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/internal.h"
29 #include "libavutil/intreadwrite.h"
34 #include "mpegutils.h"
35 #include "rectangle.h"
39 /* Deblocking filter (p153) */
40 static const uint8_t alpha_table[52*3] = {
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, 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,
46 0, 0, 0, 0, 0, 0, 4, 4, 5, 6,
47 7, 8, 9, 10, 12, 13, 15, 17, 20, 22,
48 25, 28, 32, 36, 40, 45, 50, 56, 63, 71,
49 80, 90,101,113,127,144,162,182,203,226,
51 255,255,255,255,255,255,255,255,255,255,255,255,255,
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,
56 static const uint8_t beta_table[52*3] = {
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, 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,
62 0, 0, 0, 0, 0, 0, 2, 2, 2, 3,
63 3, 3, 3, 4, 4, 4, 6, 6, 7, 7,
64 8, 8, 9, 9, 10, 10, 11, 11, 12, 12,
65 13, 13, 14, 14, 15, 15, 16, 16, 17, 17,
67 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
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,
72 static const uint8_t tc0_table[52*3][4] = {
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 }, {-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 },
82 {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-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, 1 },
85 {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },
86 {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },
87 {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },
88 {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },
89 {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },
90 {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-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 }, {-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 },
102 /* intra: 0 if this loopfilter call is guaranteed to be inter (bS < 4), 1 if it might be intra (bS == 4) */
103 static av_always_inline void filter_mb_edgev(uint8_t *pix, int stride,
105 unsigned int qp, int a, int b,
106 H264Context *h, int intra)
108 const unsigned int index_a = qp + a;
109 const int alpha = alpha_table[index_a];
110 const int beta = beta_table[qp + b];
111 if (alpha ==0 || beta == 0) return;
113 if( bS[0] < 4 || !intra ) {
115 tc[0] = tc0_table[index_a][bS[0]];
116 tc[1] = tc0_table[index_a][bS[1]];
117 tc[2] = tc0_table[index_a][bS[2]];
118 tc[3] = tc0_table[index_a][bS[3]];
119 h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
121 h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
125 static av_always_inline void filter_mb_edgecv(uint8_t *pix, int stride,
127 unsigned int qp, int a, int b,
128 H264Context *h, int intra)
130 const unsigned int index_a = qp + a;
131 const int alpha = alpha_table[index_a];
132 const int beta = beta_table[qp + b];
133 if (alpha ==0 || beta == 0) return;
135 if( bS[0] < 4 || !intra ) {
137 tc[0] = tc0_table[index_a][bS[0]]+1;
138 tc[1] = tc0_table[index_a][bS[1]]+1;
139 tc[2] = tc0_table[index_a][bS[2]]+1;
140 tc[3] = tc0_table[index_a][bS[3]]+1;
141 h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
143 h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
147 static av_always_inline void filter_mb_mbaff_edgev(H264Context *h, uint8_t *pix,
149 const int16_t bS[7], int bsi,
150 int qp, int a, int b,
153 const unsigned int index_a = qp + a;
154 const int alpha = alpha_table[index_a];
155 const int beta = beta_table[qp + b];
156 if (alpha ==0 || beta == 0) return;
158 if( bS[0] < 4 || !intra ) {
160 tc[0] = tc0_table[index_a][bS[0*bsi]];
161 tc[1] = tc0_table[index_a][bS[1*bsi]];
162 tc[2] = tc0_table[index_a][bS[2*bsi]];
163 tc[3] = tc0_table[index_a][bS[3*bsi]];
164 h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
166 h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
170 static av_always_inline void filter_mb_mbaff_edgecv(H264Context *h,
171 uint8_t *pix, int stride,
173 int bsi, int qp, int a,
176 const unsigned int index_a = qp + a;
177 const int alpha = alpha_table[index_a];
178 const int beta = beta_table[qp + b];
179 if (alpha ==0 || beta == 0) return;
181 if( bS[0] < 4 || !intra ) {
183 tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
184 tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
185 tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
186 tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
187 h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
189 h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
193 static av_always_inline void filter_mb_edgeh(uint8_t *pix, int stride,
195 unsigned int qp, int a, int b,
196 H264Context *h, int intra)
198 const unsigned int index_a = qp + a;
199 const int alpha = alpha_table[index_a];
200 const int beta = beta_table[qp + b];
201 if (alpha ==0 || beta == 0) return;
203 if( bS[0] < 4 || !intra ) {
205 tc[0] = tc0_table[index_a][bS[0]];
206 tc[1] = tc0_table[index_a][bS[1]];
207 tc[2] = tc0_table[index_a][bS[2]];
208 tc[3] = tc0_table[index_a][bS[3]];
209 h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
211 h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
215 static av_always_inline void filter_mb_edgech(uint8_t *pix, int stride,
