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 void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {
105 const unsigned int index_a = qp + a;
106 const int alpha = alpha_table[index_a];
107 const int beta = beta_table[qp + b];
108 if (alpha ==0 || beta == 0) return;
110 if( bS[0] < 4 || !intra ) {
112 tc[0] = tc0_table[index_a][bS[0]];
113 tc[1] = tc0_table[index_a][bS[1]];
114 tc[2] = tc0_table[index_a][bS[2]];
115 tc[3] = tc0_table[index_a][bS[3]];
116 h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
118 h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
121 static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {
122 const unsigned int index_a = qp + a;
123 const int alpha = alpha_table[index_a];
124 const int beta = beta_table[qp + b];
125 if (alpha ==0 || beta == 0) return;
127 if( bS[0] < 4 || !intra ) {
129 tc[0] = tc0_table[index_a][bS[0]]+1;
130 tc[1] = tc0_table[index_a][bS[1]]+1;
131 tc[2] = tc0_table[index_a][bS[2]]+1;
132 tc[3] = tc0_table[index_a][bS[3]]+1;
133 h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
135 h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
139 static void av_always_inline filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp, int a, int b, int intra ) {
140 const unsigned int index_a = qp + a;
141 const int alpha = alpha_table[index_a];
142 const int beta = beta_table[qp + b];
143 if (alpha ==0 || beta == 0) return;
145 if( bS[0] < 4 || !intra ) {
147 tc[0] = tc0_table[index_a][bS[0*bsi]];
148 tc[1] = tc0_table[index_a][bS[1*bsi]];
149 tc[2] = tc0_table[index_a][bS[2*bsi]];
150 tc[3] = tc0_table[index_a][bS[3*bsi]];
151 h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
153 h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
156 static void av_always_inline filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp, int a, int b, int intra ) {
157 const unsigned int index_a = qp + a;
158 const int alpha = alpha_table[index_a];
159 const int beta = beta_table[qp + b];
160 if (alpha ==0 || beta == 0) return;
162 if( bS[0] < 4 || !intra ) {
164 tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
165 tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
166 tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
167 tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
168 h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
170 h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
174 static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {
175 const unsigned int index_a = qp + a;
176 const int alpha = alpha_table[index_a];
177 const int beta = beta_table[qp + b];
178 if (alpha ==0 || beta == 0) return;
180 if( bS[0] < 4 || !intra ) {
182 tc[0] = tc0_table[index_a][bS[0]];
183 tc[1] = tc0_table[index_a][bS[1]];
184 tc[2] = tc0_table[index_a][bS[2]];
185 tc[3] = tc0_table[index_a][bS[3]];
186 h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
188 h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
192 static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {
193 const unsigned int index_a = qp + a;
194 const int alpha = alpha_table[index_a];
195 const int beta = beta_table[qp + b];
196 if (alpha ==0 || beta == 0) return;
198 if( bS[0] < 4 || !intra ) {
200 tc[0] = tc0_table[index_a][bS[0]]+1;
201 tc[1] = tc0_table[index_a][bS[1]]+1;
202 tc[2] = tc0_table[index_a][bS[2]]+1;
203 tc[3] = tc0_table[index_a][bS[3]]+1;
204 h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
206 h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
210 static void av_always_inline h264_filter_mb_fast_internal( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
211 unsigned int linesize, unsigned int uvlinesize, int pixel_shift) {
212 MpegEncContext * const s = &h->s;
213 int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
214 int chroma444 = CHROMA444;
216 int mb_xy = h->mb_xy;
217 int left_type= h->left_type[LTOP];
218 int top_type= h->top_type;
220 int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
221 int a = h->slice_alpha_c0_offset - qp_bd_offset;
222 int b = h->slice_beta_offset - qp_bd_offset;
224 int mb_type = s->current_picture.f.mb_type[mb_xy];
225 int qp = s->current_picture.f.qscale_table[mb_xy];
226 int qp0 = s->current_picture.f.qscale_table[mb_xy - 1];
227 int qp1 = s->current_picture.f.qscale_table[h->top_mb_xy];
228 int qpc = get_chroma_qp( h, 0, qp );
229 int qpc0 = get_chroma_qp( h, 0, qp0 );
230 int qpc1 = get_chroma_qp( h, 0, qp1 );
231 qp0 = (qp + qp0 + 1) >> 1;
232 qp1 = (qp + qp1 + 1) >> 1;
233 qpc0 = (qpc + qpc0 + 1) >> 1;
234 qpc1 = (qpc + qpc1 + 1) >> 1;
236 if( IS_INTRA(mb_type) ) {
237 static const int16_t bS4[4] = {4,4,4,4};
238 static const int16_t bS3[4] = {3,3,3,3};
239 const int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
241 filter_mb_edgev( &img_y[4*0<<pixel_shift], linesize, bS4, qp0, a, b, h, 1);
242 if( IS_8x8DCT(mb_type) ) {
243 filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
245 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
247 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
249 filter_mb_edgev( &img_y[4*1<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
250 filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
251 filter_mb_edgev( &img_y[4*3<<pixel_shift], linesize, bS3, qp, a, b, h, 0);
253 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);
255 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, a, b, h, 0);
256 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);
257 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, a, b, h, 0);
262 filter_mb_edgev( &img_cb[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
263 filter_mb_edgev( &img_cr[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);
265 if( IS_8x8DCT(mb_type) ) {
266 filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
267 filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
269 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
270 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );
272 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
273 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
275 filter_mb_edgev( &img_cb[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
276 filter_mb_edgev( &img_cr[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
277 filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
278 filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
279 filter_mb_edgev( &img_cb[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
280 filter_mb_edgev( &img_cr[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);
282 filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
283 filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);
285 filter_mb_edgeh( &img_cb[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
286 filter_mb_edgeh( &img_cr[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);
