1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... loop filter
3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * H.264 / AVC / MPEG-4 part10 loop filter.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27 
28 #include "libavutil/internal.h"
29 #include "libavutil/intreadwrite.h"
30 #include "libavutil/mem_internal.h"
31 #include "avcodec.h"
32 #include "h264dec.h"
33 #include "h264_ps.h"
34 #include "mpegutils.h"
35 
36 /* Deblocking filter (p153) */
37 static const uint8_t alpha_table[52*3] = {
38      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
39      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
40      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
41      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
42      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
43      0,  0,  0,  0,  0,  0,  4,  4,  5,  6,
44      7,  8,  9, 10, 12, 13, 15, 17, 20, 22,
45     25, 28, 32, 36, 40, 45, 50, 56, 63, 71,
46     80, 90,101,113,127,144,162,182,203,226,
47    255,255,
48    255,255,255,255,255,255,255,255,255,255,255,255,255,
49    255,255,255,255,255,255,255,255,255,255,255,255,255,
50    255,255,255,255,255,255,255,255,255,255,255,255,255,
51    255,255,255,255,255,255,255,255,255,255,255,255,255,
52 };
53 static const uint8_t beta_table[52*3] = {
54      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
55      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
56      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
57      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
58      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
59      0,  0,  0,  0,  0,  0,  2,  2,  2,  3,
60      3,  3,  3,  4,  4,  4,  6,  6,  7,  7,
61      8,  8,  9,  9, 10, 10, 11, 11, 12, 12,
62     13, 13, 14, 14, 15, 15, 16, 16, 17, 17,
63     18, 18,
64     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
65     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
66     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
67     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
68 };
69 static const uint8_t tc0_table[52*3][4] = {
70     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
71     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
72     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
73     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
74     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
75     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
76     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
77     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
78     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
79     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
80     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
81     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 1 },
82     {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },
83     {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },
84     {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },
85     {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },
86     {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },
87     {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-1,13,17,25 },
88     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
89     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
90     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
91     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
92     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
93     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
94     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
95     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
96     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
97 };
98 
99 /* intra: 0 if this loopfilter call is guaranteed to be inter (bS < 4), 1 if it might be intra (bS == 4) */
filter_mb_edgev(uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, const H264Context *h, int intra)100 static av_always_inline void filter_mb_edgev(uint8_t *pix, int stride,
101                                              const int16_t bS[4],
102                                              unsigned int qp, int a, int b,
103                                              const H264Context *h, int intra)
104 {
105     const unsigned int index_a = qp + a;
106     const int alpha = alpha_table[index_a];
107     const int beta  = beta_table[qp + b];
108     if (alpha ==0 || beta == 0) return;
109 
110     if( bS[0] < 4 || !intra ) {
111         int8_t tc[4];
112         tc[0] = tc0_table[index_a][bS[0]];
113         tc[1] = tc0_table[index_a][bS[1]];
114         tc[2] = tc0_table[index_a][bS[2]];
115         tc[3] = tc0_table[index_a][bS[3]];
116         h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
117     } else {
118         h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
119     }
120 }
121 
filter_mb_edgecv(uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, const H264Context *h, int intra)122 static av_always_inline void filter_mb_edgecv(uint8_t *pix, int stride,
123                                               const int16_t bS[4],
124                                               unsigned int qp, int a, int b,
125                                               const H264Context *h, int intra)
126 {
127     const unsigned int index_a = qp + a;
128     const int alpha = alpha_table[index_a];
129     const int beta  = beta_table[qp + b];
130     if (alpha ==0 || beta == 0) return;
131 
132     if( bS[0] < 4 || !intra ) {
133         int8_t tc[4];
134         tc[0] = tc0_table[index_a][bS[0]]+1;