217 unsigned int qp, int a, int b,
218 H264Context *h, int intra)
220 const unsigned int index_a = qp + a;
221 const int alpha = alpha_table[index_a];
222 const int beta = beta_table[qp + b];
223 if (alpha ==0 || beta == 0) return;
225 if( bS[0] < 4 || !intra ) {
227 tc[0] = tc0_table[index_a][bS[0]]+1;
228 tc[1] = tc0_table[index_a][bS[1]]+1;
229 tc[2] = tc0_table[index_a][bS[2]]+1;
230 tc[3] = tc0_table[index_a][bS[3]]+1;
231 h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
233 h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
237 static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,
242 unsigned int linesize,
243 unsigned int uvlinesize,
246 int chroma = !(CONFIG_GRAY && (h->flags&CODEC_FLAG_GRAY));
247 int chroma444 = CHROMA444(h);
248 int chroma422 = CHROMA422(h);
250 int mb_xy = h->mb_xy;
251 int left_type= h->left_type[LTOP];
252 int top_type= h->top_type;
254 int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
255 int a = 52 + h->slice_alpha_c0_offset - qp_bd_offset;
256 int b = 52 + h->slice_beta_offset - qp_bd_offset;
258 int mb_type = h->cur_pic.mb_type[mb_xy];
259 int qp = h->cur_pic.qscale_table[mb_xy];
260 int qp0 = h->cur_pic.qscale_table[mb_xy - 1];
261 int qp1 = h->cur_pic.qscale_table[h->top_mb_xy];
262 int qpc = get_chroma_qp( h, 0, qp );
263 int qpc0 = get_chroma_qp( h, 0, qp0 );
264 int qpc1 = get_chroma_qp( h, 0, qp1 );
265 qp0 = (qp + qp0 + 1) >> 1;
266 qp1 = (qp + qp1 + 1) >> 1;
267 qpc0 = (qpc + qpc0 + 1) >> 1;
268 qpc1 = (qpc + qpc1 + 1) >> 1;
270 if( IS_INTRA(mb_type) ) {
271 static const int16_t bS4[4] = {4,4,4,4};
272 static const int16_t bS3[4] = {3,3,3,3};
273 const int16_t *bSH = FIELD_PICTURE(h) ? bS3 : bS4;
275 filter_mb_edgev( &img_y[4*0<<pixel_shift], linesize, bS4, qp0, a, b, h, 1);
276 if( IS_8x8DCT(mb_type) ) {
277 filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
279 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
281 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
283 filter_mb_edgev( &img_y[4*1<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
284 filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
285 filter_mb_edgev( &img_y[4*3<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
287 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
289 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, a, b, h, 0);
290 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
291 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, a, b, h, 0);
296 filter_mb_edgev( &img_cb[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
297 filter_mb_edgev( &img_cr[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
299 if( IS_8x8DCT(mb_type) ) {
300 filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
301 filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
303 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
304 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
306 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
307 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
309 filter_mb_edgev( &img_cb[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
310 filter_mb_edgev( &img_cr[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
311 filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
312 filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
313 filter_mb_edgev( &img_cb[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
314 filter_mb_edgev( &img_cr[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
316 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
317 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
319 filter_mb_edgeh( &img_cb[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
320 filter_mb_edgeh( &img_cr[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
321 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
322 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
323 filter_mb_edgeh( &img_cb[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
324 filter_mb_edgeh( &img_cr[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
328 filter_mb_edgecv(&img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
329 filter_mb_edgecv(&img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
331 filter_mb_edgecv(&img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
332 filter_mb_edgecv(&img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
334 filter_mb_edgech(&img_cb[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
335 filter_mb_edgech(&img_cr[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
337 filter_mb_edgech(&img_cb[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
338 filter_mb_edgech(&img_cr[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
339 filter_mb_edgech(&img_cb[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
340 filter_mb_edgech(&img_cr[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
341 filter_mb_edgech(&img_cb[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
342 filter_mb_edgech(&img_cr[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
345 filter_mb_edgecv( &img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
346 filter_mb_edgecv( &img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
