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 unsigned int index_a = qp + h->slice_alpha_c0_offset;
105 const int alpha = alpha_table[index_a];
106 const int beta = beta_table[qp + h->slice_beta_offset];
107 if (alpha ==0 || beta == 0) return;
111 tc[0] = tc0_table[index_a][bS[0]];
112 tc[1] = tc0_table[index_a][bS[1]];
113 tc[2] = tc0_table[index_a][bS[2]];
114 tc[3] = tc0_table[index_a][bS[3]];
115 h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
117 h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
120 static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
121 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
122 const int alpha = alpha_table[index_a];
123 const int beta = beta_table[qp + h->slice_beta_offset];
124 if (alpha ==0 || beta == 0) return;
128 tc[0] = tc0_table[index_a][bS[0]]+1;
129 tc[1] = tc0_table[index_a][bS[1]]+1;
130 tc[2] = tc0_table[index_a][bS[2]]+1;
131 tc[3] = tc0_table[index_a][bS[3]]+1;
132 h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
134 h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
138 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
140 int index_a = qp + h->slice_alpha_c0_offset;
141 int alpha = alpha_table[index_a];
142 int beta = beta_table[qp + h->slice_beta_offset];
143 for( i = 0; i < 8; i++, pix += stride) {
144 const int bS_index = (i >> 1) * bsi;
146 if( bS[bS_index] == 0 ) {
150 if( bS[bS_index] < 4 ) {
151 const int tc0 = tc0_table[index_a][bS[bS_index]];
152 const int p0 = pix[-1];
153 const int p1 = pix[-2];
154 const int p2 = pix[-3];
155 const int q0 = pix[0];
156 const int q1 = pix[1];
157 const int q2 = pix[2];
159 if( FFABS( p0 - q0 ) < alpha &&
160 FFABS( p1 - p0 ) < beta &&
161 FFABS( q1 - q0 ) < beta ) {
165 if( FFABS( p2 - p0 ) < beta ) {
167 pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
170 if( FFABS( q2 - q0 ) < beta ) {
172 pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
176 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
177 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
178 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
179 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);
182 const int p0 = pix[-1];
183 const int p1 = pix[-2];
184 const int p2 = pix[-3];
186 const int q0 = pix[0];
187 const int q1 = pix[1];
188 const int q2 = pix[2];
190 if( FFABS( p0 - q0 ) < alpha &&
191 FFABS( p1 - p0 ) < beta &&
192 FFABS( q1 - q0 ) < beta ) {
194 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
195 if( FFABS( p2 - p0 ) < beta)
197 const int p3 = pix[-4];
199 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
200 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
201 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
204 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
206 if( FFABS( q2 - q0 ) < beta)
208 const int q3 = pix[3];
210 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
211 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
212 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
215 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
219 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
220 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
222 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]);
227 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
229 int index_a = qp + h->slice_alpha_c0_offset;
230 int alpha = alpha_table[index_a];
231 int beta = beta_table[qp + h->slice_beta_offset];
232 for( i = 0; i < 4; i++, pix += stride) {
233 const int bS_index = i*bsi;
235 if( bS[bS_index] == 0 ) {
239 if( bS[bS_index] < 4 ) {
240 const int tc = tc0_table[index_a][bS[bS_index]] + 1;
241 const int p0 = pix[-1];
242 const int p1 = pix[-2];
243 const int q0 = pix[0];
244 const int q1 = pix[1];
246 if( FFABS( p0 - q0 ) < alpha &&
247 FFABS( p1 - p0 ) < beta &&
248 FFABS( q1 - q0 ) < beta ) {
249 const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
251 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
252 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
253 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);
256 const int p0 = pix[-1];
257 const int p1 = pix[-2];
258 const int q0 = pix[0];
259 const int q1 = pix[1];
261 if( FFABS( p0 - q0 ) < alpha &&
262 FFABS( p1 - p0 ) < beta &&
263 FFABS( q1 - q0 ) < beta ) {
265 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
266 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
267 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]);
273 static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
274 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
275 const int alpha = alpha_table[index_a];
276 const int beta = beta_table[qp + h->slice_beta_offset];
277 if (alpha ==0 || beta == 0) return;
281 tc[0] = tc0_table[index_a][bS[0]];
282 tc[1] = tc0_table[index_a][bS[1]];
283 tc[2] = tc0_table[index_a][bS[2]];
284 tc[3] = tc0_table[index_a][bS[3]];
285 h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
287 h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
291 static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
292 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
293 const int alpha = alpha_table[index_a];
294 const int beta = beta_table[qp + h->slice_beta_offset];
295 if (alpha ==0 || beta == 0) return;
299 tc[0] = tc0_table[index_a][bS[0]]+1;
300 tc[1] = tc0_table[index_a][bS[1]]+1;
301 tc[2] = tc0_table[index_a][bS[2]]+1;
302 tc[3] = tc0_table[index_a][bS[3]]+1;
303 h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
305 h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
309 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) {
310 MpegEncContext * const s = &h->s;
312 int mb_type, left_type;
313 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
317 if(!h->top_type || !h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