135         tc[1] = tc0_table[index_a][bS[1]]+1;
136         tc[2] = tc0_table[index_a][bS[2]]+1;
137         tc[3] = tc0_table[index_a][bS[3]]+1;
138         h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
139     } else {
140         h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
141     }
142 }
143 
filter_mb_mbaff_edgev(const H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp, int a, int b, int intra)144 static av_always_inline void filter_mb_mbaff_edgev(const H264Context *h, uint8_t *pix,
145                                                    int stride,
146                                                    const int16_t bS[7], int bsi,
147                                                    int qp, int a, int b,
148                                                    int intra)
149 {
150     const unsigned int index_a = qp + a;
151     const int alpha = alpha_table[index_a];
152     const int beta  = beta_table[qp + b];
153     if (alpha ==0 || beta == 0) return;
154 
155     if( bS[0] < 4 || !intra ) {
156         int8_t tc[4];
157         tc[0] = tc0_table[index_a][bS[0*bsi]];
158         tc[1] = tc0_table[index_a][bS[1*bsi]];
159         tc[2] = tc0_table[index_a][bS[2*bsi]];
160         tc[3] = tc0_table[index_a][bS[3*bsi]];
161         h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
162     } else {
163         h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
164     }
165 }
166 
filter_mb_mbaff_edgecv(const H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp, int a, int b, int intra)167 static av_always_inline void filter_mb_mbaff_edgecv(const H264Context *h,
168                                                     uint8_t *pix, int stride,
169                                                     const int16_t bS[7],
170                                                     int bsi, int qp, int a,
171                                                     int b, int intra)
172 {
173     const unsigned int index_a = qp + a;
174     const int alpha = alpha_table[index_a];
175     const int beta  = beta_table[qp + b];
176     if (alpha ==0 || beta == 0) return;
177 
178     if( bS[0] < 4 || !intra ) {
179         int8_t tc[4];
180         tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
181         tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
182         tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
183         tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
184         h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
185     } else {
186         h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
187     }
188 }
189 
filter_mb_edgeh(uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, const H264Context *h, int intra)190 static av_always_inline void filter_mb_edgeh(uint8_t *pix, int stride,
191                                              const int16_t bS[4],
192                                              unsigned int qp, int a, int b,
193                                              const H264Context *h, int intra)
194 {
195     const unsigned int index_a = qp + a;
196     const int alpha = alpha_table[index_a];
197     const int beta  = beta_table[qp + b];
198     if (alpha ==0 || beta == 0) return;
199 
200     if( bS[0] < 4 || !intra ) {
201         int8_t tc[4];
202         tc[0] = tc0_table[index_a][bS[0]];
203         tc[1] = tc0_table[index_a][bS[1]];
204         tc[2] = tc0_table[index_a][bS[2]];
205         tc[3] = tc0_table[index_a][bS[3]];
206         h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
207     } else {
208         h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
209     }
210 }
211 
filter_mb_edgech(uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, const H264Context *h, int intra)212 static av_always_inline void filter_mb_edgech(uint8_t *pix, int stride,
213                                               const int16_t bS[4],
214                                               unsigned int qp, int a, int b,
215                                               const H264Context *h, int intra)
216 {
217     const unsigned int index_a = qp + a;
218     const int alpha = alpha_table[index_a];
219     const int beta  = beta_table[qp + b];
220     if (alpha ==0 || beta == 0) return;
221 
222     if( bS[0] < 4 || !intra ) {
223         int8_t tc[4];
224         tc[0] = tc0_table[index_a][bS[0]]+1;
225         tc[1] = tc0_table[index_a][bS[1]]+1;
226         tc[2] = tc0_table[index_a][bS[2]]+1;
227         tc[3] = tc0_table[index_a][bS[3]]+1;
228         h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
229     } else {
230         h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
231     }
232 }
233 
h264_filter_mb_fast_internal(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int pixel_shift)234 static av_always_inline void h264_filter_mb_fast_internal(const H264Context *h,
235                                                           H264SliceContext *sl,
236                                                           int mb_x, int mb_y,
237                                                           uint8_t *img_y,
238                                                           uint8_t *img_cb,
239                                                           uint8_t *img_cr,
240                                                           unsigned int linesize,
241                                                           unsigned int uvlinesize,
242                                                           int pixel_shift)
243 {
244     int chroma = CHROMA(h) && !(CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
245     int chroma444 = CHROMA444(h);
