Unroll loops in check_mv()
[libav.git] / libavcodec / h264_loopfilter.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... loop filter
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of FFmpeg.
6 *
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.
11 *
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.
16 *
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
20 */
21
22 /**
23 * @file libavcodec/h264_loopfilter.c
24 * H.264 / AVC / MPEG4 part10 loop filter.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #include "internal.h"
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "mathops.h"
34 #include "rectangle.h"
35
36 //#undef NDEBUG
37 #include <assert.h>
38
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,
50 255,255,
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,
55 };
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,
66 18, 18,
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,
71 };
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 },
100 };
101
102 static void av_noinline filter_mb_edgev( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h) {
103 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
104 const int alpha = alpha_table[index_a];
105 const int beta = beta_table[qp + h->slice_beta_offset];
106 if (alpha ==0 || beta == 0) return;
107
108 if( bS[0] < 4 ) {
109 int8_t tc[4];
110 tc[0] = tc0_table[index_a][bS[0]];
111 tc[1] = tc0_table[index_a][bS[1]];
112 tc[2] = tc0_table[index_a][bS[2]];
113 tc[3] = tc0_table[index_a][bS[3]];
114 h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
115 } else {
116 h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
117 }
118 }
119 static void av_noinline filter_mb_edgecv( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
120 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
121 const int alpha = alpha_table[index_a];
122 const int beta = beta_table[qp + h->slice_beta_offset];
123 if (alpha ==0 || beta == 0) return;
124
125 if( bS[0] < 4 ) {
126 int8_t tc[4];
127 tc[0] = tc0_table[index_a][bS[0]]+1;
128 tc[1] = tc0_table[index_a][bS[1]]+1;
129 tc[2] = tc0_table[index_a][bS[2]]+1;
130 tc[3] = tc0_table[index_a][bS[3]]+1;
131 h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
132 } else {
133 h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
134 }
135 }
136
137 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
138 int i;
139 int index_a = qp + h->slice_alpha_c0_offset;
140 int alpha = alpha_table[index_a];
141 int beta = beta_table[qp + h->slice_beta_offset];
142 for( i = 0; i < 8; i++, pix += stride) {
143 const int bS_index = (i >> 1) * bsi;
144
145 if( bS[bS_index] == 0 ) {
146 continue;
147 }
148
149 if( bS[bS_index] < 4 ) {
150 const int tc0 = tc0_table[index_a][bS[bS_index]];
151 const int p0 = pix[-1];
152 const int p1 = pix[-2];
153 const int p2 = pix[-3];
154 const int q0 = pix[0];
155 const int q1 = pix[1];
156 const int q2 = pix[2];
157
158 if( FFABS( p0 - q0 ) < alpha &&
159 FFABS( p1 - p0 ) < beta &&
160 FFABS( q1 - q0 ) < beta ) {
161 int tc = tc0;
162 int i_delta;
163
164 if( FFABS( p2 - p0 ) < beta ) {
165 if(tc0)
166 pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
167 tc++;
168 }
169 if( FFABS( q2 - q0 ) < beta ) {
170 if(tc0)
171 pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
172 tc++;
173 }
174
175 i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
176 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
177 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
178 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);
179 }
180 }else{
181 const int p0 = pix[-1];
182 const int p1 = pix[-2];
183 const int p2 = pix[-3];
184
185 const int q0 = pix[0];
186 const int q1 = pix[1];
187 const int q2 = pix[2];
188
189 if( FFABS( p0 - q0 ) < alpha &&
190 FFABS( p1 - p0 ) < beta &&
191 FFABS( q1 - q0 ) < beta ) {
192
193 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
194 if( FFABS( p2 - p0 ) < beta)
195 {
196 const int p3 = pix[-4];
197 /* p0', p1', p2' */
198 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
199 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
200 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
201 } else {
202 /* p0' */
203 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
204 }
205 if( FFABS( q2 - q0 ) < beta)
206 {
207 const int q3 = pix[3];
208 /* q0', q1', q2' */
209 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
210 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
211 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
