hevcdsp: split the pred functions by width
[ffmpeg.git] / libavcodec / hevcdsp_template.c
1 /*
2  * HEVC video decoder
3  *
4  * Copyright (C) 2012 - 2013 Guillaume Martres
5  *
6  * This file is part of Libav.
7  *
8  * Libav is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * Libav is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with Libav; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22
23 #include "get_bits.h"
24 #include "hevc.h"
25
26 #include "bit_depth_template.c"
27
28 static void FUNC(put_pcm)(uint8_t *_dst, ptrdiff_t stride, int size,
29                           GetBitContext *gb, int pcm_bit_depth)
30 {
31     int x, y;
32     pixel *dst = (pixel *)_dst;
33
34     stride /= sizeof(pixel);
35
36     for (y = 0; y < size; y++) {
37         for (x = 0; x < size; x++)
38             dst[x] = get_bits(gb, pcm_bit_depth) << (BIT_DEPTH - pcm_bit_depth);
39         dst += stride;
40     }
41 }
42
43 static av_always_inline void FUNC(transquant_bypass)(uint8_t *_dst, int16_t *coeffs,
44                                                      ptrdiff_t stride, int size)
45 {
46     int x, y;
47     pixel *dst = (pixel *)_dst;
48
49     stride /= sizeof(pixel);
50
51     for (y = 0; y < size; y++) {
52         for (x = 0; x < size; x++) {
53             dst[x] = av_clip_pixel(dst[x] + *coeffs);
54             coeffs++;
55         }
56         dst += stride;
57     }
58 }
59
60 static void FUNC(transquant_bypass4x4)(uint8_t *_dst, int16_t *coeffs,
61                                        ptrdiff_t stride)
62 {
63     FUNC(transquant_bypass)(_dst, coeffs, stride, 4);
64 }
65
66 static void FUNC(transquant_bypass8x8)(uint8_t *_dst, int16_t *coeffs,
67                                        ptrdiff_t stride)
68 {
69     FUNC(transquant_bypass)(_dst, coeffs, stride, 8);
70 }
71
72 static void FUNC(transquant_bypass16x16)(uint8_t *_dst, int16_t *coeffs,
73                                          ptrdiff_t stride)
74 {
75     FUNC(transquant_bypass)(_dst, coeffs, stride, 16);
76 }
77
78 static void FUNC(transquant_bypass32x32)(uint8_t *_dst, int16_t *coeffs,
79                                          ptrdiff_t stride)
80 {
81     FUNC(transquant_bypass)(_dst, coeffs, stride, 32);
82 }
83
84 static void FUNC(transform_skip)(uint8_t *_dst, int16_t *coeffs,
85                                  ptrdiff_t stride)
86 {
87     pixel *dst = (pixel *)_dst;
88     int shift  = 13 - BIT_DEPTH;
89 #if BIT_DEPTH <= 13
90     int offset = 1 << (shift - 1);
91 #else
92     int offset = 0;
93 #endif
94     int x, y;
95
96     stride /= sizeof(pixel);
97
98     for (y = 0; y < 4 * 4; y += 4) {
99         for (x = 0; x < 4; x++)
100             dst[x] = av_clip_pixel(dst[x] + ((coeffs[y + x] + offset) >> shift));
101         dst += stride;
102     }
103 }
104
105 #define SET(dst, x)   (dst) = (x)
106 #define SCALE(dst, x) (dst) = av_clip_int16(((x) + add) >> shift)
107 #define ADD_AND_SCALE(dst, x)                                           \
108     (dst) = av_clip_pixel((dst) + av_clip_int16(((x) + add) >> shift))
109
110 #define TR_4x4_LUMA(dst, src, step, assign)                             \
111     do {                                                                \
112         int c0 = src[0 * step] + src[2 * step];                         \
113         int c1 = src[2 * step] + src[3 * step];                         \
114         int c2 = src[0 * step] - src[3 * step];                         \
115         int c3 = 74 * src[1 * step];                                    \
116                                                                         \
117         assign(dst[2 * step], 74 * (src[0 * step] -                     \
118                                     src[2 * step] +                     \
119                                     src[3 * step]));                    \
120         assign(dst[0 * step], 29 * c0 + 55 * c1 + c3);                  \
121         assign(dst[1 * step], 55 * c2 - 29 * c1 + c3);                  \
122         assign(dst[3 * step], 55 * c0 + 29 * c2 - c3);                  \
123     } while (0)
124
125 static void FUNC(transform_4x4_luma_add)(uint8_t *_dst, int16_t *coeffs,
126                                          ptrdiff_t stride)
127 {
128     int i;
129     pixel *dst   = (pixel *)_dst;
130     int shift    = 7;
131     int add      = 1 << (shift - 1);
132     int16_t *src = coeffs;
133
134     stride /= sizeof(pixel);
135
136     for (i = 0; i < 4; i++) {
137         TR_4x4_LUMA(src, src, 4, SCALE);
138         src++;
139     }
140
141     shift = 20 - BIT_DEPTH;
142     add   = 1 << (shift - 1);
143     for (i = 0; i < 4; i++) {
144         TR_4x4_LUMA(dst, coeffs, 1, ADD_AND_SCALE);
145         coeffs += 4;
146         dst    += stride;
147     }
148 }
149
150 #undef TR_4x4_LUMA
151
152 #define TR_4(dst, src, dstep, sstep, assign)                            \
153     do {                                                                \
154         const int e0 = transform[8 * 0][0] * src[0 * sstep] +           \
155                        transform[8 * 2][0] * src[2 * sstep];            \
156         const int e1 = transform[8 * 0][1] * src[0 * sstep] +           \
157                        transform[8 * 2][1] * src[2 * sstep];            \
158         const int o0 = transform[8 * 1][0] * src[1 * sstep] +           \
159                        transform[8 * 3][0] * src[3 * sstep];            \
160         const int o1 = transform[8 * 1][1] * src[1 * sstep] +           \
161                        transform[8 * 3][1] * src[3 * sstep];            \
162                                                                         \
163         assign(dst[0 * dstep], e0 + o0);                                \
164         assign(dst[1 * dstep], e1 + o1);                                \
165         assign(dst[2 * dstep], e1 - o1);                                \
166         assign(dst[3 * dstep], e0 - o0);                                \
167     } while (0)
168
169 static void FUNC(transform_4x4_add)(uint8_t *_dst, int16_t *coeffs,
170                                     ptrdiff_t stride)
171 {
172     int i;
173     pixel *dst   = (pixel *)_dst;
174     int shift    = 7;
175     int add      = 1 << (shift - 1);
176     int16_t *src = coeffs;
177
178     stride /= sizeof(pixel);
179
180     for (i = 0; i < 4; i++) {
181         TR_4(src, src, 4, 4, SCALE);
182         src++;
183     }
184
185     shift = 20 - BIT_DEPTH;
186     add   = 1 << (shift - 1);
187     for (i = 0; i < 4; i++) {
188         TR_4(dst, coeffs, 1, 1, ADD_AND_SCALE);
189         coeffs += 4;
190         dst    += stride;
191     }
192 }
193
194 #define TR_8(dst, src, dstep, sstep, assign)                      \
195     do {                                                          \
196         int i, j;                                                 \
197         int e_8[4];                                               \
198         int o_8[4] = { 0 };                                       \
199         for (i = 0; i < 4; i++)                                   \
200             for (j = 1; j < 8; j += 2)                            \
201                 o_8[i] += transform[4 * j][i] * src[j * sstep];   \
202         TR_4(e_8, src, 1, 2 * sstep, SET);                        \
203                                                                   \
204         for (i = 0; i < 4; i++) {                                 \
205             assign(dst[i * dstep], e_8[i] + o_8[i]);              \
206             assign(dst[(7 - i) * dstep], e_8[i] - o_8[i]);        \
207         }                                                         \
208     } while (0)
209
210 #define TR_16(dst, src, dstep, sstep, assign)                     \
211     do {                                                          \
212         int i, j;                                                 \
213         int e_16[8];                                              \
214         int o_16[8] = { 0 };                                      \
215         for (i = 0; i < 8; i++)                                   \
216             for (j = 1; j < 16; j += 2)                           \
217                 o_16[i] += transform[2 * j][i] * src[j * sstep];  \