246     int chroma422 = CHROMA422(h);
247 
248     int mb_xy = sl->mb_xy;
249     int left_type = sl->left_type[LTOP];
250     int top_type  = sl->top_type;
251 
252     int qp_bd_offset = 6 * (h->ps.sps->bit_depth_luma - 8);
253     int a = 52 + sl->slice_alpha_c0_offset - qp_bd_offset;
254     int b = 52 + sl->slice_beta_offset - qp_bd_offset;
255 
256     int mb_type = h->cur_pic.mb_type[mb_xy];
257     int qp      = h->cur_pic.qscale_table[mb_xy];
258     int qp0     = h->cur_pic.qscale_table[mb_xy - 1];
259     int qp1     = h->cur_pic.qscale_table[sl->top_mb_xy];
260     int qpc  = get_chroma_qp(h->ps.pps, 0, qp);
261     int qpc0 = get_chroma_qp(h->ps.pps, 0, qp0);
262     int qpc1 = get_chroma_qp(h->ps.pps, 0, qp1);
263     qp0 = (qp + qp0 + 1) >> 1;
264     qp1 = (qp + qp1 + 1) >> 1;
265     qpc0 = (qpc + qpc0 + 1) >> 1;
266     qpc1 = (qpc + qpc1 + 1) >> 1;
267 
268     if( IS_INTRA(mb_type) ) {
269         static const int16_t bS4[4] = {4,4,4,4};
270         static const int16_t bS3[4] = {3,3,3,3};
271         const int16_t *bSH = FIELD_PICTURE(h) ? bS3 : bS4;
272         if(left_type)
273             filter_mb_edgev( &img_y[4*0<<pixel_shift], linesize, bS4, qp0, a, b, h, 1);
274         if( IS_8x8DCT(mb_type) ) {
275             filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
276             if(top_type){
277                 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
278             }
279             filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
280         } else {
281             filter_mb_edgev( &img_y[4*1<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
282             filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
283             filter_mb_edgev( &img_y[4*3<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
284             if(top_type){
285                 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
286             }
287             filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, a, b, h, 0);
288             filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
289             filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, a, b, h, 0);
290         }
291         if(chroma){
292             if(chroma444){
293                 if(left_type){
294                     filter_mb_edgev( &img_cb[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
295                     filter_mb_edgev( &img_cr[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
296                 }
297                 if( IS_8x8DCT(mb_type) ) {
298                     filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
299                     filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
300                     if(top_type){
301                         filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
302                         filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
303                     }
304                     filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
305                     filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
306                 } else {
307                     filter_mb_edgev( &img_cb[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
308                     filter_mb_edgev( &img_cr[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
309                     filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
310                     filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
311                     filter_mb_edgev( &img_cb[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
312                     filter_mb_edgev( &img_cr[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
313                     if(top_type){
314                         filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
315                         filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
316                     }
317                     filter_mb_edgeh( &img_cb[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
318                     filter_mb_edgeh( &img_cr[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
319                     filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
320                     filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
321                     filter_mb_edgeh( &img_cb[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
322                     filter_mb_edgeh( &img_cr[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
323                 }
324             }else if(chroma422){
325                 if(left_type){
326                     filter_mb_edgecv(&img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
327                     filter_mb_edgecv(&img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
328                 }
329                 filter_mb_edgecv(&img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
330                 filter_mb_edgecv(&img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
331                 if(top_type){
332                     filter_mb_edgech(&img_cb[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
333                     filter_mb_edgech(&img_cr[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
334                 }
335                 filter_mb_edgech(&img_cb[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
336                 filter_mb_edgech(&img_cr[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
337                 filter_mb_edgech(&img_cb[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
338                 filter_mb_edgech(&img_cr[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
