postprocess_template.c 157 KB
Newer Older
1
/*
2 3
 * Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
 *
4
 * This file is part of Libav.
5
 *
6
 * Libav is free software; you can redistribute it and/or modify
7 8 9 10
 * 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.
 *
11
 * Libav is distributed in the hope that it will be useful,
12 13 14 15 16
 * 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
17
 * along with Libav; if not, write to the Free Software
18
 * 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_cpu.h"
Michael Niedermayer's avatar
Michael Niedermayer committed
27

28
#define ALIGN_MASK "$-8"
29

30
#undef REAL_PAVGB
Michael Niedermayer's avatar
Michael Niedermayer committed
31 32 33
#undef PAVGB
#undef PMINUB
#undef PMAXUB
Arpi's avatar
Arpi committed
34

35
#if   HAVE_MMX2
36
#define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
37
#elif HAVE_AMD3DNOW
D Richard Felker III's avatar
1000l  
D Richard Felker III committed
38
#define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
Arpi's avatar
Arpi committed
39
#endif
40
#define PAVGB(a,b)  REAL_PAVGB(a,b)
41

42
#if   HAVE_MMX2
Michael Niedermayer's avatar
Michael Niedermayer committed
43
#define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
44
#elif HAVE_MMX
Michael Niedermayer's avatar
Michael Niedermayer committed
45
#define PMINUB(b,a,t) \
46 47 48
    "movq " #a ", " #t " \n\t"\
    "psubusb " #b ", " #t " \n\t"\
    "psubb " #t ", " #a " \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
49 50
#endif

51
#if   HAVE_MMX2
Michael Niedermayer's avatar
Michael Niedermayer committed
52
#define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
53
#elif HAVE_MMX
Michael Niedermayer's avatar
Michael Niedermayer committed
54
#define PMAXUB(a,b) \
55 56
    "psubusb " #a ", " #b " \n\t"\
    "paddb " #a ", " #b " \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
57 58
#endif

Diego Biurrun's avatar
Diego Biurrun committed
59
//FIXME? |255-0| = 1 (should not be a problem ...)
60
#if HAVE_MMX
61
/**
62
 * Check if the middle 8x8 Block in the given 8x16 block is flat
63
 */
64
static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
65 66
    int numEq= 0, dcOk;
    src+= stride*4; // src points to begin of the 8x8 Block
67
    __asm__ volatile(
68 69 70 71 72
        "movq %0, %%mm7                         \n\t"
        "movq %1, %%mm6                         \n\t"
        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
        );

73
    __asm__ volatile(
74
        "lea (%2, %3), %%"REG_a"                \n\t"
75 76 77
//      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

78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
        "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)
        "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
        "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"
140
#if HAVE_MMX2
141 142
        "pxor %%mm7, %%mm7                      \n\t"
        "psadbw %%mm7, %%mm0                    \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
143
#else
144 145 146 147 148 149 150 151 152
        "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
153
#endif
154 155 156 157 158 159 160 161
        "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)
162
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
163 164 165 166 167 168 169 170 171 172
        : "%"REG_a
        );

    numEq= (-numEq) &0xFF;
    if(numEq > c->ppMode.flatnessThreshold){
        if(dcOk) return 0;
        else     return 1;
    }else{
        return 2;
    }
173
}
174
#endif //HAVE_MMX
175 176

/**
177
 * 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
178
 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
179
 */
180
#if !HAVE_ALTIVEC
Michael Niedermayer's avatar
Michael Niedermayer committed
181
static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
182
{
183
#if HAVE_MMX2 || HAVE_AMD3DNOW
184
    src+= stride*3;
185
    __asm__ volatile(        //"movv %0 %1 %2\n\t"
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306
        "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"

        :
307
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
308 309
        : "%"REG_a, "%"REG_c
    );
310
#else //HAVE_MMX2 || HAVE_AMD3DNOW
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
    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++;
    }
349
#endif //HAVE_MMX2 || HAVE_AMD3DNOW
350
}
351
#endif //HAVE_ALTIVEC
352