218         TR_8(e_16, src, 1, 2 * sstep, SET);                       \
219                                                                   \
220         for (i = 0; i < 8; i++) {                                 \
221             assign(dst[i * dstep], e_16[i] + o_16[i]);            \
222             assign(dst[(15 - i) * dstep], e_16[i] - o_16[i]);     \
223         }                                                         \
224     } while (0)
225
226 #define TR_32(dst, src, dstep, sstep, assign)                     \
227     do {                                                          \
228         int i, j;                                                 \
229         int e_32[16];                                             \
230         int o_32[16] = { 0 };                                     \
231         for (i = 0; i < 16; i++)                                  \
232             for (j = 1; j < 32; j += 2)                           \
233                 o_32[i] += transform[j][i] * src[j * sstep];      \
234         TR_16(e_32, src, 1, 2 * sstep, SET);                      \
235                                                                   \
236         for (i = 0; i < 16; i++) {                                \
237             assign(dst[i * dstep], e_32[i] + o_32[i]);            \
238             assign(dst[(31 - i) * dstep], e_32[i] - o_32[i]);     \
239         }                                                         \
240     } while (0)
241
242
243
244 static void FUNC(transform_8x8_add)(uint8_t *_dst, int16_t *coeffs,
245                                     ptrdiff_t stride)
246 {
247     int i;
248     pixel *dst   = (pixel *)_dst;
249     int shift    = 7;
250     int add      = 1 << (shift - 1);
251     int16_t *src = coeffs;
252
253     stride /= sizeof(pixel);
254
255     for (i = 0; i < 8; i++) {
256         TR_8(src, src, 8, 8, SCALE);
257         src++;
258     }
259
260     shift = 20 - BIT_DEPTH;
261     add   = 1 << (shift - 1);
262     for (i = 0; i < 8; i++) {
263         TR_8(dst, coeffs, 1, 1, ADD_AND_SCALE);
264         coeffs += 8;
265         dst    += stride;
266     }
267 }
268
269 static void FUNC(transform_16x16_add)(uint8_t *_dst, int16_t *coeffs,
270                                       ptrdiff_t stride)
271 {
272     int i;
273     pixel *dst   = (pixel *)_dst;
274     int shift    = 7;
275     int add      = 1 << (shift - 1);
276     int16_t *src = coeffs;
277
278     stride /= sizeof(pixel);
279
280     for (i = 0; i < 16; i++) {
281         TR_16(src, src, 16, 16, SCALE);
282         src++;
283     }
284
285     shift = 20 - BIT_DEPTH;
286     add   = 1 << (shift - 1);
287     for (i = 0; i < 16; i++) {
288         TR_16(dst, coeffs, 1, 1, ADD_AND_SCALE);
289         coeffs += 16;
290         dst    += stride;
291     }
292 }
293
294 static void FUNC(transform_32x32_add)(uint8_t *_dst, int16_t *coeffs,
295                                       ptrdiff_t stride)
296 {
297     int i;
298     pixel *dst   = (pixel *)_dst;
299     int shift    = 7;
300     int add      = 1 << (shift - 1);
301     int16_t *src = coeffs;
302
303     stride /= sizeof(pixel);
304
305     for (i = 0; i < 32; i++) {
306         TR_32(src, src, 32, 32, SCALE);
307         src++;
308     }
309     src   = coeffs;
310     shift = 20 - BIT_DEPTH;
311     add   = 1 << (shift - 1);
312     for (i = 0; i < 32; i++) {
313         TR_32(dst, coeffs, 1, 1, ADD_AND_SCALE);
314         coeffs += 32;
315         dst    += stride;
316     }
317 }
318
319 static void FUNC(sao_band_filter)(uint8_t *_dst, uint8_t *_src,
320                                   ptrdiff_t stride, SAOParams *sao,
321                                   int *borders, int width, int height,
322                                   int c_idx, int class)
323 {
324     pixel *dst = (pixel *)_dst;
325     pixel *src = (pixel *)_src;
326     int offset_table[32] = { 0 };
327     int k, y, x;
328     int chroma = !!c_idx;
329     int shift  = BIT_DEPTH - 5;
330     int *sao_offset_val = sao->offset_val[c_idx];
331     int sao_left_class  = sao->band_position[c_idx];
332     int init_y = 0, init_x = 0;
333
334     stride /= sizeof(pixel);
335
336     switch (class) {
337     case 0:
338         if (!borders[2])
339             width -= (8 >> chroma) + 2;
340         if (!borders[3])
341             height -= (4 >> chroma) + 2;
342         break;
343     case 1:
344         init_y = -(4 >> chroma) - 2;
345         if (!borders[2])
346             width -= (8 >> chroma) + 2;
347         height = (4 >> chroma) + 2;
348         break;
349     case 2:
350         init_x = -(8 >> chroma) - 2;
351         width  =  (8 >> chroma) + 2;
352         if (!borders[3])
353             height -= (4 >> chroma) + 2;
354         break;
355     case 3:
356         init_y = -(4 >> chroma) - 2;
357         init_x = -(8 >> chroma) - 2;
358         width  =  (8 >> chroma) + 2;
359         height =  (4 >> chroma) + 2;
360         break;
361     }
362
363     dst = dst + (init_y * stride + init_x);
364     src = src + (init_y * stride + init_x);
365     for (k = 0; k < 4; k++)
366         offset_table[(k + sao_left_class) & 31] = sao_offset_val[k + 1];
367     for (y = 0; y < height; y++) {
368         for (x = 0; x < width; x++)
369             dst[x] = av_clip_pixel(src[x] + offset_table[src[x] >> shift]);
370         dst += stride;
371         src += stride;
372     }
373 }
374
375 static void FUNC(sao_band_filter_0)(uint8_t *dst, uint8_t *src,
376                                     ptrdiff_t stride, SAOParams *sao,
377                                     int *borders, int width, int height,
378                                     int c_idx)
379 {
380     FUNC(sao_band_filter)(dst, src, stride, sao, borders,
381                           width, height, c_idx, 0);
382 }
383
384 static void FUNC(sao_band_filter_1)(uint8_t *dst, uint8_t *src,
385                                     ptrdiff_t stride, SAOParams *sao,
386                                     int *borders, int width, int height,
387                                     int c_idx)
388 {
389     FUNC(sao_band_filter)(dst, src, stride, sao, borders,
390                           width, height, c_idx, 1);
391 }
392
393 static void FUNC(sao_band_filter_2)(uint8_t *dst, uint8_t *src,
394                                     ptrdiff_t stride, SAOParams *sao,
395                                     int *borders, int width, int height,
396                                     int c_idx)
397 {
398     FUNC(sao_band_filter)(dst, src, stride, sao, borders,
399                           width, height, c_idx, 2);
400 }
401
402 static void FUNC(sao_band_filter_3)(uint8_t *_dst, uint8_t *_src,
403                                     ptrdiff_t stride, SAOParams *sao,
404                                     int *borders, int width, int height,
405                                     int c_idx)
406 {
407     FUNC(sao_band_filter)(_dst, _src, stride, sao, borders,
408                           width, height, c_idx, 3);
409 }
410
411 static void FUNC(sao_edge_filter_0)(uint8_t *_dst, uint8_t *_src,
412                                     ptrdiff_t stride, SAOParams *sao,
413                                     int *borders, int _width, int _height,
414                                     int c_idx, uint8_t vert_edge,
415                                     uint8_t horiz_edge, uint8_t diag_edge)
416 {
417     int x, y;
418     pixel *dst = (pixel *)_dst;
419     pixel *src = (pixel *)_src;
420     int chroma = !!c_idx;
421     int *sao_offset_val = sao->offset_val[c_idx];
422     int sao_eo_class    = sao->eo_class[c_idx];
423     int init_x = 0, init_y = 0, width = _width, height = _height;
424
425     static const int8_t pos[4][2][2] = {
426         { { -1,  0 }, {  1, 0 } }, // horizontal
427         { {  0, -1 }, {  0, 1 } }, // vertical
428         { { -1, -1 }, {  1, 1 } }, // 45 degree
429         { {  1, -1 }, { -1, 1 } }, // 135 degree
430     };
431     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
432
433 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
434
435     stride /= sizeof(pixel);
436
437     if (!borders[2])
438         width -= (8 >> chroma) + 2;
439     if (!borders[3])
440         height -= (4 >> chroma) + 2;
