Mercurial > libavcodec.hg
annotate h264.c @ 4228:1cffc7036079 libavcodec
fix sigfpe with intra only encoding
author | bcoudurier |
---|---|
date | Wed, 22 Nov 2006 13:24:58 +0000 |
parents | 23da44e8fd05 |
children | 6f839bb47457 |
rev | line source |
---|---|
1168 | 1 /* |
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder | |
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> | |
4 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
1168 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
1168 | 11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
1168 | 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 | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3927
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3029
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
1168 | 20 * |
21 */ | |
2967 | 22 |
1168 | 23 /** |
24 * @file h264.c | |
25 * H.264 / AVC / MPEG4 part10 codec. | |
26 * @author Michael Niedermayer <michaelni@gmx.at> | |
27 */ | |
28 | |
29 #include "common.h" | |
30 #include "dsputil.h" | |
31 #include "avcodec.h" | |
32 #include "mpegvideo.h" | |
33 #include "h264data.h" | |
34 #include "golomb.h" | |
35 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
36 #include "cabac.h" |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
37 |
3284
a224d9752912
don't force asserts in release builds. 2% faster h264.
lorenm
parents:
3219
diff
changeset
|
38 //#undef NDEBUG |
1168 | 39 #include <assert.h> |
40 | |
41 #define interlaced_dct interlaced_dct_is_a_bad_name | |
42 #define mb_intra mb_intra_isnt_initalized_see_mb_type | |
43 | |
44 #define LUMA_DC_BLOCK_INDEX 25 | |
45 #define CHROMA_DC_BLOCK_INDEX 26 | |
46 | |
47 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8 | |
48 #define COEFF_TOKEN_VLC_BITS 8 | |
49 #define TOTAL_ZEROS_VLC_BITS 9 | |
50 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3 | |
51 #define RUN_VLC_BITS 3 | |
52 #define RUN7_VLC_BITS 6 | |
53 | |
54 #define MAX_SPS_COUNT 32 | |
55 #define MAX_PPS_COUNT 256 | |
56 | |
57 #define MAX_MMCO_COUNT 66 | |
58 | |
3316 | 59 /* Compiling in interlaced support reduces the speed |
60 * of progressive decoding by about 2%. */ | |
61 #define ALLOW_INTERLACE | |
62 | |
63 #ifdef ALLOW_INTERLACE | |
64 #define MB_MBAFF h->mb_mbaff | |
65 #define MB_FIELD h->mb_field_decoding_flag | |
66 #define FRAME_MBAFF h->mb_aff_frame | |
67 #else | |
68 #define MB_MBAFF 0 | |
69 #define MB_FIELD 0 | |
70 #define FRAME_MBAFF 0 | |
71 #undef IS_INTERLACED | |
72 #define IS_INTERLACED(mb_type) 0 | |
73 #endif | |
74 | |
1168 | 75 /** |
76 * Sequence parameter set | |
77 */ | |
78 typedef struct SPS{ | |
2967 | 79 |
1168 | 80 int profile_idc; |
81 int level_idc; | |
2763 | 82 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag |
1168 | 83 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4 |
84 int poc_type; ///< pic_order_cnt_type | |
85 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4 | |
86 int delta_pic_order_always_zero_flag; | |
87 int offset_for_non_ref_pic; | |
88 int offset_for_top_to_bottom_field; | |
89 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle | |
90 int ref_frame_count; ///< num_ref_frames | |
1371 | 91 int gaps_in_frame_num_allowed_flag; |
1168 | 92 int mb_width; ///< frame_width_in_mbs_minus1 + 1 |
93 int mb_height; ///< frame_height_in_mbs_minus1 + 1 | |
94 int frame_mbs_only_flag; | |
95 int mb_aff; ///<mb_adaptive_frame_field_flag | |
96 int direct_8x8_inference_flag; | |
1371 | 97 int crop; ///< frame_cropping_flag |
98 int crop_left; ///< frame_cropping_rect_left_offset | |
99 int crop_right; ///< frame_cropping_rect_right_offset | |
100 int crop_top; ///< frame_cropping_rect_top_offset | |
101 int crop_bottom; ///< frame_cropping_rect_bottom_offset | |
1168 | 102 int vui_parameters_present_flag; |
1548 | 103 AVRational sar; |
2174
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
104 int timing_info_present_flag; |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
105 uint32_t num_units_in_tick; |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
106 uint32_t time_scale; |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
107 int fixed_frame_rate_flag; |
1168 | 108 short offset_for_ref_frame[256]; //FIXME dyn aloc? |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
109 int bitstream_restriction_flag; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
110 int num_reorder_frames; |
2919 | 111 int scaling_matrix_present; |
112 uint8_t scaling_matrix4[6][16]; | |
113 uint8_t scaling_matrix8[2][64]; | |
1168 | 114 }SPS; |
115 | |
116 /** | |
117 * Picture parameter set | |
118 */ | |
119 typedef struct PPS{ | |
120 int sps_id; | |
121 int cabac; ///< entropy_coding_mode_flag | |
122 int pic_order_present; ///< pic_order_present_flag | |
123 int slice_group_count; ///< num_slice_groups_minus1 + 1 | |
124 int mb_slice_group_map_type; | |
125 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1 | |
126 int weighted_pred; ///< weighted_pred_flag | |
127 int weighted_bipred_idc; | |
128 int init_qp; ///< pic_init_qp_minus26 + 26 | |
129 int init_qs; ///< pic_init_qs_minus26 + 26 | |
130 int chroma_qp_index_offset; | |
131 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag | |
132 int constrained_intra_pred; ///< constrained_intra_pred_flag | |
133 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag | |
2755 | 134 int transform_8x8_mode; ///< transform_8x8_mode_flag |
2919 | 135 uint8_t scaling_matrix4[6][16]; |
136 uint8_t scaling_matrix8[2][64]; | |
1168 | 137 }PPS; |
138 | |
139 /** | |
140 * Memory management control operation opcode. | |
141 */ | |
142 typedef enum MMCOOpcode{ | |
143 MMCO_END=0, | |
144 MMCO_SHORT2UNUSED, | |
145 MMCO_LONG2UNUSED, | |
146 MMCO_SHORT2LONG, | |
147 MMCO_SET_MAX_LONG, | |
2967 | 148 MMCO_RESET, |
1168 | 149 MMCO_LONG, |
150 } MMCOOpcode; | |
151 | |
152 /** | |
153 * Memory management control operation. | |
154 */ | |
155 typedef struct MMCO{ | |
156 MMCOOpcode opcode; | |
157 int short_frame_num; | |
158 int long_index; | |
159 } MMCO; | |
160 | |
161 /** | |
162 * H264Context | |
163 */ | |
164 typedef struct H264Context{ | |
165 MpegEncContext s; | |
2967 | 166 int nal_ref_idc; |
1168 | 167 int nal_unit_type; |
2979 | 168 #define NAL_SLICE 1 |
169 #define NAL_DPA 2 | |
170 #define NAL_DPB 3 | |
171 #define NAL_DPC 4 | |
172 #define NAL_IDR_SLICE 5 | |
173 #define NAL_SEI 6 | |
174 #define NAL_SPS 7 | |
175 #define NAL_PPS 8 | |
176 #define NAL_AUD 9 | |
2960 | 177 #define NAL_END_SEQUENCE 10 |
178 #define NAL_END_STREAM 11 | |
179 #define NAL_FILLER_DATA 12 | |
180 #define NAL_SPS_EXT 13 | |
181 #define NAL_AUXILIARY_SLICE 19 | |
1168 | 182 uint8_t *rbsp_buffer; |
3066
04b924f8f5a5
warning fixes by Luca Abeni, lucabe72 ##@## email ##.## it
diego
parents:
3063
diff
changeset
|
183 unsigned int rbsp_buffer_size; |
1168 | 184 |
2227 | 185 /** |
186 * Used to parse AVC variant of h264 | |
187 */ | |
188 int is_avc; ///< this flag is != 0 if codec is avc1 | |
189 int got_avcC; ///< flag used to parse avcC data only once | |
190 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4) | |
191 | |
1168 | 192 int chroma_qp; //QPc |
193 | |
3316 | 194 int prev_mb_skipped; |
195 int next_mb_skipped; | |
1168 | 196 |
197 //prediction stuff | |
198 int chroma_pred_mode; | |
199 int intra16x16_pred_mode; | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
200 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
201 int top_mb_xy; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
202 int left_mb_xy[2]; |
2967 | 203 |
1168 | 204 int8_t intra4x4_pred_mode_cache[5*8]; |
205 int8_t (*intra4x4_pred_mode)[8]; | |
206 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp? | |
2755 | 207 void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride); |
1168 | 208 void (*pred8x8 [4+3])(uint8_t *src, int stride); |
209 void (*pred16x16[4+3])(uint8_t *src, int stride); | |
210 unsigned int topleft_samples_available; | |
211 unsigned int top_samples_available; | |
212 unsigned int topright_samples_available; | |
213 unsigned int left_samples_available; | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
214 uint8_t (*top_borders[2])[16+2*8]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
215 uint8_t left_border[2*(17+2*9)]; |
1168 | 216 |
217 /** | |
218 * non zero coeff count cache. | |
219 * is 64 if not available. | |
220 */ | |
3089 | 221 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]); |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
222 uint8_t (*non_zero_count)[16]; |
1168 | 223 |
224 /** | |
225 * Motion vector cache. | |
226 */ | |
3089 | 227 DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]); |
228 DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]); | |
1168 | 229 #define LIST_NOT_USED -1 //FIXME rename? |
230 #define PART_NOT_AVAILABLE -2 | |
2967 | 231 |
1168 | 232 /** |
233 * is 1 if the specific list MV&references are set to 0,0,-2. | |
234 */ | |
235 int mv_cache_clean[2]; | |
236 | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
237 /** |
2755 | 238 * number of neighbors (top and/or left) that used 8x8 dct |
239 */ | |
240 int neighbor_transform_size; | |
241 | |
242 /** | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
243 * block_offset[ 0..23] for frame macroblocks |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
244 * block_offset[24..47] for field macroblocks |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
245 */ |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
246 int block_offset[2*(16+8)]; |
2967 | 247 |
2641
c337f851d0f9
fix storage of motion vectors for frames with more than 4096 macroblocks.
lorenm
parents:
2640
diff
changeset
|
248 uint32_t *mb2b_xy; //FIXME are these 4 a good idea? |
c337f851d0f9
fix storage of motion vectors for frames with more than 4096 macroblocks.
lorenm
parents:
2640
diff
changeset
|
249 uint32_t *mb2b8_xy; |
2395 | 250 int b_stride; //FIXME use s->b4_stride |
1168 | 251 int b8_stride; |
252 | |
3316 | 253 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff |
254 int mb_uvlinesize; | |
255 | |
256 int emu_edge_width; | |
257 int emu_edge_height; | |
258 | |
1234 | 259 int halfpel_flag; |
260 int thirdpel_flag; | |
261 | |
1319 | 262 int unknown_svq3_flag; |
263 int next_slice_index; | |
264 | |
1168 | 265 SPS sps_buffer[MAX_SPS_COUNT]; |
266 SPS sps; ///< current sps | |
2967 | 267 |
1168 | 268 PPS pps_buffer[MAX_PPS_COUNT]; |
269 /** | |
270 * current pps | |
271 */ | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
272 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that? |
1168 | 273 |
2919 | 274 uint32_t dequant4_buffer[6][52][16]; |
275 uint32_t dequant8_buffer[2][52][64]; | |
276 uint32_t (*dequant4_coeff[6])[16]; | |
277 uint32_t (*dequant8_coeff[2])[64]; | |
278 int dequant_coeff_pps; ///< reinit tables when pps changes | |
2755 | 279 |
1168 | 280 int slice_num; |
281 uint8_t *slice_table_base; | |
3316 | 282 uint8_t *slice_table; ///< slice_table_base + 2*mb_stride + 1 |
1168 | 283 int slice_type; |
284 int slice_type_fixed; | |
2967 | 285 |
1168 | 286 //interlacing specific flags |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
287 int mb_aff_frame; |
1168 | 288 int mb_field_decoding_flag; |
3316 | 289 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag |
2967 | 290 |
1168 | 291 int sub_mb_type[4]; |
2967 | 292 |
1168 | 293 //POC stuff |
294 int poc_lsb; | |
295 int poc_msb; | |
296 int delta_poc_bottom; | |
297 int delta_poc[2]; | |
298 int frame_num; | |
299 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0 | |
300 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0 | |
301 int frame_num_offset; ///< for POC type 2 | |
302 int prev_frame_num_offset; ///< for POC type 2 | |
303 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2 | |
304 | |
305 /** | |
306 * frame_num for frames or 2*frame_num for field pics. | |
307 */ | |
308 int curr_pic_num; | |
2967 | 309 |
1168 | 310 /** |
311 * max_frame_num or 2*max_frame_num for field pics. | |
312 */ | |
313 int max_pic_num; | |
314 | |
315 //Weighted pred stuff | |
2415 | 316 int use_weight; |
317 int use_weight_chroma; | |
1168 | 318 int luma_log2_weight_denom; |
319 int chroma_log2_weight_denom; | |
3316 | 320 int luma_weight[2][48]; |
321 int luma_offset[2][48]; | |
322 int chroma_weight[2][48][2]; | |
323 int chroma_offset[2][48][2]; | |
324 int implicit_weight[48][48]; | |
2967 | 325 |
1168 | 326 //deblock |
2967 | 327 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0 |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
328 int slice_alpha_c0_offset; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
329 int slice_beta_offset; |
2967 | 330 |
1168 | 331 int redundant_pic_count; |
2967 | 332 |
1168 | 333 int direct_spatial_mv_pred; |
2396 | 334 int dist_scale_factor[16]; |
3316 | 335 int dist_scale_factor_field[32]; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
336 int map_col_to_list0[2][16]; |
3316 | 337 int map_col_to_list0_field[2][32]; |
1168 | 338 |
339 /** | |
340 * num_ref_idx_l0/1_active_minus1 + 1 | |
341 */ | |
3316 | 342 int ref_count[2]; ///< counts frames or fields, depending on current mb mode |
2582 | 343 Picture *short_ref[32]; |
344 Picture *long_ref[32]; | |
1168 | 345 Picture default_ref_list[2][32]; |
3316 | 346 Picture ref_list[2][48]; ///< 0..15: frame refs, 16..47: mbaff field refs |
2409 | 347 Picture *delayed_pic[16]; //FIXME size? |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
348 Picture *delayed_output_pic; |
2967 | 349 |
1168 | 350 /** |
351 * memory management control operations buffer. | |
352 */ | |
353 MMCO mmco[MAX_MMCO_COUNT]; | |
354 int mmco_index; | |
2967 | 355 |
1168 | 356 int long_ref_count; ///< number of actual long term references |
357 int short_ref_count; ///< number of actual short term references | |
2967 | 358 |
1168 | 359 //data partitioning |
360 GetBitContext intra_gb; | |
361 GetBitContext inter_gb; | |
362 GetBitContext *intra_gb_ptr; | |
363 GetBitContext *inter_gb_ptr; | |
2967 | 364 |
3089 | 365 DECLARE_ALIGNED_8(DCTELEM, mb[16*24]); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
366 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
367 /** |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
368 * Cabac |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
369 */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
370 CABACContext cabac; |
2755 | 371 uint8_t cabac_state[460]; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
372 int cabac_init_idc; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
373 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
374 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
375 uint16_t *cbp_table; |
3651 | 376 int cbp; |
2314 | 377 int top_cbp; |
378 int left_cbp; | |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
379 /* chroma_pred_mode for i4x4 or i16x16, else 0 */ |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
380 uint8_t *chroma_pred_mode_table; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
381 int last_qscale_diff; |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
382 int16_t (*mvd_table[2])[2]; |
3089 | 383 DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]); |
2396 | 384 uint8_t *direct_table; |
385 uint8_t direct_cache[5*8]; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
386 |
2748
135aa7d0bd17
avoid one transpose (730->680 dezicycles on duron)
michael
parents:
2717
diff
changeset
|
387 uint8_t zigzag_scan[16]; |
3174 | 388 uint8_t zigzag_scan8x8[64]; |
389 uint8_t zigzag_scan8x8_cavlc[64]; | |
3316 | 390 uint8_t field_scan[16]; |
391 uint8_t field_scan8x8[64]; | |
392 uint8_t field_scan8x8_cavlc[64]; | |
2763 | 393 const uint8_t *zigzag_scan_q0; |
3174 | 394 const uint8_t *zigzag_scan8x8_q0; |
395 const uint8_t *zigzag_scan8x8_cavlc_q0; | |
3316 | 396 const uint8_t *field_scan_q0; |
397 const uint8_t *field_scan8x8_q0; | |
398 const uint8_t *field_scan8x8_cavlc_q0; | |
2967 | 399 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
400 int x264_build; |
1168 | 401 }H264Context; |
402 | |
403 static VLC coeff_token_vlc[4]; | |
404 static VLC chroma_dc_coeff_token_vlc; | |
405 | |
406 static VLC total_zeros_vlc[15]; | |
407 static VLC chroma_dc_total_zeros_vlc[3]; | |
408 | |
409 static VLC run_vlc[6]; | |
410 static VLC run7_vlc; | |
411 | |
1234 | 412 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp); |
413 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc); | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
414 static void 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); |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
415 static void 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); |
1234 | 416 |
3028 | 417 static always_inline uint32_t pack16to32(int a, int b){ |
1269 | 418 #ifdef WORDS_BIGENDIAN |
419 return (b&0xFFFF) + (a<<16); | |
420 #else | |
421 return (a&0xFFFF) + (b<<16); | |
422 #endif | |
423 } | |
424 | |
1168 | 425 /** |
426 * fill a rectangle. | |
2392 | 427 * @param h height of the rectangle, should be a constant |
428 * @param w width of the rectangle, should be a constant | |
1168 | 429 * @param size the size of val (1 or 4), should be a constant |
430 */ | |
3028 | 431 static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ |
1187 | 432 uint8_t *p= (uint8_t*)vp; |
1168 | 433 assert(size==1 || size==4); |
3315 | 434 assert(w<=4); |
2967 | 435 |
1168 | 436 w *= size; |
437 stride *= size; | |
2967 | 438 |
2962 | 439 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0); |
2669 | 440 assert((stride&(w-1))==0); |
3315 | 441 if(w==2){ |
442 const uint16_t v= size==4 ? val : val*0x0101; | |
443 *(uint16_t*)(p + 0*stride)= v; | |
444 if(h==1) return; | |
445 *(uint16_t*)(p + 1*stride)= v; | |
446 if(h==2) return; | |
1168 | 447 *(uint16_t*)(p + 2*stride)= |
3315 | 448 *(uint16_t*)(p + 3*stride)= v; |
449 }else if(w==4){ | |
450 const uint32_t v= size==4 ? val : val*0x01010101; | |
451 *(uint32_t*)(p + 0*stride)= v; | |
452 if(h==1) return; | |
453 *(uint32_t*)(p + 1*stride)= v; | |
454 if(h==2) return; | |
1168 | 455 *(uint32_t*)(p + 2*stride)= |
3315 | 456 *(uint32_t*)(p + 3*stride)= v; |
457 }else if(w==8){ | |
458 //gcc can't optimize 64bit math on x86_32 | |
459 #if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64) | |
460 const uint64_t v= val*0x0100000001ULL; | |
461 *(uint64_t*)(p + 0*stride)= v; | |
462 if(h==1) return; | |
463 *(uint64_t*)(p + 1*stride)= v; | |
464 if(h==2) return; | |
1168 | 465 *(uint64_t*)(p + 2*stride)= |
3315 | 466 *(uint64_t*)(p + 3*stride)= v; |
467 }else if(w==16){ | |
468 const uint64_t v= val*0x0100000001ULL; | |
1168 | 469 *(uint64_t*)(p + 0+0*stride)= |
470 *(uint64_t*)(p + 8+0*stride)= | |
471 *(uint64_t*)(p + 0+1*stride)= | |
3315 | 472 *(uint64_t*)(p + 8+1*stride)= v; |
473 if(h==2) return; | |
1168 | 474 *(uint64_t*)(p + 0+2*stride)= |
475 *(uint64_t*)(p + 8+2*stride)= | |
476 *(uint64_t*)(p + 0+3*stride)= | |
3315 | 477 *(uint64_t*)(p + 8+3*stride)= v; |
478 #else | |
479 *(uint32_t*)(p + 0+0*stride)= | |
480 *(uint32_t*)(p + 4+0*stride)= val; | |
481 if(h==1) return; | |
482 *(uint32_t*)(p + 0+1*stride)= | |
483 *(uint32_t*)(p + 4+1*stride)= val; | |
484 if(h==2) return; | |
485 *(uint32_t*)(p + 0+2*stride)= | |
486 *(uint32_t*)(p + 4+2*stride)= | |
487 *(uint32_t*)(p + 0+3*stride)= | |
488 *(uint32_t*)(p + 4+3*stride)= val; | |
489 }else if(w==16){ | |
490 *(uint32_t*)(p + 0+0*stride)= | |
491 *(uint32_t*)(p + 4+0*stride)= | |
492 *(uint32_t*)(p + 8+0*stride)= | |
493 *(uint32_t*)(p +12+0*stride)= | |
494 *(uint32_t*)(p + 0+1*stride)= | |
495 *(uint32_t*)(p + 4+1*stride)= | |
496 *(uint32_t*)(p + 8+1*stride)= | |
497 *(uint32_t*)(p +12+1*stride)= val; | |
498 if(h==2) return; | |
499 *(uint32_t*)(p + 0+2*stride)= | |
500 *(uint32_t*)(p + 4+2*stride)= | |
501 *(uint32_t*)(p + 8+2*stride)= | |
502 *(uint32_t*)(p +12+2*stride)= | |
503 *(uint32_t*)(p + 0+3*stride)= | |
504 *(uint32_t*)(p + 4+3*stride)= | |
505 *(uint32_t*)(p + 8+3*stride)= | |
506 *(uint32_t*)(p +12+3*stride)= val; | |
507 #endif | |
1168 | 508 }else |
509 assert(0); | |
3315 | 510 assert(h==4); |
1168 | 511 } |
512 | |
3028 | 513 static void fill_caches(H264Context *h, int mb_type, int for_deblock){ |
1168 | 514 MpegEncContext * const s = &h->s; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
515 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 516 int topleft_xy, top_xy, topright_xy, left_xy[2]; |
517 int topleft_type, top_type, topright_type, left_type[2]; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
518 int left_block[8]; |
1168 | 519 int i; |
520 | |
3482 | 521 //FIXME deblocking could skip the intra and nnz parts. |
522 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF) | |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
523 return; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
524 |
2967 | 525 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it |
526 | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
527 top_xy = mb_xy - s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
528 topleft_xy = top_xy - 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
529 topright_xy= top_xy + 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
530 left_xy[1] = left_xy[0] = mb_xy-1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
531 left_block[0]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
532 left_block[1]= 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
533 left_block[2]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
534 left_block[3]= 3; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
535 left_block[4]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
536 left_block[5]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
537 left_block[6]= 8; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
538 left_block[7]= 11; |
3316 | 539 if(FRAME_MBAFF){ |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
540 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
541 const int top_pair_xy = pair_xy - s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
542 const int topleft_pair_xy = top_pair_xy - 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
543 const int topright_pair_xy = top_pair_xy + 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
544 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
545 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
546 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
547 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
548 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
549 const int bottom = (s->mb_y & 1); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
550 tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag); |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
551 if (bottom |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
552 ? !curr_mb_frame_flag // bottom macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
553 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
554 ) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
555 top_xy -= s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
556 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
557 if (bottom |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
558 ? !curr_mb_frame_flag // bottom macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
559 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
560 ) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
561 topleft_xy -= s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
562 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
563 if (bottom |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
564 ? !curr_mb_frame_flag // bottom macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
565 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
566 ) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
567 topright_xy -= s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
568 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
569 if (left_mb_frame_flag != curr_mb_frame_flag) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
570 left_xy[1] = left_xy[0] = pair_xy - 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
571 if (curr_mb_frame_flag) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
572 if (bottom) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
573 left_block[0]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
574 left_block[1]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
575 left_block[2]= 3; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
576 left_block[3]= 3; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
577 left_block[4]= 8; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
578 left_block[5]= 11; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
579 left_block[6]= 8; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
580 left_block[7]= 11; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
581 } else { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
582 left_block[0]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
583 left_block[1]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
584 left_block[2]= 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
585 left_block[3]= 1; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
586 left_block[4]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
587 left_block[5]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
588 left_block[6]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
589 left_block[7]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
590 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
591 } else { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
592 left_xy[1] += s->mb_stride; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
593 //left_block[0]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
594 left_block[1]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
595 left_block[2]= 0; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
596 left_block[3]= 2; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
597 //left_block[4]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
598 left_block[5]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
599 left_block[6]= 7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
600 left_block[7]= 10; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
601 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
602 } |
1168 | 603 } |
604 | |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
605 h->top_mb_xy = top_xy; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
606 h->left_mb_xy[0] = left_xy[0]; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
607 h->left_mb_xy[1] = left_xy[1]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
608 if(for_deblock){ |
3482 | 609 topleft_type = 0; |
610 topright_type = 0; | |
2449 | 611 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0; |
612 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0; | |
613 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0; | |
3316 | 614 |
615 if(FRAME_MBAFF && !IS_INTRA(mb_type)){ | |
616 int list; | |
617 int v = *(uint16_t*)&h->non_zero_count[mb_xy][14]; | |
618 for(i=0; i<16; i++) | |
619 h->non_zero_count_cache[scan8[i]] = (v>>i)&1; | |
620 for(list=0; list<1+(h->slice_type==B_TYPE); list++){ | |
621 if(USES_LIST(mb_type,list)){ | |
622 uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]]; | |
623 uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]]; | |
3776 | 624 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]]; |
3316 | 625 for(i=0; i<4; i++, dst+=8, src+=h->b_stride){ |
626 dst[0] = src[0]; | |
627 dst[1] = src[1]; | |
628 dst[2] = src[2]; | |
629 dst[3] = src[3]; | |
630 } | |
631 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] = | |
632 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101; | |
633 ref += h->b8_stride; | |
634 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] = | |
635 *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101; | |
636 }else{ | |
637 fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4); | |
638 fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1); | |
639 } | |
640 } | |
641 } | |
2449 | 642 }else{ |
643 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0; | |
644 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0; | |
645 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0; | |
646 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0; | |
647 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0; | |
648 } | |
1168 | 649 |
650 if(IS_INTRA(mb_type)){ | |
2967 | 651 h->topleft_samples_available= |
652 h->top_samples_available= | |
1168 | 653 h->left_samples_available= 0xFFFF; |
654 h->topright_samples_available= 0xEEEA; | |
655 | |
656 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){ | |
657 h->topleft_samples_available= 0xB3FF; | |
658 h->top_samples_available= 0x33FF; | |
659 h->topright_samples_available= 0x26EA; | |
660 } | |
661 for(i=0; i<2; i++){ | |
662 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){ | |
663 h->topleft_samples_available&= 0xDF5F; | |
664 h->left_samples_available&= 0x5F5F; | |
665 } | |
666 } | |
2967 | 667 |
1168 | 668 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred)) |
669 h->topleft_samples_available&= 0x7FFF; | |
2967 | 670 |
1168 | 671 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred)) |
672 h->topright_samples_available&= 0xFBFF; | |
2967 | 673 |
1168 | 674 if(IS_INTRA4x4(mb_type)){ |
675 if(IS_INTRA4x4(top_type)){ | |
676 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4]; | |
677 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5]; | |
678 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6]; | |
679 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3]; | |
680 }else{ | |
681 int pred; | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
682 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred)) |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
683 pred= -1; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
684 else{ |
1168 | 685 pred= 2; |
686 } | |
687 h->intra4x4_pred_mode_cache[4+8*0]= | |
688 h->intra4x4_pred_mode_cache[5+8*0]= | |
689 h->intra4x4_pred_mode_cache[6+8*0]= | |
690 h->intra4x4_pred_mode_cache[7+8*0]= pred; | |
691 } | |
692 for(i=0; i<2; i++){ | |
693 if(IS_INTRA4x4(left_type[i])){ | |
694 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]]; | |
695 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]]; | |
696 }else{ | |
697 int pred; | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
698 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred)) |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
699 pred= -1; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
700 else{ |
1168 | 701 pred= 2; |
702 } | |
703 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= | |
704 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred; | |
705 } | |
706 } | |
707 } | |
708 } | |
2967 | 709 |
710 | |
1168 | 711 /* |
2967 | 712 0 . T T. T T T T |
713 1 L . .L . . . . | |
714 2 L . .L . . . . | |
715 3 . T TL . . . . | |
716 4 L . .L . . . . | |
717 5 L . .. . . . . | |
1168 | 718 */ |
719 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec) | |
720 if(top_type){ | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
721 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
722 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
723 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
724 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3]; |
2967 | 725 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
726 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
727 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8]; |
2967 | 728 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
729 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
730 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11]; |
2967 | 731 |
1168 | 732 }else{ |
2967 | 733 h->non_zero_count_cache[4+8*0]= |
1168 | 734 h->non_zero_count_cache[5+8*0]= |
735 h->non_zero_count_cache[6+8*0]= | |
736 h->non_zero_count_cache[7+8*0]= | |
2967 | 737 |
1168 | 738 h->non_zero_count_cache[1+8*0]= |
739 h->non_zero_count_cache[2+8*0]= | |
2967 | 740 |
1168 | 741 h->non_zero_count_cache[1+8*3]= |
2314 | 742 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64; |
2967 | 743 |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
744 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
745 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
746 for (i=0; i<2; i++) { |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
747 if(left_type[i]){ |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
748 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
749 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
750 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
751 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
752 }else{ |
2967 | 753 h->non_zero_count_cache[3+8*1 + 2*8*i]= |
754 h->non_zero_count_cache[3+8*2 + 2*8*i]= | |
755 h->non_zero_count_cache[0+8*1 + 8*i]= | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
756 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64; |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
757 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
758 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
759 |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
760 if( h->pps.cabac ) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
761 // top_cbp |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
762 if(top_type) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
763 h->top_cbp = h->cbp_table[top_xy]; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
764 } else if(IS_INTRA(mb_type)) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
765 h->top_cbp = 0x1C0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
766 } else { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
767 h->top_cbp = 0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
768 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
769 // left_cbp |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
770 if (left_type[0]) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
771 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
772 } else if(IS_INTRA(mb_type)) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
773 h->left_cbp = 0x1C0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
774 } else { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
775 h->left_cbp = 0; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
776 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
777 if (left_type[0]) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
778 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
779 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
780 if (left_type[1]) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
781 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
782 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
783 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
784 |
1168 | 785 #if 1 |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
786 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){ |
1168 | 787 int list; |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
788 for(list=0; list<1+(h->slice_type==B_TYPE); list++){ |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
789 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){ |
1168 | 790 /*if(!h->mv_cache_clean[list]){ |
791 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all? | |
792 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t)); | |
793 h->mv_cache_clean[list]= 1; | |
794 }*/ | |
2396 | 795 continue; |
1168 | 796 } |
797 h->mv_cache_clean[list]= 0; | |
2967 | 798 |
3212 | 799 if(USES_LIST(top_type, list)){ |
1168 | 800 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; |
801 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride; | |
802 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0]; | |
803 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1]; | |
804 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2]; | |
805 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3]; | |
806 h->ref_cache[list][scan8[0] + 0 - 1*8]= | |
807 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0]; | |
808 h->ref_cache[list][scan8[0] + 2 - 1*8]= | |
809 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1]; | |
810 }else{ | |
2967 | 811 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= |
812 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= | |
813 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= | |
1168 | 814 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0; |
815 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101; | |
816 } | |
817 | |
818 //FIXME unify cleanup or sth | |
3212 | 819 if(USES_LIST(left_type[0], list)){ |
1168 | 820 const int b_xy= h->mb2b_xy[left_xy[0]] + 3; |
821 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1; | |
822 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]; | |
823 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]]; | |
3316 | 824 h->ref_cache[list][scan8[0] - 1 + 0*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)]; |
825 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1]>>1)]; | |
1168 | 826 }else{ |
827 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]= | |
828 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0; | |
829 h->ref_cache[list][scan8[0] - 1 + 0*8]= | |
830 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
831 } | |
2967 | 832 |
3212 | 833 if(USES_LIST(left_type[1], list)){ |
1168 | 834 const int b_xy= h->mb2b_xy[left_xy[1]] + 3; |
835 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1; | |
836 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]]; | |
837 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]]; | |
3316 | 838 h->ref_cache[list][scan8[0] - 1 + 2*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)]; |
839 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[3]>>1)]; | |
1168 | 840 }else{ |
841 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]= | |
842 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0; | |
843 h->ref_cache[list][scan8[0] - 1 + 2*8]= | |
844 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE; | |
2669 | 845 assert((!left_type[0]) == (!left_type[1])); |
1168 | 846 } |
847 | |
3519 | 848 if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF) |
2449 | 849 continue; |
850 | |
3212 | 851 if(USES_LIST(topleft_type, list)){ |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
852 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
853 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
854 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
855 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
856 }else{ |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
857 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
858 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
859 } |
2967 | 860 |
3212 | 861 if(USES_LIST(topright_type, list)){ |
2623
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
862 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
863 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
864 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
865 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy]; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
866 }else{ |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
867 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
868 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; |
aaf8e94bce00
deblocking filter doesn't need to call fill_caches again. 1.4% faster decoding.
lorenm
parents:
2597
diff
changeset
|
869 } |
2967 | 870 |
3519 | 871 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF) |
3482 | 872 continue; |
2967 | 873 |
874 h->ref_cache[list][scan8[5 ]+1] = | |
875 h->ref_cache[list][scan8[7 ]+1] = | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
876 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else) |
2967 | 877 h->ref_cache[list][scan8[4 ]] = |
1168 | 878 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE; |
879 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]= | |
880 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]= | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
881 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else) |
1168 | 882 *(uint32_t*)h->mv_cache [list][scan8[4 ]]= |
883 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0; | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
884 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
885 if( h->pps.cabac ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
886 /* XXX beurk, Load mvd */ |
3212 | 887 if(USES_LIST(top_type, list)){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
888 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
889 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
890 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
891 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
892 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
893 }else{ |
2967 | 894 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= |
895 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= | |
896 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
897 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
898 } |
3212 | 899 if(USES_LIST(left_type[0], list)){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
900 const int b_xy= h->mb2b_xy[left_xy[0]] + 3; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
901 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
902 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
903 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
904 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
905 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
906 } |
3212 | 907 if(USES_LIST(left_type[1], list)){ |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
908 const int b_xy= h->mb2b_xy[left_xy[1]] + 3; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
909 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
910 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
911 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
912 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
913 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
914 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
915 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
916 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]= |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
917 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else) |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
918 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
919 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0; |
2396 | 920 |
921 if(h->slice_type == B_TYPE){ | |
922 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1); | |
923 | |
924 if(IS_DIRECT(top_type)){ | |
925 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101; | |
926 }else if(IS_8X8(top_type)){ | |
927 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride; | |
928 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy]; | |
929 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1]; | |
930 }else{ | |
931 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0; | |
932 } | |
2967 | 933 |
3316 | 934 if(IS_DIRECT(left_type[0])) |
935 h->direct_cache[scan8[0] - 1 + 0*8]= 1; | |
936 else if(IS_8X8(left_type[0])) | |
937 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)]; | |
938 else | |
939 h->direct_cache[scan8[0] - 1 + 0*8]= 0; | |
940 | |
941 if(IS_DIRECT(left_type[1])) | |
2396 | 942 h->direct_cache[scan8[0] - 1 + 2*8]= 1; |
3316 | 943 else if(IS_8X8(left_type[1])) |
944 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)]; | |
945 else | |
2396 | 946 h->direct_cache[scan8[0] - 1 + 2*8]= 0; |
3316 | 947 } |
948 } | |
949 | |
950 if(FRAME_MBAFF){ | |
951 #define MAP_MVS\ | |
952 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\ | |
953 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\ | |
954 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\ | |
955 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\ | |
956 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\ | |
957 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\ | |
958 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\ | |
959 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\ | |
960 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\ | |
961 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1]) | |
962 if(MB_FIELD){ | |
963 #define MAP_F2F(idx, mb_type)\ | |
964 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ | |
965 h->ref_cache[list][idx] <<= 1;\ | |
966 h->mv_cache[list][idx][1] /= 2;\ | |
967 h->mvd_cache[list][idx][1] /= 2;\ | |
2396 | 968 } |
3316 | 969 MAP_MVS |
970 #undef MAP_F2F | |
971 }else{ | |
972 #define MAP_F2F(idx, mb_type)\ | |
973 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ | |
974 h->ref_cache[list][idx] >>= 1;\ | |
975 h->mv_cache[list][idx][1] <<= 1;\ | |
976 h->mvd_cache[list][idx][1] <<= 1;\ | |
977 } | |
978 MAP_MVS | |
979 #undef MAP_F2F | |
2396 | 980 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
981 } |
1168 | 982 } |
983 } | |
984 #endif | |
2755 | 985 |
986 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]); | |
1168 | 987 } |
988 | |
989 static inline void write_back_intra_pred_mode(H264Context *h){ | |
990 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
991 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 992 |
993 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1]; | |
994 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2]; | |
995 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3]; | |
996 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4]; | |
997 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4]; | |
998 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4]; | |
999 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4]; | |
1000 } | |
1001 | |
1002 /** | |
1003 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
1004 */ | |
1005 static inline int check_intra4x4_pred_mode(H264Context *h){ | |
1006 MpegEncContext * const s = &h->s; | |
1007 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0}; | |
1008 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED}; | |
1009 int i; | |
2967 | 1010 |
1168 | 1011 if(!(h->top_samples_available&0x8000)){ |
1012 for(i=0; i<4; i++){ | |
1013 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ]; | |
1014 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
1015 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); |
1168 | 1016 return -1; |
1017 } else if(status){ | |
1018 h->intra4x4_pred_mode_cache[scan8[0] + i]= status; | |
1019 } | |
1020 } | |
1021 } | |
2967 | 1022 |
1168 | 1023 if(!(h->left_samples_available&0x8000)){ |
1024 for(i=0; i<4; i++){ | |
1025 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ]; | |
1026 if(status<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
1027 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y); |
1168 | 1028 return -1; |
1029 } else if(status){ | |
1030 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status; | |
1031 } | |
1032 } | |
1033 } | |
1034 | |
1035 return 0; | |
1036 } //FIXME cleanup like next | |
1037 | |
1038 /** | |
1039 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. | |
1040 */ | |
1041 static inline int check_intra_pred_mode(H264Context *h, int mode){ | |
1042 MpegEncContext * const s = &h->s; | |
1043 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1}; | |
1044 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8}; | |
2967 | 1045 |
2392 | 1046 if(mode < 0 || mode > 6) { |
1047 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y); | |
2163 | 1048 return -1; |
2392 | 1049 } |
2967 | 1050 |
1168 | 1051 if(!(h->top_samples_available&0x8000)){ |
1052 mode= top[ mode ]; | |
1053 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
1054 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); |
1168 | 1055 return -1; |
1056 } | |
1057 } | |
2967 | 1058 |
1168 | 1059 if(!(h->left_samples_available&0x8000)){ |
1060 mode= left[ mode ]; | |
1061 if(mode<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
1062 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y); |
1168 | 1063 return -1; |
2967 | 1064 } |
1168 | 1065 } |
1066 | |
1067 return mode; | |
1068 } | |
1069 | |
1070 /** | |
1071 * gets the predicted intra4x4 prediction mode. | |
1072 */ | |
1073 static inline int pred_intra_mode(H264Context *h, int n){ | |
1074 const int index8= scan8[n]; | |
1075 const int left= h->intra4x4_pred_mode_cache[index8 - 1]; | |
1076 const int top = h->intra4x4_pred_mode_cache[index8 - 8]; | |
1077 const int min= FFMIN(left, top); | |
1078 | |
1170 | 1079 tprintf("mode:%d %d min:%d\n", left ,top, min); |
1168 | 1080 |
1081 if(min<0) return DC_PRED; | |
1082 else return min; | |
1083 } | |
1084 | |
1085 static inline void write_back_non_zero_count(H264Context *h){ | |
1086 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
1087 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
1088 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1089 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1090 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1091 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
1092 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4]; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1093 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1094 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1095 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4]; |
2967 | 1096 |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1097 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
1098 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2]; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1099 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1]; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1100 |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1101 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
1102 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5]; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
1103 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4]; |
3316 | 1104 |
1105 if(FRAME_MBAFF){ | |
1106 // store all luma nnzs, for deblocking | |
1107 int v = 0, i; | |
1108 for(i=0; i<16; i++) | |
1109 v += (!!h->non_zero_count_cache[scan8[i]]) << i; | |
1110 *(uint16_t*)&h->non_zero_count[mb_xy][14] = v; | |
1111 } | |
1168 | 1112 } |
1113 | |
1114 /** | |
1115 * gets the predicted number of non zero coefficients. | |
1116 * @param n block index | |
1117 */ | |
1118 static inline int pred_non_zero_count(H264Context *h, int n){ | |
1119 const int index8= scan8[n]; | |
1120 const int left= h->non_zero_count_cache[index8 - 1]; | |
1121 const int top = h->non_zero_count_cache[index8 - 8]; | |
1122 int i= left + top; | |
2967 | 1123 |
1168 | 1124 if(i<64) i= (i+1)>>1; |
1125 | |
1170 | 1126 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31); |
1168 | 1127 |
1128 return i&31; | |
1129 } | |
1130 | |
1169 | 1131 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){ |
1132 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ]; | |
1133 | |
3316 | 1134 /* there is no consistent mapping of mvs to neighboring locations that will |
1135 * make mbaff happy, so we can't move all this logic to fill_caches */ | |
1136 if(FRAME_MBAFF){ | |
1137 MpegEncContext *s = &h->s; | |
3776 | 1138 const uint32_t *mb_types = s->current_picture_ptr->mb_type; |
3316 | 1139 const int16_t *mv; |
1140 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0; | |
1141 *C = h->mv_cache[list][scan8[0]-2]; | |
1142 | |
1143 if(!MB_FIELD | |
1144 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){ | |
1145 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3); | |
1146 if(IS_INTERLACED(mb_types[topright_xy])){ | |
1147 #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\ | |
1148 const int x4 = X4, y4 = Y4;\ | |
1149 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\ | |
1150 if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\ | |
1151 return LIST_NOT_USED;\ | |
1152 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\ | |
1153 h->mv_cache[list][scan8[0]-2][0] = mv[0];\ | |
1154 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\ | |
1155 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP; | |
1156 | |
1157 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1); | |
1158 } | |
1159 } | |
1160 if(topright_ref == PART_NOT_AVAILABLE | |
1161 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4 | |
1162 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){ | |
1163 if(!MB_FIELD | |
1164 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){ | |
1165 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1); | |
1166 } | |
1167 if(MB_FIELD | |
1168 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]]) | |
1169 && i >= scan8[0]+8){ | |
1170 // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok. | |
1171 SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2); | |
1172 } | |
1173 } | |
1174 #undef SET_DIAG_MV | |
1175 } | |
1176 | |
1169 | 1177 if(topright_ref != PART_NOT_AVAILABLE){ |
1178 *C= h->mv_cache[list][ i - 8 + part_width ]; | |
1179 return topright_ref; | |
1180 }else{ | |
1170 | 1181 tprintf("topright MV not available\n"); |
1182 | |
1169 | 1183 *C= h->mv_cache[list][ i - 8 - 1 ]; |
1184 return h->ref_cache[list][ i - 8 - 1 ]; | |
1185 } | |
1186 } | |
1187 | |
1168 | 1188 /** |
1189 * gets the predicted MV. | |
1190 * @param n the block index | |
1191 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4) | |
1192 * @param mx the x component of the predicted motion vector | |
1193 * @param my the y component of the predicted motion vector | |
1194 */ | |
1195 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){ | |
1196 const int index8= scan8[n]; | |
1197 const int top_ref= h->ref_cache[list][ index8 - 8 ]; | |
1198 const int left_ref= h->ref_cache[list][ index8 - 1 ]; | |
1199 const int16_t * const A= h->mv_cache[list][ index8 - 1 ]; | |
1200 const int16_t * const B= h->mv_cache[list][ index8 - 8 ]; | |
1169 | 1201 const int16_t * C; |
1202 int diagonal_ref, match_count; | |
1203 | |
1168 | 1204 assert(part_width==1 || part_width==2 || part_width==4); |
1169 | 1205 |
1168 | 1206 /* mv_cache |
2967 | 1207 B . . A T T T T |
1168 | 1208 U . . L . . , . |
1209 U . . L . . . . | |
1210 U . . L . . , . | |
1211 . . . L . . . . | |
1212 */ | |
1169 | 1213 |
1214 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width); | |
1215 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref); | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
1216 tprintf("pred_motion match_count=%d\n", match_count); |
1169 | 1217 if(match_count > 1){ //most common |
1218 *mx= mid_pred(A[0], B[0], C[0]); | |
1219 *my= mid_pred(A[1], B[1], C[1]); | |
1220 }else if(match_count==1){ | |
1221 if(left_ref==ref){ | |
1222 *mx= A[0]; | |
2967 | 1223 *my= A[1]; |
1169 | 1224 }else if(top_ref==ref){ |
1225 *mx= B[0]; | |
2967 | 1226 *my= B[1]; |
1169 | 1227 }else{ |
1228 *mx= C[0]; | |
2967 | 1229 *my= C[1]; |
1169 | 1230 } |
1231 }else{ | |
1232 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ | |
1233 *mx= A[0]; | |
2967 | 1234 *my= A[1]; |
1168 | 1235 }else{ |
1236 *mx= mid_pred(A[0], B[0], C[0]); | |
1237 *my= mid_pred(A[1], B[1], C[1]); | |
1238 } | |
1169 | 1239 } |
2967 | 1240 |
1187 | 1241 tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list); |
1168 | 1242 } |
1243 | |
1244 /** | |
1245 * gets the directionally predicted 16x8 MV. | |
1246 * @param n the block index | |
1247 * @param mx the x component of the predicted motion vector | |
1248 * @param my the y component of the predicted motion vector | |
1249 */ | |
1250 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
1251 if(n==0){ | |
1252 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ]; | |
1253 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ]; | |
1254 | |
2402
f9d4e1eddbc5
- correct several errors on the deblocking accross slice boundaries.