353 354
/**
 * Experimental Filter 1
355
 * will not damage linear gradients
Diego Biurrun's avatar
Diego Biurrun committed
356
 * 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
357 358
 * 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
359
 */
Michael Niedermayer's avatar
Michael Niedermayer committed
360
static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
361
{
362
#if HAVE_MMX2 || HAVE_AMD3DNOW
363
    src+= stride*3;
364

365
    __asm__ volatile(
366 367 368
        "pxor %%mm7, %%mm7                      \n\t" // 0
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
369 370
//      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
371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444
        "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

        :
445
        : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
446 447
        : "%"REG_a, "%"REG_c
    );
448
#else //HAVE_MMX2 || HAVE_AMD3DNOW
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
    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;
479
        }
480 481
        src++;
    }
482
#endif //HAVE_MMX2 || HAVE_AMD3DNOW
483 484
}

485
#if !HAVE_ALTIVEC
Michael Niedermayer's avatar
Michael Niedermayer committed
486
static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
487
{
488
#if HAVE_MMX2 || HAVE_AMD3DNOW
489
/*
490 491 492 493 494 495 496 497 498 499 500 501
    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);
502
*/
503
    src+= stride*4;
504
    __asm__ volatile(
505

Diego Biurrun's avatar
Diego Biurrun committed
506
#if 0 //slightly more accurate and slightly slower
507 508 509
        "pxor %%mm7, %%mm7                      \n\t" // 0
        "lea (%0, %1), %%"REG_a"                \n\t"
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
510 511 512 513 514
//      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


515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532
        "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
533 534
// mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0

535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551
        "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)
552 553
// mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0

554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569
        "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
570 571 572
// mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0


573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610
        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"
611
#endif //0
612

613 614
        "lea (%0, %1), %%"REG_a"                \n\t"
        "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
615 616 617
//      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
618 619


620 621 622 623
        "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
624 625
// mm1=-l3-1, mm0=128-q

626 627 628 629 630 631 632 633 634 635
        "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
636 637
// mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1

638 639 640 641 642 643 644 645
        "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
646 647
// mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1

648 649 650 651 652 653 654 655
        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
656 657
// mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128

658 659 660 661 662 663 664
        "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
665 666 667

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

668 669 670 671 672 673 674 675 676 677 678
        "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
679 680
// mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16

681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709
        "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"

        :
710
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
711 712
        : "%"REG_a, "%"REG_c
    );
713 714

/*
715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740
    {
    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;
741
        }
742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764
        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);
            }
765
        }
766
    }
767 768
}
*/
769
#elif HAVE_MMX
770
    src+= stride*4;
771
    __asm__ volatile(
772 773 774
        "pxor %%mm7, %%mm7                      \n\t"
        "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
        "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
775 776 777 778
//      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

779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837
        "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
        "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
        "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3

        "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
        "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
        "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
        "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
838
//50 opcodes so far
839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876
        "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

        "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
        "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
Michael Niedermayer's avatar
Michael Niedermayer committed
877

878
#if HAVE_MMX2
879 880 881 882 883 884 885 886 887 888 889 890
        "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
891
#else
892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907
        "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
908
#endif
909

910
#if HAVE_MMX2
911 912
        "pminsw %%mm2, %%mm0                    \n\t"
        "pminsw %%mm3, %%mm1                    \n\t"
913
#else
914 915 916 917 918 919
        "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"
920 921
#endif

922 923
        "movd %2, %%mm2                         \n\t" // QP
        "punpcklbw %%mm7, %%mm2                 \n\t"
924

925 926 927 928 929 930 931
        "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|
932
// 100 opcodes
933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972
        "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"

        "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
        "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4

        "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"
973

974
#if HAVE_MMX2
975 976
        "pminsw %%mm0, %%mm4                    \n\t"
        "pminsw %%mm1, %%mm5                    \n\t"
977
#else
978 979 980 981 982 983
        "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"
984
#endif
985 986 987 988 989 990 991 992 993 994 995 996 997
        "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)
998
        : "r" ((x86_reg)stride), "m" (c->pQPb)
999 1000
        : "%"REG_a, "%"REG_c
    );
1001
#else //HAVE_MMX2 || HAVE_AMD3DNOW
1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
    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){
                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;
1036
        }
1037 1038
        src++;
    }
1039
#endif //HAVE_MMX2 || HAVE_AMD3DNOW
1040
}
1041
#endif //HAVE_ALTIVEC
1042

