mm1, %%mm5                     \n\t"\
        "paddd %%mm5, %%mm0                     \n\t"
2237

2238
#endif //defined (FAST_L2_DIFF)
2239 2240 2241

#define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)

2242 2243 2244 2245 2246 2247
L2_DIFF_CORE((%0)          , (%1))
L2_DIFF_CORE((%0, %2)      , (%1, %2))
L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))
L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))
L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))
L2_DIFF_CORE((%0, %%REGd)  , (%1, %%REGd))
2248
L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
2249
L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))
2250

2251
#endif //L1_DIFF
2252

2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451
        "movq %%mm0, %%mm4                      \n\t"
        "psrlq $32, %%mm0                       \n\t"
        "paddd %%mm0, %%mm4                     \n\t"
        "movd %%mm4, %%ecx                      \n\t"
        "shll $2, %%ecx                         \n\t"
        "mov %3, %%"REG_d"                      \n\t"
        "addl -4(%%"REG_d"), %%ecx              \n\t"
        "addl 4(%%"REG_d"), %%ecx               \n\t"
        "addl -1024(%%"REG_d"), %%ecx           \n\t"
        "addl $4, %%ecx                         \n\t"
        "addl 1024(%%"REG_d"), %%ecx            \n\t"
        "shrl $3, %%ecx                         \n\t"
        "movl %%ecx, (%%"REG_d")                \n\t"

//        "mov %3, %%"REG_c"                      \n\t"
//        "mov %%"REG_c", test                    \n\t"
//        "jmp 4f                                 \n\t"
        "cmpl 512(%%"REG_d"), %%ecx             \n\t"
        " jb 2f                                 \n\t"
        "cmpl 516(%%"REG_d"), %%ecx             \n\t"
        " jb 1f                                 \n\t"

        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
        "movq (%0), %%mm0                       \n\t" // L0
        "movq (%0, %2), %%mm1                   \n\t" // L1
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
        "movq (%0, %2, 4), %%mm4                \n\t" // L4
        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
        "movq %%mm0, (%1)                       \n\t" // L0
        "movq %%mm1, (%1, %2)                   \n\t" // L1
        "movq %%mm2, (%1, %2, 2)                \n\t" // L2
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
        "movq %%mm4, (%1, %2, 4)                \n\t" // L4
        "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
        "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
        "jmp 4f                                 \n\t"

        "1:                                     \n\t"
        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
        "movq (%0), %%mm0                       \n\t" // L0
        PAVGB((%1), %%mm0)                            // L0
        "movq (%0, %2), %%mm1                   \n\t" // L1
        PAVGB((%1, %2), %%mm1)                        // L1
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
        PAVGB((%1, %2, 2), %%mm2)                     // L2
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
        PAVGB((%1, %%REGa), %%mm3)                    // L3
        "movq (%0, %2, 4), %%mm4                \n\t" // L4
        PAVGB((%1, %2, 4), %%mm4)                     // L4
        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
        PAVGB((%1, %%REGd), %%mm5)                    // L5
        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
        PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
        PAVGB((%1, %%REGc), %%mm7)                    // L7
        "movq %%mm0, (%1)                       \n\t" // R0
        "movq %%mm1, (%1, %2)                   \n\t" // R1
        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
        "movq %%mm4, (%1, %2, 4)                \n\t" // R4
        "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
        "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
        "movq %%mm0, (%0)                       \n\t" // L0
        "movq %%mm1, (%0, %2)                   \n\t" // L1
        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
        "movq %%mm4, (%0, %2, 4)                \n\t" // L4
        "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
        "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
        "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
        "jmp 4f                                 \n\t"

        "2:                                     \n\t"
        "cmpl 508(%%"REG_d"), %%ecx             \n\t"
        " jb 3f                                 \n\t"

        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
        "movq (%0), %%mm0                       \n\t" // L0
        "movq (%0, %2), %%mm1                   \n\t" // L1
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
        "movq (%1), %%mm4                       \n\t" // R0
        "movq (%1, %2), %%mm5                   \n\t" // R1
        "movq (%1, %2, 2), %%mm6                \n\t" // R2
        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        "movq %%mm0, (%1)                       \n\t" // R0
        "movq %%mm1, (%1, %2)                   \n\t" // R1
        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
        "movq %%mm0, (%0)                       \n\t" // L0
        "movq %%mm1, (%0, %2)                   \n\t" // L1
        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

        "movq (%0, %2, 4), %%mm0                \n\t" // L4
        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
        "movq (%1, %2, 4), %%mm4                \n\t" // R4
        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
        "jmp 4f                                 \n\t"

        "3:                                     \n\t"
        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
        "movq (%0), %%mm0                       \n\t" // L0
        "movq (%0, %2), %%mm1                   \n\t" // L1
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
        "movq (%1), %%mm4                       \n\t" // R0
        "movq (%1, %2), %%mm5                   \n\t" // R1
        "movq (%1, %2, 2), %%mm6                \n\t" // R2
        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        "movq %%mm0, (%1)                       \n\t" // R0
        "movq %%mm1, (%1, %2)                   \n\t" // R1
        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
        "movq %%mm0, (%0)                       \n\t" // L0
        "movq %%mm1, (%0, %2)                   \n\t" // L1
        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

        "movq (%0, %2, 4), %%mm0                \n\t" // L4
        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
        "movq (%1, %2, 4), %%mm4                \n\t" // R4
        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        PAVGB(%%mm4, %%mm0)
        PAVGB(%%mm5, %%mm1)
        PAVGB(%%mm6, %%mm2)
        PAVGB(%%mm7, %%mm3)
        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7

        "4:                                     \n\t"

2452
        :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)
2453
          NAMED_CONSTRAINTS_ADD(b80)
2454 2455
        : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
    );
2456
#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
2457
{
2458 2459 2460 2461
    int y;
    int d=0;
//    int sysd=0;
    int i;
2462

2463 2464 2465
    for(y=0; y<8; y++){
        int x;
        for(x=0; x<8; x++){
2466
            int ref= tempBlurred[ x + y*stride ];
2467 2468 2469 2470 2471 2472 2473
            int cur= src[ x + y*stride ];
            int d1=ref - cur;
//            if(x==0 || x==7) d1+= d1>>1;
//            if(y==0 || y==7) d1+= d1>>1;
//            d+= FFABS(d1);
            d+= d1*d1;
//            sysd+= d1;
2474
        }
2475 2476 2477 2478
    }
    i=d;
    d=  (
        4*d
2479 2480 2481
        +(*(tempBlurredPast-256))
        +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
        +(*(tempBlurredPast+256))
2482
        +4)>>3;
2483 2484
    *tempBlurredPast=i;
//    ((*tempBlurredPast)*3 + d + 2)>>2;
Michael Niedermayer's avatar
Michael Niedermayer committed
2485

