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 /* intra: 0 if this loopfilter call is guaranteed to be inter (bS < 4), 1 if it might be intra (bS == 4) */
104 static av_always_inline void filter_mb_edgev(uint8_t *pix, int stride,
106 unsigned int qp, int a, int b,
107 H264Context *h, int intra)
109 const unsigned int index_a = qp + a;
110 const int alpha = alpha_table[index_a];
111 const int beta = beta_table[qp + b];
112 if (alpha ==0 || beta == 0) return;
114 if( bS[0] < 4 || !intra ) {
116 tc[0] = tc0_table[index_a][bS[0]];
117 tc[1] = tc0_table[index_a][bS[1]];
118 tc[2] = tc0_table[index_a][bS[2]];
119 tc[3] = tc0_table[index_a][bS[3]];
120 h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
122 h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
126 static av_always_inline void filter_mb_edgecv(uint8_t *pix, int stride,
128 unsigned int qp, int a, int b,
129 H264Context *h, int intra)
131 const unsigned int index_a = qp + a;
132 const int alpha = alpha_table[index_a];
133 const int beta = beta_table[qp + b];
134 if (alpha ==0 || beta == 0) return;
136 if( bS[0] < 4 || !intra ) {
138 tc[0] = tc0_table[index_a][bS[0]]+1;
139 tc[1] = tc0_table[index_a][bS[1]]+1;
140 tc[2] = tc0_table[index_a][bS[2]]+1;
141 tc[3] = tc0_table[index_a][bS[3]]+1;
142 h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
144 h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
148 static av_always_inline void filter_mb_mbaff_edgev(H264Context *h, uint8_t *pix,
150 const int16_t bS[7], int bsi,
151 int qp, int a, int b,
154 const unsigned int index_a = qp + a;
155 const int alpha = alpha_table[index_a];
156 const int beta = beta_table[qp + b];
157 if (alpha ==0 || beta == 0) return;
159 if( bS[0] < 4 || !intra ) {
161 tc[0] = tc0_table[index_a][bS[0*bsi]];
162 tc[1] = tc0_table[index_a][bS[1*bsi]];
163 tc[2] = tc0_table[index_a][bS[2*bsi]];
164 tc[3] = tc0_table[index_a][bS[3*bsi]];
165 h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
167 h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
171 static av_always_inline void filter_mb_mbaff_edgecv(H264Context *h,
172 uint8_t *pix, int stride,
174 int bsi, int qp, int a,
177 const unsigned int index_a = qp + a;
178 const int alpha = alpha_table[index_a];
179 const int beta = beta_table[qp + b];
180 if (alpha ==0 || beta == 0) return;
182 if( bS[0] < 4 || !intra ) {
184 tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
185 tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
186 tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
187 tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
188 h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
190 h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
194 static av_always_inline void filter_mb_edgeh(uint8_t *pix, int stride,
196 unsigned int qp, int a, int b,
197 H264Context *h, int intra)
199 const unsigned int index_a = qp + a;
200 const int alpha = alpha_table[index_a];
201 const int beta = beta_table[qp + b];
202 if (alpha ==0 || beta == 0) return;
204 if( bS[0] < 4 || !intra ) {
206 tc[0] = tc0_table[index_a][bS[0]];
207 tc[1] = tc0_table[index_a][bS[1]];
208 tc[2] = tc0_table[index_a][bS[2]];
209 tc[3] = tc0_table[index_a][bS[3]];
210 h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
212 h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
216 static av_always_inline void filter_mb_edgech(uint8_t *pix, int stride,
218 unsigned int qp, int a, int b,
219 H264Context *h, int intra)
221 const unsigned int index_a = qp + a;
222 const int alpha = alpha_table[index_a];
223 const int beta = beta_table[qp + b];
224 if (alpha ==0 || beta == 0) return;
226 if( bS[0] < 4 || !intra ) {
228 tc[0] = tc0_table[index_a][bS[0]]+1;
229 tc[1] = tc0_table[index_a][bS[1]]+1;
230 tc[2] = tc0_table[index_a][bS[2]]+1;
231 tc[3] = tc0_table[index_a][bS[3]]+1;
232 h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
234 h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
238 static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,
243 unsigned int linesize,
244 unsigned int uvlinesize,
247 MpegEncContext * const s = &h->s;
248 int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
249 int chroma444 = CHROMA444;
250 int chroma422 = CHROMA422;
252 int mb_xy = h->mb_xy;
253 int left_type= h->left_type[LTOP];
254 int top_type= h->top_type;
256 int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
257 int a = h->slice_alpha_c0_offset - qp_bd_offset;
258 int b = h->slice_beta_offset - qp_bd_offset;
260 int mb_type = s->current_picture.f.mb_type[mb_xy];
261 int qp = s->current_picture.f.qscale_table[mb_xy];
262 int qp0 = s->current_picture.f.qscale_table[mb_xy - 1];
263 int qp1 = s->current_picture.f.qscale_table[h->top_mb_xy];
264 int qpc = get_chroma_qp( h, 0, qp );
265 int qpc0 = get_chroma_qp( h, 0, qp0 );
266 int qpc1 = get_chroma_qp( h, 0, qp1 );
267 qp0 = (qp + qp0 + 1) >> 1;
268 qp1 = (qp + qp1 + 1) >> 1;
269 qpc0 = (qpc + qpc0 + 1) >> 1;
270 qpc1 = (qpc + qpc1 + 1) >> 1;
272 if( IS_INTRA(mb_type) ) {
273 static const int16_t bS4[4] = {4,4,4,4};
274 static const int16_t bS3[4] = {3,3,3,3};
275 const int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
277 filter_mb_edgev( &img_y[4*0<<pixel_shift], linesize, bS4, qp0, a, b, h, 1);
278 if( IS_8x8DCT(mb_type) ) {
279 filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
281 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
283 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
285 filter_mb_edgev( &img_y[4*1<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
286 filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
287 filter_mb_edgev( &img_y[4*3<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