212 } else {
213 /* q0' */
214 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
215 }
216 }else{
217 /* p0', q0' */
218 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
219 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
220 }
221 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]);
222 }
223 }
224 }
225 }
226 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
227 int i;
228 int index_a = qp + h->slice_alpha_c0_offset;
229 int alpha = alpha_table[index_a];
230 int beta = beta_table[qp + h->slice_beta_offset];
231 for( i = 0; i < 4; i++, pix += stride) {
232 const int bS_index = i*bsi;
233
234 if( bS[bS_index] == 0 ) {
235 continue;
236 }
237
238 if( bS[bS_index] < 4 ) {
239 const int tc = tc0_table[index_a][bS[bS_index]] + 1;
240 const int p0 = pix[-1];
241 const int p1 = pix[-2];
242 const int q0 = pix[0];
243 const int q1 = pix[1];
244
245 if( FFABS( p0 - q0 ) < alpha &&
246 FFABS( p1 - p0 ) < beta &&
247 FFABS( q1 - q0 ) < beta ) {
248 const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
249
250 pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
251 pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
252 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);
253 }
254 }else{
255 const int p0 = pix[-1];
256 const int p1 = pix[-2];
257 const int q0 = pix[0];
258 const int q1 = pix[1];
259
260 if( FFABS( p0 - q0 ) < alpha &&
261 FFABS( p1 - p0 ) < beta &&
262 FFABS( q1 - q0 ) < beta ) {
263
264 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
265 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
266 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]);
267 }
268 }
269 }
270 }
271
272 static void av_noinline filter_mb_edgeh( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
273 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
274 const int alpha = alpha_table[index_a];
275 const int beta = beta_table[qp + h->slice_beta_offset];
276 if (alpha ==0 || beta == 0) return;
277
278 if( bS[0] < 4 ) {
279 int8_t tc[4];
280 tc[0] = tc0_table[index_a][bS[0]];
281 tc[1] = tc0_table[index_a][bS[1]];
282 tc[2] = tc0_table[index_a][bS[2]];
283 tc[3] = tc0_table[index_a][bS[3]];
284 h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
285 } else {
286 h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
287 }
288 }
289
290 static void av_noinline filter_mb_edgech( uint8_t *pix, int stride, int16_t bS[4], unsigned int qp, H264Context *h ) {
291 const unsigned int index_a = qp + h->slice_alpha_c0_offset;
292 const int alpha = alpha_table[index_a];
293 const int beta = beta_table[qp + h->slice_beta_offset];
294 if (alpha ==0 || beta == 0) return;
295
296 if( bS[0] < 4 ) {
297 int8_t tc[4];
298 tc[0] = tc0_table[index_a][bS[0]]+1;
299 tc[1] = tc0_table[index_a][bS[1]]+1;
300 tc[2] = tc0_table[index_a][bS[2]]+1;
301 tc[3] = tc0_table[index_a][bS[3]]+1;
302 h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
303 } else {
304 h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
305 }
306 }
307
308 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) {
309 MpegEncContext * const s = &h->s;
310 int mb_xy;
311 int mb_type, left_type;
312 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
313
314 mb_xy = h->mb_xy;
315
316 if(!h->top_type || !s->dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
317 ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
318 return;
319 }
320 assert(!FRAME_MBAFF);
321 left_type= h->left_type[0];
322
323 mb_type = s->current_picture.mb_type[mb_xy];
324 qp = s->current_picture.qscale_table[mb_xy];
325 qp0 = s->current_picture.qscale_table[mb_xy-1];
326 qp1 = s->current_picture.qscale_table[h->top_mb_xy];
327 qpc = get_chroma_qp( h, 0, qp );
328 qpc0 = get_chroma_qp( h, 0, qp0 );
329 qpc1 = get_chroma_qp( h, 0, qp1 );
330 qp0 = (qp + qp0 + 1) >> 1;
331 qp1 = (qp + qp1 + 1) >> 1;
332 qpc0 = (qpc + qpc0 + 1) >> 1;
333 qpc1 = (qpc + qpc1 + 1) >> 1;
334 qp_thresh = 15+52 - h->slice_alpha_c0_offset;
335 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
336 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
337 return;
338
339 if( IS_INTRA(mb_type) ) {
340 int16_t bS4[4] = {4,4,4,4};
341 int16_t bS3[4] = {3,3,3,3};
342 int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
343 if( IS_8x8DCT(mb_type) ) {
344 if(left_type)
345 filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
346 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
347 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
348 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