318 ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
321 assert(!FRAME_MBAFF);
322 left_type= h->left_type[0];
324 mb_type = s->current_picture.mb_type[mb_xy];
325 qp = s->current_picture.qscale_table[mb_xy];
326 qp0 = s->current_picture.qscale_table[mb_xy-1];
327 qp1 = s->current_picture.qscale_table[h->top_mb_xy];
328 qpc = get_chroma_qp( h, 0, qp );
329 qpc0 = get_chroma_qp( h, 0, qp0 );
330 qpc1 = get_chroma_qp( h, 0, qp1 );
331 qp0 = (qp + qp0 + 1) >> 1;
332 qp1 = (qp + qp1 + 1) >> 1;
333 qpc0 = (qpc + qpc0 + 1) >> 1;
334 qpc1 = (qpc + qpc1 + 1) >> 1;
335 qp_thresh = 15+52 - h->slice_alpha_c0_offset;
336 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
337 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
340 if( IS_INTRA(mb_type) ) {
341 int16_t bS4[4] = {4,4,4,4};
342 int16_t bS3[4] = {3,3,3,3};
343 int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
345 filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
346 if( IS_8x8DCT(mb_type) ) {
347 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
348 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
349 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
351 filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
352 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
353 filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
354 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
355 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
356 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
357 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
360 filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
361 filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
363 filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
364 filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
365 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
366 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
367 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
368 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
371 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
373 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
375 AV_WN64A(bS[0][0], 0x0002000200020002ULL);
376 AV_WN64A(bS[0][2], 0x0002000200020002ULL);
377 AV_WN64A(bS[1][0], 0x0002000200020002ULL);
378 AV_WN64A(bS[1][2], 0x0002000200020002ULL);
380 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;
381 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;
382 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
383 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
384 h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
385 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
387 if( IS_INTRA(left_type) )
388 AV_WN64A(bS[0][0], 0x0004000400040004ULL);
389 if( IS_INTRA(h->top_type) )
390 AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
392 #define FILTER(hv,dir,edge)\
393 if(AV_RN64A(bS[dir][edge])) { \
394 filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
396 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
397 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
404 } else if( IS_8x8DCT(mb_type) ) {
421 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
424 v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
425 if(!v && h->ref_cache[0][b_idx]!=-1)
426 v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
427 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
429 if(h->list_count==2){
431 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
432 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
433 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
436 if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
437 h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
440 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
441 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
442 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
443 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
450 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) {
451 MpegEncContext * const s = &h->s;
453 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
454 const int mbm_type = dir == 0 ? h->left_type[0] : h->top_type;
456 // how often to recheck mv-based bS when iterating between edges
457 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
459 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
460 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
462 // how often to recheck mv-based bS when iterating along each edge
463 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
465 if(mbm_type && !first_vertical_edge_done){
467 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
468 && IS_INTERLACED(mbm_type&~mb_type)
470 // This is a special case in the norm where the filtering must
471 // be done twice (one each of the field) even if we are in a
474 unsigned int tmp_linesize = 2 * linesize;
475 unsigned int tmp_uvlinesize = 2 * uvlinesize;
476 int mbn_xy = mb_xy - 2 * s->mb_stride;
479 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
480 DECLARE_ALIGNED(8, int16_t, bS)[4];
482 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
483 AV_WN64A(bS, 0x0003000300030003ULL);
485 if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
486 bS[0]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+0]);
487 bS[1]= 1+((h->cbp_table[mbn_xy] & 4)||h->non_zero_count_cache[scan8[0]+1]);
488 bS[2]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+2]);
489 bS[3]= 1+((h->cbp_table[mbn_xy] & 8)||h->non_zero_count_cache[scan8[0]+3]);
491 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8;
493 for( i = 0; i < 4; i++ ) {
494 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
498 // Do not use s->qscale as luma quantizer because it has not the same
499 // value in IPCM macroblocks.