michael
parents:
2401
diff
changeset
|
1255 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list); |
2967 | 1256 |
1168 | 1257 if(top_ref == ref){ |
1258 *mx= B[0]; | |
1259 *my= B[1]; | |
1260 return; | |
1261 } | |
1262 }else{ | |
1263 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ]; | |
1264 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ]; | |
2967 | 1265 |
2402
f9d4e1eddbc5
- correct several errors on the deblocking accross slice boundaries.
michael
parents:
2401
diff
changeset
|
1266 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 1267 |
1268 if(left_ref == ref){ | |
1269 *mx= A[0]; | |
1270 *my= A[1]; | |
1271 return; | |
1272 } | |
1273 } | |
1274 | |
1275 //RARE | |
1276 pred_motion(h, n, 4, list, ref, mx, my); | |
1277 } | |
1278 | |
1279 /** | |
1280 * gets the directionally predicted 8x16 MV. | |
1281 * @param n the block index | |
1282 * @param mx the x component of the predicted motion vector | |
1283 * @param my the y component of the predicted motion vector | |
1284 */ | |
1285 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){ | |
1286 if(n==0){ | |
1287 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ]; | |
1288 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ]; | |
2967 | 1289 |
2402
f9d4e1eddbc5
- correct several errors on the deblocking accross slice boundaries.
michael
parents:
2401
diff
changeset
|
1290 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 1291 |
1292 if(left_ref == ref){ | |
1293 *mx= A[0]; | |
1294 *my= A[1]; | |
1295 return; | |
1296 } | |
1297 }else{ | |
1169 | 1298 const int16_t * C; |
1299 int diagonal_ref; | |
1300 | |
1301 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2); | |
2967 | 1302 |
2402
f9d4e1eddbc5
- correct several errors on the deblocking accross slice boundaries.
michael
parents:
2401
diff
changeset
|
1303 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list); |
1168 | 1304 |
2967 | 1305 if(diagonal_ref == ref){ |
1168 | 1306 *mx= C[0]; |
1307 *my= C[1]; | |
1308 return; | |
1309 } | |
1310 } | |
1311 | |
1312 //RARE | |
1313 pred_motion(h, n, 2, list, ref, mx, my); | |
1314 } | |
1315 | |
1316 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){ | |
1317 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ]; | |
1318 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ]; | |
1319 | |
2392 | 1320 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y); |
1168 | 1321 |
1322 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE | |
1323 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0) | |
1324 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){ | |
2967 | 1325 |
1168 | 1326 *mx = *my = 0; |
1327 return; | |
1328 } | |
2967 | 1329 |
1168 | 1330 pred_motion(h, 0, 4, 0, 0, mx, my); |
1331 | |
1332 return; | |
1333 } | |
1334 | |
2396 | 1335 static inline void direct_dist_scale_factor(H264Context * const h){ |
1336 const int poc = h->s.current_picture_ptr->poc; | |
1337 const int poc1 = h->ref_list[1][0].poc; | |
1338 int i; | |
1339 for(i=0; i<h->ref_count[0]; i++){ | |
1340 int poc0 = h->ref_list[0][i].poc; | |
1341 int td = clip(poc1 - poc0, -128, 127); | |
1342 if(td == 0 /* FIXME || pic0 is a long-term ref */){ | |
1343 h->dist_scale_factor[i] = 256; | |
1344 }else{ | |
1345 int tb = clip(poc - poc0, -128, 127); | |
4001 | 1346 int tx = (16384 + (FFABS(td) >> 1)) / td; |
2396 | 1347 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023); |
1348 } | |
1349 } | |
3316 | 1350 if(FRAME_MBAFF){ |
1351 for(i=0; i<h->ref_count[0]; i++){ | |
1352 h->dist_scale_factor_field[2*i] = | |
1353 h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i]; | |
1354 } | |
1355 } | |
2396 | 1356 } |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1357 static inline void direct_ref_list_init(H264Context * const h){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1358 MpegEncContext * const s = &h->s; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1359 Picture * const ref1 = &h->ref_list[1][0]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1360 Picture * const cur = s->current_picture_ptr; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1361 int list, i, j; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1362 if(cur->pict_type == I_TYPE) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1363 cur->ref_count[0] = 0; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1364 if(cur->pict_type != B_TYPE) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1365 cur->ref_count[1] = 0; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1366 for(list=0; list<2; list++){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1367 cur->ref_count[list] = h->ref_count[list]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1368 for(j=0; j<h->ref_count[list]; j++) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1369 cur->ref_poc[list][j] = h->ref_list[list][j].poc; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1370 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1371 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1372 return; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1373 for(list=0; list<2; list++){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1374 for(i=0; i<ref1->ref_count[list]; i++){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1375 const int poc = ref1->ref_poc[list][i]; |
3125 | 1376 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */ |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1377 for(j=0; j<h->ref_count[list]; j++) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1378 if(h->ref_list[list][j].poc == poc){ |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1379 h->map_col_to_list0[list][i] = j; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1380 break; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1381 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1382 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1383 } |
3316 | 1384 if(FRAME_MBAFF){ |
1385 for(list=0; list<2; list++){ | |
1386 for(i=0; i<ref1->ref_count[list]; i++){ | |
1387 j = h->map_col_to_list0[list][i]; | |
1388 h->map_col_to_list0_field[list][2*i] = 2*j; | |
1389 h->map_col_to_list0_field[list][2*i+1] = 2*j+1; | |
1390 } | |
1391 } | |
1392 } | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1393 } |
2396 | 1394 |
1395 static inline void pred_direct_motion(H264Context * const h, int *mb_type){ | |
1396 MpegEncContext * const s = &h->s; | |
1397 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride; | |
1398 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride; | |
1399 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
1400 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy]; | |
1401 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy]; | |
2808
932ec96933f5
fixing colocated mv if colocated block is L1 predicted
michael
parents:
2803
diff
changeset
|
1402 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy]; |
2396 | 1403 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy]; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1404 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy]; |
2396 | 1405 const int is_b8x8 = IS_8X8(*mb_type); |
1406 int sub_mb_type; | |
1407 int i8, i4; | |
1408 | |
3316 | 1409 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM) |
2396 | 1410 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){ |
1411 /* FIXME save sub mb types from previous frames (or derive from MVs) | |
1412 * so we know exactly what block size to use */ | |
1413 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */ | |
2536 | 1414 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1; |
3316 | 1415 }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){ |
2396 | 1416 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ |
1417 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */ | |
1418 }else{ | |
1419 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */ | |
2536 | 1420 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1; |
2396 | 1421 } |
1422 if(!is_b8x8) | |
1423 *mb_type |= MB_TYPE_DIRECT2; | |
3316 | 1424 if(MB_FIELD) |
1425 *mb_type |= MB_TYPE_INTERLACED; | |
2396 | 1426 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
1427 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col); |
2967 | 1428 |
2396 | 1429 if(h->direct_spatial_mv_pred){ |
1430 int ref[2]; | |
1431 int mv[2][2]; | |
1432 int list; | |
1433 | |
3316 | 1434 /* FIXME interlacing + spatial direct uses wrong colocated block positions */ |
1435 | |
2396 | 1436 /* ref = min(neighbors) */ |
1437 for(list=0; list<2; list++){ | |
1438 int refa = h->ref_cache[list][scan8[0] - 1]; | |
1439 int refb = h->ref_cache[list][scan8[0] - 8]; | |
1440 int refc = h->ref_cache[list][scan8[0] - 8 + 4]; | |
1441 if(refc == -2) | |
1442 refc = h->ref_cache[list][scan8[0] - 8 - 1]; | |
1443 ref[list] = refa; | |
1444 if(ref[list] < 0 || (refb < ref[list] && refb >= 0)) | |
1445 ref[list] = refb; | |
1446 if(ref[list] < 0 || (refc < ref[list] && refc >= 0)) | |
1447 ref[list] = refc; | |
1448 if(ref[list] < 0) | |
1449 ref[list] = -1; | |
1450 } | |
1451 | |
1452 if(ref[0] < 0 && ref[1] < 0){ | |
1453 ref[0] = ref[1] = 0; | |
1454 mv[0][0] = mv[0][1] = | |
1455 mv[1][0] = mv[1][1] = 0; | |
1456 }else{ | |
1457 for(list=0; list<2; list++){ | |
1458 if(ref[list] >= 0) | |
1459 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]); | |
1460 else | |
1461 mv[list][0] = mv[list][1] = 0; | |
1462 } | |
1463 } | |
1464 | |
1465 if(ref[1] < 0){ | |
1466 *mb_type &= ~MB_TYPE_P0L1; | |
1467 sub_mb_type &= ~MB_TYPE_P0L1; | |
1468 }else if(ref[0] < 0){ | |
1469 *mb_type &= ~MB_TYPE_P0L0; | |
1470 sub_mb_type &= ~MB_TYPE_P0L0; | |
1471 } | |
1472 | |
1473 if(IS_16X16(*mb_type)){ | |
3001
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1474 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1); |
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1475 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1); |
2967 | 1476 if(!IS_INTRA(mb_type_col) |
4001 | 1477 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1) |
1478 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1 | |
2834 | 1479 && (h->x264_build>33 || !h->x264_build)))){ |
2396 | 1480 if(ref[0] > 0) |
1481 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4); | |
1482 else | |
1483 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); | |
1484 if(ref[1] > 0) | |
1485 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4); | |
1486 else | |
1487 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4); | |
1488 }else{ | |
1489 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4); | |
1490 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4); | |
1491 } | |
1492 }else{ | |
1493 for(i8=0; i8<4; i8++){ | |
1494 const int x8 = i8&1; | |
1495 const int y8 = i8>>1; | |
2967 | 1496 |
2396 | 1497 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) |
1498 continue; | |
1499 h->sub_mb_type[i8] = sub_mb_type; | |
2967 | 1500 |
2396 | 1501 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4); |
1502 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4); | |
3001
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1503 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1); |
b52d8ee430f6
fix some potential arithmetic overflows in pred_direct_motion() and
lorenm
parents:
2979
diff
changeset
|
1504 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1); |
2967 | 1505 |
2396 | 1506 /* col_zero_flag */ |
2967 | 1507 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0 |
1508 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0 | |
2834 | 1509 && (h->x264_build>33 || !h->x264_build)))){ |
2808
932ec96933f5
fixing colocated mv if colocated block is L1 predicted
michael
parents:
2803
diff
changeset
|
1510 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1; |
3002 | 1511 if(IS_SUB_8X8(sub_mb_type)){ |
1512 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride]; | |
4001 | 1513 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ |
3002 | 1514 if(ref[0] == 0) |
1515 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1516 if(ref[1] == 0) | |
1517 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1518 } | |
1519 }else | |
2396 | 1520 for(i4=0; i4<4; i4++){ |
2808
932ec96933f5
fixing colocated mv if colocated block is L1 predicted
michael
parents:
2803
diff
changeset
|
1521 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride]; |
4001 | 1522 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){ |
2396 | 1523 if(ref[0] == 0) |
1524 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0; | |
1525 if(ref[1] == 0) | |
1526 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0; | |
1527 } | |
1528 } | |
1529 } | |
1530 } | |
1531 } | |
1532 }else{ /* direct temporal mv pred */ | |
3316 | 1533 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]}; |
1534 const int *dist_scale_factor = h->dist_scale_factor; | |
1535 | |
1536 if(FRAME_MBAFF){ | |
1537 if(IS_INTERLACED(*mb_type)){ | |
1538 map_col_to_list0[0] = h->map_col_to_list0_field[0]; | |
1539 map_col_to_list0[1] = h->map_col_to_list0_field[1]; | |
1540 dist_scale_factor = h->dist_scale_factor_field; | |
1541 } | |
1542 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){ | |
1543 /* FIXME assumes direct_8x8_inference == 1 */ | |
1544 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride; | |
1545 int mb_types_col[2]; | |
1546 int y_shift; | |
1547 | |
1548 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1 | |
1549 | (is_b8x8 ? 0 : MB_TYPE_DIRECT2) | |
1550 | (*mb_type & MB_TYPE_INTERLACED); | |
1551 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16; | |
1552 | |
1553 if(IS_INTERLACED(*mb_type)){ | |
1554 /* frame to field scaling */ | |
1555 mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy]; | |
1556 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride]; | |
1557 if(s->mb_y&1){ | |
1558 l1ref0 -= 2*h->b8_stride; | |
1559 l1ref1 -= 2*h->b8_stride; | |
1560 l1mv0 -= 4*h->b_stride; | |
1561 l1mv1 -= 4*h->b_stride; | |
1562 } | |
1563 y_shift = 0; | |
1564 | |
1565 if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA) | |
1566 && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA) | |
1567 && !is_b8x8) | |
1568 *mb_type |= MB_TYPE_16x8; | |
1569 else | |
1570 *mb_type |= MB_TYPE_8x8; | |
1571 }else{ | |
1572 /* field to frame scaling */ | |
1573 /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1) | |
1574 * but in MBAFF, top and bottom POC are equal */ | |
1575 int dy = (s->mb_y&1) ? 1 : 2; | |
1576 mb_types_col[0] = | |
1577 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride]; | |
1578 l1ref0 += dy*h->b8_stride; | |
1579 l1ref1 += dy*h->b8_stride; | |
1580 l1mv0 += 2*dy*h->b_stride; | |
1581 l1mv1 += 2*dy*h->b_stride; | |
1582 y_shift = 2; | |
1583 | |
1584 if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8)) | |
1585 && !is_b8x8) | |
1586 *mb_type |= MB_TYPE_16x16; | |
1587 else | |
1588 *mb_type |= MB_TYPE_8x8; | |
1589 } | |
1590 | |
1591 for(i8=0; i8<4; i8++){ | |
1592 const int x8 = i8&1; | |
1593 const int y8 = i8>>1; | |
1594 int ref0, scale; | |
1595 const int16_t (*l1mv)[2]= l1mv0; | |
1596 | |
1597 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) | |
1598 continue; | |
1599 h->sub_mb_type[i8] = sub_mb_type; | |
1600 | |
1601 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); | |
1602 if(IS_INTRA(mb_types_col[y8])){ | |
1603 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); | |
1604 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1605 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1606 continue; | |
1607 } | |
1608 | |
1609 ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride]; | |
1610 if(ref0 >= 0) | |
1611 ref0 = map_col_to_list0[0][ref0*2>>y_shift]; | |
1612 else{ | |
1613 ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift]; | |
1614 l1mv= l1mv1; | |
1615 } | |
1616 scale = dist_scale_factor[ref0]; | |
1617 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); | |
1618 | |
1619 { | |
1620 const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride]; | |
1621 int my_col = (mv_col[1]<<y_shift)/2; | |
1622 int mx = (scale * mv_col[0] + 128) >> 8; | |
1623 int my = (scale * my_col + 128) >> 8; | |
1624 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); | |
1625 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4); | |
1626 } | |
1627 } | |
1628 return; | |
1629 } | |
1630 } | |
1631 | |
1632 /* one-to-one mv scaling */ | |
1633 | |
2396 | 1634 if(IS_16X16(*mb_type)){ |
1635 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1); | |
1636 if(IS_INTRA(mb_type_col)){ | |
1637 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); | |
1638 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4); | |
1639 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4); | |
1640 }else{ | |
3316 | 1641 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]] |
1642 : map_col_to_list0[1][l1ref1[0]]; | |
1643 const int scale = dist_scale_factor[ref0]; | |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1644 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0]; |
2396 | 1645 int mv_l0[2]; |
3316 | 1646 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; |
1647 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; | |
2396 | 1648 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1); |
1649 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4); | |
1650 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4); | |
1651 } | |
1652 }else{ | |
1653 for(i8=0; i8<4; i8++){ | |
1654 const int x8 = i8&1; | |
1655 const int y8 = i8>>1; | |
3316 | 1656 int ref0, scale; |
2834 | 1657 const int16_t (*l1mv)[2]= l1mv0; |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1658 |
2396 | 1659 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8])) |
1660 continue; | |
1661 h->sub_mb_type[i8] = sub_mb_type; | |
3316 | 1662 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1); |
2396 | 1663 if(IS_INTRA(mb_type_col)){ |
1664 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1); | |
1665 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1666 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4); | |
1667 continue; | |
1668 } | |
2967 | 1669 |
2396 | 1670 ref0 = l1ref0[x8 + y8*h->b8_stride]; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
1671 if(ref0 >= 0) |
3316 | 1672 ref0 = map_col_to_list0[0][ref0]; |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1673 else{ |
3316 | 1674 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]]; |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1675 l1mv= l1mv1; |
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1676 } |
3316 | 1677 scale = dist_scale_factor[ref0]; |
2967 | 1678 |
2396 | 1679 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1); |
3002 | 1680 if(IS_SUB_8X8(sub_mb_type)){ |
1681 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride]; | |
3316 | 1682 int mx = (scale * mv_col[0] + 128) >> 8; |
1683 int my = (scale * mv_col[1] + 128) >> 8; | |
3002 | 1684 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4); |
1685 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4); | |
1686 }else | |
2396 | 1687 for(i4=0; i4<4; i4++){ |
2809
75400dfbe117
fixing colocated mv if colocated block is L1 predicted for the temporal direct case
michael
parents:
2808
diff
changeset
|
1688 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride]; |
2396 | 1689 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]]; |
3316 | 1690 mv_l0[0] = (scale * mv_col[0] + 128) >> 8; |
1691 mv_l0[1] = (scale * mv_col[1] + 128) >> 8; | |
2396 | 1692 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = |
1693 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]); | |
1694 } | |
1695 } | |
1696 } | |
1697 } | |
1698 } | |
1699 | |
1168 | 1700 static inline void write_back_motion(H264Context *h, int mb_type){ |
1701 MpegEncContext * const s = &h->s; | |
1702 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; | |
1703 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride; | |
1704 int list; | |
1705 | |
3219
cf233efbcece
10l in 1.199 (write_back_motion with B-pyramid + spatial)
lorenm
parents:
3215
diff
changeset
|
1706 if(!USES_LIST(mb_type, 0)) |
cf233efbcece
10l in 1.199 (write_back_motion with B-pyramid + spatial)
lorenm
parents:
3215
diff
changeset
|
1707 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1); |
cf233efbcece
10l in 1.199 (write_back_motion with B-pyramid + spatial)
lorenm
parents:
3215
diff
changeset
|
1708 |
1168 | 1709 for(list=0; list<2; list++){ |
1710 int y; | |
3212 | 1711 if(!USES_LIST(mb_type, list)) |
2396 | 1712 continue; |
2967 | 1713 |
1168 | 1714 for(y=0; y<4; y++){ |
1715 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y]; | |
1716 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y]; | |
1717 } | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1718 if( h->pps.cabac ) { |
3927 | 1719 if(IS_SKIP(mb_type)) |
1720 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4); | |
1721 else | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1722 for(y=0; y<4; y++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1723 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1724 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1725 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
1726 } |
3212 | 1727 |
1728 { | |
3776 | 1729 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy]; |
3212 | 1730 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]]; |
1731 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]]; | |
1732 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]]; | |
1733 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]]; | |
1168 | 1734 } |
1735 } | |
2967 | 1736 |
2396 | 1737 if(h->slice_type == B_TYPE && h->pps.cabac){ |
1738 if(IS_8X8(mb_type)){ | |
3212 | 1739 uint8_t *direct_table = &h->direct_table[b8_xy]; |
1740 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0; | |
1741 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0; | |
1742 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0; | |
2396 | 1743 } |
1744 } | |
1168 | 1745 } |
1746 | |
1747 /** | |
1748 * Decodes a network abstraction layer unit. | |
1749 * @param consumed is the number of bytes used as input | |
1750 * @param length is the length of the array | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
1751 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing? |
2967 | 1752 * @returns decoded bytes, might be src+1 if no escapes |
1168 | 1753 */ |
1754 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){ | |
1755 int i, si, di; | |
1756 uint8_t *dst; | |
1757 | |
2979 | 1758 // src[0]&0x80; //forbidden bit |
1168 | 1759 h->nal_ref_idc= src[0]>>5; |
1760 h->nal_unit_type= src[0]&0x1F; | |
1761 | |
1762 src++; length--; | |
2967 | 1763 #if 0 |
1168 | 1764 for(i=0; i<length; i++) |
1765 printf("%2X ", src[i]); | |
1766 #endif | |
1767 for(i=0; i+1<length; i+=2){ | |
1768 if(src[i]) continue; | |
1769 if(i>0 && src[i-1]==0) i--; | |
1770 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
1771 if(src[i+2]!=3){ | |
1772 /* startcode, so we must be past the end */ | |
1773 length=i; | |
1774 } | |
1775 break; | |
1776 } | |
1777 } | |
1778 | |
1779 if(i>=length-1){ //no escaped 0 | |
1780 *dst_length= length; | |
1781 *consumed= length+1; //+1 for the header | |
2967 | 1782 return src; |
1168 | 1783 } |
1784 | |
1785 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length); | |
1786 dst= h->rbsp_buffer; | |
1787 | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
1788 //printf("decoding esc\n"); |
1168 | 1789 si=di=0; |
2967 | 1790 while(si<length){ |
1168 | 1791 //remove escapes (very rare 1:2^22) |
1792 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
1793 if(src[si+2]==3){ //escape | |
1794 dst[di++]= 0; | |
1795 dst[di++]= 0; | |
1796 si+=3; | |
1957
54411768fa38
h264 nal decoding fix by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1956
diff
changeset
|
1797 continue; |
1168 | 1798 }else //next start code |
1799 break; | |
1800 } | |
1801 | |
1802 dst[di++]= src[si++]; | |
1803 } | |
1804 | |
1805 *dst_length= di; | |
1806 *consumed= si + 1;//+1 for the header | |
1807 //FIXME store exact number of bits in the getbitcontext (its needed for decoding) | |
1808 return dst; | |
1809 } | |
1810 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1811 #if 0 |
1168 | 1812 /** |
1813 * @param src the data which should be escaped | |
1814 * @param dst the target buffer, dst+1 == src is allowed as a special case | |
1815 * @param length the length of the src data | |
1816 * @param dst_length the length of the dst array | |
1817 * @returns length of escaped data in bytes or -1 if an error occured | |
1818 */ | |
1819 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){ | |
1820 int i, escape_count, si, di; | |
1821 uint8_t *temp; | |
2967 | 1822 |
1168 | 1823 assert(length>=0); |
1824 assert(dst_length>0); | |
2967 | 1825 |
1168 | 1826 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type; |
1827 | |
1828 if(length==0) return 1; | |
1829 | |
1830 escape_count= 0; | |
1831 for(i=0; i<length; i+=2){ | |
1832 if(src[i]) continue; | |
2967 | 1833 if(i>0 && src[i-1]==0) |
1168 | 1834 i--; |
1835 if(i+2<length && src[i+1]==0 && src[i+2]<=3){ | |
1836 escape_count++; | |
1837 i+=2; | |
1838 } | |
1839 } | |
2967 | 1840 |
1841 if(escape_count==0){ | |
1168 | 1842 if(dst+1 != src) |
1843 memcpy(dst+1, src, length); | |
1844 return length + 1; | |
1845 } | |
2967 | 1846 |
1168 | 1847 if(length + escape_count + 1> dst_length) |
1848 return -1; | |
1849 | |
1850 //this should be damn rare (hopefully) | |
1851 | |
1852 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count); | |
1853 temp= h->rbsp_buffer; | |
1854 //printf("encoding esc\n"); | |
2967 | 1855 |
1168 | 1856 si= 0; |
1857 di= 0; | |
1858 while(si < length){ | |
1859 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){ | |
1860 temp[di++]= 0; si++; | |
1861 temp[di++]= 0; si++; | |
2967 | 1862 temp[di++]= 3; |
1168 | 1863 temp[di++]= src[si++]; |
1864 } | |
1865 else | |
1866 temp[di++]= src[si++]; | |
1867 } | |
1868 memcpy(dst+1, temp, length+escape_count); | |
2967 | 1869 |
1168 | 1870 assert(di == length+escape_count); |
2967 | 1871 |
1168 | 1872 return di + 1; |
1873 } | |
1874 | |
1875 /** | |
1876 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4 | |
1877 */ | |
1878 static void encode_rbsp_trailing(PutBitContext *pb){ | |
1879 int length; | |
1880 put_bits(pb, 1, 1); | |
1786 | 1881 length= (-put_bits_count(pb))&7; |
1168 | 1882 if(length) put_bits(pb, length, 0); |
1883 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1884 #endif |
1168 | 1885 |
1886 /** | |
1887 * identifies the exact end of the bitstream | |
1888 * @return the length of the trailing, or 0 if damaged | |
1889 */ | |
1890 static int decode_rbsp_trailing(uint8_t *src){ | |
1891 int v= *src; | |
1892 int r; | |
1893 | |
1170 | 1894 tprintf("rbsp trailing %X\n", v); |
1168 | 1895 |
1896 for(r=1; r<9; r++){ | |
1897 if(v&1) return r; | |
1898 v>>=1; | |
1899 } | |
1900 return 0; | |
1901 } | |
1902 | |
1903 /** | |
1904 * idct tranforms the 16 dc values and dequantize them. | |
1905 * @param qp quantization parameter | |
1906 */ | |
2919 | 1907 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){ |
1168 | 1908 #define stride 16 |
1909 int i; | |
1910 int temp[16]; //FIXME check if this is a good idea | |
1911 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1912 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1913 | |
1914 //memset(block, 64, 2*256); | |
1915 //return; | |
1916 for(i=0; i<4; i++){ | |
1917 const int offset= y_offset[i]; | |
1918 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1919 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1920 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1921 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1922 | |
1923 temp[4*i+0]= z0+z3; | |
1924 temp[4*i+1]= z1+z2; | |
1925 temp[4*i+2]= z1-z2; | |
1926 temp[4*i+3]= z0-z3; | |
1927 } | |
1928 | |
1929 for(i=0; i<4; i++){ | |
1930 const int offset= x_offset[i]; | |
1931 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1932 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1933 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1934 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1935 | |
2919 | 1936 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual |
1937 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8)); | |
1938 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8)); | |
1939 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8)); | |
1168 | 1940 } |
1941 } | |
1942 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1943 #if 0 |
1168 | 1944 /** |
1945 * dct tranforms the 16 dc values. | |
1946 * @param qp quantization parameter ??? FIXME | |
1947 */ | |
1948 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){ | |
1949 // const int qmul= dequant_coeff[qp][0]; | |
1950 int i; | |
1951 int temp[16]; //FIXME check if this is a good idea | |
1952 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride}; | |
1953 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride}; | |
1954 | |
1955 for(i=0; i<4; i++){ | |
1956 const int offset= y_offset[i]; | |
1957 const int z0= block[offset+stride*0] + block[offset+stride*4]; | |
1958 const int z1= block[offset+stride*0] - block[offset+stride*4]; | |
1959 const int z2= block[offset+stride*1] - block[offset+stride*5]; | |
1960 const int z3= block[offset+stride*1] + block[offset+stride*5]; | |
1961 | |
1962 temp[4*i+0]= z0+z3; | |
1963 temp[4*i+1]= z1+z2; | |
1964 temp[4*i+2]= z1-z2; | |
1965 temp[4*i+3]= z0-z3; | |
1966 } | |
1967 | |
1968 for(i=0; i<4; i++){ | |
1969 const int offset= x_offset[i]; | |
1970 const int z0= temp[4*0+i] + temp[4*2+i]; | |
1971 const int z1= temp[4*0+i] - temp[4*2+i]; | |
1972 const int z2= temp[4*1+i] - temp[4*3+i]; | |
1973 const int z3= temp[4*1+i] + temp[4*3+i]; | |
1974 | |
1975 block[stride*0 +offset]= (z0 + z3)>>1; | |
1976 block[stride*2 +offset]= (z1 + z2)>>1; | |
1977 block[stride*8 +offset]= (z1 - z2)>>1; | |
1978 block[stride*10+offset]= (z0 - z3)>>1; | |
1979 } | |
1980 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1981 #endif |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
1982 |
1168 | 1983 #undef xStride |
1984 #undef stride | |
1985 | |
2919 | 1986 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){ |
1168 | 1987 const int stride= 16*2; |
1988 const int xStride= 16; | |
1989 int a,b,c,d,e; | |
1990 | |
1991 a= block[stride*0 + xStride*0]; | |
1992 b= block[stride*0 + xStride*1]; | |
1993 c= block[stride*1 + xStride*0]; | |
1994 d= block[stride*1 + xStride*1]; | |
1995 | |
1996 e= a-b; | |
1997 a= a+b; | |
1998 b= c-d; | |
1999 c= c+d; | |
2000 | |
2919 | 2001 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7; |
2002 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7; | |
2003 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7; | |
2004 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7; | |
1168 | 2005 } |
2006 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2007 #if 0 |
1168 | 2008 static void chroma_dc_dct_c(DCTELEM *block){ |
2009 const int stride= 16*2; | |
2010 const int xStride= 16; | |
2011 int a,b,c,d,e; | |
2012 | |
2013 a= block[stride*0 + xStride*0]; | |
2014 b= block[stride*0 + xStride*1]; | |
2015 c= block[stride*1 + xStride*0]; | |
2016 d= block[stride*1 + xStride*1]; | |
2017 | |
2018 e= a-b; | |
2019 a= a+b; | |
2020 b= c-d; | |
2021 c= c+d; | |
2022 | |
2023 block[stride*0 + xStride*0]= (a+c); | |
2024 block[stride*0 + xStride*1]= (e+b); | |
2025 block[stride*1 + xStride*0]= (a-c); | |
2026 block[stride*1 + xStride*1]= (e-b); | |
2027 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2028 #endif |
1168 | 2029 |
2030 /** | |
2031 * gets the chroma qp. | |
2032 */ | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
2033 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){ |
2967 | 2034 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
2035 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)]; |
1168 | 2036 } |
2037 | |
2038 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2039 #if 0 |
1168 | 2040 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){ |
2041 int i; | |
2042 //FIXME try int temp instead of block | |
2967 | 2043 |
1168 | 2044 for(i=0; i<4; i++){ |
2045 const int d0= src1[0 + i*stride] - src2[0 + i*stride]; | |
2046 const int d1= src1[1 + i*stride] - src2[1 + i*stride]; | |
2047 const int d2= src1[2 + i*stride] - src2[2 + i*stride]; | |
2048 const int d3= src1[3 + i*stride] - src2[3 + i*stride]; | |
2049 const int z0= d0 + d3; | |
2050 const int z3= d0 - d3; | |
2051 const int z1= d1 + d2; | |
2052 const int z2= d1 - d2; | |
2967 | 2053 |
1168 | 2054 block[0 + 4*i]= z0 + z1; |
2055 block[1 + 4*i]= 2*z3 + z2; | |
2056 block[2 + 4*i]= z0 - z1; | |
2057 block[3 + 4*i]= z3 - 2*z2; | |
2967 | 2058 } |
1168 | 2059 |
2060 for(i=0; i<4; i++){ | |
2061 const int z0= block[0*4 + i] + block[3*4 + i]; | |
2062 const int z3= block[0*4 + i] - block[3*4 + i]; | |
2063 const int z1= block[1*4 + i] + block[2*4 + i]; | |
2064 const int z2= block[1*4 + i] - block[2*4 + i]; | |
2967 | 2065 |
1168 | 2066 block[0*4 + i]= z0 + z1; |
2067 block[1*4 + i]= 2*z3 + z2; | |
2068 block[2*4 + i]= z0 - z1; | |
2069 block[3*4 + i]= z3 - 2*z2; | |
2070 } | |
2071 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
2072 #endif |
1168 | 2073 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
2074 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close |
1168 | 2075 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away) |
2076 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){ | |
2077 int i; | |
2078 const int * const quant_table= quant_coeff[qscale]; | |
2079 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6; | |
2080 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1; | |
2081 const unsigned int threshold2= (threshold1<<1); | |
2082 int last_non_zero; | |
2083 | |
2084 if(seperate_dc){ | |
2085 if(qscale<=18){ | |
2086 //avoid overflows | |
2087 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6; | |
2088 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1; | |
2089 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
2090 | |
2091 int level= block[0]*quant_coeff[qscale+18][0]; | |
2092 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
2093 if(level>0){ | |
2094 level= (dc_bias + level)>>(QUANT_SHIFT-2); | |
2095 block[0]= level; | |
2096 }else{ | |
2097 level= (dc_bias - level)>>(QUANT_SHIFT-2); | |
2098 block[0]= -level; | |
2099 } | |
2100 // last_non_zero = i; | |
2101 }else{ | |
2102 block[0]=0; | |
2103 } | |
2104 }else{ | |
2105 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6; | |
2106 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1; | |
2107 const unsigned int dc_threshold2= (dc_threshold1<<1); | |
2108 | |
2109 int level= block[0]*quant_table[0]; | |
2110 if(((unsigned)(level+dc_threshold1))>dc_threshold2){ | |
2111 if(level>0){ | |
2112 level= (dc_bias + level)>>(QUANT_SHIFT+1); | |
2113 block[0]= level; | |
2114 }else{ | |
2115 level= (dc_bias - level)>>(QUANT_SHIFT+1); | |
2116 block[0]= -level; | |
2117 } | |
2118 // last_non_zero = i; | |
2119 }else{ | |
2120 block[0]=0; | |
2121 } | |
2122 } | |
2123 last_non_zero= 0; | |
2124 i=1; | |
2125 }else{ | |
2126 last_non_zero= -1; | |
2127 i=0; | |
2128 } | |
2129 | |
2130 for(; i<16; i++){ | |
2131 const int j= scantable[i]; | |
2132 int level= block[j]*quant_table[j]; | |
2133 | |
2134 // if( bias+level >= (1<<(QMAT_SHIFT - 3)) | |
2135 // || bias-level >= (1<<(QMAT_SHIFT - 3))){ | |
2136 if(((unsigned)(level+threshold1))>threshold2){ | |
2137 if(level>0){ | |
2138 level= (bias + level)>>QUANT_SHIFT; | |
2139 block[j]= level; | |
2140 }else{ | |
2141 level= (bias - level)>>QUANT_SHIFT; | |
2142 block[j]= -level; | |
2143 } | |
2144 last_non_zero = i; | |
2145 }else{ | |
2146 block[j]=0; | |
2147 } | |
2148 } | |
2149 | |
2150 return last_non_zero; | |
2151 } | |
2152 | |
2153 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){ | |
2154 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
2155 ((uint32_t*)(src+0*stride))[0]= a; | |
2156 ((uint32_t*)(src+1*stride))[0]= a; | |
2157 ((uint32_t*)(src+2*stride))[0]= a; | |
2158 ((uint32_t*)(src+3*stride))[0]= a; | |
2159 } | |
2160 | |
2161 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){ | |
2162 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101; | |
2163 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101; | |
2164 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101; | |
2165 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101; | |
2166 } | |
2167 | |
2168 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
2169 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] | |
2170 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3; | |
2967 | 2171 |
2172 ((uint32_t*)(src+0*stride))[0]= | |
2173 ((uint32_t*)(src+1*stride))[0]= | |
2174 ((uint32_t*)(src+2*stride))[0]= | |
2175 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1168 | 2176 } |
2177 | |
2178 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
2179 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2; | |
2967 | 2180 |
2181 ((uint32_t*)(src+0*stride))[0]= | |
2182 ((uint32_t*)(src+1*stride))[0]= | |
2183 ((uint32_t*)(src+2*stride))[0]= | |
2184 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1168 | 2185 } |
2186 | |
2187 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
2188 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2; | |
2967 | 2189 |
2190 ((uint32_t*)(src+0*stride))[0]= | |
2191 ((uint32_t*)(src+1*stride))[0]= | |
2192 ((uint32_t*)(src+2*stride))[0]= | |
2193 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; | |
1168 | 2194 } |
2195 | |
2196 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 2197 ((uint32_t*)(src+0*stride))[0]= |
2198 ((uint32_t*)(src+1*stride))[0]= | |
2199 ((uint32_t*)(src+2*stride))[0]= | |
1168 | 2200 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U; |
2201 } | |
2202 | |
2203 | |
2204 #define LOAD_TOP_RIGHT_EDGE\ | |
2205 const int t4= topright[0];\ | |
2206 const int t5= topright[1];\ | |
2207 const int t6= topright[2];\ | |
2208 const int t7= topright[3];\ | |
2209 | |
2210 #define LOAD_LEFT_EDGE\ | |
2211 const int l0= src[-1+0*stride];\ | |
2212 const int l1= src[-1+1*stride];\ | |
2213 const int l2= src[-1+2*stride];\ | |
2214 const int l3= src[-1+3*stride];\ | |
2215 | |
2216 #define LOAD_TOP_EDGE\ | |
2217 const int t0= src[ 0-1*stride];\ | |
2218 const int t1= src[ 1-1*stride];\ | |
2219 const int t2= src[ 2-1*stride];\ | |
2220 const int t3= src[ 3-1*stride];\ | |
2221 | |
2222 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
2223 const int lt= src[-1-1*stride]; | |
2224 LOAD_TOP_EDGE | |
2225 LOAD_LEFT_EDGE | |
2226 | |
2967 | 2227 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; |
1168 | 2228 src[0+2*stride]= |
2967 | 2229 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; |
1168 | 2230 src[0+1*stride]= |
2231 src[1+2*stride]= | |
2967 | 2232 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; |
1168 | 2233 src[0+0*stride]= |
2234 src[1+1*stride]= | |
2235 src[2+2*stride]= | |
2967 | 2236 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; |
1168 | 2237 src[1+0*stride]= |
2238 src[2+1*stride]= | |
2239 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
2240 src[2+0*stride]= | |
2241 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
2242 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
1282 | 2243 } |
1168 | 2244 |
2245 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 2246 LOAD_TOP_EDGE |
2247 LOAD_TOP_RIGHT_EDGE | |
2248 // LOAD_LEFT_EDGE | |
1168 | 2249 |
2250 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2; | |
2251 src[1+0*stride]= | |
2252 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2; | |
2253 src[2+0*stride]= | |
2254 src[1+1*stride]= | |
2255 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2; | |
2256 src[3+0*stride]= | |
2257 src[2+1*stride]= | |
2258 src[1+2*stride]= | |
2259 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2; | |
2260 src[3+1*stride]= | |
2261 src[2+2*stride]= | |
2262 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2; | |
2263 src[3+2*stride]= | |
2264 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2; | |
2265 src[3+3*stride]=(t6 + 3*t7 + 2)>>2; | |
1282 | 2266 } |
1168 | 2267 |
2268 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){ | |
2269 const int lt= src[-1-1*stride]; | |
2967 | 2270 LOAD_TOP_EDGE |
2271 LOAD_LEFT_EDGE | |
1168 | 2272 const __attribute__((unused)) int unu= l3; |
2273 | |
2274 src[0+0*stride]= | |
2275 src[1+2*stride]=(lt + t0 + 1)>>1; | |
2276 src[1+0*stride]= | |
2277 src[2+2*stride]=(t0 + t1 + 1)>>1; | |
2278 src[2+0*stride]= | |
2279 src[3+2*stride]=(t1 + t2 + 1)>>1; | |
2280 src[3+0*stride]=(t2 + t3 + 1)>>1; | |
2281 src[0+1*stride]= | |
2282 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
2283 src[1+1*stride]= | |
2284 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
2285 src[2+1*stride]= | |
2286 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
2287 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
2288 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
2289 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
1282 | 2290 } |
1168 | 2291 |
2292 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 2293 LOAD_TOP_EDGE |
2294 LOAD_TOP_RIGHT_EDGE | |
1168 | 2295 const __attribute__((unused)) int unu= t7; |
2296 | |
2297 src[0+0*stride]=(t0 + t1 + 1)>>1; | |
2298 src[1+0*stride]= | |
2299 src[0+2*stride]=(t1 + t2 + 1)>>1; | |
2300 src[2+0*stride]= | |
2301 src[1+2*stride]=(t2 + t3 + 1)>>1; | |
2302 src[3+0*stride]= | |
2303 src[2+2*stride]=(t3 + t4+ 1)>>1; | |
2304 src[3+2*stride]=(t4 + t5+ 1)>>1; | |
2305 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
2306 src[1+1*stride]= | |
2307 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2; | |
2308 src[2+1*stride]= | |
2309 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2; | |
2310 src[3+1*stride]= | |
2311 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2; | |
2312 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2; | |
1282 | 2313 } |
1168 | 2314 |
2315 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){ | |
2967 | 2316 LOAD_LEFT_EDGE |
1168 | 2317 |
2318 src[0+0*stride]=(l0 + l1 + 1)>>1; | |
2319 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
2320 src[2+0*stride]= | |
2321 src[0+1*stride]=(l1 + l2 + 1)>>1; | |
2322 src[3+0*stride]= | |
2323 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
2324 src[2+1*stride]= | |
2325 src[0+2*stride]=(l2 + l3 + 1)>>1; | |
2326 src[3+1*stride]= | |
2327 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2; | |
2328 src[3+2*stride]= | |
2329 src[1+3*stride]= | |
2330 src[0+3*stride]= | |
2331 src[2+2*stride]= | |
2332 src[2+3*stride]= | |
2333 src[3+3*stride]=l3; | |
1282 | 2334 } |
2967 | 2335 |
1168 | 2336 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){ |
2337 const int lt= src[-1-1*stride]; | |
2967 | 2338 LOAD_TOP_EDGE |
2339 LOAD_LEFT_EDGE | |
1168 | 2340 const __attribute__((unused)) int unu= t3; |
2341 | |
2342 src[0+0*stride]= | |
2343 src[2+1*stride]=(lt + l0 + 1)>>1; | |
2344 src[1+0*stride]= | |
2345 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2; | |
2346 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2; | |
2347 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2; | |
2348 src[0+1*stride]= | |
2349 src[2+2*stride]=(l0 + l1 + 1)>>1; | |
2350 src[1+1*stride]= | |
2351 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2; | |
2352 src[0+2*stride]= | |
2353 src[2+3*stride]=(l1 + l2+ 1)>>1; | |
2354 src[1+2*stride]= | |
2355 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2; | |
2356 src[0+3*stride]=(l2 + l3 + 1)>>1; | |
2357 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2; | |
1282 | 2358 } |
1168 | 2359 |
2360 static void pred16x16_vertical_c(uint8_t *src, int stride){ | |
2361 int i; | |
2362 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
2363 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
2364 const uint32_t c= ((uint32_t*)(src-stride))[2]; | |
2365 const uint32_t d= ((uint32_t*)(src-stride))[3]; | |
2967 | 2366 |
1168 | 2367 for(i=0; i<16; i++){ |
2368 ((uint32_t*)(src+i*stride))[0]= a; | |
2369 ((uint32_t*)(src+i*stride))[1]= b; | |
2370 ((uint32_t*)(src+i*stride))[2]= c; | |
2371 ((uint32_t*)(src+i*stride))[3]= d; | |
2372 } | |
2373 } | |
2374 | |
2375 static void pred16x16_horizontal_c(uint8_t *src, int stride){ | |
2376 int i; | |
2377 | |
2378 for(i=0; i<16; i++){ | |
2379 ((uint32_t*)(src+i*stride))[0]= | |
2380 ((uint32_t*)(src+i*stride))[1]= | |
2381 ((uint32_t*)(src+i*stride))[2]= | |
2382 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101; | |
2383 } | |
2384 } | |
2385 | |
2386 static void pred16x16_dc_c(uint8_t *src, int stride){ | |
2387 int i, dc=0; | |
2388 | |
2389 for(i=0;i<16; i++){ | |
2390 dc+= src[-1+i*stride]; | |
2391 } | |
2967 | 2392 |
1168 | 2393 for(i=0;i<16; i++){ |
2394 dc+= src[i-stride]; | |
2395 } | |
2396 | |
2397 dc= 0x01010101*((dc + 16)>>5); | |
2398 | |
2399 for(i=0; i<16; i++){ | |
2400 ((uint32_t*)(src+i*stride))[0]= | |
2401 ((uint32_t*)(src+i*stride))[1]= | |
2402 ((uint32_t*)(src+i*stride))[2]= | |
2403 ((uint32_t*)(src+i*stride))[3]= dc; | |
2404 } | |
2405 } | |
2406 | |
2407 static void pred16x16_left_dc_c(uint8_t *src, int stride){ | |
2408 int i, dc=0; | |
2409 | |
2410 for(i=0;i<16; i++){ | |
2411 dc+= src[-1+i*stride]; | |
2412 } | |
2967 | 2413 |
1168 | 2414 dc= 0x01010101*((dc + 8)>>4); |
2415 | |
2416 for(i=0; i<16; i++){ | |
2417 ((uint32_t*)(src+i*stride))[0]= | |
2418 ((uint32_t*)(src+i*stride))[1]= | |
2419 ((uint32_t*)(src+i*stride))[2]= | |
2420 ((uint32_t*)(src+i*stride))[3]= dc; | |
2421 } | |
2422 } | |
2423 | |
2424 static void pred16x16_top_dc_c(uint8_t *src, int stride){ | |
2425 int i, dc=0; | |
2426 | |
2427 for(i=0;i<16; i++){ | |
2428 dc+= src[i-stride]; | |
2429 } | |
2430 dc= 0x01010101*((dc + 8)>>4); | |
2431 | |
2432 for(i=0; i<16; i++){ | |
2433 ((uint32_t*)(src+i*stride))[0]= | |
2434 ((uint32_t*)(src+i*stride))[1]= | |
2435 ((uint32_t*)(src+i*stride))[2]= | |
2436 ((uint32_t*)(src+i*stride))[3]= dc; | |
2437 } | |
2438 } | |
2439 | |
2440 static void pred16x16_128_dc_c(uint8_t *src, int stride){ | |
2441 int i; | |
2442 | |
2443 for(i=0; i<16; i++){ | |
2444 ((uint32_t*)(src+i*stride))[0]= | |
2445 ((uint32_t*)(src+i*stride))[1]= | |
2446 ((uint32_t*)(src+i*stride))[2]= | |
2447 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U; | |
2448 } | |
2449 } | |
2450 | |
1234 | 2451 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){ |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2452 int i, j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2453 int a; |
4176 | 2454 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2455 const uint8_t * const src0 = src+7-stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2456 const uint8_t *src1 = src+8*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2457 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2458 int H = src0[1] - src0[-1]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2459 int V = src1[0] - src2[ 0]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2460 for(k=2; k<=8; ++k) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2461 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2462 H += k*(src0[k] - src0[-k]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2463 V += k*(src1[0] - src2[ 0]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2464 } |
1234 | 2465 if(svq3){ |
2466 H = ( 5*(H/4) ) / 16; | |
2467 V = ( 5*(V/4) ) / 16; | |
1330
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
2468 |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
2469 /* required for 100% accuracy */ |
c05c381a9c47
- fix PLANE_PRED8x8 prediction (H/V are swapped, this is correct!)
