CoCalc -- postprocess

05. Matplotlib / ffmpeg-3.0 / libpostproc / postprocess_template.c
⁵²⁸⁶⁷ views
1
/*
2
 * Copyright (C) 2001-2002 Michael Niedermayer ([email protected])
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or modify
7
 * it under the terms of the GNU General Public License as published by
8
 * the Free Software Foundation; either version 2 of the License, or
9
 * (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14
 * GNU General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU General Public License
17
 * along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20

21
/**
22
 * @file
23
 * mmx/mmx2/3dnow postprocess code.
24
 */
25

26
#include "libavutil/x86/asm.h"
27

28
/* A single TEMPLATE_PP_* should be defined (to 1) when this template is
29
 * included. The following macros will define its dependencies to 1 as well
30
 * (like MMX2 depending on MMX), and will define to 0 all the others. Every
31
 * TEMPLATE_PP_* need to be undef at the end. */
32

33
#ifdef TEMPLATE_PP_C
34
#   define RENAME(a) a ## _C
35
#else
36
#   define TEMPLATE_PP_C 0
37
#endif
38

39
#ifdef TEMPLATE_PP_ALTIVEC
40
#   define RENAME(a) a ## _altivec
41
#else
42
#   define TEMPLATE_PP_ALTIVEC 0
43
#endif
44

45
#ifdef TEMPLATE_PP_MMX
46
#   define RENAME(a) a ## _MMX
47
#else
48
#   define TEMPLATE_PP_MMX 0
49
#endif
50

51
#ifdef TEMPLATE_PP_MMXEXT
52
#   undef  TEMPLATE_PP_MMX
53
#   define TEMPLATE_PP_MMX 1
54
#   define RENAME(a) a ## _MMX2
55
#else
56
#   define TEMPLATE_PP_MMXEXT 0
57
#endif
58

59
#ifdef TEMPLATE_PP_3DNOW
60
#   undef  TEMPLATE_PP_MMX
61
#   define TEMPLATE_PP_MMX 1
62
#   define RENAME(a) a ## _3DNow
63
#else
64
#   define TEMPLATE_PP_3DNOW 0
65
#endif
66

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

77
#undef REAL_PAVGB
78
#undef PAVGB
79
#undef PMINUB
80
#undef PMAXUB
81

82
#if   TEMPLATE_PP_MMXEXT
83
#define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
84
#elif TEMPLATE_PP_3DNOW
85
#define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
86
#endif
87
#define PAVGB(a,b)  REAL_PAVGB(a,b)
88

89
#if   TEMPLATE_PP_MMXEXT
90
#define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
91
#elif TEMPLATE_PP_MMX
92
#define PMINUB(b,a,t) \
93
    "movq " #a ", " #t " \n\t"\
94
    "psubusb " #b ", " #t " \n\t"\
95
    "psubb " #t ", " #a " \n\t"
96
#endif
97

98
#if   TEMPLATE_PP_MMXEXT
99
#define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
100
#elif TEMPLATE_PP_MMX
101
#define PMAXUB(a,b) \
102
    "psubusb " #a ", " #b " \n\t"\
103
    "paddb " #a ", " #b " \n\t"
104
#endif
105

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
    int numEq= 0, dcOk;
113
    src+= stride*4; // src points to begin of the 8x8 Block
114
    __asm__ volatile(
115
        "movq %0, %%mm7                         \n\t"
116
        "movq %1, %%mm6                         \n\t"
117
        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
118
        );
119

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

125
        "movq (%2), %%mm0                       \n\t"
126
        "movq (%%"REG_a"), %%mm1                \n\t"
127
        "movq %%mm0, %%mm3                      \n\t"
128
        "movq %%mm0, %%mm4                      \n\t"
129
        PMAXUB(%%mm1, %%mm4)
130
        PMINUB(%%mm1, %%mm3, %%mm5)
131
        "psubb %%mm1, %%mm0                     \n\t" // mm0 = difference
132
        "paddb %%mm7, %%mm0                     \n\t"
133
        "pcmpgtb %%mm6, %%mm0                   \n\t"
134

135
        "movq (%%"REG_a",%3), %%mm2             \n\t"
136
        PMAXUB(%%mm2, %%mm4)
137
        PMINUB(%%mm2, %%mm3, %%mm5)
138
        "psubb %%mm2, %%mm1                     \n\t"
139
        "paddb %%mm7, %%mm1                     \n\t"
140
        "pcmpgtb %%mm6, %%mm1                   \n\t"
141
        "paddb %%mm1, %%mm0                     \n\t"
142

143
        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
144
        PMAXUB(%%mm1, %%mm4)
145
        PMINUB(%%mm1, %%mm3, %%mm5)
146
        "psubb %%mm1, %%mm2                     \n\t"
147
        "paddb %%mm7, %%mm2                     \n\t"
148
        "pcmpgtb %%mm6, %%mm2                   \n\t"
149
        "paddb %%mm2, %%mm0                     \n\t"
150

151
        "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
152

153
        "movq (%2, %3, 4), %%mm2                \n\t"
154
        PMAXUB(%%mm2, %%mm4)
155
        PMINUB(%%mm2, %%mm3, %%mm5)
156
        "psubb %%mm2, %%mm1                     \n\t"
157
        "paddb %%mm7, %%mm1                     \n\t"
158
        "pcmpgtb %%mm6, %%mm1                   \n\t"
159
        "paddb %%mm1, %%mm0                     \n\t"
160

161
        "movq (%%"REG_a"), %%mm1                \n\t"
162
        PMAXUB(%%mm1, %%mm4)
163
        PMINUB(%%mm1, %%mm3, %%mm5)
164
        "psubb %%mm1, %%mm2                     \n\t"
165
        "paddb %%mm7, %%mm2                     \n\t"
166
        "pcmpgtb %%mm6, %%mm2                   \n\t"
167
        "paddb %%mm2, %%mm0                     \n\t"
168

169
        "movq (%%"REG_a", %3), %%mm2            \n\t"
170
        PMAXUB(%%mm2, %%mm4)
171
        PMINUB(%%mm2, %%mm3, %%mm5)
172
        "psubb %%mm2, %%mm1                     \n\t"
173
        "paddb %%mm7, %%mm1                     \n\t"
174
        "pcmpgtb %%mm6, %%mm1                   \n\t"
175
        "paddb %%mm1, %%mm0                     \n\t"
176

177
        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
178
        PMAXUB(%%mm1, %%mm4)
179
        PMINUB(%%mm1, %%mm3, %%mm5)
180
        "psubb %%mm1, %%mm2                     \n\t"
181
        "paddb %%mm7, %%mm2                     \n\t"
182
        "pcmpgtb %%mm6, %%mm2                   \n\t"
183
        "paddb %%mm2, %%mm0                     \n\t"
184
        "psubusb %%mm3, %%mm4                   \n\t"
185

186
        "                                       \n\t"
187
#if TEMPLATE_PP_MMXEXT
188
        "pxor %%mm7, %%mm7                      \n\t"
189
        "psadbw %%mm7, %%mm0                    \n\t"
190
#else
191
        "movq %%mm0, %%mm1                      \n\t"
192
        "psrlw $8, %%mm0                        \n\t"
193
        "paddb %%mm1, %%mm0                     \n\t"
194
        "movq %%mm0, %%mm1                      \n\t"
195
        "psrlq $16, %%mm0                       \n\t"
196
        "paddb %%mm1, %%mm0                     \n\t"
197
        "movq %%mm0, %%mm1                      \n\t"
198
        "psrlq $32, %%mm0                       \n\t"
199
        "paddb %%mm1, %%mm0                     \n\t"
200
#endif
201
        "movq %4, %%mm7                         \n\t" // QP,..., QP
202
        "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
203
        "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
204
        "packssdw %%mm4, %%mm4                  \n\t"
205
        "movd %%mm0, %0                         \n\t"
206
        "movd %%mm4, %1                         \n\t"
207

208
        : "=r" (numEq), "=r" (dcOk)
209
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
210
        : "%"REG_a
211
        );
212

213
    numEq= (-numEq) &0xFF;
214
    if(numEq > c->ppMode.flatnessThreshold){
215
        if(dcOk) return 0;
216
        else     return 1;
217
    }else{
218
        return 2;
219
    }
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)
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
        "movq %2, %%mm0                         \n\t"  // QP,..., QP
234
        "pxor %%mm4, %%mm4                      \n\t"
235

236
        "movq (%0), %%mm6                       \n\t"
237
        "movq (%0, %1), %%mm5                   \n\t"
238
        "movq %%mm5, %%mm1                      \n\t"
239
        "movq %%mm6, %%mm2                      \n\t"
240
        "psubusb %%mm6, %%mm5                   \n\t"
241
        "psubusb %%mm1, %%mm2                   \n\t"
242
        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
243
        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
244
        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
245

246
        "pand %%mm2, %%mm6                      \n\t"
247
        "pandn %%mm1, %%mm2                     \n\t"
248
        "por %%mm2, %%mm6                       \n\t"// First Line to Filter
249

250
        "movq (%0, %1, 8), %%mm5                \n\t"
251
        "lea (%0, %1, 4), %%"REG_a"             \n\t"
252
        "lea (%0, %1, 8), %%"REG_c"             \n\t"
253
        "sub %1, %%"REG_c"                      \n\t"
254
        "add %1, %0                             \n\t" // %0 points to line 1 not 0
255
        "movq (%0, %1, 8), %%mm7                \n\t"
256
        "movq %%mm5, %%mm1                      \n\t"
257
        "movq %%mm7, %%mm2                      \n\t"
258
        "psubusb %%mm7, %%mm5                   \n\t"
259
        "psubusb %%mm1, %%mm2                   \n\t"
260
        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
261
        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
262
        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
263

264
        "pand %%mm2, %%mm7                      \n\t"
265
        "pandn %%mm1, %%mm2                     \n\t"
266
        "por %%mm2, %%mm7                       \n\t" // First Line to Filter
267

268

269
        //      1       2       3       4       5       6       7       8
270
        //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
271
        // 6 4 2 2 1 1
272
        // 6 4 4 2
273
        // 6 8 2
274

275
        "movq (%0, %1), %%mm0                   \n\t" //  1
276
        "movq %%mm0, %%mm1                      \n\t" //  1
277
        PAVGB(%%mm6, %%mm0)                           //1 1        /2
278
        PAVGB(%%mm6, %%mm0)                           //3 1        /4
279

280
        "movq (%0, %1, 4), %%mm2                \n\t" //     1
281
        "movq %%mm2, %%mm5                      \n\t" //     1
282
        PAVGB((%%REGa), %%mm2)                        //    11        /2
283
        PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
284
        "movq %%mm2, %%mm3                      \n\t" //   211        /4
285
        "movq (%0), %%mm4                       \n\t" // 1
286
        PAVGB(%%mm4, %%mm3)                           // 4 211        /8
287
        PAVGB(%%mm0, %%mm3)                           //642211        /16
288
        "movq %%mm3, (%0)                       \n\t" // X
289
        // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
290
        "movq %%mm1, %%mm0                      \n\t" //  1
291
        PAVGB(%%mm6, %%mm0)                           //1 1        /2
292
        "movq %%mm4, %%mm3                      \n\t" // 1
293
        PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
294
        PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
295
        PAVGB((%%REGa), %%mm5)                        //    211 /4
296
        PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
297
        PAVGB(%%mm0, %%mm3)                           //4242211 /16
298
        "movq %%mm3, (%0,%1)                    \n\t" //  X
299
        // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
300
        PAVGB(%%mm4, %%mm6)                                   //11        /2
301
        "movq (%%"REG_c"), %%mm0                \n\t" //       1
302
        PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
303
        "movq %%mm0, %%mm3                      \n\t" //      11/2
304
        PAVGB(%%mm1, %%mm0)                           //  2   11/4
305
        PAVGB(%%mm6, %%mm0)                           //222   11/8
306
        PAVGB(%%mm2, %%mm0)                           //22242211/16
307
        "movq (%0, %1, 2), %%mm2                \n\t" //   1
308
        "movq %%mm0, (%0, %1, 2)                \n\t" //   X
309
        // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
310
        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
311
        PAVGB((%%REGc), %%mm0)                        //       11        /2
312
        PAVGB(%%mm0, %%mm6)                           //11     11        /4
313
        PAVGB(%%mm1, %%mm4)                           // 11                /2
314
        PAVGB(%%mm2, %%mm1)                           //  11                /2
315
        PAVGB(%%mm1, %%mm6)                           //1122   11        /8
316
        PAVGB(%%mm5, %%mm6)                           //112242211        /16
317
        "movq (%%"REG_a"), %%mm5                \n\t" //    1
318
        "movq %%mm6, (%%"REG_a")                \n\t" //    X
319
        // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
320
        "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
321
        PAVGB(%%mm7, %%mm6)                           //        11        /2
322
        PAVGB(%%mm4, %%mm6)                           // 11     11        /4
323
        PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
324
        PAVGB(%%mm5, %%mm2)                           //   11                /2
325
        "movq (%0, %1, 4), %%mm4                \n\t" //     1
326
        PAVGB(%%mm4, %%mm2)                           //   112                /4
327
        PAVGB(%%mm2, %%mm6)                           // 112242211        /16
328
        "movq %%mm6, (%0, %1, 4)                \n\t" //     X
329
        // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
330
        PAVGB(%%mm7, %%mm1)                           //  11     2        /4
331
        PAVGB(%%mm4, %%mm5)                           //    11                /2
332
        PAVGB(%%mm5, %%mm0)                           //    11 11        /4
333
        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
334
        PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
335
        PAVGB(%%mm0, %%mm1)                           //  11224222        /16
336
        "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
337
        // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
338
        PAVGB((%%REGc), %%mm2)                        //   112 4        /8
339
        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
340
        PAVGB(%%mm0, %%mm6)                           //      1 1        /2
341
        PAVGB(%%mm7, %%mm6)                           //      1 12        /4
342
        PAVGB(%%mm2, %%mm6)                           //   1122424        /4
343
        "movq %%mm6, (%%"REG_c")                \n\t" //       X
344
        // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
345
        PAVGB(%%mm7, %%mm5)                           //    11   2        /4
346
        PAVGB(%%mm7, %%mm5)                           //    11   6        /8
347

348
        PAVGB(%%mm3, %%mm0)                           //      112        /4
349
        PAVGB(%%mm0, %%mm5)                           //    112246        /16
350
        "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
351
        "sub %1, %0                             \n\t"
352

353
        :
354
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
355
        : "%"REG_a, "%"REG_c
356
    );
357
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
358
    const int l1= stride;
359
    const int l2= stride + l1;
360
    const int l3= stride + l2;
361
    const int l4= stride + l3;
362
    const int l5= stride + l4;
363
    const int l6= stride + l5;
364
    const int l7= stride + l6;
365
    const int l8= stride + l7;
366
    const int l9= stride + l8;
367
    int x;
368
    src+= stride*3;
369
    for(x=0; x<BLOCK_SIZE; x++){
370
        const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
371
        const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
372