441
442     dst = dst + (init_y * stride + init_x);
443     src = src + (init_y * stride + init_x);
444     init_y = init_x = 0;
445     if (sao_eo_class != SAO_EO_VERT) {
446         if (borders[0]) {
447             int offset_val = sao_offset_val[0];
448             int y_stride   = 0;
449             for (y = 0; y < height; y++) {
450                 dst[y_stride] = av_clip_pixel(src[y_stride] + offset_val);
451                 y_stride     += stride;
452             }
453             init_x = 1;
454         }
455         if (borders[2]) {
456             int offset_val = sao_offset_val[0];
457             int x_stride   = width - 1;
458             for (x = 0; x < height; x++) {
459                 dst[x_stride] = av_clip_pixel(src[x_stride] + offset_val);
460                 x_stride     += stride;
461             }
462             width--;
463         }
464     }
465     if (sao_eo_class != SAO_EO_HORIZ) {
466         if (borders[1]) {
467             int offset_val = sao_offset_val[0];
468             for (x = init_x; x < width; x++)
469                 dst[x] = av_clip_pixel(src[x] + offset_val);
470             init_y = 1;
471         }
472         if (borders[3]) {
473             int offset_val = sao_offset_val[0];
474             int y_stride   = stride * (height - 1);
475             for (x = init_x; x < width; x++)
476                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + offset_val);
477             height--;
478         }
479     }
480     {
481         int y_stride = init_y * stride;
482         int pos_0_0  = pos[sao_eo_class][0][0];
483         int pos_0_1  = pos[sao_eo_class][0][1];
484         int pos_1_0  = pos[sao_eo_class][1][0];
485         int pos_1_1  = pos[sao_eo_class][1][1];
486
487         int y_stride_0_1 = (init_y + pos_0_1) * stride;
488         int y_stride_1_1 = (init_y + pos_1_1) * stride;
489         for (y = init_y; y < height; y++) {
490             for (x = init_x; x < width; x++) {
491                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
492                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
493                 int offset_val    = edge_idx[2 + diff0 + diff1];
494                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
495             }
496             y_stride     += stride;
497             y_stride_0_1 += stride;
498             y_stride_1_1 += stride;
499         }
500     }
501
502     {
503         // Restore pixels that can't be modified
504         int save_upper_left = !diag_edge && sao_eo_class == SAO_EO_135D && !borders[0] && !borders[1];
505         if (vert_edge && sao_eo_class != SAO_EO_VERT)
506             for (y = init_y+save_upper_left; y< height; y++)
507                 dst[y*stride] = src[y*stride];
508         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
509             for(x = init_x+save_upper_left; x<width; x++)
510                 dst[x] = src[x];
511         if(diag_edge && sao_eo_class == SAO_EO_135D)
512             dst[0] = src[0];
513     }
514
515 #undef CMP
516 }
517
518 static void FUNC(sao_edge_filter_1)(uint8_t *_dst, uint8_t *_src,
519                                     ptrdiff_t stride, SAOParams *sao,
520                                     int *borders, int _width, int _height,
521                                     int c_idx, uint8_t vert_edge,
522                                     uint8_t horiz_edge, uint8_t diag_edge)
523 {
524     int x, y;
525     pixel *dst = (pixel *)_dst;
526     pixel *src = (pixel *)_src;
527     int chroma = !!c_idx;
528     int *sao_offset_val = sao->offset_val[c_idx];
529     int sao_eo_class    = sao->eo_class[c_idx];
530     int init_x = 0, init_y = 0, width = _width, height = _height;
531
532     static const int8_t pos[4][2][2] = {
533         { { -1, 0  }, { 1,  0 } }, // horizontal
534         { { 0,  -1 }, { 0,  1 } }, // vertical
535         { { -1, -1 }, { 1,  1 } }, // 45 degree
536         { { 1,  -1 }, { -1, 1 } }, // 135 degree
537     };
538     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
539
540 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
541
542     stride /= sizeof(pixel);
543
544     init_y = -(4 >> chroma) - 2;
545     if (!borders[2])
546         width -= (8 >> chroma) + 2;
547     height = (4 >> chroma) + 2;
548
549     dst = dst + (init_y * stride + init_x);
550     src = src + (init_y * stride + init_x);
551     init_y = init_x = 0;
552     if (sao_eo_class != SAO_EO_VERT) {
553         if (borders[0]) {
554             int offset_val = sao_offset_val[0];
555             int y_stride   = 0;
556             for (y = 0; y < height; y++) {
557                 dst[y_stride] = av_clip_pixel(src[y_stride] + offset_val);
558                 y_stride     += stride;
559             }
560             init_x = 1;
561         }
562         if (borders[2]) {
563             int offset_val = sao_offset_val[0];
564             int x_stride   = width - 1;
565             for (x = 0; x < height; x++) {
566                 dst[x_stride] = av_clip_pixel(src[x_stride] + offset_val);
567                 x_stride     += stride;
568             }
569             width--;
570         }
571     }
572     {
573         int y_stride = init_y * stride;
574         int pos_0_0  = pos[sao_eo_class][0][0];
575         int pos_0_1  = pos[sao_eo_class][0][1];
576         int pos_1_0  = pos[sao_eo_class][1][0];
577         int pos_1_1  = pos[sao_eo_class][1][1];
578
579         int y_stride_0_1 = (init_y + pos_0_1) * stride;
580         int y_stride_1_1 = (init_y + pos_1_1) * stride;
581         for (y = init_y; y < height; y++) {
582             for (x = init_x; x < width; x++) {
583                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
584                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
585                 int offset_val    = edge_idx[2 + diff0 + diff1];
586                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
587             }
588             y_stride     += stride;
589             y_stride_0_1 += stride;
590             y_stride_1_1 += stride;
591         }
592     }
593
594     {
595         // Restore pixels that can't be modified
596         int save_lower_left = !diag_edge && sao_eo_class == SAO_EO_45D && !borders[0];
597         if(vert_edge && sao_eo_class != SAO_EO_VERT)
598             for(y = init_y; y< height-save_lower_left; y++)
599                 dst[y*stride] = src[y*stride];
600         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
601             for(x = init_x+save_lower_left; x<width; x++)
602                 dst[(height-1)*stride+x] = src[(height-1)*stride+x];
603         if(diag_edge && sao_eo_class == SAO_EO_45D)
604             dst[stride*(height-1)] = src[stride*(height-1)];
605     }
606
607 #undef CMP
608 }
609
610 static void FUNC(sao_edge_filter_2)(uint8_t *_dst, uint8_t *_src,
611                                     ptrdiff_t stride, SAOParams *sao,
612                                     int *borders, int _width, int _height,
613                                     int c_idx, uint8_t vert_edge,
614                                     uint8_t horiz_edge, uint8_t diag_edge)
615 {
616     int x, y;
617     pixel *dst = (pixel *)_dst;
618     pixel *src = (pixel *)_src;
619     int chroma = !!c_idx;
620     int *sao_offset_val = sao->offset_val[c_idx];
621     int sao_eo_class    = sao->eo_class[c_idx];
622     int init_x = 0, init_y = 0, width = _width, height = _height;
623
624     static const int8_t pos[4][2][2] = {
625         { { -1,  0 }, {  1, 0 } }, // horizontal
626         { {  0, -1 }, {  0, 1 } }, // vertical
627         { { -1, -1 }, {  1, 1 } }, // 45 degree
628         { {  1, -1 }, { -1, 1 } }, // 135 degree
629     };
630     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
631
632 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
633
634     stride /= sizeof(pixel);
635
636     init_x = -(8 >> chroma) - 2;
637     width  =  (8 >> chroma) + 2;
638     if (!borders[3])
639         height -= (4 >> chroma) + 2;
640
641     dst = dst + (init_y * stride + init_x);
642     src = src + (init_y * stride + init_x);
643     init_y = init_x = 0;
644     if (sao_eo_class != SAO_EO_HORIZ) {
645         if (borders[1]) {