1043
#if !HAVE_ALTIVEC
Michael Niedermayer's avatar
Michael Niedermayer committed
1044
static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1045
{
1046
#if HAVE_MMX2 || HAVE_AMD3DNOW
1047
    __asm__ volatile(
1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058
        "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"
1059 1060 1061

//        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
1062

1063
#undef REAL_FIND_MIN_MAX
Michael Niedermayer's avatar
Michael Niedermayer committed
1064
#undef FIND_MIN_MAX
1065
#if HAVE_MMX2
1066
#define REAL_FIND_MIN_MAX(addr)\
1067 1068 1069
        "movq " #addr ", %%mm0                  \n\t"\
        "pminub %%mm0, %%mm7                    \n\t"\
        "pmaxub %%mm0, %%mm6                    \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1070
#else
1071
#define REAL_FIND_MIN_MAX(addr)\
1072 1073 1074 1075 1076 1077
        "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
1078
#endif
1079
#define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
1080

1081 1082 1083
FIND_MIN_MAX((%%REGa))
FIND_MIN_MAX((%%REGa, %1))
FIND_MIN_MAX((%%REGa, %1, 2))
Michael Niedermayer's avatar
Michael Niedermayer committed
1084
FIND_MIN_MAX((%0, %1, 4))
1085 1086 1087
FIND_MIN_MAX((%%REGd))
FIND_MIN_MAX((%%REGd, %1))
FIND_MIN_MAX((%%REGd, %1, 2))
Michael Niedermayer's avatar
Michael Niedermayer committed
1088
FIND_MIN_MAX((%0, %1, 8))
1089

1090 1091
        "movq %%mm7, %%mm4                      \n\t"
        "psrlq $8, %%mm7                        \n\t"
1092
#if HAVE_MMX2
1093 1094 1095 1096 1097
        "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"
1098
#else
1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111
        "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"
1112
#endif
Michael Niedermayer's avatar
Michael Niedermayer committed
1113 1114


1115 1116
        "movq %%mm6, %%mm4                      \n\t"
        "psrlq $8, %%mm6                        \n\t"
1117
#if HAVE_MMX2
1118 1119 1120 1121 1122
        "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
1123
#else
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133
        "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
1134
#endif
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
        "movq %%mm6, %%mm0                      \n\t" // max
        "psubb %%mm7, %%mm6                     \n\t" // max - min
        "movd %%mm6, %%ecx                      \n\t"
        "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
        " jb 1f                                 \n\t"
        "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
        "and "ALIGN_MASK", %%"REG_c"            \n\t"
        PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "punpcklbw %%mm7, %%mm7                 \n\t"
        "movq %%mm7, (%%"REG_c")                \n\t"

        "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
1193
// 0, 2, 3, 1
1194
#define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
        "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)                                     \
        "movq " #lx ", 8(%%"REG_c")             \n\t"\
        "movq (%%"REG_c"), " #lx "              \n\t"\
        "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
1221
\
1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
        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"\
        "movq 8(%%"REG_c"), " #lx "             \n\t"
Michael Niedermayer's avatar
Michael Niedermayer committed
1239

1240 1241
#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
1242 1243 1244
/*
0000000
1111111
1245

Michael Niedermayer's avatar
Michael Niedermayer committed
1246 1247 1248 1249 1250 1251
1111110
1111101
1111100
1111011
1111010
1111001
1252

Michael Niedermayer's avatar
Michael Niedermayer committed
1253 1254
1111000
1110111
1255

Michael Niedermayer's avatar
Michael Niedermayer committed
1256
*/
1257 1258 1259 1260 1261 1262 1263 1264 1265 1266
//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)

Diego Biurrun's avatar