500 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
501 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);
502 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
503 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
504 filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
505 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
506 filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
507 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
510 DECLARE_ALIGNED(8, int16_t, bS)[4];
513 if( IS_INTRA(mb_type|mbm_type)) {
514 AV_WN64A(bS, 0x0003000300030003ULL);
515 if ( (!IS_INTERLACED(mb_type|mbm_type))
516 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
518 AV_WN64A(bS, 0x0004000400040004ULL);
523 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
524 AV_WN64A(bS, 0x0001000100010001ULL);
527 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
529 int bn_idx= b_idx - (dir ? 8:1);
531 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
537 for( i = 0; i < 4; i++ ) {
538 int x = dir == 0 ? 0 : i;
539 int y = dir == 0 ? i : 0;
540 int b_idx= 8 + 4 + x + 8*y;
541 int bn_idx= b_idx - (dir ? 8:1);
543 if( h->non_zero_count_cache[b_idx] |
544 h->non_zero_count_cache[bn_idx] ) {
549 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
555 // Do not use s->qscale as luma quantizer because it has not the same
556 // value in IPCM macroblocks.
557 if(bS[0]+bS[1]+bS[2]+bS[3]){
558 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbm_xy] + 1 ) >> 1;
559 //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]);
560 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);
561 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
563 filter_mb_edgev( &img_y[0], linesize, bS, qp, h );
565 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
566 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, qp, h);
567 if(h->pps.chroma_qp_diff)
568 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
569 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, qp, h);
572 filter_mb_edgeh( &img_y[0], linesize, bS, qp, h );
574 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
575 filter_mb_edgech( &img_cb[0], uvlinesize, bS, qp, h);
576 if(h->pps.chroma_qp_diff)
577 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
578 filter_mb_edgech( &img_cr[0], uvlinesize, bS, qp, h);
586 for( edge = 1; edge < edges; edge++ ) {
587 DECLARE_ALIGNED(8, int16_t, bS)[4];
590 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
593 if( IS_INTRA(mb_type)) {
594 AV_WN64A(bS, 0x0003000300030003ULL);
599 if( edge & mask_edge ) {
603 else if( mask_par0 ) {
604 int b_idx= 8 + 4 + edge * (dir ? 8:1);
605 int bn_idx= b_idx - (dir ? 8:1);
607 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
613 for( i = 0; i < 4; i++ ) {
614 int x = dir == 0 ? edge : i;
615 int y = dir == 0 ? i : edge;
616 int b_idx= 8 + 4 + x + 8*y;
617 int bn_idx= b_idx - (dir ? 8:1);
619 if( h->non_zero_count_cache[b_idx] |
620 h->non_zero_count_cache[bn_idx] ) {
625 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
629 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
634 // Do not use s->qscale as luma quantizer because it has not the same
635 // value in IPCM macroblocks.
636 qp = s->current_picture.qscale_table[mb_xy];
637 //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]);
638 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);
639 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
641 filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h );
642 if( (edge&1) == 0 ) {
643 filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, h->chroma_qp[0], h);
644 filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, h->chroma_qp[1], h);
647 filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
648 if( (edge&1) == 0 ) {
649 filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
650 filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
656 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) {
657 MpegEncContext * const s = &h->s;
658 const int mb_xy= mb_x + mb_y*s->mb_stride;
659 const int mb_type = s->current_picture.mb_type[mb_xy];
660 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
661 int first_vertical_edge_done = 0;
665 // and current and left pair do not have the same interlaced type
666 && IS_INTERLACED(mb_type^h->left_type[0])
667 // and left mb is in available to us
668 && h->left_type[0]) {
669 /* First vertical edge is different in MBAFF frames
670 * There are 8 different bS to compute and 2 different Qp
672 DECLARE_ALIGNED(8, int16_t, bS)[8];
676 int mb_qp, mbn0_qp, mbn1_qp;
678 first_vertical_edge_done = 1;
680 if( IS_INTRA(mb_type) ) {
681 AV_WN64A(&bS[0], 0x0004000400040004ULL);
682 AV_WN64A(&bS[4], 0x0004000400040004ULL);
684 static const uint8_t offset[2][2][8]={
686 {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
687 {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3},
689 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
690 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
693 const uint8_t *off= offset[MB_FIELD][mb_y&1];
694 for( i = 0; i < 8; i++ ) {
695 int j= MB_FIELD ? i>>2 : i&1;
696 int mbn_xy = h->left_mb_xy[j];
697 int mbn_type= h->left_type[j];
699 if( IS_INTRA( mbn_type ) )
702 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
703 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
704 (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
706 h->non_zero_count[mbn_xy][ off[i] ]));
711 mb_qp = s->current_picture.qscale_table[mb_xy];
712 mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]];
713 mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]];
714 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
715 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
716 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
717 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
718 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
719 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
720 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
721 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
722 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
723 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
726 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);
727 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
729 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
730 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
731 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
732 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
733 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
734 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
736 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
737 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
738 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
739 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
740 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
741 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
746 for( dir = 0; dir < 2; dir++ )
747 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);
749 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);
750 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);