tmmm
parents:
1322
diff
changeset
|
2470 i = H; H = V; V = i; |
1234 | 2471 }else{ |
2472 H = ( 5*H+32 ) >> 6; | |
2473 V = ( 5*V+32 ) >> 6; | |
2474 } | |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2475 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2476 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2477 for(j=16; j>0; --j) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2478 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2479 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2480 for(i=-16; i<0; i+=4) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2481 src[16+i] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2482 src[17+i] = cm[ (b+ H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2483 src[18+i] = cm[ (b+2*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2484 src[19+i] = cm[ (b+3*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2485 b += 4*H; |
1168 | 2486 } |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2487 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2488 } |
1168 | 2489 } |
2490 | |
1234 | 2491 static void pred16x16_plane_c(uint8_t *src, int stride){ |
2492 pred16x16_plane_compat_c(src, stride, 0); | |
2493 } | |
2494 | |
1168 | 2495 static void pred8x8_vertical_c(uint8_t *src, int stride){ |
2496 int i; | |
2497 const uint32_t a= ((uint32_t*)(src-stride))[0]; | |
2498 const uint32_t b= ((uint32_t*)(src-stride))[1]; | |
2967 | 2499 |
1168 | 2500 for(i=0; i<8; i++){ |
2501 ((uint32_t*)(src+i*stride))[0]= a; | |
2502 ((uint32_t*)(src+i*stride))[1]= b; | |
2503 } | |
2504 } | |
2505 | |
2506 static void pred8x8_horizontal_c(uint8_t *src, int stride){ | |
2507 int i; | |
2508 | |
2509 for(i=0; i<8; i++){ | |
2510 ((uint32_t*)(src+i*stride))[0]= | |
2511 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101; | |
2512 } | |
2513 } | |
2514 | |
2515 static void pred8x8_128_dc_c(uint8_t *src, int stride){ | |
2516 int i; | |
2517 | |
2755 | 2518 for(i=0; i<8; i++){ |
2967 | 2519 ((uint32_t*)(src+i*stride))[0]= |
1168 | 2520 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U; |
2521 } | |
2522 } | |
2523 | |
2524 static void pred8x8_left_dc_c(uint8_t *src, int stride){ | |
2525 int i; | |
2526 int dc0, dc2; | |
2527 | |
2528 dc0=dc2=0; | |
2529 for(i=0;i<4; i++){ | |
2530 dc0+= src[-1+i*stride]; | |
2531 dc2+= src[-1+(i+4)*stride]; | |
2532 } | |
2533 dc0= 0x01010101*((dc0 + 2)>>2); | |
2534 dc2= 0x01010101*((dc2 + 2)>>2); | |
2535 | |
2536 for(i=0; i<4; i++){ | |
2537 ((uint32_t*)(src+i*stride))[0]= | |
2538 ((uint32_t*)(src+i*stride))[1]= dc0; | |
2539 } | |
2540 for(i=4; i<8; i++){ | |
2541 ((uint32_t*)(src+i*stride))[0]= | |
2542 ((uint32_t*)(src+i*stride))[1]= dc2; | |
2543 } | |
2544 } | |
2545 | |
2546 static void pred8x8_top_dc_c(uint8_t *src, int stride){ | |
2547 int i; | |
2548 int dc0, dc1; | |
2549 | |
2550 dc0=dc1=0; | |
2551 for(i=0;i<4; i++){ | |
2552 dc0+= src[i-stride]; | |
2553 dc1+= src[4+i-stride]; | |
2554 } | |
2555 dc0= 0x01010101*((dc0 + 2)>>2); | |
2556 dc1= 0x01010101*((dc1 + 2)>>2); | |
2557 | |
2558 for(i=0; i<4; i++){ | |
2559 ((uint32_t*)(src+i*stride))[0]= dc0; | |
2560 ((uint32_t*)(src+i*stride))[1]= dc1; | |
2561 } | |
2562 for(i=4; i<8; i++){ | |
2563 ((uint32_t*)(src+i*stride))[0]= dc0; | |
2564 ((uint32_t*)(src+i*stride))[1]= dc1; | |
2565 } | |
2566 } | |
2567 | |
2568 | |
2569 static void pred8x8_dc_c(uint8_t *src, int stride){ | |
2570 int i; | |
2571 int dc0, dc1, dc2, dc3; | |
2572 | |
2573 dc0=dc1=dc2=0; | |
2574 for(i=0;i<4; i++){ | |
2575 dc0+= src[-1+i*stride] + src[i-stride]; | |
2576 dc1+= src[4+i-stride]; | |
2577 dc2+= src[-1+(i+4)*stride]; | |
2578 } | |
2579 dc3= 0x01010101*((dc1 + dc2 + 4)>>3); | |
2580 dc0= 0x01010101*((dc0 + 4)>>3); | |
2581 dc1= 0x01010101*((dc1 + 2)>>2); | |
2582 dc2= 0x01010101*((dc2 + 2)>>2); | |
2583 | |
2584 for(i=0; i<4; i++){ | |
2585 ((uint32_t*)(src+i*stride))[0]= dc0; | |
2586 ((uint32_t*)(src+i*stride))[1]= dc1; | |
2587 } | |
2588 for(i=4; i<8; i++){ | |
2589 ((uint32_t*)(src+i*stride))[0]= dc2; | |
2590 ((uint32_t*)(src+i*stride))[1]= dc3; | |
2591 } | |
2592 } | |
2593 | |
2594 static void pred8x8_plane_c(uint8_t *src, int stride){ | |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2595 int j, k; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2596 int a; |
4176 | 2597 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; |
1184
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2598 const uint8_t * const src0 = src+3-stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2599 const uint8_t *src1 = src+4*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2600 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2601 int H = src0[1] - src0[-1]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2602 int V = src1[0] - src2[ 0]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2603 for(k=2; k<=4; ++k) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2604 src1 += stride; src2 -= stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2605 H += k*(src0[k] - src0[-k]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2606 V += k*(src1[0] - src2[ 0]); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2607 } |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2608 H = ( 17*H+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2609 V = ( 17*V+16 ) >> 5; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2610 |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2611 a = 16*(src1[0] + src2[8]+1) - 3*(V+H); |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2612 for(j=8; j>0; --j) { |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2613 int b = a; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2614 a += V; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2615 src[0] = cm[ (b ) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2616 src[1] = cm[ (b+ H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2617 src[2] = cm[ (b+2*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2618 src[3] = cm[ (b+3*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2619 src[4] = cm[ (b+4*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2620 src[5] = cm[ (b+5*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2621 src[6] = cm[ (b+6*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2622 src[7] = cm[ (b+7*H) >> 5 ]; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2623 src += stride; |
05a2ac8978ad
faster 8x8 & 16x16 plane prediction by skal (massimin at planet-d dot net)
michaelni
parents:
1177
diff
changeset
|
2624 } |
1168 | 2625 } |
2626 | |
2755 | 2627 #define SRC(x,y) src[(x)+(y)*stride] |
2628 #define PL(y) \ | |
2629 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2; | |
2630 #define PREDICT_8x8_LOAD_LEFT \ | |
2631 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \ | |
2632 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \ | |
2633 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \ | |
2834 | 2634 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2 |
2755 | 2635 |
2636 #define PT(x) \ | |
2637 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2; | |
2638 #define PREDICT_8x8_LOAD_TOP \ | |
2639 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \ | |
2640 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \ | |
2641 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \ | |
2834 | 2642 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \ |
2757 | 2643 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2 |
2755 | 2644 |
2645 #define PTR(x) \ | |
2646 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2; | |
2647 #define PREDICT_8x8_LOAD_TOPRIGHT \ | |
2648 int t8, t9, t10, t11, t12, t13, t14, t15; \ | |
2649 if(has_topright) { \ | |
2650 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \ | |
2651 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \ | |
2652 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1); | |
2653 | |
2654 #define PREDICT_8x8_LOAD_TOPLEFT \ | |
2757 | 2655 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2 |
2755 | 2656 |
2657 #define PREDICT_8x8_DC(v) \ | |
2658 int y; \ | |
2659 for( y = 0; y < 8; y++ ) { \ | |
2660 ((uint32_t*)src)[0] = \ | |
2661 ((uint32_t*)src)[1] = v; \ | |
2662 src += stride; \ | |
2663 } | |
2664 | |
2665 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2666 { | |
2667 PREDICT_8x8_DC(0x80808080); | |
2668 } | |
2669 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2670 { | |
2671 PREDICT_8x8_LOAD_LEFT; | |
2672 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101; | |
2673 PREDICT_8x8_DC(dc); | |
2674 } | |
2675 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2676 { | |
2677 PREDICT_8x8_LOAD_TOP; | |
2678 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101; | |
2679 PREDICT_8x8_DC(dc); | |
2680 } | |
2681 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2682 { | |
2683 PREDICT_8x8_LOAD_LEFT; | |
2684 PREDICT_8x8_LOAD_TOP; | |
2685 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7 | |
2686 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101; | |
2687 PREDICT_8x8_DC(dc); | |
2688 } | |
2689 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2690 { | |
2691 PREDICT_8x8_LOAD_LEFT; | |
2692 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\ | |
2693 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y | |
2694 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7); | |
2695 #undef ROW | |
2696 } | |
2697 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2698 { | |
2699 int y; | |
2700 PREDICT_8x8_LOAD_TOP; | |
2701 src[0] = t0; | |
2702 src[1] = t1; | |
2703 src[2] = t2; | |
2704 src[3] = t3; | |
2705 src[4] = t4; | |
2706 src[5] = t5; | |
2707 src[6] = t6; | |
2708 src[7] = t7; | |
2709 for( y = 1; y < 8; y++ ) | |
2710 *(uint64_t*)(src+y*stride) = *(uint64_t*)src; | |
2711 } | |
2712 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2713 { | |
2714 PREDICT_8x8_LOAD_TOP; | |
2715 PREDICT_8x8_LOAD_TOPRIGHT; | |
2716 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2717 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2718 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2719 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2720 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2721 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2722 SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2; | |
2723 SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2; | |
2724 SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2; | |
2725 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2; | |
2726 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2; | |
2727 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2; | |
2728 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2; | |
2729 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2; | |
2730 SRC(7,7)= (t14 + 3*t15 + 2) >> 2; | |
2731 } | |
2732 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2733 { | |
2734 PREDICT_8x8_LOAD_TOP; | |
2735 PREDICT_8x8_LOAD_LEFT; | |
2736 PREDICT_8x8_LOAD_TOPLEFT; | |
2737 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2; | |
2738 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2; | |
2739 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2; | |
2740 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2; | |
2741 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2; | |
2742 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2; | |
2743 SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2; | |
2744 SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2; | |
2745 SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2; | |
2746 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2747 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2748 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2749 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2750 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2751 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2967 | 2752 |
2755 | 2753 } |
2754 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2755 { | |
2756 PREDICT_8x8_LOAD_TOP; | |
2757 PREDICT_8x8_LOAD_LEFT; | |
2758 PREDICT_8x8_LOAD_TOPLEFT; | |
2759 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2; | |
2760 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2; | |
2761 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2; | |
2762 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2; | |
2763 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2; | |
2764 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2; | |
2765 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2; | |
2766 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1; | |
2767 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2; | |
2768 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1; | |
2769 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2770 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1; | |
2771 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2772 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1; | |
2773 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2774 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1; | |
2775 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2776 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1; | |
2777 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2778 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1; | |
2779 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2780 SRC(7,0)= (t6 + t7 + 1) >> 1; | |
2781 } | |
2782 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2783 { | |
2784 PREDICT_8x8_LOAD_TOP; | |
2785 PREDICT_8x8_LOAD_LEFT; | |
2786 PREDICT_8x8_LOAD_TOPLEFT; | |
2787 SRC(0,7)= (l6 + l7 + 1) >> 1; | |
2788 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2; | |
2789 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1; | |
2790 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2; | |
2791 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1; | |
2792 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2; | |
2793 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1; | |
2794 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2; | |
2795 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1; | |
2796 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2; | |
2797 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1; | |
2798 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2; | |
2799 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1; | |
2800 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2; | |
2801 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1; | |
2802 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2; | |
2803 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2; | |
2804 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2; | |
2805 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2; | |
2806 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2; | |
2807 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2; | |
2808 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2; | |
2809 } | |
2810 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2811 { | |
2812 PREDICT_8x8_LOAD_TOP; | |
2813 PREDICT_8x8_LOAD_TOPRIGHT; | |
2814 SRC(0,0)= (t0 + t1 + 1) >> 1; | |
2815 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2; | |
2816 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1; | |
2817 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2; | |
2818 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1; | |
2819 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2; | |
2820 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1; | |
2821 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2; | |
2822 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1; | |
2823 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2; | |
2824 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1; | |
2825 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2; | |
2826 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1; | |
2827 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2; | |
2828 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1; | |
2829 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2; | |
2830 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1; | |
2831 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2; | |
2832 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1; | |
2833 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2; | |
2834 SRC(7,6)= (t10 + t11 + 1) >> 1; | |
2835 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2; | |
2836 } | |
2837 static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride) | |
2838 { | |
2839 PREDICT_8x8_LOAD_LEFT; | |
2840 SRC(0,0)= (l0 + l1 + 1) >> 1; | |
2841 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2; | |
2842 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1; | |
2843 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2; | |
2844 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1; | |
2845 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2; | |
2846 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1; | |
2847 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2; | |
2848 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1; | |
2849 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2; | |
2850 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1; | |
2851 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2; | |
2852 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1; | |
2853 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2; | |
2854 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)= | |
2855 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)= | |
2856 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)= | |
2857 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7; | |
2858 } | |
2859 #undef PREDICT_8x8_LOAD_LEFT | |
2860 #undef PREDICT_8x8_LOAD_TOP | |
2861 #undef PREDICT_8x8_LOAD_TOPLEFT | |
2862 #undef PREDICT_8x8_LOAD_TOPRIGHT | |
2863 #undef PREDICT_8x8_DC | |
2864 #undef PTR | |
2865 #undef PT | |
2866 #undef PL | |
2867 #undef SRC | |
2868 | |
1168 | 2869 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list, |
2870 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
2871 int src_x_offset, int src_y_offset, | |
2872 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){ | |
2873 MpegEncContext * const s = &h->s; | |
2874 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8; | |
3316 | 2875 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8; |
1168 | 2876 const int luma_xy= (mx&3) + ((my&3)<<2); |
3316 | 2877 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize; |
2878 uint8_t * src_cb, * src_cr; | |
2879 int extra_width= h->emu_edge_width; | |
2880 int extra_height= h->emu_edge_height; | |
1168 | 2881 int emu=0; |
2882 const int full_mx= mx>>2; | |
2883 const int full_my= my>>2; | |
2861
3b920e274b26
fix emulated edge motion compensation when resolution%16!=0
lorenm
parents:
2846
diff
changeset
|
2884 const int pic_width = 16*s->mb_width; |
3316 | 2885 const int pic_height = 16*s->mb_height >> MB_MBAFF; |
2967 | 2886 |
3125 | 2887 if(!pic->data[0]) |
2888 return; | |
2967 | 2889 |
1168 | 2890 if(mx&7) extra_width -= 3; |
2891 if(my&7) extra_height -= 3; | |
2967 | 2892 |
2893 if( full_mx < 0-extra_width | |
2894 || full_my < 0-extra_height | |
2895 || full_mx + 16/*FIXME*/ > pic_width + extra_width | |
2861
3b920e274b26
fix emulated edge motion compensation when resolution%16!=0
lorenm
parents:
2846
diff
changeset
|
2896 || full_my + 16/*FIXME*/ > pic_height + extra_height){ |
3316 | 2897 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height); |
2898 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize; | |
1168 | 2899 emu=1; |
2900 } | |
2967 | 2901 |
3316 | 2902 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps? |
1168 | 2903 if(!square){ |
3316 | 2904 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize); |
1168 | 2905 } |
2967 | 2906 |
1168 | 2907 if(s->flags&CODEC_FLAG_GRAY) return; |
2967 | 2908 |
3316 | 2909 if(MB_MBAFF){ |
2910 // chroma offset when predicting from a field of opposite parity | |
2911 my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1)); | |
2912 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1); | |
2913 } | |
2914 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize; | |
2915 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize; | |
1168 | 2916 |
2917 if(emu){ | |
3316 | 2918 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1); |
2919 src_cb= s->edge_emu_buffer; | |
2920 } | |
2921 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7); | |
2922 | |
2923 if(emu){ | |
2924 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1); | |
1168 | 2925 src_cr= s->edge_emu_buffer; |
2926 } | |
3316 | 2927 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7); |
1168 | 2928 } |
2929 | |
2415 | 2930 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta, |
1168 | 2931 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
2932 int x_offset, int y_offset, | |
2933 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
2934 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg, | |
2935 int list0, int list1){ | |
2936 MpegEncContext * const s = &h->s; | |
2937 qpel_mc_func *qpix_op= qpix_put; | |
2938 h264_chroma_mc_func chroma_op= chroma_put; | |
2967 | 2939 |
3316 | 2940 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize; |
2941 dest_cb += x_offset + y_offset*h->mb_uvlinesize; | |
2942 dest_cr += x_offset + y_offset*h->mb_uvlinesize; | |
1168 | 2943 x_offset += 8*s->mb_x; |
3316 | 2944 y_offset += 8*(s->mb_y >> MB_MBAFF); |
2967 | 2945 |
1168 | 2946 if(list0){ |
1169 | 2947 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ]; |
1168 | 2948 mc_dir_part(h, ref, n, square, chroma_height, delta, 0, |
2949 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2950 qpix_op, chroma_op); | |
2951 | |
2952 qpix_op= qpix_avg; | |
2953 chroma_op= chroma_avg; | |
2954 } | |
2955 | |
2956 if(list1){ | |
1169 | 2957 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ]; |
1168 | 2958 mc_dir_part(h, ref, n, square, chroma_height, delta, 1, |
2959 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
2960 qpix_op, chroma_op); | |
2961 } | |
2962 } | |
2963 | |
2415 | 2964 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta, |
2965 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
2966 int x_offset, int y_offset, | |
2967 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
2968 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op, | |
2969 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg, | |
2970 int list0, int list1){ | |
2971 MpegEncContext * const s = &h->s; | |
2972 | |
3316 | 2973 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize; |
2974 dest_cb += x_offset + y_offset*h->mb_uvlinesize; | |
2975 dest_cr += x_offset + y_offset*h->mb_uvlinesize; | |
2415 | 2976 x_offset += 8*s->mb_x; |
3316 | 2977 y_offset += 8*(s->mb_y >> MB_MBAFF); |
2967 | 2978 |
2415 | 2979 if(list0 && list1){ |
2980 /* don't optimize for luma-only case, since B-frames usually | |
2981 * use implicit weights => chroma too. */ | |
2982 uint8_t *tmp_cb = s->obmc_scratchpad; | |
3316 | 2983 uint8_t *tmp_cr = s->obmc_scratchpad + 8; |
2984 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize; | |
2415 | 2985 int refn0 = h->ref_cache[0][ scan8[n] ]; |
2986 int refn1 = h->ref_cache[1][ scan8[n] ]; | |
2987 | |
2988 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0, | |
2989 dest_y, dest_cb, dest_cr, | |
2990 x_offset, y_offset, qpix_put, chroma_put); | |
2991 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1, | |
2992 tmp_y, tmp_cb, tmp_cr, | |
2993 x_offset, y_offset, qpix_put, chroma_put); | |
2994 | |
2995 if(h->use_weight == 2){ | |
2996 int weight0 = h->implicit_weight[refn0][refn1]; | |
2997 int weight1 = 64 - weight0; | |
3316 | 2998 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0); |
2999 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0); | |
3000 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0); | |
2415 | 3001 }else{ |
3316 | 3002 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom, |
2967 | 3003 h->luma_weight[0][refn0], h->luma_weight[1][refn1], |
3029 | 3004 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]); |
3316 | 3005 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2967 | 3006 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], |
3029 | 3007 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]); |
3316 | 3008 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2967 | 3009 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], |
3029 | 3010 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]); |
2415 | 3011 } |
3012 }else{ | |
3013 int list = list1 ? 1 : 0; | |
3014 int refn = h->ref_cache[list][ scan8[n] ]; | |
3015 Picture *ref= &h->ref_list[list][refn]; | |
3016 mc_dir_part(h, ref, n, square, chroma_height, delta, list, | |
3017 dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
3018 qpix_put, chroma_put); | |
3019 | |
3316 | 3020 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom, |
2415 | 3021 h->luma_weight[list][refn], h->luma_offset[list][refn]); |
3022 if(h->use_weight_chroma){ | |
3316 | 3023 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2415 | 3024 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]); |
3316 | 3025 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom, |
2415 | 3026 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]); |
3027 } | |
3028 } | |
3029 } | |
3030 | |
3031 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta, | |
3032 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, | |
3033 int x_offset, int y_offset, | |
3034 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put, | |
3035 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg, | |
2967 | 3036 h264_weight_func *weight_op, h264_biweight_func *weight_avg, |
2415 | 3037 int list0, int list1){ |
3038 if((h->use_weight==2 && list0 && list1 | |
3039 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32)) | |
3040 || h->use_weight==1) | |
3041 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr, | |
3042 x_offset, y_offset, qpix_put, chroma_put, | |
3043 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1); | |
3044 else | |
3045 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr, | |
3046 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1); | |
3047 } | |
3048 | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3049 static inline void prefetch_motion(H264Context *h, int list){ |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3050 /* fetch pixels for estimated mv 4 macroblocks ahead |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3051 * optimized for 64byte cache lines */ |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3052 MpegEncContext * const s = &h->s; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3053 const int refn = h->ref_cache[list][scan8[0]]; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3054 if(refn >= 0){ |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3055 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3056 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3057 uint8_t **src= h->ref_list[list][refn].data; |
3316 | 3058 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64; |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3059 s->dsp.prefetch(src[0]+off, s->linesize, 4); |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3060 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64; |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3061 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2); |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3062 } |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3063 } |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3064 |
1168 | 3065 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, |
3066 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put), | |
2415 | 3067 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg), |
3068 h264_weight_func *weight_op, h264_biweight_func *weight_avg){ | |
1168 | 3069 MpegEncContext * const s = &h->s; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
3070 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 3071 const int mb_type= s->current_picture.mb_type[mb_xy]; |
2967 | 3072 |
1168 | 3073 assert(IS_INTER(mb_type)); |
2967 | 3074 |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3075 prefetch_motion(h, 0); |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3076 |
1168 | 3077 if(IS_16X16(mb_type)){ |
3078 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0, | |
3079 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0], | |
2415 | 3080 &weight_op[0], &weight_avg[0], |
1168 | 3081 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); |
3082 }else if(IS_16X8(mb_type)){ | |
3083 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0, | |
3084 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
2415 | 3085 &weight_op[1], &weight_avg[1], |
1168 | 3086 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); |
3087 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4, | |
3088 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0], | |
2415 | 3089 &weight_op[1], &weight_avg[1], |
1168 | 3090 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); |
3091 }else if(IS_8X16(mb_type)){ | |
3316 | 3092 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0, |
1168 | 3093 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], |
2415 | 3094 &weight_op[2], &weight_avg[2], |
1168 | 3095 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1)); |
3316 | 3096 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0, |
1168 | 3097 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], |
2415 | 3098 &weight_op[2], &weight_avg[2], |
1168 | 3099 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1)); |
3100 }else{ | |
3101 int i; | |
2967 | 3102 |
1168 | 3103 assert(IS_8X8(mb_type)); |
3104 | |
3105 for(i=0; i<4; i++){ | |
3106 const int sub_mb_type= h->sub_mb_type[i]; | |
3107 const int n= 4*i; | |
3108 int x_offset= (i&1)<<2; | |
3109 int y_offset= (i&2)<<1; | |
3110 | |
3111 if(IS_SUB_8X8(sub_mb_type)){ | |
3112 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
3113 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1], | |
2415 | 3114 &weight_op[3], &weight_avg[3], |
1168 | 3115 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
3116 }else if(IS_SUB_8X4(sub_mb_type)){ | |
3117 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset, | |
3118 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2415 | 3119 &weight_op[4], &weight_avg[4], |
1168 | 3120 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
3121 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2, | |
3122 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1], | |
2415 | 3123 &weight_op[4], &weight_avg[4], |
1168 | 3124 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
3125 }else if(IS_SUB_4X8(sub_mb_type)){ | |
3316 | 3126 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset, |
1168 | 3127 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], |
2415 | 3128 &weight_op[5], &weight_avg[5], |
1168 | 3129 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
3316 | 3130 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset, |
1168 | 3131 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], |
2415 | 3132 &weight_op[5], &weight_avg[5], |
1168 | 3133 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
3134 }else{ | |
3135 int j; | |
3136 assert(IS_SUB_4X4(sub_mb_type)); | |
3137 for(j=0; j<4; j++){ | |
3138 int sub_x_offset= x_offset + 2*(j&1); | |
3139 int sub_y_offset= y_offset + (j&2); | |
3140 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset, | |
3141 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2], | |
2415 | 3142 &weight_op[6], &weight_avg[6], |
1168 | 3143 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1)); |
3144 } | |
3145 } | |
3146 } | |
3147 } | |
3215
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3148 |
06f98047ff26
prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents:
3212
diff
changeset
|
3149 prefetch_motion(h, 1); |
1168 | 3150 } |
3151 | |
4164
171e768324cd
Remove the H264Context parameter from decode_init_vlc() as it is not being used.
takis
parents:
4118
diff
changeset
|
3152 static void decode_init_vlc(){ |
1168 | 3153 static int done = 0; |
3154 | |
3155 if (!done) { | |
3156 int i; | |
3157 done = 1; | |
3158 | |
2967 | 3159 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, |
1168 | 3160 &chroma_dc_coeff_token_len [0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
3161 &chroma_dc_coeff_token_bits[0], 1, 1, 1); |
1168 | 3162 |
3163 for(i=0; i<4; i++){ | |
2967 | 3164 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, |
1168 | 3165 &coeff_token_len [i][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
3166 &coeff_token_bits[i][0], 1, 1, 1); |
1168 | 3167 } |
3168 | |
3169 for(i=0; i<3; i++){ | |
3170 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4, | |
3171 &chroma_dc_total_zeros_len [i][0], 1, 1, | |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
3172 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1); |
1168 | 3173 } |
3174 for(i=0; i<15; i++){ | |
2967 | 3175 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, |
1168 | 3176 &total_zeros_len [i][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
3177 &total_zeros_bits[i][0], 1, 1, 1); |
1168 | 3178 } |
3179 | |
3180 for(i=0; i<6; i++){ | |
2967 | 3181 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, |
1168 | 3182 &run_len [i][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
3183 &run_bits[i][0], 1, 1, 1); |
1168 | 3184 } |
2967 | 3185 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, |
1168 | 3186 &run_len [6][0], 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2365
diff
changeset
|
3187 &run_bits[6][0], 1, 1, 1); |
1168 | 3188 } |
3189 } | |
3190 | |
3191 /** | |
3192 * Sets the intra prediction function pointers. | |
3193 */ | |
3194 static void init_pred_ptrs(H264Context *h){ | |
3195 // MpegEncContext * const s = &h->s; | |
3196 | |
3197 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c; | |
3198 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c; | |
3199 h->pred4x4[DC_PRED ]= pred4x4_dc_c; | |
3200 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c; | |
3201 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c; | |
3202 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c; | |
3203 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c; | |
3204 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c; | |
3205 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c; | |
3206 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c; | |
3207 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c; | |
3208 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c; | |
3209 | |
2755 | 3210 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c; |
3211 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c; | |
3212 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c; | |
3213 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c; | |
3214 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c; | |
3215 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c; | |
3216 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c; | |
3217 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c; | |
3218 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c; | |
3219 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c; | |
3220 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c; | |
3221 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c; | |
3222 | |
1168 | 3223 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c; |
3224 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c; | |
3225 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c; | |
3226 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c; | |
3227 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c; | |
3228 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c; | |
3229 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c; | |
3230 | |
3231 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c; | |
3232 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c; | |
3233 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c; | |
3234 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c; | |
3235 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c; | |
3236 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c; | |
3237 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c; | |
3238 } | |
3239 | |
3240 static void free_tables(H264Context *h){ | |
3241 av_freep(&h->intra4x4_pred_mode); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3242 av_freep(&h->chroma_pred_mode_table); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3243 av_freep(&h->cbp_table); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
3244 av_freep(&h->mvd_table[0]); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
3245 av_freep(&h->mvd_table[1]); |
2396 | 3246 av_freep(&h->direct_table); |
1168 | 3247 av_freep(&h->non_zero_count); |
3248 av_freep(&h->slice_table_base); | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3249 av_freep(&h->top_borders[1]); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3250 av_freep(&h->top_borders[0]); |
1168 | 3251 h->slice_table= NULL; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3252 |
1168 | 3253 av_freep(&h->mb2b_xy); |
3254 av_freep(&h->mb2b8_xy); | |
2415 | 3255 |
3256 av_freep(&h->s.obmc_scratchpad); | |
1168 | 3257 } |
3258 | |
2919 | 3259 static void init_dequant8_coeff_table(H264Context *h){ |
3260 int i,q,x; | |
3174 | 3261 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly |
2919 | 3262 h->dequant8_coeff[0] = h->dequant8_buffer[0]; |
3263 h->dequant8_coeff[1] = h->dequant8_buffer[1]; | |
3264 | |
3265 for(i=0; i<2; i++ ){ | |
3266 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){ | |
3267 h->dequant8_coeff[1] = h->dequant8_buffer[0]; | |
3268 break; | |
3269 } | |
3270 | |
3271 for(q=0; q<52; q++){ | |
3272 int shift = div6[q]; | |
3273 int idx = rem6[q]; | |
3274 for(x=0; x<64; x++) | |
3174 | 3275 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] = |
3276 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] * | |
3277 h->pps.scaling_matrix8[i][x]) << shift; | |
2919 | 3278 } |
3279 } | |
3280 } | |
3281 | |
3282 static void init_dequant4_coeff_table(H264Context *h){ | |
3283 int i,j,q,x; | |
3005 | 3284 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly |
2919 | 3285 for(i=0; i<6; i++ ){ |
3286 h->dequant4_coeff[i] = h->dequant4_buffer[i]; | |
3287 for(j=0; j<i; j++){ | |
3288 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){ | |
3289 h->dequant4_coeff[i] = h->dequant4_buffer[j]; | |
3290 break; | |
3291 } | |
3292 } | |
3293 if(j<i) | |
3294 continue; | |
3295 | |
3296 for(q=0; q<52; q++){ | |
3297 int shift = div6[q] + 2; | |
3298 int idx = rem6[q]; | |
3299 for(x=0; x<16; x++) | |
3005 | 3300 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] = |
3301 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] * | |
2919 | 3302 h->pps.scaling_matrix4[i][x]) << shift; |
3303 } | |
3304 } | |
3305 } | |
3306 | |
3307 static void init_dequant_tables(H264Context *h){ | |
3308 int i,x; | |
3309 init_dequant4_coeff_table(h); | |
3310 if(h->pps.transform_8x8_mode) | |
3311 init_dequant8_coeff_table(h); | |
3312 if(h->sps.transform_bypass){ | |
3313 for(i=0; i<6; i++) | |
3314 for(x=0; x<16; x++) | |
3315 h->dequant4_coeff[i][0][x] = 1<<6; | |
3316 if(h->pps.transform_8x8_mode) | |
3317 for(i=0; i<2; i++) | |
3318 for(x=0; x<64; x++) | |
3319 h->dequant8_coeff[i][0][x] = 1<<6; | |
3320 } | |
3321 } | |
3322 | |
3323 | |
1168 | 3324 /** |
3325 * allocates tables. | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3326 * needs width/height |
1168 | 3327 */ |
3328 static int alloc_tables(H264Context *h){ | |
3329 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
3330 const int big_mb_num= s->mb_stride * (s->mb_height+1); |
2919 | 3331 int x,y; |
1168 | 3332 |
3333 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t)) | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3334 |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3335 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t)) |
3316 | 3336 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(uint8_t)) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3337 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t)) |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3338 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t)) |
2336 | 3339 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t)) |
1168 | 3340 |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3341 if( h->pps.cabac ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3342 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t)) |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
3343 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t)); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
3344 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t)); |
2396 | 3345 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t)); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3346 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
3347 |
3316 | 3348 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(uint8_t)); |
3349 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1; | |
1168 | 3350 |
2641
c337f851d0f9
fix storage of motion vectors for frames with more than 4096 macroblocks.