646             int offset_val = sao_offset_val[0];
647             for (x = init_x; x < width; x++)
648                 dst[x] = av_clip_pixel(src[x] + offset_val);
649             init_y = 1;
650         }
651         if (borders[3]) {
652             int offset_val = sao_offset_val[0];
653             int y_stride   = stride * (height - 1);
654             for (x = init_x; x < width; x++)
655                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + offset_val);
656             height--;
657         }
658     }
659     {
660         int y_stride = init_y * stride;
661         int pos_0_0  = pos[sao_eo_class][0][0];
662         int pos_0_1  = pos[sao_eo_class][0][1];
663         int pos_1_0  = pos[sao_eo_class][1][0];
664         int pos_1_1  = pos[sao_eo_class][1][1];
665
666         int y_stride_0_1 = (init_y + pos_0_1) * stride;
667         int y_stride_1_1 = (init_y + pos_1_1) * stride;
668         for (y = init_y; y < height; y++) {
669             for (x = init_x; x < width; x++) {
670                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
671                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
672                 int offset_val    = edge_idx[2 + diff0 + diff1];
673                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
674             }
675             y_stride     += stride;
676             y_stride_0_1 += stride;
677             y_stride_1_1 += stride;
678         }
679     }
680
681     {
682         // Restore pixels that can't be modified
683         int save_upper_right = !diag_edge && sao_eo_class == SAO_EO_45D && !borders[1];
684         if(vert_edge && sao_eo_class != SAO_EO_VERT)
685             for(y = init_y+save_upper_right; y< height; y++)
686                 dst[y*stride+width-1] = src[y*stride+width-1];
687         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
688             for(x = init_x; x<width-save_upper_right; x++)
689                 dst[x] = src[x];
690         if(diag_edge && sao_eo_class == SAO_EO_45D)
691             dst[width-1] = src[width-1];
692     }
693 #undef CMP
694 }
695
696 static void FUNC(sao_edge_filter_3)(uint8_t *_dst, uint8_t *_src,
697                                     ptrdiff_t stride, SAOParams *sao,
698                                     int *borders, int _width, int _height,
699                                     int c_idx, uint8_t vert_edge,
700                                     uint8_t horiz_edge, uint8_t diag_edge)
701 {
702     int x, y;
703     pixel *dst = (pixel *)_dst;
704     pixel *src = (pixel *)_src;
705     int chroma = !!c_idx;
706     int *sao_offset_val = sao->offset_val[c_idx];
707     int sao_eo_class    = sao->eo_class[c_idx];
708     int init_x = 0, init_y = 0, width = _width, height = _height;
709
710     static const int8_t pos[4][2][2] = {
711         { { -1,  0 }, {  1, 0 } }, // horizontal
712         { {  0, -1 }, {  0, 1 } }, // vertical
713         { { -1, -1 }, {  1, 1 } }, // 45 degree
714         { {  1, -1 }, { -1, 1 } }, // 135 degree
715     };
716     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
717
718 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
719
720     stride /= sizeof(pixel);
721
722     init_y = -(4 >> chroma) - 2;
723     init_x = -(8 >> chroma) - 2;
724     width  =  (8 >> chroma) + 2;
725     height =  (4 >> chroma) + 2;
726
727
728     dst    = dst + (init_y * stride + init_x);
729     src    = src + (init_y * stride + init_x);
730     init_y = init_x = 0;
731
732     {
733         int y_stride = init_y * stride;
734         int pos_0_0  = pos[sao_eo_class][0][0];
735         int pos_0_1  = pos[sao_eo_class][0][1];
736         int pos_1_0  = pos[sao_eo_class][1][0];
737         int pos_1_1  = pos[sao_eo_class][1][1];
738
739         int y_stride_0_1 = (init_y + pos_0_1) * stride;
740         int y_stride_1_1 = (init_y + pos_1_1) * stride;
741
742         for (y = init_y; y < height; y++) {
743             for (x = init_x; x < width; x++) {
744                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
745                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
746                 int offset_val    = edge_idx[2 + diff0 + diff1];
747                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
748             }
749             y_stride     += stride;
750             y_stride_0_1 += stride;
751             y_stride_1_1 += stride;
752         }
753     }
754
755     {
756         // Restore pixels that can't be modified
757         int save_lower_right = !diag_edge && sao_eo_class == SAO_EO_135D;
758         if(vert_edge && sao_eo_class != SAO_EO_VERT)
759             for(y = init_y; y< height-save_lower_right; y++)
760                 dst[y*stride+width-1] = src[y*stride+width-1];
761         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
762             for(x = init_x; x<width-save_lower_right; x++)
763                 dst[(height-1)*stride+x] = src[(height-1)*stride+x];
764         if(diag_edge && sao_eo_class == SAO_EO_135D)
765             dst[stride*(height-1)+width-1] = src[stride*(height-1)+width-1];
766     }
767 #undef CMP
768 }
769
770 #undef SET
771 #undef SCALE
772 #undef ADD_AND_SCALE
773 #undef TR_4
774 #undef TR_8
775 #undef TR_16
776 #undef TR_32
777
778 static av_always_inline void
779 FUNC(put_hevc_qpel_pixels)(int16_t *dst, ptrdiff_t dststride,
780                            uint8_t *_src, ptrdiff_t _srcstride,
781                            int width, int height, int mx, int my,
782                            int16_t* mcbuffer)
783 {
784     int x, y;
785     pixel *src          = (pixel *)_src;
786     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
787
788     for (y = 0; y < height; y++) {
789         for (x = 0; x < width; x++)
790             dst[x] = src[x] << (14 - BIT_DEPTH);
791         src += srcstride;
792         dst += dststride;
793     }
794 }
795
796 #define QPEL_FILTER_1(src, stride)      \
797     (1 * -src[x - 3 * stride] +         \
798      4 *  src[x - 2 * stride] -         \
799     10 *  src[x -     stride] +         \
800     58 *  src[x]              +         \
801     17 *  src[x +     stride] -         \
802      5 *  src[x + 2 * stride] +         \
803      1 *  src[x + 3 * stride])
804
805 #define QPEL_FILTER_2(src, stride)      \
806     (1  * -src[x - 3 * stride] +        \
807      4  *  src[x - 2 * stride] -        \
808     11  *  src[x -     stride] +        \
809     40  *  src[x]              +        \
810     40  *  src[x +     stride] -        \
811     11  *  src[x + 2 * stride] +        \
812      4  *  src[x + 3 * stride] -        \
813      1  *  src[x + 4 * stride])
814
815 #define QPEL_FILTER_3(src, stride)      \
816     (1  * src[x - 2 * stride] -         \
817      5  * src[x -     stride] +         \
818     17  * src[x]              +         \
819     58  * src[x + stride]     -         \
820     10  * src[x + 2 * stride] +         \
821      4  * src[x + 3 * stride] -         \
822      1  * src[x + 4 * stride])
823
824
825 #define PUT_HEVC_QPEL_H(H)                                                     \
826 static void FUNC(put_hevc_qpel_h ## H)(int16_t *dst,  ptrdiff_t dststride,     \
827                                        uint8_t *_src, ptrdiff_t _srcstride,    \
828                                        int width, int height,                  \
829                                        int16_t* mcbuffer)                      \
830 {                                                                              \
831     int x, y;                                                                  \
832     pixel *src = (pixel*)_src;                                                 \
833     ptrdiff_t srcstride = _srcstride / sizeof(pixel);                          \
834                                                                                \
835     for (y = 0; y < height; y++) {                                             \
836         for (x = 0; x < width; x++)                                            \
837             dst[x] = QPEL_FILTER_ ## H(src, 1) >> (BIT_DEPTH - 8);             \
838         src += srcstride;                                                      \
