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 int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
105 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
106 const int alpha = alpha_table[index_a];
107 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
108 if (alpha ==0 || beta == 0) return;
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);
118 h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
121 static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
122 const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
123 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
124 const int alpha = alpha_table[index_a];
125 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
126 if (alpha ==0 || beta == 0) return;
130 tc[0] = tc0_table[index_a][bS[0]]+1;
131 tc[1] = tc0_table[index_a][bS[1]]+1;
132 tc[2] = tc0_table[index_a][bS[2]]+1;
133 tc[3] = tc0_table[index_a][bS[3]]+1;
134 h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
136 h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
140 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
141 const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
142 int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
143 int alpha = alpha_table[index_a];
144 int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
145 if (alpha ==0 || beta == 0) return;
149 tc[0] = tc0_table[index_a][bS[0*bsi]];
150 tc[1] = tc0_table[index_a][bS[1*bsi]];
151 tc[2] = tc0_table[index_a][bS[2*bsi]];
152 tc[3] = tc0_table[index_a][bS[3*bsi]];
153 h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
155 h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
158 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[7], int bsi, int qp ) {
159 const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
160 int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
161 int alpha = alpha_table[index_a];
162 int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
163 if (alpha ==0 || beta == 0) return;
167 tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
168 tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
169 tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
170 tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
171 h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
173 h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
177 static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
178 const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
179 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
180 const int alpha = alpha_table[index_a];
181 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
182 if (alpha ==0 || beta == 0) return;
186 tc[0] = tc0_table[index_a][bS[0]];
187 tc[1] = tc0_table[index_a][bS[1]];
188 tc[2] = tc0_table[index_a][bS[2]];
189 tc[3] = tc0_table[index_a][bS[3]];
190 h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
192 h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
196 static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
197 const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
198 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
199 const int alpha = alpha_table[index_a];
200 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
201 if (alpha ==0 || beta == 0) return;
205 tc[0] = tc0_table[index_a][bS[0]]+1;
206 tc[1] = tc0_table[index_a][bS[1]]+1;
207 tc[2] = tc0_table[index_a][bS[2]]+1;
208 tc[3] = tc0_table[index_a][bS[3]]+1;
209 h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
211 h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
215 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) {
216 MpegEncContext * const s = &h->s;
218 int mb_type, left_type;
219 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
220 int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
221 int chroma444 = CHROMA444;
225 if(!h->top_type || !h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
226 ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
229 assert(!FRAME_MBAFF);
230 left_type= h->left_type[0];
232 mb_type = s->current_picture.mb_type[mb_xy];
233 qp = s->current_picture.qscale_table[mb_xy];
234 qp0 = s->current_picture.qscale_table[mb_xy-1];
235 qp1 = s->current_picture.qscale_table[h->top_mb_xy];
236 qpc = get_chroma_qp( h, 0, qp );
237 qpc0 = get_chroma_qp( h, 0, qp0 );
238 qpc1 = get_chroma_qp( h, 0, qp1 );
239 qp0 = (qp + qp0 + 1) >> 1;
240 qp1 = (qp + qp1 + 1) >> 1;
241 qpc0 = (qpc + qpc0 + 1) >> 1;
242 qpc1 = (qpc + qpc1 + 1) >> 1;
243 qp_thresh = 15+52 - h->slice_alpha_c0_offset;
244 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
245 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
248 if( IS_INTRA(mb_type) ) {
249 int16_t bS4[4] = {4,4,4,4};
250 int16_t bS3[4] = {3,3,3,3};
251 int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
253 filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
254 if( IS_8x8DCT(mb_type) ) {
255 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
256 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
257 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
259 filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
260 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
261 filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
262 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
263 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
264 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
265 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
270 filter_mb_edgev( &img_cb[4*0], linesize, bS4, qpc0, h);
271 filter_mb_edgev( &img_cr[4*0], linesize, bS4, qpc0, h);
273 if( IS_8x8DCT(mb_type) ) {
274 filter_mb_edgev( &img_cb[4*2], linesize, bS3, qpc, h);
275 filter_mb_edgev( &img_cr[4*2], linesize, bS3, qpc, h);
276 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, h);
277 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, h);
278 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, h);
279 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, h);
281 filter_mb_edgev( &img_cb[4*1], linesize, bS3, qpc, h);
282 filter_mb_edgev( &img_cr[4*1], linesize, bS3, qpc, h);
283 filter_mb_edgev( &img_cb[4*2], linesize, bS3, qpc, h);
284 filter_mb_edgev( &img_cr[4*2], linesize, bS3, qpc, h);
285 filter_mb_edgev( &img_cb[4*3], linesize, bS3, qpc, h);
286 filter_mb_edgev( &img_cr[4*3], linesize, bS3, qpc, h);
287 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, h);