289 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
291 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, a, b, h, 0);
292 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
293 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, a, b, h, 0);
298 filter_mb_edgev( &img_cb[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
299 filter_mb_edgev( &img_cr[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
301 if( IS_8x8DCT(mb_type) ) {
302 filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
303 filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
305 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
306 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
308 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
309 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
311 filter_mb_edgev( &img_cb[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
312 filter_mb_edgev( &img_cr[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
313 filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
314 filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
315 filter_mb_edgev( &img_cb[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
316 filter_mb_edgev( &img_cr[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
318 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
319 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
321 filter_mb_edgeh( &img_cb[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
322 filter_mb_edgeh( &img_cr[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
323 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
324 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
325 filter_mb_edgeh( &img_cb[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
326 filter_mb_edgeh( &img_cr[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
330 filter_mb_edgecv(&img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
331 filter_mb_edgecv(&img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
333 filter_mb_edgecv(&img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
334 filter_mb_edgecv(&img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
336 filter_mb_edgech(&img_cb[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
337 filter_mb_edgech(&img_cr[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
339 filter_mb_edgech(&img_cb[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
340 filter_mb_edgech(&img_cr[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
341 filter_mb_edgech(&img_cb[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
342 filter_mb_edgech(&img_cr[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
343 filter_mb_edgech(&img_cb[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
344 filter_mb_edgech(&img_cr[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
347 filter_mb_edgecv( &img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
348 filter_mb_edgecv( &img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
350 filter_mb_edgecv( &img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
351 filter_mb_edgecv( &img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
353 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
354 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
356 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
357 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
362 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
364 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) {
366 AV_WN64A(bS[0][0], 0x0002000200020002ULL);
367 AV_WN64A(bS[0][2], 0x0002000200020002ULL);
368 AV_WN64A(bS[1][0], 0x0002000200020002ULL);
369 AV_WN64A(bS[1][2], 0x0002000200020002ULL);
371 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;
372 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;
373 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
374 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
375 h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
376 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
378 if( IS_INTRA(left_type) )
379 AV_WN64A(bS[0][0], 0x0004000400040004ULL);
380 if( IS_INTRA(top_type) )
381 AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
383 #define FILTER(hv,dir,edge,intra)\
384 if(AV_RN64A(bS[dir][edge])) { \
385 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 );\
388 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 );\
389 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 );\
390 } else if(!(edge&1)) {\
391 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 );\
392 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 );\
401 } else if( IS_8x8DCT(mb_type) ) {
420 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) {
421 assert(!FRAME_MBAFF);
422 if(!h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
423 ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
428 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift);
431 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1);
433 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0);
438 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
441 v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
442 if(!v && h->ref_cache[0][b_idx]!=-1)
443 v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
444 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
446 if(h->list_count==2){
448 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
449 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
450 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
453 if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
454 h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