349 } else {
350 if(left_type)
351 filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
352 filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
353 filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
354 filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
355 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
356 filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
357 filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
358 filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
359 }
360 if(left_type){
361 filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
362 filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
363 }
364 filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
365 filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
366 filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
367 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
368 filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
369 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
370 return;
371 } else {
372 DECLARE_ALIGNED_8(int16_t, bS)[2][4][4];
373 uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
374 int edges;
375 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
376 edges = 4;
377 bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;
378 } else {
379 int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 :
380 (mb_type & MB_TYPE_16x8) ? 1 : 0;
381 int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16))
382 && (h->left_type[0] & (MB_TYPE_16x16 | MB_TYPE_8x16))
383 ? 3 : 0;
384 int step = IS_8x8DCT(mb_type) ? 2 : 1;
385 edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
386 s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
387 h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
388 }
389 if( IS_INTRA(h->left_type[0]) )
390 bSv[0][0] = 0x0004000400040004ULL;
391 if( IS_INTRA(h->top_type) )
392 bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;
393
394 #define FILTER(hv,dir,edge)\
395 if(bSv[dir][edge]) {\
396 filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
397 if(!(edge&1)) {\
398 filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
399 filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
400 }\
401 }
402 if( edges == 1 ) {
403 if(left_type)
404 FILTER(v,0,0);
405 FILTER(h,1,0);
406 } else if( IS_8x8DCT(mb_type) ) {
407 if(left_type)
408 FILTER(v,0,0);
409 FILTER(v,0,2);
410 FILTER(h,1,0);
411 FILTER(h,1,2);
412 } else {
413 if(left_type)
414 FILTER(v,0,0);
415 FILTER(v,0,1);
416 FILTER(v,0,2);
417 FILTER(v,0,3);
418 FILTER(h,1,0);
419 FILTER(h,1,1);
420 FILTER(h,1,2);
421 FILTER(h,1,3);
422 }
423 #undef FILTER
424 }
425 }
426
427 static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
428 int v;
429
430 v = h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx] |
431 h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
432 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
433 if(h->list_count==2 && !v)
434 v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
435 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
436 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
437
438 if(h->list_count==2 && v){
439 if(h->ref_cache[0][b_idx] != h->ref_cache[1][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 return 1;
443 if(h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx] |
444 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
445 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit)
446 return 1;
447 return 0;
448 }
449
450 return v;
451 }
452
453 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) {
454 MpegEncContext * const s = &h->s;
455 int edge;
456 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
457 const int mbm_type = dir == 0 ? h->left_type[0] : h->top_type;
458
459 // how often to recheck mv-based bS when iterating between edges
460 static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
461 {0,3,1,1,3,3,3,3}};
462 const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
463 const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
464
465 // how often to recheck mv-based bS when iterating along each edge
466 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
467
468 if(mbm_type && !first_vertical_edge_done){
469
470 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
471 && IS_INTERLACED(mbm_type&~mb_type)
472 ) {
473 // This is a special case in the norm where the filtering must
474 // be done twice (one each of the field) even if we are in a
475 // frame macroblock.