373
        int sums[10];
374
        sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
375
        sums[1] = sums[0] - first  + src[l4];
376
        sums[2] = sums[1] - first  + src[l5];
377
        sums[3] = sums[2] - first  + src[l6];
378
        sums[4] = sums[3] - first  + src[l7];
379
        sums[5] = sums[4] - src[l1] + src[l8];
380
        sums[6] = sums[5] - src[l2] + last;
381
        sums[7] = sums[6] - src[l3] + last;
382
        sums[8] = sums[7] - src[l4] + last;
383
        sums[9] = sums[8] - src[l5] + last;
384

385
        src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
386
        src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
387
        src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
388
        src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
389
        src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
390
        src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
391
        src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
392
        src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
393

394
        src++;
395
    }
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
403
 * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
404
 * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
405
 * 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
        "pxor %%mm7, %%mm7                      \n\t" // 0
414
        "lea (%0, %1), %%"REG_a"                \n\t"
415
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
416
//      0       1       2       3       4       5       6       7       8       9
417
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
418
        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
419
        "movq (%0, %1, 4), %%mm1                \n\t" // line 4
420
        "movq %%mm1, %%mm2                      \n\t" // line 4
421
        "psubusb %%mm0, %%mm1                   \n\t"
422
        "psubusb %%mm2, %%mm0                   \n\t"
423
        "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
424
        "movq (%%"REG_c"), %%mm3                \n\t" // line 5
425
        "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
426
        "movq %%mm3, %%mm5                      \n\t" // line 5
427
        "psubusb %%mm4, %%mm3                   \n\t"
428
        "psubusb %%mm5, %%mm4                   \n\t"
429
        "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
430
        PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
431
        "movq %%mm2, %%mm1                      \n\t" // line 4
432
        "psubusb %%mm5, %%mm2                   \n\t"
433
        "movq %%mm2, %%mm4                      \n\t"
434
        "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
435
        "psubusb %%mm1, %%mm5                   \n\t"
436
        "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
437
        "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
438
        "movq %%mm4, %%mm3                      \n\t" // d
439
        "movq %2, %%mm0                         \n\t"
440
        "paddusb %%mm0, %%mm0                   \n\t"
441
        "psubusb %%mm0, %%mm4                   \n\t"
442
        "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
443
        "psubusb "MANGLE(b01)", %%mm3           \n\t"
444
        "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
445

446
        PAVGB(%%mm7, %%mm3)                           // d/2
447
        "movq %%mm3, %%mm1                      \n\t" // d/2
448
        PAVGB(%%mm7, %%mm3)                           // d/4
449
        PAVGB(%%mm1, %%mm3)                           // 3*d/8
450

451
        "movq (%0, %1, 4), %%mm0                \n\t" // line 4
452
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
453
        "psubusb %%mm3, %%mm0                   \n\t"
454
        "pxor %%mm2, %%mm0                      \n\t"
455
        "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
456

457
        "movq (%%"REG_c"), %%mm0                \n\t" // line 5
458
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
459
        "paddusb %%mm3, %%mm0                   \n\t"
460
        "pxor %%mm2, %%mm0                      \n\t"
461
        "movq %%mm0, (%%"REG_c")                \n\t" // line 5
462

463
        PAVGB(%%mm7, %%mm1)                           // d/4
464

465
        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
466
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
467
        "psubusb %%mm1, %%mm0                   \n\t"
468
        "pxor %%mm2, %%mm0                      \n\t"
469
        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
470

471
        "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
472
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
473
        "paddusb %%mm1, %%mm0                   \n\t"
474
        "pxor %%mm2, %%mm0                      \n\t"
475
        "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
476

477
        PAVGB(%%mm7, %%mm1)                           // d/8
478

479
        "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
480
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
481
        "psubusb %%mm1, %%mm0                   \n\t"
482
        "pxor %%mm2, %%mm0                      \n\t"
483
        "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
484

485
        "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
486
        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
487
        "paddusb %%mm1, %%mm0                   \n\t"
488
        "pxor %%mm2, %%mm0                      \n\t"
489
        "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
490

491
        :
492
        : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
493
          NAMED_CONSTRAINTS_ADD(b01)
494
        : "%"REG_a, "%"REG_c
495
    );
496
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
497

498
    const int l1= stride;
499
    const int l2= stride + l1;
500
    const int l3= stride + l2;
501
    const int l4= stride + l3;
502
    const int l5= stride + l4;
503
    const int l6= stride + l5;
504
    const int l7= stride + l6;
505
//    const int l8= stride + l7;
506
//    const int l9= stride + l8;
507
    int x;
508

509
    src+= stride*3;
510
    for(x=0; x<BLOCK_SIZE; x++){
511
        int a= src[l3] - src[l4];
512
        int b= src[l4] - src[l5];
513
        int c= src[l5] - src[l6];
514

515
        int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
516
        d= FFMAX(d, 0);
517

518
        if(d < co->QP*2){
519
            int v = d * FFSIGN(-b);
520

521
            src[l2] +=v>>3;
522
            src[l3] +=v>>2;
523
            src[l4] +=(3*v)>>3;
524
            src[l5] -=(3*v)>>3;
525
            src[l6] -=v>>2;
526
            src[l7] -=v>>3;
527
        }
528
        src++;
529
    }
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
    uint8_t tmp[16];
539
    const int l1= stride;
540
    const int l2= stride + l1;
541
    const int l3= stride + l2;
542
    const int l4= (int)tmp - (int)src - stride*3;
543
    const int l5= (int)tmp - (int)src - stride*3 + 8;
544
    const int l6= stride*3 + l3;
545
    const int l7= stride + l6;
546
    const int l8= stride + l7;
547

548
    memcpy(tmp, src+stride*7, 8);
549
    memcpy(tmp+8, src+stride*8, 8);
550
*/
551
    src+= stride*4;
552
    __asm__ volatile(
553

554
#if 0 //slightly more accurate and slightly slower
555
        "pxor %%mm7, %%mm7                      \n\t" // 0
556
        "lea (%0, %1), %%"REG_a"                \n\t"
557
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
558
//      0       1       2       3       4       5       6       7
559
//      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
560
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
561

562

563
        "movq (%0, %1, 2), %%mm0                \n\t" // l2
564
        "movq (%0), %%mm1                       \n\t" // l0
565
        "movq %%mm0, %%mm2                      \n\t" // l2
566
        PAVGB(%%mm7, %%mm0)                           // ~l2/2
567
        PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
568
        PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
569

570
        "movq (%%"REG_a"), %%mm1                \n\t" // l1
571
        "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
572
        "movq %%mm1, %%mm4                      \n\t" // l1
573
        PAVGB(%%mm7, %%mm1)                           // ~l1/2
574
        PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
575
        PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
576

577
        "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
578
        "psubusb %%mm1, %%mm0                   \n\t"
579
        "psubusb %%mm4, %%mm1                   \n\t"
580
        "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
581
// mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
582

583
        "movq (%0, %1, 4), %%mm0                \n\t" // l4
584
        "movq %%mm0, %%mm4                      \n\t" // l4
585
        PAVGB(%%mm7, %%mm0)                           // ~l4/2
586
        PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
587
        PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
588

589
        "movq (%%"REG_c"), %%mm2                \n\t" // l5
590
        "movq %%mm3, %%mm5                      \n\t" // l3
591
        PAVGB(%%mm7, %%mm3)                           // ~l3/2
592
        PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
593
        PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
594

595
        "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
596
        "psubusb %%mm3, %%mm0                   \n\t"
597
        "psubusb %%mm6, %%mm3                   \n\t"
598
        "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
599
        "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
600
// mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
601

602
        "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
603
        "movq %%mm6, %%mm5                      \n\t" // l6
604
        PAVGB(%%mm7, %%mm6)                           // ~l6/2
605
        PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
606
        PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
607

608
        "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
609
        "movq %%mm2, %%mm4                      \n\t" // l5
610
        PAVGB(%%mm7, %%mm2)                           // ~l5/2
611
        PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
612
        PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
613

614
        "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
615
        "psubusb %%mm2, %%mm6                   \n\t"
616
        "psubusb %%mm4, %%mm2                   \n\t"
617
        "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
618
// mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
619

620

621
        PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
622
        "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
623
        "paddusb "MANGLE(b01)", %%mm4           \n\t"
624
        "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
625
        "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
626
        "pand %%mm4, %%mm3                      \n\t"
627

628
        "movq %%mm3, %%mm1                      \n\t"
629
//        "psubusb "MANGLE(b01)", %%mm3           \n\t"
630
        PAVGB(%%mm7, %%mm3)
631
        PAVGB(%%mm7, %%mm3)
632
        "paddusb %%mm1, %%mm3                   \n\t"
633
//        "paddusb "MANGLE(b01)", %%mm3           \n\t"
634

635
        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
636
        "movq (%0, %1, 4), %%mm5                \n\t" //l4
637
        "movq (%0, %1, 4), %%mm4                \n\t" //l4
638
        "psubusb %%mm6, %%mm5                   \n\t"
639
        "psubusb %%mm4, %%mm6                   \n\t"
640
        "por %%mm6, %%mm5                       \n\t" // |l3-l4|
641
        "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
642
        "pxor %%mm6, %%mm0                      \n\t"
643
        "pand %%mm0, %%mm3                      \n\t"
644
        PMINUB(%%mm5, %%mm3, %%mm0)
645

646
        "psubusb "MANGLE(b01)", %%mm3           \n\t"
647
        PAVGB(%%mm7, %%mm3)
648

649
        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
650
        "movq (%0, %1, 4), %%mm2                \n\t"
651
        "pxor %%mm6, %%mm0                      \n\t"
652
        "pxor %%mm6, %%mm2                      \n\t"
653
        "psubb %%mm3, %%mm0                     \n\t"
654
        "paddb %%mm3, %%mm2                     \n\t"
655
        "pxor %%mm6, %%mm0                      \n\t"
656
        "pxor %%mm6, %%mm2                      \n\t"
657
        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
658
        "movq %%mm2, (%0, %1, 4)                \n\t"
659
#endif //0
660

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

667

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

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

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

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

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

714
// mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
715

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

721
        "movq %%mm4, %%mm1                      \n\t"
722
        "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
723
        "pxor %%mm1, %%mm4                      \n\t"
724
        "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
725
        "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
726
        "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
727
// mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
728

729
        "movq %%mm4, %%mm3                      \n\t" // d
730
        "psubusb "MANGLE(b01)", %%mm4           \n\t"
731
        PAVGB(%%mm7, %%mm4)                           // d/32
732
        PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
733
        "paddb %%mm3, %%mm4                     \n\t" // 5d/64
734
        "pand %%mm2, %%mm4                      \n\t"
735

736
        "movq "MANGLE(b80)", %%mm5              \n\t" // 128
737
        "psubb %%mm0, %%mm5                     \n\t" // q
738
        "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
739
        "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
740
        "pxor %%mm7, %%mm5                      \n\t"
741

742
        PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
743
        "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
744

745
        "pand %%mm7, %%mm4                      \n\t"
746
        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
747
        "movq (%0, %1, 4), %%mm2                \n\t"
748
        "pxor %%mm1, %%mm0                      \n\t"
749
        "pxor %%mm1, %%mm2                      \n\t"
750
        "paddb %%mm4, %%mm0                     \n\t"
751
        "psubb %%mm4, %%mm2                     \n\t"
752
        "pxor %%mm1, %%mm0                      \n\t"
753
        "pxor %%mm1, %%mm2                      \n\t"
754
        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
755
        "movq %%mm2, (%0, %1, 4)                \n\t"
756

757
        :
758
        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
759
          NAMED_CONSTRAINTS_ADD(b80,b00,b01)
760
        : "%"REG_a, "%"REG_c
761
    );
762

763
/*
764
    {
765
    int x;
766
    src-= stride;
767
    for(x=0; x<BLOCK_SIZE; x++){
768
        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
769
        if(FFABS(middleEnergy)< 8*QP){
770
            const int q=(src[l4] - src[l5])/2;
771
            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
772
            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
773

774
            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
775
            d= FFMAX(d, 0);
776

777
            d= (5*d + 32) >> 6;
778
            d*= FFSIGN(-middleEnergy);
779

780
            if(q>0){
781
                d= d<0 ? 0 : d;
782
                d= d>q ? q : d;
783
            }else{
784
                d= d>0 ? 0 : d;
785
                d= d<q ? q : d;
786
            }
787

788
            src[l4]-= d;
789
            src[l5]+= d;
790
        }
791
        src++;
792
    }
793
    src-=8;
794
    for(x=0; x<8; x++){
795
        int y;
796
        for(y=4; y<6; y++){
797
            int d= src[x+y*stride] - tmp[x+(y-4)*8];
798
            int ad= FFABS(d);
799
            static int max=0;
800
            static int sum=0;
801
            static int num=0;
802
            static int bias=0;
803

804
            if(max<ad) max=ad;
805
            sum+= ad>3 ? 1 : 0;
806
            if(ad>3){
807
                src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
808
            }
809
            if(y==4) bias+=d;
810
            num++;
811
            if(num%1000000 == 0){
812
                av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
813
            }
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
//      0       1       2       3       4       5       6       7
824
//      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
825
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1
826

827
        "movq (%0), %%mm0                       \n\t"
828
        "movq %%mm0, %%mm1                      \n\t"
829
        "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
830
        "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
831

832
        "movq (%0, %1), %%mm2                   \n\t"
833
        "lea (%0, %1, 2), %%"REG_a"             \n\t"
834
        "movq %%mm2, %%mm3                      \n\t"
835
        "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
836
        "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
837

838
        "movq (%%"REG_a"), %%mm4                \n\t"
839
        "movq %%mm4, %%mm5                      \n\t"
840
        "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
841
        "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
842

843
        "paddw %%mm0, %%mm0                     \n\t" // 2L0
844
        "paddw %%mm1, %%mm1                     \n\t" // 2H0
845
        "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
846
        "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
847
        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
848
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
849

850
        "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
851
        "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
852
        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
853
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
854

855
        "movq (%%"REG_a", %1), %%mm2            \n\t"
856
        "movq %%mm2, %%mm3                      \n\t"
857
        "punpcklbw %%mm7, %%mm2                 \n\t" // L3
858
        "punpckhbw %%mm7, %%mm3                 \n\t" // H3
859

860
        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
861
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
862
        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
863
        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
864
        "movq %%mm0, (%3)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
865
        "movq %%mm1, 8(%3)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
866

867
        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
868
        "movq %%mm0, %%mm1                      \n\t"
869
        "punpcklbw %%mm7, %%mm0                 \n\t" // L4
870
        "punpckhbw %%mm7, %%mm1                 \n\t" // H4
871

872
        "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
873
        "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
874
        "movq %%mm2, 16(%3)                     \n\t" // L3 - L4
875
        "movq %%mm3, 24(%3)                     \n\t" // H3 - H4
876
        "paddw %%mm4, %%mm4                     \n\t" // 2L2
877
        "paddw %%mm5, %%mm5                     \n\t" // 2H2
878
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
879
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
880