339                 filter_mb_edgech(&img_cb[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
340                 filter_mb_edgech(&img_cr[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
341             }else{
342                 if(left_type){
343                     filter_mb_edgecv( &img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
344                     filter_mb_edgecv( &img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
345                 }
346                 filter_mb_edgecv( &img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
347                 filter_mb_edgecv( &img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
348                 if(top_type){
349                     filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
350                     filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
351                 }
352                 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
353                 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
354             }
355         }
356         return;
357     } else {
358         LOCAL_ALIGNED(8, int16_t, bS, [2], [4][4]);
359         int edges;
360         if( IS_8x8DCT(mb_type) && (sl->cbp&7) == 7 && !chroma444 ) {
361             edges = 4;
362             AV_WN64A(bS[0][0], 0x0002000200020002ULL);
363             AV_WN64A(bS[0][2], 0x0002000200020002ULL);
364             AV_WN64A(bS[1][0], 0x0002000200020002ULL);
365             AV_WN64A(bS[1][2], 0x0002000200020002ULL);
366         } else {
367             int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0;
368             int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[LTOP] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0;
369             int step =  1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
370             edges = 4 - 3*((mb_type>>3) & !(sl->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
371             h->h264dsp.h264_loop_filter_strength(bS, sl->non_zero_count_cache, sl->ref_cache, sl->mv_cache,
372                                                  sl->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE(h));
373         }
374         if( IS_INTRA(left_type) )
375             AV_WN64A(bS[0][0], 0x0004000400040004ULL);
376         if( IS_INTRA(top_type) )
377             AV_WN64A(bS[1][0], FIELD_PICTURE(h) ? 0x0003000300030003ULL : 0x0004000400040004ULL);
378 
379 #define FILTER(hv,dir,edge,intra)\
380         if(AV_RN64A(bS[dir][edge])) {                                   \
381             filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qp : qp##dir, a, b, h, intra );\
382             if(chroma){\
383                 if(chroma444){\
384                     filter_mb_edge##hv( &img_cb[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
385                     filter_mb_edge##hv( &img_cr[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
386                 } else if(!(edge&1)) {\
387                     filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1<<pixel_shift)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
388                     filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1<<pixel_shift)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\
389                 }\
390             }\
391         }
392         if(left_type)
393             FILTER(v,0,0,1);
394         if( edges == 1 ) {
395             if(top_type)
396                 FILTER(h,1,0,1);
397         } else if( IS_8x8DCT(mb_type) ) {
398             FILTER(v,0,2,0);
399             if(top_type)
400                 FILTER(h,1,0,1);
401             FILTER(h,1,2,0);
402         } else {
403             FILTER(v,0,1,0);
404             FILTER(v,0,2,0);
405             FILTER(v,0,3,0);
406             if(top_type)
407                 FILTER(h,1,0,1);
408             FILTER(h,1,1,0);
409             FILTER(h,1,2,0);
410             FILTER(h,1,3,0);
411         }
412 #undef FILTER
413     }
414 }
415 
ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize)416 void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl,
417                             int mb_x, int mb_y, uint8_t *img_y,
418                             uint8_t *img_cb, uint8_t *img_cr,
419                             unsigned int linesize, unsigned int uvlinesize)
420 {
421     av_assert2(!FRAME_MBAFF(h));
422     if(!h->h264dsp.h264_loop_filter_strength || h->ps.pps->chroma_qp_diff) {
423         ff_h264_filter_mb(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
424         return;
425     }
426 
427 #if CONFIG_SMALL
428     h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift);
429 #else
430     if(h->pixel_shift){
431         h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1);
432     }else{
433         h264_filter_mb_fast_internal(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0);
434     }
435 #endif
436 }
437 
check_mv(H264SliceContext *sl, long b_idx, long bn_idx, int mvy_limit)438 static int check_mv(H264SliceContext *sl, long b_idx, long bn_idx, int mvy_limit)
439 {
440     int v;
441 
442     v = sl->ref_cache[0][b_idx] != sl->ref_cache[0][bn_idx];
443     if (!v && sl->ref_cache[0][b_idx] != -1)
444         v = sl->mv_cache[0][b_idx][0] - sl->mv_cache[0][bn_idx][0] + 3 >= 7U |
445            FFABS(sl->mv_cache[0][b_idx][1] - sl->mv_cache[0][bn_idx][1]) >= mvy_limit;
446 
447     if (sl->list_count == 2) {
448         if(!v)
449             v = sl->ref_cache[1][b_idx] != sl->ref_cache[1][bn_idx] |
450                 sl->mv_cache[1][b_idx][0] - sl->mv_cache[1][bn_idx][0] + 3 >= 7U |
451                 FFABS(sl->mv_cache[1][b_idx][1] - sl->mv_cache[1][bn_idx][1]) >= mvy_limit;
452 