288 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, h);
289 filter_mb_edgeh( &img_cb[4*1*linesize], linesize, bS3, qpc, h);
290 filter_mb_edgeh( &img_cr[4*1*linesize], linesize, bS3, qpc, h);
291 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, h);
292 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, h);
293 filter_mb_edgeh( &img_cb[4*3*linesize], linesize, bS3, qpc, h);
294 filter_mb_edgeh( &img_cr[4*3*linesize], linesize, bS3, qpc, h);
298 filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
299 filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
301 filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
302 filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
303 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
304 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
305 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
306 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
311 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
313 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
315 AV_WN64A(bS[0][0], 0x0002000200020002ULL);
316 AV_WN64A(bS[0][2], 0x0002000200020002ULL);
317 AV_WN64A(bS[1][0], 0x0002000200020002ULL);
318 AV_WN64A(bS[1][2], 0x0002000200020002ULL);
320 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;
321 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;
322 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
323 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
324 h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
325 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
327 if( IS_INTRA(left_type) )
328 AV_WN64A(bS[0][0], 0x0004000400040004ULL);
329 if( IS_INTRA(h->top_type) )
330 AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
332 #define FILTER(hv,dir,edge)\
333 if(AV_RN64A(bS[dir][edge])) { \
334 filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
337 filter_mb_edge##hv( &img_cb[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
338 filter_mb_edge##hv( &img_cr[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
339 } else if(!(edge&1)) {\
340 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
341 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
349 } else if( IS_8x8DCT(mb_type) ) {
366 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
369 v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
370 if(!v && h->ref_cache[0][b_idx]!=-1)
371 v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
372 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
374 if(h->list_count==2){
376 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
377 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
378 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
381 if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
382 h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
385 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
386 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
387 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
388 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
395 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 chroma, int chroma444, int dir) {
396 MpegEncContext * const s = &h->s;
398 int chroma_qp_avg[2];
399 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
400 const int mbm_type = dir == 0 ? h->left_type[0] : h->top_type;
402 // how often to recheck mv-based bS when iterating between edges
403 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
405 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
406 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
408 // how often to recheck mv-based bS when iterating along each edge
409 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
411 if(mbm_type && !first_vertical_edge_done){
413 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
414 && IS_INTERLACED(mbm_type&~mb_type)
416 // This is a special case in the norm where the filtering must
417 // be done twice (one each of the field) even if we are in a
420 unsigned int tmp_linesize = 2 * linesize;
421 unsigned int tmp_uvlinesize = 2 * uvlinesize;
422 int mbn_xy = mb_xy - 2 * s->mb_stride;
425 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
426 DECLARE_ALIGNED(8, int16_t, bS)[4];
428 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
429 AV_WN64A(bS, 0x0003000300030003ULL);
431 if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
432 bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+0]);
433 bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+1]);
434 bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+2]);
435 bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+3]);
437 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4;
439 for( i = 0; i < 4; i++ ) {
440 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
444 // Do not use s->qscale as luma quantizer because it has not the same
445 // value in IPCM macroblocks.
446 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
447 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);
448 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
449 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
450 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
451 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
454 filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
455 filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
457 filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
458 filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
463 DECLARE_ALIGNED(8, int16_t, bS)[4];
466 if( IS_INTRA(mb_type|mbm_type)) {
467 AV_WN64A(bS, 0x0003000300030003ULL);
468 if ( (!IS_INTERLACED(mb_type|mbm_type))
469 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
471 AV_WN64A(bS, 0x0004000400040004ULL);
476 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
477 AV_WN64A(bS, 0x0001000100010001ULL);
480 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
482 int bn_idx= b_idx - (dir ? 8:1);
484 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
490 for( i = 0; i < 4; i++ ) {
491 int x = dir == 0 ? 0 : i;
492 int y = dir == 0 ? i : 0;
493 int b_idx= 8 + 4 + x + 8*y;
494 int bn_idx= b_idx - (dir ? 8:1);
496 if( h->non_zero_count_cache[b_idx] |
497 h->non_zero_count_cache[bn_idx] ) {
502 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
508 // Do not use s->qscale as luma quantizer because it has not the same
509 // value in IPCM macroblocks.