457 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
458 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
459 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
460 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
467 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 a, int b, int chroma, int dir) {
468 MpegEncContext * const s = &h->s;
470 int chroma_qp_avg[2];
471 int chroma444 = CHROMA444;
472 int chroma422 = CHROMA422;
473 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
474 const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type;
476 // how often to recheck mv-based bS when iterating between edges
477 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
479 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
480 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
482 // how often to recheck mv-based bS when iterating along each edge
483 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
485 if(mbm_type && !first_vertical_edge_done){
487 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
488 && IS_INTERLACED(mbm_type&~mb_type)
490 // This is a special case in the norm where the filtering must
491 // be done twice (one each of the field) even if we are in a
494 unsigned int tmp_linesize = 2 * linesize;
495 unsigned int tmp_uvlinesize = 2 * uvlinesize;
496 int mbn_xy = mb_xy - 2 * s->mb_stride;
499 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
500 DECLARE_ALIGNED(8, int16_t, bS)[4];
502 if (IS_INTRA(mb_type | s->current_picture.f.mb_type[mbn_xy])) {
503 AV_WN64A(bS, 0x0003000300030003ULL);
505 if (!CABAC && IS_8x8DCT(s->current_picture.f.mb_type[mbn_xy])) {
506 bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+0]);
507 bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+1]);
508 bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+2]);
509 bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+3]);
511 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4;
513 for( i = 0; i < 4; i++ ) {
514 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
518 // Do not use s->qscale as luma quantizer because it has not the same
519 // value in IPCM macroblocks.
520 qp = (s->current_picture.f.qscale_table[mb_xy] + s->current_picture.f.qscale_table[mbn_xy] + 1) >> 1;
521 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);
522 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
523 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, a, b, h, 0 );
524 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1;
525 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1;
528 filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
529 filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
531 filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
532 filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
537 DECLARE_ALIGNED(8, int16_t, bS)[4];
540 if( IS_INTRA(mb_type|mbm_type)) {
541 AV_WN64A(bS, 0x0003000300030003ULL);
542 if ( (!IS_INTERLACED(mb_type|mbm_type))
543 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
545 AV_WN64A(bS, 0x0004000400040004ULL);
550 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
551 AV_WN64A(bS, 0x0001000100010001ULL);
554 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
556 int bn_idx= b_idx - (dir ? 8:1);
558 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
564 for( i = 0; i < 4; i++ ) {
565 int x = dir == 0 ? 0 : i;
566 int y = dir == 0 ? i : 0;
567 int b_idx= 8 + 4 + x + 8*y;
568 int bn_idx= b_idx - (dir ? 8:1);
570 if( h->non_zero_count_cache[b_idx] |
571 h->non_zero_count_cache[bn_idx] ) {
576 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
582 // Do not use s->qscale as luma quantizer because it has not the same
583 // value in IPCM macroblocks.
584 if(bS[0]+bS[1]+bS[2]+bS[3]){
585 qp = (s->current_picture.f.qscale_table[mb_xy] + s->current_picture.f.qscale_table[mbm_xy] + 1) >> 1;
586 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);
587 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1;
588 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1;
590 filter_mb_edgev( &img_y[0], linesize, bS, qp, a, b, h, 1 );
593 filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
594 filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
596 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
597 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
601 filter_mb_edgeh( &img_y[0], linesize, bS, qp, a, b, h, 1 );
604 filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
605 filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
607 filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
608 filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
617 for( edge = 1; edge < edges; edge++ ) {
618 DECLARE_ALIGNED(8, int16_t, bS)[4];
620 const int deblock_edge = !IS_8x8DCT(mb_type & (edge<<24)); // (edge&1) && IS_8x8DCT(mb_type)
622 if (!deblock_edge && (!chroma422 || dir == 0))
625 if( IS_INTRA(mb_type)) {
626 AV_WN64A(bS, 0x0003000300030003ULL);
631 if( edge & mask_edge ) {
635 else if( mask_par0 ) {
636 int b_idx= 8 + 4 + edge * (dir ? 8:1);
637 int bn_idx= b_idx - (dir ? 8:1);
639 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
645 for( i = 0; i < 4; i++ ) {
646 int x = dir == 0 ? edge : i;
647 int y = dir == 0 ? i : edge;
648 int b_idx= 8 + 4 + x + 8*y;
649 int bn_idx= b_idx - (dir ? 8:1);
651 if( h->non_zero_count_cache[b_idx] |
652 h->non_zero_count_cache[bn_idx] ) {
657 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
661 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
666 // Do not use s->qscale as luma quantizer because it has not the same
667 // value in IPCM macroblocks.