348 filter_mb_edgecv( &img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
349 filter_mb_edgecv( &img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
351 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
352 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
354 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
355 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
360 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
362 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) {
364 AV_WN64A(bS[0][0], 0x0002000200020002ULL);
365 AV_WN64A(bS[0][2], 0x0002000200020002ULL);
366 AV_WN64A(bS[1][0], 0x0002000200020002ULL);
367 AV_WN64A(bS[1][2], 0x0002000200020002ULL);
369 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;
370 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;
371 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
372 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
373 h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
374 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE(h));
376 if( IS_INTRA(left_type) )
377 AV_WN64A(bS[0][0], 0x0004000400040004ULL);
378 if( IS_INTRA(top_type) )
379 AV_WN64A(bS[1][0], FIELD_PICTURE(h) ? 0x0003000300030003ULL : 0x0004000400040004ULL);
381 #define FILTER(hv,dir,edge,intra)\
382 if(AV_RN64A(bS[dir][edge])) { \
383 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 );\
386 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 );\
387 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 );\
388 } else if(!(edge&1)) {\
389 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 );\
390 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 );\
399 } else if( IS_8x8DCT(mb_type) ) {
418 void ff_h264_filter_mb_fast(H264Context *h, H264SliceContext *sl,
419 int mb_x, int mb_y, uint8_t *img_y,
420 uint8_t *img_cb, uint8_t *img_cr,
421 unsigned int linesize, unsigned int uvlinesize)
423 assert(!FRAME_MBAFF(h));
424 if(!h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
425 ff_h264_filter_mb(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
430 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift);
433 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1);
435 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0);
440 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
443 v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
444 if(!v && h->ref_cache[0][b_idx]!=-1)
445 v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
446 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
448 if(h->list_count==2){
450 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
451 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
452 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
455 if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
456 h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
459 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
460 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
461 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
462 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
469 static av_always_inline void filter_mb_dir(H264Context *h, H264SliceContext *sl,
471 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
472 unsigned int linesize, unsigned int uvlinesize,
473 int mb_xy, int mb_type, int mvy_limit,
474 int first_vertical_edge_done, int a, int b,
478 int chroma_qp_avg[2];
479 int chroma444 = CHROMA444(h);
480 int chroma422 = CHROMA422(h);
481 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
482 const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type;
484 // how often to recheck mv-based bS when iterating between edges
485 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
487 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
488 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
490 // how often to recheck mv-based bS when iterating along each edge
491 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
493 if(mbm_type && !first_vertical_edge_done){
495 if (FRAME_MBAFF(h) && (dir == 1) && ((mb_y&1) == 0)
496 && IS_INTERLACED(mbm_type&~mb_type)
498 // This is a special case in the norm where the filtering must
499 // be done twice (one each of the field) even if we are in a
502 unsigned int tmp_linesize = 2 * linesize;
503 unsigned int tmp_uvlinesize = 2 * uvlinesize;
504 int mbn_xy = mb_xy - 2 * h->mb_stride;
507 for(j=0; j<2; j++, mbn_xy += h->mb_stride){
508 DECLARE_ALIGNED(8, int16_t, bS)[4];
510 if (IS_INTRA(mb_type | h->cur_pic.mb_type[mbn_xy])) {
511 AV_WN64A(bS, 0x0003000300030003ULL);
513 if (!CABAC(h) && IS_8x8DCT(h->cur_pic.mb_type[mbn_xy])) {
514 bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+0]);
515 bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+1]);
516 bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+2]);
517 bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+3]);
519 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4;
521 for( i = 0; i < 4; i++ ) {
522 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
526 // Do not use s->qscale as luma quantizer because it has not the same
527 // value in IPCM macroblocks.