453         if(v){
454             if (sl->ref_cache[0][b_idx] != sl->ref_cache[1][bn_idx] |
455                 sl->ref_cache[1][b_idx] != sl->ref_cache[0][bn_idx])
456                 return 1;
457             return
458                 sl->mv_cache[0][b_idx][0] - sl->mv_cache[1][bn_idx][0] + 3 >= 7U |
459                 FFABS(sl->mv_cache[0][b_idx][1] - sl->mv_cache[1][bn_idx][1]) >= mvy_limit |
460                 sl->mv_cache[1][b_idx][0] - sl->mv_cache[0][bn_idx][0] + 3 >= 7U |
461                 FFABS(sl->mv_cache[1][b_idx][1] - sl->mv_cache[0][bn_idx][1]) >= mvy_limit;
462         }
463     }
464 
465     return v;
466 }
467 
filter_mb_dir(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int a, int b, int chroma, int dir)468 static av_always_inline void filter_mb_dir(const H264Context *h, H264SliceContext *sl,
469                                            int mb_x, int mb_y,
470                                            uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
471                                            unsigned int linesize, unsigned int uvlinesize,
472                                            int mb_xy, int mb_type, int mvy_limit,
473                                            int first_vertical_edge_done, int a, int b,
474                                            int chroma, int dir)
475 {
476     int edge;
477     int chroma_qp_avg[2];
478     int chroma444 = CHROMA444(h);
479     int chroma422 = CHROMA422(h);
480     const int mbm_xy = dir == 0 ? mb_xy -1 : sl->top_mb_xy;
481     const int mbm_type = dir == 0 ? sl->left_type[LTOP] : sl->top_type;
482 
483     // how often to recheck mv-based bS when iterating between edges
484     static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
485                                               {0,3,1,1,3,3,3,3}};
486     const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
487     const int edges = mask_edge== 3 && !(sl->cbp&15) ? 1 : 4;
488 
489     // how often to recheck mv-based bS when iterating along each edge
490     const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
491 
492     if(mbm_type && !first_vertical_edge_done){
493 
494         if (FRAME_MBAFF(h) && (dir == 1) && ((mb_y&1) == 0)
495             && IS_INTERLACED(mbm_type&~mb_type)
496             ) {
497             // This is a special case in the norm where the filtering must
498             // be done twice (one each of the field) even if we are in a
499             // frame macroblock.
500             //
501             unsigned int tmp_linesize   = 2 *   linesize;
502             unsigned int tmp_uvlinesize = 2 * uvlinesize;
503             int mbn_xy = mb_xy - 2 * h->mb_stride;
504             int j;
505 
506             for(j=0; j<2; j++, mbn_xy += h->mb_stride){
507                 LOCAL_ALIGNED(8, int16_t, bS, [4]);
508                 int qp;
509                 if (IS_INTRA(mb_type | h->cur_pic.mb_type[mbn_xy])) {
510                     AV_WN64A(bS, 0x0003000300030003ULL);
511                 } else {
512                     if (!CABAC(h) && IS_8x8DCT(h->cur_pic.mb_type[mbn_xy])) {
513                         bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000) || sl->non_zero_count_cache[scan8[0]+0]);
514                         bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000) || sl->non_zero_count_cache[scan8[0]+1]);
515                         bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000) || sl->non_zero_count_cache[scan8[0]+2]);
516                         bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000) || sl->non_zero_count_cache[scan8[0]+3]);
517                     }else{
518                     const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4;
519                     int i;
520                     for( i = 0; i < 4; i++ ) {
521                         bS[i] = 1 + !!(sl->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
522                     }
523                     }
524                 }
525                 // Do not use s->qscale as luma quantizer because it has not the same
526                 // value in IPCM macroblocks.
527                 qp = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbn_xy] + 1) >> 1;
528                 ff_tlog(h->avctx, "filter mb:%d/%d dir:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, qp, tmp_linesize, tmp_uvlinesize);
529                 { int i; for (i = 0; i < 4; i++) ff_tlog(h->avctx, " bS[%d]:%d", i, bS[i]); ff_tlog(h->avctx, "\n"); }
530                 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, a, b, h, 0 );
531                 chroma_qp_avg[0] = (sl->chroma_qp[0] + get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mbn_xy]) + 1) >> 1;
532                 chroma_qp_avg[1] = (sl->chroma_qp[1] + get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mbn_xy]) + 1) >> 1;
533                 if (chroma) {
534                     if (chroma444) {
535                         filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
536                         filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
537                     } else {
538                         filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
539                         filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
540                     }
541                 }
542             }
543         }else{
544             LOCAL_ALIGNED(8, int16_t, bS, [4]);
545             int qp;
546 
547             if( IS_INTRA(mb_type|mbm_type)) {
548                 AV_WN64A(bS, 0x0003000300030003ULL);
549                 if (   (!IS_INTERLACED(mb_type|mbm_type))
