2 * H.26L/H.264/AVC/JVT/14496-10/... loop filter
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of Libav.
7 * Libav 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.
12 * Libav 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 loop filter.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/intreadwrite.h"
32 #include "mpegvideo.h"
35 #include "rectangle.h"
40 /* Deblocking filter (p153) */
41 static const uint8_t alpha_table[52*3] = {
42 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
43 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
44 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
45 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
46 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
47 0, 0, 0, 0, 0, 0, 4, 4, 5, 6,
48 7, 8, 9, 10, 12, 13, 15, 17, 20, 22,
49 25, 28, 32, 36, 40, 45, 50, 56, 63, 71,
50 80, 90,101,113,127,144,162,182,203,226,
52 255,255,255,255,255,255,255,255,255,255,255,255,255,
53 255,255,255,255,255,255,255,255,255,255,255,255,255,
54 255,255,255,255,255,255,255,255,255,255,255,255,255,
55 255,255,255,255,255,255,255,255,255,255,255,255,255,
57 static const uint8_t beta_table[52*3] = {
58 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
59 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
60 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
61 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
62 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
63 0, 0, 0, 0, 0, 0, 2, 2, 2, 3,
64 3, 3, 3, 4, 4, 4, 6, 6, 7, 7,
65 8, 8, 9, 9, 10, 10, 11, 11, 12, 12,
66 13, 13, 14, 14, 15, 15, 16, 16, 17, 17,
68 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
69 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
70 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
71 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
73 static const uint8_t tc0_table[52*3][4] = {
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 }, {-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, 0 },
82 {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
83 {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
84 {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
85 {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 1 },
86 {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },
87 {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },
88 {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },
89 {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },
90 {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },
91 {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-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 }, {-1,13,17,25 }, {-1,13,17,25 },
97 {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
98 {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
99 {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
100 {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
103 static void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) {
104 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
105 const int alpha = alpha_table[index_a];
106 const int beta = beta_table[qp + h->slice_beta_offset];
107 if (alpha ==0 || beta == 0) return;
111 tc[0] = tc0_table[index_a][bS[0]];
112 tc[1] = tc0_table[index_a][bS[1]];
113 tc[2] = tc0_table[index_a][bS[2]];
114 tc[3] = tc0_table[index_a][bS[3]];
115 h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
117 h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
120 static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
121 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
122 const int alpha = alpha_table[index_a];
123 const int beta = beta_table[qp + h->slice_beta_offset];
124 if (alpha ==0 || beta == 0) return;
128 tc[0] = tc0_table[index_a][bS[0]]+1;
129 tc[1] = tc0_table[index_a][bS[1]]+1;
130 tc[2] = tc0_table[index_a][bS[2]]+1;
131 tc[3] = tc0_table[index_a][bS[3]]+1;
132 h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
134 h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
138 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
139 int index_a = qp + h->slice_alpha_c0_offset;
140 int alpha = alpha_table[index_a];
141 int beta = beta_table[qp + h->slice_beta_offset];
142 if (alpha ==0 || beta == 0) return;
146 tc[0] = tc0_table[index_a][bS[0*bsi]];
147 tc[1] = tc0_table[index_a][bS[1*bsi]];
148 tc[2] = tc0_table[index_a][bS[2*bsi]];
149 tc[3] = tc0_table[index_a][bS[3*bsi]];
150 h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
152 h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
155 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
156 int index_a = qp + h->slice_alpha_c0_offset;
157 int alpha = alpha_table[index_a];
158 int beta = beta_table[qp + h->slice_beta_offset];
159 if (alpha ==0 || beta == 0) return;
163 tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
164 tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
165 tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
166 tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
167 h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
169 h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
173 static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
174 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
175 const int alpha = alpha_table[index_a];
176 const int beta = beta_table[qp + h->slice_beta_offset];
177 if (alpha ==0 || beta == 0) return;
181 tc[0] = tc0_table[index_a][bS[0]];
182 tc[1] = tc0_table[index_a][bS[1]];
183 tc[2] = tc0_table[index_a][bS[2]];
184 tc[3] = tc0_table[index_a][bS[3]];
185 h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
187 h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
191 static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
192 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
193 const int alpha = alpha_table[index_a];
194 const int beta = beta_table[qp + h->slice_beta_offset];
195 if (alpha ==0 || beta == 0) return;
199 tc[0] = tc0_table[index_a][bS[0]]+1;
200 tc[1] = tc0_table[index_a][bS[1]]+1;
201 tc[2] = tc0_table[index_a][bS[2]]+1;
202 tc[3] = tc0_table[index_a][bS[3]]+1;
203 h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
205 h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
209 void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
210 MpegEncContext * const s = &h->s;
212 int mb_type, left_type;
213 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
217 if(!h->top_type || !h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
218 ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
221 assert(!FRAME_MBAFF);
222 left_type= h->left_type[0];
224 mb_type = s->current_picture.mb_type[mb_xy];
225 qp = s->current_picture.qscale_table[mb_xy];
226 qp0 = s->current_picture.qscale_table[mb_xy-1];
227 qp1 = s->current_picture.qscale_table[h->top_mb_xy];
228 qpc = get_chroma_qp( h, 0, qp );