2486 2487 2488 2489 2490 2491 2492
/*
Switch between
 1  0  0  0  0  0  0  (0)
64 32 16  8  4  2  1  (1)
64 48 36 27 20 15 11 (33) (approx)
64 56 49 43 37 33 29 (200) (approx)
*/
2493 2494 2495 2496 2497
    if(d > maxNoise[1]){
        if(d < maxNoise[2]){
            for(y=0; y<8; y++){
                int x;
                for(x=0; x<8; x++){
2498
                    int ref= tempBlurred[ x + y*stride ];
2499
                    int cur= src[ x + y*stride ];
2500
                    tempBlurred[ x + y*stride ]=
2501 2502
                    src[ x + y*stride ]=
                        (ref + cur + 1)>>1;
2503
                }
2504 2505 2506 2507 2508
            }
        }else{
            for(y=0; y<8; y++){
                int x;
                for(x=0; x<8; x++){
2509
                    tempBlurred[ x + y*stride ]= src[ x + y*stride ];
2510
                }
2511
            }
2512
        }
2513 2514 2515 2516 2517
    }else{
        if(d < maxNoise[0]){
            for(y=0; y<8; y++){
                int x;
                for(x=0; x<8; x++){
2518
                    int ref= tempBlurred[ x + y*stride ];
2519
                    int cur= src[ x + y*stride ];
2520
                    tempBlurred[ x + y*stride ]=
2521 2522
                    src[ x + y*stride ]=
                        (ref*7 + cur + 4)>>3;
2523
                }
2524 2525 2526 2527 2528
            }
        }else{
            for(y=0; y<8; y++){
                int x;
                for(x=0; x<8; x++){
2529
                    int ref= tempBlurred[ x + y*stride ];
2530
                    int cur= src[ x + y*stride ];
2531
                    tempBlurred[ x + y*stride ]=
2532 2533
                    src[ x + y*stride ]=
                        (ref*3 + cur + 2)>>2;
2534
                }
2535
            }
2536
        }
2537
    }
2538
}
2539
#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
2540
}
2541
#endif //TEMPLATE_PP_ALTIVEC
2542

2543
#if TEMPLATE_PP_MMX
2544 2545 2546
/**
 * accurate deblock filter
 */
2547
static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, const PPContext *c, int mode){
2548 2549 2550
    int64_t dc_mask, eq_mask, both_masks;
    int64_t sums[10*8*2];
    src+= step*3; // src points to begin of the 8x8 Block
2551
    //{ START_TIMER
2552
    __asm__ volatile(
2553 2554 2555 2556 2557
        "movq %0, %%mm7                         \n\t"
        "movq %1, %%mm6                         \n\t"
        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
        );

2558
    __asm__ volatile(
2559
        "lea (%2, %3), %%"REG_a"                \n\t"
2560 2561 2562
//      0       1       2       3       4       5       6       7       8       9
//      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2

2563 2564 2565 2566
        "movq (%2), %%mm0                       \n\t"
        "movq (%%"REG_a"), %%mm1                \n\t"
        "movq %%mm1, %%mm3                      \n\t"
        "movq %%mm1, %%mm4                      \n\t"
2567
        "psubb %%mm1, %%mm0                     \n\t" // mm0 = difference
2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652
        "paddb %%mm7, %%mm0                     \n\t"
        "pcmpgtb %%mm6, %%mm0                   \n\t"

        "movq (%%"REG_a",%3), %%mm2             \n\t"
        PMAXUB(%%mm2, %%mm4)
        PMINUB(%%mm2, %%mm3, %%mm5)
        "psubb %%mm2, %%mm1                     \n\t"
        "paddb %%mm7, %%mm1                     \n\t"
        "pcmpgtb %%mm6, %%mm1                   \n\t"
        "paddb %%mm1, %%mm0                     \n\t"

        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
        PMAXUB(%%mm1, %%mm4)
        PMINUB(%%mm1, %%mm3, %%mm5)
        "psubb %%mm1, %%mm2                     \n\t"
        "paddb %%mm7, %%mm2                     \n\t"
        "pcmpgtb %%mm6, %%mm2                   \n\t"
        "paddb %%mm2, %%mm0                     \n\t"

        "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"

        "movq (%2, %3, 4), %%mm2                \n\t"
        PMAXUB(%%mm2, %%mm4)
        PMINUB(%%mm2, %%mm3, %%mm5)
        "psubb %%mm2, %%mm1                     \n\t"
        "paddb %%mm7, %%mm1                     \n\t"
        "pcmpgtb %%mm6, %%mm1                   \n\t"
        "paddb %%mm1, %%mm0                     \n\t"

        "movq (%%"REG_a"), %%mm1                \n\t"
        PMAXUB(%%mm1, %%mm4)
        PMINUB(%%mm1, %%mm3, %%mm5)
        "psubb %%mm1, %%mm2                     \n\t"
        "paddb %%mm7, %%mm2                     \n\t"
        "pcmpgtb %%mm6, %%mm2                   \n\t"
        "paddb %%mm2, %%mm0                     \n\t"

        "movq (%%"REG_a", %3), %%mm2            \n\t"
        PMAXUB(%%mm2, %%mm4)
        PMINUB(%%mm2, %%mm3, %%mm5)
        "psubb %%mm2, %%mm1                     \n\t"
        "paddb %%mm7, %%mm1                     \n\t"
        "pcmpgtb %%mm6, %%mm1                   \n\t"
        "paddb %%mm1, %%mm0                     \n\t"

        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
        PMAXUB(%%mm1, %%mm4)
        PMINUB(%%mm1, %%mm3, %%mm5)
        "psubb %%mm1, %%mm2                     \n\t"
        "paddb %%mm7, %%mm2                     \n\t"
        "pcmpgtb %%mm6, %%mm2                   \n\t"
        "paddb %%mm2, %%mm0                     \n\t"

        "movq (%2, %3, 8), %%mm2                \n\t"
        PMAXUB(%%mm2, %%mm4)
        PMINUB(%%mm2, %%mm3, %%mm5)
        "psubb %%mm2, %%mm1                     \n\t"
        "paddb %%mm7, %%mm1                     \n\t"
        "pcmpgtb %%mm6, %%mm1                   \n\t"
        "paddb %%mm1, %%mm0                     \n\t"

        "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
        "psubb %%mm1, %%mm2                     \n\t"
        "paddb %%mm7, %%mm2                     \n\t"
        "pcmpgtb %%mm6, %%mm2                   \n\t"
        "paddb %%mm2, %%mm0                     \n\t"
        "psubusb %%mm3, %%mm4                   \n\t"

        "pxor %%mm6, %%mm6                      \n\t"
        "movq %4, %%mm7                         \n\t" // QP,..., QP
        "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
        "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
        "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
        "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
        "movq %%mm7, %1                         \n\t"

        "movq %5, %%mm7                         \n\t"
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "psubb %%mm0, %%mm6                     \n\t"
        "pcmpgtb %%mm7, %%mm6                   \n\t"
        "movq %%mm6, %0                         \n\t"

        : "=m" (eq_mask), "=m" (dc_mask)
2653
        : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2654 2655 2656 2657 2658 2659
        : "%"REG_a
    );

    both_masks = dc_mask & eq_mask;

    if(both_masks){
2660
        x86_reg offset= -8*step;
2661 2662
        int64_t *temp_sums= sums;

2663
        __asm__ volatile(
2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708
            "movq %2, %%mm0                         \n\t"  // QP,..., QP
            "pxor %%mm4, %%mm4                      \n\t"

            "movq (%0), %%mm6                       \n\t"
            "movq (%0, %1), %%mm5                   \n\t"
            "movq %%mm5, %%mm1                      \n\t"
            "movq %%mm6, %%mm2                      \n\t"
            "psubusb %%mm6, %%mm5                   \n\t"
            "psubusb %%mm1, %%mm2                   \n\t"
            "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
            "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
            "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF

            "pxor %%mm6, %%mm1                      \n\t"
            "pand %%mm0, %%mm1                      \n\t"
            "pxor %%mm1, %%mm6                      \n\t"
            // 0:QP  6:First

            "movq (%0, %1, 8), %%mm5                \n\t"
            "add %1, %0                             \n\t" // %0 points to line 1 not 0
            "movq (%0, %1, 8), %%mm7                \n\t"
            "movq %%mm5, %%mm1                      \n\t"
            "movq %%mm7, %%mm2                      \n\t"
            "psubusb %%mm7, %%mm5                   \n\t"
            "psubusb %%mm1, %%mm2                   \n\t"
            "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
            "movq %2, %%mm0                         \n\t"  // QP,..., QP
            "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
            "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF

            "pxor %%mm7, %%mm1                      \n\t"
            "pand %%mm0, %%mm1                      \n\t"
            "pxor %%mm1, %%mm7                      \n\t"

            "movq %%mm6, %%mm5                      \n\t"
            "punpckhbw %%mm4, %%mm6                 \n\t"
            "punpcklbw %%mm4, %%mm5                 \n\t"
            // 4:0 5/6:First 7:Last

            "movq %%mm5, %%mm0                      \n\t"
            "movq %%mm6, %%mm1                      \n\t"
            "psllw $2, %%mm0                        \n\t"
            "psllw $2, %%mm1                        \n\t"
            "paddw "MANGLE(w04)", %%mm0             \n\t"
            "paddw "MANGLE(w04)", %%mm1             \n\t"
2709 2710

#define NEXT\
2711 2712 2713 2714 2715 2716 2717
            "movq (%0), %%mm2                       \n\t"\
            "movq (%0), %%mm3                       \n\t"\
            "add %1, %0                             \n\t"\
            "punpcklbw %%mm4, %%mm2                 \n\t"\
            "punpckhbw %%mm4, %%mm3                 \n\t"\
            "paddw %%mm2, %%mm0                     \n\t"\
            "paddw %%mm3, %%mm1                     \n\t"
2718 2719

#define PREV\
2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796
            "movq (%0), %%mm2                       \n\t"\
            "movq (%0), %%mm3                       \n\t"\
            "add %1, %0                             \n\t"\
            "punpcklbw %%mm4, %%mm2                 \n\t"\
            "punpckhbw %%mm4, %%mm3                 \n\t"\
            "psubw %%mm2, %%mm0                     \n\t"\
            "psubw %%mm3, %%mm1                     \n\t"


            NEXT //0
            NEXT //1
            NEXT //2
            "movq %%mm0, (%3)                       \n\t"
            "movq %%mm1, 8(%3)                      \n\t"

            NEXT //3
            "psubw %%mm5, %%mm0                     \n\t"
            "psubw %%mm6, %%mm1                     \n\t"
            "movq %%mm0, 16(%3)                     \n\t"
            "movq %%mm1, 24(%3)                     \n\t"

            NEXT //4
            "psubw %%mm5, %%mm0                     \n\t"
            "psubw %%mm6, %%mm1                     \n\t"
            "movq %%mm0, 32(%3)                     \n\t"
            "movq %%mm1, 40(%3)                     \n\t"

            NEXT //5
            "psubw %%mm5, %%mm0                     \n\t"
            "psubw %%mm6, %%mm1                     \n\t"
            "movq %%mm0, 48(%3)                     \n\t"
            "movq %%mm1, 56(%3)                     \n\t"

            NEXT //6
            "psubw %%mm5, %%mm0                     \n\t"
            "psubw %%mm6, %%mm1                     \n\t"
            "movq %%mm0, 64(%3)                     \n\t"
            "movq %%mm1, 72(%3)                     \n\t"

            "movq %%mm7, %%mm6                      \n\t"
            "punpckhbw %%mm4, %%mm7                 \n\t"
            "punpcklbw %%mm4, %%mm6                 \n\t"

            NEXT //7
            "mov %4, %0                             \n\t"
            "add %1, %0                             \n\t"
            PREV //0
            "movq %%mm0, 80(%3)                     \n\t"
            "movq %%mm1, 88(%3)                     \n\t"

            PREV //1
            "paddw %%mm6, %%mm0                     \n\t"
            "paddw %%mm7, %%mm1                     \n\t"
            "movq %%mm0, 96(%3)                     \n\t"
            "movq %%mm1, 104(%3)                    \n\t"

            PREV //2
            "paddw %%mm6, %%mm0                     \n\t"
            "paddw %%mm7, %%mm1                     \n\t"
            "movq %%mm0, 112(%3)                    \n\t"
            "movq %%mm1, 120(%3)                    \n\t"

            PREV //3
            "paddw %%mm6, %%mm0                     \n\t"
            "paddw %%mm7, %%mm1                     \n\t"
            "movq %%mm0, 128(%3)                    \n\t"
            "movq %%mm1, 136(%3)                    \n\t"

            PREV //4
            "paddw %%mm6, %%mm0                     \n\t"
            "paddw %%mm7, %%mm1                     \n\t"
            "movq %%mm0, 144(%3)                    \n\t"
            "movq %%mm1, 152(%3)                    \n\t"

            "mov %4, %0                             \n\t" //FIXME

            : "+&r"(src)
2797
            : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)
2798
              NAMED_CONSTRAINTS_ADD(w04)
2799 2800 2801 2802
        );

        src+= step; // src points to begin of the 8x8 Block

2803
        __asm__ volatile(
2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834
            "movq %4, %%mm6                         \n\t"
            "pcmpeqb %%mm5, %%mm5                   \n\t"
            "pxor %%mm6, %%mm5                      \n\t"
            "pxor %%mm7, %%mm7                      \n\t"

            "1:                                     \n\t"
            "movq (%1), %%mm0                       \n\t"
            "movq 8(%1), %%mm1                      \n\t"
            "paddw 32(%1), %%mm0                    \n\t"
            "paddw 40(%1), %%mm1                    \n\t"
            "movq (%0, %3), %%mm2                   \n\t"
            "movq %%mm2, %%mm3                      \n\t"
            "movq %%mm2, %%mm4                      \n\t"
            "punpcklbw %%mm7, %%mm2                 \n\t"
            "punpckhbw %%mm7, %%mm3                 \n\t"
            "paddw %%mm2, %%mm0                     \n\t"
            "paddw %%mm3, %%mm1                     \n\t"
            "paddw %%mm2, %%mm0                     \n\t"
            "paddw %%mm3, %%mm1                     \n\t"
            "psrlw $4, %%mm0                        \n\t"
            "psrlw $4, %%mm1                        \n\t"
            "packuswb %%mm1, %%mm0                  \n\t"
            "pand %%mm6, %%mm0                      \n\t"
            "pand %%mm5, %%mm4                      \n\t"
            "por %%mm4, %%mm0                       \n\t"
            "movq %%mm0, (%0, %3)                   \n\t"
            "add $16, %1                            \n\t"
            "add %2, %0                             \n\t"
            " js 1b                                 \n\t"

            : "+r"(offset), "+r"(temp_sums)
2835
            : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)
2836 2837 2838 2839 2840 2841
        );
    }else
        src+= step; // src points to begin of the 8x8 Block

    if(eq_mask != -1LL){
        uint8_t *temp_src= src;
2842
        DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars
2843
        __asm__ volatile(
2844
            "pxor %%mm7, %%mm7                      \n\t"
2845 2846 2847
//      0       1       2       3       4       5       6       7       8       9
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1