550                     || ((FRAME_MBAFF(h) || (h->picture_structure != PICT_FRAME)) && (dir == 0))
551                 )
552                     AV_WN64A(bS, 0x0004000400040004ULL);
553             } else {
554                 int i;
555                 int mv_done;
556 
557                 if( dir && FRAME_MBAFF(h) && IS_INTERLACED(mb_type ^ mbm_type)) {
558                     AV_WN64A(bS, 0x0001000100010001ULL);
559                     mv_done = 1;
560                 }
561                 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
562                     int b_idx= 8 + 4;
563                     int bn_idx= b_idx - (dir ? 8:1);
564 
565                     bS[0] = bS[1] = bS[2] = bS[3] = check_mv(sl, 8 + 4, bn_idx, mvy_limit);
566                     mv_done = 1;
567                 }
568                 else
569                     mv_done = 0;
570 
571                 for( i = 0; i < 4; i++ ) {
572                     int x = dir == 0 ? 0 : i;
573                     int y = dir == 0 ? i    : 0;
574                     int b_idx= 8 + 4 + x + 8*y;
575                     int bn_idx= b_idx - (dir ? 8:1);
576 
577                     if (sl->non_zero_count_cache[b_idx] |
578                         sl->non_zero_count_cache[bn_idx]) {
579                         bS[i] = 2;
580                     }
581                     else if(!mv_done)
582                     {
583                         bS[i] = check_mv(sl, b_idx, bn_idx, mvy_limit);
584                     }
585                 }
586             }
587 
588             /* Filter edge */
589             // Do not use s->qscale as luma quantizer because it has not the same
590             // value in IPCM macroblocks.
591             if(bS[0]+bS[1]+bS[2]+bS[3]){
592                 qp = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbm_xy] + 1) >> 1;
593                 //ff_tlog(h->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp[0], h->cur_pic.qscale_table[mbn_xy]);
594                 ff_tlog(h->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
595                 //{ int i; for (i = 0; i < 4; i++) ff_tlog(h->avctx, " bS[%d]:%d", i, bS[i]); ff_tlog(h->avctx, "\n"); }
596                 chroma_qp_avg[0] = (sl->chroma_qp[0] + get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mbm_xy]) + 1) >> 1;
597                 chroma_qp_avg[1] = (sl->chroma_qp[1] + get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mbm_xy]) + 1) >> 1;
598                 if( dir == 0 ) {
599                     filter_mb_edgev( &img_y[0], linesize, bS, qp, a, b, h, 1 );
600                     if (chroma) {
601                         if (chroma444) {
602                             filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
603                             filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
604                         } else {
605                             filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
606                             filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
607                         }
608                     }
609                 } else {
610                     filter_mb_edgeh( &img_y[0], linesize, bS, qp, a, b, h, 1 );
611                     if (chroma) {
612                         if (chroma444) {
613                             filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
614                             filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
615                         } else {
616                             filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
617                             filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
618                         }
619                     }
620                 }
621             }
622         }
623     }
624 
625     /* Calculate bS */
626     for( edge = 1; edge < edges; edge++ ) {
627         LOCAL_ALIGNED(8, int16_t, bS, [4]);
628         int qp;
629         const int deblock_edge = !IS_8x8DCT(mb_type & (edge<<24)); // (edge&1) && IS_8x8DCT(mb_type)
630 
631         if (!deblock_edge && (!chroma422 || dir == 0))
632             continue;
633 
634         if( IS_INTRA(mb_type)) {
635             AV_WN64A(bS, 0x0003000300030003ULL);
636         } else {
637             int i;
638             int mv_done;
639 
640             if( edge & mask_edge ) {
641                 AV_ZERO64(bS);
642                 mv_done = 1;
643             }
644             else if( mask_par0 ) {
645                 int b_idx= 8 + 4 + edge * (dir ? 8:1);
646                 int bn_idx= b_idx - (dir ? 8:1);
647 
648                 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(sl, b_idx, bn_idx, mvy_limit);
649                 mv_done = 1;
650             }
651             else
652                 mv_done = 0;
653 
654             for( i = 0; i < 4; i++ ) {
655                 int x = dir == 0 ? edge : i;
656                 int y = dir == 0 ? i    : edge;
657                 int b_idx= 8 + 4 + x + 8*y;
658                 int bn_idx= b_idx - (dir ? 8:1);
659 
660                 if (sl->non_zero_count_cache[b_idx] |
661                     sl->non_zero_count_cache[bn_idx]) {
662                     bS[i] = 2;
663                 }
664                 else if(!mv_done)
665                 {
666                     bS[i] = check_mv(sl, b_idx, bn_idx, mvy_limit);
667                 }
668             }
669 
670             if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
671                 continue;
672         }
673 
674         /* Filter edge */
675         // Do not use s->qscale as luma quantizer because it has not the same
676         // value in IPCM macroblocks.