528 qp = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbn_xy] + 1) >> 1;
529 tprintf(h->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);
530 { int i; for (i = 0; i < 4; i++) tprintf(h->avctx, " bS[%d]:%d", i, bS[i]); tprintf(h->avctx, "\n"); }
531 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, a, b, h, 0 );
532 chroma_qp_avg[0] = (sl->chroma_qp[0] + get_chroma_qp(h, 0, h->cur_pic.qscale_table[mbn_xy]) + 1) >> 1;
533 chroma_qp_avg[1] = (sl->chroma_qp[1] + get_chroma_qp(h, 1, h->cur_pic.qscale_table[mbn_xy]) + 1) >> 1;
536 filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
537 filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
539 filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
540 filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
545 DECLARE_ALIGNED(8, int16_t, bS)[4];
548 if( IS_INTRA(mb_type|mbm_type)) {
549 AV_WN64A(bS, 0x0003000300030003ULL);
550 if ( (!IS_INTERLACED(mb_type|mbm_type))
551 || ((FRAME_MBAFF(h) || (h->picture_structure != PICT_FRAME)) && (dir == 0))
553 AV_WN64A(bS, 0x0004000400040004ULL);
558 if( dir && FRAME_MBAFF(h) && IS_INTERLACED(mb_type ^ mbm_type)) {
559 AV_WN64A(bS, 0x0001000100010001ULL);
562 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
564 int bn_idx= b_idx - (dir ? 8:1);
566 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
572 for( i = 0; i < 4; i++ ) {
573 int x = dir == 0 ? 0 : i;
574 int y = dir == 0 ? i : 0;
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);
590 // Do not use s->qscale as luma quantizer because it has not the same
591 // value in IPCM macroblocks.
592 if(bS[0]+bS[1]+bS[2]+bS[3]){
593 qp = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbm_xy] + 1) >> 1;
594 tprintf(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 chroma_qp_avg[0] = (sl->chroma_qp[0] + get_chroma_qp(h, 0, h->cur_pic.qscale_table[mbm_xy]) + 1) >> 1;
596 chroma_qp_avg[1] = (sl->chroma_qp[1] + get_chroma_qp(h, 1, h->cur_pic.qscale_table[mbm_xy]) + 1) >> 1;
598 filter_mb_edgev( &img_y[0], linesize, bS, qp, a, b, h, 1 );
601 filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
602 filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
604 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
605 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
609 filter_mb_edgeh( &img_y[0], linesize, bS, qp, a, b, h, 1 );
612 filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
613 filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
615 filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
616 filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
625 for( edge = 1; edge < edges; edge++ ) {
626 DECLARE_ALIGNED(8, int16_t, bS)[4];
628 const int deblock_edge = !IS_8x8DCT(mb_type & (edge<<24)); // (edge&1) && IS_8x8DCT(mb_type)
630 if (!deblock_edge && (!chroma422 || dir == 0))
633 if( IS_INTRA(mb_type)) {
634 AV_WN64A(bS, 0x0003000300030003ULL);
639 if( edge & mask_edge ) {
643 else if( mask_par0 ) {
644 int b_idx= 8 + 4 + edge * (dir ? 8:1);
645 int bn_idx= b_idx - (dir ? 8:1);
647 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
653 for( i = 0; i < 4; i++ ) {
654 int x = dir == 0 ? edge : i;
655 int y = dir == 0 ? i : edge;
656 int b_idx= 8 + 4 + x + 8*y;
657 int bn_idx= b_idx - (dir ? 8:1);
659 if( h->non_zero_count_cache[b_idx] |
660 h->non_zero_count_cache[bn_idx] ) {
665 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
669 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
674 // Do not use s->qscale as luma quantizer because it has not the same
675 // value in IPCM macroblocks.
