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 FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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 bit_depth = h->sps.bit_depth_luma;
105 const int qp_bd_offset = 6*(bit_depth-8);
106 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
107 const int alpha = alpha_table[index_a] << (bit_depth-8);
108 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset] << (bit_depth-8);
109 if (alpha ==0 || beta == 0) return;
113 tc[0] = tc0_table[index_a][bS[0]] << (bit_depth-8);
114 tc[1] = tc0_table[index_a][bS[1]] << (bit_depth-8);
115 tc[2] = tc0_table[index_a][bS[2]] << (bit_depth-8);
116 tc[3] = tc0_table[index_a][bS[3]] << (bit_depth-8);
117 h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
119 h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
122 static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
123 const int bit_depth = h->sps.bit_depth_luma;
124 const int qp_bd_offset = 6*(bit_depth-8);
125 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
126 const int alpha = alpha_table[index_a] << (bit_depth-8);
127 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset] << (bit_depth-8);
128 if (alpha ==0 || beta == 0) return;
132 tc[0] = (tc0_table[index_a][bS[0]] << (bit_depth-8))+1;
133 tc[1] = (tc0_table[index_a][bS[1]] << (bit_depth-8))+1;
134 tc[2] = (tc0_table[index_a][bS[2]] << (bit_depth-8))+1;
135 tc[3] = (tc0_table[index_a][bS[3]] << (bit_depth-8))+1;
136 h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
138 h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
142 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
144 const int bit_depth = h->sps.bit_depth_luma;
145 const int qp_bd_offset = 6*(bit_depth-8);
146 int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
147 int alpha = alpha_table[index_a] << (bit_depth-8);
148 int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset] << (bit_depth-8);
149 for( i = 0; i < 8; i++, pix += stride) {
150 const int bS_index = (i >> 1) * bsi;
152 if( bS[bS_index] == 0 ) {
156 if( bS[bS_index] < 4 ) {
157 const int tc0 = tc0_table[index_a][bS[bS_index]] << (bit_depth-8);
158 const int p0 = pix[-1];
159 const int p1 = pix[-2];
160 const int p2 = pix[-3];
161 const int q0 = pix[0];
162 const int q1 = pix[1];
163 const int q2 = pix[2];
165 if( FFABS( p0 - q0 ) < alpha &&
166 FFABS( p1 - p0 ) < beta &&
167 FFABS( q1 - q0 ) < beta ) {
171 if( FFABS( p2 - p0 ) < beta ) {
173 pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
176 if( FFABS( q2 - q0 ) < beta ) {
178 pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
182 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
183 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
184 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
185 tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
188 const int p0 = pix[-1];
189 const int p1 = pix[-2];
190 const int p2 = pix[-3];
192 const int q0 = pix[0];
193 const int q1 = pix[1];
194 const int q2 = pix[2];
196 if( FFABS( p0 - q0 ) < alpha &&
197 FFABS( p1 - p0 ) < beta &&
198 FFABS( q1 - q0 ) < beta ) {
200 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
201 if( FFABS( p2 - p0 ) < beta)
203 const int p3 = pix[-4];
205 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
206 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
207 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
210 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
212 if( FFABS( q2 - q0 ) < beta)
214 const int q3 = pix[3];
216 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
217 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
218 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
221 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
225 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
226 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
228 tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
233 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
235 const int bit_depth = h->sps.bit_depth_luma;
236 const int qp_bd_offset = 6*(bit_depth-8);
237 int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
238 int alpha = alpha_table[index_a] << (bit_depth-8);
239 int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset] << (bit_depth-8);
240 for( i = 0; i < 4; i++, pix += stride) {
241 const int bS_index = i*bsi;
243 if( bS[bS_index] == 0 ) {
247 if( bS[bS_index] < 4 ) {
248 const int tc = (tc0_table[index_a][bS[bS_index]] << (bit_depth-8)) + 1;
249 const int p0 = pix[-1];
250 const int p1 = pix[-2];
251 const int q0 = pix[0];
252 const int q1 = pix[1];
254 if( FFABS( p0 - q0 ) < alpha &&
255 FFABS( p1 - p0 ) < beta &&
256 FFABS( q1 - q0 ) < beta ) {
257 const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
259 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
260 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
261 tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
264 const int p0 = pix[-1];
265 const int p1 = pix[-2];
266 const int q0 = pix[0];
267 const int q1 = pix[1];
269 if( FFABS( p0 - q0 ) < alpha &&
270 FFABS( p1 - p0 ) < beta &&
271 FFABS( q1 - q0 ) < beta ) {
273 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
274 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
275 tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
281 static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
282 const int bit_depth = h->sps.bit_depth_luma;
283 const int qp_bd_offset = 6*(bit_depth-8);
284 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
285 const int alpha = alpha_table[index_a] << (bit_depth-8);
286 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset] << (bit_depth-8);
287 if (alpha ==0 || beta == 0) return;
291 tc[0] = tc0_table[index_a][bS[0]] << (bit_depth-8);
292 tc[1] = tc0_table[index_a][bS[1]] << (bit_depth-8);
293 tc[2] = tc0_table[index_a][bS[2]] << (bit_depth-8);
294 tc[3] = tc0_table[index_a][bS[3]] << (bit_depth-8);
295 h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
297 h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
301 static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
302 const int bit_depth = h->sps.bit_depth_luma;
303 const int qp_bd_offset = 6*(bit_depth-8);
304 const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
305 const int alpha = alpha_table[index_a] << (bit_depth-8);
306 const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset] << (bit_depth-8);