510 if(bS[0]+bS[1]+bS[2]+bS[3]){
511 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbm_xy] + 1 ) >> 1;
512 //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]);
513 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);
514 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
515 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
516 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
518 filter_mb_edgev( &img_y[0], linesize, bS, qp, h );
521 filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
522 filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
524 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
525 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
529 filter_mb_edgeh( &img_y[0], linesize, bS, qp, h );
532 filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
533 filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
535 filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
536 filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
545 for( edge = 1; edge < edges; edge++ ) {
546 DECLARE_ALIGNED(8, int16_t, bS)[4];
549 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
552 if( IS_INTRA(mb_type)) {
553 AV_WN64A(bS, 0x0003000300030003ULL);
558 if( edge & mask_edge ) {
562 else if( mask_par0 ) {
563 int b_idx= 8 + 4 + edge * (dir ? 8:1);
564 int bn_idx= b_idx - (dir ? 8:1);
566 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
572 for( i = 0; i < 4; i++ ) {
573 int x = dir == 0 ? edge : i;
574 int y = dir == 0 ? i : edge;
575 int b_idx= 8 + 4 + x + 8*y;
576 int bn_idx= b_idx - (dir ? 8:1);
578 if( h->non_zero_count_cache[b_idx] |
579 h->non_zero_count_cache[bn_idx] ) {
584 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
588 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
593 // Do not use s->qscale as luma quantizer because it has not the same
594 // value in IPCM macroblocks.
595 qp = s->current_picture.qscale_table[mb_xy];
596 //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]);
597 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);
598 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
600 filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, h );
603 filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
604 filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
605 } else if( (edge&1) == 0 ) {
606 filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
607 filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
611 filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
614 filter_mb_edgeh ( &img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
615 filter_mb_edgeh ( &img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
616 } else if( (edge&1) == 0 ) {
617 filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
618 filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
625 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) {
626 MpegEncContext * const s = &h->s;
627 const int mb_xy= mb_x + mb_y*s->mb_stride;
628 const int mb_type = s->current_picture.mb_type[mb_xy];
629 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
630 int first_vertical_edge_done = 0;
632 int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
635 // and current and left pair do not have the same interlaced type
636 && IS_INTERLACED(mb_type^h->left_type[0])
637 // and left mb is in available to us
638 && h->left_type[0]) {
639 /* First vertical edge is different in MBAFF frames
640 * There are 8 different bS to compute and 2 different Qp
642 DECLARE_ALIGNED(8, int16_t, bS)[8];
646 int mb_qp, mbn0_qp, mbn1_qp;
648 first_vertical_edge_done = 1;
650 if( IS_INTRA(mb_type) ) {
651 AV_WN64A(&bS[0], 0x0004000400040004ULL);
652 AV_WN64A(&bS[4], 0x0004000400040004ULL);
654 static const uint8_t offset[2][2][8]={
656 {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},
657 {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},
659 {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
660 {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
663 const uint8_t *off= offset[MB_FIELD][mb_y&1];
664 for( i = 0; i < 8; i++ ) {
665 int j= MB_FIELD ? i>>2 : i&1;
666 int mbn_xy = h->left_mb_xy[j];
667 int mbn_type= h->left_type[j];
669 if( IS_INTRA( mbn_type ) )
672 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
673 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
674 (h->cbp_table[mbn_xy] & (((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2) << 12))
676 h->non_zero_count[mbn_xy][ off[i] ]));
681 mb_qp = s->current_picture.qscale_table[mb_xy];
682 mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]];
683 mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]];
684 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
685 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
686 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
687 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
688 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
689 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
690 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
691 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
692 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
693 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
696 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);
697 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
699 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
700 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
703 filter_mb_mbaff_edgev ( h, img_cb, uvlinesize, bS , 1, bqp[0] );
704 filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
705 filter_mb_mbaff_edgev ( h, img_cr, uvlinesize, bS , 1, rqp[0] );
706 filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
708 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
709 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
710 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
711 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
715 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
716 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
719 filter_mb_mbaff_edgev ( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
720 filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
721 filter_mb_mbaff_edgev ( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
722 filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
724 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
725 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
726 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
727 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
734 for( dir = 0; dir < 2; dir++ )
735 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, chroma, CHROMA444, dir);
737 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, chroma, CHROMA444, 0);
738 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, chroma, CHROMA444, 1);