881
        "lea (%%"REG_a", %1), %0                \n\t"
882
        "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
883
        "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
884
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
885
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
886
//50 opcodes so far
887
        "movq (%0, %1, 2), %%mm2                \n\t"
888
        "movq %%mm2, %%mm3                      \n\t"
889
        "punpcklbw %%mm7, %%mm2                 \n\t" // L5
890
        "punpckhbw %%mm7, %%mm3                 \n\t" // H5
891
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
892
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
893
        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
894
        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
895

896
        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
897
        "punpcklbw %%mm7, %%mm6                 \n\t" // L6
898
        "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
899
        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
900
        "punpckhbw %%mm7, %%mm6                 \n\t" // H6
901
        "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
902

903
        "paddw %%mm0, %%mm0                     \n\t" // 2L4
904
        "paddw %%mm1, %%mm1                     \n\t" // 2H4
905
        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
906
        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
907

908
        "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
909
        "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
910
        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
911
        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
912

913
        "movq (%0, %1, 4), %%mm2                \n\t"
914
        "movq %%mm2, %%mm3                      \n\t"
915
        "punpcklbw %%mm7, %%mm2                 \n\t" // L7
916
        "punpckhbw %%mm7, %%mm3                 \n\t" // H7
917

918
        "paddw %%mm2, %%mm2                     \n\t" // 2L7
919
        "paddw %%mm3, %%mm3                     \n\t" // 2H7
920
        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
921
        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
922

923
        "movq (%3), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
924
        "movq 8(%3), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
925

926
#if TEMPLATE_PP_MMXEXT
927
        "movq %%mm7, %%mm6                      \n\t" // 0
928
        "psubw %%mm0, %%mm6                     \n\t"
929
        "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
930
        "movq %%mm7, %%mm6                      \n\t" // 0
931
        "psubw %%mm1, %%mm6                     \n\t"
932
        "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
933
        "movq %%mm7, %%mm6                      \n\t" // 0
934
        "psubw %%mm2, %%mm6                     \n\t"
935
        "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
936
        "movq %%mm7, %%mm6                      \n\t" // 0
937
        "psubw %%mm3, %%mm6                     \n\t"
938
        "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
939
#else
940
        "movq %%mm7, %%mm6                      \n\t" // 0
941
        "pcmpgtw %%mm0, %%mm6                   \n\t"
942
        "pxor %%mm6, %%mm0                      \n\t"
943
        "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
944
        "movq %%mm7, %%mm6                      \n\t" // 0
945
        "pcmpgtw %%mm1, %%mm6                   \n\t"
946
        "pxor %%mm6, %%mm1                      \n\t"
947
        "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
948
        "movq %%mm7, %%mm6                      \n\t" // 0
949
        "pcmpgtw %%mm2, %%mm6                   \n\t"
950
        "pxor %%mm6, %%mm2                      \n\t"
951
        "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
952
        "movq %%mm7, %%mm6                      \n\t" // 0
953
        "pcmpgtw %%mm3, %%mm6                   \n\t"
954
        "pxor %%mm6, %%mm3                      \n\t"
955
        "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
956
#endif
957

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

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

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

988

989
        "psubusw %%mm0, %%mm4                   \n\t" // hd
990
        "psubusw %%mm1, %%mm5                   \n\t" // ld
991

992

993
        "movq "MANGLE(w05)", %%mm2              \n\t" // 5
994
        "pmullw %%mm2, %%mm4                    \n\t"
995
        "pmullw %%mm2, %%mm5                    \n\t"
996
        "movq "MANGLE(w20)", %%mm2              \n\t" // 32
997
        "paddw %%mm2, %%mm4                     \n\t"
998
        "paddw %%mm2, %%mm5                     \n\t"
999
        "psrlw $6, %%mm4                        \n\t"
1000
        "psrlw $6, %%mm5                        \n\t"
1001

1002
        "movq 16(%3), %%mm0                     \n\t" // L3 - L4
1003
        "movq 24(%3), %%mm1                     \n\t" // H3 - H4
1004

1005
        "pxor %%mm2, %%mm2                      \n\t"
1006
        "pxor %%mm3, %%mm3                      \n\t"
1007

1008
        "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
1009
        "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
1010
        "pxor %%mm2, %%mm0                      \n\t"
1011
        "pxor %%mm3, %%mm1                      \n\t"
1012
        "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
1013
        "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
1014
        "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
1015
        "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
1016

1017
        "pxor %%mm6, %%mm2                      \n\t"
1018
        "pxor %%mm7, %%mm3                      \n\t"
1019
        "pand %%mm2, %%mm4                      \n\t"
1020
        "pand %%mm3, %%mm5                      \n\t"
1021

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

1045
        : "+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
    const int l1= stride;
1052
    const int l2= stride + l1;
1053
    const int l3= stride + l2;
1054
    const int l4= stride + l3;
1055
    const int l5= stride + l4;
1056
    const int l6= stride + l5;
1057
    const int l7= stride + l6;
1058
    const int l8= stride + l7;
1059
//    const int l9= stride + l8;
1060
    int x;
1061
    src+= stride*3;
1062
    for(x=0; x<BLOCK_SIZE; x++){
1063
        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1064
        if(FFABS(middleEnergy) < 8*c->QP){
1065
            const int q=(src[l4] - src[l5])/2;
1066
            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1067
            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1068

1069
            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
1070
            d= FFMAX(d, 0);
1071

1072
            d= (5*d + 32) >> 6;
1073
            d*= FFSIGN(-middleEnergy);
1074

1075
            if(q>0){
1076
                d = FFMAX(d, 0);
1077
                d = FFMIN(d, q);
1078
            }else{
1079
                d = FFMIN(d, 0);
1080
                d = FFMAX(d, q);
1081
            }
1082

1083
            src[l4]-= d;
1084
            src[l5]+= d;
1085
        }
1086
        src++;
1087
    }
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
        "pxor %%mm6, %%mm6                      \n\t"
1099
        "pcmpeqb %%mm7, %%mm7                   \n\t"
1100
        "movq %2, %%mm0                         \n\t"
1101
        "punpcklbw %%mm6, %%mm0                 \n\t"
1102
        "psrlw $1, %%mm0                        \n\t"
1103
        "psubw %%mm7, %%mm0                     \n\t"
1104
        "packuswb %%mm0, %%mm0                  \n\t"
1105
        "movq %%mm0, %3                         \n\t"
1106

1107
        "lea (%0, %1), %%"REG_a"                \n\t"
1108
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1109

1110
//        0        1        2        3        4        5        6        7        8        9
1111
//        %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
        "movq " #addr ", %%mm0                  \n\t"\
1118
        "pminub %%mm0, %%mm7                    \n\t"\
1119
        "pmaxub %%mm0, %%mm6                    \n\t"
1120
#else
1121
#define REAL_FIND_MIN_MAX(addr)\
1122
        "movq " #addr ", %%mm0                  \n\t"\
1123
        "movq %%mm7, %%mm1                      \n\t"\
1124
        "psubusb %%mm0, %%mm6                   \n\t"\
1125
        "paddb %%mm0, %%mm6                     \n\t"\
1126
        "psubusb %%mm0, %%mm1                   \n\t"\
1127
        "psubb %%mm1, %%mm7                     \n\t"
1128
#endif
1129
#define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
1130

1131
FIND_MIN_MAX((%%REGa))
1132
FIND_MIN_MAX((%%REGa, %1))
1133
FIND_MIN_MAX((%%REGa, %1, 2))
1134
FIND_MIN_MAX((%0, %1, 4))
1135
FIND_MIN_MAX((%%REGd))
1136
FIND_MIN_MAX((%%REGd, %1))
1137
FIND_MIN_MAX((%%REGd, %1, 2))
1138
FIND_MIN_MAX((%0, %1, 8))
1139

1140
        "movq %%mm7, %%mm4                      \n\t"
1141
        "psrlq $8, %%mm7                        \n\t"
1142
#if TEMPLATE_PP_MMXEXT
1143
        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1144
        "pshufw $0xF9, %%mm7, %%mm4             \n\t"
1145
        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1146
        "pshufw $0xFE, %%mm7, %%mm4             \n\t"
1147
        "pminub %%mm4, %%mm7                    \n\t"
1148
#else
1149
        "movq %%mm7, %%mm1                      \n\t"
1150
        "psubusb %%mm4, %%mm1                   \n\t"
1151
        "psubb %%mm1, %%mm7                     \n\t"
1152
        "movq %%mm7, %%mm4                      \n\t"
1153
        "psrlq $16, %%mm7                       \n\t"
1154
        "movq %%mm7, %%mm1                      \n\t"
1155
        "psubusb %%mm4, %%mm1                   \n\t"
1156
        "psubb %%mm1, %%mm7                     \n\t"
1157
        "movq %%mm7, %%mm4                      \n\t"
1158
        "psrlq $32, %%mm7                       \n\t"
1159
        "movq %%mm7, %%mm1                      \n\t"
1160
        "psubusb %%mm4, %%mm1                   \n\t"
1161
        "psubb %%mm1, %%mm7                     \n\t"
1162
#endif
1163

1164

1165
        "movq %%mm6, %%mm4                      \n\t"
1166
        "psrlq $8, %%mm6                        \n\t"
1167
#if TEMPLATE_PP_MMXEXT
1168
        "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
1169
        "pshufw $0xF9, %%mm6, %%mm4             \n\t"
1170
        "pmaxub %%mm4, %%mm6                    \n\t"
1171
        "pshufw $0xFE, %%mm6, %%mm4             \n\t"
1172
        "pmaxub %%mm4, %%mm6                    \n\t"
1173
#else
1174
        "psubusb %%mm4, %%mm6                   \n\t"
1175
        "paddb %%mm4, %%mm6                     \n\t"
1176
        "movq %%mm6, %%mm4                      \n\t"
1177
        "psrlq $16, %%mm6                       \n\t"
1178
        "psubusb %%mm4, %%mm6                   \n\t"
1179
        "paddb %%mm4, %%mm6                     \n\t"
1180
        "movq %%mm6, %%mm4                      \n\t"
1181
        "psrlq $32, %%mm6                       \n\t"
1182
        "psubusb %%mm4, %%mm6                   \n\t"
1183
        "paddb %%mm4, %%mm6                     \n\t"
1184
#endif
1185
        "movq %%mm6, %%mm0                      \n\t" // max
1186
        "psubb %%mm7, %%mm6                     \n\t" // max - min
1187
        "push %4                              \n\t"
1188
        "movd %%mm6, %k4                        \n\t"
1189
        "cmpb "MANGLE(deringThreshold)", %b4    \n\t"
1190
        "pop %4                               \n\t"
1191
        " jb 1f                                 \n\t"
1192
        PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
1193
        "punpcklbw %%mm7, %%mm7                 \n\t"
1194
        "punpcklbw %%mm7, %%mm7                 \n\t"
1195
        "punpcklbw %%mm7, %%mm7                 \n\t"
1196
        "movq %%mm7, (%4)                       \n\t"
1197

1198
        "movq (%0), %%mm0                       \n\t" // L10
1199
        "movq %%mm0, %%mm1                      \n\t" // L10
1200
        "movq %%mm0, %%mm2                      \n\t" // L10
1201
        "psllq $8, %%mm1                        \n\t"
1202
        "psrlq $8, %%mm2                        \n\t"
1203
        "movd -4(%0), %%mm3                     \n\t"
1204
        "movd 8(%0), %%mm4                      \n\t"
1205
        "psrlq $24, %%mm3                       \n\t"
1206
        "psllq $56, %%mm4                       \n\t"
1207
        "por %%mm3, %%mm1                       \n\t" // L00
1208
        "por %%mm4, %%mm2                       \n\t" // L20
1209
        "movq %%mm1, %%mm3                      \n\t" // L00
1210
        PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
1211
        PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
1212
        "psubusb %%mm7, %%mm0                   \n\t"
1213
        "psubusb %%mm7, %%mm2                   \n\t"
1214
        "psubusb %%mm7, %%mm3                   \n\t"
1215
        "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
1216
        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
1217
        "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
1218
        "paddb %%mm2, %%mm0                     \n\t"
1219
        "paddb %%mm3, %%mm0                     \n\t"
1220

1221
        "movq (%%"REG_a"), %%mm2                \n\t" // L11
1222
        "movq %%mm2, %%mm3                      \n\t" // L11
1223
        "movq %%mm2, %%mm4                      \n\t" // L11
1224
        "psllq $8, %%mm3                        \n\t"
1225
        "psrlq $8, %%mm4                        \n\t"
1226
        "movd -4(%%"REG_a"), %%mm5              \n\t"
1227
        "movd 8(%%"REG_a"), %%mm6               \n\t"
1228
        "psrlq $24, %%mm5                       \n\t"
1229
        "psllq $56, %%mm6                       \n\t"
1230
        "por %%mm5, %%mm3                       \n\t" // L01
1231
        "por %%mm6, %%mm4                       \n\t" // L21
1232
        "movq %%mm3, %%mm5                      \n\t" // L01
1233
        PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
1234
        PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
1235
        "psubusb %%mm7, %%mm2                   \n\t"
1236
        "psubusb %%mm7, %%mm4                   \n\t"
1237
        "psubusb %%mm7, %%mm5                   \n\t"
1238
        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
1239
        "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
1240
        "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
1241
        "paddb %%mm4, %%mm2                     \n\t"
1242
        "paddb %%mm5, %%mm2                     \n\t"
1243
// 0, 2, 3, 1
1244
#define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1245
        "movq " #src ", " #sx "                 \n\t" /* src[0] */\
1246
        "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
1247
        "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
1248
        "psllq $8, " #lx "                      \n\t"\
1249
        "psrlq $8, " #t0 "                      \n\t"\
1250
        "movd -4" #src ", " #t1 "               \n\t"\
1251
        "psrlq $24, " #t1 "                     \n\t"\
1252
        "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
1253
        "movd 8" #src ", " #t1 "                \n\t"\
1254
        "psllq $56, " #t1 "                     \n\t"\
1255
        "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
1256
        "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
1257
        PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
1258
        PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
1259
        PAVGB(lx, pplx)                                     \
1260
        "movq " #lx ", 8(%4)                    \n\t"\
1261
        "movq (%4), " #lx "                     \n\t"\
1262
        "psubusb " #lx ", " #t1 "               \n\t"\
1263
        "psubusb " #lx ", " #t0 "               \n\t"\
1264
        "psubusb " #lx ", " #sx "               \n\t"\
1265
        "movq "MANGLE(b00)", " #lx "            \n\t"\
1266
        "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
1267
        "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
1268
        "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
1269
        "paddb " #t1 ", " #t0 "                 \n\t"\
1270
        "paddb " #t0 ", " #sx "                 \n\t"\
1271
\
1272
        PAVGB(plx, pplx)                              /* filtered */\
1273
        "movq " #dst ", " #t0 "                 \n\t" /* dst */\
1274
        "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
1275
        "psubusb %3, " #t0 "                    \n\t"\
1276
        "paddusb %3, " #t1 "                    \n\t"\
1277
        PMAXUB(t0, pplx)\
1278
        PMINUB(t1, pplx, t0)\
1279
        "paddb " #sx ", " #ppsx "               \n\t"\
1280
        "paddb " #psx ", " #ppsx "              \n\t"\
1281
        "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
1282
        "pand "MANGLE(b08)", " #ppsx "          \n\t"\
1283
        "pcmpeqb " #lx ", " #ppsx "             \n\t"\
1284
        "pand " #ppsx ", " #pplx "              \n\t"\
1285
        "pandn " #dst ", " #ppsx "              \n\t"\
1286
        "por " #pplx ", " #ppsx "               \n\t"\
1287
        "movq " #ppsx ", " #dst "               \n\t"\
1288
        "movq 8(%4), " #lx "                    \n\t"
1289