476 //
477 unsigned int tmp_linesize = 2 * linesize;
478 unsigned int tmp_uvlinesize = 2 * uvlinesize;
479 int mbn_xy = mb_xy - 2 * s->mb_stride;
480 int j;
481
482 for(j=0; j<2; j++, mbn_xy += s->mb_stride){
483 DECLARE_ALIGNED_8(int16_t, bS)[4];
484 int qp;
485 if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
486 *(uint64_t*)bS= 0x0003000300030003ULL;
487 } else {
488 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 4+3*8; //FIXME 8x8dct?
489 int i;
490 for( i = 0; i < 4; i++ ) {
491 bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
492 }
493 }
494 // Do not use s->qscale as luma quantizer because it has not the same
495 // value in IPCM macroblocks.
496 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
497 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);
498 { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
499 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
500 filter_mb_edgech( &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
501 ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
502 filter_mb_edgech( &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
503 ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1, h);
504 }
505 }else{
506 DECLARE_ALIGNED_8(int16_t, bS)[4];
507 int qp;
508
509 if( IS_INTRA(mb_type|mbm_type)) {
510 *(uint64_t*)bS= 0x0003000300030003ULL;
511 if ( (!IS_INTERLACED(mb_type|mbm_type))
512 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
513 )
514 *(uint64_t*)bS= 0x0004000400040004ULL;
515 } else {
516 int i, l;
517 int mv_done;
518
519 if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) { //FIXME not posible left
520 *(uint64_t*)bS= 0x0001000100010001ULL;
521 mv_done = 1;
522 }
523 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
524 int b_idx= 8 + 4;
525 int bn_idx= b_idx - (dir ? 8:1);
526
527 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
528 mv_done = 1;
529 }
530 else
531 mv_done = 0;
532
533 for( i = 0; i < 4; i++ ) {
534 int x = dir == 0 ? 0 : i;
535 int y = dir == 0 ? i : 0;
536 int b_idx= 8 + 4 + x + 8*y;
537 int bn_idx= b_idx - (dir ? 8:1);
538
539 if( h->non_zero_count_cache[b_idx] |
540 h->non_zero_count_cache[bn_idx] ) {
541 bS[i] = 2;
542 }
543 else if(!mv_done)
544 {
545 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
546 }
547 }
548 }
549
550 /* Filter edge */
551 // Do not use s->qscale as luma quantizer because it has not the same
552 // value in IPCM macroblocks.
553 if(bS[0]+bS[1]+bS[2]+bS[3]){
554 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbm_xy] + 1 ) >> 1;
555 //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]);
556 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);
557 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
558 if( dir == 0 ) {
559 filter_mb_edgev( &img_y[0], linesize, bS, qp, h );
560 {
561 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
562 filter_mb_edgecv( &img_cb[0], uvlinesize, bS, qp, h);
563 if(h->pps.chroma_qp_diff)
564 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
565 filter_mb_edgecv( &img_cr[0], uvlinesize, bS, qp, h);
566 }
567 } else {
568 filter_mb_edgeh( &img_y[0], linesize, bS, qp, h );
569 {
570 int qp= ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
571 filter_mb_edgech( &img_cb[0], uvlinesize, bS, qp, h);
572 if(h->pps.chroma_qp_diff)
573 qp= ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
574 filter_mb_edgech( &img_cr[0], uvlinesize, bS, qp, h);
575 }
576 }
577 }
578 }
579 }
580
581 /* Calculate bS */
582 for( edge = 1; edge < edges; edge++ ) {
583 DECLARE_ALIGNED_8(int16_t, bS)[4];
584 int qp;
585
586 if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
587 continue;
588
589 if( IS_INTRA(mb_type)) {
590 *(uint64_t*)bS= 0x0003000300030003ULL;
591 } else {
592 int i, l;
593 int mv_done;
594
595 if( edge & mask_edge ) {
596 *(uint64_t*)bS= 0;
597 mv_done = 1;
598 }
599 else if( mask_par0 ) {
600 int b_idx= 8 + 4 + edge * (dir ? 8:1);
601 int bn_idx= b_idx - (dir ? 8:1);
602
603 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
604 mv_done = 1;
605 }
606 else
607 mv_done = 0;
608
609 for( i = 0; i < 4; i++ ) {
610 int x = dir == 0 ? edge : i;
611 int y = dir == 0 ? i : edge;
612 int b_idx= 8 + 4 + x + 8*y;
613 int bn_idx= b_idx - (dir ? 8:1);
614
615 if( h->non_zero_count_cache[b_idx] |
616 h->non_zero_count_cache[bn_idx] ) {
617 bS[i] = 2;
618 }
619 else if(!mv_done)
620 {
621 bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
622 }
623 }
624
625 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
626 continue;
627 }
628
629 /* Filter edge */
630 // Do not use s->qscale as luma quantizer because it has not the same
631 // value in IPCM macroblocks.