229 qpc0 = get_chroma_qp( h, 0, qp0 );
230 qpc1 = get_chroma_qp( h, 0, qp1 );
231 qp0 = (qp + qp0 + 1) >> 1;
232 qp1 = (qp + qp1 + 1) >> 1;
233 qpc0 = (qpc + qpc0 + 1) >> 1;
234 qpc1 = (qpc + qpc1 + 1) >> 1;
235 qp_thresh = 15+52 - h->slice_alpha_c0_offset;
236 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
237 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
240 if( IS_INTRA(mb_type) ) {
241 int16_t bS4[4] = {4,4,4,4};
242 int16_t bS3[4] = {3,3,3,3};
243 int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
245 filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
246 if( IS_8x8DCT(mb_type) ) {
247 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
248 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
249 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
251 filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
252 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
253 filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
254 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
255 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
256 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
257 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
260 filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
261 filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
263 filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
264 filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
265 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
266 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
267 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
268 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
271 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
273 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
275 AV_WN64A(bS[0][0], 0x0002000200020002ULL);
276 AV_WN64A(bS[0][2], 0x0002000200020002ULL);
277 AV_WN64A(bS[1][0], 0x0002000200020002ULL);
278 AV_WN64A(bS[1][2], 0x0002000200020002ULL);
280 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;
281 int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0;
282 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
283 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
284 h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
285 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
287 if( IS_INTRA(left_type) )
288 AV_WN64A(bS[0][0], 0x0004000400040004ULL);
289 if( IS_INTRA(h->top_type) )
290 AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
292 #define FILTER(hv,dir,edge)\
293 if(AV_RN64A(bS[dir][edge])) { \
294 filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
296 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
297 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
304 } else if( IS_8x8DCT(mb_type) ) {
321 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
324 v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
325 if(!v && h->ref_cache[0][b_idx]!=-1)
326 v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
327 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
329 if(h->list_count==2){
331 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
332 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
333 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
336 if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
337 h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
340 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
341 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
342 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
343 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
350 static av_always_inline void filter_mb_dir(H264Context *h, 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 dir) {
351 MpegEncContext * const s = &h->s;
353 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
354 const int mbm_type = dir == 0 ? h->left_type[0] : h->top_type;
356 // how often to recheck mv-based bS when iterating between edges
357 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
359 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
360 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
362 // how often to recheck mv-based bS when iterating along each edge
363 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
365 if(mbm_type && !first_vertical_edge_done){
367 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
368 && IS_INTERLACED(mbm_type&~mb_type)
370 // This is a special case in the norm where the filtering must
371 // be done twice (one each of the field) even if we are in a
374 unsigned int tmp_linesize = 2 * linesize;
375 unsigned int tmp_uvlinesize = 2 * uvlinesize;
376 int mbn_xy = mb_xy - 2 * s->mb_stride;
379 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
380 DECLARE_ALIGNED(8, int16_t, bS)[4];
382 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
383 AV_WN64A(bS, 0x0003000300030003ULL);
385 if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
386 bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]);
387 bS[1]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+1]);
388 bS[2]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+2]);
389 bS[3]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+3]);
391 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8;
393 for( i = 0; i < 4; i++ ) {
394 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
398 // Do not use s->qscale as luma quantizer because it has not the same
399 // value in IPCM macroblocks.
400 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
401 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
402 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
403 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
404 filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
405 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
406 filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
407 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
410 DECLARE_ALIGNED(8, int16_t, bS)[4];
413 if( IS_INTRA(mb_type|mbm_type)) {
414 AV_WN64A(bS, 0x0003000300030003ULL);
415 if ( (!IS_INTERLACED(mb_type|mbm_type))
416 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
418 AV_WN64A(bS, 0x0004000400040004ULL);
423 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
424 AV_WN64A(bS, 0x0001000100010001ULL);
427 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
429 int bn_idx= b_idx - (dir ? 8:1);
431 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
437 for( i = 0; i < 4; i++ ) {
438 int x = dir == 0 ? 0 : i;
439 int y = dir == 0 ? i : 0;
440 int b_idx= 8 + 4 + x + 8*y;
441 int bn_idx= b_idx - (dir ? 8:1);
443 if( h->non_zero_count_cache[b_idx] |
444 h->non_zero_count_cache[bn_idx] ) {
449 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
455 // Do not use s->qscale as luma quantizer because it has not the same
456 // value in IPCM macroblocks.