2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884
            "movq (%0), %%mm0                       \n\t"
            "movq %%mm0, %%mm1                      \n\t"
            "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
            "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0

            "movq (%0, %1), %%mm2                   \n\t"
            "lea (%0, %1, 2), %%"REG_a"             \n\t"
            "movq %%mm2, %%mm3                      \n\t"
            "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
            "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1

            "movq (%%"REG_a"), %%mm4                \n\t"
            "movq %%mm4, %%mm5                      \n\t"
            "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
            "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2

            "paddw %%mm0, %%mm0                     \n\t" // 2L0
            "paddw %%mm1, %%mm1                     \n\t" // 2H0
            "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
            "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2

            "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
            "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2

            "movq (%%"REG_a", %1), %%mm2            \n\t"
            "movq %%mm2, %%mm3                      \n\t"
            "punpcklbw %%mm7, %%mm2                 \n\t" // L3
            "punpckhbw %%mm7, %%mm3                 \n\t" // H3

            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2885 2886
            "movq %%mm0, (%4)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
            "movq %%mm1, 8(%4)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2887 2888 2889 2890 2891 2892 2893 2894

            "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
            "movq %%mm0, %%mm1                      \n\t"
            "punpcklbw %%mm7, %%mm0                 \n\t" // L4
            "punpckhbw %%mm7, %%mm1                 \n\t" // H4

            "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
            "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
2895 2896
            "movq %%mm2, 16(%4)                     \n\t" // L3 - L4
            "movq %%mm3, 24(%4)                     \n\t" // H3 - H4
2897 2898 2899 2900 2901 2902 2903 2904 2905 2906
            "paddw %%mm4, %%mm4                     \n\t" // 2L2
            "paddw %%mm5, %%mm5                     \n\t" // 2H2
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4

            "lea (%%"REG_a", %1), %0                \n\t"
            "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
            "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
2907
//50 opcodes so far
2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943
            "movq (%0, %1, 2), %%mm2                \n\t"
            "movq %%mm2, %%mm3                      \n\t"
            "punpcklbw %%mm7, %%mm2                 \n\t" // L5
            "punpckhbw %%mm7, %%mm3                 \n\t" // H5
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5

            "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
            "punpcklbw %%mm7, %%mm6                 \n\t" // L6
            "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
            "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
            "punpckhbw %%mm7, %%mm6                 \n\t" // H6
            "psubw %%mm6, %%mm3                     \n\t" // H5 - H6

            "paddw %%mm0, %%mm0                     \n\t" // 2L4
            "paddw %%mm1, %%mm1                     \n\t" // 2H4
            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6

            "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
            "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6

            "movq (%0, %1, 4), %%mm2                \n\t"
            "movq %%mm2, %%mm3                      \n\t"
            "punpcklbw %%mm7, %%mm2                 \n\t" // L7
            "punpckhbw %%mm7, %%mm3                 \n\t" // H7

            "paddw %%mm2, %%mm2                     \n\t" // 2L7
            "paddw %%mm3, %%mm3                     \n\t" // 2H7
            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7

2944 2945
            "movq (%4), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
            "movq 8(%4), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2946

2947
#if TEMPLATE_PP_MMXEXT
2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959
            "movq %%mm7, %%mm6                      \n\t" // 0
            "psubw %%mm0, %%mm6                     \n\t"
            "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
            "movq %%mm7, %%mm6                      \n\t" // 0
            "psubw %%mm1, %%mm6                     \n\t"
            "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
            "movq %%mm7, %%mm6                      \n\t" // 0
            "psubw %%mm2, %%mm6                     \n\t"
            "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
            "movq %%mm7, %%mm6                      \n\t" // 0
            "psubw %%mm3, %%mm6                     \n\t"
            "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
2960
#else
2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976
            "movq %%mm7, %%mm6                      \n\t" // 0
            "pcmpgtw %%mm0, %%mm6                   \n\t"
            "pxor %%mm6, %%mm0                      \n\t"
            "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
            "movq %%mm7, %%mm6                      \n\t" // 0
            "pcmpgtw %%mm1, %%mm6                   \n\t"
            "pxor %%mm6, %%mm1                      \n\t"
            "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
            "movq %%mm7, %%mm6                      \n\t" // 0
            "pcmpgtw %%mm2, %%mm6                   \n\t"
            "pxor %%mm6, %%mm2                      \n\t"
            "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
            "movq %%mm7, %%mm6                      \n\t" // 0
            "pcmpgtw %%mm3, %%mm6                   \n\t"
            "pxor %%mm6, %%mm3                      \n\t"
            "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
2977 2978
#endif

2979
#if TEMPLATE_PP_MMXEXT
2980 2981
            "pminsw %%mm2, %%mm0                    \n\t"
            "pminsw %%mm3, %%mm1                    \n\t"
2982
#else
2983 2984 2985 2986 2987 2988
            "movq %%mm0, %%mm6                      \n\t"
            "psubusw %%mm2, %%mm6                   \n\t"
            "psubw %%mm6, %%mm0                     \n\t"
            "movq %%mm1, %%mm6                      \n\t"
            "psubusw %%mm3, %%mm6                   \n\t"
            "psubw %%mm6, %%mm1                     \n\t"
2989 2990
#endif

2991 2992
            "movd %2, %%mm2                         \n\t" // QP
            "punpcklbw %%mm7, %%mm2                 \n\t"
2993

2994 2995 2996 2997 2998 2999 3000
            "movq %%mm7, %%mm6                      \n\t" // 0
            "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
            "pxor %%mm6, %%mm4                      \n\t"
            "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
            "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
            "pxor %%mm7, %%mm5                      \n\t"
            "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
3001
// 100 opcodes
3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022
            "psllw $3, %%mm2                        \n\t" // 8QP
            "movq %%mm2, %%mm3                      \n\t" // 8QP
            "pcmpgtw %%mm4, %%mm2                   \n\t"
            "pcmpgtw %%mm5, %%mm3                   \n\t"
            "pand %%mm2, %%mm4                      \n\t"
            "pand %%mm3, %%mm5                      \n\t"


            "psubusw %%mm0, %%mm4                   \n\t" // hd
            "psubusw %%mm1, %%mm5                   \n\t" // ld


            "movq "MANGLE(w05)", %%mm2              \n\t" // 5
            "pmullw %%mm2, %%mm4                    \n\t"
            "pmullw %%mm2, %%mm5                    \n\t"
            "movq "MANGLE(w20)", %%mm2              \n\t" // 32
            "paddw %%mm2, %%mm4                     \n\t"
            "paddw %%mm2, %%mm5                     \n\t"
            "psrlw $6, %%mm4                        \n\t"
            "psrlw $6, %%mm5                        \n\t"

3023 3024
            "movq 16(%4), %%mm0                     \n\t" // L3 - L4
            "movq 24(%4), %%mm1                     \n\t" // H3 - H4
3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041

            "pxor %%mm2, %%mm2                      \n\t"
            "pxor %%mm3, %%mm3                      \n\t"

            "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
            "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
            "pxor %%mm2, %%mm0                      \n\t"
            "pxor %%mm3, %%mm1                      \n\t"
            "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
            "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
            "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
            "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2