632 qp = s->current_picture.qscale_table[mb_xy];
633 //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]);
634 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);
635 //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
636 if( dir == 0 ) {
637 filter_mb_edgev( &img_y[4*edge], linesize, bS, qp, h );
638 if( (edge&1) == 0 ) {
639 filter_mb_edgecv( &img_cb[2*edge], uvlinesize, bS, h->chroma_qp[0], h);
640 filter_mb_edgecv( &img_cr[2*edge], uvlinesize, bS, h->chroma_qp[1], h);
641 }
642 } else {
643 filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
644 if( (edge&1) == 0 ) {
645 filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
646 filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
647 }
648 }
649 }
650 }
651
652 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) {
653 MpegEncContext * const s = &h->s;
654 const int mb_xy= mb_x + mb_y*s->mb_stride;
655 const int mb_type = s->current_picture.mb_type[mb_xy];
656 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
657 int first_vertical_edge_done = 0;
658 av_unused int dir;
659 int list;
660
661 if (FRAME_MBAFF
662 // and current and left pair do not have the same interlaced type
663 && IS_INTERLACED(mb_type^h->left_type[0])
664 // and left mb is in available to us
665 && h->left_type[0]) {
666 /* First vertical edge is different in MBAFF frames
667 * There are 8 different bS to compute and 2 different Qp
668 */
669 DECLARE_ALIGNED_8(int16_t, bS)[8];
670 int qp[2];
671 int bqp[2];
672 int rqp[2];
673 int mb_qp, mbn0_qp, mbn1_qp;
674 int i;
675 first_vertical_edge_done = 1;
676
677 if( IS_INTRA(mb_type) )
678 *(uint64_t*)&bS[0]=
679 *(uint64_t*)&bS[4]= 0x0004000400040004ULL;
680 else {
681 static const uint8_t offset[2][2][8]={
682 {
683 {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
684 {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3},
685 },{
686 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
687 {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
688 }
689 };
690 const uint8_t *off= offset[MB_FIELD][mb_y&1];
691 for( i = 0; i < 8; i++ ) {
692 int j= MB_FIELD ? i>>2 : i&1;
693 int mbn_xy = h->left_mb_xy[j];
694 int mbn_type= h->left_type[j];
695
696 if( IS_INTRA( mbn_type ) )
697 bS[i] = 4;
698 else{
699 bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
700 ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
701 (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
702 :
703 h->non_zero_count[mbn_xy][ off[i] ]));
704 }
705 }
706 }
707
708 mb_qp = s->current_picture.qscale_table[mb_xy];
709 mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]];
710 mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]];
711 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
712 bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
713 get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
714 rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
715 get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
716 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
717 bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
718 get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
719 rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
720 get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
721
722 /* Filter edge */
723 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);
724 { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
725 if(MB_FIELD){
726 filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
727 filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
728 filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
729 filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
730 filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
731 filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
732 }else{
733 filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
734 filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
735 filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
736 filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
737 filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
738 filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
739 }
740 }
741
742 #if CONFIG_SMALL
743 for( dir = 0; dir < 2; dir++ )
744 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);
745 #else
746 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);
747 filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);
748 #endif
749 }