307 if (alpha ==0 || beta == 0) return;
311 tc[0] = (tc0_table[index_a][bS[0]] << (bit_depth-8))+1;
312 tc[1] = (tc0_table[index_a][bS[1]] << (bit_depth-8))+1;
313 tc[2] = (tc0_table[index_a][bS[2]] << (bit_depth-8))+1;
314 tc[3] = (tc0_table[index_a][bS[3]] << (bit_depth-8))+1;
315 h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
317 h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
321 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) {
322 MpegEncContext * const s = &h->s;
324 int mb_type, left_type;
325 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
329 if(!h->top_type || !h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
330 ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
333 assert(!FRAME_MBAFF);
334 left_type= h->left_type[0];
336 mb_type = s->current_picture.mb_type[mb_xy];
337 qp = s->current_picture.qscale_table[mb_xy];
338 qp0 = s->current_picture.qscale_table[mb_xy-1];
339 qp1 = s->current_picture.qscale_table[h->top_mb_xy];
340 qpc = get_chroma_qp( h, 0, qp );
341 qpc0 = get_chroma_qp( h, 0, qp0 );
342 qpc1 = get_chroma_qp( h, 0, qp1 );
343 qp0 = (qp + qp0 + 1) >> 1;
344 qp1 = (qp + qp1 + 1) >> 1;
345 qpc0 = (qpc + qpc0 + 1) >> 1;
346 qpc1 = (qpc + qpc1 + 1) >> 1;
347 qp_thresh = 15+52 - h->slice_alpha_c0_offset;
348 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
349 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
352 if( IS_INTRA(mb_type) ) {
353 int16_t bS4[4] = {4,4,4,4};
354 int16_t bS3[4] = {3,3,3,3};
355 int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
357 filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
358 if( IS_8x8DCT(mb_type) ) {
359 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
360 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
361 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
363 filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
364 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
365 filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
366 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
367 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
368 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
369 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
372 filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
373 filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
375 filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
376 filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
377 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
378 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
379 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
380 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
383 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
385 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
387 AV_WN64A(bS[0][0], 0x0002000200020002ULL);
388 AV_WN64A(bS[0][2], 0x0002000200020002ULL);
389 AV_WN64A(bS[1][0], 0x0002000200020002ULL);
390 AV_WN64A(bS[1][2], 0x0002000200020002ULL);
392 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;
393 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;
394 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
395 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
396 h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
397 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
399 if( IS_INTRA(left_type) )
400 AV_WN64A(bS[0][0], 0x0004000400040004ULL);
401 if( IS_INTRA(h->top_type) )
402 AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
404 #define FILTER(hv,dir,edge)\
405 if(AV_RN64A(bS[dir][edge])) { \
406 filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
408 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
409 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
416 } else if( IS_8x8DCT(mb_type) ) {
433 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
436 v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
437 if(!v && h->ref_cache[0][b_idx]!=-1)
438 v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
439 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
441 if(h->list_count==2){
443 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
444 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
445 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
448 if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
449 h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
452 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
453 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
454 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
455 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
462 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) {
463 MpegEncContext * const s = &h->s;
465 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
466 const int mbm_type = dir == 0 ? h->left_type[0] : h->top_type;
468 // how often to recheck mv-based bS when iterating between edges
469 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
471 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
472 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
474 // how often to recheck mv-based bS when iterating along each edge
475 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
477 if(mbm_type && !first_vertical_edge_done){
479 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
480 && IS_INTERLACED(mbm_type&~mb_type)
482 // This is a special case in the norm where the filtering must
483 // be done twice (one each of the field) even if we are in a
486 unsigned int tmp_linesize = 2 * linesize;
487 unsigned int tmp_uvlinesize = 2 * uvlinesize;
488 int mbn_xy = mb_xy - 2 * s->mb_stride;
491 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
492 DECLARE_ALIGNED(8, int16_t, bS)[4];
494 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
495 AV_WN64A(bS, 0x0003000300030003ULL);
497 if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
498 bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]);
499 bS[1]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+1]);
500 bS[2]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+2]);
501 bS[3]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+3]);
503 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8;
505 for( i = 0; i < 4; i++ ) {
506 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
510 // Do not use s->qscale as luma quantizer because it has not the same
511 // value in IPCM macroblocks.