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

1296
1111110
1297
1111101
1298
1111100
1299
1111011
1300
1111010
1301
1111001
1302

1303
1111000
1304
1110111
1305

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

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
    int y;
1324
    int min=255;
1325
    int max=0;
1326
    int avg;
1327
    uint8_t *p;
1328
    int s[10];
1329
    const int QP2= c->QP/2 + 1;
1330

1331
    src --;
1332
    for(y=1; y<9; y++){
1333
        int x;
1334
        p= src + stride*y;
1335
        for(x=1; x<9; x++){
1336
            p++;
1337
            if(*p > max) max= *p;
1338
            if(*p < min) min= *p;
1339
        }
1340
    }
1341
    avg= (min + max + 1)>>1;
1342

1343
    if(max - min <deringThreshold) return;
1344

1345
    for(y=0; y<10; y++){
1346
        int t = 0;
1347

1348
        if(src[stride*y + 0] > avg) t+= 1;
1349
        if(src[stride*y + 1] > avg) t+= 2;
1350
        if(src[stride*y + 2] > avg) t+= 4;
1351
        if(src[stride*y + 3] > avg) t+= 8;
1352
        if(src[stride*y + 4] > avg) t+= 16;
1353
        if(src[stride*y + 5] > avg) t+= 32;
1354
        if(src[stride*y + 6] > avg) t+= 64;
1355
        if(src[stride*y + 7] > avg) t+= 128;
1356
        if(src[stride*y + 8] > avg) t+= 256;
1357
        if(src[stride*y + 9] > avg) t+= 512;
1358

1359
        t |= (~t)<<16;
1360
        t &= (t<<1) & (t>>1);
1361
        s[y] = t;
1362
    }
1363

1364
    for(y=1; y<9; y++){
1365
        int t = s[y-1] & s[y] & s[y+1];
1366
        t|= t>>16;
1367
        s[y-1]= t;
1368
    }
1369

1370
    for(y=1; y<9; y++){
1371
        int x;
1372
        int t = s[y-1];
1373

1374
        p= src + stride*y;
1375
        for(x=1; x<9; x++){
1376
            p++;
1377
            if(t & (1<<x)){
1378
                int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1379
                      +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
1380
                      +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1381
                f= (f + 8)>>4;
1382

1383
#ifdef DEBUG_DERING_THRESHOLD
1384
                    __asm__ volatile("emms\n\t":);
1385
                    {
1386
                    static uint64_t numPixels=0;
1387
                    if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1388
//                    if((max-min)<20 || (max-min)*QP<200)
1389
//                    if((max-min)*QP < 500)
1390
//                    if(max-min<QP/2)
1391
                    if(max-min < 20){
1392
                        static int numSkipped=0;
1393
                        static int errorSum=0;
1394
                        static int worstQP=0;
1395
                        static int worstRange=0;
1396
                        static int worstDiff=0;
1397
                        int diff= (f - *p);
1398
                        int absDiff= FFABS(diff);
1399
                        int error= diff*diff;
1400

1401
                        if(x==1 || x==8 || y==1 || y==8) continue;
1402

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

1411
                        if(1024LL*1024LL*1024LL % numSkipped == 0){
1412
                            av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
1413
                                   "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1414
                                   (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
1415
                                   worstDiff, (float)numSkipped/numPixels);
1416
                        }
1417
                    }
1418
                    }
1419
#endif
1420
                    if     (*p + QP2 < f) *p= *p + QP2;
1421
                    else if(*p - QP2 > f) *p= *p - QP2;
1422
                    else *p=f;
1423
            }
1424
        }
1425
    }
1426
#ifdef DEBUG_DERING_THRESHOLD
1427
    if(max-min < 20){
1428
        for(y=1; y<9; y++){
1429
            int x;
1430
            int t = 0;
1431
            p= src + stride*y;
1432
            for(x=1; x<9; x++){
1433
                p++;
1434
                *p = FFMIN(*p + 20, 255);
1435
            }
1436
        }
1437
//        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1438
    }
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.
1446
 * will be called for every 8x8 block and can read & write from line 4-15
1447
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
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
        "lea (%0, %1), %%"REG_a"                \n\t"
1456
        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
1457
//      0       1       2       3       4       5       6       7       8       9
1458
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
1459

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

1474
        : : "r" (src), "r" ((x86_reg)stride)
1475
        : "%"REG_a, "%"REG_c
1476
    );
1477
#else
1478
    int a, b, x;
1479
    src+= 4*stride;
1480

1481
    for(x=0; x<2; x++){
1482
        a= *(uint32_t*)&src[stride*0];
1483
        b= *(uint32_t*)&src[stride*2];
1484
        *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1485
        a= *(uint32_t*)&src[stride*4];
1486
        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1487
        b= *(uint32_t*)&src[stride*6];
1488
        *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1489
        a= *(uint32_t*)&src[stride*8];
1490
        *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1491
        src += 4;
1492
    }
1493
#endif
1494
}
1495

1496
/**
1497
 * Deinterlace the given block by cubic interpolating every second line.
1498
 * will be called for every 8x8 block and can read & write from line 4-15
1499
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1500
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1501
 * 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
        "lea (%0, %1), %%"REG_a"                \n\t"
1509
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1510
        "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
1511
        "add %1, %%"REG_c"                      \n\t"
1512
#if TEMPLATE_PP_SSE2
1513
        "pxor %%xmm7, %%xmm7                    \n\t"
1514
#define REAL_DEINT_CUBIC(a,b,c,d,e)\
1515
        "movq " #a ", %%xmm0                    \n\t"\
1516
        "movq " #b ", %%xmm1                    \n\t"\
1517
        "movq " #d ", %%xmm2                    \n\t"\
1518
        "movq " #e ", %%xmm3                    \n\t"\
1519
        "pavgb %%xmm2, %%xmm1                   \n\t"\
1520
        "pavgb %%xmm3, %%xmm0                   \n\t"\
1521
        "punpcklbw %%xmm7, %%xmm0               \n\t"\
1522
        "punpcklbw %%xmm7, %%xmm1               \n\t"\
1523
        "psubw %%xmm1, %%xmm0                   \n\t"\
1524
        "psraw $3, %%xmm0                       \n\t"\
1525
        "psubw %%xmm0, %%xmm1                   \n\t"\
1526
        "packuswb %%xmm1, %%xmm1                \n\t"\
1527
        "movlps %%xmm1, " #c "                  \n\t"
1528
#else //TEMPLATE_PP_SSE2
1529
        "pxor %%mm7, %%mm7                      \n\t"
1530
//      0       1       2       3       4       5       6       7       8       9       10
1531
//      %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
        "movq " #a ", %%mm0                     \n\t"\
1535
        "movq " #b ", %%mm1                     \n\t"\
1536
        "movq " #d ", %%mm2                     \n\t"\
1537
        "movq " #e ", %%mm3                     \n\t"\
1538
        PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
1539
        PAVGB(%%mm3, %%mm0)                             /* (a+e) /2 */\
1540
        "movq %%mm0, %%mm2                      \n\t"\
1541
        "punpcklbw %%mm7, %%mm0                 \n\t"\
1542
        "punpckhbw %%mm7, %%mm2                 \n\t"\
1543
        "movq %%mm1, %%mm3                      \n\t"\
1544
        "punpcklbw %%mm7, %%mm1                 \n\t"\
1545
        "punpckhbw %%mm7, %%mm3                 \n\t"\
1546
        "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
1547
        "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
1548
        "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
1549
        "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
1550
        "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
1551
        "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
1552
        "packuswb %%mm3, %%mm1                  \n\t"\
1553
        "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
DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
1558
DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))
1559
DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
1560
DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
1561

1562
        : : "r" (src), "r" ((x86_reg)stride)
1563
        :
1564
#if TEMPLATE_PP_SSE2
1565
        XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm7",)
1566
#endif
1567
        "%"REG_a, "%"REG_d, "%"REG_c
1568
    );
1569
#undef REAL_DEINT_CUBIC
1570
#else //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1571
    int x;
1572
    src+= stride*3;
1573
    for(x=0; x<8; x++){
1574
        src[stride*3] = av_clip_uint8((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1575
        src[stride*5] = av_clip_uint8((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1576
        src[stride*7] = av_clip_uint8((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1577
        src[stride*9] = av_clip_uint8((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1578
        src++;
1579
    }
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
1586
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1587
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1588
 * this filter will read lines 4-13 and write 5-11
1589
 */
1590
static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
1591
{
1592
#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1593
    src+= stride*4;
1594
    __asm__ volatile(
1595
        "lea (%0, %1), %%"REG_a"                \n\t"
1596
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1597
        "pxor %%mm7, %%mm7                      \n\t"
1598
        "movq (%2), %%mm0                       \n\t"
1599
//      0       1       2       3       4       5       6       7       8       9       10
1600
//      %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
        "movq " #a ", %%mm1                     \n\t"\
1604
        "movq " #b ", %%mm2                     \n\t"\
1605
        "movq " #c ", %%mm3                     \n\t"\
1606
        "movq " #d ", %%mm4                     \n\t"\
1607
        PAVGB(%%mm3, %%mm1)                          \
1608
        PAVGB(%%mm4, %%mm0)                          \
1609
        "movq %%mm0, %%mm3                      \n\t"\
1610
        "punpcklbw %%mm7, %%mm0                 \n\t"\
1611
        "punpckhbw %%mm7, %%mm3                 \n\t"\
1612
        "movq %%mm1, %%mm4                      \n\t"\
1613
        "punpcklbw %%mm7, %%mm1                 \n\t"\
1614
        "punpckhbw %%mm7, %%mm4                 \n\t"\
1615
        "psllw $2, %%mm1                        \n\t"\
1616
        "psllw $2, %%mm4                        \n\t"\
1617
        "psubw %%mm0, %%mm1                     \n\t"\
1618
        "psubw %%mm3, %%mm4                     \n\t"\
1619
        "movq %%mm2, %%mm5                      \n\t"\
1620
        "movq %%mm2, %%mm0                      \n\t"\
1621
        "punpcklbw %%mm7, %%mm2                 \n\t"\
1622
        "punpckhbw %%mm7, %%mm5                 \n\t"\
1623
        "paddw %%mm2, %%mm1                     \n\t"\
1624
        "paddw %%mm5, %%mm4                     \n\t"\
1625
        "psraw $2, %%mm1                        \n\t"\
1626
        "psraw $2, %%mm4                        \n\t"\
1627
        "packuswb %%mm4, %%mm1                  \n\t"\
1628
        "movq %%mm1, " #b "                     \n\t"\
1629

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

1632
DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))
1633
DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
1634
DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
1635
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
        : "%"REG_a, "%"REG_d
1640
    );
1641
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1642
    int x;
1643
    src+= stride*4;
1644
    for(x=0; x<8; x++){
1645
        int t1= tmp[x];
1646
        int t2= src[stride*1];
1647

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
        tmp[x]= t1;
1656

1657
        src++;
1658
    }
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
1665
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1666
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1667
 * this filter will read lines 4-13 and write 4-11
1668
 */
1669
static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
1670
{
1671
#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
1672
    src+= stride*4;
1673
    __asm__ volatile(
1674
        "lea (%0, %1), %%"REG_a"                \n\t"
1675
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1676
        "pxor %%mm7, %%mm7                      \n\t"
1677
        "movq (%2), %%mm0                       \n\t"
1678
        "movq (%3), %%mm1                       \n\t"
1679
//      0       1       2       3       4       5       6       7       8       9       10
1680
//      %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
        "movq " #a ", %%mm2                     \n\t"\
1684
        "movq " #b ", %%mm3                     \n\t"\
1685
        "movq " #c ", %%mm4                     \n\t"\
1686
        PAVGB(t2, %%mm3)                             \
1687
        PAVGB(t1, %%mm4)                             \
1688
        "movq %%mm2, %%mm5                      \n\t"\
1689
        "movq %%mm2, " #t1 "                    \n\t"\
1690
        "punpcklbw %%mm7, %%mm2                 \n\t"\
1691
        "punpckhbw %%mm7, %%mm5                 \n\t"\
1692
        "movq %%mm2, %%mm6                      \n\t"\
1693
        "paddw %%mm2, %%mm2                     \n\t"\
1694
        "paddw %%mm6, %%mm2                     \n\t"\
1695
        "movq %%mm5, %%mm6                      \n\t"\
1696
        "paddw %%mm5, %%mm5                     \n\t"\
1697
        "paddw %%mm6, %%mm5                     \n\t"\
1698
        "movq %%mm3, %%mm6                      \n\t"\
1699
        "punpcklbw %%mm7, %%mm3                 \n\t"\
1700
        "punpckhbw %%mm7, %%mm6                 \n\t"\
1701
        "paddw %%mm3, %%mm3                     \n\t"\
1702
        "paddw %%mm6, %%mm6                     \n\t"\
1703
        "paddw %%mm3, %%mm2                     \n\t"\
1704
        "paddw %%mm6, %%mm5                     \n\t"\
1705
        "movq %%mm4, %%mm6                      \n\t"\
1706
        "punpcklbw %%mm7, %%mm4                 \n\t"\
1707
        "punpckhbw %%mm7, %%mm6                 \n\t"\
1708
        "psubw %%mm4, %%mm2                     \n\t"\
1709
        "psubw %%mm6, %%mm5                     \n\t"\
1710
        "psraw $2, %%mm2                        \n\t"\
1711
        "psraw $2, %%mm5                        \n\t"\
1712
        "packuswb %%mm5, %%mm2                  \n\t"\
1713
        "movq %%mm2, " #a "                     \n\t"\
1714

1715
#define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)
1716

1717
DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )
1718
DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))
1719
DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )
1720
DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )
1721
DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )
1722
DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
1723
DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
1724
DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
1725

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

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
        tmp[x]= t3;
1756
        tmp2[x]= t1;
1757