lorenm
parents:
2640
diff
changeset
|
3351 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t)); |
c337f851d0f9
fix storage of motion vectors for frames with more than 4096 macroblocks.
lorenm
parents:
2640
diff
changeset
|
3352 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t)); |
1168 | 3353 for(y=0; y<s->mb_height; y++){ |
3354 for(x=0; x<s->mb_width; x++){ | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
3355 const int mb_xy= x + y*s->mb_stride; |
1168 | 3356 const int b_xy = 4*x + 4*y*h->b_stride; |
3357 const int b8_xy= 2*x + 2*y*h->b8_stride; | |
2967 | 3358 |
1168 | 3359 h->mb2b_xy [mb_xy]= b_xy; |
3360 h->mb2b8_xy[mb_xy]= b8_xy; | |
3361 } | |
3362 } | |
2415 | 3363 |
2417 | 3364 s->obmc_scratchpad = NULL; |
3365 | |
2920 | 3366 if(!h->dequant4_coeff[0]) |
3367 init_dequant_tables(h); | |
3368 | |
1168 | 3369 return 0; |
3370 fail: | |
3371 free_tables(h); | |
3372 return -1; | |
3373 } | |
3374 | |
3375 static void common_init(H264Context *h){ | |
3376 MpegEncContext * const s = &h->s; | |
3377 | |
3378 s->width = s->avctx->width; | |
3379 s->height = s->avctx->height; | |
3380 s->codec_id= s->avctx->codec->id; | |
2967 | 3381 |
1168 | 3382 init_pred_ptrs(h); |
3383 | |
2919 | 3384 h->dequant_coeff_pps= -1; |
1698 | 3385 s->unrestricted_mv=1; |
1168 | 3386 s->decode=1; //FIXME |
2920 | 3387 |
3388 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t)); | |
3389 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t)); | |
1168 | 3390 } |
3391 | |
3392 static int decode_init(AVCodecContext *avctx){ | |
3393 H264Context *h= avctx->priv_data; | |
3394 MpegEncContext * const s = &h->s; | |
3395 | |
1892 | 3396 MPV_decode_defaults(s); |
2967 | 3397 |
1168 | 3398 s->avctx = avctx; |
3399 common_init(h); | |
3400 | |
3401 s->out_format = FMT_H264; | |
3402 s->workaround_bugs= avctx->workaround_bugs; | |
3403 | |
3404 // set defaults | |
3405 // s->decode_mb= ff_h263_decode_mb; | |
3406 s->low_delay= 1; | |
3407 avctx->pix_fmt= PIX_FMT_YUV420P; | |
3408 | |
4164
171e768324cd
Remove the H264Context parameter from decode_init_vlc() as it is not being used.
takis
parents:
4118
diff
changeset
|
3409 decode_init_vlc(); |
2967 | 3410 |
2547
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
3411 if(avctx->extradata_size > 0 && avctx->extradata && |
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
3412 *(char *)avctx->extradata == 1){ |
2227 | 3413 h->is_avc = 1; |
3414 h->got_avcC = 0; | |
2547
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
3415 } else { |
c5781912ad8a
improved detection of "AVC1" style H.264 patch by (Mns Rullgrd <mru inprovide com>)
michael
parents:
2538
diff
changeset
|
3416 h->is_avc = 0; |
2227 | 3417 } |
3418 | |
1168 | 3419 return 0; |
3420 } | |
3421 | |
2935 | 3422 static int frame_start(H264Context *h){ |
1168 | 3423 MpegEncContext * const s = &h->s; |
3424 int i; | |
3425 | |
2935 | 3426 if(MPV_frame_start(s, s->avctx) < 0) |
3427 return -1; | |
1168 | 3428 ff_er_frame_start(s); |
3429 | |
3430 assert(s->linesize && s->uvlinesize); | |
3431 | |
3432 for(i=0; i<16; i++){ | |
3433 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3); | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3434 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3); |
1168 | 3435 } |
3436 for(i=0; i<4; i++){ | |
3437 h->block_offset[16+i]= | |
3438 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3); | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3439 h->block_offset[24+16+i]= |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3440 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3); |
1168 | 3441 } |
3442 | |
2416
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
3443 /* can't be in alloc_tables because linesize isn't known there. |
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
3444 * FIXME: redo bipred weight to not require extra buffer? */ |
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
3445 if(!s->obmc_scratchpad) |
3316 | 3446 s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize); |
3447 | |
3448 /* some macroblocks will be accessed before they're available */ | |
3449 if(FRAME_MBAFF) | |
3450 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t)); | |
2416
06b7d678b582
10l: scratchpad could be allocated before its size was known.
lorenm
parents:
2415
diff
changeset
|
3451 |
1168 | 3452 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1; |
2935 | 3453 return 0; |
1168 | 3454 } |
3455 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3456 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3457 MpegEncContext * const s = &h->s; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3458 int i; |
2967 | 3459 |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3460 src_y -= linesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3461 src_cb -= uvlinesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3462 src_cr -= uvlinesize; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3463 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3464 // There are two lines saved, the line above the the top macroblock of a pair, |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3465 // and the line above the bottom macroblock |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3466 h->left_border[0]= h->top_borders[0][s->mb_x][15]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3467 for(i=1; i<17; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3468 h->left_border[i]= src_y[15+i* linesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3469 } |
2967 | 3470 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3471 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3472 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize); |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3473 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3474 if(!(s->flags&CODEC_FLAG_GRAY)){ |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3475 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3476 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7]; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3477 for(i=1; i<9; i++){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3478 h->left_border[i+17 ]= src_cb[7+i*uvlinesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3479 h->left_border[i+17+9]= src_cr[7+i*uvlinesize]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3480 } |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3481 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3482 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize); |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3483 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3484 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3485 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3486 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3487 MpegEncContext * const s = &h->s; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3488 int temp8, i; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3489 uint64_t temp64; |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3490 int deblock_left = (s->mb_x > 0); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3491 int deblock_top = (s->mb_y > 0); |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3492 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3493 src_y -= linesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3494 src_cb -= uvlinesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3495 src_cr -= uvlinesize + 1; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3496 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3497 #define XCHG(a,b,t,xchg)\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3498 t= a;\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3499 if(xchg)\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3500 a= b;\ |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3501 b= t; |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3502 |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3503 if(deblock_left){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3504 for(i = !deblock_top; i<17; i++){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3505 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3506 } |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3507 } |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3508 |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3509 if(deblock_top){ |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3510 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3511 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1); |
2803 | 3512 if(s->mb_x+1 < s->mb_width){ |
2755 | 3513 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1); |
3514 } | |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3515 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3516 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3517 if(!(s->flags&CODEC_FLAG_GRAY)){ |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3518 if(deblock_left){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3519 for(i = !deblock_top; i<9; i++){ |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3520 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3521 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg); |
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3522 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3523 } |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3524 if(deblock_top){ |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3525 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3526 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3527 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3528 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3529 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3530 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3531 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3532 MpegEncContext * const s = &h->s; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3533 int i; |
2967 | 3534 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3535 src_y -= 2 * linesize; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3536 src_cb -= 2 * uvlinesize; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3537 src_cr -= 2 * uvlinesize; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3538 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3539 // There are two lines saved, the line above the the top macroblock of a pair, |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3540 // and the line above the bottom macroblock |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3541 h->left_border[0]= h->top_borders[0][s->mb_x][15]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3542 h->left_border[1]= h->top_borders[1][s->mb_x][15]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3543 for(i=2; i<34; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3544 h->left_border[i]= src_y[15+i* linesize]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3545 } |
2967 | 3546 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3547 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3548 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3549 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3550 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3551 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3552 if(!(s->flags&CODEC_FLAG_GRAY)){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3553 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3554 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3555 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3556 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3557 for(i=2; i<18; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3558 h->left_border[i+34 ]= src_cb[7+i*uvlinesize]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3559 h->left_border[i+34+18]= src_cr[7+i*uvlinesize]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3560 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3561 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3562 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3563 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3564 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3565 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3566 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3567 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3568 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3569 MpegEncContext * const s = &h->s; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3570 int temp8, i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3571 uint64_t temp64; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3572 int deblock_left = (s->mb_x > 0); |
3316 | 3573 int deblock_top = (s->mb_y > 1); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3574 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3575 tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3576 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3577 src_y -= 2 * linesize + 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3578 src_cb -= 2 * uvlinesize + 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3579 src_cr -= 2 * uvlinesize + 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3580 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3581 #define XCHG(a,b,t,xchg)\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3582 t= a;\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3583 if(xchg)\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3584 a= b;\ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3585 b= t; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3586 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3587 if(deblock_left){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3588 for(i = (!deblock_top)<<1; i<34; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3589 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3590 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3591 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3592 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3593 if(deblock_top){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3594 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3595 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3596 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3597 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1); |
3316 | 3598 if(s->mb_x+1 < s->mb_width){ |
3599 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1); | |
3600 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1); | |
3601 } | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3602 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3603 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3604 if(!(s->flags&CODEC_FLAG_GRAY)){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3605 if(deblock_left){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3606 for(i = (!deblock_top) << 1; i<18; i++){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3607 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3608 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3609 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3610 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3611 if(deblock_top){ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3612 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3613 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3614 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3615 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1); |
2200
733c60a6e30c
h264 deblocking crash patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2175
diff
changeset
|
3616 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3617 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3618 } |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3619 |
1168 | 3620 static void hl_decode_mb(H264Context *h){ |
3621 MpegEncContext * const s = &h->s; | |
3622 const int mb_x= s->mb_x; | |
3623 const int mb_y= s->mb_y; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
3624 const int mb_xy= mb_x + mb_y*s->mb_stride; |
1168 | 3625 const int mb_type= s->current_picture.mb_type[mb_xy]; |
3626 uint8_t *dest_y, *dest_cb, *dest_cr; | |
3627 int linesize, uvlinesize /*dct_offset*/; | |
3628 int i; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3629 int *block_offset = &h->block_offset[0]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3630 const unsigned int bottom = mb_y & 1; |
2763 | 3631 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass); |
3632 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride); | |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3633 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride); |
1168 | 3634 |
3635 if(!s->decode) | |
3636 return; | |
3637 | |
3638 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; | |
3639 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
3640 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; | |
3641 | |
4000 | 3642 s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + 64, s->linesize, 4); |
3643 s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + 64, dest_cr - dest_cb, 2); | |
3644 | |
3316 | 3645 if (MB_FIELD) { |
3646 linesize = h->mb_linesize = s->linesize * 2; | |
3647 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3648 block_offset = &h->block_offset[24]; |
1168 | 3649 if(mb_y&1){ //FIXME move out of this func? |
3650 dest_y -= s->linesize*15; | |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3651 dest_cb-= s->uvlinesize*7; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3652 dest_cr-= s->uvlinesize*7; |
1168 | 3653 } |
3316 | 3654 if(FRAME_MBAFF) { |
3655 int list; | |
3656 for(list=0; list<2; list++){ | |
3657 if(!USES_LIST(mb_type, list)) | |
3658 continue; | |
3659 if(IS_16X16(mb_type)){ | |
3660 int8_t *ref = &h->ref_cache[list][scan8[0]]; | |
3661 fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1); | |
3662 }else{ | |
3663 for(i=0; i<16; i+=4){ | |
3664 //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ? | |
3665 int ref = h->ref_cache[list][scan8[i]]; | |
3666 if(ref >= 0) | |
3667 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1); | |
3668 } | |
3669 } | |
3670 } | |
3671 } | |
1168 | 3672 } else { |
3316 | 3673 linesize = h->mb_linesize = s->linesize; |
3674 uvlinesize = h->mb_uvlinesize = s->uvlinesize; | |
1168 | 3675 // dct_offset = s->linesize * 16; |
3676 } | |
2967 | 3677 |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3678 if(transform_bypass){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3679 idct_dc_add = |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3680 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3681 }else if(IS_8x8DCT(mb_type)){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3682 idct_dc_add = s->dsp.h264_idct8_dc_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3683 idct_add = s->dsp.h264_idct8_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3684 }else{ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3685 idct_dc_add = s->dsp.h264_idct_dc_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3686 idct_add = s->dsp.h264_idct_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3687 } |
1168 | 3688 |
3316 | 3689 if(FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type) |
3690 && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){ | |
3691 int mbt_y = mb_y&~1; | |
3692 uint8_t *top_y = s->current_picture.data[0] + (mbt_y * 16* s->linesize ) + mb_x * 16; | |
3693 uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8; | |
3694 uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8; | |
3695 xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1); | |
3696 } | |
3697 | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3698 if (IS_INTRA_PCM(mb_type)) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3699 unsigned int x, y; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3700 |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3701 // The pixels are stored in h->mb array in the same order as levels, |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3702 // copy them in output in the correct order. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3703 for(i=0; i<16; i++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3704 for (y=0; y<4; y++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3705 for (x=0; x<4; x++) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3706 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x]; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3707 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3708 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3709 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3710 for(i=16; i<16+4; i++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3711 for (y=0; y<4; y++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3712 for (x=0; x<4; x++) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3713 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x]; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3714 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3715 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3716 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3717 for(i=20; i<20+4; i++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3718 for (y=0; y<4; y++) { |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3719 for (x=0; x<4; x++) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3720 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x]; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3721 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3722 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
3723 } |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3724 } else { |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3725 if(IS_INTRA(mb_type)){ |
3316 | 3726 if(h->deblocking_filter && !FRAME_MBAFF) |
3727 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1); | |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3728 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3729 if(!(s->flags&CODEC_FLAG_GRAY)){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3730 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3731 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3732 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3733 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3734 if(IS_INTRA4x4(mb_type)){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3735 if(!s->encoding){ |
2755 | 3736 if(IS_8x8DCT(mb_type)){ |
3737 for(i=0; i<16; i+=4){ | |
3738 uint8_t * const ptr= dest_y + block_offset[i]; | |
3739 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ]; | |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3740 const int nnz = h->non_zero_count_cache[ scan8[i] ]; |
2755 | 3741 h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000, |
3742 (h->topright_samples_available<<(i+1))&0x8000, linesize); | |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3743 if(nnz){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3744 if(nnz == 1 && h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3745 idct_dc_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3746 else |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3747 idct_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3748 } |
2755 | 3749 } |
3750 }else | |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3751 for(i=0; i<16; i++){ |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3752 uint8_t * const ptr= dest_y + block_offset[i]; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3753 uint8_t *topright; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3754 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ]; |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3755 int nnz, tr; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3756 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3757 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3758 const int topright_avail= (h->topright_samples_available<<i)&0x8000; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3759 assert(mb_y || linesize <= block_offset[i]); |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3760 if(!topright_avail){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3761 tr= ptr[3 - linesize]*0x01010101; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3762 topright= (uint8_t*) &tr; |
2967 | 3763 }else |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3764 topright= ptr + 4 - linesize; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3765 }else |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3766 topright= NULL; |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3767 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3768 h->pred4x4[ dir ](ptr, topright, linesize); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3769 nnz = h->non_zero_count_cache[ scan8[i] ]; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3770 if(nnz){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3771 if(s->codec_id == CODEC_ID_H264){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3772 if(nnz == 1 && h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3773 idct_dc_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3774 else |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3775 idct_add(ptr, h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3776 }else |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3777 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3778 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3779 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3780 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3781 }else{ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3782 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize); |
2763 | 3783 if(s->codec_id == CODEC_ID_H264){ |
3784 if(!transform_bypass) | |
2919 | 3785 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]); |
2763 | 3786 }else |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3787 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3788 } |
3316 | 3789 if(h->deblocking_filter && !FRAME_MBAFF) |
3790 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0); | |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3791 }else if(s->codec_id == CODEC_ID_H264){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3792 hl_motion(h, dest_y, dest_cb, dest_cr, |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
3793 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab, |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
3794 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab, |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3795 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3796 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3797 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3798 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3799 if(!IS_INTRA4x4(mb_type)){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3800 if(s->codec_id == CODEC_ID_H264){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3801 if(IS_INTRA16x16(mb_type)){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3802 for(i=0; i<16; i++){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3803 if(h->non_zero_count_cache[ scan8[i] ]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3804 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3805 else if(h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3806 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3807 } |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3808 }else{ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3809 const int di = IS_8x8DCT(mb_type) ? 4 : 1; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3810 for(i=0; i<16; i+=di){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3811 int nnz = h->non_zero_count_cache[ scan8[i] ]; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3812 if(nnz){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3813 if(nnz==1 && h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3814 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3815 else |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3816 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3817 } |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3818 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3819 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3820 }else{ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3821 for(i=0; i<16; i++){ |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3822 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
3823 uint8_t * const ptr= dest_y + block_offset[i]; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3824 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0); |
1234 | 3825 } |
1168 | 3826 } |
3827 } | |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3828 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3829 |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3830 if(!(s->flags&CODEC_FLAG_GRAY)){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3831 uint8_t *dest[2] = {dest_cb, dest_cr}; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3832 if(transform_bypass){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3833 idct_add = idct_dc_add = s->dsp.add_pixels4; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3834 }else{ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3835 idct_add = s->dsp.h264_idct_add; |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3836 idct_dc_add = s->dsp.h264_idct_dc_add; |
2919 | 3837 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp][0]); |
3838 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp][0]); | |
2763 | 3839 } |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3840 if(s->codec_id == CODEC_ID_H264){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3841 for(i=16; i<16+8; i++){ |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3842 if(h->non_zero_count_cache[ scan8[i] ]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3843 idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize); |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3844 else if(h->mb[i*16]) |
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3845 idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize); |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3846 } |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3847 }else{ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3848 for(i=16; i<16+8; i++){ |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3849 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
3850 uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i]; |
2509
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3851 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2); |
b3b06ba8787c
remove goto and reindent patch by (Loic Le Loarer <lll+ffmpeg m4x org>)
michael
parents:
2504
diff
changeset
|
3852 } |
1250 | 3853 } |
1168 | 3854 } |
3855 } | |
3856 } | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3857 if(h->deblocking_filter) { |
3316 | 3858 if (FRAME_MBAFF) { |
3859 //FIXME try deblocking one mb at a time? | |
3860 // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3861 const int mb_y = s->mb_y - 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3862 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3863 const int mb_xy= mb_x + mb_y*s->mb_stride; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3864 const int mb_type_top = s->current_picture.mb_type[mb_xy]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3865 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3866 if (!bottom) return; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3867 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3868 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3869 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3870 |
3316 | 3871 if(IS_INTRA(mb_type_top | mb_type_bottom)) |
3872 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0); | |
3873 | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3874 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize); |
3316 | 3875 // deblock a pair |
2967 | 3876 // top |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3877 s->mb_y--; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3878 tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3879 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb |
3316 | 3880 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3881 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3882 // bottom |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3883 s->mb_y++; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3884 tprintf("call mbaff filter_mb\n"); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3885 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb |
3316 | 3886 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3887 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3888 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3889 tprintf("call filter_mb\n"); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3890 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
3891 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
3892 filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
3893 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
3894 } |
1168 | 3895 } |
3896 | |
3897 /** | |
3898 * fills the default_ref_list. | |
3899 */ | |
3900 static int fill_default_ref_list(H264Context *h){ | |
3901 MpegEncContext * const s = &h->s; | |
3902 int i; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3903 int smallest_poc_greater_than_current = -1; |
2582 | 3904 Picture sorted_short_ref[32]; |
2967 | 3905 |
1168 | 3906 if(h->slice_type==B_TYPE){ |
3907 int out_i; | |
2713 | 3908 int limit= INT_MIN; |
1168 | 3909 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3910 /* sort frame according to poc in B slice */ |
1168 | 3911 for(out_i=0; out_i<h->short_ref_count; out_i++){ |
2713 | 3912 int best_i=INT_MIN; |
2255
507690ff49a2
assertion when playing AVC/H.264 streams fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2254
diff
changeset
|
3913 int best_poc=INT_MAX; |
1168 | 3914 |
3915 for(i=0; i<h->short_ref_count; i++){ | |
3916 const int poc= h->short_ref[i]->poc; | |
3917 if(poc > limit && poc < best_poc){ | |
3918 best_poc= poc; | |
3919 best_i= i; | |
3920 } | |
3921 } | |
2967 | 3922 |
2713 | 3923 assert(best_i != INT_MIN); |
2967 | 3924 |
1168 | 3925 limit= best_poc; |
3926 sorted_short_ref[out_i]= *h->short_ref[best_i]; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3927 tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3928 if (-1 == smallest_poc_greater_than_current) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3929 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3930 smallest_poc_greater_than_current = out_i; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3931 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3932 } |
1168 | 3933 } |
3934 } | |
3935 | |
3936 if(s->picture_structure == PICT_FRAME){ | |
3937 if(h->slice_type==B_TYPE){ | |
3938 int list; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3939 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3940 |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3941 // find the largest poc |
1168 | 3942 for(list=0; list<2; list++){ |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3943 int index = 0; |
2447 | 3944 int j= -99; |
3945 int step= list ? -1 : 1; | |
3946 | |
3947 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) { | |
3948 while(j<0 || j>= h->short_ref_count){ | |
2680
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3949 if(j != -99 && step == (list ? -1 : 1)) |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
3950 return -1; |
2447 | 3951 step = -step; |
3952 j= smallest_poc_greater_than_current + (step>>1); | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3953 } |
2447 | 3954 if(sorted_short_ref[j].reference != 3) continue; |
3955 h->default_ref_list[list][index ]= sorted_short_ref[j]; | |
3956 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num; | |
1168 | 3957 } |
3958 | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3959 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3960 if(h->long_ref[i] == NULL) continue; |
1168 | 3961 if(h->long_ref[i]->reference != 3) continue; |
3962 | |
3963 h->default_ref_list[ list ][index ]= *h->long_ref[i]; | |
3964 h->default_ref_list[ list ][index++].pic_id= i;; | |
3965 } | |
2967 | 3966 |
2447 | 3967 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){ |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3968 // swap the two first elements of L1 when |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3969 // L0 and L1 are identical |
1168 | 3970 Picture temp= h->default_ref_list[1][0]; |
3971 h->default_ref_list[1][0] = h->default_ref_list[1][1]; | |
2673 | 3972 h->default_ref_list[1][1] = temp; |
1168 | 3973 } |
3974 | |
3975 if(index < h->ref_count[ list ]) | |
3976 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index)); | |
3977 } | |
3978 }else{ | |
3979 int index=0; | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
3980 for(i=0; i<h->short_ref_count; i++){ |
1168 | 3981 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit |
3982 h->default_ref_list[0][index ]= *h->short_ref[i]; | |
3983 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num; | |
3984 } | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
3985 for(i = 0; i < 16; i++){ |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
3986 if(h->long_ref[i] == NULL) continue; |
1168 | 3987 if(h->long_ref[i]->reference != 3) continue; |
3988 h->default_ref_list[0][index ]= *h->long_ref[i]; | |
3989 h->default_ref_list[0][index++].pic_id= i;; | |
3990 } | |
3991 if(index < h->ref_count[0]) | |
3992 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index)); | |
3993 } | |
3994 }else{ //FIELD | |
3995 if(h->slice_type==B_TYPE){ | |
3996 }else{ | |
3997 //FIXME second field balh | |
3998 } | |
3999 } | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4000 #ifdef TRACE |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4001 for (i=0; i<h->ref_count[0]; i++) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4002 tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4003 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4004 if(h->slice_type==B_TYPE){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4005 for (i=0; i<h->ref_count[1]; i++) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4006 tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4007 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4008 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4009 #endif |
1168 | 4010 return 0; |
4011 } | |
4012 | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4013 static void print_short_term(H264Context *h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4014 static void print_long_term(H264Context *h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4015 |
1168 | 4016 static int decode_ref_pic_list_reordering(H264Context *h){ |
4017 MpegEncContext * const s = &h->s; | |
2680
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4018 int list, index; |
2967 | 4019 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4020 print_short_term(h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4021 print_long_term(h); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4022 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func |
2967 | 4023 |
1168 | 4024 for(list=0; list<2; list++){ |
4025 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]); | |
4026 | |
4027 if(get_bits1(&s->gb)){ | |
4028 int pred= h->curr_pic_num; | |
4029 | |
4030 for(index=0; ; index++){ | |
4031 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb); | |
4032 int pic_id; | |
4033 int i; | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
4034 Picture *ref = NULL; |
2967 | 4035 |
4036 if(reordering_of_pic_nums_idc==3) | |
2284
6d26e105f68f
h.264 ref list reordering bugfix patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2272
diff
changeset
|
4037 break; |
2967 | 4038 |
1168 | 4039 if(index >= h->ref_count[list]){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4040 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n"); |
1168 | 4041 return -1; |
4042 } | |
2967 | 4043 |
1168 | 4044 if(reordering_of_pic_nums_idc<3){ |
4045 if(reordering_of_pic_nums_idc<2){ | |
4046 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1; | |
4047 | |
4048 if(abs_diff_pic_num >= h->max_pic_num){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4049 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n"); |
1168 | 4050 return -1; |
4051 } | |
4052 | |
4053 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num; | |
4054 else pred+= abs_diff_pic_num; | |
4055 pred &= h->max_pic_num - 1; | |
2967 | 4056 |
2683 | 4057 for(i= h->short_ref_count-1; i>=0; i--){ |
4058 ref = h->short_ref[i]; | |
4059 assert(ref->reference == 3); | |
4060 assert(!ref->long_ref); | |
4061 if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer | |
1168 | 4062 break; |
4063 } | |
2683 | 4064 if(i>=0) |
4065 ref->pic_id= ref->frame_num; | |
1168 | 4066 }else{ |
4067 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx | |
2683 | 4068 ref = h->long_ref[pic_id]; |
4069 ref->pic_id= pic_id; | |
4070 assert(ref->reference == 3); | |
4071 assert(ref->long_ref); | |
4072 i=0; | |
1168 | 4073 } |
4074 | |
2485
2844b8cf4e11
H.264 multiplce instance in reference list patch by (Loic <lll+ffmpeg m4x org )
michael
parents:
2484
diff
changeset
|
4075 if (i < 0) { |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4076 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n"); |
1168 | 4077 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME |
2683 | 4078 } else { |
4079 for(i=index; i+1<h->ref_count[list]; i++){ | |
4080 if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id) | |
4081 break; | |
2650
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
4082 } |
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
4083 for(; i > index; i--){ |
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
4084 h->ref_list[list][i]= h->ref_list[list][i-1]; |
d321fbf0b2ce
reverse 1 hunk from 1.96->1.97, fixes decoding of (MR1_BT_A.h264 and MR1_MW_A.264)
michael
parents:
2649
diff
changeset
|
4085 } |
2683 | 4086 h->ref_list[list][index]= *ref; |
1168 | 4087 } |
2284
6d26e105f68f
h.264 ref list reordering bugfix patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2272
diff
changeset
|
4088 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4089 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n"); |
1168 | 4090 return -1; |
4091 } | |
4092 } | |
4093 } | |
4094 | |
4095 if(h->slice_type!=B_TYPE) break; | |
4096 } | |
2680
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4097 for(list=0; list<2; list++){ |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4098 for(index= 0; index < h->ref_count[list]; index++){ |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4099 if(!h->ref_list[list][index].data[0]) |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4100 h->ref_list[list][index]= s->current_picture; |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4101 } |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4102 if(h->slice_type!=B_TYPE) break; |
1a6192e0b1f9
fill missing reference pictures with something to avoid assertion failure
michael
parents:
2675
diff
changeset
|
4103 } |
2967 | 4104 |
2396 | 4105 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred) |
4106 direct_dist_scale_factor(h); | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
4107 direct_ref_list_init(h); |
2967 | 4108 return 0; |
1168 | 4109 } |
4110 | |
3379
69901769c811
fill_mbaff_ref_list is missing a return statement, its return value
mru
parents:
3341
diff
changeset
|
4111 static void fill_mbaff_ref_list(H264Context *h){ |
3316 | 4112 int list, i, j; |
4113 for(list=0; list<2; list++){ | |
4114 for(i=0; i<h->ref_count[list]; i++){ | |
4115 Picture *frame = &h->ref_list[list][i]; | |
4116 Picture *field = &h->ref_list[list][16+2*i]; | |
4117 field[0] = *frame; | |
4118 for(j=0; j<3; j++) | |
4119 field[0].linesize[j] <<= 1; | |
4120 field[1] = field[0]; | |
4121 for(j=0; j<3; j++) | |
4122 field[1].data[j] += frame->linesize[j]; | |
4123 | |
4124 h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i]; | |
4125 h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i]; | |
4126 for(j=0; j<2; j++){ | |
4127 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j]; | |
4128 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j]; | |
4129 } | |
4130 } | |
4131 } | |
4132 for(j=0; j<h->ref_count[1]; j++){ | |
4133 for(i=0; i<h->ref_count[0]; i++) | |
4134 h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i]; | |
4135 memcpy(h->implicit_weight[16+2*j], h->implicit_weight[j], sizeof(*h->implicit_weight)); | |
4136 memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight)); | |
4137 } | |
4138 } | |
4139 | |
1168 | 4140 static int pred_weight_table(H264Context *h){ |
4141 MpegEncContext * const s = &h->s; | |
4142 int list, i; | |
2415 | 4143 int luma_def, chroma_def; |
2967 | 4144 |
2415 | 4145 h->use_weight= 0; |
4146 h->use_weight_chroma= 0; | |
1168 | 4147 h->luma_log2_weight_denom= get_ue_golomb(&s->gb); |
4148 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb); | |
2415 | 4149 luma_def = 1<<h->luma_log2_weight_denom; |
4150 chroma_def = 1<<h->chroma_log2_weight_denom; | |
1168 | 4151 |
4152 for(list=0; list<2; list++){ | |
4153 for(i=0; i<h->ref_count[list]; i++){ | |
4154 int luma_weight_flag, chroma_weight_flag; | |
2967 | 4155 |
1168 | 4156 luma_weight_flag= get_bits1(&s->gb); |
4157 if(luma_weight_flag){ | |
4158 h->luma_weight[list][i]= get_se_golomb(&s->gb); | |
4159 h->luma_offset[list][i]= get_se_golomb(&s->gb); | |
2415 | 4160 if( h->luma_weight[list][i] != luma_def |
4161 || h->luma_offset[list][i] != 0) | |
4162 h->use_weight= 1; | |
4163 }else{ | |
4164 h->luma_weight[list][i]= luma_def; | |
4165 h->luma_offset[list][i]= 0; | |
1168 | 4166 } |
4167 | |
4168 chroma_weight_flag= get_bits1(&s->gb); | |
4169 if(chroma_weight_flag){ | |
4170 int j; | |
4171 for(j=0; j<2; j++){ | |
4172 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb); | |
4173 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb); | |
2415 | 4174 if( h->chroma_weight[list][i][j] != chroma_def |
4175 || h->chroma_offset[list][i][j] != 0) | |
4176 h->use_weight_chroma= 1; | |
4177 } | |
4178 }else{ | |
4179 int j; | |
4180 for(j=0; j<2; j++){ | |
4181 h->chroma_weight[list][i][j]= chroma_def; | |
4182 h->chroma_offset[list][i][j]= 0; | |
1168 | 4183 } |
4184 } | |
4185 } | |
4186 if(h->slice_type != B_TYPE) break; | |
4187 } | |
2415 | 4188 h->use_weight= h->use_weight || h->use_weight_chroma; |
1168 | 4189 return 0; |
4190 } | |
4191 | |
2415 | 4192 static void implicit_weight_table(H264Context *h){ |
4193 MpegEncContext * const s = &h->s; | |
4194 int ref0, ref1; | |
4195 int cur_poc = s->current_picture_ptr->poc; | |
4196 | |
4197 if( h->ref_count[0] == 1 && h->ref_count[1] == 1 | |
4198 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){ | |
4199 h->use_weight= 0; | |
4200 h->use_weight_chroma= 0; | |
4201 return; | |
4202 } | |
4203 | |
4204 h->use_weight= 2; | |
4205 h->use_weight_chroma= 2; | |
4206 h->luma_log2_weight_denom= 5; | |
4207 h->chroma_log2_weight_denom= 5; | |
4208 | |
4209 for(ref0=0; ref0 < h->ref_count[0]; ref0++){ | |
4210 int poc0 = h->ref_list[0][ref0].poc; | |
4211 for(ref1=0; ref1 < h->ref_count[1]; ref1++){ | |
2519 | 4212 int poc1 = h->ref_list[1][ref1].poc; |
2415 | 4213 int td = clip(poc1 - poc0, -128, 127); |
4214 if(td){ | |
4215 int tb = clip(cur_poc - poc0, -128, 127); | |
4001 | 4216 int tx = (16384 + (FFABS(td) >> 1)) / td; |
2415 | 4217 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2; |
4218 if(dist_scale_factor < -64 || dist_scale_factor > 128) | |
4219 h->implicit_weight[ref0][ref1] = 32; | |
4220 else | |
4221 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor; | |
4222 }else | |
4223 h->implicit_weight[ref0][ref1] = 32; | |
4224 } | |
4225 } | |
4226 } | |
4227 | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4228 static inline void unreference_pic(H264Context *h, Picture *pic){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4229 int i; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4230 pic->reference=0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4231 if(pic == h->delayed_output_pic) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4232 pic->reference=1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4233 else{ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4234 for(i = 0; h->delayed_pic[i]; i++) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4235 if(pic == h->delayed_pic[i]){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4236 pic->reference=1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4237 break; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4238 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4239 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4240 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4241 |
1168 | 4242 /** |
2392 | 4243 * instantaneous decoder refresh. |
1168 | 4244 */ |
4245 static void idr(H264Context *h){ | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4246 int i; |
1168 | 4247 |
2484
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
4248 for(i=0; i<16; i++){ |
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
4249 if (h->long_ref[i] != NULL) { |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4250 unreference_pic(h, h->long_ref[i]); |
2484
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
4251 h->long_ref[i]= NULL; |
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
4252 } |
1168 | 4253 } |
4254 h->long_ref_count=0; | |
4255 | |
4256 for(i=0; i<h->short_ref_count; i++){ | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4257 unreference_pic(h, h->short_ref[i]); |
1168 | 4258 h->short_ref[i]= NULL; |
4259 } | |
4260 h->short_ref_count=0; | |
4261 } | |
4262 | |
2640 | 4263 /* forget old pics after a seek */ |
4264 static void flush_dpb(AVCodecContext *avctx){ | |
4265 H264Context *h= avctx->priv_data; | |
4266 int i; | |
3144 | 4267 for(i=0; i<16; i++) { |
4268 if(h->delayed_pic[i]) | |
4269 h->delayed_pic[i]->reference= 0; | |
2640 | 4270 h->delayed_pic[i]= NULL; |
3144 | 4271 } |
4272 if(h->delayed_output_pic) | |
4273 h->delayed_output_pic->reference= 0; | |
2640 | 4274 h->delayed_output_pic= NULL; |
4275 idr(h); | |
2751
3247049dfc7a
fix segfault if flush is called before the first frame
mru
parents:
2748
diff
changeset
|
4276 if(h->s.current_picture_ptr) |
3247049dfc7a
fix segfault if flush is called before the first frame
mru
parents:
2748
diff
changeset
|
4277 h->s.current_picture_ptr->reference= 0; |
2640 | 4278 } |
4279 | |
1168 | 4280 /** |
4281 * | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4282 * @return the removed picture or NULL if an error occurs |
1168 | 4283 */ |
4284 static Picture * remove_short(H264Context *h, int frame_num){ | |
1169 | 4285 MpegEncContext * const s = &h->s; |
1168 | 4286 int i; |
2967 | 4287 |
1169 | 4288 if(s->avctx->debug&FF_DEBUG_MMCO) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4289 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count); |
2967 | 4290 |
1168 | 4291 for(i=0; i<h->short_ref_count; i++){ |
4292 Picture *pic= h->short_ref[i]; | |
1169 | 4293 if(s->avctx->debug&FF_DEBUG_MMCO) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4294 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic); |
1168 | 4295 if(pic->frame_num == frame_num){ |
4296 h->short_ref[i]= NULL; | |
4297 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*)); | |
4298 h->short_ref_count--; | |
4299 return pic; | |
4300 } | |
4301 } | |
4302 return NULL; | |
4303 } | |
4304 | |
4305 /** | |
4306 * | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4307 * @return the removed picture or NULL if an error occurs |
1168 | 4308 */ |
4309 static Picture * remove_long(H264Context *h, int i){ | |
4310 Picture *pic; | |
4311 | |
4312 pic= h->long_ref[i]; | |
4313 h->long_ref[i]= NULL; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4314 if(pic) h->long_ref_count--; |
1168 | 4315 |
4316 return pic; | |
4317 } | |
4318 | |
4319 /** | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4320 * print short term list |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4321 */ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4322 static void print_short_term(H264Context *h) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4323 uint32_t i; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4324 if(h->s.avctx->debug&FF_DEBUG_MMCO) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4325 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n"); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4326 for(i=0; i<h->short_ref_count; i++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4327 Picture *pic= h->short_ref[i]; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4328 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4329 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4330 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4331 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4332 |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4333 /** |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4334 * print long term list |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4335 */ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4336 static void print_long_term(H264Context *h) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4337 uint32_t i; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4338 if(h->s.avctx->debug&FF_DEBUG_MMCO) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4339 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n"); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4340 for(i = 0; i < 16; i++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4341 Picture *pic= h->long_ref[i]; |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4342 if (pic) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4343 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4344 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4345 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4346 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4347 } |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4348 |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4349 /** |
1168 | 4350 * Executes the reference picture marking (memory management control operations). |
4351 */ | |
4352 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){ | |
4353 MpegEncContext * const s = &h->s; | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4354 int i, j; |
1168 | 4355 int current_is_long=0; |
4356 Picture *pic; | |
2967 | 4357 |
1168 | 4358 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0) |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4359 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n"); |
2967 | 4360 |
1168 | 4361 for(i=0; i<mmco_count; i++){ |
4362 if(s->avctx->debug&FF_DEBUG_MMCO) | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4363 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index); |
1168 | 4364 |
4365 switch(mmco[i].opcode){ | |
4366 case MMCO_SHORT2UNUSED: | |
4367 pic= remove_short(h, mmco[i].short_frame_num); | |
2766
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4368 if(pic) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4369 unreference_pic(h, pic); |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4370 else if(s->avctx->debug&FF_DEBUG_MMCO) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4371 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n"); |
1168 | 4372 break; |
4373 case MMCO_SHORT2LONG: | |
4374 pic= remove_long(h, mmco[i].long_index); | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4375 if(pic) unreference_pic(h, pic); |
2967 | 4376 |
1168 | 4377 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num); |
4378 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
2484
a3188eb4266c
correct long term picture management patch by (Loic <lll+ffmpeg m4x org>)
michael
parents:
2471
diff
changeset
|
4379 h->long_ref_count++; |
1168 | 4380 break; |
4381 case MMCO_LONG2UNUSED: | |
4382 pic= remove_long(h, mmco[i].long_index); | |
2766
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4383 if(pic) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4384 unreference_pic(h, pic); |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4385 else if(s->avctx->debug&FF_DEBUG_MMCO) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4386 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n"); |
1168 | 4387 break; |
4388 case MMCO_LONG: | |
4389 pic= remove_long(h, mmco[i].long_index); | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4390 if(pic) unreference_pic(h, pic); |
2967 | 4391 |
1168 | 4392 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr; |
4393 h->long_ref[ mmco[i].long_index ]->long_ref=1; | |
4394 h->long_ref_count++; | |
2967 | 4395 |
1168 | 4396 current_is_long=1; |
4397 break; | |
4398 case MMCO_SET_MAX_LONG: | |
4399 assert(mmco[i].long_index <= 16); | |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4400 // just remove the long term which index is greater than new max |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4401 for(j = mmco[i].long_index; j<16; j++){ |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4402 pic = remove_long(h, j); |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4403 if (pic) unreference_pic(h, pic); |
1168 | 4404 } |
4405 break; | |
4406 case MMCO_RESET: | |
4407 while(h->short_ref_count){ | |
4408 pic= remove_short(h, h->short_ref[0]->frame_num); | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4409 unreference_pic(h, pic); |
1168 | 4410 } |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4411 for(j = 0; j < 16; j++) { |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4412 pic= remove_long(h, j); |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4413 if(pic) unreference_pic(h, pic); |
1168 | 4414 } |
4415 break; | |
4416 default: assert(0); | |
4417 } | |
4418 } | |
2967 | 4419 |
1168 | 4420 if(!current_is_long){ |
4421 pic= remove_short(h, s->current_picture_ptr->frame_num); | |
4422 if(pic){ | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