839         dst += dststride;                                                      \
840     }                                                                          \
841 }
842
843 #define PUT_HEVC_QPEL_V(V)                                                     \
844 static void FUNC(put_hevc_qpel_v ## V)(int16_t *dst,  ptrdiff_t dststride,     \
845                                        uint8_t *_src, ptrdiff_t _srcstride,    \
846                                        int width, int height,                  \
847                                        int16_t* mcbuffer)                      \
848 {                                                                              \
849     int x, y;                                                                  \
850     pixel *src = (pixel*)_src;                                                 \
851     ptrdiff_t srcstride = _srcstride / sizeof(pixel);                          \
852                                                                                \
853     for (y = 0; y < height; y++)  {                                            \
854         for (x = 0; x < width; x++)                                            \
855             dst[x] = QPEL_FILTER_ ## V(src, srcstride) >> (BIT_DEPTH - 8);     \
856         src += srcstride;                                                      \
857         dst += dststride;                                                      \
858     }                                                                          \
859 }
860
861 #define PUT_HEVC_QPEL_HV(H, V)                                                 \
862 static void FUNC(put_hevc_qpel_h ## H ## v ## V)(int16_t *dst,                 \
863                                                  ptrdiff_t dststride,          \
864                                                  uint8_t *_src,                \
865                                                  ptrdiff_t _srcstride,         \
866                                                  int width, int height,        \
867                                                  int16_t* mcbuffer)            \
868 {                                                                              \
869     int x, y;                                                                  \
870     pixel *src = (pixel*)_src;                                                 \
871     ptrdiff_t srcstride = _srcstride / sizeof(pixel);                          \
872                                                                                \
873     int16_t tmp_array[(MAX_PB_SIZE + 7) * MAX_PB_SIZE];                        \
874     int16_t *tmp = tmp_array;                                                  \
875                                                                                \
876     src -= ff_hevc_qpel_extra_before[V] * srcstride;                           \
877                                                                                \
878     for (y = 0; y < height + ff_hevc_qpel_extra[V]; y++) {                     \
879         for (x = 0; x < width; x++)                                            \
880             tmp[x] = QPEL_FILTER_ ## H(src, 1) >> (BIT_DEPTH - 8);             \
881         src += srcstride;                                                      \
882         tmp += MAX_PB_SIZE;                                                    \
883     }                                                                          \
884                                                                                \
885     tmp = tmp_array + ff_hevc_qpel_extra_before[V] * MAX_PB_SIZE;              \
886                                                                                \
887     for (y = 0; y < height; y++) {                                             \
888         for (x = 0; x < width; x++)                                            \
889             dst[x] = QPEL_FILTER_ ## V(tmp, MAX_PB_SIZE) >> 6;                 \
890         tmp += MAX_PB_SIZE;                                                    \
891         dst += dststride;                                                      \
892     }                                                                          \
893 }
894
895 PUT_HEVC_QPEL_H(1)
896 PUT_HEVC_QPEL_H(2)
897 PUT_HEVC_QPEL_H(3)
898 PUT_HEVC_QPEL_V(1)
899 PUT_HEVC_QPEL_V(2)
900 PUT_HEVC_QPEL_V(3)
901 PUT_HEVC_QPEL_HV(1, 1)
902 PUT_HEVC_QPEL_HV(1, 2)
903 PUT_HEVC_QPEL_HV(1, 3)
904 PUT_HEVC_QPEL_HV(2, 1)
905 PUT_HEVC_QPEL_HV(2, 2)
906 PUT_HEVC_QPEL_HV(2, 3)
907 PUT_HEVC_QPEL_HV(3, 1)
908 PUT_HEVC_QPEL_HV(3, 2)
909 PUT_HEVC_QPEL_HV(3, 3)
910
911 #define QPEL(W)                                                                             \
912 static void FUNC(put_hevc_qpel_pixels_ ## W)(int16_t *dst, ptrdiff_t dststride,             \
913                                              uint8_t *src, ptrdiff_t srcstride,             \
914                                              int height, int mx, int my,                    \
915                                              int16_t *mcbuffer)                             \
916 {                                                                                           \
917     FUNC(put_hevc_qpel_pixels)(dst, dststride, src, srcstride, W, height,                   \
918                                mx, my, mcbuffer);                                           \
919 }                                                                                           \
920                                                                                             \
921 static void FUNC(put_hevc_qpel_h_ ## W)(int16_t *dst, ptrdiff_t dststride,                  \
922                                         uint8_t *src, ptrdiff_t srcstride,                  \
923                                         int height, int mx, int my,                         \
924                                         int16_t *mcbuffer)                                  \
925 {                                                                                           \
926     if (mx == 1)                                                                            \
927         FUNC(put_hevc_qpel_h1)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
928     else if (mx == 2)                                                                       \
929         FUNC(put_hevc_qpel_h2)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
930     else                                                                                    \
931         FUNC(put_hevc_qpel_h3)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
932 }                                                                                           \
933                                                                                             \
934 static void FUNC(put_hevc_qpel_v_ ## W)(int16_t *dst, ptrdiff_t dststride,                  \
935                                              uint8_t *src, ptrdiff_t srcstride,             \
936                                              int height, int mx, int my,                    \
937                                              int16_t *mcbuffer)                             \
938 {                                                                                           \
939     if (my == 1)                                                                            \
940         FUNC(put_hevc_qpel_v1)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
941     else if (my == 2)                                                                       \
942         FUNC(put_hevc_qpel_v2)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
943     else                                                                                    \
944         FUNC(put_hevc_qpel_v3)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
945 }                                                                                           \
946                                                                                             \
947 static void FUNC(put_hevc_qpel_hv_ ## W)(int16_t *dst, ptrdiff_t dststride,                 \
948                                              uint8_t *src, ptrdiff_t srcstride,             \