1758
        src++;
1759
    }
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.
1765
 * will be called for every 8x8 block and can read & write from line 4-15
1766
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1767
 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1768
 * 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
        "lea (%0, %1), %%"REG_a"                \n\t"
1776
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1777
//      0       1       2       3       4       5       6       7       8       9
1778
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1779

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

1816
        : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
1817
        : "%"REG_a, "%"REG_d
1818
    );
1819
#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
1820
    int a, b, c, x;
1821
    src+= 4*stride;
1822

1823
    for(x=0; x<2; x++){
1824
        a= *(uint32_t*)&tmp[stride*0];
1825
        b= *(uint32_t*)&src[stride*0];
1826
        c= *(uint32_t*)&src[stride*1];
1827
        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1828
        *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1829

1830
        a= *(uint32_t*)&src[stride*2];
1831
        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1832
        *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1833

1834
        b= *(uint32_t*)&src[stride*3];
1835
        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1836
        *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1837

1838
        c= *(uint32_t*)&src[stride*4];
1839
        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1840
        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1841

1842
        a= *(uint32_t*)&src[stride*5];
1843
        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1844
        *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1845

1846
        b= *(uint32_t*)&src[stride*6];
1847
        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1848
        *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1849

1850
        c= *(uint32_t*)&src[stride*7];
1851
        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1852
        *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1853

1854
        a= *(uint32_t*)&src[stride*8];
1855
        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1856
        *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1857

1858
        *(uint32_t*)&tmp[stride*0]= c;
1859
        src += 4;
1860
        tmp += 4;
1861
    }
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.
1867
 * will be called for every 8x8 block and can read & write from line 4-15,
1868
 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
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
        "lea (%0, %1), %%"REG_a"                \n\t"
1878
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1879
//      0       1       2       3       4       5       6       7       8       9
1880
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1881

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

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

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

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

1919

1920
        : : "r" (src), "r" ((x86_reg)stride)
1921
        : "%"REG_a, "%"REG_d
1922
    );
1923

1924
#else // MMX without MMX2
1925
    __asm__ volatile(
1926
        "lea (%0, %1), %%"REG_a"                \n\t"
1927
        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1928
//      0       1       2       3       4       5       6       7       8       9
1929
//      %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"
1931

1932
#define REAL_MEDIAN(a,b,c)\
1933
        "movq " #a ", %%mm0                     \n\t"\
1934
        "movq " #b ", %%mm2                     \n\t"\
1935
        "movq " #c ", %%mm1                     \n\t"\
1936
        "movq %%mm0, %%mm3                      \n\t"\
1937
        "movq %%mm1, %%mm4                      \n\t"\
1938
        "movq %%mm2, %%mm5                      \n\t"\
1939
        "psubusb %%mm1, %%mm3                   \n\t"\
1940
        "psubusb %%mm2, %%mm4                   \n\t"\
1941
        "psubusb %%mm0, %%mm5                   \n\t"\
1942
        "pcmpeqb %%mm7, %%mm3                   \n\t"\
1943
        "pcmpeqb %%mm7, %%mm4                   \n\t"\
1944
        "pcmpeqb %%mm7, %%mm5                   \n\t"\
1945
        "movq %%mm3, %%mm6                      \n\t"\
1946
        "pxor %%mm4, %%mm3                      \n\t"\
1947
        "pxor %%mm5, %%mm4                      \n\t"\
1948
        "pxor %%mm6, %%mm5                      \n\t"\
1949
        "por %%mm3, %%mm1                       \n\t"\
1950
        "por %%mm4, %%mm2                       \n\t"\
1951
        "por %%mm5, %%mm0                       \n\t"\
1952
        "pand %%mm2, %%mm0                      \n\t"\
1953
        "pand %%mm1, %%mm0                      \n\t"\
1954
        "movq %%mm0, " #b "                     \n\t"
1955
#define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
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))
1961

1962
        : : "r" (src), "r" ((x86_reg)stride)
1963
        : "%"REG_a, "%"REG_d
1964
    );
1965
#endif //TEMPLATE_PP_MMXEXT
1966
#else //TEMPLATE_PP_MMX
1967
    int x, y;
1968
    src+= 4*stride;
1969
    // FIXME - there should be a way to do a few columns in parallel like w/mmx
1970
    for(x=0; x<8; x++){
1971
        uint8_t *colsrc = src;
1972
        for (y=0; y<4; y++){
1973
            int a, b, c, d, e, f;
1974
            a = colsrc[0       ];
1975
            b = colsrc[stride  ];
1976
            c = colsrc[stride*2];
1977
            d = (a-b)>>31;
1978
            e = (b-c)>>31;
1979
            f = (c-a)>>31;
1980
            colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
1981
            colsrc += stride*2;
1982
        }
1983
        src++;
1984
    }
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
//      0       1       2       3       4       5       6       7       8       9
1997
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1998
        "movq (%0), %%mm0                       \n\t" // 12345678
1999
        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2000
        "movq %%mm0, %%mm2                      \n\t" // 12345678
2001
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2002
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2003

2004
        "movq (%%"REG_a", %1), %%mm1            \n\t"
2005
        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2006
        "movq %%mm1, %%mm4                      \n\t"
2007
        "punpcklbw %%mm3, %%mm1                 \n\t"
2008
        "punpckhbw %%mm3, %%mm4                 \n\t"
2009

2010
        "movq %%mm0, %%mm3                      \n\t"
2011
        "punpcklwd %%mm1, %%mm0                 \n\t"
2012
        "punpckhwd %%mm1, %%mm3                 \n\t"
2013
        "movq %%mm2, %%mm1                      \n\t"
2014
        "punpcklwd %%mm4, %%mm2                 \n\t"
2015
        "punpckhwd %%mm4, %%mm1                 \n\t"
2016

2017
        "movd %%mm0, 128(%2)                    \n\t"
2018
        "psrlq $32, %%mm0                       \n\t"
2019
        "movd %%mm0, 144(%2)                    \n\t"
2020
        "movd %%mm3, 160(%2)                    \n\t"
2021
        "psrlq $32, %%mm3                       \n\t"
2022
        "movd %%mm3, 176(%2)                    \n\t"
2023
        "movd %%mm3, 48(%3)                     \n\t"
2024
        "movd %%mm2, 192(%2)                    \n\t"
2025
        "movd %%mm2, 64(%3)                     \n\t"
2026
        "psrlq $32, %%mm2                       \n\t"
2027
        "movd %%mm2, 80(%3)                     \n\t"
2028
        "movd %%mm1, 96(%3)                     \n\t"
2029
        "psrlq $32, %%mm1                       \n\t"
2030
        "movd %%mm1, 112(%3)                    \n\t"
2031

2032
        "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
2033

2034
        "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
2035
        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2036
        "movq %%mm0, %%mm2                      \n\t" // 12345678
2037
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2038
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2039

2040
        "movq (%%"REG_a", %1), %%mm1            \n\t"
2041
        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2042
        "movq %%mm1, %%mm4                      \n\t"
2043
        "punpcklbw %%mm3, %%mm1                 \n\t"
2044
        "punpckhbw %%mm3, %%mm4                 \n\t"
2045

2046
        "movq %%mm0, %%mm3                      \n\t"
2047
        "punpcklwd %%mm1, %%mm0                 \n\t"
2048
        "punpckhwd %%mm1, %%mm3                 \n\t"
2049
        "movq %%mm2, %%mm1                      \n\t"
2050
        "punpcklwd %%mm4, %%mm2                 \n\t"
2051
        "punpckhwd %%mm4, %%mm1                 \n\t"
2052

2053
        "movd %%mm0, 132(%2)                    \n\t"
2054
        "psrlq $32, %%mm0                       \n\t"
2055
        "movd %%mm0, 148(%2)                    \n\t"
2056
        "movd %%mm3, 164(%2)                    \n\t"
2057
        "psrlq $32, %%mm3                       \n\t"
2058
        "movd %%mm3, 180(%2)                    \n\t"
2059
        "movd %%mm3, 52(%3)                     \n\t"
2060
        "movd %%mm2, 196(%2)                    \n\t"
2061
        "movd %%mm2, 68(%3)                     \n\t"
2062
        "psrlq $32, %%mm2                       \n\t"
2063
        "movd %%mm2, 84(%3)                     \n\t"
2064
        "movd %%mm1, 100(%3)                    \n\t"
2065
        "psrlq $32, %%mm1                       \n\t"
2066
        "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
        "lea (%0, %1), %%"REG_a"                \n\t"
2081
        "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
2082
//      0       1       2       3       4       5       6       7       8       9
2083
//      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
2084
        "movq (%2), %%mm0                       \n\t" // 12345678
2085
        "movq 16(%2), %%mm1                     \n\t" // abcdefgh
2086
        "movq %%mm0, %%mm2                      \n\t" // 12345678
2087
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2088
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2089

2090
        "movq 32(%2), %%mm1                     \n\t"
2091
        "movq 48(%2), %%mm3                     \n\t"
2092
        "movq %%mm1, %%mm4                      \n\t"
2093
        "punpcklbw %%mm3, %%mm1                 \n\t"
2094
        "punpckhbw %%mm3, %%mm4                 \n\t"
2095

2096
        "movq %%mm0, %%mm3                      \n\t"
2097
        "punpcklwd %%mm1, %%mm0                 \n\t"
2098
        "punpckhwd %%mm1, %%mm3                 \n\t"
2099
        "movq %%mm2, %%mm1                      \n\t"
2100
        "punpcklwd %%mm4, %%mm2                 \n\t"
2101
        "punpckhwd %%mm4, %%mm1                 \n\t"
2102

2103
        "movd %%mm0, (%0)                       \n\t"
2104
        "psrlq $32, %%mm0                       \n\t"
2105
        "movd %%mm0, (%%"REG_a")                \n\t"
2106
        "movd %%mm3, (%%"REG_a", %1)            \n\t"
2107
        "psrlq $32, %%mm3                       \n\t"
2108
        "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
2109
        "movd %%mm2, (%0, %1, 4)                \n\t"
2110
        "psrlq $32, %%mm2                       \n\t"
2111
        "movd %%mm2, (%%"REG_d")                \n\t"
2112
        "movd %%mm1, (%%"REG_d", %1)            \n\t"
2113
        "psrlq $32, %%mm1                       \n\t"
2114
        "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
2115

2116

2117
        "movq 64(%2), %%mm0                     \n\t" // 12345678
2118
        "movq 80(%2), %%mm1                     \n\t" // abcdefgh
2119
        "movq %%mm0, %%mm2                      \n\t" // 12345678
2120
        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2121
        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2122

2123
        "movq 96(%2), %%mm1                     \n\t"
2124
        "movq 112(%2), %%mm3                    \n\t"
2125
        "movq %%mm1, %%mm4                      \n\t"
2126
        "punpcklbw %%mm3, %%mm1                 \n\t"
2127
        "punpckhbw %%mm3, %%mm4                 \n\t"
2128

2129
        "movq %%mm0, %%mm3                      \n\t"
2130
        "punpcklwd %%mm1, %%mm0                 \n\t"
2131
        "punpckhwd %%mm1, %%mm3                 \n\t"
2132
        "movq %%mm2, %%mm1                      \n\t"
2133
        "punpcklwd %%mm4, %%mm2                 \n\t"
2134
        "punpckhwd %%mm4, %%mm1                 \n\t"
2135

2136
        "movd %%mm0, 4(%0)                      \n\t"
2137
        "psrlq $32, %%mm0                       \n\t"
2138
        "movd %%mm0, 4(%%"REG_a")               \n\t"
2139
        "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
2140
        "psrlq $32, %%mm3                       \n\t"
2141
        "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
2142
        "movd %%mm2, 4(%0, %1, 4)               \n\t"
2143
        "psrlq $32, %%mm2                       \n\t"
2144
        "movd %%mm2, 4(%%"REG_d")               \n\t"
2145
        "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
2146
        "psrlq $32, %%mm1                       \n\t"
2147
        "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
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
    tempBlurredPast[127]= maxNoise[0];
2162
    tempBlurredPast[128]= maxNoise[1];
2163
    tempBlurredPast[129]= maxNoise[2];
2164

2165
#define FAST_L2_DIFF
2166
//#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
        "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
2170
        "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
2171
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2172
//      0       1       2       3       4       5       6       7       8       9
2173
//      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
2174
//FIXME reorder?
2175
#ifdef L1_DIFF //needs mmx2
2176
        "movq (%0), %%mm0                       \n\t" // L0
2177
        "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
2178
        "movq (%0, %2), %%mm1                   \n\t" // L1
2179
        "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
2180
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2181
        "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
2182
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2183
        "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
2184

2185
        "movq (%0, %2, 4), %%mm4                \n\t" // L4
2186
        "paddw %%mm1, %%mm0                     \n\t"
2187
        "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
2188
        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2189
        "paddw %%mm2, %%mm0                     \n\t"
2190
        "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
2191
        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2192
        "paddw %%mm3, %%mm0                     \n\t"
2193
        "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
2194
        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2195
        "paddw %%mm4, %%mm0                     \n\t"
2196
        "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
2197
        "paddw %%mm5, %%mm6                     \n\t"
2198
        "paddw %%mm7, %%mm6                     \n\t"
2199
        "paddw %%mm6, %%mm0                     \n\t"
2200
#else //L1_DIFF
2201
#if defined (FAST_L2_DIFF)
2202
        "pcmpeqb %%mm7, %%mm7                   \n\t"
2203
        "movq "MANGLE(b80)", %%mm6              \n\t"
2204
        "pxor %%mm0, %%mm0                      \n\t"
2205
#define REAL_L2_DIFF_CORE(a, b)\
2206
        "movq " #a ", %%mm5                     \n\t"\
2207
        "movq " #b ", %%mm2                     \n\t"\
2208
        "pxor %%mm7, %%mm2                      \n\t"\
2209
        PAVGB(%%mm2, %%mm5)\
2210
        "paddb %%mm6, %%mm5                     \n\t"\
2211
        "movq %%mm5, %%mm2                      \n\t"\
2212
        "psllw $8, %%mm5                        \n\t"\
2213
        "pmaddwd %%mm5, %%mm5                   \n\t"\
2214
        "pmaddwd %%mm2, %%mm2                   \n\t"\
2215
        "paddd %%mm2, %%mm5                     \n\t"\
2216
        "psrld $14, %%mm5                       \n\t"\
2217
        "paddd %%mm5, %%mm0                     \n\t"
2218

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

2238
#endif //defined (FAST_L2_DIFF)
2239

2240
#define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)
2241

2242
L2_DIFF_CORE((%0)          , (%1))
2243
L2_DIFF_CORE((%0, %2)      , (%1, %2))
2244
L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))
2245
L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))
2246
L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))
2247
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
        "movq %%mm0, %%mm4                      \n\t"
2254
        "psrlq $32, %%mm0                       \n\t"
2255
        "paddd %%mm0, %%mm4                     \n\t"
2256
        "movd %%mm4, %%ecx                      \n\t"
2257
        "shll $2, %%ecx                         \n\t"
2258
        "mov %3, %%"REG_d"                      \n\t"
2259
        "addl -4(%%"REG_d"), %%ecx              \n\t"
2260
        "addl 4(%%"REG_d"), %%ecx               \n\t"
2261
        "addl -1024(%%"REG_d"), %%ecx           \n\t"
2262
        "addl $4, %%ecx                         \n\t"
2263
        "addl 1024(%%"REG_d"), %%ecx            \n\t"
2264
        "shrl $3, %%ecx                         \n\t"
2265
        "movl %%ecx, (%%"REG_d")                \n\t"
2266