4423 unreference_pic(h, pic); |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4424 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n"); |
1168 | 4425 } |
2967 | 4426 |
1168 | 4427 if(h->short_ref_count) |
1169 | 4428 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*)); |
4429 | |
4430 h->short_ref[0]= s->current_picture_ptr; | |
1168 | 4431 h->short_ref[0]->long_ref=0; |
4432 h->short_ref_count++; | |
4433 } | |
2967 | 4434 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4435 print_short_term(h); |
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
4436 print_long_term(h); |
2967 | 4437 return 0; |
1168 | 4438 } |
4439 | |
4440 static int decode_ref_pic_marking(H264Context *h){ | |
4441 MpegEncContext * const s = &h->s; | |
4442 int i; | |
2967 | 4443 |
1168 | 4444 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields |
4445 s->broken_link= get_bits1(&s->gb) -1; | |
4446 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx | |
4447 if(h->mmco[0].long_index == -1) | |
4448 h->mmco_index= 0; | |
4449 else{ | |
4450 h->mmco[0].opcode= MMCO_LONG; | |
4451 h->mmco_index= 1; | |
2967 | 4452 } |
1168 | 4453 }else{ |
4454 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag | |
2967 | 4455 for(i= 0; i<MAX_MMCO_COUNT; i++) { |
1168 | 4456 MMCOOpcode opcode= get_ue_golomb(&s->gb);; |
4457 | |
4458 h->mmco[i].opcode= opcode; | |
4459 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){ | |
4460 h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields | |
4461 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){ | |
2846
40765c51a7a9
Compilation fixes part 1 patch by (Arvind R. and Burkhard Plaum, plaum, ipf uni-stuttgart de)
michael
parents:
2844
diff
changeset
|
4462 av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco); |
1168 | 4463 return -1; |
4464 }*/ | |
4465 } | |
4466 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){ | |
4467 h->mmco[i].long_index= get_ue_golomb(&s->gb); | |
4468 if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ h->mmco[i].long_index >= 16){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4469 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode); |
1168 | 4470 return -1; |
4471 } | |
4472 } | |
2967 | 4473 |
1168 | 4474 if(opcode > MMCO_LONG){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4475 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode); |
1168 | 4476 return -1; |
4477 } | |
2255
507690ff49a2
assertion when playing AVC/H.264 streams fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2254
diff
changeset
|
4478 if(opcode == MMCO_END) |
507690ff49a2
assertion when playing AVC/H.264 streams fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2254
diff
changeset
|
4479 break; |
1168 | 4480 } |
4481 h->mmco_index= i; | |
4482 }else{ | |
4483 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count); | |
4484 | |
4485 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields | |
4486 h->mmco[0].opcode= MMCO_SHORT2UNUSED; | |
4487 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num; | |
4488 h->mmco_index= 1; | |
4489 }else | |
4490 h->mmco_index= 0; | |
4491 } | |
4492 } | |
2967 | 4493 |
4494 return 0; | |
1168 | 4495 } |
4496 | |
4497 static int init_poc(H264Context *h){ | |
4498 MpegEncContext * const s = &h->s; | |
4499 const int max_frame_num= 1<<h->sps.log2_max_frame_num; | |
4500 int field_poc[2]; | |
4501 | |
4502 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
4503 h->frame_num_offset= 0; | |
4504 }else{ | |
4505 if(h->frame_num < h->prev_frame_num) | |
4506 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num; | |
4507 else | |
4508 h->frame_num_offset= h->prev_frame_num_offset; | |
4509 } | |
4510 | |
4511 if(h->sps.poc_type==0){ | |
4512 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb; | |
4513 | |
2649
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4514 if(h->nal_unit_type == NAL_IDR_SLICE){ |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4515 h->prev_poc_msb= |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4516 h->prev_poc_lsb= 0; |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4517 } |
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
4518 |
1168 | 4519 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2) |
4520 h->poc_msb = h->prev_poc_msb + max_poc_lsb; | |
4521 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2) | |
4522 h->poc_msb = h->prev_poc_msb - max_poc_lsb; | |
4523 else | |
4524 h->poc_msb = h->prev_poc_msb; | |
4525 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb); | |
2967 | 4526 field_poc[0] = |
1168 | 4527 field_poc[1] = h->poc_msb + h->poc_lsb; |
2967 | 4528 if(s->picture_structure == PICT_FRAME) |
1168 | 4529 field_poc[1] += h->delta_poc_bottom; |
4530 }else if(h->sps.poc_type==1){ | |
4531 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc; | |
4532 int i; | |
4533 | |
4534 if(h->sps.poc_cycle_length != 0) | |
4535 abs_frame_num = h->frame_num_offset + h->frame_num; | |
4536 else | |
4537 abs_frame_num = 0; | |
4538 | |
4539 if(h->nal_ref_idc==0 && abs_frame_num > 0) | |
4540 abs_frame_num--; | |
2967 | 4541 |
1168 | 4542 expected_delta_per_poc_cycle = 0; |
4543 for(i=0; i < h->sps.poc_cycle_length; i++) | |
4544 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse | |
4545 | |
4546 if(abs_frame_num > 0){ | |
4547 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length; | |
4548 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length; | |
4549 | |
4550 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle; | |
4551 for(i = 0; i <= frame_num_in_poc_cycle; i++) | |
4552 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ]; | |
4553 } else | |
4554 expectedpoc = 0; | |
4555 | |
2967 | 4556 if(h->nal_ref_idc == 0) |
1168 | 4557 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic; |
2967 | 4558 |
1168 | 4559 field_poc[0] = expectedpoc + h->delta_poc[0]; |
4560 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field; | |
4561 | |
4562 if(s->picture_structure == PICT_FRAME) | |
4563 field_poc[1] += h->delta_poc[1]; | |
4564 }else{ | |
4565 int poc; | |
4566 if(h->nal_unit_type == NAL_IDR_SLICE){ | |
4567 poc= 0; | |
4568 }else{ | |
4569 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num); | |
4570 else poc= 2*(h->frame_num_offset + h->frame_num) - 1; | |
4571 } | |
4572 field_poc[0]= poc; | |
4573 field_poc[1]= poc; | |
4574 } | |
2967 | 4575 |
1168 | 4576 if(s->picture_structure != PICT_BOTTOM_FIELD) |
4577 s->current_picture_ptr->field_poc[0]= field_poc[0]; | |
4578 if(s->picture_structure != PICT_TOP_FIELD) | |
4579 s->current_picture_ptr->field_poc[1]= field_poc[1]; | |
4580 if(s->picture_structure == PICT_FRAME) // FIXME field pix? | |
4581 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]); | |
4582 | |
4583 return 0; | |
4584 } | |
4585 | |
4586 /** | |
4587 * decodes a slice header. | |
4588 * this will allso call MPV_common_init() and frame_start() as needed | |
4589 */ | |
4590 static int decode_slice_header(H264Context *h){ | |
4591 MpegEncContext * const s = &h->s; | |
4592 int first_mb_in_slice, pps_id; | |
4593 int num_ref_idx_active_override_flag; | |
4594 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE}; | |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4595 int slice_type; |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4596 int default_ref_list_done = 0; |
1168 | 4597 |
4598 s->current_picture.reference= h->nal_ref_idc != 0; | |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
4599 s->dropable= h->nal_ref_idc == 0; |
1168 | 4600 |
4601 first_mb_in_slice= get_ue_golomb(&s->gb); | |
4602 | |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4603 slice_type= get_ue_golomb(&s->gb); |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4604 if(slice_type > 9){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4605 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y); |
2392 | 4606 return -1; |
1168 | 4607 } |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4608 if(slice_type > 4){ |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4609 slice_type -= 5; |
1168 | 4610 h->slice_type_fixed=1; |
4611 }else | |
4612 h->slice_type_fixed=0; | |
2967 | 4613 |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4614 slice_type= slice_type_map[ slice_type ]; |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4615 if (slice_type == I_TYPE |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4616 || (h->slice_num != 0 && slice_type == h->slice_type) ) { |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4617 default_ref_list_done = 1; |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4618 } |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4619 h->slice_type= slice_type; |
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4620 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
4621 s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though |
2967 | 4622 |
1168 | 4623 pps_id= get_ue_golomb(&s->gb); |
4624 if(pps_id>255){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4625 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n"); |
1168 | 4626 return -1; |
4627 } | |
4628 h->pps= h->pps_buffer[pps_id]; | |
1174 | 4629 if(h->pps.slice_group_count == 0){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4630 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n"); |
1174 | 4631 return -1; |
4632 } | |
4633 | |
1168 | 4634 h->sps= h->sps_buffer[ h->pps.sps_id ]; |
1174 | 4635 if(h->sps.log2_max_frame_num == 0){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4636 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n"); |
1174 | 4637 return -1; |
4638 } | |
2919 | 4639 |
4640 if(h->dequant_coeff_pps != pps_id){ | |
4641 h->dequant_coeff_pps = pps_id; | |
4642 init_dequant_tables(h); | |
4643 } | |
2967 | 4644 |
1168 | 4645 s->mb_width= h->sps.mb_width; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4646 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag); |
2967 | 4647 |
3178 | 4648 h->b_stride= s->mb_width*4; |
4649 h->b8_stride= s->mb_width*2; | |
1168 | 4650 |
1371 | 4651 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right ); |
1168 | 4652 if(h->sps.frame_mbs_only_flag) |
1371 | 4653 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom); |
1168 | 4654 else |
1371 | 4655 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck |
2967 | 4656 |
4657 if (s->context_initialized | |
1548 | 4658 && ( s->width != s->avctx->width || s->height != s->avctx->height)) { |
1168 | 4659 free_tables(h); |
4660 MPV_common_end(s); | |
4661 } | |
4662 if (!s->context_initialized) { | |
4663 if (MPV_common_init(s) < 0) | |
4664 return -1; | |
2967 | 4665 |
2758
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4666 if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4667 memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t)); |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4668 memcpy(h-> field_scan, field_scan, 16*sizeof(uint8_t)); |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4669 }else{ |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4670 int i; |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4671 for(i=0; i<16; i++){ |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4672 #define T(x) (x>>2) | ((x<<2) & 0xF) |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4673 h->zigzag_scan[i] = T(zigzag_scan[i]); |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4674 h-> field_scan[i] = T( field_scan[i]); |
3174 | 4675 #undef T |
4676 } | |
4677 } | |
4678 if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){ | |
3316 | 4679 memcpy(h->zigzag_scan8x8, zigzag_scan8x8, 64*sizeof(uint8_t)); |
3174 | 4680 memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t)); |
3316 | 4681 memcpy(h->field_scan8x8, field_scan8x8, 64*sizeof(uint8_t)); |
4682 memcpy(h->field_scan8x8_cavlc, field_scan8x8_cavlc, 64*sizeof(uint8_t)); | |
3174 | 4683 }else{ |
4684 int i; | |
4685 for(i=0; i<64; i++){ | |
4686 #define T(x) (x>>3) | ((x&7)<<3) | |
3316 | 4687 h->zigzag_scan8x8[i] = T(zigzag_scan8x8[i]); |
3174 | 4688 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]); |
3316 | 4689 h->field_scan8x8[i] = T(field_scan8x8[i]); |
4690 h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]); | |
3174 | 4691 #undef T |
2758
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4692 } |
a7925aaeea1e
10l (dont read variables before they have been set)
michael
parents:
2757
diff
changeset
|
4693 } |
2763 | 4694 if(h->sps.transform_bypass){ //FIXME same ugly |
3316 | 4695 h->zigzag_scan_q0 = zigzag_scan; |
4696 h->zigzag_scan8x8_q0 = zigzag_scan8x8; | |
3174 | 4697 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc; |
3316 | 4698 h->field_scan_q0 = field_scan; |
4699 h->field_scan8x8_q0 = field_scan8x8; | |
4700 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc; | |
2763 | 4701 }else{ |
3316 | 4702 h->zigzag_scan_q0 = h->zigzag_scan; |
4703 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8; | |
3174 | 4704 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc; |
3316 | 4705 h->field_scan_q0 = h->field_scan; |
4706 h->field_scan8x8_q0 = h->field_scan8x8; | |
4707 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc; | |
2763 | 4708 } |
1168 | 4709 |
4710 alloc_tables(h); | |
4711 | |
4712 s->avctx->width = s->width; | |
4713 s->avctx->height = s->height; | |
1548 | 4714 s->avctx->sample_aspect_ratio= h->sps.sar; |
2440 | 4715 if(!s->avctx->sample_aspect_ratio.den) |
4716 s->avctx->sample_aspect_ratio.den = 1; | |
2174
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
4717 |
2648 | 4718 if(h->sps.timing_info_present_flag){ |
3052 | 4719 s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale}; |
3063
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4720 if(h->x264_build > 0 && h->x264_build < 44) |
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4721 s->avctx->time_base.den *= 2; |
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4722 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den, |
f02d0b59279c
Remove all stray tabs and trailing whitespace, this time for good.
diego
parents:
3052
diff
changeset
|
4723 s->avctx->time_base.num, s->avctx->time_base.den, 1<<30); |
2174
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
4724 } |
1168 | 4725 } |
4726 | |
2392 | 4727 if(h->slice_num == 0){ |
2935 | 4728 if(frame_start(h) < 0) |
4729 return -1; | |
1168 | 4730 } |
4731 | |
1169 | 4732 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup |
1168 | 4733 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num); |
4734 | |
3316 | 4735 h->mb_mbaff = 0; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4736 h->mb_aff_frame = 0; |
1168 | 4737 if(h->sps.frame_mbs_only_flag){ |
4738 s->picture_structure= PICT_FRAME; | |
4739 }else{ | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4740 if(get_bits1(&s->gb)) { //field_pic_flag |
1168 | 4741 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag |
3316 | 4742 av_log(h->s.avctx, AV_LOG_ERROR, "PAFF interlacing is not implemented\n"); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4743 } else { |
1168 | 4744 s->picture_structure= PICT_FRAME; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4745 h->mb_aff_frame = h->sps.mb_aff; |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4746 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4747 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4748 |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4749 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width; |
3316 | 4750 s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << h->mb_aff_frame; |
2772 | 4751 if(s->mb_y >= s->mb_height){ |
4752 return -1; | |
4753 } | |
2967 | 4754 |
1168 | 4755 if(s->picture_structure==PICT_FRAME){ |
4756 h->curr_pic_num= h->frame_num; | |
4757 h->max_pic_num= 1<< h->sps.log2_max_frame_num; | |
4758 }else{ | |
4759 h->curr_pic_num= 2*h->frame_num; | |
4760 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1); | |
4761 } | |
2967 | 4762 |
1168 | 4763 if(h->nal_unit_type == NAL_IDR_SLICE){ |
1453 | 4764 get_ue_golomb(&s->gb); /* idr_pic_id */ |
1168 | 4765 } |
2967 | 4766 |
1168 | 4767 if(h->sps.poc_type==0){ |
4768 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb); | |
2967 | 4769 |
1168 | 4770 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){ |
4771 h->delta_poc_bottom= get_se_golomb(&s->gb); | |
4772 } | |
4773 } | |
2967 | 4774 |
1168 | 4775 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){ |
4776 h->delta_poc[0]= get_se_golomb(&s->gb); | |
2967 | 4777 |
1168 | 4778 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME) |
4779 h->delta_poc[1]= get_se_golomb(&s->gb); | |
4780 } | |
2967 | 4781 |
1168 | 4782 init_poc(h); |
2967 | 4783 |
1168 | 4784 if(h->pps.redundant_pic_cnt_present){ |
4785 h->redundant_pic_count= get_ue_golomb(&s->gb); | |
4786 } | |
4787 | |
4788 //set defaults, might be overriden a few line later | |
4789 h->ref_count[0]= h->pps.ref_count[0]; | |
4790 h->ref_count[1]= h->pps.ref_count[1]; | |
4791 | |
4792 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){ | |
4793 if(h->slice_type == B_TYPE){ | |
4794 h->direct_spatial_mv_pred= get_bits1(&s->gb); | |
3316 | 4795 if(h->sps.mb_aff && h->direct_spatial_mv_pred) |
4796 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + spatial direct mode is not implemented\n"); | |
1168 | 4797 } |
4798 num_ref_idx_active_override_flag= get_bits1(&s->gb); | |
2967 | 4799 |
1168 | 4800 if(num_ref_idx_active_override_flag){ |
4801 h->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
4802 if(h->slice_type==B_TYPE) | |
4803 h->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
4804 | |
4805 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
4806 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n"); |
1168 | 4807 return -1; |
4808 } | |
4809 } | |
4810 } | |
4811 | |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
4812 if(!default_ref_list_done){ |
1168 | 4813 fill_default_ref_list(h); |
4814 } | |
4815 | |
2766
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4816 if(decode_ref_pic_list_reordering(h) < 0) |
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
4817 return -1; |
1168 | 4818 |
2967 | 4819 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) |
1168 | 4820 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) ) |
4821 pred_weight_table(h); | |
2415 | 4822 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE) |
4823 implicit_weight_table(h); | |
4824 else | |
4825 h->use_weight = 0; | |
2967 | 4826 |
2580
b4f6f89ec2f6
The cvs version 1.103 of h264.c brokes 13 conformance streams, this
michael
parents:
2561
diff
changeset
|
4827 if(s->current_picture.reference) |
1168 | 4828 decode_ref_pic_marking(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4829 |
3316 | 4830 if(FRAME_MBAFF) |
4831 fill_mbaff_ref_list(h); | |
4832 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4833 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4834 h->cabac_init_idc = get_ue_golomb(&s->gb); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4835 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4836 h->last_qscale_diff = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
4837 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); |
1898 | 4838 if(s->qscale<0 || s->qscale>51){ |
4839 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale); | |
4840 return -1; | |
4841 } | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
4842 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale); |
1168 | 4843 //FIXME qscale / qp ... stuff |
4844 if(h->slice_type == SP_TYPE){ | |
1453 | 4845 get_bits1(&s->gb); /* sp_for_switch_flag */ |
1168 | 4846 } |
4847 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){ | |
1453 | 4848 get_se_golomb(&s->gb); /* slice_qs_delta */ |
1168 | 4849 } |
4850 | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4851 h->deblocking_filter = 1; |
1898 | 4852 h->slice_alpha_c0_offset = 0; |
4853 h->slice_beta_offset = 0; | |
1168 | 4854 if( h->pps.deblocking_filter_parameters_present ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4855 h->deblocking_filter= get_ue_golomb(&s->gb); |
2967 | 4856 if(h->deblocking_filter < 2) |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4857 h->deblocking_filter^= 1; // 1<->0 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4858 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
4859 if( h->deblocking_filter ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4860 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4861 h->slice_beta_offset = get_se_golomb(&s->gb) << 1; |
1168 | 4862 } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
4863 } |
2792 | 4864 if( s->avctx->skip_loop_filter >= AVDISCARD_ALL |
4865 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type != I_TYPE) | |
4866 ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR && h->slice_type == B_TYPE) | |
4867 ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0)) | |
4868 h->deblocking_filter= 0; | |
1168 | 4869 |
4870 #if 0 //FMO | |
4871 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5) | |
4872 slice_group_change_cycle= get_bits(&s->gb, ?); | |
4873 #endif | |
4874 | |
2392 | 4875 h->slice_num++; |
4876 | |
3316 | 4877 h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; |
4878 h->emu_edge_height= FRAME_MBAFF ? 0 : h->emu_edge_width; | |
4879 | |
1168 | 4880 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ |
2967 | 4881 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s\n", |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4882 h->slice_num, |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
4883 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"), |
2967 | 4884 first_mb_in_slice, |
1264 | 4885 av_get_pict_type_char(h->slice_type), |
1168 | 4886 pps_id, h->frame_num, |
4887 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1], | |
4888 h->ref_count[0], h->ref_count[1], | |
4889 s->qscale, | |
2583 | 4890 h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2, |
2415 | 4891 h->use_weight, |
4892 h->use_weight==1 && h->use_weight_chroma ? "c" : "" | |
1168 | 4893 ); |
4894 } | |
4895 | |
3807
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4896 if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !s->current_picture.reference){ |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4897 s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4898 s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4899 }else{ |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4900 s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4901 s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab; |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4902 } |
6a40092eb9e6
approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents:
3776
diff
changeset
|
4903 |
1168 | 4904 return 0; |
4905 } | |
4906 | |
4907 /** | |
4908 * | |
4909 */ | |
4910 static inline int get_level_prefix(GetBitContext *gb){ | |
4911 unsigned int buf; | |
4912 int log; | |
2967 | 4913 |
1168 | 4914 OPEN_READER(re, gb); |
4915 UPDATE_CACHE(re, gb); | |
4916 buf=GET_CACHE(re, gb); | |
2967 | 4917 |
1168 | 4918 log= 32 - av_log2(buf); |
4919 #ifdef TRACE | |
4920 print_bin(buf>>(32-log), log); | |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2255
diff
changeset
|
4921 av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__); |
1168 | 4922 #endif |
4923 | |
4924 LAST_SKIP_BITS(re, gb, log); | |
4925 CLOSE_READER(re, gb); | |
4926 | |
4927 return log-1; | |
4928 } | |
4929 | |
2755 | 4930 static inline int get_dct8x8_allowed(H264Context *h){ |
4931 int i; | |
4932 for(i=0; i<4; i++){ | |
4933 if(!IS_SUB_8X8(h->sub_mb_type[i]) | |
2781
899a2ea0907c
parenthesing to avoid compiler errors in the future - compared asm outputs, behaviour didnt changed
alex
parents:
2774
diff
changeset
|
4934 || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i]))) |
2755 | 4935 return 0; |
4936 } | |
4937 return 1; | |
4938 } | |
4939 | |
1168 | 4940 /** |
4941 * decodes a residual block. | |
4942 * @param n block index | |
4943 * @param scantable scantable | |
4944 * @param max_coeff number of coefficients in the block | |
4945 * @return <0 if an error occured | |
4946 */ | |
2919 | 4947 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){ |
1168 | 4948 MpegEncContext * const s = &h->s; |
4949 static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3}; | |
2897 | 4950 int level[16]; |
4951 int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before; | |
1168 | 4952 |
4953 //FIXME put trailing_onex into the context | |
4954 | |
4955 if(n == CHROMA_DC_BLOCK_INDEX){ | |
4956 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1); | |
4957 total_coeff= coeff_token>>2; | |
2967 | 4958 }else{ |
1168 | 4959 if(n == LUMA_DC_BLOCK_INDEX){ |
4960 total_coeff= pred_non_zero_count(h, 0); | |
4961 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
4962 total_coeff= coeff_token>>2; | |
4963 }else{ | |
4964 total_coeff= pred_non_zero_count(h, n); | |
4965 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2); | |
4966 total_coeff= coeff_token>>2; | |
4967 h->non_zero_count_cache[ scan8[n] ]= total_coeff; | |
4968 } | |
4969 } | |
4970 | |
4971 //FIXME set last_non_zero? | |
4972 | |
4973 if(total_coeff==0) | |
4974 return 0; | |
2967 | 4975 |
1168 | 4976 trailing_ones= coeff_token&3; |
1170 | 4977 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff); |
1168 | 4978 assert(total_coeff<=16); |
2967 | 4979 |
1168 | 4980 for(i=0; i<trailing_ones; i++){ |
4981 level[i]= 1 - 2*get_bits1(gb); | |
4982 } | |
4983 | |
2897 | 4984 if(i<total_coeff) { |
1168 | 4985 int level_code, mask; |
2897 | 4986 int suffix_length = total_coeff > 10 && trailing_ones < 3; |
4987 int prefix= get_level_prefix(gb); | |
4988 | |
4989 //first coefficient has suffix_length equal to 0 or 1 | |
1168 | 4990 if(prefix<14){ //FIXME try to build a large unified VLC table for all this |
4991 if(suffix_length) | |
4992 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
4993 else | |
4994 level_code= (prefix<<suffix_length); //part | |
4995 }else if(prefix==14){ | |
4996 if(suffix_length) | |
4997 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part | |
4998 else | |
4999 level_code= prefix + get_bits(gb, 4); //part | |
5000 }else if(prefix==15){ | |
5001 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5002 if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense |
1168 | 5003 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
5004 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y); |
1168 | 5005 return -1; |
5006 } | |
5007 | |
2897 | 5008 if(trailing_ones < 3) level_code += 2; |
5009 | |
5010 suffix_length = 1; | |
5011 if(level_code > 5) | |
5012 suffix_length++; | |
1168 | 5013 mask= -(level_code&1); |
5014 level[i]= (((2+level_code)>>1) ^ mask) - mask; | |
2897 | 5015 i++; |
5016 | |
5017 //remaining coefficients have suffix_length > 0 | |
5018 for(;i<total_coeff;i++) { | |
5019 static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX }; | |
5020 prefix = get_level_prefix(gb); | |
5021 if(prefix<15){ | |
5022 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length); | |
5023 }else if(prefix==15){ | |
5024 level_code = (prefix<<suffix_length) + get_bits(gb, 12); | |
5025 }else{ | |
5026 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y); | |
5027 return -1; | |
5028 } | |
5029 mask= -(level_code&1); | |
5030 level[i]= (((2+level_code)>>1) ^ mask) - mask; | |
5031 if(level_code > suffix_limit[suffix_length]) | |
5032 suffix_length++; | |
5033 } | |
1168 | 5034 } |
5035 | |
5036 if(total_coeff == max_coeff) | |
5037 zeros_left=0; | |
5038 else{ | |
5039 if(n == CHROMA_DC_BLOCK_INDEX) | |
5040 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1); | |
5041 else | |
5042 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1); | |
5043 } | |
2897 | 5044 |
5045 coeff_num = zeros_left + total_coeff - 1; | |
5046 j = scantable[coeff_num]; | |
5047 if(n > 24){ | |
5048 block[j] = level[0]; | |
5049 for(i=1;i<total_coeff;i++) { | |
5050 if(zeros_left <= 0) | |
5051 run_before = 0; | |
5052 else if(zeros_left < 7){ | |
5053 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1); | |
5054 }else{ | |
5055 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); | |
5056 } | |
5057 zeros_left -= run_before; | |
5058 coeff_num -= 1 + run_before; | |
5059 j= scantable[ coeff_num ]; | |
5060 | |
5061 block[j]= level[i]; | |
5062 } | |
5063 }else{ | |
2919 | 5064 block[j] = (level[0] * qmul[j] + 32)>>6; |
2897 | 5065 for(i=1;i<total_coeff;i++) { |
5066 if(zeros_left <= 0) | |
5067 run_before = 0; | |
5068 else if(zeros_left < 7){ | |
5069 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1); | |
5070 }else{ | |
5071 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2); | |
5072 } | |
5073 zeros_left -= run_before; | |
5074 coeff_num -= 1 + run_before; | |
5075 j= scantable[ coeff_num ]; | |
5076 | |
2919 | 5077 block[j]= (level[i] * qmul[j] + 32)>>6; |
2897 | 5078 } |
1168 | 5079 } |
5080 | |
5081 if(zeros_left<0){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
5082 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y); |
1168 | 5083 return -1; |
5084 } | |
2897 | 5085 |
1168 | 5086 return 0; |
5087 } | |
5088 | |
3316 | 5089 static void predict_field_decoding_flag(H264Context *h){ |
5090 MpegEncContext * const s = &h->s; | |
5091 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; | |
5092 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num) | |
5093 ? s->current_picture.mb_type[mb_xy-1] | |
5094 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num) | |
5095 ? s->current_picture.mb_type[mb_xy-s->mb_stride] | |
5096 : 0; | |
5097 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0; | |
5098 } | |
5099 | |
1168 | 5100 /** |
2396 | 5101 * decodes a P_SKIP or B_SKIP macroblock |
5102 */ | |
5103 static void decode_mb_skip(H264Context *h){ | |
5104 MpegEncContext * const s = &h->s; | |
5105 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; | |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
5106 int mb_type=0; |
2967 | 5107 |
2396 | 5108 memset(h->non_zero_count[mb_xy], 0, 16); |
5109 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui | |
5110 | |
3316 | 5111 if(MB_FIELD) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
5112 mb_type|= MB_TYPE_INTERLACED; |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
5113 |
2396 | 5114 if( h->slice_type == B_TYPE ) |
5115 { | |
5116 // just for fill_caches. pred_direct_motion will set the real mb_type | |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
5117 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP; |
2396 | 5118 |
2449 | 5119 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ... |
2396 | 5120 pred_direct_motion(h, &mb_type); |
3927 | 5121 mb_type|= MB_TYPE_SKIP; |
2396 | 5122 } |
5123 else | |
5124 { | |
5125 int mx, my; | |
2674
58fb8e1efcae
10l (mb_type uninitalized and then changed and overwritten)
michael
parents:
2673
diff
changeset
|
5126 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP; |
2396 | 5127 |
2449 | 5128 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ... |
2396 | 5129 pred_pskip_motion(h, &mx, &my); |
5130 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); | |
5131 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4); | |
5132 } | |
5133 | |
5134 write_back_motion(h, mb_type); | |
3927 | 5135 s->current_picture.mb_type[mb_xy]= mb_type; |
2396 | 5136 s->current_picture.qscale_table[mb_xy]= s->qscale; |
5137 h->slice_table[ mb_xy ]= h->slice_num; | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
5138 h->prev_mb_skipped= 1; |
2396 | 5139 } |
5140 | |
5141 /** | |
1168 | 5142 * decodes a macroblock |
5143 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed | |
5144 */ | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5145 static int decode_mb_cavlc(H264Context *h){ |
1168 | 5146 MpegEncContext * const s = &h->s; |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
5147 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1169 | 5148 int mb_type, partition_count, cbp; |
2755 | 5149 int dct8x8_allowed= h->pps.transform_8x8_mode; |
1168 | 5150 |
2967 | 5151 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong? |
1168 | 5152 |
1170 | 5153 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y); |
1453 | 5154 cbp = 0; /* avoid warning. FIXME: find a solution without slowing |
5155 down the code */ | |
1168 | 5156 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){ |
5157 if(s->mb_skip_run==-1) | |
5158 s->mb_skip_run= get_ue_golomb(&s->gb); | |
2967 | 5159 |
1168 | 5160 if (s->mb_skip_run--) { |
3316 | 5161 if(FRAME_MBAFF && (s->mb_y&1) == 0){ |
5162 if(s->mb_skip_run==0) | |
5163 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb); | |
5164 else | |
5165 predict_field_decoding_flag(h); | |
5166 } | |
2396 | 5167 decode_mb_skip(h); |
1168 | 5168 return 0; |
5169 } | |
5170 } | |
3316 | 5171 if(FRAME_MBAFF){ |
5172 if( (s->mb_y&1) == 0 ) | |
5173 h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb); | |
1168 | 5174 }else |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
5175 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME); |
2967 | 5176 |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
5177 h->prev_mb_skipped= 0; |
2967 | 5178 |
1168 | 5179 mb_type= get_ue_golomb(&s->gb); |
5180 if(h->slice_type == B_TYPE){ | |
5181 if(mb_type < 23){ | |
5182 partition_count= b_mb_type_info[mb_type].partition_count; | |
5183 mb_type= b_mb_type_info[mb_type].type; | |
5184 }else{ | |
5185 mb_type -= 23; | |
5186 goto decode_intra_mb; | |
5187 } | |
5188 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){ | |
5189 if(mb_type < 5){ | |
5190 partition_count= p_mb_type_info[mb_type].partition_count; | |
5191 mb_type= p_mb_type_info[mb_type].type; | |
5192 }else{ | |
5193 mb_type -= 5; | |
5194 goto decode_intra_mb; | |
5195 } | |
5196 }else{ | |
5197 assert(h->slice_type == I_TYPE); | |
5198 decode_intra_mb: | |
5199 if(mb_type > 25){ | |
3954 | 5200 av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice too large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y); |
1168 | 5201 return -1; |
5202 } | |
5203 partition_count=0; | |
5204 cbp= i_mb_type_info[mb_type].cbp; | |
5205 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode; | |
5206 mb_type= i_mb_type_info[mb_type].type; | |
5207 } | |
5208 | |
3316 | 5209 if(MB_FIELD) |
1168 | 5210 mb_type |= MB_TYPE_INTERLACED; |
5211 | |
5212 h->slice_table[ mb_xy ]= h->slice_num; | |
2967 | 5213 |
1168 | 5214 if(IS_INTRA_PCM(mb_type)){ |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5215 unsigned int x, y; |
2967 | 5216 |
1168 | 5217 // we assume these blocks are very rare so we dont optimize it |
5218 align_get_bits(&s->gb); | |
2967 | 5219 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5220 // The pixels are stored in the same order as levels in h->mb array. |
1168 | 5221 for(y=0; y<16; y++){ |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5222 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3); |
1168 | 5223 for(x=0; x<16; x++){ |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5224 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8)); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5225 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8); |
1168 | 5226 } |
5227 } | |
5228 for(y=0; y<8; y++){ | |
5229 const int index= 256 + 4*(y&3) + 32*(y>>2); | |
5230 for(x=0; x<8; x++){ | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5231 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8)); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5232 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8); |
1168 | 5233 } |
5234 } | |
5235 for(y=0; y<8; y++){ | |
5236 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2); | |
5237 for(x=0; x<8; x++){ | |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5238 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8)); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5239 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8); |
1168 | 5240 } |
5241 } | |
2967 | 5242 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5243 // In deblocking, the quantizer is 0 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
5244 s->current_picture.qscale_table[mb_xy]= 0; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
5245 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5246 // All coeffs are present |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5247 memset(h->non_zero_count[mb_xy], 16, 16); |
2967 | 5248 |
2755 | 5249 s->current_picture.mb_type[mb_xy]= mb_type; |
1168 | 5250 return 0; |
5251 } | |
2967 | 5252 |
3316 | 5253 if(MB_MBAFF){ |
5254 h->ref_count[0] <<= 1; | |
5255 h->ref_count[1] <<= 1; | |
5256 } | |
5257 | |
2449 | 5258 fill_caches(h, mb_type, 0); |
1168 | 5259 |
5260 //mb_pred | |
5261 if(IS_INTRA(mb_type)){ | |
5262 // init_top_left_availability(h); | |
5263 if(IS_INTRA4x4(mb_type)){ | |
5264 int i; | |
2755 | 5265 int di = 1; |
5266 if(dct8x8_allowed && get_bits1(&s->gb)){ | |
5267 mb_type |= MB_TYPE_8x8DCT; | |
5268 di = 4; | |
5269 } | |
1168 | 5270 |
5271 // fill_intra4x4_pred_table(h); | |
2755 | 5272 for(i=0; i<16; i+=di){ |
3409 | 5273 int mode= pred_intra_mode(h, i); |
5274 | |
5275 if(!get_bits1(&s->gb)){ | |
1168 | 5276 const int rem_mode= get_bits(&s->gb, 3); |
3409 | 5277 mode = rem_mode + (rem_mode >= mode); |
1168 | 5278 } |
2967 | 5279 |
2755 | 5280 if(di==4) |
5281 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 ); | |
5282 else | |
5283 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode; | |
1168 | 5284 } |
5285 write_back_intra_pred_mode(h); | |
5286 if( check_intra4x4_pred_mode(h) < 0) | |
5287 return -1; | |
5288 }else{ | |
5289 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode); | |
5290 if(h->intra16x16_pred_mode < 0) | |
5291 return -1; | |
5292 } | |
5293 h->chroma_pred_mode= get_ue_golomb(&s->gb); | |
5294 | |
5295 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode); | |
5296 if(h->chroma_pred_mode < 0) | |
5297 return -1; | |
5298 }else if(partition_count==4){ | |
5299 int i, j, sub_partition_count[4], list, ref[2][4]; | |
2967 | 5300 |
1168 | 5301 if(h->slice_type == B_TYPE){ |
5302 for(i=0; i<4; i++){ | |
5303 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
5304 if(h->sub_mb_type[i] >=13){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
5305 av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); |
1168 | 5306 return -1; |
5307 } | |
5308 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
5309 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
5310 } | |
2396 | 5311 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1]) |
3003 | 5312 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) { |
2396 | 5313 pred_direct_motion(h, &mb_type); |
3003 | 5314 h->ref_cache[0][scan8[4]] = |
5315 h->ref_cache[1][scan8[4]] = | |
5316 h->ref_cache[0][scan8[12]] = | |
5317 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE; | |
5318 } | |
1168 | 5319 }else{ |
5320 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ? | |
5321 for(i=0; i<4; i++){ | |
5322 h->sub_mb_type[i]= get_ue_golomb(&s->gb); | |
5323 if(h->sub_mb_type[i] >=4){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
5324 av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y); |
1168 | 5325 return -1; |
5326 } | |
5327 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; | |
5328 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type; | |
5329 } | |
5330 } | |
2967 | 5331 |
1168 | 5332 for(list=0; list<2; list++){ |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5333 int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; |
1168 | 5334 if(ref_count == 0) continue; |
5335 for(i=0; i<4; i++){ | |
2396 | 5336 if(IS_DIRECT(h->sub_mb_type[i])) continue; |
5337 if(IS_DIR(h->sub_mb_type[i], 0, list)){ | |
1168 | 5338 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip? |
5339 }else{ | |
5340 //FIXME | |
5341 ref[list][i] = -1; | |
5342 } | |
5343 } | |
5344 } | |
2967 | 5345 |
2755 | 5346 if(dct8x8_allowed) |
5347 dct8x8_allowed = get_dct8x8_allowed(h); | |
2967 | 5348 |
1168 | 5349 for(list=0; list<2; list++){ |
5350 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list]; | |
5351 if(ref_count == 0) continue; | |
5352 | |
5353 for(i=0; i<4; i++){ | |
3003 | 5354 if(IS_DIRECT(h->sub_mb_type[i])) { |
5355 h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ]; | |
5356 continue; | |
5357 } | |
1168 | 5358 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]= |
5359 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i]; | |
5360 | |
2396 | 5361 if(IS_DIR(h->sub_mb_type[i], 0, list)){ |
1168 | 5362 const int sub_mb_type= h->sub_mb_type[i]; |
5363 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1; | |
5364 for(j=0; j<sub_partition_count[i]; j++){ | |
5365 int mx, my; | |
5366 const int index= 4*i + block_width*j; | |
5367 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; | |
5368 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my); | |
5369 mx += get_se_golomb(&s->gb); | |
5370 my += get_se_golomb(&s->gb); | |
1170 | 5371 tprintf("final mv:%d %d\n", mx, my); |
5372 | |
1168 | 5373 if(IS_SUB_8X8(sub_mb_type)){ |
2967 | 5374 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= |
1168 | 5375 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx; |
2967 | 5376 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= |
1168 | 5377 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my; |
5378 }else if(IS_SUB_8X4(sub_mb_type)){ | |
5379 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx; | |
5380 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my; | |
5381 }else if(IS_SUB_4X8(sub_mb_type)){ | |
5382 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx; | |
5383 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my; | |
5384 }else{ | |
5385 assert(IS_SUB_4X4(sub_mb_type)); | |
5386 mv_cache[ 0 ][0]= mx; | |
5387 mv_cache[ 0 ][1]= my; | |
5388 } | |
5389 } | |
5390 }else{ | |
5391 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0]; | |
5392 p[0] = p[1]= | |
5393 p[8] = p[9]= 0; | |
5394 } | |
5395 } | |
5396 } | |
2396 | 5397 }else if(IS_DIRECT(mb_type)){ |
5398 pred_direct_motion(h, &mb_type); | |
2755 | 5399 dct8x8_allowed &= h->sps.direct_8x8_inference_flag; |
2396 | 5400 }else{ |
1168 | 5401 int list, mx, my, i; |
5402 //FIXME we should set ref_idx_l? to 0 if we use that later ... | |
5403 if(IS_16X16(mb_type)){ | |
5404 for(list=0; list<2; list++){ | |
2392 | 5405 if(h->ref_count[list]>0){ |
1168 | 5406 if(IS_DIR(mb_type, 0, list)){ |
5407 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
5408 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1); | |
2523 | 5409 }else |
5410 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1); | |
1168 | 5411 } |
5412 } | |
5413 for(list=0; list<2; list++){ | |
5414 if(IS_DIR(mb_type, 0, list)){ | |
5415 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my); | |
5416 mx += get_se_golomb(&s->gb); | |
5417 my += get_se_golomb(&s->gb); | |
1170 | 5418 tprintf("final mv:%d %d\n", mx, my); |
5419 | |
1269 | 5420 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4); |
2523 | 5421 }else |
5422 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4); | |
1168 | 5423 } |
5424 } | |
5425 else if(IS_16X8(mb_type)){ | |
5426 for(list=0; list<2; list++){ | |
5427 if(h->ref_count[list]>0){ | |
5428 for(i=0; i<2; i++){ | |
5429 if(IS_DIR(mb_type, i, list)){ | |
5430 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
5431 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1); | |
2523 | 5432 }else |
2396 | 5433 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1); |
1168 | 5434 } |
5435 } | |
5436 } | |
5437 for(list=0; list<2; list++){ | |
5438 for(i=0; i<2; i++){ | |
5439 if(IS_DIR(mb_type, i, list)){ | |
5440 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my); | |
5441 mx += get_se_golomb(&s->gb); | |
5442 my += get_se_golomb(&s->gb); | |
1170 | 5443 tprintf("final mv:%d %d\n", mx, my); |
5444 | |
1269 | 5445 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4); |
2396 | 5446 }else |
5447 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4); | |
1168 | 5448 } |
5449 } | |
5450 }else{ | |
5451 assert(IS_8X16(mb_type)); | |
5452 for(list=0; list<2; list++){ | |
5453 if(h->ref_count[list]>0){ | |
5454 for(i=0; i<2; i++){ | |
5455 if(IS_DIR(mb_type, i, list)){ //FIXME optimize | |
5456 const int val= get_te0_golomb(&s->gb, h->ref_count[list]); | |
5457 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1); | |
2523 | 5458 }else |
2396 | 5459 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1); |
1168 | 5460 } |
5461 } | |
5462 } | |
5463 for(list=0; list<2; list++){ | |
5464 for(i=0; i<2; i++){ | |
5465 if(IS_DIR(mb_type, i, list)){ | |
5466 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my); | |
5467 mx += get_se_golomb(&s->gb); | |
5468 my += get_se_golomb(&s->gb); | |
1170 | 5469 tprintf("final mv:%d %d\n", mx, my); |
5470 | |
1269 | 5471 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4); |
2396 | 5472 }else |
5473 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4); | |
1168 | 5474 } |
5475 } | |
5476 } | |
5477 } | |
2967 | 5478 |
1168 | 5479 if(IS_INTER(mb_type)) |
5480 write_back_motion(h, mb_type); | |
2967 | 5481 |
1168 | 5482 if(!IS_INTRA16x16(mb_type)){ |
5483 cbp= get_ue_golomb(&s->gb); | |
5484 if(cbp > 47){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
5485 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y); |
1168 | 5486 return -1; |
5487 } | |
2967 | 5488 |
1168 | 5489 if(IS_INTRA4x4(mb_type)) |
5490 cbp= golomb_to_intra4x4_cbp[cbp]; | |
5491 else | |
5492 cbp= golomb_to_inter_cbp[cbp]; | |
5493 } | |
3651 | 5494 h->cbp = cbp; |
1168 | 5495 |
2755 | 5496 if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){ |
5497 if(get_bits1(&s->gb)) | |
5498 mb_type |= MB_TYPE_8x8DCT; | |
5499 } | |
5500 s->current_picture.mb_type[mb_xy]= mb_type; | |
5501 | |
1168 | 5502 if(cbp || IS_INTRA16x16(mb_type)){ |
5503 int i8x8, i4x4, chroma_idx; | |
5504 int chroma_qp, dquant; | |
5505 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr; | |
3174 | 5506 const uint8_t *scan, *scan8x8, *dc_scan; |
2967 | 5507 |
1168 | 5508 // fill_non_zero_count_cache(h); |
5509 | |
5510 if(IS_INTERLACED(mb_type)){ | |
3316 | 5511 scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0; |
2763 | 5512 scan= s->qscale ? h->field_scan : h->field_scan_q0; |
1168 | 5513 dc_scan= luma_dc_field_scan; |
5514 }else{ | |
3316 | 5515 scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0; |
2763 | 5516 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0; |
1168 | 5517 dc_scan= luma_dc_zigzag_scan; |
5518 } | |
5519 | |
5520 dquant= get_se_golomb(&s->gb); | |
5521 | |
5522 if( dquant > 25 || dquant < -26 ){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
5523 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y); |
1168 | 5524 return -1; |
5525 } | |
2967 | 5526 |
1168 | 5527 s->qscale += dquant; |
5528 if(((unsigned)s->qscale) > 51){ | |
5529 if(s->qscale<0) s->qscale+= 52; | |
5530 else s->qscale-= 52; | |
5531 } | |
2967 | 5532 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
5533 h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale); |
1168 | 5534 if(IS_INTRA16x16(mb_type)){ |
2919 | 5535 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, h->dequant4_coeff[0][s->qscale], 16) < 0){ |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
5536 return -1; //FIXME continue if partitioned and other return -1 too |
1168 | 5537 } |
5538 | |
5539 assert((cbp&15) == 0 || (cbp&15) == 15); | |
5540 | |
5541 if(cbp&15){ | |
5542 for(i8x8=0; i8x8<4; i8x8++){ | |
5543 for(i4x4=0; i4x4<4; i4x4++){ | |
5544 const int index= i4x4 + 4*i8x8; | |
2919 | 5545 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){ |
1168 | 5546 return -1; |
5547 } | |
5548 } | |
5549 } | |
5550 }else{ | |
1636 | 5551 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1); |
1168 | 5552 } |
5553 }else{ | |
5554 for(i8x8=0; i8x8<4; i8x8++){ | |
5555 if(cbp & (1<<i8x8)){ | |
2755 | 5556 if(IS_8x8DCT(mb_type)){ |
5557 DCTELEM *buf = &h->mb[64*i8x8]; | |
2757 | 5558 uint8_t *nnz; |
2755 | 5559 for(i4x4=0; i4x4<4; i4x4++){ |
3174 | 5560 if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4, |
2919 | 5561 h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 ) |
2755 | 5562 return -1; |
5563 } | |
2757 | 5564 nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
5565 nnz[0] += nnz[1] + nnz[8] + nnz[9]; |
2755 | 5566 }else{ |
5567 for(i4x4=0; i4x4<4; i4x4++){ | |
5568 const int index= i4x4 + 4*i8x8; | |
2967 | 5569 |
2919 | 5570 if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){ |
2755 | 5571 return -1; |
5572 } | |
1168 | 5573 } |
5574 } | |
5575 }else{ | |
5576 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; | |
5577 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0; | |
5578 } | |
5579 } | |
5580 } | |
2967 | 5581 |
1168 | 5582 if(cbp&0x30){ |
5583 for(chroma_idx=0; chroma_idx<2; chroma_idx++) | |
2919 | 5584 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, NULL, 4) < 0){ |
1168 | 5585 return -1; |
5586 } | |
5587 } | |
5588 | |
5589 if(cbp&0x20){ | |
5590 for(chroma_idx=0; chroma_idx<2; chroma_idx++){ | |
5591 for(i4x4=0; i4x4<4; i4x4++){ | |
5592 const int index= 16 + 4*chroma_idx + i4x4; | |
2919 | 5593 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][chroma_qp], 15) < 0){ |
1168 | 5594 return -1; |
5595 } | |
5596 } | |
5597 } | |
5598 }else{ | |
5599 uint8_t * const nnz= &h->non_zero_count_cache[0]; | |
5600 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = | |
5601 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; | |
5602 } | |
5603 }else{ | |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5604 uint8_t * const nnz= &h->non_zero_count_cache[0]; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5605 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5606 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
5607 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; |
1168 | 5608 } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
5609 s->current_picture.qscale_table[mb_xy]= s->qscale; |
1168 | 5610 write_back_non_zero_count(h); |
5611 | |
3316 | 5612 if(MB_MBAFF){ |
5613 h->ref_count[0] >>= 1; | |
5614 h->ref_count[1] >>= 1; | |
5615 } | |
5616 | |
1168 | 5617 return 0; |
5618 } | |
5619 | |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5620 static int decode_cabac_field_decoding_flag(H264Context *h) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5621 MpegEncContext * const s = &h->s; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5622 const int mb_x = s->mb_x; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5623 const int mb_y = s->mb_y & ~1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5624 const int mba_xy = mb_x - 1 + mb_y *s->mb_stride; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5625 const int mbb_xy = mb_x + (mb_y-2)*s->mb_stride; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5626 |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5627 unsigned int ctx = 0; |
2967 | 5628 |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5629 if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5630 ctx += 1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5631 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5632 if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) { |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5633 ctx += 1; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5634 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5635 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5636 return get_cabac_noinline( &h->cabac, &h->cabac_state[70 + ctx] ); |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5637 } |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5638 |
2312 | 5639 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) { |
5640 uint8_t *state= &h->cabac_state[ctx_base]; | |
5641 int mb_type; | |
2967 | 5642 |
2312 | 5643 if(intra_slice){ |
5644 MpegEncContext * const s = &h->s; | |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5645 const int mba_xy = h->left_mb_xy[0]; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5646 const int mbb_xy = h->top_mb_xy; |
2312 | 5647 int ctx=0; |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5648 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5649 ctx++; |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5650 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5651 ctx++; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5652 if( get_cabac_noinline( &h->cabac, &state[ctx] ) == 0 ) |
2312 | 5653 return 0; /* I4x4 */ |
5654 state += 2; | |
5655 }else{ | |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5656 if( get_cabac_noinline( &h->cabac, &state[0] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5657 return 0; /* I4x4 */ |
2312 | 5658 } |
5659 | |
5660 if( get_cabac_terminate( &h->cabac ) ) | |
5661 return 25; /* PCM */ | |
5662 | |
5663 mb_type = 1; /* I16x16 */ | |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5664 mb_type += 12 * get_cabac_noinline( &h->cabac, &state[1] ); /* cbp_luma != 0 */ |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5665 if( get_cabac_noinline( &h->cabac, &state[2] ) ) /* cbp_chroma */ |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5666 mb_type += 4 + 4 * get_cabac_noinline( &h->cabac, &state[2+intra_slice] ); |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5667 mb_type += 2 * get_cabac_noinline( &h->cabac, &state[3+intra_slice] ); |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5668 mb_type += 1 * get_cabac_noinline( &h->cabac, &state[3+2*intra_slice] ); |
2312 | 5669 return mb_type; |
5670 } | |
5671 | |
5672 static int decode_cabac_mb_type( H264Context *h ) { | |
5673 MpegEncContext * const s = &h->s; | |
5674 | |
5675 if( h->slice_type == I_TYPE ) { | |
5676 return decode_cabac_intra_mb_type(h, 3, 1); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5677 } else if( h->slice_type == P_TYPE ) { |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5678 if( get_cabac_noinline( &h->cabac, &h->cabac_state[14] ) == 0 ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5679 /* P-type */ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5680 if( get_cabac_noinline( &h->cabac, &h->cabac_state[15] ) == 0 ) { |
3138 | 5681 /* P_L0_D16x16, P_8x8 */ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5682 return 3 * get_cabac_noinline( &h->cabac, &h->cabac_state[16] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5683 } else { |
3138 | 5684 /* P_L0_D8x16, P_L0_D16x8 */ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5685 return 2 - get_cabac_noinline( &h->cabac, &h->cabac_state[17] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5686 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5687 } else { |
2312 | 5688 return decode_cabac_intra_mb_type(h, 17, 0) + 5; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5689 } |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5690 } else if( h->slice_type == B_TYPE ) { |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5691 const int mba_xy = h->left_mb_xy[0]; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
5692 const int mbb_xy = h->top_mb_xy; |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5693 int ctx = 0; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5694 int bits; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5695 |
3138 | 5696 if( h->slice_table[mba_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) ) |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5697 ctx++; |
3138 | 5698 if( h->slice_table[mbb_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) ) |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5699 ctx++; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5700 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5701 if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+ctx] ) ) |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5702 return 0; /* B_Direct_16x16 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5703 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5704 if( !get_cabac_noinline( &h->cabac, &h->cabac_state[27+3] ) ) { |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5705 return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */ |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5706 } |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5707 |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5708 bits = get_cabac_noinline( &h->cabac, &h->cabac_state[27+4] ) << 3; |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5709 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 2; |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5710 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ) << 1; |
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5711 bits|= get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5712 if( bits < 8 ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5713 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5714 else if( bits == 13 ) { |
2312 | 5715 return decode_cabac_intra_mb_type(h, 32, 0) + 23; |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5716 } else if( bits == 14 ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5717 return 11; /* B_L1_L0_8x16 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5718 else if( bits == 15 ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5719 return 22; /* B_8x8 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5720 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5721 bits= ( bits<<1 ) | get_cabac_noinline( &h->cabac, &h->cabac_state[27+5] ); |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5722 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */ |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5723 } else { |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5724 /* TODO SI/SP frames? */ |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5725 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5726 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5727 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5728 |
3316 | 5729 static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5730 MpegEncContext * const s = &h->s; |
3316 | 5731 int mba_xy, mbb_xy; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5732 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5733 |
3316 | 5734 if(FRAME_MBAFF){ //FIXME merge with the stuff in fill_caches? |
5735 int mb_xy = mb_x + (mb_y&~1)*s->mb_stride; | |
5736 mba_xy = mb_xy - 1; | |
5737 if( (mb_y&1) | |
5738 && h->slice_table[mba_xy] == h->slice_num | |
5739 && MB_FIELD == !!IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) | |
5740 mba_xy += s->mb_stride; | |
5741 if( MB_FIELD ){ | |