287 filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
288 filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);
289 filter_mb_edgeh( &img_cb[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
290 filter_mb_edgeh( &img_cr[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);
294 filter_mb_edgecv( &img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
295 filter_mb_edgecv( &img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);
297 filter_mb_edgecv( &img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
298 filter_mb_edgecv( &img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);
300 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
301 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);
303 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
304 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);
309 LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
311 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) {
313 AV_WN64A(bS[0][0], 0x0002000200020002ULL);
314 AV_WN64A(bS[0][2], 0x0002000200020002ULL);
315 AV_WN64A(bS[1][0], 0x0002000200020002ULL);
316 AV_WN64A(bS[1][2], 0x0002000200020002ULL);
318 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;
319 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;
320 int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
321 edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
322 h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
323 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
325 if( IS_INTRA(left_type) )
326 AV_WN64A(bS[0][0], 0x0004000400040004ULL);
327 if( IS_INTRA(top_type) )
328 AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
330 #define FILTER(hv,dir,edge,intra)\
331 if(AV_RN64A(bS[dir][edge])) { \
332 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 );\
335 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 );\
336 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 );\
337 } else if(!(edge&1)) {\
338 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 );\
339 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 );\
348 } else if( IS_8x8DCT(mb_type) ) {
367 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) {
368 assert(!FRAME_MBAFF);
369 if(!h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
370 ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
375 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift);
378 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1);
380 h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0);
385 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
388 v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
389 if(!v && h->ref_cache[0][b_idx]!=-1)
390 v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
391 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
393 if(h->list_count==2){
395 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
396 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
397 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
400 if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
401 h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
404 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
405 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
406 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
407 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
414 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 chroma444, int dir) {
415 MpegEncContext * const s = &h->s;
417 int chroma_qp_avg[2];
418 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
419 const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type;
421 // how often to recheck mv-based bS when iterating between edges
422 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
424 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
425 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
427 // how often to recheck mv-based bS when iterating along each edge
428 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
430 if(mbm_type && !first_vertical_edge_done){
432 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
433 && IS_INTERLACED(mbm_type&~mb_type)
435 // This is a special case in the norm where the filtering must
436 // be done twice (one each of the field) even if we are in a
439 unsigned int tmp_linesize = 2 * linesize;
440 unsigned int tmp_uvlinesize = 2 * uvlinesize;
441 int mbn_xy = mb_xy - 2 * s->mb_stride;
444 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
445 DECLARE_ALIGNED(8, int16_t, bS)[4];
447 if (IS_INTRA(mb_type | s->current_picture.f.mb_type[mbn_xy])) {
448 AV_WN64A(bS, 0x0003000300030003ULL);
450 if (!CABAC && IS_8x8DCT(s->current_picture.f.mb_type[mbn_xy])) {
451 bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+0]);
452 bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+1]);
453 bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+2]);
454 bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+3]);
456 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4;
458 for( i = 0; i < 4; i++ ) {
459 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
463 // Do not use s->qscale as luma quantizer because it has not the same
464 // value in IPCM macroblocks.
465 qp = (s->current_picture.f.qscale_table[mb_xy] + s->current_picture.f.qscale_table[mbn_xy] + 1) >> 1;
466 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);
467 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
468 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, a, b, h, 0 );
469 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1;
470 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1;
473 filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
474 filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
476 filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);
477 filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);
482 DECLARE_ALIGNED(8, int16_t, bS)[4];
485 if( IS_INTRA(mb_type|mbm_type)) {
486 AV_WN64A(bS, 0x0003000300030003ULL);
487 if ( (!IS_INTERLACED(mb_type|mbm_type))
488 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
490 AV_WN64A(bS, 0x0004000400040004ULL);
495 if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
496 AV_WN64A(bS, 0x0001000100010001ULL);
499 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
501 int bn_idx= b_idx - (dir ? 8:1);
503 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
509 for( i = 0; i < 4; i++ ) {
510 int x = dir == 0 ? 0 : i;
511 int y = dir == 0 ? i : 0;
512 int b_idx= 8 + 4 + x + 8*y;
513 int bn_idx= b_idx - (dir ? 8:1);
515 if( h->non_zero_count_cache[b_idx] |
516 h->non_zero_count_cache[bn_idx] ) {
521 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
527 // Do not use s->qscale as luma quantizer because it has not the same
528 // value in IPCM macroblocks.