2267
//        "mov %3, %%"REG_c"                      \n\t"
2268
//        "mov %%"REG_c", test                    \n\t"
2269
//        "jmp 4f                                 \n\t"
2270
        "cmpl 512(%%"REG_d"), %%ecx             \n\t"
2271
        " jb 2f                                 \n\t"
2272
        "cmpl 516(%%"REG_d"), %%ecx             \n\t"
2273
        " jb 1f                                 \n\t"
2274

2275
        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2276
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2277
        "movq (%0), %%mm0                       \n\t" // L0
2278
        "movq (%0, %2), %%mm1                   \n\t" // L1
2279
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2280
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2281
        "movq (%0, %2, 4), %%mm4                \n\t" // L4
2282
        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2283
        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2284
        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2285
        "movq %%mm0, (%1)                       \n\t" // L0
2286
        "movq %%mm1, (%1, %2)                   \n\t" // L1
2287
        "movq %%mm2, (%1, %2, 2)                \n\t" // L2
2288
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
2289
        "movq %%mm4, (%1, %2, 4)                \n\t" // L4
2290
        "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
2291
        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
2292
        "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
2293
        "jmp 4f                                 \n\t"
2294

2295
        "1:                                     \n\t"
2296
        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2297
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2298
        "movq (%0), %%mm0                       \n\t" // L0
2299
        PAVGB((%1), %%mm0)                            // L0
2300
        "movq (%0, %2), %%mm1                   \n\t" // L1
2301
        PAVGB((%1, %2), %%mm1)                        // L1
2302
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2303
        PAVGB((%1, %2, 2), %%mm2)                     // L2
2304
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2305
        PAVGB((%1, %%REGa), %%mm3)                    // L3
2306
        "movq (%0, %2, 4), %%mm4                \n\t" // L4
2307
        PAVGB((%1, %2, 4), %%mm4)                     // L4
2308
        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2309
        PAVGB((%1, %%REGd), %%mm5)                    // L5
2310
        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2311
        PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
2312
        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2313
        PAVGB((%1, %%REGc), %%mm7)                    // L7
2314
        "movq %%mm0, (%1)                       \n\t" // R0
2315
        "movq %%mm1, (%1, %2)                   \n\t" // R1
2316
        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2317
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2318
        "movq %%mm4, (%1, %2, 4)                \n\t" // R4
2319
        "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
2320
        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
2321
        "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
2322
        "movq %%mm0, (%0)                       \n\t" // L0
2323
        "movq %%mm1, (%0, %2)                   \n\t" // L1
2324
        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2325
        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2326
        "movq %%mm4, (%0, %2, 4)                \n\t" // L4
2327
        "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
2328
        "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
2329
        "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
2330
        "jmp 4f                                 \n\t"
2331

2332
        "2:                                     \n\t"
2333
        "cmpl 508(%%"REG_d"), %%ecx             \n\t"
2334
        " jb 3f                                 \n\t"
2335

2336
        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2337
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2338
        "movq (%0), %%mm0                       \n\t" // L0
2339
        "movq (%0, %2), %%mm1                   \n\t" // L1
2340
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2341
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2342
        "movq (%1), %%mm4                       \n\t" // R0
2343
        "movq (%1, %2), %%mm5                   \n\t" // R1
2344
        "movq (%1, %2, 2), %%mm6                \n\t" // R2
2345
        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2346
        PAVGB(%%mm4, %%mm0)
2347
        PAVGB(%%mm5, %%mm1)
2348
        PAVGB(%%mm6, %%mm2)
2349
        PAVGB(%%mm7, %%mm3)
2350
        PAVGB(%%mm4, %%mm0)
2351
        PAVGB(%%mm5, %%mm1)
2352
        PAVGB(%%mm6, %%mm2)
2353
        PAVGB(%%mm7, %%mm3)
2354
        "movq %%mm0, (%1)                       \n\t" // R0
2355
        "movq %%mm1, (%1, %2)                   \n\t" // R1
2356
        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2357
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2358
        "movq %%mm0, (%0)                       \n\t" // L0
2359
        "movq %%mm1, (%0, %2)                   \n\t" // L1
2360
        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2361
        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2362

2363
        "movq (%0, %2, 4), %%mm0                \n\t" // L4
2364
        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2365
        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2366
        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2367
        "movq (%1, %2, 4), %%mm4                \n\t" // R4
2368
        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2369
        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2370
        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2371
        PAVGB(%%mm4, %%mm0)
2372
        PAVGB(%%mm5, %%mm1)
2373
        PAVGB(%%mm6, %%mm2)
2374
        PAVGB(%%mm7, %%mm3)
2375
        PAVGB(%%mm4, %%mm0)
2376
        PAVGB(%%mm5, %%mm1)
2377
        PAVGB(%%mm6, %%mm2)
2378
        PAVGB(%%mm7, %%mm3)
2379
        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2380
        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2381
        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2382
        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2383
        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2384
        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2385
        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2386
        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2387
        "jmp 4f                                 \n\t"
2388

2389
        "3:                                     \n\t"
2390
        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2391
        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2392
        "movq (%0), %%mm0                       \n\t" // L0
2393
        "movq (%0, %2), %%mm1                   \n\t" // L1
2394
        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2395
        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2396
        "movq (%1), %%mm4                       \n\t" // R0
2397
        "movq (%1, %2), %%mm5                   \n\t" // R1
2398
        "movq (%1, %2, 2), %%mm6                \n\t" // R2
2399
        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2400
        PAVGB(%%mm4, %%mm0)
2401
        PAVGB(%%mm5, %%mm1)
2402
        PAVGB(%%mm6, %%mm2)
2403
        PAVGB(%%mm7, %%mm3)
2404
        PAVGB(%%mm4, %%mm0)
2405
        PAVGB(%%mm5, %%mm1)
2406
        PAVGB(%%mm6, %%mm2)
2407
        PAVGB(%%mm7, %%mm3)
2408
        PAVGB(%%mm4, %%mm0)
2409
        PAVGB(%%mm5, %%mm1)
2410
        PAVGB(%%mm6, %%mm2)
2411
        PAVGB(%%mm7, %%mm3)
2412
        "movq %%mm0, (%1)                       \n\t" // R0
2413
        "movq %%mm1, (%1, %2)                   \n\t" // R1
2414
        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2415
        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2416
        "movq %%mm0, (%0)                       \n\t" // L0
2417
        "movq %%mm1, (%0, %2)                   \n\t" // L1
2418
        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2419
        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2420

2421
        "movq (%0, %2, 4), %%mm0                \n\t" // L4
2422
        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2423
        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2424
        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2425
        "movq (%1, %2, 4), %%mm4                \n\t" // R4
2426
        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2427
        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2428
        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2429
        PAVGB(%%mm4, %%mm0)
2430
        PAVGB(%%mm5, %%mm1)
2431
        PAVGB(%%mm6, %%mm2)
2432
        PAVGB(%%mm7, %%mm3)
2433
        PAVGB(%%mm4, %%mm0)
2434
        PAVGB(%%mm5, %%mm1)
2435
        PAVGB(%%mm6, %%mm2)
2436
        PAVGB(%%mm7, %%mm3)
2437
        PAVGB(%%mm4, %%mm0)
2438
        PAVGB(%%mm5, %%mm1)
2439
        PAVGB(%%mm6, %%mm2)
2440
        PAVGB(%%mm7, %%mm3)
2441
        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2442
        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2443
        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2444
        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2445
        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2446
        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2447
        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2448
        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2449

2450
        "4:                                     \n\t"
2451

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

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

2486
/*
2487
Switch between
2488
 1  0  0  0  0  0  0  (0)
2489
64 32 16  8  4  2  1  (1)
2490
64 48 36 27 20 15 11 (33) (approx)
2491
64 56 49 43 37 33 29 (200) (approx)
2492
*/
2493
    if(d > maxNoise[1]){
2494
        if(d < maxNoise[2]){
2495
            for(y=0; y<8; y++){
2496
                int x;
2497
                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
                    src[ x + y*stride ]=
2502
                        (ref + cur + 1)>>1;
2503
                }
2504
            }
2505
        }else{
2506
            for(y=0; y<8; y++){
2507
                int x;
2508
                for(x=0; x<8; x++){
2509
                    tempBlurred[ x + y*stride ]= src[ x + y*stride ];
2510
                }
2511
            }
2512
        }
2513
    }else{
2514
        if(d < maxNoise[0]){
2515
            for(y=0; y<8; y++){
2516
                int x;
2517
                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
                    src[ x + y*stride ]=
2522
                        (ref*7 + cur + 4)>>3;
2523
                }
2524
            }
2525
        }else{
2526
            for(y=0; y<8; y++){
2527
                int x;
2528
                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
                    src[ x + y*stride ]=
2533
                        (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
 * accurate deblock filter
2546
 */
2547
static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, const PPContext *c, int mode){
2548
    int64_t dc_mask, eq_mask, both_masks;
2549
    int64_t sums[10*8*2];
2550
    src+= step*3; // src points to begin of the 8x8 Block
2551
    //{ START_TIMER
2552
    __asm__ volatile(
2553
        "movq %0, %%mm7                         \n\t"
2554
        "movq %1, %%mm6                         \n\t"
2555
        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
2556
        );
2557

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

2563
        "movq (%2), %%mm0                       \n\t"
2564
        "movq (%%"REG_a"), %%mm1                \n\t"
2565
        "movq %%mm1, %%mm3                      \n\t"
2566
        "movq %%mm1, %%mm4                      \n\t"
2567
        "psubb %%mm1, %%mm0                     \n\t" // mm0 = difference
2568
        "paddb %%mm7, %%mm0                     \n\t"
2569
        "pcmpgtb %%mm6, %%mm0                   \n\t"
2570

2571
        "movq (%%"REG_a",%3), %%mm2             \n\t"
2572
        PMAXUB(%%mm2, %%mm4)
2573
        PMINUB(%%mm2, %%mm3, %%mm5)
2574
        "psubb %%mm2, %%mm1                     \n\t"
2575
        "paddb %%mm7, %%mm1                     \n\t"
2576
        "pcmpgtb %%mm6, %%mm1                   \n\t"
2577
        "paddb %%mm1, %%mm0                     \n\t"
2578

2579
        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2580
        PMAXUB(%%mm1, %%mm4)
2581
        PMINUB(%%mm1, %%mm3, %%mm5)
2582
        "psubb %%mm1, %%mm2                     \n\t"
2583
        "paddb %%mm7, %%mm2                     \n\t"
2584
        "pcmpgtb %%mm6, %%mm2                   \n\t"
2585
        "paddb %%mm2, %%mm0                     \n\t"
2586

2587
        "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
2588

2589
        "movq (%2, %3, 4), %%mm2                \n\t"
2590
        PMAXUB(%%mm2, %%mm4)
2591
        PMINUB(%%mm2, %%mm3, %%mm5)
2592
        "psubb %%mm2, %%mm1                     \n\t"
2593
        "paddb %%mm7, %%mm1                     \n\t"
2594
        "pcmpgtb %%mm6, %%mm1                   \n\t"
2595
        "paddb %%mm1, %%mm0                     \n\t"
2596

2597
        "movq (%%"REG_a"), %%mm1                \n\t"
2598
        PMAXUB(%%mm1, %%mm4)
2599
        PMINUB(%%mm1, %%mm3, %%mm5)
2600
        "psubb %%mm1, %%mm2                     \n\t"
2601
        "paddb %%mm7, %%mm2                     \n\t"
2602
        "pcmpgtb %%mm6, %%mm2                   \n\t"
2603
        "paddb %%mm2, %%mm0                     \n\t"
2604

2605
        "movq (%%"REG_a", %3), %%mm2            \n\t"
2606
        PMAXUB(%%mm2, %%mm4)
2607
        PMINUB(%%mm2, %%mm3, %%mm5)
2608
        "psubb %%mm2, %%mm1                     \n\t"
2609
        "paddb %%mm7, %%mm1                     \n\t"
2610
        "pcmpgtb %%mm6, %%mm1                   \n\t"
2611
        "paddb %%mm1, %%mm0                     \n\t"
2612

2613
        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2614
        PMAXUB(%%mm1, %%mm4)
2615
        PMINUB(%%mm1, %%mm3, %%mm5)
2616
        "psubb %%mm1, %%mm2                     \n\t"
2617
        "paddb %%mm7, %%mm2                     \n\t"
2618
        "pcmpgtb %%mm6, %%mm2                   \n\t"
2619
        "paddb %%mm2, %%mm0                     \n\t"
2620

2621
        "movq (%2, %3, 8), %%mm2                \n\t"
2622
        PMAXUB(%%mm2, %%mm4)
2623
        PMINUB(%%mm2, %%mm3, %%mm5)
2624
        "psubb %%mm2, %%mm1                     \n\t"
2625
        "paddb %%mm7, %%mm1                     \n\t"
2626
        "pcmpgtb %%mm6, %%mm1                   \n\t"
2627
        "paddb %%mm1, %%mm0                     \n\t"
2628

2629
        "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
2630
        "psubb %%mm1, %%mm2                     \n\t"
2631
        "paddb %%mm7, %%mm2                     \n\t"
2632
        "pcmpgtb %%mm6, %%mm2                   \n\t"
2633
        "paddb %%mm2, %%mm0                     \n\t"
2634
        "psubusb %%mm3, %%mm4                   \n\t"
2635

2636
        "pxor %%mm6, %%mm6                      \n\t"
2637
        "movq %4, %%mm7                         \n\t" // QP,..., QP
2638
        "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
2639
        "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
2640
        "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2641
        "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2642
        "movq %%mm7, %1                         \n\t"
2643

2644
        "movq %5, %%mm7                         \n\t"
2645
        "punpcklbw %%mm7, %%mm7                 \n\t"
2646
        "punpcklbw %%mm7, %%mm7                 \n\t"
2647
        "punpcklbw %%mm7, %%mm7                 \n\t"
2648
        "psubb %%mm0, %%mm6                     \n\t"
2649
        "pcmpgtb %%mm7, %%mm6                   \n\t"
2650
        "movq %%mm6, %0                         \n\t"
2651

2652
        : "=m" (eq_mask), "=m" (dc_mask)
2653
        : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2654
        : "%"REG_a
2655
    );
2656

2657
    both_masks = dc_mask & eq_mask;
2658

2659
    if(both_masks){
2660
        x86_reg offset= -8*step;
2661
        int64_t *temp_sums= sums;
2662