949                                              int height, int mx, int my,                    \
950                                              int16_t *mcbuffer)                             \
951 {                                                                                           \
952     if (my == 1) {                                                                          \
953         if (mx == 1)                                                                        \
954             FUNC(put_hevc_qpel_h1v1)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
955         else if (mx == 2)                                                                   \
956             FUNC(put_hevc_qpel_h2v1)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
957         else                                                                                \
958             FUNC(put_hevc_qpel_h3v1)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
959     } else if (my == 2) {                                                                   \
960         if (mx == 1)                                                                        \
961             FUNC(put_hevc_qpel_h1v2)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
962         else if (mx == 2)                                                                   \
963             FUNC(put_hevc_qpel_h2v2)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
964         else                                                                                \
965             FUNC(put_hevc_qpel_h3v2)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
966     } else {                                                                                \
967         if (mx == 1)                                                                        \
968             FUNC(put_hevc_qpel_h1v3)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
969         else if (mx == 2)                                                                   \
970             FUNC(put_hevc_qpel_h2v3)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
971         else                                                                                \
972             FUNC(put_hevc_qpel_h3v3)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
973     }                                                                                       \
974 }
975
976 QPEL(64)
977 QPEL(48)
978 QPEL(32)
979 QPEL(24)
980 QPEL(16)
981 QPEL(12)
982 QPEL(8)
983 QPEL(4)
984
985 static inline void FUNC(put_hevc_epel_pixels)(int16_t *dst, ptrdiff_t dststride,
986                                               uint8_t *_src, ptrdiff_t _srcstride,
987                                               int width, int height, int mx, int my,
988                                               int16_t* mcbuffer)
989 {
990     int x, y;
991     pixel *src          = (pixel *)_src;
992     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
993
994     for (y = 0; y < height; y++) {
995         for (x = 0; x < width; x++)
996             dst[x] = src[x] << (14 - BIT_DEPTH);
997         src += srcstride;
998         dst += dststride;
999     }
1000 }
1001
1002 #define EPEL_FILTER(src, stride)                \
1003     (filter_0 * src[x - stride] +               \
1004      filter_1 * src[x]          +               \
1005      filter_2 * src[x + stride] +               \
1006      filter_3 * src[x + 2 * stride])
1007
1008 static inline void FUNC(put_hevc_epel_h)(int16_t *dst, ptrdiff_t dststride,
1009                                          uint8_t *_src, ptrdiff_t _srcstride,
1010                                          int width, int height, int mx, int my,
1011                                          int16_t* mcbuffer)
1012 {
1013     int x, y;
1014     pixel *src = (pixel *)_src;
1015     ptrdiff_t srcstride  = _srcstride / sizeof(pixel);
1016     const int8_t *filter = ff_hevc_epel_filters[mx - 1];
1017     int8_t filter_0 = filter[0];
1018     int8_t filter_1 = filter[1];
1019     int8_t filter_2 = filter[2];
1020     int8_t filter_3 = filter[3];
1021     for (y = 0; y < height; y++) {
1022         for (x = 0; x < width; x++)
1023             dst[x] = EPEL_FILTER(src, 1) >> (BIT_DEPTH - 8);
1024         src += srcstride;
1025         dst += dststride;
1026     }
1027 }
1028
1029 static inline void FUNC(put_hevc_epel_v)(int16_t *dst, ptrdiff_t dststride,
1030                                          uint8_t *_src, ptrdiff_t _srcstride,
1031                                          int width, int height, int mx, int my,
1032                                          int16_t* mcbuffer)
1033 {
1034     int x, y;
1035     pixel *src = (pixel *)_src;
1036     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
1037     const int8_t *filter = ff_hevc_epel_filters[my - 1];
1038     int8_t filter_0 = filter[0];
1039     int8_t filter_1 = filter[1];
1040     int8_t filter_2 = filter[2];
1041     int8_t filter_3 = filter[3];
1042
1043     for (y = 0; y < height; y++) {
1044         for (x = 0; x < width; x++)
1045             dst[x] = EPEL_FILTER(src, srcstride) >> (BIT_DEPTH - 8);
1046         src += srcstride;
1047         dst += dststride;
1048     }
1049 }
1050
1051 static inline void FUNC(put_hevc_epel_hv)(int16_t *dst, ptrdiff_t dststride,
1052                                           uint8_t *_src, ptrdiff_t _srcstride,
1053                                           int width, int height, int mx, int my,
1054                                           int16_t* mcbuffer)
1055 {
1056     int x, y;
1057     pixel *src = (pixel *)_src;
1058     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
1059     const int8_t *filter_h = ff_hevc_epel_filters[mx - 1];
1060     const int8_t *filter_v = ff_hevc_epel_filters[my - 1];
1061     int8_t filter_0 = filter_h[0];
1062     int8_t filter_1 = filter_h[1];
1063     int8_t filter_2 = filter_h[2];
1064     int8_t filter_3 = filter_h[3];
1065     int16_t tmp_array[(MAX_PB_SIZE + 3) * MAX_PB_SIZE];
1066     int16_t *tmp = tmp_array;
1067
1068     src -= EPEL_EXTRA_BEFORE * srcstride;
1069
1070     for (y = 0; y < height + EPEL_EXTRA; y++) {
1071         for (x = 0; x < width; x++)
1072             tmp[x] = EPEL_FILTER(src, 1) >> (BIT_DEPTH - 8);
1073         src += srcstride;
1074         tmp += MAX_PB_SIZE;
1075     }
1076
1077     tmp      = tmp_array + EPEL_EXTRA_BEFORE * MAX_PB_SIZE;
1078     filter_0 = filter_v[0];
1079     filter_1 = filter_v[1];
1080     filter_2 = filter_v[2];
1081     filter_3 = filter_v[3];
1082     for (y = 0; y < height; y++) {
1083         for (x = 0; x < width; x++)
1084             dst[x] = EPEL_FILTER(tmp, MAX_PB_SIZE) >> 6;
1085         tmp += MAX_PB_SIZE;
1086         dst += dststride;
1087     }
1088 }
1089
1090 #define EPEL(W)                                                                 \
1091 static void FUNC(put_hevc_epel_pixels_ ## W)(int16_t *dst, ptrdiff_t dststride, \
1092                                              uint8_t *src, ptrdiff_t srcstride, \
1093                                              int height, int mx, int my,        \
1094                                              int16_t *mcbuffer)                 \
1095 {                                                                               \
1096     FUNC(put_hevc_epel_pixels)(dst, dststride, src, srcstride,                  \
1097                                W, height, mx, my, mcbuffer);                    \
1098 }                                                                               \
1099 static void FUNC(put_hevc_epel_h_ ## W)(int16_t *dst, ptrdiff_t dststride,      \
1100                                         uint8_t *src, ptrdiff_t srcstride,      \
1101                                         int height, int mx, int my,             \
1102                                         int16_t *mcbuffer)                      \
1103 {                                                                               \
1104     FUNC(put_hevc_epel_h)(dst, dststride, src, srcstride,                       \
1105                           W, height, mx, my, mcbuffer);                         \
1106 }                                                                               \