677         qp = h->cur_pic.qscale_table[mb_xy];
678         //ff_tlog(h->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp[0], h->cur_pic.qscale_table[mbn_xy]);
679         ff_tlog(h->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
680         //{ int i; for (i = 0; i < 4; i++) ff_tlog(h->avctx, " bS[%d]:%d", i, bS[i]); ff_tlog(h->avctx, "\n"); }
681         if( dir == 0 ) {
682             filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, a, b, h, 0 );
683             if (chroma) {
684                 if (chroma444) {
685                     filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
686                     filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
687                 } else if( (edge&1) == 0 ) {
688                     filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
689                     filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
690                 }
691             }
692         } else {
693             if (chroma422) {
694                 if (deblock_edge)
695                     filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);
696                 if (chroma) {
697                     filter_mb_edgech(&img_cb[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
698                     filter_mb_edgech(&img_cr[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
699                 }
700             } else {
701                 filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);
702                 if (chroma) {
703                     if (chroma444) {
704                         filter_mb_edgeh (&img_cb[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
705                         filter_mb_edgeh (&img_cr[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
706                     } else if ((edge&1) == 0) {
707                         filter_mb_edgech(&img_cb[2*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
708                         filter_mb_edgech(&img_cr[2*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
709                     }
710                 }
711             }
712         }
713     }
714 }
715 
ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize)716 void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl,
717                        int mb_x, int mb_y,
718                        uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
719                        unsigned int linesize, unsigned int uvlinesize)
720 {
721     const int mb_xy= mb_x + mb_y*h->mb_stride;
722     const int mb_type = h->cur_pic.mb_type[mb_xy];
723     const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
724     int first_vertical_edge_done = 0;
725     int chroma = CHROMA(h) && !(CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
726     int qp_bd_offset = 6 * (h->ps.sps->bit_depth_luma - 8);
727     int a = 52 + sl->slice_alpha_c0_offset - qp_bd_offset;
728     int b = 52 + sl->slice_beta_offset - qp_bd_offset;
729 
730     if (FRAME_MBAFF(h)
731             // and current and left pair do not have the same interlaced type
732             && IS_INTERLACED(mb_type ^ sl->left_type[LTOP])
733             // and left mb is in available to us
734             && sl->left_type[LTOP]) {
735         /* First vertical edge is different in MBAFF frames
736          * There are 8 different bS to compute and 2 different Qp
737          */
738         LOCAL_ALIGNED(8, int16_t, bS, [8]);
739         int qp[2];
740         int bqp[2];
741         int rqp[2];
742         int mb_qp, mbn0_qp, mbn1_qp;
743         int i;
744         first_vertical_edge_done = 1;
745 
746         if( IS_INTRA(mb_type) ) {
747             AV_WN64A(&bS[0], 0x0004000400040004ULL);
748             AV_WN64A(&bS[4], 0x0004000400040004ULL);
749         } else {
750             static const uint8_t offset[2][2][8]={
751                 {
752                     {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},
753                     {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},
754                 },{
755                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
756                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
757                 }
758             };
759             const uint8_t *off= offset[MB_FIELD(sl)][mb_y&1];
760             for( i = 0; i < 8; i++ ) {
761                 int j= MB_FIELD(sl) ? i>>2 : i&1;
762                 int mbn_xy = sl->left_mb_xy[LEFT(j)];
763                 int mbn_type = sl->left_type[LEFT(j)];
764 
765                 if( IS_INTRA( mbn_type ) )
766                     bS[i] = 4;
767                 else{
768                     bS[i] = 1 + !!(sl->non_zero_count_cache[12+8*(i>>1)] |
769                          ((!h->ps.pps->cabac && IS_8x8DCT(mbn_type)) ?