            "pxor %%mm6, %%mm2                      \n\t"
            "pxor %%mm7, %%mm3                      \n\t"
            "pand %%mm2, %%mm4                      \n\t"
            "pand %%mm3, %%mm5                      \n\t"
3042

3043
#if TEMPLATE_PP_MMXEXT
3044 3045
            "pminsw %%mm0, %%mm4                    \n\t"
            "pminsw %%mm1, %%mm5                    \n\t"
3046
#else
3047 3048 3049 3050 3051 3052
            "movq %%mm4, %%mm2                      \n\t"
            "psubusw %%mm0, %%mm2                   \n\t"
            "psubw %%mm2, %%mm4                     \n\t"
            "movq %%mm5, %%mm2                      \n\t"
            "psubusw %%mm1, %%mm2                   \n\t"
            "psubw %%mm2, %%mm5                     \n\t"
3053
#endif
3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068
            "pxor %%mm6, %%mm4                      \n\t"
            "pxor %%mm7, %%mm5                      \n\t"
            "psubw %%mm6, %%mm4                     \n\t"
            "psubw %%mm7, %%mm5                     \n\t"
            "packsswb %%mm5, %%mm4                  \n\t"
            "movq %3, %%mm1                         \n\t"
            "pandn %%mm4, %%mm1                     \n\t"
            "movq (%0), %%mm0                       \n\t"
            "paddb   %%mm1, %%mm0                   \n\t"
            "movq %%mm0, (%0)                       \n\t"
            "movq (%0, %1), %%mm0                   \n\t"
            "psubb %%mm1, %%mm0                     \n\t"
            "movq %%mm0, (%0, %1)                   \n\t"

            : "+r" (temp_src)
3069
            : "r" ((x86_reg)step), "m" (c->pQPb), "m"(eq_mask), "r"(tmp)
3070
              NAMED_CONSTRAINTS_ADD(w05,w20)
3071
            : "%"REG_a
3072 3073
        );
    }
3074 3075 3076 3077
/*if(step==16){
    STOP_TIMER("step16")
}else{
    STOP_TIMER("stepX")
3078 3079
}
    } */
3080
}
3081
#endif //TEMPLATE_PP_MMX
3082

Michael Niedermayer's avatar
Michael Niedermayer committed
3083
static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3084
                                const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
3085

3086
/**
3087
 * Copy a block from src to dst and fixes the blacklevel.
3088
 * levelFix == 0 -> do not touch the brightness & contrast
3089
 */
3090
#undef REAL_SCALED_CPY
3091 3092
#undef SCALED_CPY

Michael Niedermayer's avatar
Michael Niedermayer committed
3093
static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
3094
                                     int levelFix, int64_t *packedOffsetAndScale)
3095
{
3096
#if !TEMPLATE_PP_MMX || !HAVE_6REGS
3097
    int i;
3098
#endif
3099
    if(levelFix){
3100
#if TEMPLATE_PP_MMX && HAVE_6REGS
3101
    __asm__ volatile(
3102 3103 3104 3105 3106
        "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
        "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
        "lea (%2,%4), %%"REG_a"         \n\t"
        "lea (%3,%5), %%"REG_d"         \n\t"
        "pxor %%mm4, %%mm4              \n\t"
3107
#if TEMPLATE_PP_MMXEXT
3108
#define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \
3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128
        "movq " #src1 ", %%mm0          \n\t"\
        "movq " #src1 ", %%mm5          \n\t"\
        "movq " #src2 ", %%mm1          \n\t"\
        "movq " #src2 ", %%mm6          \n\t"\
        "punpcklbw %%mm0, %%mm0         \n\t"\
        "punpckhbw %%mm5, %%mm5         \n\t"\
        "punpcklbw %%mm1, %%mm1         \n\t"\
        "punpckhbw %%mm6, %%mm6         \n\t"\
        "pmulhuw %%mm3, %%mm0           \n\t"\
        "pmulhuw %%mm3, %%mm5           \n\t"\
        "pmulhuw %%mm3, %%mm1           \n\t"\
        "pmulhuw %%mm3, %%mm6           \n\t"\
        "psubw %%mm2, %%mm0             \n\t"\
        "psubw %%mm2, %%mm5             \n\t"\
        "psubw %%mm2, %%mm1             \n\t"\
        "psubw %%mm2, %%mm6             \n\t"\
        "packuswb %%mm5, %%mm0          \n\t"\
        "packuswb %%mm6, %%mm1          \n\t"\
        "movq %%mm0, " #dst1 "          \n\t"\
        "movq %%mm1, " #dst2 "          \n\t"\
3129

3130
#else //TEMPLATE_PP_MMXEXT
3131
#define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \
3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155
        "movq " #src1 ", %%mm0          \n\t"\
        "movq " #src1 ", %%mm5          \n\t"\
        "punpcklbw %%mm4, %%mm0         \n\t"\
        "punpckhbw %%mm4, %%mm5         \n\t"\
        "psubw %%mm2, %%mm0             \n\t"\
        "psubw %%mm2, %%mm5             \n\t"\
        "movq " #src2 ", %%mm1          \n\t"\
        "psllw $6, %%mm0                \n\t"\
        "psllw $6, %%mm5                \n\t"\
        "pmulhw %%mm3, %%mm0            \n\t"\
        "movq " #src2 ", %%mm6          \n\t"\
        "pmulhw %%mm3, %%mm5            \n\t"\
        "punpcklbw %%mm4, %%mm1         \n\t"\
        "punpckhbw %%mm4, %%mm6         \n\t"\
        "psubw %%mm2, %%mm1             \n\t"\
        "psubw %%mm2, %%mm6             \n\t"\
        "psllw $6, %%mm1                \n\t"\
        "psllw $6, %%mm6                \n\t"\
        "pmulhw %%mm3, %%mm1            \n\t"\
        "pmulhw %%mm3, %%mm6            \n\t"\
        "packuswb %%mm5, %%mm0          \n\t"\
        "packuswb %%mm6, %%mm1          \n\t"\
        "movq %%mm0, " #dst1 "          \n\t"\
        "movq %%mm1, " #dst2 "          \n\t"\
Michael Niedermayer's avatar
Michael Niedermayer committed
3156

3157
#endif //TEMPLATE_PP_MMXEXT
3158 3159
#define SCALED_CPY(src1, src2, dst1, dst2)\
   REAL_SCALED_CPY(src1, src2, dst1, dst2)
3160

3161
SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
3162 3163
SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
3164 3165
        "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
        "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
3166
SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
Michael Niedermayer's avatar
Michael Niedermayer committed
3167 3168


3169 3170 3171 3172
        : "=&a" (packedOffsetAndScale)
        : "0" (packedOffsetAndScale),
        "r"(src),
        "r"(dst),
3173 3174
        "r" ((x86_reg)srcStride),
        "r" ((x86_reg)dstStride)
3175 3176
        : "%"REG_d
    );
3177
#else //TEMPLATE_PP_MMX && HAVE_6REGS
3178 3179 3180
    for(i=0; i<8; i++)
        memcpy( &(dst[dstStride*i]),
                &(src[srcStride*i]), BLOCK_SIZE);
3181
#endif //TEMPLATE_PP_MMX && HAVE_6REGS
3182
    }else{
3183
#if TEMPLATE_PP_MMX && HAVE_6REGS
3184
    __asm__ volatile(
3185 3186
        "lea (%0,%2), %%"REG_a"                 \n\t"
        "lea (%1,%3), %%"REG_d"                 \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
3187