5742 mbb_xy = mb_xy - s->mb_stride; | |
5743 if( !(mb_y&1) | |
5744 && h->slice_table[mbb_xy] == h->slice_num | |
5745 && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) | |
5746 mbb_xy -= s->mb_stride; | |
5747 }else | |
5748 mbb_xy = mb_x + (mb_y-1)*s->mb_stride; | |
5749 }else{ | |
5750 int mb_xy = mb_x + mb_y*s->mb_stride; | |
5751 mba_xy = mb_xy - 1; | |
5752 mbb_xy = mb_xy - s->mb_stride; | |
5753 } | |
5754 | |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5755 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] )) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5756 ctx++; |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5757 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] )) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5758 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5759 |
3138 | 5760 if( h->slice_type == B_TYPE ) |
5761 ctx += 13; | |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5762 return get_cabac_noinline( &h->cabac, &h->cabac_state[11+ctx] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5763 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5764 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5765 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5766 int mode = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5767 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5768 if( get_cabac( &h->cabac, &h->cabac_state[68] ) ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5769 return pred_mode; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5770 |
2928 | 5771 mode += 1 * get_cabac( &h->cabac, &h->cabac_state[69] ); |
5772 mode += 2 * get_cabac( &h->cabac, &h->cabac_state[69] ); | |
5773 mode += 4 * get_cabac( &h->cabac, &h->cabac_state[69] ); | |
5774 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5775 if( mode >= pred_mode ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5776 return mode + 1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5777 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5778 return mode; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5779 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5780 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5781 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) { |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5782 const int mba_xy = h->left_mb_xy[0]; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5783 const int mbb_xy = h->top_mb_xy; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5784 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5785 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5786 |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5787 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */ |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5788 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5789 ctx++; |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5790 |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5791 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5792 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5793 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5794 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+ctx] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5795 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5796 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5797 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5798 return 1; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5799 if( get_cabac_noinline( &h->cabac, &h->cabac_state[64+3] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5800 return 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5801 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5802 return 3; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5803 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5804 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5805 static const uint8_t block_idx_x[16] = { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5806 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5807 }; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5808 static const uint8_t block_idx_y[16] = { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5809 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5810 }; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5811 static const uint8_t block_idx_xy[4][4] = { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5812 { 0, 2, 8, 10}, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5813 { 1, 3, 9, 11}, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5814 { 4, 6, 12, 14}, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5815 { 5, 7, 13, 15} |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5816 }; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5817 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5818 static int decode_cabac_mb_cbp_luma( H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5819 int cbp = 0; |
3138 | 5820 int cbp_b = -1; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5821 int i8x8; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5822 |
3138 | 5823 if( h->slice_table[h->top_mb_xy] == h->slice_num ) { |
5824 cbp_b = h->top_cbp; | |
5825 tprintf("cbp_b = top_cbp = %x\n", cbp_b); | |
5826 } | |
5827 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5828 for( i8x8 = 0; i8x8 < 4; i8x8++ ) { |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5829 int cbp_a = -1; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5830 int x, y; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5831 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5832 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5833 x = block_idx_x[4*i8x8]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5834 y = block_idx_y[4*i8x8]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5835 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5836 if( x > 0 ) |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5837 cbp_a = cbp; |
3138 | 5838 else if( h->slice_table[h->left_mb_xy[0]] == h->slice_num ) { |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5839 cbp_a = h->left_cbp; |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5840 tprintf("cbp_a = left_cbp = %x\n", cbp_a); |
2498
4d6d056a00c6
H.264 multiple slice support in CABAC patch by (Loic (lll+ffmpeg m4x org)
michael
parents:
2485
diff
changeset
|
5841 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5842 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5843 if( y > 0 ) |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5844 cbp_b = cbp; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5845 |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
5846 /* No need to test for skip as we put 0 for skip block */ |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
5847 /* No need to test for IPCM as we put 1 for IPCM block */ |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5848 if( cbp_a >= 0 ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5849 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4; |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5850 if( ((cbp_a >> i8x8a)&0x01) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5851 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5852 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5853 |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5854 if( cbp_b >= 0 ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5855 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4; |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
5856 if( ((cbp_b >> i8x8b)&0x01) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5857 ctx += 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5858 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5859 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5860 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5861 cbp |= 1 << i8x8; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5862 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5863 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5864 return cbp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5865 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5866 static int decode_cabac_mb_cbp_chroma( H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5867 int ctx; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5868 int cbp_a, cbp_b; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5869 |
2314 | 5870 cbp_a = (h->left_cbp>>4)&0x03; |
5871 cbp_b = (h-> top_cbp>>4)&0x03; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5872 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5873 ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5874 if( cbp_a > 0 ) ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5875 if( cbp_b > 0 ) ctx += 2; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5876 if( get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5877 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5878 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5879 ctx = 4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5880 if( cbp_a == 2 ) ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5881 if( cbp_b == 2 ) ctx += 2; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5882 return 1 + get_cabac_noinline( &h->cabac, &h->cabac_state[77 + ctx] ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5883 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5884 static int decode_cabac_mb_dqp( H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5885 MpegEncContext * const s = &h->s; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5886 int mbn_xy; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5887 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5888 int val = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5889 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5890 if( s->mb_x > 0 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5891 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5892 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5893 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5894 |
3316 | 5895 if( h->last_qscale_diff != 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5896 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5897 |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5898 while( get_cabac_noinline( &h->cabac, &h->cabac_state[60 + ctx] ) ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5899 if( ctx < 2 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5900 ctx = 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5901 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5902 ctx = 3; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5903 val++; |
3128 | 5904 if(val > 102) //prevent infinite loop |
2844
5f20ab245501
prevent assert failures and infinite loops with broken streams
michael
parents:
2834
diff
changeset
|
5905 return INT_MIN; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5906 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5907 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5908 if( val&0x01 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5909 return (val + 1)/2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5910 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5911 return -(val + 1)/2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
5912 } |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5913 static int decode_cabac_p_mb_sub_type( H264Context *h ) { |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5914 if( get_cabac( &h->cabac, &h->cabac_state[21] ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5915 return 0; /* 8x8 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5916 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5917 return 1; /* 8x4 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5918 if( get_cabac( &h->cabac, &h->cabac_state[23] ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5919 return 2; /* 4x8 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5920 return 3; /* 4x4 */ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5921 } |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5922 static int decode_cabac_b_mb_sub_type( H264Context *h ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5923 int type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5924 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5925 return 0; /* B_Direct_8x8 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5926 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) ) |
2311
cdbb2f30e08b
small typo patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
2310
diff
changeset
|
5927 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */ |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5928 type = 3; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5929 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5930 if( get_cabac( &h->cabac, &h->cabac_state[39] ) ) |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5931 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5932 type += 4; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5933 } |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5934 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5935 type += get_cabac( &h->cabac, &h->cabac_state[39] ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5936 return type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
5937 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5938 |
2755 | 5939 static inline int decode_cabac_mb_transform_size( H264Context *h ) { |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
5940 return get_cabac_noinline( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] ); |
2755 | 5941 } |
5942 | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5943 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5944 int refa = h->ref_cache[list][scan8[n] - 1]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5945 int refb = h->ref_cache[list][scan8[n] - 8]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5946 int ref = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5947 int ctx = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5948 |
2396 | 5949 if( h->slice_type == B_TYPE) { |
5950 if( refa > 0 && !h->direct_cache[scan8[n] - 1] ) | |
5951 ctx++; | |
5952 if( refb > 0 && !h->direct_cache[scan8[n] - 8] ) | |
5953 ctx += 2; | |
5954 } else { | |
5955 if( refa > 0 ) | |
5956 ctx++; | |
5957 if( refb > 0 ) | |
5958 ctx += 2; | |
5959 } | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5960 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5961 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5962 ref++; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5963 if( ctx < 4 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5964 ctx = 4; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5965 else |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5966 ctx = 5; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5967 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5968 return ref; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5969 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5970 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5971 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5972 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) + |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5973 abs( h->mvd_cache[list][scan8[n] - 8][l] ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5974 int ctxbase = (l == 0) ? 40 : 47; |
2317 | 5975 int ctx, mvd; |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5976 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5977 if( amvd < 3 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5978 ctx = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5979 else if( amvd > 32 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5980 ctx = 2; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5981 else |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5982 ctx = 1; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5983 |
2317 | 5984 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx])) |
5985 return 0; | |
5986 | |
5987 mvd= 1; | |
5988 ctx= 3; | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5989 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5990 mvd++; |
2317 | 5991 if( ctx < 6 ) |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5992 ctx++; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5993 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5994 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5995 if( mvd >= 9 ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5996 int k = 3; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5997 while( get_cabac_bypass( &h->cabac ) ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5998 mvd += 1 << k; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
5999 k++; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6000 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6001 while( k-- ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6002 if( get_cabac_bypass( &h->cabac ) ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6003 mvd += 1 << k; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6004 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6005 } |
4042 | 6006 return get_cabac_bypass_sign( &h->cabac, -mvd ); |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6007 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6008 |
2316 | 6009 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) { |
2314 | 6010 int nza, nzb; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6011 int ctx = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6012 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6013 if( cat == 0 ) { |
2314 | 6014 nza = h->left_cbp&0x100; |
6015 nzb = h-> top_cbp&0x100; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6016 } else if( cat == 1 || cat == 2 ) { |
2314 | 6017 nza = h->non_zero_count_cache[scan8[idx] - 1]; |
6018 nzb = h->non_zero_count_cache[scan8[idx] - 8]; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6019 } else if( cat == 3 ) { |
2314 | 6020 nza = (h->left_cbp>>(6+idx))&0x01; |
6021 nzb = (h-> top_cbp>>(6+idx))&0x01; | |
6022 } else { | |
6023 assert(cat == 4); | |
6024 nza = h->non_zero_count_cache[scan8[16+idx] - 1]; | |
6025 nzb = h->non_zero_count_cache[scan8[16+idx] - 8]; | |
6026 } | |
6027 | |
6028 if( nza > 0 ) | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6029 ctx++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6030 |
2314 | 6031 if( nzb > 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6032 ctx += 2; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6033 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6034 return ctx + 4 * cat; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6035 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6036 |
4051 | 6037 static const __attribute((used)) uint8_t last_coeff_flag_offset_8x8[63] = { |
6038 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
6039 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
6040 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, | |
6041 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8 | |
6042 }; | |
6043 | |
3028 | 6044 static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6045 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride; |
3316 | 6046 static const int significant_coeff_flag_offset[2][6] = { |
6047 { 105+0, 105+15, 105+29, 105+44, 105+47, 402 }, | |
6048 { 277+0, 277+15, 277+29, 277+44, 277+47, 436 } | |
6049 }; | |
6050 static const int last_coeff_flag_offset[2][6] = { | |
6051 { 166+0, 166+15, 166+29, 166+44, 166+47, 417 }, | |
6052 { 338+0, 338+15, 338+29, 338+44, 338+47, 451 } | |
6053 }; | |
6054 static const int coeff_abs_level_m1_offset[6] = { | |
6055 227+0, 227+10, 227+20, 227+30, 227+39, 426 | |
6056 }; | |
4007 | 6057 static const uint8_t significant_coeff_flag_offset_8x8[2][63] = { |
3316 | 6058 { 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5, |
2755 | 6059 4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7, |
6060 7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11, | |
3316 | 6061 12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 }, |
6062 { 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5, | |
6063 6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11, | |
6064 9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9, | |
6065 9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 } | |
2755 | 6066 }; |
6067 | |
6068 int index[64]; | |
2313 | 6069 |
4041
0113ceb0f953
1 variable less in decode_cabac_residual() no speed change (but less vars means less things gcc can load and store redundantly)
michael
parents:
4040
diff
changeset
|
6070 int last; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6071 int coeff_count = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6072 |
2316 | 6073 int abslevel1 = 1; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6074 int abslevelgt1 = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6075 |
2898 | 6076 uint8_t *significant_coeff_ctx_base; |
6077 uint8_t *last_coeff_ctx_base; | |
6078 uint8_t *abs_level_m1_ctx_base; | |
2755 | 6079 |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6080 #ifndef ARCH_X86 |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6081 #define CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6082 #endif |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6083 #ifdef CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6084 #define CC &cc |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6085 CABACContext cc; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6086 cc.range = h->cabac.range; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6087 cc.low = h->cabac.low; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6088 cc.bytestream= h->cabac.bytestream; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6089 #else |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6090 #define CC &h->cabac |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6091 #endif |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6092 |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6093 |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6094 /* cat: 0-> DC 16x16 n = 0 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6095 * 1-> AC 16x16 n = luma4x4idx |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6096 * 2-> Luma4x4 n = luma4x4idx |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6097 * 3-> DC Chroma n = iCbCr |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6098 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx |
2755 | 6099 * 5-> Luma8x8 n = 4 * luma8x8idx |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6100 */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6101 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6102 /* read coded block flag */ |
2898 | 6103 if( cat != 5 ) { |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6104 if( get_cabac( CC, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) { |
2755 | 6105 if( cat == 1 || cat == 2 ) |
6106 h->non_zero_count_cache[scan8[n]] = 0; | |
6107 else if( cat == 4 ) | |
6108 h->non_zero_count_cache[scan8[16+n]] = 0; | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6109 #ifdef CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6110 h->cabac.range = cc.range ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6111 h->cabac.low = cc.low ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6112 h->cabac.bytestream= cc.bytestream; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6113 #endif |
2755 | 6114 return 0; |
6115 } | |
2898 | 6116 } |
6117 | |
6118 significant_coeff_ctx_base = h->cabac_state | |
3316 | 6119 + significant_coeff_flag_offset[MB_FIELD][cat]; |
2898 | 6120 last_coeff_ctx_base = h->cabac_state |
3316 | 6121 + last_coeff_flag_offset[MB_FIELD][cat]; |
2898 | 6122 abs_level_m1_ctx_base = h->cabac_state |
6123 + coeff_abs_level_m1_offset[cat]; | |
6124 | |
6125 if( cat == 5 ) { | |
6126 #define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \ | |
6127 for(last= 0; last < coefs; last++) { \ | |
6128 uint8_t *sig_ctx = significant_coeff_ctx_base + sig_off; \ | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6129 if( get_cabac( CC, sig_ctx )) { \ |
2898 | 6130 uint8_t *last_ctx = last_coeff_ctx_base + last_off; \ |
6131 index[coeff_count++] = last; \ | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6132 if( get_cabac( CC, last_ctx ) ) { \ |
2898 | 6133 last= max_coeff; \ |
6134 break; \ | |
6135 } \ | |
6136 } \ | |
4037
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
6137 }\ |
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
6138 if( last == max_coeff -1 ) {\ |
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
6139 index[coeff_count++] = last;\ |
2898 | 6140 } |
4009 | 6141 const uint8_t *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD]; |
4113
4ce3923d5806
Reenable AMD64 optimizations for cabac accidentially disabled in r6852
reimar
parents:
4112
diff
changeset
|
6142 #if defined(ARCH_X86) && !(defined(PIC) && defined(__GNUC__)) |
4051 | 6143 coeff_count= decode_significance_8x8_x86(CC, significant_coeff_ctx_base, index, sig_off); |
6144 } else { | |
6145 coeff_count= decode_significance_x86(CC, max_coeff, significant_coeff_ctx_base, index); | |
6146 #else | |
3316 | 6147 DECODE_SIGNIFICANCE( 63, sig_off[last], last_coeff_flag_offset_8x8[last] ); |
2898 | 6148 } else { |
6149 DECODE_SIGNIFICANCE( max_coeff - 1, last, last ); | |
4037
53be304c7f54
x86 asm version of the decode significance loop (not 8x8) of decode_residual() 5% faster decode_residual() on P3
michael
parents:
4016
diff
changeset
|
6150 #endif |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6151 } |
2316 | 6152 assert(coeff_count > 0); |
6153 | |
6154 if( cat == 0 ) | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6155 h->cbp_table[mb_xy] |= 0x100; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6156 else if( cat == 1 || cat == 2 ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6157 h->non_zero_count_cache[scan8[n]] = coeff_count; |
2316 | 6158 else if( cat == 3 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6159 h->cbp_table[mb_xy] |= 0x40 << n; |
2755 | 6160 else if( cat == 4 ) |
6161 h->non_zero_count_cache[scan8[16+n]] = coeff_count; | |
2316 | 6162 else { |
2755 | 6163 assert( cat == 5 ); |
3105
2d35fb3cb940
h264: special case dc-only idct. ~1% faster overall
lorenm
parents:
3101
diff
changeset
|
6164 fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1); |
2316 | 6165 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6166 |
4041
0113ceb0f953
1 variable less in decode_cabac_residual() no speed change (but less vars means less things gcc can load and store redundantly)
michael
parents:
4040
diff
changeset
|
6167 for( coeff_count--; coeff_count >= 0; coeff_count-- ) { |
2898 | 6168 uint8_t *ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + abs_level_m1_ctx_base; |
4041
0113ceb0f953
1 variable less in decode_cabac_residual() no speed change (but less vars means less things gcc can load and store redundantly)
michael
parents:
4040
diff
changeset
|
6169 int j= scantable[index[coeff_count]]; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6170 |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6171 if( get_cabac( CC, ctx ) == 0 ) { |
2898 | 6172 if( !qmul ) { |
4040 | 6173 block[j] = get_cabac_bypass_sign( CC, -1); |
2316 | 6174 }else{ |
4040 | 6175 block[j] = (get_cabac_bypass_sign( CC, -qmul[j]) + 32) >> 6;; |
2316 | 6176 } |
2967 | 6177 |
2313 | 6178 abslevel1++; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6179 } else { |
2313 | 6180 int coeff_abs = 2; |
2898 | 6181 ctx = 5 + FFMIN( 4, abslevelgt1 ) + abs_level_m1_ctx_base; |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6182 while( coeff_abs < 15 && get_cabac( CC, ctx ) ) { |
2313 | 6183 coeff_abs++; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6184 } |
2313 | 6185 |
6186 if( coeff_abs >= 15 ) { | |
6187 int j = 0; | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6188 while( get_cabac_bypass( CC ) ) { |
2313 | 6189 j++; |
6190 } | |
2967 | 6191 |
4016 | 6192 coeff_abs=1; |
2313 | 6193 while( j-- ) { |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6194 coeff_abs += coeff_abs + get_cabac_bypass( CC ); |
2313 | 6195 } |
4016 | 6196 coeff_abs+= 14; |
2313 | 6197 } |
6198 | |
2898 | 6199 if( !qmul ) { |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6200 if( get_cabac_bypass( CC ) ) block[j] = -coeff_abs; |
2316 | 6201 else block[j] = coeff_abs; |
6202 }else{ | |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6203 if( get_cabac_bypass( CC ) ) block[j] = (-coeff_abs * qmul[j] + 32) >> 6; |
2919 | 6204 else block[j] = ( coeff_abs * qmul[j] + 32) >> 6; |
2316 | 6205 } |
2967 | 6206 |
2313 | 6207 abslevelgt1++; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6208 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6209 } |
4038
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6210 #ifdef CABAC_ON_STACK |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6211 h->cabac.range = cc.range ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6212 h->cabac.low = cc.low ; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6213 h->cabac.bytestream= cc.bytestream; |
16697bdf4ac2
copy cabac context onto the stack for the c code in decode_cabac_residual() (15% faster on P3 but still slower then the current asm)
michael
parents:
4037
diff
changeset
|
6214 #endif |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6215 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6216 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6217 |
3077 | 6218 static void inline compute_mb_neighbors(H264Context *h) |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6219 { |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6220 MpegEncContext * const s = &h->s; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6221 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6222 h->top_mb_xy = mb_xy - s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6223 h->left_mb_xy[0] = mb_xy - 1; |
3316 | 6224 if(FRAME_MBAFF){ |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6225 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6226 const int top_pair_xy = pair_xy - s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6227 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]); |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6228 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]); |
3316 | 6229 const int curr_mb_frame_flag = !MB_FIELD; |
2597
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6230 const int bottom = (s->mb_y & 1); |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6231 if (bottom |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6232 ? !curr_mb_frame_flag // bottom macroblock |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6233 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6234 ) { |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6235 h->top_mb_xy -= s->mb_stride; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6236 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6237 if (left_mb_frame_flag != curr_mb_frame_flag) { |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6238 h->left_mb_xy[0] = pair_xy - 1; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6239 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6240 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6241 return; |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6242 } |
b5b09255f7c3
CABAC support for MBAFF I frames patch by (Loc Le Loarer < lll+ffmpeg m4x org)
michael
parents:
2594
diff
changeset
|
6243 |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6244 /** |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6245 * decodes a macroblock |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6246 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6247 */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6248 static int decode_mb_cabac(H264Context *h) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6249 MpegEncContext * const s = &h->s; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6250 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6251 int mb_type, partition_count, cbp = 0; |
2755 | 6252 int dct8x8_allowed= h->pps.transform_8x8_mode; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6253 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
6254 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6255 |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
6256 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6257 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) { |
3316 | 6258 int skip; |
6259 /* a skipped mb needs the aff flag from the following mb */ | |
6260 if( FRAME_MBAFF && s->mb_x==0 && (s->mb_y&1)==0 ) | |
6261 predict_field_decoding_flag(h); | |
6262 if( FRAME_MBAFF && (s->mb_y&1)==1 && h->prev_mb_skipped ) | |
6263 skip = h->next_mb_skipped; | |
6264 else | |
6265 skip = decode_cabac_mb_skip( h, s->mb_x, s->mb_y ); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6266 /* read skip flags */ |
3316 | 6267 if( skip ) { |
6268 if( FRAME_MBAFF && (s->mb_y&1)==0 ){ | |
6269 s->current_picture.mb_type[mb_xy] = MB_TYPE_SKIP; | |
6270 h->next_mb_skipped = decode_cabac_mb_skip( h, s->mb_x, s->mb_y+1 ); | |
6271 if(h->next_mb_skipped) | |
6272 predict_field_decoding_flag(h); | |
6273 else | |
6274 h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h); | |
6275 } | |
6276 | |
2396 | 6277 decode_mb_skip(h); |
6278 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6279 h->cbp_table[mb_xy] = 0; |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6280 h->chroma_pred_mode_table[mb_xy] = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6281 h->last_qscale_diff = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6282 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6283 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6284 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6285 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6286 } |
3316 | 6287 if(FRAME_MBAFF){ |
6288 if( (s->mb_y&1) == 0 ) | |
6289 h->mb_mbaff = | |
2594
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
6290 h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h); |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
6291 }else |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
6292 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME); |
5357b214eda0
CABAC support for MBAFF I frames patch by (Loc Le Loarer | lll+ffmpeg m4x org)
michael
parents:
2583
diff
changeset
|
6293 |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
6294 h->prev_mb_skipped = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6295 |
3077 | 6296 compute_mb_neighbors(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6297 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6298 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6299 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6300 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6301 |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6302 if( h->slice_type == B_TYPE ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6303 if( mb_type < 23 ){ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6304 partition_count= b_mb_type_info[mb_type].partition_count; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6305 mb_type= b_mb_type_info[mb_type].type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6306 }else{ |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6307 mb_type -= 23; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6308 goto decode_intra_mb; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6309 } |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6310 } else if( h->slice_type == P_TYPE ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6311 if( mb_type < 5) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6312 partition_count= p_mb_type_info[mb_type].partition_count; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6313 mb_type= p_mb_type_info[mb_type].type; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6314 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6315 mb_type -= 5; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6316 goto decode_intra_mb; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6317 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6318 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6319 assert(h->slice_type == I_TYPE); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6320 decode_intra_mb: |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6321 partition_count = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6322 cbp= i_mb_type_info[mb_type].cbp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6323 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6324 mb_type= i_mb_type_info[mb_type].type; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6325 } |
3316 | 6326 if(MB_FIELD) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6327 mb_type |= MB_TYPE_INTERLACED; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6328 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6329 h->slice_table[ mb_xy ]= h->slice_num; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6330 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6331 if(IS_INTRA_PCM(mb_type)) { |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6332 const uint8_t *ptr; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6333 unsigned int x, y; |
2967 | 6334 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6335 // We assume these blocks are very rare so we dont optimize it. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6336 // FIXME The two following lines get the bitstream position in the cabac |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6337 // decode, I think it should be done by a function in cabac.h (or cabac.c). |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6338 ptr= h->cabac.bytestream; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6339 if (h->cabac.low&0x1) ptr-=CABAC_BITS/8; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6340 |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6341 // The pixels are stored in the same order as levels in h->mb array. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6342 for(y=0; y<16; y++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6343 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6344 for(x=0; x<16; x++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6345 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6346 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6347 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6348 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6349 for(y=0; y<8; y++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6350 const int index= 256 + 4*(y&3) + 32*(y>>2); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6351 for(x=0; x<8; x++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6352 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6353 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6354 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6355 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6356 for(y=0; y<8; y++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6357 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6358 for(x=0; x<8; x++){ |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6359 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6360 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++; |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6361 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6362 } |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6363 |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6364 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr); |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6365 |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
6366 // All blocks are present |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6367 h->cbp_table[mb_xy] = 0x1ef; |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6368 h->chroma_pred_mode_table[mb_xy] = 0; |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
6369 // In deblocking, the quantizer is 0 |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6370 s->current_picture.qscale_table[mb_xy]= 0; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6371 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0); |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
6372 // All coeffs are present |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6373 memset(h->non_zero_count[mb_xy], 16, 16); |
2755 | 6374 s->current_picture.mb_type[mb_xy]= mb_type; |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
6375 return 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6376 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6377 |
3316 | 6378 if(MB_MBAFF){ |
6379 h->ref_count[0] <<= 1; | |
6380 h->ref_count[1] <<= 1; | |
6381 } | |
6382 | |
2449 | 6383 fill_caches(h, mb_type, 0); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6384 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6385 if( IS_INTRA( mb_type ) ) { |
2755 | 6386 int i; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6387 if( IS_INTRA4x4( mb_type ) ) { |
2755 | 6388 if( dct8x8_allowed && decode_cabac_mb_transform_size( h ) ) { |
6389 mb_type |= MB_TYPE_8x8DCT; | |
6390 for( i = 0; i < 16; i+=4 ) { | |
6391 int pred = pred_intra_mode( h, i ); | |
6392 int mode = decode_cabac_mb_intra4x4_pred_mode( h, pred ); | |
6393 fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 ); | |
6394 } | |
6395 } else { | |
6396 for( i = 0; i < 16; i++ ) { | |
6397 int pred = pred_intra_mode( h, i ); | |
6398 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred ); | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6399 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6400 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] ); |
2755 | 6401 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6402 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6403 write_back_intra_pred_mode(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6404 if( check_intra4x4_pred_mode(h) < 0 ) return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6405 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6406 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6407 if( h->intra16x16_pred_mode < 0 ) return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6408 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6409 h->chroma_pred_mode_table[mb_xy] = |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6410 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6411 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6412 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6413 if( h->chroma_pred_mode < 0 ) return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6414 } else if( partition_count == 4 ) { |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6415 int i, j, sub_partition_count[4], list, ref[2][4]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6416 |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6417 if( h->slice_type == B_TYPE ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6418 for( i = 0; i < 4; i++ ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6419 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6420 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6421 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6422 } |
3482 | 6423 if( IS_DIRECT(h->sub_mb_type[0] | h->sub_mb_type[1] | |
6424 h->sub_mb_type[2] | h->sub_mb_type[3]) ) { | |
2396 | 6425 pred_direct_motion(h, &mb_type); |
6426 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) { | |
6427 for( i = 0; i < 4; i++ ) | |
6428 if( IS_DIRECT(h->sub_mb_type[i]) ) | |
6429 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 ); | |
6430 } | |
6431 } | |
2310
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6432 } else { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6433 for( i = 0; i < 4; i++ ) { |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6434 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h ); |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6435 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6436 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type; |
c5cd8a064c34
H.264 CABAC + B-frames patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2284
diff
changeset
|
6437 } |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6438 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6439 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6440 for( list = 0; list < 2; list++ ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6441 if( h->ref_count[list] > 0 ) { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6442 for( i = 0; i < 4; i++ ) { |
2396 | 6443 if(IS_DIRECT(h->sub_mb_type[i])) continue; |
6444 if(IS_DIR(h->sub_mb_type[i], 0, list)){ | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6445 if( h->ref_count[list] > 1 ) |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6446 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6447 else |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6448 ref[list][i] = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6449 } else { |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6450 ref[list][i] = -1; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6451 } |
2110 | 6452 h->ref_cache[list][ scan8[4*i]+1 ]= |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6453 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6454 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6455 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6456 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6457 |
2755 | 6458 if(dct8x8_allowed) |
6459 dct8x8_allowed = get_dct8x8_allowed(h); | |
6460 | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6461 for(list=0; list<2; list++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6462 for(i=0; i<4; i++){ |
2396 | 6463 if(IS_DIRECT(h->sub_mb_type[i])){ |
6464 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4); | |
6465 continue; | |
6466 } | |
2110 | 6467 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]; |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6468 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6469 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6470 const int sub_mb_type= h->sub_mb_type[i]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6471 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6472 for(j=0; j<sub_partition_count[i]; j++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6473 int mpx, mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6474 int mx, my; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6475 const int index= 4*i + block_width*j; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6476 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6477 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6478 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6479 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6480 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6481 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6482 tprintf("final mv:%d %d\n", mx, my); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6483 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6484 if(IS_SUB_8X8(sub_mb_type)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6485 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6486 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6487 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6488 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6489 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6490 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6491 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6492 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6493 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6494 }else if(IS_SUB_8X4(sub_mb_type)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6495 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6496 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6497 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6498 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6499 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6500 }else if(IS_SUB_4X8(sub_mb_type)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6501 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6502 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6503 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6504 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6505 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6506 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6507 assert(IS_SUB_4X4(sub_mb_type)); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6508 mv_cache[ 0 ][0]= mx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6509 mv_cache[ 0 ][1]= my; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6510 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6511 mvd_cache[ 0 ][0]= mx - mpx; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6512 mvd_cache[ 0 ][1]= my - mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6513 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6514 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6515 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6516 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6517 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0]; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6518 p[0] = p[1] = p[8] = p[9] = 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6519 pd[0]= pd[1]= pd[8]= pd[9]= 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6520 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6521 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6522 } |
2396 | 6523 } else if( IS_DIRECT(mb_type) ) { |
6524 pred_direct_motion(h, &mb_type); | |
6525 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4); | |
6526 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4); | |
2755 | 6527 dct8x8_allowed &= h->sps.direct_8x8_inference_flag; |
2396 | 6528 } else { |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6529 int list, mx, my, i, mpx, mpy; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6530 if(IS_16X16(mb_type)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6531 for(list=0; list<2; list++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6532 if(IS_DIR(mb_type, 0, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6533 if(h->ref_count[list] > 0 ){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6534 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6535 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6536 } |
2523 | 6537 }else |
6538 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6539 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6540 for(list=0; list<2; list++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6541 if(IS_DIR(mb_type, 0, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6542 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6543 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6544 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6545 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6546 tprintf("final mv:%d %d\n", mx, my); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6547 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6548 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6549 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4); |
2523 | 6550 }else |
6551 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6552 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6553 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6554 else if(IS_16X8(mb_type)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6555 for(list=0; list<2; list++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6556 if(h->ref_count[list]>0){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6557 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6558 if(IS_DIR(mb_type, i, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6559 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6560 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1); |
2396 | 6561 }else |
6562 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6563 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6564 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6565 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6566 for(list=0; list<2; list++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6567 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6568 if(IS_DIR(mb_type, i, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6569 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6570 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6571 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6572 tprintf("final mv:%d %d\n", mx, my); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6573 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6574 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6575 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4); |
2523 | 6576 }else{ |
2396 | 6577 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4); |
6578 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6579 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6580 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6581 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6582 }else{ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6583 assert(IS_8X16(mb_type)); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6584 for(list=0; list<2; list++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6585 if(h->ref_count[list]>0){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6586 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6587 if(IS_DIR(mb_type, i, list)){ //FIXME optimize |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6588 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0; |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6589 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1); |
2396 | 6590 }else |
6591 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6592 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6593 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6594 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6595 for(list=0; list<2; list++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6596 for(i=0; i<2; i++){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6597 if(IS_DIR(mb_type, i, list)){ |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6598 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6599 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6600 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 ); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6601 |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6602 tprintf("final mv:%d %d\n", mx, my); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6603 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4); |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6604 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4); |
2523 | 6605 }else{ |
2396 | 6606 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4); |
6607 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4); | |
1935
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6608 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6609 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6610 } |
e935af90767b
progressive P frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1909
diff
changeset
|
6611 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6612 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6613 |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6614 if( IS_INTER( mb_type ) ) { |
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6615 h->chroma_pred_mode_table[mb_xy] = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6616 write_back_motion( h, mb_type ); |
1956
0eb2947f56f6
h264 hurry up fix and a tiny cabac clean patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1935
diff
changeset
|
6617 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6618 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6619 if( !IS_INTRA16x16( mb_type ) ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6620 cbp = decode_cabac_mb_cbp_luma( h ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6621 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6622 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6623 |
3651 | 6624 h->cbp_table[mb_xy] = h->cbp = cbp; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6625 |
2755 | 6626 if( dct8x8_allowed && (cbp&15) && !IS_INTRA( mb_type ) ) { |
6627 if( decode_cabac_mb_transform_size( h ) ) | |
6628 mb_type |= MB_TYPE_8x8DCT; | |
6629 } | |
6630 s->current_picture.mb_type[mb_xy]= mb_type; | |
6631 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6632 if( cbp || IS_INTRA16x16( mb_type ) ) { |
3174 | 6633 const uint8_t *scan, *scan8x8, *dc_scan; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6634 int dqp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6635 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6636 if(IS_INTERLACED(mb_type)){ |
3316 | 6637 scan8x8= s->qscale ? h->field_scan8x8 : h->field_scan8x8_q0; |
2763 | 6638 scan= s->qscale ? h->field_scan : h->field_scan_q0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6639 dc_scan= luma_dc_field_scan; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6640 }else{ |
3316 | 6641 scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0; |
2763 | 6642 scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6643 dc_scan= luma_dc_zigzag_scan; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6644 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6645 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6646 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h ); |
2883
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6647 if( dqp == INT_MIN ){ |
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6648 av_log(h->s.avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", s->mb_x, s->mb_y); |
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6649 return -1; |
a95e55679dfa
check return value of decode_cabac_mb_dqp, it is INT_MIN on error.