529 if(bS[0]+bS[1]+bS[2]+bS[3]){
530 qp = (s->current_picture.f.qscale_table[mb_xy] + s->current_picture.f.qscale_table[mbm_xy] + 1) >> 1;
531 //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]);
532 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);
533 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
534 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1;
535 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1;
537 filter_mb_edgev( &img_y[0], linesize, bS, qp, a, b, h, 1 );
540 filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
541 filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
543 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
544 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
548 filter_mb_edgeh( &img_y[0], linesize, bS, qp, a, b, h, 1 );
551 filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
552 filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
554 filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);
555 filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);
564 for( edge = 1; edge < edges; edge++ ) {
565 DECLARE_ALIGNED(8, int16_t, bS)[4];
568 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
571 if( IS_INTRA(mb_type)) {
572 AV_WN64A(bS, 0x0003000300030003ULL);
577 if( edge & mask_edge ) {
581 else if( mask_par0 ) {
582 int b_idx= 8 + 4 + edge * (dir ? 8:1);
583 int bn_idx= b_idx - (dir ? 8:1);
585 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
591 for( i = 0; i < 4; i++ ) {
592 int x = dir == 0 ? edge : i;
593 int y = dir == 0 ? i : edge;
594 int b_idx= 8 + 4 + x + 8*y;
595 int bn_idx= b_idx - (dir ? 8:1);
597 if( h->non_zero_count_cache[b_idx] |
598 h->non_zero_count_cache[bn_idx] ) {
603 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
607 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
612 // Do not use s->qscale as luma quantizer because it has not the same
613 // value in IPCM macroblocks.
614 qp = s->current_picture.f.qscale_table[mb_xy];
615 //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]);
616 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);
617 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
619 filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, a, b, h, 0 );
622 filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
623 filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
624 } else if( (edge&1) == 0 ) {
625 filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
626 filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
630 filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0 );
633 filter_mb_edgeh ( &img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
634 filter_mb_edgeh ( &img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
635 } else if( (edge&1) == 0 ) {
636 filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], a, b, h, 0);
637 filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], a, b, h, 0);
644 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) {
645 MpegEncContext * const s = &h->s;
646 const int mb_xy= mb_x + mb_y*s->mb_stride;
647 const int mb_type = s->current_picture.f.mb_type[mb_xy];
648 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
649 int first_vertical_edge_done = 0;
651 int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
652 int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
653 int a = h->slice_alpha_c0_offset - qp_bd_offset;
654 int b = h->slice_beta_offset - qp_bd_offset;
657 // and current and left pair do not have the same interlaced type
658 && IS_INTERLACED(mb_type^h->left_type[LTOP])
659 // and left mb is in available to us
660 && h->left_type[LTOP]) {
661 /* First vertical edge is different in MBAFF frames
662 * There are 8 different bS to compute and 2 different Qp
664 DECLARE_ALIGNED(8, int16_t, bS)[8];
668 int mb_qp, mbn0_qp, mbn1_qp;
670 first_vertical_edge_done = 1;
672 if( IS_INTRA(mb_type) ) {
673 AV_WN64A(&bS[0], 0x0004000400040004ULL);
674 AV_WN64A(&bS[4], 0x0004000400040004ULL);
676 static const uint8_t offset[2][2][8]={
678 {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},
679 {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},
681 {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
682 {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
685 const uint8_t *off= offset[MB_FIELD][mb_y&1];
686 for( i = 0; i < 8; i++ ) {
687 int j= MB_FIELD ? i>>2 : i&1;
688 int mbn_xy = h->left_mb_xy[LEFT(j)];
689 int mbn_type= h->left_type[LEFT(j)];
691 if( IS_INTRA( mbn_type ) )
694 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
695 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
696 (h->cbp_table[mbn_xy] & (((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2) << 12))
698 h->non_zero_count[mbn_xy][ off[i] ]));
703 mb_qp = s->current_picture.f.qscale_table[mb_xy];
704 mbn0_qp = s->current_picture.f.qscale_table[h->left_mb_xy[0]];
705 mbn1_qp = s->current_picture.f.qscale_table[h->left_mb_xy[1]];
706 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
707 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
708 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
709 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
710 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
711 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
712 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
713 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
714 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
715 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
718 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);
719 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
721 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0], a, b, 1 );
722 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1], a, b, 1 );
725 filter_mb_mbaff_edgev ( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 );
726 filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
727 filter_mb_mbaff_edgev ( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 );
728 filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
730 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0], a, b, 1 );
731 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );
732 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0], a, b, 1 );
733 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );
737 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0], a, b, 1 );
738 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1], a, b, 1 );
741 filter_mb_mbaff_edgev ( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 );
742 filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
743 filter_mb_mbaff_edgev ( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 );
744 filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
746 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0], a, b, 1 );
747 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );
748 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0], a, b, 1 );
749 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );
756 for( dir = 0; dir < 2; dir++ )
757 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, CHROMA444, dir);
759 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, CHROMA444, 0);
760 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, CHROMA444, 1);