2663
        __asm__ volatile(
2664
            "movq %2, %%mm0                         \n\t"  // QP,..., QP
2665
            "pxor %%mm4, %%mm4                      \n\t"
2666

2667
            "movq (%0), %%mm6                       \n\t"
2668
            "movq (%0, %1), %%mm5                   \n\t"
2669
            "movq %%mm5, %%mm1                      \n\t"
2670
            "movq %%mm6, %%mm2                      \n\t"
2671
            "psubusb %%mm6, %%mm5                   \n\t"
2672
            "psubusb %%mm1, %%mm2                   \n\t"
2673
            "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2674
            "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2675
            "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2676

2677
            "pxor %%mm6, %%mm1                      \n\t"
2678
            "pand %%mm0, %%mm1                      \n\t"
2679
            "pxor %%mm1, %%mm6                      \n\t"
2680
            // 0:QP  6:First
2681

2682
            "movq (%0, %1, 8), %%mm5                \n\t"
2683
            "add %1, %0                             \n\t" // %0 points to line 1 not 0
2684
            "movq (%0, %1, 8), %%mm7                \n\t"
2685
            "movq %%mm5, %%mm1                      \n\t"
2686
            "movq %%mm7, %%mm2                      \n\t"
2687
            "psubusb %%mm7, %%mm5                   \n\t"
2688
            "psubusb %%mm1, %%mm2                   \n\t"
2689
            "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2690
            "movq %2, %%mm0                         \n\t"  // QP,..., QP
2691
            "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2692
            "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2693

2694
            "pxor %%mm7, %%mm1                      \n\t"
2695
            "pand %%mm0, %%mm1                      \n\t"
2696
            "pxor %%mm1, %%mm7                      \n\t"
2697

2698
            "movq %%mm6, %%mm5                      \n\t"
2699
            "punpckhbw %%mm4, %%mm6                 \n\t"
2700
            "punpcklbw %%mm4, %%mm5                 \n\t"
2701
            // 4:0 5/6:First 7:Last
2702

2703
            "movq %%mm5, %%mm0                      \n\t"
2704
            "movq %%mm6, %%mm1                      \n\t"
2705
            "psllw $2, %%mm0                        \n\t"
2706
            "psllw $2, %%mm1                        \n\t"
2707
            "paddw "MANGLE(w04)", %%mm0             \n\t"
2708
            "paddw "MANGLE(w04)", %%mm1             \n\t"
2709

2710
#define NEXT\
2711
            "movq (%0), %%mm2                       \n\t"\
2712
            "movq (%0), %%mm3                       \n\t"\
2713
            "add %1, %0                             \n\t"\
2714
            "punpcklbw %%mm4, %%mm2                 \n\t"\
2715
            "punpckhbw %%mm4, %%mm3                 \n\t"\
2716
            "paddw %%mm2, %%mm0                     \n\t"\
2717
            "paddw %%mm3, %%mm1                     \n\t"
2718

2719
#define PREV\
2720
            "movq (%0), %%mm2                       \n\t"\
2721
            "movq (%0), %%mm3                       \n\t"\
2722
            "add %1, %0                             \n\t"\
2723
            "punpcklbw %%mm4, %%mm2                 \n\t"\
2724
            "punpckhbw %%mm4, %%mm3                 \n\t"\
2725
            "psubw %%mm2, %%mm0                     \n\t"\
2726
            "psubw %%mm3, %%mm1                     \n\t"
2727

2728

2729
            NEXT //0
2730
            NEXT //1
2731
            NEXT //2
2732
            "movq %%mm0, (%3)                       \n\t"
2733
            "movq %%mm1, 8(%3)                      \n\t"
2734

2735
            NEXT //3
2736
            "psubw %%mm5, %%mm0                     \n\t"
2737
            "psubw %%mm6, %%mm1                     \n\t"
2738
            "movq %%mm0, 16(%3)                     \n\t"
2739
            "movq %%mm1, 24(%3)                     \n\t"
2740

2741
            NEXT //4
2742
            "psubw %%mm5, %%mm0                     \n\t"
2743
            "psubw %%mm6, %%mm1                     \n\t"
2744
            "movq %%mm0, 32(%3)                     \n\t"
2745
            "movq %%mm1, 40(%3)                     \n\t"
2746

2747
            NEXT //5
2748
            "psubw %%mm5, %%mm0                     \n\t"
2749
            "psubw %%mm6, %%mm1                     \n\t"
2750
            "movq %%mm0, 48(%3)                     \n\t"
2751
            "movq %%mm1, 56(%3)                     \n\t"
2752

2753
            NEXT //6
2754
            "psubw %%mm5, %%mm0                     \n\t"
2755
            "psubw %%mm6, %%mm1                     \n\t"
2756
            "movq %%mm0, 64(%3)                     \n\t"
2757
            "movq %%mm1, 72(%3)                     \n\t"
2758

2759
            "movq %%mm7, %%mm6                      \n\t"
2760
            "punpckhbw %%mm4, %%mm7                 \n\t"
2761
            "punpcklbw %%mm4, %%mm6                 \n\t"
2762

2763
            NEXT //7
2764
            "mov %4, %0                             \n\t"
2765
            "add %1, %0                             \n\t"
2766
            PREV //0
2767
            "movq %%mm0, 80(%3)                     \n\t"
2768
            "movq %%mm1, 88(%3)                     \n\t"
2769

2770
            PREV //1
2771
            "paddw %%mm6, %%mm0                     \n\t"
2772
            "paddw %%mm7, %%mm1                     \n\t"
2773
            "movq %%mm0, 96(%3)                     \n\t"
2774
            "movq %%mm1, 104(%3)                    \n\t"
2775

2776
            PREV //2
2777
            "paddw %%mm6, %%mm0                     \n\t"
2778
            "paddw %%mm7, %%mm1                     \n\t"
2779
            "movq %%mm0, 112(%3)                    \n\t"
2780
            "movq %%mm1, 120(%3)                    \n\t"
2781

2782
            PREV //3
2783
            "paddw %%mm6, %%mm0                     \n\t"
2784
            "paddw %%mm7, %%mm1                     \n\t"
2785
            "movq %%mm0, 128(%3)                    \n\t"
2786
            "movq %%mm1, 136(%3)                    \n\t"
2787

2788
            PREV //4
2789
            "paddw %%mm6, %%mm0                     \n\t"
2790
            "paddw %%mm7, %%mm1                     \n\t"
2791
            "movq %%mm0, 144(%3)                    \n\t"
2792
            "movq %%mm1, 152(%3)                    \n\t"
2793

2794
            "mov %4, %0                             \n\t" //FIXME
2795

2796
            : "+&r"(src)
2797
            : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)
2798
              NAMED_CONSTRAINTS_ADD(w04)
2799
        );
2800

2801
        src+= step; // src points to begin of the 8x8 Block
2802

2803
        __asm__ volatile(
2804
            "movq %4, %%mm6                         \n\t"
2805
            "pcmpeqb %%mm5, %%mm5                   \n\t"
2806
            "pxor %%mm6, %%mm5                      \n\t"
2807
            "pxor %%mm7, %%mm7                      \n\t"
2808

2809
            "1:                                     \n\t"
2810
            "movq (%1), %%mm0                       \n\t"
2811
            "movq 8(%1), %%mm1                      \n\t"
2812
            "paddw 32(%1), %%mm0                    \n\t"
2813
            "paddw 40(%1), %%mm1                    \n\t"
2814
            "movq (%0, %3), %%mm2                   \n\t"
2815
            "movq %%mm2, %%mm3                      \n\t"
2816
            "movq %%mm2, %%mm4                      \n\t"
2817
            "punpcklbw %%mm7, %%mm2                 \n\t"
2818
            "punpckhbw %%mm7, %%mm3                 \n\t"
2819
            "paddw %%mm2, %%mm0                     \n\t"
2820
            "paddw %%mm3, %%mm1                     \n\t"
2821
            "paddw %%mm2, %%mm0                     \n\t"
2822
            "paddw %%mm3, %%mm1                     \n\t"
2823
            "psrlw $4, %%mm0                        \n\t"
2824
            "psrlw $4, %%mm1                        \n\t"
2825
            "packuswb %%mm1, %%mm0                  \n\t"
2826
            "pand %%mm6, %%mm0                      \n\t"
2827
            "pand %%mm5, %%mm4                      \n\t"
2828
            "por %%mm4, %%mm0                       \n\t"
2829
            "movq %%mm0, (%0, %3)                   \n\t"
2830
            "add $16, %1                            \n\t"
2831
            "add %2, %0                             \n\t"
2832
            " js 1b                                 \n\t"
2833

2834
            : "+r"(offset), "+r"(temp_sums)
2835
            : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)
2836
        );
2837
    }else
2838
        src+= step; // src points to begin of the 8x8 Block
2839

2840
    if(eq_mask != -1LL){
2841
        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
//      0       1       2       3       4       5       6       7       8       9
2846
//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1
2847

2848
            "movq (%0), %%mm0                       \n\t"
2849
            "movq %%mm0, %%mm1                      \n\t"
2850
            "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
2851
            "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
2852

2853
            "movq (%0, %1), %%mm2                   \n\t"
2854
            "lea (%0, %1, 2), %%"REG_a"             \n\t"
2855
            "movq %%mm2, %%mm3                      \n\t"
2856
            "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
2857
            "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
2858

2859
            "movq (%%"REG_a"), %%mm4                \n\t"
2860
            "movq %%mm4, %%mm5                      \n\t"
2861
            "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
2862
            "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
2863

2864
            "paddw %%mm0, %%mm0                     \n\t" // 2L0
2865
            "paddw %%mm1, %%mm1                     \n\t" // 2H0
2866
            "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
2867
            "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
2868
            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
2869
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
2870

2871
            "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
2872
            "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
2873
            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
2874
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
2875

2876
            "movq (%%"REG_a", %1), %%mm2            \n\t"
2877
            "movq %%mm2, %%mm3                      \n\t"
2878
            "punpcklbw %%mm7, %%mm2                 \n\t" // L3
2879
            "punpckhbw %%mm7, %%mm3                 \n\t" // H3
2880

2881
            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
2882
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
2883
            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2884
            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2885
            "movq %%mm0, (%4)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2886
            "movq %%mm1, 8(%4)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2887

2888
            "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
2889
            "movq %%mm0, %%mm1                      \n\t"
2890
            "punpcklbw %%mm7, %%mm0                 \n\t" // L4
2891
            "punpckhbw %%mm7, %%mm1                 \n\t" // H4
2892

2893
            "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
2894
            "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
2895
            "movq %%mm2, 16(%4)                     \n\t" // L3 - L4
2896
            "movq %%mm3, 24(%4)                     \n\t" // H3 - H4
2897
            "paddw %%mm4, %%mm4                     \n\t" // 2L2
2898
            "paddw %%mm5, %%mm5                     \n\t" // 2H2
2899
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
2900
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
2901

2902
            "lea (%%"REG_a", %1), %0                \n\t"
2903
            "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
2904
            "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
2905
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
2906
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
2907
//50 opcodes so far
2908
            "movq (%0, %1, 2), %%mm2                \n\t"
2909
            "movq %%mm2, %%mm3                      \n\t"
2910
            "punpcklbw %%mm7, %%mm2                 \n\t" // L5
2911
            "punpckhbw %%mm7, %%mm3                 \n\t" // H5
2912
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
2913
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
2914
            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
2915
            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
2916

2917
            "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
2918
            "punpcklbw %%mm7, %%mm6                 \n\t" // L6
2919
            "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
2920
            "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
2921
            "punpckhbw %%mm7, %%mm6                 \n\t" // H6
2922
            "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
2923

2924
            "paddw %%mm0, %%mm0                     \n\t" // 2L4
2925
            "paddw %%mm1, %%mm1                     \n\t" // 2H4
2926
            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
2927
            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
2928

2929
            "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
2930
            "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
2931
            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
2932
            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
2933

2934
            "movq (%0, %1, 4), %%mm2                \n\t"
2935
            "movq %%mm2, %%mm3                      \n\t"
2936
            "punpcklbw %%mm7, %%mm2                 \n\t" // L7
2937
            "punpckhbw %%mm7, %%mm3                 \n\t" // H7
2938

2939
            "paddw %%mm2, %%mm2                     \n\t" // 2L7
2940
            "paddw %%mm3, %%mm3                     \n\t" // 2H7
2941
            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
2942
            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
2943

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

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

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

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

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

3009

3010
            "psubusw %%mm0, %%mm4                   \n\t" // hd
3011
            "psubusw %%mm1, %%mm5                   \n\t" // ld
3012

3013

3014
            "movq "MANGLE(w05)", %%mm2              \n\t" // 5
3015
            "pmullw %%mm2, %%mm4                    \n\t"
3016
            "pmullw %%mm2, %%mm5                    \n\t"
3017
            "movq "MANGLE(w20)", %%mm2              \n\t" // 32
3018
            "paddw %%mm2, %%mm4                     \n\t"
3019
            "paddw %%mm2, %%mm5                     \n\t"
3020
            "psrlw $6, %%mm4                        \n\t"
3021
            "psrlw $6, %%mm5                        \n\t"
3022

3023
            "movq 16(%4), %%mm0                     \n\t" // L3 - L4
3024
            "movq 24(%4), %%mm1                     \n\t" // H3 - H4
3025

3026
            "pxor %%mm2, %%mm2                      \n\t"
3027
            "pxor %%mm3, %%mm3                      \n\t"
3028

3029
            "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
3030
            "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
3031
            "pxor %%mm2, %%mm0                      \n\t"
3032
            "pxor %%mm3, %%mm1                      \n\t"
3033
            "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
3034
            "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
3035
            "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
3036
            "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
3037

3038
            "pxor %%mm6, %%mm2                      \n\t"
3039
            "pxor %%mm7, %%mm3                      \n\t"
3040
            "pand %%mm2, %%mm4                      \n\t"
3041
            "pand %%mm3, %%mm5                      \n\t"
3042

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

3068
            : "+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
/*if(step==16){
3075
    STOP_TIMER("step16")
3076
}else{
3077
    STOP_TIMER("stepX")
3078
}
3079
    } */
3080
}
3081
#endif //TEMPLATE_PP_MMX
3082