1107 static void FUNC(put_hevc_epel_v_ ## W)(int16_t *dst, ptrdiff_t dststride,      \
1108                                         uint8_t *src, ptrdiff_t srcstride,      \
1109                                         int height, int mx, int my,             \
1110                                         int16_t *mcbuffer)                      \
1111 {                                                                               \
1112     FUNC(put_hevc_epel_v)(dst, dststride, src, srcstride,                       \
1113                           W, height, mx, my, mcbuffer);                         \
1114 }                                                                               \
1115 static void FUNC(put_hevc_epel_hv_ ## W)(int16_t *dst, ptrdiff_t dststride,     \
1116                                          uint8_t *src, ptrdiff_t srcstride,     \
1117                                          int height, int mx, int my,            \
1118                                          int16_t *mcbuffer)                     \
1119 {                                                                               \
1120     FUNC(put_hevc_epel_hv)(dst, dststride, src, srcstride,                      \
1121                            W, height, mx, my, mcbuffer);                        \
1122 }
1123
1124 EPEL(32)
1125 EPEL(24)
1126 EPEL(16)
1127 EPEL(12)
1128 EPEL(8)
1129 EPEL(6)
1130 EPEL(4)
1131 EPEL(2)
1132
1133 static av_always_inline void
1134 FUNC(put_unweighted_pred)(uint8_t *_dst, ptrdiff_t _dststride,
1135                           int16_t *src, ptrdiff_t srcstride,
1136                           int width, int height)
1137 {
1138     int x, y;
1139     pixel *dst          = (pixel *)_dst;
1140     ptrdiff_t dststride = _dststride / sizeof(pixel);
1141
1142     int shift = 14 - BIT_DEPTH;
1143 #if BIT_DEPTH < 14
1144     int offset = 1 << (shift - 1);
1145 #else
1146     int offset = 0;
1147 #endif
1148     for (y = 0; y < height; y++) {
1149         for (x = 0; x < width; x++)
1150             dst[x] = av_clip_pixel((src[x] + offset) >> shift);
1151         dst += dststride;
1152         src += srcstride;
1153     }
1154 }
1155
1156 static av_always_inline void
1157 FUNC(put_unweighted_pred_avg)(uint8_t *_dst, ptrdiff_t _dststride,
1158                               int16_t *src1, int16_t *src2,
1159                               ptrdiff_t srcstride,
1160                               int width, int height)
1161 {
1162     int x, y;
1163     pixel *dst          = (pixel *)_dst;
1164     ptrdiff_t dststride = _dststride / sizeof(pixel);
1165
1166     int shift = 14 + 1 - BIT_DEPTH;
1167 #if BIT_DEPTH < 14
1168     int offset = 1 << (shift - 1);
1169 #else
1170     int offset = 0;
1171 #endif
1172
1173     for (y = 0; y < height; y++) {
1174         for (x = 0; x < width; x++)
1175             dst[x] = av_clip_pixel((src1[x] + src2[x] + offset) >> shift);
1176         dst  += dststride;
1177         src1 += srcstride;
1178         src2 += srcstride;
1179     }
1180 }
1181
1182 static av_always_inline void
1183 FUNC(weighted_pred)(uint8_t denom, int16_t wlxFlag, int16_t olxFlag,
1184                     uint8_t *_dst, ptrdiff_t _dststride,
1185                     int16_t *src, ptrdiff_t srcstride,
1186                     int width, int height)
1187 {
1188     int shift, log2Wd, wx, ox, x, y, offset;
1189     pixel *dst          = (pixel *)_dst;
1190     ptrdiff_t dststride = _dststride / sizeof(pixel);
1191
1192     shift  = 14 - BIT_DEPTH;
1193     log2Wd = denom + shift;
1194     offset = 1 << (log2Wd - 1);
1195     wx     = wlxFlag;
1196     ox     = olxFlag * (1 << (BIT_DEPTH - 8));
1197
1198     for (y = 0; y < height; y++) {
1199         for (x = 0; x < width; x++) {
1200             if (log2Wd >= 1) {
1201                 dst[x] = av_clip_pixel(((src[x] * wx + offset) >> log2Wd) + ox);
1202             } else {
1203                 dst[x] = av_clip_pixel(src[x] * wx + ox);
1204             }
1205         }
1206         dst += dststride;
1207         src += srcstride;
1208     }
1209 }
1210
1211 static av_always_inline void
1212 FUNC(weighted_pred_avg)(uint8_t denom,
1213                         int16_t wl0Flag, int16_t wl1Flag,
1214                         int16_t ol0Flag, int16_t ol1Flag,
1215                         uint8_t *_dst, ptrdiff_t _dststride,
1216                         int16_t *src1, int16_t *src2,
1217                         ptrdiff_t srcstride,
1218                         int width, int height)
1219 {
1220     int shift, log2Wd, w0, w1, o0, o1, x, y;
1221     pixel *dst = (pixel *)_dst;
1222     ptrdiff_t dststride = _dststride / sizeof(pixel);
1223
1224     shift  = 14 - BIT_DEPTH;
1225     log2Wd = denom + shift;
1226     w0     = wl0Flag;
1227     w1     = wl1Flag;
1228     o0     = ol0Flag * (1 << (BIT_DEPTH - 8));
1229     o1     = ol1Flag * (1 << (BIT_DEPTH - 8));
1230
1231     for (y = 0; y < height; y++) {
1232         for (x = 0; x < width; x++)
1233             dst[x] = av_clip_pixel((src1[x] * w0 + src2[x] * w1 +
1234                                     ((o0 + o1 + 1) << log2Wd)) >> (log2Wd + 1));
1235         dst  += dststride;
1236         src1 += srcstride;
1237         src2 += srcstride;
1238     }
1239 }
1240
1241 #define PUT_PRED(w)                                                                            \
1242 static void FUNC(put_unweighted_pred_ ## w)(uint8_t *dst, ptrdiff_t dststride,                 \
1243                                             int16_t *src, ptrdiff_t srcstride,                 \
1244                                             int height)                                        \
1245 {                                                                                              \
1246     FUNC(put_unweighted_pred)(dst, dststride, src, srcstride, w, height);                      \
1247 }                                                                                              \
1248 static void FUNC(put_unweighted_pred_avg_ ## w)(uint8_t *dst, ptrdiff_t dststride,             \
1249                                                 int16_t *src1, int16_t *src2,                  \
1250                                                 ptrdiff_t srcstride, int height)               \
1251 {                                                                                              \
1252     FUNC(put_unweighted_pred_avg)(dst, dststride, src1, src2, srcstride, w, height);           \
1253 }                                                                                              \
1254 static void FUNC(put_weighted_pred_ ## w)(uint8_t denom, int16_t weight, int16_t offset,       \
1255                                           uint8_t *dst, ptrdiff_t dststride,                   \
1256                                           int16_t *src, ptrdiff_t srcstride, int height)       \
1257 {                                                                                              \
1258     FUNC(weighted_pred)(denom, weight, offset,                                                 \
1259                         dst, dststride, src, srcstride, w, height);                            \
1260 }                                                                                              \
1261 static void FUNC(put_weighted_pred_avg_ ## w)(uint8_t denom, int16_t weight0, int16_t weight1, \
1262                                               int16_t offset0, int16_t offset1,                \
1263                                               uint8_t *dst, ptrdiff_t dststride,               \
1264                                               int16_t *src1, int16_t *src2,                    \
1265                                               ptrdiff_t srcstride, int height)                 \
1266 {                                                                                              \
1267     FUNC(weighted_pred_avg)(denom, weight0, weight1, offset0, offset1,                         \
1268                             dst, dststride, src1, src2, srcstride, w, height);                 \
1269 }
1270
1271 PUT_PRED(64)
1272 PUT_PRED(48)
1273 PUT_PRED(32)
1274 PUT_PRED(24)
1275 PUT_PRED(16)
1276 PUT_PRED(12)
1277 PUT_PRED(8)
1278 PUT_PRED(6)