3188
#define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \
3189 3190 3191 3192
        "movq " #src1 ", %%mm0          \n\t"\
        "movq " #src2 ", %%mm1          \n\t"\
        "movq %%mm0, " #dst1 "          \n\t"\
        "movq %%mm1, " #dst2 "          \n\t"\
Michael Niedermayer's avatar
Michael Niedermayer committed
3193

3194 3195 3196
#define SIMPLE_CPY(src1, src2, dst1, dst2)\
   REAL_SIMPLE_CPY(src1, src2, dst1, dst2)

3197
SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))
3198 3199
SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
3200 3201
        "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
        "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
3202
SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
Michael Niedermayer's avatar
Michael Niedermayer committed
3203

3204 3205
        : : "r" (src),
        "r" (dst),
3206 3207
        "r" ((x86_reg)srcStride),
        "r" ((x86_reg)dstStride)
3208 3209
        : "%"REG_a, "%"REG_d
    );
3210
#else //TEMPLATE_PP_MMX && HAVE_6REGS
3211 3212 3213
    for(i=0; i<8; i++)
        memcpy( &(dst[dstStride*i]),
                &(src[srcStride*i]), BLOCK_SIZE);
3214
#endif //TEMPLATE_PP_MMX && HAVE_6REGS
3215
    }
3216 3217
}

Michael Niedermayer's avatar
Michael Niedermayer committed
3218
/**
3219
 * Duplicate the given 8 src pixels ? times upward
Michael Niedermayer's avatar
Michael Niedermayer committed
3220 3221 3222
 */
static inline void RENAME(duplicate)(uint8_t src[], int stride)
{
3223
#if TEMPLATE_PP_MMX
3224
    __asm__ volatile(
3225
        "movq (%0), %%mm0               \n\t"
3226
        "movq %%mm0, (%0, %1, 4)        \n\t"
3227 3228 3229 3230
        "add %1, %0                     \n\t"
        "movq %%mm0, (%0)               \n\t"
        "movq %%mm0, (%0, %1)           \n\t"
        "movq %%mm0, (%0, %1, 2)        \n\t"
3231
        "movq %%mm0, (%0, %1, 4)        \n\t"
3232
        : "+r" (src)
3233
        : "r" ((x86_reg)-stride)
3234
    );
Michael Niedermayer's avatar
Michael Niedermayer committed
3235
#else
3236 3237
    int i;
    uint8_t *p=src;
3238
    for(i=0; i<5; i++){
3239 3240 3241
        p-= stride;
        memcpy(p, src, 8);
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
3242 3243
#endif
}
3244

3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307
#if ARCH_X86 && TEMPLATE_PP_MMXEXT
static inline void RENAME(prefetchnta)(const void *p)
{
    __asm__ volatile(   "prefetchnta (%0)\n\t"
        : : "r" (p)
    );
}

static inline void RENAME(prefetcht0)(const void *p)
{
    __asm__ volatile(   "prefetcht0 (%0)\n\t"
        : : "r" (p)
    );
}

static inline void RENAME(prefetcht1)(const void *p)
{
    __asm__ volatile(   "prefetcht1 (%0)\n\t"
        : : "r" (p)
    );
}

static inline void RENAME(prefetcht2)(const void *p)
{
    __asm__ volatile(   "prefetcht2 (%0)\n\t"
        : : "r" (p)
    );
}
#elif !ARCH_X86 && AV_GCC_VERSION_AT_LEAST(3,2)
static inline void RENAME(prefetchnta)(const void *p)
{
    __builtin_prefetch(p,0,0);
}
static inline void RENAME(prefetcht0)(const void *p)
{
    __builtin_prefetch(p,0,1);
}
static inline void RENAME(prefetcht1)(const void *p)
{
    __builtin_prefetch(p,0,2);
}
static inline void RENAME(prefetcht2)(const void *p)
{
    __builtin_prefetch(p,0,3);
}
#else
static inline void RENAME(prefetchnta)(const void *p)
{
    return;
}
static inline void RENAME(prefetcht0)(const void *p)
{
    return;
}
static inline void RENAME(prefetcht1)(const void *p)
{
    return;
}
static inline void RENAME(prefetcht2)(const void *p)
{
    return;
}
#endif
3308
/**
3309
 * Filter array of bytes (Y or U or V values)
3310
 */
Michael Niedermayer's avatar
Michael Niedermayer committed
3311
static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3312
                                const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
3313
{
3314 3315
    DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
    int x,y;
3316 3317
#ifdef TEMPLATE_PP_TIME_MODE
    const int mode= TEMPLATE_PP_TIME_MODE;
3318
#else
3319
    const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3320
#endif
3321 3322
    int black=0, white=255; // blackest black and whitest white in the picture
    int QPCorrecture= 256*256;
3323

3324
    int copyAhead;
3325
#if TEMPLATE_PP_MMX
3326
    int i;
3327
#endif
Michael Niedermayer's avatar
Michael Niedermayer committed
3328

3329 3330
    const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
    const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
3331

3332 3333 3334
    //FIXME remove
    uint64_t * const yHistogram= c.yHistogram;
    uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
3335
    uint8_t * const tempDst= (dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride) + 32;
3336
    //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
3337

3338 3339 3340 3341 3342 3343
    if (mode & VISUALIZE){
        if(!(mode & (V_A_DEBLOCK | H_A_DEBLOCK)) || TEMPLATE_PP_MMX) {
            av_log(c2, AV_LOG_WARNING, "Visualization is currently only supported with the accurate deblock filter without SIMD\n");
        }
    }

3344
#if TEMPLATE_PP_MMX
3345 3346 3347 3348 3349 3350 3351 3352
    for(i=0; i<57; i++){
        int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
        int threshold= offset*2 + 1;
        c.mmxDcOffset[i]= 0x7F - offset;
        c.mmxDcThreshold[i]= 0x7F - threshold;
        c.mmxDcOffset[i]*= 0x0101010101010101LL;
        c.mmxDcThreshold[i]*= 0x0101010101010101LL;
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
3353
#endif
3354

3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374
    if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
    else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
            || (mode & FFMPEG_DEINT_FILTER)
            || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
    else if(   (mode & V_DEBLOCK)
            || (mode & LINEAR_IPOL_DEINT_FILTER)
            || (mode & MEDIAN_DEINT_FILTER)
            || (mode & V_A_DEBLOCK)) copyAhead=13;
    else if(mode & V_X1_FILTER) copyAhead=11;
//    else if(mode & V_RK1_FILTER) copyAhead=10;
    else if(mode & DERING) copyAhead=9;
    else copyAhead=8;

    copyAhead-= 8;

    if(!isColor){
        uint64_t sum= 0;
        int i;
        uint64_t maxClipped;
        uint64_t clipped;
3375
        AVRational scale;
3376

3377 3378
        c.frameNum++;
        // first frame is fscked so we ignore it
3379
        if(c.frameNum == 1) yHistogram[0]= width*(uint64_t)height/64*15/256;
3380

3381 3382 3383
        for(i=0; i<256; i++){
            sum+= yHistogram[i];
        }
3384