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
#undef SCALED_CPY
3092

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
        "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
3103
        "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
3104
        "lea (%2,%4), %%"REG_a"         \n\t"
3105
        "lea (%3,%5), %%"REG_d"         \n\t"
3106
        "pxor %%mm4, %%mm4              \n\t"
3107
#if TEMPLATE_PP_MMXEXT
3108
#define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \
3109
        "movq " #src1 ", %%mm0          \n\t"\
3110
        "movq " #src1 ", %%mm5          \n\t"\
3111
        "movq " #src2 ", %%mm1          \n\t"\
3112
        "movq " #src2 ", %%mm6          \n\t"\
3113
        "punpcklbw %%mm0, %%mm0         \n\t"\
3114
        "punpckhbw %%mm5, %%mm5         \n\t"\
3115
        "punpcklbw %%mm1, %%mm1         \n\t"\
3116
        "punpckhbw %%mm6, %%mm6         \n\t"\
3117
        "pmulhuw %%mm3, %%mm0           \n\t"\
3118
        "pmulhuw %%mm3, %%mm5           \n\t"\
3119
        "pmulhuw %%mm3, %%mm1           \n\t"\
3120
        "pmulhuw %%mm3, %%mm6           \n\t"\
3121
        "psubw %%mm2, %%mm0             \n\t"\
3122
        "psubw %%mm2, %%mm5             \n\t"\
3123
        "psubw %%mm2, %%mm1             \n\t"\
3124
        "psubw %%mm2, %%mm6             \n\t"\
3125
        "packuswb %%mm5, %%mm0          \n\t"\
3126
        "packuswb %%mm6, %%mm1          \n\t"\
3127
        "movq %%mm0, " #dst1 "          \n\t"\
3128
        "movq %%mm1, " #dst2 "          \n\t"\
3129

3130
#else //TEMPLATE_PP_MMXEXT
3131
#define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \
3132
        "movq " #src1 ", %%mm0          \n\t"\
3133
        "movq " #src1 ", %%mm5          \n\t"\
3134
        "punpcklbw %%mm4, %%mm0         \n\t"\
3135
        "punpckhbw %%mm4, %%mm5         \n\t"\
3136
        "psubw %%mm2, %%mm0             \n\t"\
3137
        "psubw %%mm2, %%mm5             \n\t"\
3138
        "movq " #src2 ", %%mm1          \n\t"\
3139
        "psllw $6, %%mm0                \n\t"\
3140
        "psllw $6, %%mm5                \n\t"\
3141
        "pmulhw %%mm3, %%mm0            \n\t"\
3142
        "movq " #src2 ", %%mm6          \n\t"\
3143
        "pmulhw %%mm3, %%mm5            \n\t"\
3144
        "punpcklbw %%mm4, %%mm1         \n\t"\
3145
        "punpckhbw %%mm4, %%mm6         \n\t"\
3146
        "psubw %%mm2, %%mm1             \n\t"\
3147
        "psubw %%mm2, %%mm6             \n\t"\
3148
        "psllw $6, %%mm1                \n\t"\
3149
        "psllw $6, %%mm6                \n\t"\
3150
        "pmulhw %%mm3, %%mm1            \n\t"\
3151
        "pmulhw %%mm3, %%mm6            \n\t"\
3152
        "packuswb %%mm5, %%mm0          \n\t"\
3153
        "packuswb %%mm6, %%mm1          \n\t"\
3154
        "movq %%mm0, " #dst1 "          \n\t"\
3155
        "movq %%mm1, " #dst2 "          \n\t"\
3156

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

3161
SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
3162
SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
3163
SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
3164
        "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
3165
        "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
3166
SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
3167

3168

3169
        : "=&a" (packedOffsetAndScale)
3170
        : "0" (packedOffsetAndScale),
3171
        "r"(src),
3172
        "r"(dst),
3173
        "r" ((x86_reg)srcStride),
3174
        "r" ((x86_reg)dstStride)
3175
        : "%"REG_d
3176
    );
3177
#else //TEMPLATE_PP_MMX && HAVE_6REGS
3178
    for(i=0; i<8; i++)
3179
        memcpy( &(dst[dstStride*i]),
3180
                &(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
        "lea (%0,%2), %%"REG_a"                 \n\t"
3186
        "lea (%1,%3), %%"REG_d"                 \n\t"
3187

3188
#define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \
3189
        "movq " #src1 ", %%mm0          \n\t"\
3190
        "movq " #src2 ", %%mm1          \n\t"\
3191
        "movq %%mm0, " #dst1 "          \n\t"\
3192
        "movq %%mm1, " #dst2 "          \n\t"\
3193

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

3197
SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))
3198
SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
3199
SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
3200
        "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
3201
        "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
3202
SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
3203

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

3218
/**
3219
 * Duplicate the given 8 src pixels ? times upward
3220
 */
3221
static inline void RENAME(duplicate)(uint8_t src[], int stride)
3222
{
3223
#if TEMPLATE_PP_MMX
3224
    __asm__ volatile(
3225
        "movq (%0), %%mm0               \n\t"
3226
        "movq %%mm0, (%0, %1, 4)        \n\t"
3227
        "add %1, %0                     \n\t"
3228
        "movq %%mm0, (%0)               \n\t"
3229
        "movq %%mm0, (%0, %1)           \n\t"
3230
        "movq %%mm0, (%0, %1, 2)        \n\t"
3231
        "movq %%mm0, (%0, %1, 4)        \n\t"
3232
        : "+r" (src)
3233
        : "r" ((x86_reg)-stride)
3234
    );
3235
#else
3236
    int i;
3237
    uint8_t *p=src;
3238
    for(i=0; i<5; i++){
3239
        p-= stride;
3240
        memcpy(p, src, 8);
3241
    }
3242
#endif
3243
}
3244

3245
#if ARCH_X86 && TEMPLATE_PP_MMXEXT
3246
static inline void RENAME(prefetchnta)(const void *p)
3247
{
3248
    __asm__ volatile(   "prefetchnta (%0)\n\t"
3249
        : : "r" (p)
3250
    );
3251
}
3252

3253
static inline void RENAME(prefetcht0)(const void *p)
3254
{
3255
    __asm__ volatile(   "prefetcht0 (%0)\n\t"
3256
        : : "r" (p)
3257
    );
3258
}
3259

3260
static inline void RENAME(prefetcht1)(const void *p)
3261
{
3262
    __asm__ volatile(   "prefetcht1 (%0)\n\t"
3263
        : : "r" (p)
3264
    );
3265
}
3266

3267
static inline void RENAME(prefetcht2)(const void *p)
3268
{
3269
    __asm__ volatile(   "prefetcht2 (%0)\n\t"
3270
        : : "r" (p)
3271
    );
3272
}
3273
#elif !ARCH_X86 && AV_GCC_VERSION_AT_LEAST(3,2)
3274
static inline void RENAME(prefetchnta)(const void *p)
3275
{
3276
    __builtin_prefetch(p,0,0);
3277
}
3278
static inline void RENAME(prefetcht0)(const void *p)
3279
{
3280
    __builtin_prefetch(p,0,1);
3281
}
3282
static inline void RENAME(prefetcht1)(const void *p)
3283
{
3284
    __builtin_prefetch(p,0,2);
3285
}
3286
static inline void RENAME(prefetcht2)(const void *p)
3287
{
3288
    __builtin_prefetch(p,0,3);
3289
}
3290
#else
3291
static inline void RENAME(prefetchnta)(const void *p)
3292
{
3293
    return;
3294
}
3295
static inline void RENAME(prefetcht0)(const void *p)
3296
{
3297
    return;
3298
}
3299
static inline void RENAME(prefetcht1)(const void *p)
3300
{
3301
    return;
3302
}
3303
static inline void RENAME(prefetcht2)(const void *p)
3304
{
3305
    return;
3306
}
3307
#endif
3308
/**
3309
 * Filter array of bytes (Y or U or V values)
3310
 */
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
    DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
3315
    int x,y;
3316
#ifdef TEMPLATE_PP_TIME_MODE
3317
    const int mode= TEMPLATE_PP_TIME_MODE;
3318
#else
3319
    const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3320
#endif
3321
    int black=0, white=255; // blackest black and whitest white in the picture
3322
    int QPCorrecture= 256*256;
3323

3324
    int copyAhead;
3325
#if TEMPLATE_PP_MMX
3326
    int i;
3327
#endif
3328

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

3332
    //FIXME remove
3333
    uint64_t * const yHistogram= c.yHistogram;
3334
    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
    if (mode & VISUALIZE){
3339
        if(!(mode & (V_A_DEBLOCK | H_A_DEBLOCK)) || TEMPLATE_PP_MMX) {
3340
            av_log(c2, AV_LOG_WARNING, "Visualization is currently only supported with the accurate deblock filter without SIMD\n");
3341
        }
3342
    }
3343

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

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

3368
    copyAhead-= 8;
3369

3370
    if(!isColor){
3371
        uint64_t sum= 0;
3372
        int i;
3373
        uint64_t maxClipped;
3374
        uint64_t clipped;
3375
        AVRational scale;
3376

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

3381
        for(i=0; i<256; i++){
3382
            sum+= yHistogram[i];
3383
        }
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
        clipped= sum;
3389
        for(black=255; black>0; black--){
3390
            if(clipped < maxClipped) break;
3391
            clipped-= yHistogram[black];
3392
        }
3393

3394
        clipped= sum;
3395
        for(white=0; white<256; white++){
3396
            if(clipped < maxClipped) break;
3397
            clipped-= yHistogram[white];
3398
        }
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
#endif
3409

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

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

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

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

3430
        // From this point on it is guaranteed that we can read and write 16 lines downward
3431
        // 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
            RENAME(prefetchnta)(srcBlock + (((x>>2)&6) + copyAhead)*srcStride + 32);
3435
            RENAME(prefetchnta)(srcBlock + (((x>>2)&6) + copyAhead+1)*srcStride + 32);
3436
            RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead)*dstStride + 32);
3437
            RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead+1)*dstStride + 32);
3438

3439
            RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
3440
                              srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3441

3442
            RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
3443

3444
            if(mode & LINEAR_IPOL_DEINT_FILTER)
3445
                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3446
            else if(mode & LINEAR_BLEND_DEINT_FILTER)
3447
                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3448
            else if(mode & MEDIAN_DEINT_FILTER)
3449
                RENAME(deInterlaceMedian)(dstBlock, dstStride);
3450
            else if(mode & CUBIC_IPOL_DEINT_FILTER)
3451
                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3452
            else if(mode & FFMPEG_DEINT_FILTER)
3453
                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3454
            else if(mode & LOWPASS5_DEINT_FILTER)
3455
                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3456
/*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
3457
                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3458
*/
3459
            dstBlock+=8;
3460
            srcBlock+=8;
3461
        }
3462
        if(width==FFABS(dstStride))
3463
            linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
3464
        else{
3465
            int i;
3466
            for(i=0; i<copyAhead; i++){
3467
                memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
3468
            }
3469
        }
3470
    }
3471

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

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

3496
            /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
3497
            linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
3498

3499
            /* duplicate last line of dst to fill the void up to line (copyAhead) */
3500
            for(i=height-y+1; i<=copyAhead; i++)
3501
                    memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
3502

3503
            dstBlock= tempDst + dstStride;
3504
            srcBlock= tempSrc;
3505
        }
3506

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

3544
            RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3545
                              srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3546

3547
            if(mode & LINEAR_IPOL_DEINT_FILTER)
3548
                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3549
            else if(mode & LINEAR_BLEND_DEINT_FILTER)
3550
                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3551
            else if(mode & MEDIAN_DEINT_FILTER)
3552
                RENAME(deInterlaceMedian)(dstBlock, dstStride);
3553
            else if(mode & CUBIC_IPOL_DEINT_FILTER)
3554
                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3555
            else if(mode & FFMPEG_DEINT_FILTER)
3556
                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3557
            else if(mode & LOWPASS5_DEINT_FILTER)
3558
                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3559
/*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
3560
                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3561
*/
3562
            dstBlock+=8;
3563
            srcBlock+=8;
3564
          }
3565

3566
          dstBlock = dstBlockStart;
3567
          srcBlock = srcBlockStart;
3568

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

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

3585
                    if(t==1)
3586
                        RENAME(doVertLowPass)(dstBlock, stride, &c);
3587
                    else if(t==2)
3588
                        RENAME(doVertDefFilter)(dstBlock, stride, &c);
3589
                }else if(mode & V_A_DEBLOCK){
3590
                    RENAME(do_a_deblock)(dstBlock, stride, 1, &c, mode);
3591
                }
3592
            }
3593

3594
            dstBlock+=8;
3595
            srcBlock+=8;
3596
          }
3597

3598
          dstBlock = dstBlockStart;
3599
          srcBlock = srcBlockStart;
3600

3601
          for(x = startx, qp_index=0; x < endx; x+=BLOCK_SIZE, qp_index++){
3602
            const int stride= dstStride;
3603
            av_unused uint8_t *tmpXchg;
3604
            c.QP = c.QP_block[qp_index];
3605
            c.nonBQP = c.nonBQP_block[qp_index];
3606
            c.pQPb = c.pQPb_block[qp_index];
3607
            c.pQPb2 = c.pQPb2_block[qp_index];
3608
#if TEMPLATE_PP_MMX
3609
            RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3610
#endif
3611
            /* check if we have a previous block to deblock it with dstBlock */
3612
            if(x - 8 >= 0){
3613
#if TEMPLATE_PP_MMX
3614
                if(mode & H_X1_FILTER)
3615
                        RENAME(vertX1Filter)(tempBlock1, 16, &c);
3616
                else if(mode & H_DEBLOCK){
3617
                    const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3618
                    if(t==1)
3619
                        RENAME(doVertLowPass)(tempBlock1, 16, &c);
3620
                    else if(t==2)
3621
                        RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3622
                }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
                    transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
3636

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

3649
                    if(t==1)
3650
                        RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
3651
                    else if(t==2)
3652
                        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
                if(mode & DERING){
3659
                //FIXME filter first line
3660
                    if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3661
                }
3662

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

3672
            dstBlock+=8;
3673
            srcBlock+=8;
3674

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

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

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

3694
        /* did we use a tmp buffer for the last lines*/
3695
        if(y+15 >= height){
3696
            uint8_t *dstBlock= &(dst[y*dstStride]);
3697
            if(width==FFABS(dstStride))
3698
                linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
3699
            else{
3700
                int i;
3701
                for(i=0; i<height-y; i++){
3702
                    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
#endif
3712

3713
#ifdef DEBUG_BRIGHTNESS
3714
    if(!isColor){
3715
        int max=1;
3716
        int i;
3717
        for(i=0; i<256; i++)
3718
            if(yHistogram[i] > max) max=yHistogram[i];
3719

3720
        for(i=1; i<256; i++){
3721
            int x;
3722
            int start=yHistogram[i-1]/(max/256+1);
3723
            int end=yHistogram[i]/(max/256+1);
3724
            int inc= end > start ? 1 : -1;
3725
            for(x=start; x!=end+inc; x+=inc)
3726
                dst[ i*dstStride + x]+=128;
3727
        }
3728

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

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

3738
}
3739

3740
#undef RENAME
3741
#undef TEMPLATE_PP_C
3742
#undef TEMPLATE_PP_ALTIVEC
3743
#undef TEMPLATE_PP_MMX
3744
#undef TEMPLATE_PP_MMXEXT
3745
#undef TEMPLATE_PP_3DNOW
3746
#undef TEMPLATE_PP_SSE2
3747

3748
Product

Resources

Company