1279 PUT_PRED(4)
1280 PUT_PRED(2)
1281
1282 // line zero
1283 #define P3 pix[-4 * xstride]
1284 #define P2 pix[-3 * xstride]
1285 #define P1 pix[-2 * xstride]
1286 #define P0 pix[-1 * xstride]
1287 #define Q0 pix[0 * xstride]
1288 #define Q1 pix[1 * xstride]
1289 #define Q2 pix[2 * xstride]
1290 #define Q3 pix[3 * xstride]
1291
1292 // line three. used only for deblocking decision
1293 #define TP3 pix[-4 * xstride + 3 * ystride]
1294 #define TP2 pix[-3 * xstride + 3 * ystride]
1295 #define TP1 pix[-2 * xstride + 3 * ystride]
1296 #define TP0 pix[-1 * xstride + 3 * ystride]
1297 #define TQ0 pix[0  * xstride + 3 * ystride]
1298 #define TQ1 pix[1  * xstride + 3 * ystride]
1299 #define TQ2 pix[2  * xstride + 3 * ystride]
1300 #define TQ3 pix[3  * xstride + 3 * ystride]
1301
1302 static void FUNC(hevc_loop_filter_luma)(uint8_t *_pix,
1303                                         ptrdiff_t _xstride, ptrdiff_t _ystride,
1304                                         int beta, int *_tc,
1305                                         uint8_t *_no_p, uint8_t *_no_q)
1306 {
1307     int d, j;
1308     pixel *pix        = (pixel *)_pix;
1309     ptrdiff_t xstride = _xstride / sizeof(pixel);
1310     ptrdiff_t ystride = _ystride / sizeof(pixel);
1311
1312     beta <<= BIT_DEPTH - 8;
1313
1314     for (j = 0; j < 2; j++) {
1315         const int dp0  = abs(P2  - 2 * P1  + P0);
1316         const int dq0  = abs(Q2  - 2 * Q1  + Q0);
1317         const int dp3  = abs(TP2 - 2 * TP1 + TP0);
1318         const int dq3  = abs(TQ2 - 2 * TQ1 + TQ0);
1319         const int d0   = dp0 + dq0;
1320         const int d3   = dp3 + dq3;
1321         const int tc   = _tc[j]   << (BIT_DEPTH - 8);
1322         const int no_p = _no_p[j];
1323         const int no_q = _no_q[j];
1324
1325         if (d0 + d3 >= beta) {
1326             pix += 4 * ystride;
1327             continue;
1328         } else {
1329             const int beta_3 = beta >> 3;
1330             const int beta_2 = beta >> 2;
1331             const int tc25   = ((tc * 5 + 1) >> 1);
1332
1333             if (abs(P3  -  P0) + abs(Q3  -  Q0) < beta_3 && abs(P0  -  Q0) < tc25 &&
1334                 abs(TP3 - TP0) + abs(TQ3 - TQ0) < beta_3 && abs(TP0 - TQ0) < tc25 &&
1335                                       (d0 << 1) < beta_2 &&      (d3 << 1) < beta_2) {
1336                 // strong filtering
1337                 const int tc2 = tc << 1;
1338                 for (d = 0; d < 4; d++) {
1339                     const int p3 = P3;
1340                     const int p2 = P2;
1341                     const int p1 = P1;
1342                     const int p0 = P0;
1343                     const int q0 = Q0;
1344                     const int q1 = Q1;
1345                     const int q2 = Q2;
1346                     const int q3 = Q3;
1347                     if (!no_p) {
1348                         P0 = p0 + av_clip(((p2 + 2 * p1 + 2 * p0 + 2 * q0 + q1 + 4) >> 3) - p0, -tc2, tc2);
1349                         P1 = p1 + av_clip(((p2 + p1 + p0 + q0 + 2) >> 2) - p1, -tc2, tc2);
1350                         P2 = p2 + av_clip(((2 * p3 + 3 * p2 + p1 + p0 + q0 + 4) >> 3) - p2, -tc2, tc2);
1351                     }
1352                     if (!no_q) {
1353                         Q0 = q0 + av_clip(((p1 + 2 * p0 + 2 * q0 + 2 * q1 + q2 + 4) >> 3) - q0, -tc2, tc2);
1354                         Q1 = q1 + av_clip(((p0 + q0 + q1 + q2 + 2) >> 2) - q1, -tc2, tc2);
1355                         Q2 = q2 + av_clip(((2 * q3 + 3 * q2 + q1 + q0 + p0 + 4) >> 3) - q2, -tc2, tc2);
1356                     }
1357                     pix += ystride;
1358                 }
1359             } else { // normal filtering
1360                 int nd_p = 1;
1361                 int nd_q = 1;
1362                 const int tc_2 = tc >> 1;
1363                 if (dp0 + dp3 < ((beta + (beta >> 1)) >> 3))
1364                     nd_p = 2;
1365                 if (dq0 + dq3 < ((beta + (beta >> 1)) >> 3))
1366                     nd_q = 2;
1367
1368                 for (d = 0; d < 4; d++) {
1369                     const int p2 = P2;
1370                     const int p1 = P1;
1371                     const int p0 = P0;
1372                     const int q0 = Q0;
1373                     const int q1 = Q1;
1374                     const int q2 = Q2;
1375                     int delta0   = (9 * (q0 - p0) - 3 * (q1 - p1) + 8) >> 4;
1376                     if (abs(delta0) < 10 * tc) {
1377                         delta0 = av_clip(delta0, -tc, tc);
1378                         if (!no_p)
1379                             P0 = av_clip_pixel(p0 + delta0);
1380                         if (!no_q)
1381                             Q0 = av_clip_pixel(q0 - delta0);
1382                         if (!no_p && nd_p > 1) {
1383                             const int deltap1 = av_clip((((p2 + p0 + 1) >> 1) - p1 + delta0) >> 1, -tc_2, tc_2);
1384                             P1 = av_clip_pixel(p1 + deltap1);
1385                         }
1386                         if (!no_q && nd_q > 1) {
1387                             const int deltaq1 = av_clip((((q2 + q0 + 1) >> 1) - q1 - delta0) >> 1, -tc_2, tc_2);
1388                             Q1 = av_clip_pixel(q1 + deltaq1);
1389                         }
1390                     }
1391                     pix += ystride;
1392                 }
1393             }
1394         }
1395     }
1396 }
1397
1398 static void FUNC(hevc_loop_filter_chroma)(uint8_t *_pix, ptrdiff_t _xstride,
1399                                           ptrdiff_t _ystride, int *_tc,
1400                                           uint8_t *_no_p, uint8_t *_no_q)
1401 {
1402     int d, j, no_p, no_q;
1403     pixel *pix        = (pixel *)_pix;
1404     ptrdiff_t xstride = _xstride / sizeof(pixel);
1405     ptrdiff_t ystride = _ystride / sizeof(pixel);
1406
1407     for (j = 0; j < 2; j++) {
1408         const int tc = _tc[j] << (BIT_DEPTH - 8);
1409         if (tc <= 0) {
1410             pix += 4 * ystride;
1411             continue;
1412         }
1413         no_p = _no_p[j];
1414         no_q = _no_q[j];
1415
1416         for (d = 0; d < 4; d++) {
1417             int delta0;
1418             const int p1 = P1;
1419             const int p0 = P0;
1420             const int q0 = Q0;
1421             const int q1 = Q1;
1422             delta0 = av_clip((((q0 - p0) * 4) + p1 - q1 + 4) >> 3, -tc, tc);
1423             if (!no_p)
1424                 P0 = av_clip_pixel(p0 + delta0);
1425             if (!no_q)
1426                 Q0 = av_clip_pixel(q0 - delta0);
1427             pix += ystride;
1428         }
1429     }
1430 }
1431
1432 static void FUNC(hevc_h_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
1433                                             int *tc, uint8_t *no_p,
1434                                             uint8_t *no_q)
1435 {
1436     FUNC(hevc_loop_filter_chroma)(pix, stride, sizeof(pixel), tc, no_p, no_q);
1437 }
1438
1439 static void FUNC(hevc_v_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
1440                                             int *tc, uint8_t *no_p,
1441                                             uint8_t *no_q)
1442 {
1443     FUNC(hevc_loop_filter_chroma)(pix, sizeof(pixel), stride, tc, no_p, no_q);
1444 }
1445
1446 static void FUNC(hevc_h_loop_filter_luma)(uint8_t *pix, ptrdiff_t stride,
1447                                           int beta, int *tc, uint8_t *no_p,
1448                                           uint8_t *no_q)
1449 {
1450     FUNC(hevc_loop_filter_luma)(pix, stride, sizeof(pixel),
1451                                 beta, tc, no_p, no_q);
1452 }
1453
1454 static void FUNC(hevc_v_loop_filter_luma)(uint8_t *pix, ptrdiff_t stride,
1455                                           int beta, int *tc, uint8_t *no_p,
1456                                           uint8_t *no_q)
1457 {
1458     FUNC(hevc_loop_filter_luma)(pix, sizeof(pixel), stride,
1459                                 beta, tc, no_p, no_q);
1460 }
1461
1462 #undef P3
1463 #undef P2
1464 #undef P1
1465 #undef P0
1466 #undef Q0
1467 #undef Q1
1468 #undef Q2
1469 #undef Q3
1470
1471 #undef TP3
1472 #undef TP2
1473 #undef TP1
1474 #undef TP0
1475 #undef TQ0
1476 #undef TQ1
1477 #undef TQ2
1478 #undef TQ3