668 qp = s->current_picture.f.qscale_table[mb_xy];
669 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);
671 filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, a, b, h, 0 );
674 filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
675 filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
676 } else if( (edge&1) == 0 ) {
677 filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
678 filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
684 filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);
686 filter_mb_edgech(&img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
687 filter_mb_edgech(&img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
690 filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);
693 filter_mb_edgeh (&img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
694 filter_mb_edgeh (&img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
695 } else if ((edge&1) == 0) {
696 filter_mb_edgech(&img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
697 filter_mb_edgech(&img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
705 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) {
706 MpegEncContext * const s = &h->s;
707 const int mb_xy= mb_x + mb_y*s->mb_stride;
708 const int mb_type = s->current_picture.f.mb_type[mb_xy];
709 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
710 int first_vertical_edge_done = 0;
712 int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
713 int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
714 int a = h->slice_alpha_c0_offset - qp_bd_offset;
715 int b = h->slice_beta_offset - qp_bd_offset;
718 // and current and left pair do not have the same interlaced type
719 && IS_INTERLACED(mb_type^h->left_type[LTOP])
720 // and left mb is in available to us
721 && h->left_type[LTOP]) {
722 /* First vertical edge is different in MBAFF frames
723 * There are 8 different bS to compute and 2 different Qp
725 DECLARE_ALIGNED(8, int16_t, bS)[8];
729 int mb_qp, mbn0_qp, mbn1_qp;
731 first_vertical_edge_done = 1;
733 if( IS_INTRA(mb_type) ) {
734 AV_WN64A(&bS[0], 0x0004000400040004ULL);
735 AV_WN64A(&bS[4], 0x0004000400040004ULL);
737 static const uint8_t offset[2][2][8]={
739 {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},
740 {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},
742 {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
743 {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
746 const uint8_t *off= offset[MB_FIELD][mb_y&1];
747 for( i = 0; i < 8; i++ ) {
748 int j= MB_FIELD ? i>>2 : i&1;
749 int mbn_xy = h->left_mb_xy[LEFT(j)];
750 int mbn_type= h->left_type[LEFT(j)];
752 if( IS_INTRA( mbn_type ) )
755 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
756 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
757 (h->cbp_table[mbn_xy] & (((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2) << 12))
759 h->non_zero_count[mbn_xy][ off[i] ]));
764 mb_qp = s->current_picture.f.qscale_table[mb_xy];
765 mbn0_qp = s->current_picture.f.qscale_table[h->left_mb_xy[0]];
766 mbn1_qp = s->current_picture.f.qscale_table[h->left_mb_xy[1]];
767 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
768 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
769 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
770 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
771 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
772 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
773 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
774 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
775 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
776 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
779 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);
780 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
782 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0], a, b, 1 );
783 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1], a, b, 1 );
786 filter_mb_mbaff_edgev ( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 );
787 filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
788 filter_mb_mbaff_edgev ( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 );
789 filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
790 } else if (CHROMA422) {
791 filter_mb_mbaff_edgecv(h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1);
792 filter_mb_mbaff_edgecv(h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1);
793 filter_mb_mbaff_edgecv(h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1);
794 filter_mb_mbaff_edgecv(h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1);
796 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 );
797 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
798 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 );
799 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
803 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0], a, b, 1 );
804 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1], a, b, 1 );
807 filter_mb_mbaff_edgev ( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 );
808 filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
809 filter_mb_mbaff_edgev ( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 );
810 filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
812 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 );
813 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
814 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 );
815 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
822 for( dir = 0; dir < 2; dir++ )
823 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, a, b, chroma, dir);
825 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, a, b, chroma, 0);
826 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, a, b, chroma, 1);