676 qp = h->cur_pic.qscale_table[mb_xy];
677 tprintf(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);
679 filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, a, b, h, 0 );
682 filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
683 filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
684 } else if( (edge&1) == 0 ) {
685 filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
686 filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
692 filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);
694 filter_mb_edgech(&img_cb[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
695 filter_mb_edgech(&img_cr[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
698 filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);
701 filter_mb_edgeh (&img_cb[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
702 filter_mb_edgeh (&img_cr[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
703 } else if ((edge&1) == 0) {
704 filter_mb_edgech(&img_cb[2*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);
705 filter_mb_edgech(&img_cr[2*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);
713 void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl,
715 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
716 unsigned int linesize, unsigned int uvlinesize)
718 const int mb_xy= mb_x + mb_y*h->mb_stride;
719 const int mb_type = h->cur_pic.mb_type[mb_xy];
720 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
721 int first_vertical_edge_done = 0;
722 int chroma = !(CONFIG_GRAY && (h->flags&CODEC_FLAG_GRAY));
723 int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
724 int a = 52 + h->slice_alpha_c0_offset - qp_bd_offset;
725 int b = 52 + h->slice_beta_offset - qp_bd_offset;
728 // and current and left pair do not have the same interlaced type
729 && IS_INTERLACED(mb_type^h->left_type[LTOP])
730 // and left mb is in available to us
731 && h->left_type[LTOP]) {
732 /* First vertical edge is different in MBAFF frames
733 * There are 8 different bS to compute and 2 different Qp
735 DECLARE_ALIGNED(8, int16_t, bS)[8];
739 int mb_qp, mbn0_qp, mbn1_qp;
741 first_vertical_edge_done = 1;
743 if( IS_INTRA(mb_type) ) {
744 AV_WN64A(&bS[0], 0x0004000400040004ULL);
745 AV_WN64A(&bS[4], 0x0004000400040004ULL);
747 static const uint8_t offset[2][2][8]={
749 {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},
750 {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},
752 {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
753 {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 const uint8_t *off= offset[MB_FIELD(h)][mb_y&1];
757 for( i = 0; i < 8; i++ ) {
758 int j= MB_FIELD(h) ? i>>2 : i&1;
759 int mbn_xy = h->left_mb_xy[LEFT(j)];
760 int mbn_type= h->left_type[LEFT(j)];
762 if( IS_INTRA( mbn_type ) )
765 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
766 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
767 (h->cbp_table[mbn_xy] & (((MB_FIELD(h) ? (i&2) : (mb_y&1)) ? 8 : 2) << 12))
769 h->non_zero_count[mbn_xy][ off[i] ]));
774 mb_qp = h->cur_pic.qscale_table[mb_xy];
775 mbn0_qp = h->cur_pic.qscale_table[h->left_mb_xy[0]];
776 mbn1_qp = h->cur_pic.qscale_table[h->left_mb_xy[1]];
777 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
778 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
779 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
780 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
781 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
782 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
783 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
784 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
785 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
786 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
789 tprintf(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);
790 { int i; for (i = 0; i < 8; i++) tprintf(h->avctx, " bS[%d]:%d", i, bS[i]); tprintf(h->avctx, "\n"); }
792 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0], a, b, 1 );
793 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1], a, b, 1 );
796 filter_mb_mbaff_edgev ( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 );
797 filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
798 filter_mb_mbaff_edgev ( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 );
799 filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
800 } else if (CHROMA422(h)) {
801 filter_mb_mbaff_edgecv(h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1);
802 filter_mb_mbaff_edgecv(h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1);
803 filter_mb_mbaff_edgecv(h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1);
804 filter_mb_mbaff_edgecv(h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1);
806 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 );
807 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
808 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 );
809 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
813 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0], a, b, 1 );
814 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1], a, b, 1 );
817 filter_mb_mbaff_edgev ( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 );
818 filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
819 filter_mb_mbaff_edgev ( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 );
820 filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
822 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 );
823 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
824 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 );
825 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
834 for (dir = 0; dir < 2; dir++)
835 filter_mb_dir(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize,
836 uvlinesize, mb_xy, mb_type, mvy_limit,
837 dir ? 0 : first_vertical_edge_done, a, b,
841 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);
842 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);