3385
        /* We always get a completely black picture first. */
3386
        maxClipped= av_rescale(sum, c.ppMode.maxClippedThreshold.num, c.ppMode.maxClippedThreshold.den);
3387

3388 3389 3390 3391 3392
        clipped= sum;
        for(black=255; black>0; black--){
            if(clipped < maxClipped) break;
            clipped-= yHistogram[black];
        }
3393

3394 3395 3396 3397 3398
        clipped= sum;
        for(white=0; white<256; white++){
            if(clipped < maxClipped) break;
            clipped-= yHistogram[white];
        }
3399

3400
        scale = (AVRational){c.ppMode.maxAllowedY - c.ppMode.minAllowedY, white - black};
3401

3402
#if TEMPLATE_PP_MMXEXT
3403
        c.packedYScale = (uint16_t)av_rescale(scale.num, 256, scale.den);
3404
        c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
3405
#else
3406
        c.packedYScale = (uint16_t)av_rescale(scale.num, 1024, scale.den);
3407
        c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
3408 3409
#endif

3410 3411
        c.packedYOffset|= c.packedYOffset<<32;
        c.packedYOffset|= c.packedYOffset<<16;
3412

3413 3414
        c.packedYScale|= c.packedYScale<<32;
        c.packedYScale|= c.packedYScale<<16;
3415

3416
        if(mode & LEVEL_FIX)        QPCorrecture= (int)av_rescale(scale.num, 256*256, scale.den);
3417 3418 3419 3420 3421 3422
        else                        QPCorrecture= 256*256;
    }else{
        c.packedYScale= 0x0100010001000100LL;
        c.packedYOffset= 0;
        QPCorrecture= 256*256;
    }
3423

3424 3425 3426 3427 3428
    /* copy & deinterlace first row of blocks */
    y=-BLOCK_SIZE;
    {
        const uint8_t *srcBlock= &(src[y*srcStride]);
        uint8_t *dstBlock= tempDst + dstStride;
3429

3430 3431
        // From this point on it is guaranteed that we can read and write 16 lines downward
        // finish 1 block before the next otherwise we might have a problem
3432
        // with the L1 Cache of the P4 ... or only a few blocks at a time or something
3433
        for(x=0; x<width; x+=BLOCK_SIZE){
3434 3435 3436 3437
            RENAME(prefetchnta)(srcBlock + (((x>>2)&6) + copyAhead)*srcStride + 32);
            RENAME(prefetchnta)(srcBlock + (((x>>2)&6) + copyAhead+1)*srcStride + 32);
            RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead)*dstStride + 32);
            RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead+1)*dstStride + 32);
Michael Niedermayer's avatar
Michael Niedermayer committed
3438

3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457
            RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
                              srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

            RENAME(duplicate)(dstBlock + dstStride*8, dstStride);

            if(mode & LINEAR_IPOL_DEINT_FILTER)
                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
            else if(mode & LINEAR_BLEND_DEINT_FILTER)
                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
            else if(mode & MEDIAN_DEINT_FILTER)
                RENAME(deInterlaceMedian)(dstBlock, dstStride);
            else if(mode & CUBIC_IPOL_DEINT_FILTER)
                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
            else if(mode & FFMPEG_DEINT_FILTER)
                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
            else if(mode & LOWPASS5_DEINT_FILTER)
                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
/*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
Michael Niedermayer's avatar
Michael Niedermayer committed
3458
*/
3459 3460 3461 3462 3463 3464 3465 3466 3467 3468
            dstBlock+=8;
            srcBlock+=8;
        }
        if(width==FFABS(dstStride))
            linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
        else{
            int i;
            for(i=0; i<copyAhead; i++){
                memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
            }
3469
        }
3470
    }
3471

3472 3473 3474 3475
    for(y=0; y<height; y+=BLOCK_SIZE){
        //1% speedup if these are here instead of the inner loop
        const uint8_t *srcBlock= &(src[y*srcStride]);
        uint8_t *dstBlock= &(dst[y*dstStride]);
3476
#if TEMPLATE_PP_MMX
3477 3478
        uint8_t *tempBlock1= c.tempBlocks;
        uint8_t *tempBlock2= c.tempBlocks + 8;
3479
#endif
3480 3481
        const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
        int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
3482
        int QP=0, nonBQP=0;
3483 3484 3485 3486 3487 3488 3489 3490 3491
        /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
           if not than use a temporary buffer */
        if(y+15 >= height){
            int i;
            /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
               blockcopy to dst later */
            linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
                    FFMAX(height-y-copyAhead, 0), srcStride);

Diego Biurrun's avatar
Diego Biurrun committed
3492
            /* duplicate last line of src to fill the void up to line (copyAhead+7) */
3493 3494 3495 3496 3497 3498
            for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
                    memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));

            /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
            linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);

Diego Biurrun's avatar
Diego Biurrun committed
3499
            /* duplicate last line of dst to fill the void up to line (copyAhead) */
3500 3501 3502 3503 3504 3505
            for(i=height-y+1; i<=copyAhead; i++)
                    memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));

            dstBlock= tempDst + dstStride;
            srcBlock= tempSrc;
        }
3506

3507 3508
        // From this point on it is guaranteed that we can read and write 16 lines downward
        // finish 1 block before the next otherwise we might have a problem
3509
        // with the L1 Cache of the P4 ... or only a few blocks at a time or something
3510 3511 3512
        for(x=0; x<width; ){
            int startx = x;
            int endx = FFMIN(width, x+32);
3513 3514
            uint8_t *dstBlockStart = dstBlock;
            const uint8_t *srcBlockStart = srcBlock;
3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537
            int qp_index = 0;
            for(qp_index=0; qp_index < (endx-startx)/BLOCK_SIZE; qp_index++){
                QP = QPptr[(x+qp_index*BLOCK_SIZE)>>qpHShift];
                nonBQP = nonBQPptr[(x+qp_index*BLOCK_SIZE)>>qpHShift];
            if(!isColor){
                QP= (QP* QPCorrecture + 256*128)>>16;
                nonBQP= (nonBQP* QPCorrecture + 256*128)>>16;
                yHistogram[(srcBlock+qp_index*8)[srcStride*12 + 4]]++;
            }
            c.QP_block[qp_index] = QP;
            c.nonBQP_block[qp_index] = nonBQP;
#if TEMPLATE_PP_MMX
            __asm__ volatile(
                "movd %1, %%mm7         \n\t"
                "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
                "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
                "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP
                "movq %%mm7, %0         \n\t"
                : "=m" (c.pQPb_block[qp_index])
                : "r" (QP)
            );
#endif
            }
3538
          for(; x < endx; x+=BLOCK_SIZE){
3539 3540 3541 3542
            RENAME(prefetchnta)(srcBlock + (((x>>2)&6) + copyAhead)*srcStride + 32);
            RENAME(prefetchnta)(srcBlock + (((x>>2)&6) + copyAhead+1)*srcStride + 32);
            RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead)*dstStride + 32);
            RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead+1)*dstStride + 32);
3543