770                             (h->cbp_table[mbn_xy] & (((MB_FIELD(sl) ? (i&2) : (mb_y&1)) ? 8 : 2) << 12))
771                                                                        :
772                             h->non_zero_count[mbn_xy][ off[i] ]));
773                 }
774             }
775         }
776 
777         mb_qp   = h->cur_pic.qscale_table[mb_xy];
778         mbn0_qp = h->cur_pic.qscale_table[sl->left_mb_xy[0]];
779         mbn1_qp = h->cur_pic.qscale_table[sl->left_mb_xy[1]];
780         qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
781         bqp[0] = (get_chroma_qp(h->ps.pps, 0, mb_qp) +
782                   get_chroma_qp(h->ps.pps, 0, mbn0_qp) + 1) >> 1;
783         rqp[0] = (get_chroma_qp(h->ps.pps, 1, mb_qp) +
784                   get_chroma_qp(h->ps.pps, 1, mbn0_qp) + 1) >> 1;
785         qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
786         bqp[1] = (get_chroma_qp(h->ps.pps, 0, mb_qp) +
787                   get_chroma_qp(h->ps.pps, 0, mbn1_qp) + 1 ) >> 1;
788         rqp[1] = (get_chroma_qp(h->ps.pps, 1, mb_qp) +
789                   get_chroma_qp(h->ps.pps, 1, mbn1_qp) + 1 ) >> 1;
790 
791         /* Filter edge */
792         ff_tlog(h->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPb:%d/%d QPr:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], bqp[0], bqp[1], rqp[0], rqp[1], linesize, uvlinesize);
793         { int i; for (i = 0; i < 8; i++) ff_tlog(h->avctx, " bS[%d]:%d", i, bS[i]); ff_tlog(h->avctx, "\n"); }
794         if (MB_FIELD(sl)) {
795             filter_mb_mbaff_edgev ( h, img_y                ,   linesize, bS  , 1, qp [0], a, b, 1 );
796             filter_mb_mbaff_edgev ( h, img_y  + 8*  linesize,   linesize, bS+4, 1, qp [1], a, b, 1 );
797             if (chroma){
798                 if (CHROMA444(h)) {
799                     filter_mb_mbaff_edgev ( h, img_cb,                uvlinesize, bS  , 1, bqp[0], a, b, 1 );
800                     filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
801                     filter_mb_mbaff_edgev ( h, img_cr,                uvlinesize, bS  , 1, rqp[0], a, b, 1 );
802                     filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
803                 } else if (CHROMA422(h)) {
804                     filter_mb_mbaff_edgecv(h, img_cb,                uvlinesize, bS  , 1, bqp[0], a, b, 1);
805                     filter_mb_mbaff_edgecv(h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1);
806                     filter_mb_mbaff_edgecv(h, img_cr,                uvlinesize, bS  , 1, rqp[0], a, b, 1);
807                     filter_mb_mbaff_edgecv(h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1);
808                 }else{
809                     filter_mb_mbaff_edgecv( h, img_cb,                uvlinesize, bS  , 1, bqp[0], a, b, 1 );
810                     filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
811                     filter_mb_mbaff_edgecv( h, img_cr,                uvlinesize, bS  , 1, rqp[0], a, b, 1 );
812                     filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
813                 }
814             }
815         }else{
816             filter_mb_mbaff_edgev ( h, img_y              , 2*  linesize, bS  , 2, qp [0], a, b, 1 );
817             filter_mb_mbaff_edgev ( h, img_y  +   linesize, 2*  linesize, bS+1, 2, qp [1], a, b, 1 );
818             if (chroma){
819                 if (CHROMA444(h)) {
820                     filter_mb_mbaff_edgev ( h, img_cb,              2*uvlinesize, bS  , 2, bqp[0], a, b, 1 );
821                     filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
822                     filter_mb_mbaff_edgev ( h, img_cr,              2*uvlinesize, bS  , 2, rqp[0], a, b, 1 );
823                     filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
824                 }else{
825                     filter_mb_mbaff_edgecv( h, img_cb,              2*uvlinesize, bS  , 2, bqp[0], a, b, 1 );
826                     filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
827                     filter_mb_mbaff_edgecv( h, img_cr,              2*uvlinesize, bS  , 2, rqp[0], a, b, 1 );
828                     filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
829                 }
830             }
831         }
832     }
833 
834 #if CONFIG_SMALL
835     {
836         int dir;
837         for (dir = 0; dir < 2; dir++)
838             filter_mb_dir(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize,
839                           uvlinesize, mb_xy, mb_type, mvy_limit,
840                           dir ? 0 : first_vertical_edge_done, a, b,
841                           chroma, dir);
842     }
843 #else
844     filter_mb_dir(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, a, b, chroma, 0);
845     filter_mb_dir(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0,                        a, b, chroma, 1);
846 #endif
847 }
848