reimar
parents:
2867
diff
changeset
|
6650 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6651 s->qscale += dqp; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6652 if(((unsigned)s->qscale) > 51){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6653 if(s->qscale<0) s->qscale+= 52; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6654 else s->qscale-= 52; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6655 } |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6656 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6657 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6658 if( IS_INTRA16x16( mb_type ) ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6659 int i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6660 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" ); |
2898 | 6661 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, NULL, 16) < 0) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6662 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6663 if( cbp&15 ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6664 for( i = 0; i < 16; i++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6665 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i ); |
2919 | 6666 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6667 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6668 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6669 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6670 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6671 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6672 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6673 int i8x8, i4x4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6674 for( i8x8 = 0; i8x8 < 4; i8x8++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6675 if( cbp & (1<<i8x8) ) { |
2755 | 6676 if( IS_8x8DCT(mb_type) ) { |
6677 if( decode_cabac_residual(h, h->mb + 64*i8x8, 5, 4*i8x8, | |
3174 | 6678 scan8x8, h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 64) < 0 ) |
2755 | 6679 return -1; |
6680 } else | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6681 for( i4x4 = 0; i4x4 < 4; i4x4++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6682 const int index = 4*i8x8 + i4x4; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6683 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index ); |
3972
f5f1c9af095d
move outcommented START/STOP_TIMER to a hopefully better place for benchmarking ...
michael
parents:
3955
diff
changeset
|
6684 //START_TIMER |
2919 | 6685 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) < 0 ) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6686 return -1; |
3972
f5f1c9af095d
move outcommented START/STOP_TIMER to a hopefully better place for benchmarking ...
michael
parents:
3955
diff
changeset
|
6687 //STOP_TIMER("decode_residual") |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6688 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6689 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6690 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6691 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6692 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6693 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6694 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6695 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6696 if( cbp&0x30 ){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6697 int c; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6698 for( c = 0; c < 2; c++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6699 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c ); |
2898 | 6700 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, NULL, 4) < 0) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6701 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6702 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6703 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6704 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6705 if( cbp&0x20 ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6706 int c, i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6707 for( c = 0; c < 2; c++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6708 for( i = 0; i < 4; i++ ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6709 const int index = 16 + 4 * c + i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6710 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 ); |
2919 | 6711 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->dequant4_coeff[c+1+(IS_INTRA( mb_type ) ? 0:3)][h->chroma_qp], 15) < 0) |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6712 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6713 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6714 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6715 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6716 uint8_t * const nnz= &h->non_zero_count_cache[0]; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6717 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6718 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6719 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6720 } else { |
2315 | 6721 uint8_t * const nnz= &h->non_zero_count_cache[0]; |
6722 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1); | |
6723 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] = | |
6724 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0; | |
3316 | 6725 h->last_qscale_diff = 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6726 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6727 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6728 s->current_picture.qscale_table[mb_xy]= s->qscale; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6729 write_back_non_zero_count(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6730 |
3316 | 6731 if(MB_MBAFF){ |
6732 h->ref_count[0] >>= 1; | |
6733 h->ref_count[1] >>= 1; | |
6734 } | |
6735 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6736 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6737 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6738 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
6739 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6740 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6741 int i, d; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6742 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6743 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6744 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6745 |
2633 | 6746 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6747 int8_t tc[4]; |
2633 | 6748 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6749 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] : -1; |
2633 | 6750 h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc); |
6751 } else { | |
6752 /* 16px edge length, because bS=4 is triggered by being at | |
6753 * the edge of an intra MB, so all 4 bS are the same */ | |
6754 for( d = 0; d < 16; d++ ) { | |
1898 | 6755 const int p0 = pix[-1]; |
6756 const int p1 = pix[-2]; | |
6757 const int p2 = pix[-3]; | |
6758 | |
6759 const int q0 = pix[0]; | |
6760 const int q1 = pix[1]; | |
6761 const int q2 = pix[2]; | |
6762 | |
4001 | 6763 if( FFABS( p0 - q0 ) < alpha && |
6764 FFABS( p1 - p0 ) < beta && | |
6765 FFABS( q1 - q0 ) < beta ) { | |
6766 | |
6767 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
6768 if( FFABS( p2 - p0 ) < beta) | |
1898 | 6769 { |
6770 const int p3 = pix[-4]; | |
6771 /* p0', p1', p2' */ | |
6772 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
6773 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
6774 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
6775 } else { | |
6776 /* p0' */ | |
6777 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
6778 } | |
4001 | 6779 if( FFABS( q2 - q0 ) < beta) |
1898 | 6780 { |
6781 const int q3 = pix[3]; | |
6782 /* q0', q1', q2' */ | |
6783 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
6784 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
6785 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
6786 } else { | |
6787 /* q0' */ | |
6788 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
6789 } | |
6790 }else{ | |
6791 /* p0', q0' */ | |
6792 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
6793 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
6794 } | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6795 tprintf("filter_mb_edgev i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, p2, p1, p0, q0, q1, q2, pix[-2], pix[-1], pix[0], pix[1]); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6796 } |
1898 | 6797 pix += stride; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6798 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6799 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6800 } |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6801 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
2834 | 6802 int i; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6803 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6804 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6805 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6806 |
2633 | 6807 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6808 int8_t tc[4]; |
2633 | 6809 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6810 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] + 1 : 0; |
2633 | 6811 h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc); |
6812 } else { | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6813 h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6814 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6815 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6816 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6817 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6818 int i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6819 for( i = 0; i < 16; i++, pix += stride) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6820 int index_a; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6821 int alpha; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6822 int beta; |
2967 | 6823 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6824 int qp_index; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6825 int bS_index = (i >> 1); |
3316 | 6826 if (!MB_FIELD) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6827 bS_index &= ~1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6828 bS_index |= (i & 1); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6829 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6830 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6831 if( bS[bS_index] == 0 ) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6832 continue; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6833 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6834 |
3316 | 6835 qp_index = MB_FIELD ? (i >> 3) : (i & 1); |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6836 index_a = qp[qp_index] + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6837 alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6838 beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6839 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6840 if( bS[bS_index] < 4 ) { |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6841 const int tc0 = (tc0_table+52)[index_a][bS[bS_index] - 1]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6842 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6843 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6844 const int p2 = pix[-3]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6845 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6846 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6847 const int q2 = pix[2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6848 |
4001 | 6849 if( FFABS( p0 - q0 ) < alpha && |
6850 FFABS( p1 - p0 ) < beta && | |
6851 FFABS( q1 - q0 ) < beta ) { | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6852 int tc = tc0; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6853 int i_delta; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6854 |
4001 | 6855 if( FFABS( p2 - p0 ) < beta ) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6856 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 ); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6857 tc++; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6858 } |
4001 | 6859 if( FFABS( q2 - q0 ) < beta ) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6860 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 ); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6861 tc++; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6862 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6863 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6864 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6865 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6866 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6867 tprintf("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); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6868 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6869 }else{ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6870 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6871 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6872 const int p2 = pix[-3]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6873 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6874 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6875 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6876 const int q2 = pix[2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6877 |
4001 | 6878 if( FFABS( p0 - q0 ) < alpha && |
6879 FFABS( p1 - p0 ) < beta && | |
6880 FFABS( q1 - q0 ) < beta ) { | |
6881 | |
6882 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ | |
6883 if( FFABS( p2 - p0 ) < beta) | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6884 { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6885 const int p3 = pix[-4]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6886 /* p0', p1', p2' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6887 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6888 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6889 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6890 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6891 /* p0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6892 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6893 } |
4001 | 6894 if( FFABS( q2 - q0 ) < beta) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6895 { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6896 const int q3 = pix[3]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6897 /* q0', q1', q2' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6898 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6899 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6900 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6901 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6902 /* q0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6903 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6904 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6905 }else{ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6906 /* p0', q0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6907 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6908 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6909 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6910 tprintf("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]); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6911 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6912 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6913 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6914 } |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6915 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) { |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6916 int i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6917 for( i = 0; i < 8; i++, pix += stride) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6918 int index_a; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6919 int alpha; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6920 int beta; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6921 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6922 int qp_index; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6923 int bS_index = i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6924 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6925 if( bS[bS_index] == 0 ) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6926 continue; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6927 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6928 |
3316 | 6929 qp_index = MB_FIELD ? (i >> 2) : (i & 1); |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6930 index_a = qp[qp_index] + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6931 alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6932 beta = (beta_table+52)[qp[qp_index] + h->slice_beta_offset]; |
3316 | 6933 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6934 if( bS[bS_index] < 4 ) { |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6935 const int tc = (tc0_table+52)[index_a][bS[bS_index] - 1] + 1; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6936 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6937 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6938 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6939 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6940 |
4001 | 6941 if( FFABS( p0 - q0 ) < alpha && |
6942 FFABS( p1 - p0 ) < beta && | |
6943 FFABS( q1 - q0 ) < beta ) { | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6944 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6945 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6946 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6947 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6948 tprintf("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); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6949 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6950 }else{ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6951 const int p0 = pix[-1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6952 const int p1 = pix[-2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6953 const int q0 = pix[0]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6954 const int q1 = pix[1]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6955 |
4001 | 6956 if( FFABS( p0 - q0 ) < alpha && |
6957 FFABS( p1 - p0 ) < beta && | |
6958 FFABS( q1 - q0 ) < beta ) { | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6959 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6960 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6961 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */ |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6962 tprintf("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]); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6963 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6964 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6965 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6966 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
6967 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
6968 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6969 int i, d; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6970 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6971 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6972 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6973 const int pix_next = stride; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
6974 |
2633 | 6975 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
6976 int8_t tc[4]; |
2633 | 6977 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
6978 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] : -1; |
2633 | 6979 h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc); |
6980 } else { | |
6981 /* 16px edge length, see filter_mb_edgev */ | |
6982 for( d = 0; d < 16; d++ ) { | |
1898 | 6983 const int p0 = pix[-1*pix_next]; |
6984 const int p1 = pix[-2*pix_next]; | |
6985 const int p2 = pix[-3*pix_next]; | |
6986 const int q0 = pix[0]; | |
6987 const int q1 = pix[1*pix_next]; | |
6988 const int q2 = pix[2*pix_next]; | |
6989 | |
4001 | 6990 if( FFABS( p0 - q0 ) < alpha && |
6991 FFABS( p1 - p0 ) < beta && | |
6992 FFABS( q1 - q0 ) < beta ) { | |
1898 | 6993 |
6994 const int p3 = pix[-4*pix_next]; | |
6995 const int q3 = pix[ 3*pix_next]; | |
6996 | |
4001 | 6997 if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){ |
6998 if( FFABS( p2 - p0 ) < beta) { | |
1898 | 6999 /* p0', p1', p2' */ |
7000 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3; | |
7001 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2; | |
7002 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3; | |
7003 } else { | |
7004 /* p0' */ | |
7005 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
7006 } | |
4001 | 7007 if( FFABS( q2 - q0 ) < beta) { |
1898 | 7008 /* q0', q1', q2' */ |
7009 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3; | |
7010 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2; | |
7011 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3; | |
7012 } else { | |
7013 /* q0' */ | |
7014 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
7015 } | |
7016 }else{ | |
7017 /* p0', q0' */ | |
7018 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; | |
7019 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2; | |
7020 } | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7021 tprintf("filter_mb_edgeh i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, bS[i], p2, p1, p0, q0, q1, q2, pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7022 } |
1898 | 7023 pix++; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7024 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7025 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7026 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7027 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7028 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) { |
2834 | 7029 int i; |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
7030 const int index_a = qp + h->slice_alpha_c0_offset; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
7031 const int alpha = (alpha_table+52)[index_a]; |
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
7032 const int beta = (beta_table+52)[qp + h->slice_beta_offset]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7033 |
2633 | 7034 if( bS[0] < 4 ) { |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
7035 int8_t tc[4]; |
2633 | 7036 for(i=0; i<4; i++) |
4118
a1786732cd62
merge clip() into the alpha/beta/tc0 tables (10% faster filter_mb_fast() on P3)
michael
parents:
4113
diff
changeset
|
7037 tc[i] = bS[i] ? (tc0_table+52)[index_a][bS[i] - 1] + 1 : 0; |
2633 | 7038 h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc); |
7039 } else { | |
2707
360024d31dab
H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents:
2683
diff
changeset
|
7040 h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta); |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7041 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7042 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7043 |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7044 static void 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) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7045 MpegEncContext * const s = &h->s; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7046 int mb_xy, mb_type; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7047 int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7048 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7049 if(mb_x==0 || mb_y==0 || !s->dsp.h264_loop_filter_strength) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7050 filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7051 return; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7052 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7053 assert(!FRAME_MBAFF); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7054 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7055 mb_xy = mb_x + mb_y*s->mb_stride; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7056 mb_type = s->current_picture.mb_type[mb_xy]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7057 qp = s->current_picture.qscale_table[mb_xy]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7058 qp0 = s->current_picture.qscale_table[mb_xy-1]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7059 qp1 = s->current_picture.qscale_table[h->top_mb_xy]; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7060 qpc = get_chroma_qp( h->pps.chroma_qp_index_offset, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7061 qpc0 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7062 qpc1 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7063 qp0 = (qp + qp0 + 1) >> 1; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7064 qp1 = (qp + qp1 + 1) >> 1; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7065 qpc0 = (qpc + qpc0 + 1) >> 1; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7066 qpc1 = (qpc + qpc1 + 1) >> 1; |
3651 | 7067 qp_thresh = 15 - h->slice_alpha_c0_offset; |
7068 if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh && | |
7069 qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh) | |
7070 return; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7071 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7072 if( IS_INTRA(mb_type) ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7073 int16_t bS4[4] = {4,4,4,4}; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7074 int16_t bS3[4] = {3,3,3,3}; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7075 if( IS_8x8DCT(mb_type) ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7076 filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7077 filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7078 filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7079 filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7080 } else { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7081 filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7082 filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7083 filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7084 filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7085 filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7086 filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7087 filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7088 filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7089 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7090 filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7091 filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7092 filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7093 filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7094 filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bS4, qpc1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7095 filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7096 filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bS4, qpc1 ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7097 filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc ); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7098 return; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7099 } else { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7100 DECLARE_ALIGNED_8(int16_t, bS[2][4][4]); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7101 uint64_t (*bSv)[4] = (uint64_t(*)[4])bS; |
3651 | 7102 int edges; |
7103 if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) { | |
7104 edges = 4; | |
7105 bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL; | |
7106 } else { | |
7107 int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : | |
7108 (mb_type & MB_TYPE_16x8) ? 1 : 0; | |
7109 int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) | |
7110 && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16)) | |
7111 ? 3 : 0; | |
7112 int step = IS_8x8DCT(mb_type) ? 2 : 1; | |
7113 edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4; | |
7114 s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache, | |
7115 (h->slice_type == B_TYPE), edges, step, mask_edge0, mask_edge1 ); | |
7116 } | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7117 if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) ) |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7118 bSv[0][0] = 0x0004000400040004ULL; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7119 if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) ) |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7120 bSv[1][0] = 0x0004000400040004ULL; |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7121 |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7122 #define FILTER(hv,dir,edge)\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7123 if(bSv[dir][edge]) {\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7124 filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7125 if(!(edge&1)) {\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7126 filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7127 filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7128 }\ |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7129 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7130 if( edges == 1 ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7131 FILTER(v,0,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7132 FILTER(h,1,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7133 } else if( IS_8x8DCT(mb_type) ) { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7134 FILTER(v,0,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7135 FILTER(v,0,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7136 FILTER(h,1,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7137 FILTER(h,1,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7138 } else { |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7139 FILTER(v,0,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7140 FILTER(v,0,1); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7141 FILTER(v,0,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7142 FILTER(v,0,3); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7143 FILTER(h,1,0); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7144 FILTER(h,1,1); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7145 FILTER(h,1,2); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7146 FILTER(h,1,3); |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7147 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7148 #undef FILTER |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7149 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7150 } |
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7151 |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7152 static void 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) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7153 MpegEncContext * const s = &h->s; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7154 const int mb_xy= mb_x + mb_y*s->mb_stride; |
3316 | 7155 const int mb_type = s->current_picture.mb_type[mb_xy]; |
7156 const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4; | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7157 int first_vertical_edge_done = 0; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7158 int dir; |
2629
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
7159 /* FIXME: A given frame may occupy more than one position in |
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
7160 * the reference list. So ref2frm should be populated with |
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
7161 * frame numbers, not indices. */ |
3316 | 7162 static const int ref2frm[34] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, |
7163 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31}; | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7164 |
3101 | 7165 //for sufficiently low qp, filtering wouldn't do anything |
7166 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp | |
3316 | 7167 if(!FRAME_MBAFF){ |
3101 | 7168 int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX(0, h->pps.chroma_qp_index_offset); |
7169 int qp = s->current_picture.qscale_table[mb_xy]; | |
7170 if(qp <= qp_thresh | |
7171 && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh) | |
7172 && (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){ | |
7173 return; | |
7174 } | |
7175 } | |
7176 | |
3316 | 7177 if (FRAME_MBAFF |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7178 // left mb is in picture |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7179 && h->slice_table[mb_xy-1] != 255 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7180 // and current and left pair do not have the same interlaced type |
3316 | 7181 && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1])) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7182 // and left mb is in the same slice if deblocking_filter == 2 |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7183 && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7184 /* First vertical edge is different in MBAFF frames |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
7185 * There are 8 different bS to compute and 2 different Qp |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7186 */ |
3316 | 7187 const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride; |
7188 const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride }; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7189 int16_t bS[8]; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7190 int qp[2]; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7191 int chroma_qp[2]; |
3316 | 7192 int mb_qp, mbn0_qp, mbn1_qp; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7193 int i; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7194 first_vertical_edge_done = 1; |
3316 | 7195 |
7196 if( IS_INTRA(mb_type) ) | |
7197 bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4; | |
7198 else { | |
7199 for( i = 0; i < 8; i++ ) { | |
7200 int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1]; | |
7201 | |
7202 if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) | |
7203 bS[i] = 4; | |
7204 else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 || | |
7205 /* FIXME: with 8x8dct + cavlc, should check cbp instead of nnz */ | |
7206 h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2] ) | |
7207 bS[i] = 2; | |
7208 else | |
7209 bS[i] = 1; | |
7210 } | |
7211 } | |
7212 | |
7213 mb_qp = s->current_picture.qscale_table[mb_xy]; | |
7214 mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]]; | |
7215 mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]]; | |
7216 qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1; | |
7217 chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) + | |
7218 get_chroma_qp( h->pps.chroma_qp_index_offset, mbn0_qp ) + 1 ) >> 1; | |
7219 qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1; | |
7220 chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) + | |
7221 get_chroma_qp( h->pps.chroma_qp_index_offset, mbn1_qp ) + 1 ) >> 1; | |
7222 | |
7223 /* Filter edge */ | |
7224 tprintf("filter mb:%d/%d MBAFF, QPy:%d/%d, QPc:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], chroma_qp[0], chroma_qp[1], linesize, uvlinesize); | |
7225 { int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); } | |
7226 filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp ); | |
7227 filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp ); | |
7228 filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp ); | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7229 } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7230 /* dir : 0 -> vertical edge, 1 -> horizontal edge */ |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7231 for( dir = 0; dir < 2; dir++ ) |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7232 { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7233 int edge; |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7234 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy; |
2924 | 7235 const int mbm_type = s->current_picture.mb_type[mbm_xy]; |
2454 | 7236 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0; |
7237 | |
3101 | 7238 const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP)) |
7239 == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4; | |
2926 | 7240 // how often to recheck mv-based bS when iterating between edges |
7241 const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 : | |
7242 (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0; | |
7243 // how often to recheck mv-based bS when iterating along each edge | |
7244 const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)); | |
7245 | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7246 if (first_vertical_edge_done) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7247 start = 1; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7248 first_vertical_edge_done = 0; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7249 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7250 |
2454 | 7251 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy]) |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7252 start = 1; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7253 |
3316 | 7254 if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0 |
7255 && !IS_INTERLACED(mb_type) | |
7256 && IS_INTERLACED(mbm_type) | |
7257 ) { | |
7258 // This is a special case in the norm where the filtering must | |
7259 // be done twice (one each of the field) even if we are in a | |
7260 // frame macroblock. | |
7261 // | |
7262 static const int nnz_idx[4] = {4,5,6,3}; | |
7263 unsigned int tmp_linesize = 2 * linesize; | |
7264 unsigned int tmp_uvlinesize = 2 * uvlinesize; | |
7265 int mbn_xy = mb_xy - 2 * s->mb_stride; | |
7266 int qp, chroma_qp; | |
7267 int i, j; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7268 int16_t bS[4]; |
3316 | 7269 |
7270 for(j=0; j<2; j++, mbn_xy += s->mb_stride){ | |
7271 if( IS_INTRA(mb_type) || | |
7272 IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) { | |
7273 bS[0] = bS[1] = bS[2] = bS[3] = 3; | |
7274 } else { | |
7275 const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy]; | |
7276 for( i = 0; i < 4; i++ ) { | |
7277 if( h->non_zero_count_cache[scan8[0]+i] != 0 || | |
7278 mbn_nnz[nnz_idx[i]] != 0 ) | |
7279 bS[i] = 2; | |
7280 else | |
7281 bS[i] = 1; | |
7282 } | |
7283 } | |
7284 // Do not use s->qscale as luma quantizer because it has not the same | |
7285 // value in IPCM macroblocks. | |
7286 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; | |
7287 tprintf("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); | |
7288 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); } | |
7289 filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp ); | |
7290 chroma_qp = ( h->chroma_qp + | |
7291 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; | |
7292 filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp ); | |
7293 filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp ); | |
7294 } | |
7295 | |
7296 start = 1; | |
7297 } | |
7298 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7299 /* Calculate bS */ |
2924 | 7300 for( edge = start; edge < edges; edge++ ) { |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
7301 /* mbn_xy: neighbor macroblock */ |
2924 | 7302 const int mbn_xy = edge > 0 ? mb_xy : mbm_xy; |
7303 const int mbn_type = s->current_picture.mb_type[mbn_xy]; | |
3645
47821be55b6c
mmx implementation of deblocking strength decision.
lorenm
parents:
3519
diff
changeset
|
7304 int16_t bS[4]; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7305 int qp; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7306 |
2924 | 7307 if( (edge&1) && IS_8x8DCT(mb_type) ) |
2755 | 7308 continue; |
7309 | |
2924 | 7310 if( IS_INTRA(mb_type) || |
7311 IS_INTRA(mbn_type) ) { | |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7312 int value; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7313 if (edge == 0) { |
2924 | 7314 if ( (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type)) |
3316 | 7315 || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0)) |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7316 ) { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7317 value = 4; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7318 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7319 value = 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7320 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7321 } else { |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7322 value = 3; |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7323 } |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7324 bS[0] = bS[1] = bS[2] = bS[3] = value; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7325 } else { |
2924 | 7326 int i, l; |
7327 int mv_done; | |
7328 | |
7329 if( edge & mask_edge ) { | |
7330 bS[0] = bS[1] = bS[2] = bS[3] = 0; | |
7331 mv_done = 1; | |
7332 } | |
3316 | 7333 else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) { |
7334 bS[0] = bS[1] = bS[2] = bS[3] = 1; | |
7335 mv_done = 1; | |
7336 } | |
2924 | 7337 else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) { |
7338 int b_idx= 8 + 4 + edge * (dir ? 8:1); | |
7339 int bn_idx= b_idx - (dir ? 8:1); | |
7340 int v = 0; | |
7341 for( l = 0; !v && l < 1 + (h->slice_type == B_TYPE); l++ ) { | |
7342 v |= ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] || | |
4001 | 7343 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 || |
7344 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit; | |
2924 | 7345 } |
7346 bS[0] = bS[1] = bS[2] = bS[3] = v; | |
7347 mv_done = 1; | |
7348 } | |
7349 else | |
7350 mv_done = 0; | |
7351 | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7352 for( i = 0; i < 4; i++ ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7353 int x = dir == 0 ? edge : i; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7354 int y = dir == 0 ? i : edge; |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7355 int b_idx= 8 + 4 + x + 8*y; |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7356 int bn_idx= b_idx - (dir ? 8:1); |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7357 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7358 if( h->non_zero_count_cache[b_idx] != 0 || |
2449 | 7359 h->non_zero_count_cache[bn_idx] != 0 ) { |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7360 bS[i] = 2; |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7361 } |
2924 | 7362 else if(!mv_done) |
2523 | 7363 { |
7364 bS[i] = 0; | |
7365 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) { | |
2629
202cd69d8d2e
fix ref comparison in B-frame deblocker ('unused' and 'unavailable' were treated as different)
lorenm
parents:
2628
diff
changeset
|
7366 if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] || |
4001 | 7367 FFABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 || |
7368 FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) { | |
2523 | 7369 bS[i] = 1; |
7370 break; | |
7371 } | |
7372 } | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7373 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7374 } |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7375 |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7376 if(bS[0]+bS[1]+bS[2]+bS[3] == 0) |
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7377 continue; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7378 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7379 |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7380 /* Filter edge */ |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
7381 // Do not use s->qscale as luma quantizer because it has not the same |
2504
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
7382 // value in IPCM macroblocks. |
f12657081093
INTRA PCM macroblocks support patch by (Loic )lll+ffmpeg m4x org)
michael
parents:
2498
diff
changeset
|
7383 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1; |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
7384 //tprintf("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, s->current_picture.qscale_table[mbn_xy]); |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7385 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize); |
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7386 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); } |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7387 if( dir == 0 ) { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7388 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp ); |
1898 | 7389 if( (edge&1) == 0 ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7390 int chroma_qp = ( h->chroma_qp + |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7391 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7392 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7393 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7394 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7395 } else { |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7396 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp ); |
1898 | 7397 if( (edge&1) == 0 ) { |
1899
379a18a7131f
do loop filter immediatly after each macroblock is decoded instead of after a frame is decoded
michael
parents:
1898
diff
changeset
|
7398 int chroma_qp = ( h->chroma_qp + |
2581
ae72796e722f
This is the second patch for MBAFF support, this adds the deblocking
michael
parents:
2580
diff
changeset
|
7399 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1; |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7400 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7401 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp ); |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7402 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7403 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7404 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7405 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7406 } |
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
7407 |
1168 | 7408 static int decode_slice(H264Context *h){ |
7409 MpegEncContext * const s = &h->s; | |
7410 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F; | |
7411 | |
7412 s->mb_skip_run= -1; | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7413 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7414 if( h->pps.cabac ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7415 int i; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7416 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7417 /* realign */ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7418 align_get_bits( &s->gb ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7419 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7420 /* init cabac */ |
3993
8b7c59b7af01
make state transition tables global as they are constant and the code is slightly faster that way
michael
parents:
3991
diff
changeset
|
7421 ff_init_cabac_states( &h->cabac); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7422 ff_init_cabac_decoder( &h->cabac, |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7423 s->gb.buffer + get_bits_count(&s->gb)/8, |
2116 | 7424 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7425 /* calculate pre-state */ |
2755 | 7426 for( i= 0; i < 460; i++ ) { |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7427 int pre; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7428 if( h->slice_type == I_TYPE ) |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7429 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7430 else |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7431 pre = clip( ((cabac_context_init_PB[h->cabac_init_idc][i][0] * s->qscale) >>4 ) + cabac_context_init_PB[h->cabac_init_idc][i][1], 1, 126 ); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7432 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7433 if( pre <= 63 ) |
3976
27e90123b346
reverse remainder of the failed attempt to optimize *state=c->mps_state[s]
michael
parents:
3972
diff
changeset
|
7434 h->cabac_state[i] = 2 * ( 63 - pre ) + 0; |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7435 else |
3976
27e90123b346
reverse remainder of the failed attempt to optimize *state=c->mps_state[s]
michael
parents:
3972
diff
changeset
|
7436 h->cabac_state[i] = 2 * ( pre - 64 ) + 1; |
1168 | 7437 } |
7438 | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7439 for(;;){ |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
7440 //START_TIMER |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7441 int ret = decode_mb_cabac(h); |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7442 int eos; |
4008
b636f3d59283
prevent "mb level" get_cabac() calls from being inlined (3% faster decode_mb_cabac() on P3)
michael
parents:
4007
diff
changeset
|
7443 //STOP_TIMER("decode_mb_cabac") |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7444 |
2163 | 7445 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7446 |
3316 | 7447 if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ? |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7448 s->mb_y++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7449 |
2163 | 7450 if(ret>=0) ret = decode_mb_cabac(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7451 |
2844
5f20ab245501
prevent assert failures and infinite loops with broken streams
michael
parents:
2834
diff
changeset
|
7452 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7453 s->mb_y--; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7454 } |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7455 eos = get_cabac_terminate( &h->cabac ); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7456 |
3948
3edbf131ee44
refill cabac variables in 16bit steps, 3% faster get_cabac()
michael
parents:
3947
diff
changeset
|
7457 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) { |
3297
9ae9653d706c
decoders should (IMHO) not duplicate frames, the application can do this if it wants
michael
parents:
3291
diff
changeset
|
7458 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%d)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7459 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7460 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7461 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7462 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7463 if( ++s->mb_x >= s->mb_width ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7464 s->mb_x = 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7465 ff_draw_horiz_band(s, 16*s->mb_y, 16); |
2392 | 7466 ++s->mb_y; |
3316 | 7467 if(FRAME_MBAFF) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7468 ++s->mb_y; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7469 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7470 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7471 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7472 if( eos || s->mb_y >= s->mb_height ) { |
2392 | 7473 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7474 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7475 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7476 } |
1168 | 7477 } |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7478 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7479 } else { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7480 for(;;){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7481 int ret = decode_mb_cavlc(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7482 |