3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560
            RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
                              srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

            if(mode & LINEAR_IPOL_DEINT_FILTER)
                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
            else if(mode & LINEAR_BLEND_DEINT_FILTER)
                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
            else if(mode & MEDIAN_DEINT_FILTER)
                RENAME(deInterlaceMedian)(dstBlock, dstStride);
            else if(mode & CUBIC_IPOL_DEINT_FILTER)
                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
            else if(mode & FFMPEG_DEINT_FILTER)
                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
            else if(mode & LOWPASS5_DEINT_FILTER)
                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
/*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3561
*/
3562 3563 3564 3565 3566 3567 3568
            dstBlock+=8;
            srcBlock+=8;
          }

          dstBlock = dstBlockStart;
          srcBlock = srcBlockStart;

3569
          for(x = startx, qp_index = 0; x < endx; x+=BLOCK_SIZE, qp_index++){
3570
            const int stride= dstStride;
3571 3572 3573 3574 3575 3576 3577
            //temporary while changing QP stuff to make things continue to work
            //eventually QP,nonBQP,etc will be arrays and this will be unnecessary
            c.QP = c.QP_block[qp_index];
            c.nonBQP = c.nonBQP_block[qp_index];
            c.pQPb = c.pQPb_block[qp_index];
            c.pQPb2 = c.pQPb2_block[qp_index];

3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589
            /* only deblock if we have 2 blocks */
            if(y + 8 < height){
                if(mode & V_X1_FILTER)
                    RENAME(vertX1Filter)(dstBlock, stride, &c);
                else if(mode & V_DEBLOCK){
                    const int t= RENAME(vertClassify)(dstBlock, stride, &c);

                    if(t==1)
                        RENAME(doVertLowPass)(dstBlock, stride, &c);
                    else if(t==2)
                        RENAME(doVertDefFilter)(dstBlock, stride, &c);
                }else if(mode & V_A_DEBLOCK){
3590
                    RENAME(do_a_deblock)(dstBlock, stride, 1, &c, mode);
3591 3592
                }
            }
Michael Niedermayer's avatar
Michael Niedermayer committed
3593

3594 3595 3596 3597 3598 3599 3600
            dstBlock+=8;
            srcBlock+=8;
          }

          dstBlock = dstBlockStart;
          srcBlock = srcBlockStart;

3601
          for(x = startx, qp_index=0; x < endx; x+=BLOCK_SIZE, qp_index++){
3602 3603
            const int stride= dstStride;
            av_unused uint8_t *tmpXchg;
3604 3605 3606 3607
            c.QP = c.QP_block[qp_index];
            c.nonBQP = c.nonBQP_block[qp_index];
            c.pQPb = c.pQPb_block[qp_index];
            c.pQPb2 = c.pQPb2_block[qp_index];
3608
#if TEMPLATE_PP_MMX
3609
            RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3610
#endif
3611 3612
            /* check if we have a previous block to deblock it with dstBlock */
            if(x - 8 >= 0){
3613
#if TEMPLATE_PP_MMX
3614 3615 3616 3617 3618 3619 3620 3621 3622
                if(mode & H_X1_FILTER)
                        RENAME(vertX1Filter)(tempBlock1, 16, &c);
                else if(mode & H_DEBLOCK){
                    const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
                    if(t==1)
                        RENAME(doVertLowPass)(tempBlock1, 16, &c);
                    else if(t==2)
                        RENAME(doVertDefFilter)(tempBlock1, 16, &c);
                }else if(mode & H_A_DEBLOCK){
3623
                        RENAME(do_a_deblock)(tempBlock1, 16, 1, &c, mode);
3624
                }
3625

3626
                RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3627 3628

#else
3629
                if(mode & H_X1_FILTER)
3630
                    horizX1Filter(dstBlock-4, stride, c.QP);
3631
                else if(mode & H_DEBLOCK){
3632
#if TEMPLATE_PP_ALTIVEC
3633
                    DECLARE_ALIGNED(16, unsigned char, tempBlock)[272];
3634
                    int t;
3635 3636
                    transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);

3637
                    t = vertClassify_altivec(tempBlock-48, 16, &c);
3638 3639 3640 3641 3642 3643 3644 3645
                    if(t==1) {
                        doVertLowPass_altivec(tempBlock-48, 16, &c);
                        transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
                    }
                    else if(t==2) {
                        doVertDefFilter_altivec(tempBlock-48, 16, &c);
                        transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
                    }
3646
#else
3647
                    const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
3648

3649 3650 3651 3652
                    if(t==1)
                        RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
                    else if(t==2)
                        RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
3653
#endif
3654
                }else if(mode & H_A_DEBLOCK){
3655
                    RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c, mode);
3656
                }
3657
#endif //TEMPLATE_PP_MMX
3658 3659 3660 3661
                if(mode & DERING){
                //FIXME filter first line
                    if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
                }
3662

3663 3664 3665
                if(mode & TEMP_NOISE_FILTER)
                {
                    RENAME(tempNoiseReducer)(dstBlock-8, stride,
3666
                            c.tempBlurred[isColor] + y*dstStride + x,
3667
                            c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256,
3668 3669 3670 3671 3672 3673
                            c.ppMode.maxTmpNoise);
                }
            }

            dstBlock+=8;
            srcBlock+=8;
3674

3675
#if TEMPLATE_PP_MMX
3676 3677 3678
            tmpXchg= tempBlock1;
            tempBlock1= tempBlock2;
            tempBlock2 = tmpXchg;
3679
#endif
3680
          }
3681
        }
3682

3683 3684 3685
        if(mode & DERING){
            if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
        }
3686

3687 3688
        if((mode & TEMP_NOISE_FILTER)){
            RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3689
                    c.tempBlurred[isColor] + y*dstStride + x,
3690
                    c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256,
3691 3692
                    c.ppMode.maxTmpNoise);
        }
3693

3694 3695 3696 3697 3698 3699 3700 3701 3702
        /* did we use a tmp buffer for the last lines*/
        if(y+15 >= height){
            uint8_t *dstBlock= &(dst[y*dstStride]);
            if(width==FFABS(dstStride))
                linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
            else{
                int i;
                for(i=0; i<height-y; i++){
                    memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3703
                }
3704
            }
3705
        }
3706
    }
3707
#if   TEMPLATE_PP_3DNOW
3708
    __asm__ volatile("femms");
3709
#elif TEMPLATE_PP_MMX
3710
    __asm__ volatile("emms");
3711 3712
#endif

3713
#ifdef DEBUG_BRIGHTNESS
3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727
    if(!isColor){
        int max=1;
        int i;
        for(i=0; i<256; i++)
            if(yHistogram[i] > max) max=yHistogram[i];

        for(i=1; i<256; i++){
            int x;
            int start=yHistogram[i-1]/(max/256+1);
            int end=yHistogram[i]/(max/256+1);
            int inc= end > start ? 1 : -1;
            for(x=start; x!=end+inc; x+=inc)
                dst[ i*dstStride + x]+=128;
        }
3728

3729 3730 3731
        for(i=0; i<100; i+=2){
            dst[ (white)*dstStride + i]+=128;
            dst[ (black)*dstStride + i]+=128;
3732
        }
3733
    }
3734 3735
#endif

3736
    *c2= c; //copy local context back
3737

3738
}
3739 3740 3741 3742 3743 3744 3745

#undef RENAME
#undef TEMPLATE_PP_C
#undef TEMPLATE_PP_ALTIVEC
#undef TEMPLATE_PP_MMX
#undef TEMPLATE_PP_MMXEXT
#undef TEMPLATE_PP_3DNOW
3746
#undef TEMPLATE_PP_SSE2