512 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
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, tmp_linesize, tmp_uvlinesize);
514 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
515 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
516 filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
517 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
518 filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
519 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
522 DECLARE_ALIGNED(8, int16_t, bS)[4];
525 if( IS_INTRA(mb_type|mbm_type)) {
526 AV_WN64A(bS, 0x0003000300030003ULL);
527 if ( (!IS_INTERLACED(mb_type|mbm_type))
528 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
530 AV_WN64A(bS, 0x0004000400040004ULL);
535 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
536 AV_WN64A(bS, 0x0001000100010001ULL);
539 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
541 int bn_idx= b_idx - (dir ? 8:1);
543 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
549 for( i = 0; i < 4; i++ ) {
550 int x = dir == 0 ? 0 : i;
551 int y = dir == 0 ? i : 0;
552 int b_idx= 8 + 4 + x + 8*y;
553 int bn_idx= b_idx - (dir ? 8:1);
555 if( h->non_zero_count_cache[b_idx] |
556 h->non_zero_count_cache[bn_idx] ) {
561 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
567 // Do not use s->qscale as luma quantizer because it has not the same
568 // value in IPCM macroblocks.
569 if(bS[0]+bS[1]+bS[2]+bS[3]){
570 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbm_xy] + 1 ) >> 1;
571 //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]);
572 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);
573 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
575 filter_mb_edgev( &img_y[0], linesize, bS, qp, h );
577 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
578 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, qp, h);
579 if(h->pps.chroma_qp_diff)
580 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
581 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, qp, h);
584 filter_mb_edgeh( &img_y[0], linesize, bS, qp, h );
586 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
587 filter_mb_edgech( &img_cb[0], uvlinesize, bS, qp, h);
588 if(h->pps.chroma_qp_diff)
589 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
590 filter_mb_edgech( &img_cr[0], uvlinesize, bS, qp, h);
598 for( edge = 1; edge < edges; edge++ ) {
599 DECLARE_ALIGNED(8, int16_t, bS)[4];
602 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
605 if( IS_INTRA(mb_type)) {
606 AV_WN64A(bS, 0x0003000300030003ULL);
611 if( edge & mask_edge ) {
615 else if( mask_par0 ) {
616 int b_idx= 8 + 4 + edge * (dir ? 8:1);
617 int bn_idx= b_idx - (dir ? 8:1);
619 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
625 for( i = 0; i < 4; i++ ) {
626 int x = dir == 0 ? edge : i;
627 int y = dir == 0 ? i : edge;
628 int b_idx= 8 + 4 + x + 8*y;
629 int bn_idx= b_idx - (dir ? 8:1);
631 if( h->non_zero_count_cache[b_idx] |
632 h->non_zero_count_cache[bn_idx] ) {
637 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
641 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
646 // Do not use s->qscale as luma quantizer because it has not the same
647 // value in IPCM macroblocks.
648 qp = s->current_picture.qscale_table[mb_xy];
649 //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]);
650 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);
651 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
653 filter_mb_edgev( &img_y[4*edge<<h->pixel_shift], linesize, bS, qp, h );
654 if( (edge&1) == 0 ) {
655 filter_mb_edgecv( &img_cb[2*edge<<h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
656 filter_mb_edgecv( &img_cr[2*edge<<h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
659 filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
660 if( (edge&1) == 0 ) {
661 filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
662 filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
668 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) {
669 MpegEncContext * const s = &h->s;
670 const int mb_xy= mb_x + mb_y*s->mb_stride;
671 const int mb_type = s->current_picture.mb_type[mb_xy];
672 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
673 int first_vertical_edge_done = 0;
677 // and current and left pair do not have the same interlaced type
678 && IS_INTERLACED(mb_type^h->left_type[0])
679 // and left mb is in available to us
680 && h->left_type[0]) {
681 /* First vertical edge is different in MBAFF frames
682 * There are 8 different bS to compute and 2 different Qp
684 DECLARE_ALIGNED(8, int16_t, bS)[8];
688 int mb_qp, mbn0_qp, mbn1_qp;
690 first_vertical_edge_done = 1;
692 if( IS_INTRA(mb_type) ) {
693 AV_WN64A(&bS[0], 0x0004000400040004ULL);
694 AV_WN64A(&bS[4], 0x0004000400040004ULL);
696 static const uint8_t offset[2][2][8]={
698 {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
699 {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3},
701 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
702 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
705 const uint8_t *off= offset[MB_FIELD][mb_y&1];
706 for( i = 0; i < 8; i++ ) {
707 int j= MB_FIELD ? i>>2 : i&1;
708 int mbn_xy = h->left_mb_xy[j];
709 int mbn_type= h->left_type[j];
711 if( IS_INTRA( mbn_type ) )
714 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
715 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
716 (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
718 h->non_zero_count[mbn_xy][ off[i] ]));
723 mb_qp = s->current_picture.qscale_table[mb_xy];
724 mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]];
725 mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]];
726 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
727 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
728 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
729 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
730 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
731 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
732 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
733 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
734 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
735 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
738 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);
739 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
741 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
742 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
743 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
744 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
745 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
746 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
748 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
749 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
750 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
751 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
752 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
753 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
758 for( dir = 0; dir < 2; dir++ )
759 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);
761 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);
762 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);