457 if(bS[0]+bS[1]+bS[2]+bS[3]){
458 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbm_xy] + 1 ) >> 1;
459 //tprintf(s->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], s->current_picture.qscale_table[mbn_xy]);
460 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
461 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
463 filter_mb_edgev( &img_y[0], linesize, bS, qp, h );
465 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
466 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, qp, h);
467 if(h->pps.chroma_qp_diff)
468 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
469 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, qp, h);
472 filter_mb_edgeh( &img_y[0], linesize, bS, qp, h );
474 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
475 filter_mb_edgech( &img_cb[0], uvlinesize, bS, qp, h);
476 if(h->pps.chroma_qp_diff)
477 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
478 filter_mb_edgech( &img_cr[0], uvlinesize, bS, qp, h);
486 for( edge = 1; edge < edges; edge++ ) {
487 DECLARE_ALIGNED(8, int16_t, bS)[4];
490 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
493 if( IS_INTRA(mb_type)) {
494 AV_WN64A(bS, 0x0003000300030003ULL);
499 if( edge & mask_edge ) {
503 else if( mask_par0 ) {
504 int b_idx= 8 + 4 + edge * (dir ? 8:1);
505 int bn_idx= b_idx - (dir ? 8:1);
507 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
513 for( i = 0; i < 4; i++ ) {
514 int x = dir == 0 ? edge : i;
515 int y = dir == 0 ? i : edge;
516 int b_idx= 8 + 4 + x + 8*y;
517 int bn_idx= b_idx - (dir ? 8:1);
519 if( h->non_zero_count_cache[b_idx] |
520 h->non_zero_count_cache[bn_idx] ) {
525 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
529 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
534 // Do not use s->qscale as luma quantizer because it has not the same
535 // value in IPCM macroblocks.
536 qp = s->current_picture.qscale_table[mb_xy];
537 //tprintf(s->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], s->current_picture.qscale_table[mbn_xy]);
538 tprintf(s->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
539 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
541 filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h );
542 if( (edge&1) == 0 ) {
543 filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, h->chroma_qp[0], h);
544 filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, h->chroma_qp[1], h);
547 filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
548 if( (edge&1) == 0 ) {
549 filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
550 filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
556 void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
557 MpegEncContext * const s = &h->s;
558 const int mb_xy= mb_x + mb_y*s->mb_stride;
559 const int mb_type = s->current_picture.mb_type[mb_xy];
560 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
561 int first_vertical_edge_done = 0;
565 // and current and left pair do not have the same interlaced type
566 && IS_INTERLACED(mb_type^h->left_type[0])
567 // and left mb is in available to us
568 && h->left_type[0]) {
569 /* First vertical edge is different in MBAFF frames
570 * There are 8 different bS to compute and 2 different Qp
572 DECLARE_ALIGNED(8, int16_t, bS)[8];
576 int mb_qp, mbn0_qp, mbn1_qp;
578 first_vertical_edge_done = 1;
580 if( IS_INTRA(mb_type) ) {
581 AV_WN64A(&bS[0], 0x0004000400040004ULL);
582 AV_WN64A(&bS[4], 0x0004000400040004ULL);
584 static const uint8_t offset[2][2][8]={
586 {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
587 {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3},
589 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
590 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
593 const uint8_t *off= offset[MB_FIELD][mb_y&1];
594 for( i = 0; i < 8; i++ ) {
595 int j= MB_FIELD ? i>>2 : i&1;
596 int mbn_xy = h->left_mb_xy[j];
597 int mbn_type= h->left_type[j];
599 if( IS_INTRA( mbn_type ) )
602 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
603 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
604 (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
606 h->non_zero_count[mbn_xy][ off[i] ]));
611 mb_qp = s->current_picture.qscale_table[mb_xy];
612 mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]];
613 mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]];
614 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
615 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
616 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
617 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
618 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
619 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
620 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
621 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
622 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
623 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
626 tprintf(s->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);
627 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
629 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
630 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
631 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
632 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
633 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
634 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
636 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
637 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
638 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
639 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
640 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
641 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
646 for( dir = 0; dir < 2; dir++ )
647 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, dir);
649 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, 0);
650 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);