2163 | 7483 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7484 |
3316 | 7485 if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ? |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7486 s->mb_y++; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7487 ret = decode_mb_cavlc(h); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7488 |
2163 | 7489 if(ret>=0) hl_decode_mb(h); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7490 s->mb_y--; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7491 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7492 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7493 if(ret<0){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7494 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7495 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7496 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7497 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7498 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7499 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7500 if(++s->mb_x >= s->mb_width){ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7501 s->mb_x=0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7502 ff_draw_horiz_band(s, 16*s->mb_y, 16); |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7503 ++s->mb_y; |
3316 | 7504 if(FRAME_MBAFF) { |
2551
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7505 ++s->mb_y; |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7506 } |
615995277bc5
MBAFF I slice no deblocking patch by (Loic >>lll+ffmpeg m4x org<<)
michael
parents:
2548
diff
changeset
|
7507 if(s->mb_y >= s->mb_height){ |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7508 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7509 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7510 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) { |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7511 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7512 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7513 return 0; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7514 }else{ |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7515 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7516 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7517 return -1; |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7518 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7519 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7520 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7521 |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7522 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){ |
2392 | 7523 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits); |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7524 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){ |
1168 | 7525 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); |
7526 | |
7527 return 0; | |
7528 }else{ | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7529 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
1168 | 7530 |
7531 return -1; | |
7532 } | |
7533 } | |
7534 } | |
1908
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7535 } |
e20fd60b215c
h264 - progressive I frame CABAC support patch by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1899
diff
changeset
|
7536 |
1168 | 7537 #if 0 |
7538 for(;s->mb_y < s->mb_height; s->mb_y++){ | |
7539 for(;s->mb_x < s->mb_width; s->mb_x++){ | |
7540 int ret= decode_mb(h); | |
2967 | 7541 |
1168 | 7542 hl_decode_mb(h); |
7543 | |
7544 if(ret<0){ | |
3177 | 7545 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y); |
1168 | 7546 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); |
7547 | |
7548 return -1; | |
7549 } | |
2967 | 7550 |
1168 | 7551 if(++s->mb_x >= s->mb_width){ |
7552 s->mb_x=0; | |
7553 if(++s->mb_y >= s->mb_height){ | |
7554 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
7555 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
7556 | |
7557 return 0; | |
7558 }else{ | |
7559 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
7560 | |
7561 return -1; | |
7562 } | |
7563 } | |
7564 } | |
2967 | 7565 |
1168 | 7566 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){ |
7567 if(get_bits_count(s->gb) == s->gb.size_in_bits){ | |
7568 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask); | |
7569 | |
7570 return 0; | |
7571 }else{ | |
7572 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask); | |
7573 | |
7574 return -1; | |
7575 } | |
7576 } | |
7577 } | |
7578 s->mb_x=0; | |
7579 ff_draw_horiz_band(s, 16*s->mb_y, 16); | |
7580 } | |
7581 #endif | |
7582 return -1; //not reached | |
7583 } | |
7584 | |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7585 static int decode_unregistered_user_data(H264Context *h, int size){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7586 MpegEncContext * const s = &h->s; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7587 uint8_t user_data[16+256]; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7588 int e, build, i; |
2967 | 7589 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7590 if(size<16) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7591 return -1; |
2967 | 7592 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7593 for(i=0; i<sizeof(user_data)-1 && i<size; i++){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7594 user_data[i]= get_bits(&s->gb, 8); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7595 } |
2967 | 7596 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7597 user_data[i]= 0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7598 e= sscanf(user_data+16, "x264 - core %d"/*%s - H.264/MPEG-4 AVC codec - Copyleft 2005 - http://www.videolan.org/x264.html*/, &build); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7599 if(e==1 && build>=0) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7600 h->x264_build= build; |
2967 | 7601 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7602 if(s->avctx->debug & FF_DEBUG_BUGS) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7603 av_log(s->avctx, AV_LOG_DEBUG, "user data:\"%s\"\n", user_data+16); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7604 |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7605 for(; i<size; i++) |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7606 skip_bits(&s->gb, 8); |
2967 | 7607 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7608 return 0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7609 } |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7610 |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7611 static int decode_sei(H264Context *h){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7612 MpegEncContext * const s = &h->s; |
2967 | 7613 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7614 while(get_bits_count(&s->gb) + 16 < s->gb.size_in_bits){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7615 int size, type; |
2967 | 7616 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7617 type=0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7618 do{ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7619 type+= show_bits(&s->gb, 8); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7620 }while(get_bits(&s->gb, 8) == 255); |
2967 | 7621 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7622 size=0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7623 do{ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7624 size+= show_bits(&s->gb, 8); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7625 }while(get_bits(&s->gb, 8) == 255); |
2967 | 7626 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7627 switch(type){ |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7628 case 5: |
3318 | 7629 if(decode_unregistered_user_data(h, size) < 0) |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7630 return -1; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7631 break; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7632 default: |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7633 skip_bits(&s->gb, 8*size); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7634 } |
2967 | 7635 |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7636 //FIXME check bits here |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7637 align_get_bits(&s->gb); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7638 } |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7639 |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7640 return 0; |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7641 } |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
7642 |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7643 static inline void decode_hrd_parameters(H264Context *h, SPS *sps){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7644 MpegEncContext * const s = &h->s; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7645 int cpb_count, i; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7646 cpb_count = get_ue_golomb(&s->gb) + 1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7647 get_bits(&s->gb, 4); /* bit_rate_scale */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7648 get_bits(&s->gb, 4); /* cpb_size_scale */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7649 for(i=0; i<cpb_count; i++){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7650 get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7651 get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7652 get_bits1(&s->gb); /* cbr_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7653 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7654 get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7655 get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7656 get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7657 get_bits(&s->gb, 5); /* time_offset_length */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7658 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7659 |
1168 | 7660 static inline int decode_vui_parameters(H264Context *h, SPS *sps){ |
7661 MpegEncContext * const s = &h->s; | |
7662 int aspect_ratio_info_present_flag, aspect_ratio_idc; | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7663 int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag; |
1168 | 7664 |
7665 aspect_ratio_info_present_flag= get_bits1(&s->gb); | |
2967 | 7666 |
1168 | 7667 if( aspect_ratio_info_present_flag ) { |
7668 aspect_ratio_idc= get_bits(&s->gb, 8); | |
7669 if( aspect_ratio_idc == EXTENDED_SAR ) { | |
1548 | 7670 sps->sar.num= get_bits(&s->gb, 16); |
7671 sps->sar.den= get_bits(&s->gb, 16); | |
3196
15157293beea
fixing coverity warnig CID: 253 (reading over the end of the aspect_ratio array for illegal aspect ratios)
michael
parents:
3178
diff
changeset
|
7672 }else if(aspect_ratio_idc < 14){ |
1548 | 7673 sps->sar= pixel_aspect[aspect_ratio_idc]; |
1168 | 7674 }else{ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7675 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n"); |
1168 | 7676 return -1; |
7677 } | |
7678 }else{ | |
2967 | 7679 sps->sar.num= |
1548 | 7680 sps->sar.den= 0; |
1168 | 7681 } |
7682 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height); | |
2174
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7683 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7684 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7685 get_bits1(&s->gb); /* overscan_appropriate_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7686 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7687 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7688 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7689 get_bits(&s->gb, 3); /* video_format */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7690 get_bits1(&s->gb); /* video_full_range_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7691 if(get_bits1(&s->gb)){ /* colour_description_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7692 get_bits(&s->gb, 8); /* colour_primaries */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7693 get_bits(&s->gb, 8); /* transfer_characteristics */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7694 get_bits(&s->gb, 8); /* matrix_coefficients */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7695 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7696 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7697 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7698 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7699 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7700 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7701 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7702 |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7703 sps->timing_info_present_flag = get_bits1(&s->gb); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7704 if(sps->timing_info_present_flag){ |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7705 sps->num_units_in_tick = get_bits_long(&s->gb, 32); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7706 sps->time_scale = get_bits_long(&s->gb, 32); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7707 sps->fixed_frame_rate_flag = get_bits1(&s->gb); |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7708 } |
6d614374d907
Get H.264 frame rate from SPS/VUI patch by (Mns Rullgrd <mru at kth dot se>)
michael
parents:
2163
diff
changeset
|
7709 |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7710 nal_hrd_parameters_present_flag = get_bits1(&s->gb); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7711 if(nal_hrd_parameters_present_flag) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7712 decode_hrd_parameters(h, sps); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7713 vcl_hrd_parameters_present_flag = get_bits1(&s->gb); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7714 if(vcl_hrd_parameters_present_flag) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7715 decode_hrd_parameters(h, sps); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7716 if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7717 get_bits1(&s->gb); /* low_delay_hrd_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7718 get_bits1(&s->gb); /* pic_struct_present_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7719 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7720 sps->bitstream_restriction_flag = get_bits1(&s->gb); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7721 if(sps->bitstream_restriction_flag){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7722 get_bits1(&s->gb); /* motion_vectors_over_pic_boundaries_flag */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7723 get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7724 get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7725 get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7726 get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7727 sps->num_reorder_frames = get_ue_golomb(&s->gb); |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7728 get_ue_golomb(&s->gb); /* max_dec_frame_buffering */ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7729 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
7730 |
1168 | 7731 return 0; |
7732 } | |
7733 | |
3004 | 7734 static void decode_scaling_list(H264Context *h, uint8_t *factors, int size, |
7735 const uint8_t *jvt_list, const uint8_t *fallback_list){ | |
2919 | 7736 MpegEncContext * const s = &h->s; |
7737 int i, last = 8, next = 8; | |
7738 const uint8_t *scan = size == 16 ? zigzag_scan : zigzag_scan8x8; | |
3004 | 7739 if(!get_bits1(&s->gb)) /* matrix not written, we use the predicted one */ |
7740 memcpy(factors, fallback_list, size*sizeof(uint8_t)); | |
2919 | 7741 else |
7742 for(i=0;i<size;i++){ | |
7743 if(next) | |
7744 next = (last + get_se_golomb(&s->gb)) & 0xff; | |
3004 | 7745 if(!i && !next){ /* matrix not written, we use the preset one */ |
7746 memcpy(factors, jvt_list, size*sizeof(uint8_t)); | |
2919 | 7747 break; |
7748 } | |
7749 last = factors[scan[i]] = next ? next : last; | |
7750 } | |
7751 } | |
7752 | |
7753 static void decode_scaling_matrices(H264Context *h, SPS *sps, PPS *pps, int is_sps, | |
7754 uint8_t (*scaling_matrix4)[16], uint8_t (*scaling_matrix8)[64]){ | |
7755 MpegEncContext * const s = &h->s; | |
7756 int fallback_sps = !is_sps && sps->scaling_matrix_present; | |
7757 const uint8_t *fallback[4] = { | |
7758 fallback_sps ? sps->scaling_matrix4[0] : default_scaling4[0], | |
7759 fallback_sps ? sps->scaling_matrix4[3] : default_scaling4[1], | |
7760 fallback_sps ? sps->scaling_matrix8[0] : default_scaling8[0], | |
7761 fallback_sps ? sps->scaling_matrix8[1] : default_scaling8[1] | |
7762 }; | |
7763 if(get_bits1(&s->gb)){ | |
7764 sps->scaling_matrix_present |= is_sps; | |
3004 | 7765 decode_scaling_list(h,scaling_matrix4[0],16,default_scaling4[0],fallback[0]); // Intra, Y |
7766 decode_scaling_list(h,scaling_matrix4[1],16,default_scaling4[0],scaling_matrix4[0]); // Intra, Cr | |
7767 decode_scaling_list(h,scaling_matrix4[2],16,default_scaling4[0],scaling_matrix4[1]); // Intra, Cb | |
7768 decode_scaling_list(h,scaling_matrix4[3],16,default_scaling4[1],fallback[1]); // Inter, Y | |
7769 decode_scaling_list(h,scaling_matrix4[4],16,default_scaling4[1],scaling_matrix4[3]); // Inter, Cr | |
7770 decode_scaling_list(h,scaling_matrix4[5],16,default_scaling4[1],scaling_matrix4[4]); // Inter, Cb | |
2919 | 7771 if(is_sps || pps->transform_8x8_mode){ |
3004 | 7772 decode_scaling_list(h,scaling_matrix8[0],64,default_scaling8[0],fallback[2]); // Intra, Y |
7773 decode_scaling_list(h,scaling_matrix8[1],64,default_scaling8[1],fallback[3]); // Inter, Y | |
2919 | 7774 } |
7775 } else if(fallback_sps) { | |
7776 memcpy(scaling_matrix4, sps->scaling_matrix4, 6*16*sizeof(uint8_t)); | |
7777 memcpy(scaling_matrix8, sps->scaling_matrix8, 2*64*sizeof(uint8_t)); | |
7778 } | |
7779 } | |
7780 | |
1168 | 7781 static inline int decode_seq_parameter_set(H264Context *h){ |
7782 MpegEncContext * const s = &h->s; | |
1371 | 7783 int profile_idc, level_idc; |
1168 | 7784 int sps_id, i; |
7785 SPS *sps; | |
2967 | 7786 |
1168 | 7787 profile_idc= get_bits(&s->gb, 8); |
1371 | 7788 get_bits1(&s->gb); //constraint_set0_flag |
7789 get_bits1(&s->gb); //constraint_set1_flag | |
7790 get_bits1(&s->gb); //constraint_set2_flag | |
2312 | 7791 get_bits1(&s->gb); //constraint_set3_flag |
7792 get_bits(&s->gb, 4); // reserved | |
1168 | 7793 level_idc= get_bits(&s->gb, 8); |
7794 sps_id= get_ue_golomb(&s->gb); | |
2967 | 7795 |
1168 | 7796 sps= &h->sps_buffer[ sps_id ]; |
7797 sps->profile_idc= profile_idc; | |
7798 sps->level_idc= level_idc; | |
2312 | 7799 |
2755 | 7800 if(sps->profile_idc >= 100){ //high profile |
7801 if(get_ue_golomb(&s->gb) == 3) //chroma_format_idc | |
7802 get_bits1(&s->gb); //residual_color_transform_flag | |
7803 get_ue_golomb(&s->gb); //bit_depth_luma_minus8 | |
7804 get_ue_golomb(&s->gb); //bit_depth_chroma_minus8 | |
2763 | 7805 sps->transform_bypass = get_bits1(&s->gb); |
2919 | 7806 decode_scaling_matrices(h, sps, NULL, 1, sps->scaling_matrix4, sps->scaling_matrix8); |
7807 }else | |
7808 sps->scaling_matrix_present = 0; | |
2755 | 7809 |
1168 | 7810 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4; |
7811 sps->poc_type= get_ue_golomb(&s->gb); | |
2967 | 7812 |
1168 | 7813 if(sps->poc_type == 0){ //FIXME #define |
7814 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4; | |
7815 } else if(sps->poc_type == 1){//FIXME #define | |
7816 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb); | |
7817 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb); | |
7818 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb); | |
7819 sps->poc_cycle_length= get_ue_golomb(&s->gb); | |
2967 | 7820 |
1168 | 7821 for(i=0; i<sps->poc_cycle_length; i++) |
7822 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb); | |
7823 } | |
7824 if(sps->poc_type > 2){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7825 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type); |
1168 | 7826 return -1; |
7827 } | |
7828 | |
7829 sps->ref_frame_count= get_ue_golomb(&s->gb); | |
2254
0dfe4e32b19c
H.264 max reference pictures fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2227
diff
changeset
|
7830 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){ |
0dfe4e32b19c
H.264 max reference pictures fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2227
diff
changeset
|
7831 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n"); |
0dfe4e32b19c
H.264 max reference pictures fix by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2227
diff
changeset
|
7832 } |
1371 | 7833 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb); |
1168 | 7834 sps->mb_width= get_ue_golomb(&s->gb) + 1; |
7835 sps->mb_height= get_ue_golomb(&s->gb) + 1; | |
2967 | 7836 if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 || |
2422 | 7837 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height)) |
7838 return -1; | |
7839 | |
1168 | 7840 sps->frame_mbs_only_flag= get_bits1(&s->gb); |
7841 if(!sps->frame_mbs_only_flag) | |
7842 sps->mb_aff= get_bits1(&s->gb); | |
7843 else | |
7844 sps->mb_aff= 0; | |
7845 | |
7846 sps->direct_8x8_inference_flag= get_bits1(&s->gb); | |
7847 | |
3316 | 7848 #ifndef ALLOW_INTERLACE |
7849 if(sps->mb_aff) | |
3954 | 7850 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF support not included; enable it at compile-time.\n"); |
3316 | 7851 #endif |
7852 if(!sps->direct_8x8_inference_flag && sps->mb_aff) | |
7853 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + !direct_8x8_inference is not implemented\n"); | |
7854 | |
1371 | 7855 sps->crop= get_bits1(&s->gb); |
7856 if(sps->crop){ | |
7857 sps->crop_left = get_ue_golomb(&s->gb); | |
7858 sps->crop_right = get_ue_golomb(&s->gb); | |
7859 sps->crop_top = get_ue_golomb(&s->gb); | |
7860 sps->crop_bottom= get_ue_golomb(&s->gb); | |
7861 if(sps->crop_left || sps->crop_top){ | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
7862 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completely supported, this could look slightly wrong ...\n"); |
1371 | 7863 } |
7864 }else{ | |
2967 | 7865 sps->crop_left = |
7866 sps->crop_right = | |
7867 sps->crop_top = | |
1371 | 7868 sps->crop_bottom= 0; |
7869 } | |
7870 | |
1168 | 7871 sps->vui_parameters_present_flag= get_bits1(&s->gb); |
7872 if( sps->vui_parameters_present_flag ) | |
7873 decode_vui_parameters(h, sps); | |
2967 | 7874 |
1168 | 7875 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ |
2967 | 7876 av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n", |
1168 | 7877 sps_id, sps->profile_idc, sps->level_idc, |
7878 sps->poc_type, | |
7879 sps->ref_frame_count, | |
7880 sps->mb_width, sps->mb_height, | |
7881 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"), | |
7882 sps->direct_8x8_inference_flag ? "8B8" : "", | |
2967 | 7883 sps->crop_left, sps->crop_right, |
7884 sps->crop_top, sps->crop_bottom, | |
1168 | 7885 sps->vui_parameters_present_flag ? "VUI" : "" |
7886 ); | |
7887 } | |
7888 return 0; | |
7889 } | |
7890 | |
2755 | 7891 static inline int decode_picture_parameter_set(H264Context *h, int bit_length){ |
1168 | 7892 MpegEncContext * const s = &h->s; |
7893 int pps_id= get_ue_golomb(&s->gb); | |
7894 PPS *pps= &h->pps_buffer[pps_id]; | |
2967 | 7895 |
1168 | 7896 pps->sps_id= get_ue_golomb(&s->gb); |
7897 pps->cabac= get_bits1(&s->gb); | |
7898 pps->pic_order_present= get_bits1(&s->gb); | |
7899 pps->slice_group_count= get_ue_golomb(&s->gb) + 1; | |
7900 if(pps->slice_group_count > 1 ){ | |
7901 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb); | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7902 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n"); |
1168 | 7903 switch(pps->mb_slice_group_map_type){ |
7904 case 0: | |
7905 #if 0 | |
7906 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | | | |
7907 | run_length[ i ] |1 |ue(v) | | |
7908 #endif | |
7909 break; | |
7910 case 2: | |
7911 #if 0 | |
7912 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | | | |
7913 |{ | | | | |
7914 | top_left_mb[ i ] |1 |ue(v) | | |
7915 | bottom_right_mb[ i ] |1 |ue(v) | | |
7916 | } | | | | |
7917 #endif | |
7918 break; | |
7919 case 3: | |
7920 case 4: | |
7921 case 5: | |
7922 #if 0 | |
7923 | slice_group_change_direction_flag |1 |u(1) | | |
7924 | slice_group_change_rate_minus1 |1 |ue(v) | | |
7925 #endif | |
7926 break; | |
7927 case 6: | |
7928 #if 0 | |
7929 | slice_group_id_cnt_minus1 |1 |ue(v) | | |
7930 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | | | |
7931 |) | | | | |
7932 | slice_group_id[ i ] |1 |u(v) | | |
7933 #endif | |
1214 | 7934 break; |
1168 | 7935 } |
7936 } | |
7937 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1; | |
7938 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1; | |
7939 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){ | |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
7940 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n"); |
1168 | 7941 return -1; |
7942 } | |
2967 | 7943 |
1168 | 7944 pps->weighted_pred= get_bits1(&s->gb); |
7945 pps->weighted_bipred_idc= get_bits(&s->gb, 2); | |
7946 pps->init_qp= get_se_golomb(&s->gb) + 26; | |
7947 pps->init_qs= get_se_golomb(&s->gb) + 26; | |
7948 pps->chroma_qp_index_offset= get_se_golomb(&s->gb); | |
7949 pps->deblocking_filter_parameters_present= get_bits1(&s->gb); | |
7950 pps->constrained_intra_pred= get_bits1(&s->gb); | |
7951 pps->redundant_pic_cnt_present = get_bits1(&s->gb); | |
2967 | 7952 |
3291
454de57e45cf
reinit quant matrices if pps is overwritten by another of the same pps_id.
lorenm
parents:
3285
diff
changeset
|
7953 pps->transform_8x8_mode= 0; |
454de57e45cf
reinit quant matrices if pps is overwritten by another of the same pps_id.
lorenm
parents:
3285
diff
changeset
|
7954 h->dequant_coeff_pps= -1; //contents of sps/pps can change even if id doesn't, so reinit |
2919 | 7955 memset(pps->scaling_matrix4, 16, 6*16*sizeof(uint8_t)); |
7956 memset(pps->scaling_matrix8, 16, 2*64*sizeof(uint8_t)); | |
2755 | 7957 |
7958 if(get_bits_count(&s->gb) < bit_length){ | |
7959 pps->transform_8x8_mode= get_bits1(&s->gb); | |
2919 | 7960 decode_scaling_matrices(h, &h->sps_buffer[pps->sps_id], pps, 0, pps->scaling_matrix4, pps->scaling_matrix8); |
2755 | 7961 get_se_golomb(&s->gb); //second_chroma_qp_index_offset |
7962 } | |
2967 | 7963 |
1168 | 7964 if(s->avctx->debug&FF_DEBUG_PICT_INFO){ |
2967 | 7965 av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s %s\n", |
1168 | 7966 pps_id, pps->sps_id, |
7967 pps->cabac ? "CABAC" : "CAVLC", | |
7968 pps->slice_group_count, | |
7969 pps->ref_count[0], pps->ref_count[1], | |
7970 pps->weighted_pred ? "weighted" : "", | |
7971 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset, | |
7972 pps->deblocking_filter_parameters_present ? "LPAR" : "", | |
7973 pps->constrained_intra_pred ? "CONSTR" : "", | |
2755 | 7974 pps->redundant_pic_cnt_present ? "REDU" : "", |
7975 pps->transform_8x8_mode ? "8x8DCT" : "" | |
1168 | 7976 ); |
7977 } | |
2967 | 7978 |
1168 | 7979 return 0; |
7980 } | |
7981 | |
7982 /** | |
7983 * finds the end of the current frame in the bitstream. | |
7984 * @return the position of the first byte of the next frame, or -1 | |
7985 */ | |
2392 | 7986 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){ |
1187 | 7987 int i; |
1168 | 7988 uint32_t state; |
2392 | 7989 ParseContext *pc = &(h->s.parse_context); |
1168 | 7990 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]); |
7991 // mb_addr= pc->mb_addr - 1; | |
7992 state= pc->state; | |
2392 | 7993 for(i=0; i<=buf_size; i++){ |
1168 | 7994 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){ |
2392 | 7995 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i); |
1168 | 7996 if(pc->frame_start_found){ |
2392 | 7997 // If there isn't one more byte in the buffer |
7998 // the test on first_mb_in_slice cannot be done yet | |
7999 // do it at next call. | |
8000 if (i >= buf_size) break; | |
8001 if (buf[i] & 0x80) { | |
8002 // first_mb_in_slice is 0, probably the first nal of a new | |
8003 // slice | |
8004 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i); | |
2967 | 8005 pc->state=-1; |
2392 | 8006 pc->frame_start_found= 0; |
8007 return i-4; | |
8008 } | |
1168 | 8009 } |
2392 | 8010 pc->frame_start_found = 1; |
1168 | 8011 } |
2774
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8012 if((state&0xFFFFFF1F) == 0x107 || (state&0xFFFFFF1F) == 0x108 || (state&0xFFFFFF1F) == 0x109){ |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8013 if(pc->frame_start_found){ |
2967 | 8014 pc->state=-1; |
2774
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8015 pc->frame_start_found= 0; |
2967 | 8016 return i-4; |
2774
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8017 } |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8018 } |
2392 | 8019 if (i<buf_size) |
8020 state= (state<<8) | buf[i]; | |
1168 | 8021 } |
2967 | 8022 |
1168 | 8023 pc->state= state; |
1219 | 8024 return END_NOT_FOUND; |
1168 | 8025 } |
8026 | |
3455
cc4b4ea83e29
--enable/disable parsers. Warning: some combinations are broken.
mru
parents:
3409
diff
changeset
|
8027 #ifdef CONFIG_H264_PARSER |
1988 | 8028 static int h264_parse(AVCodecParserContext *s, |
8029 AVCodecContext *avctx, | |
2967 | 8030 uint8_t **poutbuf, int *poutbuf_size, |
1988 | 8031 const uint8_t *buf, int buf_size) |
8032 { | |
2392 | 8033 H264Context *h = s->priv_data; |
8034 ParseContext *pc = &h->s.parse_context; | |
1988 | 8035 int next; |
2967 | 8036 |
2392 | 8037 next= find_frame_end(h, buf, buf_size); |
1988 | 8038 |
8039 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) { | |
8040 *poutbuf = NULL; | |
8041 *poutbuf_size = 0; | |
8042 return buf_size; | |
8043 } | |
8044 | |
8045 *poutbuf = (uint8_t *)buf; | |
8046 *poutbuf_size = buf_size; | |
8047 return next; | |
8048 } | |
8049 | |
2774
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8050 static int h264_split(AVCodecContext *avctx, |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8051 const uint8_t *buf, int buf_size) |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8052 { |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8053 int i; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8054 uint32_t state = -1; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8055 int has_sps= 0; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8056 |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8057 for(i=0; i<=buf_size; i++){ |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8058 if((state&0xFFFFFF1F) == 0x107) |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8059 has_sps=1; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8060 /* if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){ |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8061 }*/ |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8062 if((state&0xFFFFFF00) == 0x100 && (state&0xFFFFFF1F) != 0x107 && (state&0xFFFFFF1F) != 0x108 && (state&0xFFFFFF1F) != 0x109){ |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8063 if(has_sps){ |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8064 while(i>4 && buf[i-5]==0) i--; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8065 return i-4; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8066 } |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8067 } |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8068 if (i<buf_size) |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8069 state= (state<<8) | buf[i]; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8070 } |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8071 return 0; |
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8072 } |
3455
cc4b4ea83e29
--enable/disable parsers. Warning: some combinations are broken.
mru
parents:
3409
diff
changeset
|
8073 #endif /* CONFIG_H264_PARSER */ |
2774
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8074 |
1168 | 8075 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){ |
8076 MpegEncContext * const s = &h->s; | |
8077 AVCodecContext * const avctx= s->avctx; | |
8078 int buf_index=0; | |
1322 | 8079 #if 0 |
1168 | 8080 int i; |
2867
7c7edddaa835
dont be so picky with .mp4 ... fixes Mr&MrsSmith.mp4
michael
parents:
2861
diff
changeset
|
8081 for(i=0; i<50; i++){ |
7c7edddaa835
dont be so picky with .mp4 ... fixes Mr&MrsSmith.mp4
michael
parents:
2861
diff
changeset
|
8082 av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]); |
1168 | 8083 } |
8084 #endif | |
2392 | 8085 h->slice_num = 0; |
2766
0609f405e000
make decoder a little bit more tolerant to missing NAL units
michael
parents:
2763
diff
changeset
|
8086 s->current_picture_ptr= NULL; |
1168 | 8087 for(;;){ |
8088 int consumed; | |
8089 int dst_length; | |
8090 int bit_length; | |
8091 uint8_t *ptr; | |
2227 | 8092 int i, nalsize = 0; |
2967 | 8093 |
2227 | 8094 if(h->is_avc) { |
8095 if(buf_index >= buf_size) break; | |
8096 nalsize = 0; | |
8097 for(i = 0; i < h->nal_length_size; i++) | |
8098 nalsize = (nalsize << 8) | buf[buf_index++]; | |
3126 | 8099 if(nalsize <= 1){ |
8100 if(nalsize == 1){ | |
8101 buf_index++; | |
8102 continue; | |
8103 }else{ | |
8104 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize); | |
8105 break; | |
8106 } | |
8107 } | |
2227 | 8108 } else { |
1168 | 8109 // start code prefix search |
8110 for(; buf_index + 3 < buf_size; buf_index++){ | |
8111 // this should allways succeed in the first iteration | |
8112 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1) | |
8113 break; | |
8114 } | |
2967 | 8115 |
1168 | 8116 if(buf_index+3 >= buf_size) break; |
2967 | 8117 |
1168 | 8118 buf_index+=3; |
2967 | 8119 } |
8120 | |
2401
46898a9fd6dc
Fix avc1 if there is nore than one nal per mov frame
rtognimp
parents:
2396
diff
changeset
|
8121 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index); |
3341 | 8122 while(ptr[dst_length - 1] == 0 && dst_length > 1) |
8123 dst_length--; | |
1168 | 8124 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1); |
8125 | |
8126 if(s->avctx->debug&FF_DEBUG_STARTCODE){ | |
2402
f9d4e1eddbc5
- correct several errors on the deblocking accross slice boundaries.
michael
parents:
2401
diff
changeset
|
8127 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", h->nal_unit_type, buf_index, buf_size, dst_length); |
1168 | 8128 } |
2967 | 8129 |
2227 | 8130 if (h->is_avc && (nalsize != consumed)) |
8131 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize); | |
8132 | |
1168 | 8133 buf_index += consumed; |
8134 | |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
8135 if( (s->hurry_up == 1 && h->nal_ref_idc == 0) //FIXME dont discard SEI id |
2792 | 8136 ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)) |
1168 | 8137 continue; |
2967 | 8138 |
1168 | 8139 switch(h->nal_unit_type){ |
8140 case NAL_IDR_SLICE: | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
8141 idr(h); //FIXME ensure we don't loose some frames if there is reordering |
1168 | 8142 case NAL_SLICE: |
8143 init_get_bits(&s->gb, ptr, bit_length); | |
8144 h->intra_gb_ptr= | |
8145 h->inter_gb_ptr= &s->gb; | |
8146 s->data_partitioning = 0; | |
2967 | 8147 |
2771 | 8148 if(decode_slice_header(h) < 0){ |
8149 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n"); | |
8150 break; | |
8151 } | |
3161
e3e94632c6e9
keyframe flag was wrong if SEIs followed the keyframe
lorenm
parents:
3145
diff
changeset
|
8152 s->current_picture_ptr->key_frame= (h->nal_unit_type == NAL_IDR_SLICE); |
2967 | 8153 if(h->redundant_pic_count==0 && s->hurry_up < 5 |
2793 | 8154 && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc) |
8155 && (avctx->skip_frame < AVDISCARD_BIDIR || h->slice_type!=B_TYPE) | |
8156 && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE) | |
8157 && avctx->skip_frame < AVDISCARD_ALL) | |
1168 | 8158 decode_slice(h); |
8159 break; | |
8160 case NAL_DPA: | |
8161 init_get_bits(&s->gb, ptr, bit_length); | |
8162 h->intra_gb_ptr= | |
8163 h->inter_gb_ptr= NULL; | |
8164 s->data_partitioning = 1; | |
2967 | 8165 |
2771 | 8166 if(decode_slice_header(h) < 0){ |
8167 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n"); | |
8168 } | |
1168 | 8169 break; |
8170 case NAL_DPB: | |
8171 init_get_bits(&h->intra_gb, ptr, bit_length); | |
8172 h->intra_gb_ptr= &h->intra_gb; | |
8173 break; | |
8174 case NAL_DPC: | |
8175 init_get_bits(&h->inter_gb, ptr, bit_length); | |
8176 h->inter_gb_ptr= &h->inter_gb; | |
1174 | 8177 |
2967 | 8178 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning |
2793 | 8179 && s->hurry_up < 5 |
8180 && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc) | |
8181 && (avctx->skip_frame < AVDISCARD_BIDIR || h->slice_type!=B_TYPE) | |
8182 && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE) | |
8183 && avctx->skip_frame < AVDISCARD_ALL) | |
1168 | 8184 decode_slice(h); |
8185 break; | |
8186 case NAL_SEI: | |
2815
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
8187 init_get_bits(&s->gb, ptr, bit_length); |
636133fccbdc
workaround 'colocated mv if colocated block is L1 predicted' bug in x264
michael
parents:
2809
diff
changeset
|
8188 decode_sei(h); |
1168 | 8189 break; |
8190 case NAL_SPS: | |
8191 init_get_bits(&s->gb, ptr, bit_length); | |
8192 decode_seq_parameter_set(h); | |
2967 | 8193 |
1168 | 8194 if(s->flags& CODEC_FLAG_LOW_DELAY) |
8195 s->low_delay=1; | |
2967 | 8196 |
2538 | 8197 if(avctx->has_b_frames < 2) |
8198 avctx->has_b_frames= !s->low_delay; | |
1168 | 8199 break; |
8200 case NAL_PPS: | |
8201 init_get_bits(&s->gb, ptr, bit_length); | |
2967 | 8202 |
2755 | 8203 decode_picture_parameter_set(h, bit_length); |
1168 | 8204 |
8205 break; | |
2960 | 8206 case NAL_AUD: |
8207 case NAL_END_SEQUENCE: | |
8208 case NAL_END_STREAM: | |
8209 case NAL_FILLER_DATA: | |
8210 case NAL_SPS_EXT: | |
8211 case NAL_AUXILIARY_SLICE: | |
1168 | 8212 break; |
2979 | 8213 default: |
8214 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type); | |
2967 | 8215 } |
8216 } | |
8217 | |
1174 | 8218 if(!s->current_picture_ptr) return buf_index; //no frame |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8219 |
2971 | 8220 s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8221 s->current_picture_ptr->pict_type= s->pict_type; |
2967 | 8222 |
1168 | 8223 h->prev_frame_num_offset= h->frame_num_offset; |
8224 h->prev_frame_num= h->frame_num; | |
8225 if(s->current_picture_ptr->reference){ | |
8226 h->prev_poc_msb= h->poc_msb; | |
8227 h->prev_poc_lsb= h->poc_lsb; | |
8228 } | |
8229 if(s->current_picture_ptr->reference) | |
8230 execute_ref_pic_marking(h, h->mmco, h->mmco_index); | |
8231 | |
8232 ff_er_frame_end(s); | |
1897
4e8ed93524f6
h264 loop filter for progressive I&P frames by (Laurent Aimar <fenrir at via dot ecp dot fr>)
michael
parents:
1892
diff
changeset
|
8233 |
1168 | 8234 MPV_frame_end(s); |
8235 | |
8236 return buf_index; | |
8237 } | |
8238 | |
8239 /** | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
8240 * returns the number of bytes consumed for building the current frame |
1168 | 8241 */ |
8242 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){ | |
8243 if(s->flags&CODEC_FLAG_TRUNCATED){ | |
8244 pos -= s->parse_context.last_index; | |
2645
42528c1f0246
More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)
michael
parents:
2641
diff
changeset
|
8245 if(pos<0) pos=0; // FIXME remove (unneeded?) |
2967 | 8246 |
1168 | 8247 return pos; |
8248 }else{ | |
8249 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...) | |
8250 if(pos+10>buf_size) pos=buf_size; // oops ;) | |
8251 | |
8252 return pos; | |
8253 } | |
8254 } | |
8255 | |
2967 | 8256 static int decode_frame(AVCodecContext *avctx, |
1168 | 8257 void *data, int *data_size, |
8258 uint8_t *buf, int buf_size) | |
8259 { | |
8260 H264Context *h = avctx->priv_data; | |
8261 MpegEncContext *s = &h->s; | |
2967 | 8262 AVFrame *pict = data; |
1168 | 8263 int buf_index; |
2967 | 8264 |
1168 | 8265 s->flags= avctx->flags; |
1754
bdf3927bf8c5
closed gop support & flags2 as all bits in flags are used
michael
parents:
1706
diff
changeset
|
8266 s->flags2= avctx->flags2; |
1168 | 8267 |
8268 /* no supplementary picture */ | |
8269 if (buf_size == 0) { | |
8270 return 0; | |
8271 } | |
2967 | 8272 |
1168 | 8273 if(s->flags&CODEC_FLAG_TRUNCATED){ |
2392 | 8274 int next= find_frame_end(h, buf, buf_size); |
2967 | 8275 |
1988 | 8276 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 ) |
1168 | 8277 return buf_size; |
8278 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index); | |
8279 } | |
8280 | |
2227 | 8281 if(h->is_avc && !h->got_avcC) { |
8282 int i, cnt, nalsize; | |
8283 unsigned char *p = avctx->extradata; | |
8284 if(avctx->extradata_size < 7) { | |
8285 av_log(avctx, AV_LOG_ERROR, "avcC too short\n"); | |
8286 return -1; | |
8287 } | |
8288 if(*p != 1) { | |
8289 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p); | |
8290 return -1; | |
8291 } | |
8292 /* sps and pps in the avcC always have length coded with 2 bytes, | |
8293 so put a fake nal_length_size = 2 while parsing them */ | |
8294 h->nal_length_size = 2; | |
8295 // Decode sps from avcC | |
8296 cnt = *(p+5) & 0x1f; // Number of sps | |
8297 p += 6; | |
8298 for (i = 0; i < cnt; i++) { | |
8299 nalsize = BE_16(p) + 2; | |
2867
7c7edddaa835
dont be so picky with .mp4 ... fixes Mr&MrsSmith.mp4
michael
parents:
2861
diff
changeset
|
8300 if(decode_nal_units(h, p, nalsize) < 0) { |
2227 | 8301 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i); |
8302 return -1; | |
8303 } | |
8304 p += nalsize; | |
2967 | 8305 } |
2227 | 8306 // Decode pps from avcC |
8307 cnt = *(p++); // Number of pps | |
8308 for (i = 0; i < cnt; i++) { | |
8309 nalsize = BE_16(p) + 2; | |
8310 if(decode_nal_units(h, p, nalsize) != nalsize) { | |
8311 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i); | |
8312 return -1; | |
8313 } | |
8314 p += nalsize; | |
2967 | 8315 } |
2227 | 8316 // Now store right nal length size, that will be use to parse all other nals |
8317 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1; | |
8318 // Do not reparse avcC | |
8319 h->got_avcC = 1; | |
8320 } | |
8321 | |
8322 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){ | |
2967 | 8323 if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0) |
1168 | 8324 return -1; |
8325 } | |
8326 | |
8327 buf_index=decode_nal_units(h, buf, buf_size); | |
2967 | 8328 if(buf_index < 0) |
1168 | 8329 return -1; |
8330 | |
2967 | 8331 //FIXME do something with unavailable reference frames |
8332 | |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2623
diff
changeset
|
8333 // if(ret==FRAME_SKIPPED) return get_consumed_bytes(s, buf_index, buf_size); |
1174 | 8334 if(!s->current_picture_ptr){ |
1598
932d306bf1dc
av_log() patch by (Michel Bardiaux <mbardiaux at peaktime dot be>)
michael
parents:
1548
diff
changeset
|
8335 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n"); |
1174 | 8336 return -1; |
8337 } | |
8338 | |
2409 | 8339 { |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
8340 Picture *out = s->current_picture_ptr; |
2561 | 8341 #if 0 //decode order |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8342 *data_size = sizeof(AVFrame); |
2561 | 8343 #else |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8344 /* Sort B-frames into display order */ |
2409 | 8345 Picture *cur = s->current_picture_ptr; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8346 Picture *prev = h->delayed_output_pic; |
3298
9637da0a9c1b
cosmetics (initalize variables where they are used)
michael
parents:
3297
diff
changeset
|
8347 int i, pics, cross_idr, out_of_order, out_idx; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8348 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8349 if(h->sps.bitstream_restriction_flag |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8350 && s->avctx->has_b_frames < h->sps.num_reorder_frames){ |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8351 s->avctx->has_b_frames = h->sps.num_reorder_frames; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8352 s->low_delay = 0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8353 } |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8354 |
3298
9637da0a9c1b
cosmetics (initalize variables where they are used)
michael
parents:
3297
diff
changeset
|
8355 pics = 0; |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8356 while(h->delayed_pic[pics]) pics++; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8357 h->delayed_pic[pics++] = cur; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8358 if(cur->reference == 0) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8359 cur->reference = 1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8360 |
3298
9637da0a9c1b
cosmetics (initalize variables where they are used)
michael
parents:
3297
diff
changeset
|
8361 cross_idr = 0; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8362 for(i=0; h->delayed_pic[i]; i++) |
2649
b2e6c5e9240b
fix decoding of conformace streams AUD_MW_E.264 and BA1_FT_C.264 again (there where last correctly decoded with h264.c rev 1.96)
michael
parents:
2648
diff
changeset
|
8363 if(h->delayed_pic[i]->key_frame || h->delayed_pic[i]->poc==0) |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8364 cross_idr = 1; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8365 |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8366 out = h->delayed_pic[0]; |
3298
9637da0a9c1b
cosmetics (initalize variables where they are used)
michael
parents:
3297
diff
changeset
|
8367 out_idx = 0; |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8368 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8369 if(h->delayed_pic[i]->poc < out->poc){ |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8370 out = h->delayed_pic[i]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8371 out_idx = i; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8372 } |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8373 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8374 out_of_order = !cross_idr && prev && out->poc < prev->poc; |
3125 | 8375 if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames) |
8376 { } | |
8377 else if(prev && pics <= s->avctx->has_b_frames) | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8378 out = prev; |
2923 | 8379 else if((out_of_order && pics-1 == s->avctx->has_b_frames && pics < 15) |
2967 | 8380 || (s->low_delay && |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8381 ((!cross_idr && prev && out->poc > prev->poc + 2) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8382 || cur->pict_type == B_TYPE))) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8383 { |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8384 s->low_delay = 0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8385 s->avctx->has_b_frames++; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8386 out = prev; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8387 } |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8388 else if(out_of_order) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8389 out = prev; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8390 |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8391 if(out_of_order || pics > s->avctx->has_b_frames){ |
2537
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8392 for(i=out_idx; h->delayed_pic[i]; i++) |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8393 h->delayed_pic[i] = h->delayed_pic[i+1]; |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8394 } |
14fef0f3f532
H.264: decode arbitrary frame orders and allow B-frames as references.
lorenm
parents:
2536
diff
changeset
|
8395 |
3297
9ae9653d706c
decoders should (IMHO) not duplicate frames, the application can do this if it wants
michael
parents:
3291
diff
changeset
|
8396 if(prev == out) |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8397 *data_size = 0; |
2561 | 8398 else |
8399 *data_size = sizeof(AVFrame); | |
2560
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8400 if(prev && prev != out && prev->reference == 1) |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8401 prev->reference = 0; |
bfba825ee300
Set keyframe flag only on IDR-frames (needed for reordering across I-frames).
lorenm
parents:
2553
diff
changeset
|
8402 h->delayed_output_pic = out; |
2441
358813ec4ca2
H.264 b ref pic list order and long term pictures patch by (Loic Le Loarer <loic.le-loarer polytechnique org>)
michael
parents:
2440
diff
changeset
|
8403 #endif |
2409 | 8404 |
2937 | 8405 if(out) |
8406 *pict= *(AVFrame*)out; | |
8407 else | |
8408 av_log(avctx, AV_LOG_DEBUG, "no picture\n"); | |
8409 } | |
8410 | |
8411 assert(pict->data[0] || !*data_size); | |
1706
3ba5c493db6f
motion vector vissualization improvements patch by (Wolfgang Hesseler <qv at multimediaware dot com>)
michael
parents:
1698
diff
changeset
|
8412 ff_print_debug_info(s, pict); |
1168 | 8413 //printf("out %d\n", (int)pict->data[0]); |
8414 #if 0 //? | |
8415 | |
8416 /* Return the Picture timestamp as the frame number */ | |
8417 /* we substract 1 because it is added on utils.c */ | |
8418 avctx->frame_number = s->picture_number - 1; | |
8419 #endif | |
8420 return get_consumed_bytes(s, buf_index, buf_size); | |
8421 } | |
8422 #if 0 | |
8423 static inline void fill_mb_avail(H264Context *h){ | |
8424 MpegEncContext * const s = &h->s; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8425 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride; |
1168 | 8426 |
8427 if(s->mb_y){ | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8428 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8429 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1174
diff
changeset
|
8430 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num; |
1168 | 8431 }else{ |
8432 h->mb_avail[0]= | |
8433 h->mb_avail[1]= | |
8434 h->mb_avail[2]= 0; | |
8435 } | |
8436 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num; | |
8437 h->mb_avail[4]= 1; //FIXME move out | |
8438 h->mb_avail[5]= 0; //FIXME move out | |
8439 } | |
8440 #endif | |
8441 | |
8442 #if 0 //selftest | |
8443 #define COUNT 8000 | |
8444 #define SIZE (COUNT*40) | |
8445 int main(){ | |
8446 int i; | |
8447 uint8_t temp[SIZE]; | |
8448 PutBitContext pb; | |
8449 GetBitContext gb; | |
8450 // int int_temp[10000]; | |
8451 DSPContext dsp; | |
8452 AVCodecContext avctx; | |
2967 | 8453 |
1168 | 8454 dsputil_init(&dsp, &avctx); |
8455 | |
1522
79dddc5cd990
removed the obsolete and unused parameters of init_put_bits
alex
parents:
1453
diff
changeset
|
8456 init_put_bits(&pb, temp, SIZE); |
1168 | 8457 printf("testing unsigned exp golomb\n"); |
8458 for(i=0; i<COUNT; i++){ | |
8459 START_TIMER | |
8460 set_ue_golomb(&pb, i); | |
8461 STOP_TIMER("set_ue_golomb"); | |
8462 } | |
8463 flush_put_bits(&pb); | |
2967 | 8464 |
1168 | 8465 init_get_bits(&gb, temp, 8*SIZE); |
8466 for(i=0; i<COUNT; i++){ | |
8467 int j, s; | |
2967 | 8468 |
1168 | 8469 s= show_bits(&gb, 24); |
2967 | 8470 |
1168 | 8471 START_TIMER |
8472 j= get_ue_golomb(&gb); | |
8473 if(j != i){ | |
8474 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
8475 // return -1; | |
8476 } | |
8477 STOP_TIMER("get_ue_golomb"); | |
8478 } | |
2967 | 8479 |
8480 | |
1524 | 8481 init_put_bits(&pb, temp, SIZE); |
1168 | 8482 printf("testing signed exp golomb\n"); |
8483 for(i=0; i<COUNT; i++){ | |
8484 START_TIMER | |
8485 set_se_golomb(&pb, i - COUNT/2); | |
8486 STOP_TIMER("set_se_golomb"); | |
8487 } | |
8488 flush_put_bits(&pb); | |
2967 | 8489 |
1168 | 8490 init_get_bits(&gb, temp, 8*SIZE); |
8491 for(i=0; i<COUNT; i++){ | |
8492 int j, s; | |
2967 | 8493 |
1168 | 8494 s= show_bits(&gb, 24); |
2967 | 8495 |
1168 | 8496 START_TIMER |
8497 j= get_se_golomb(&gb); | |
8498 if(j != i - COUNT/2){ | |
8499 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s); | |
8500 // return -1; | |
8501 } | |
8502 STOP_TIMER("get_se_golomb"); | |
8503 } | |
8504 | |
8505 printf("testing 4x4 (I)DCT\n"); | |
2967 | 8506 |
1168 | 8507 DCTELEM block[16]; |
8508 uint8_t src[16], ref[16]; | |
8509 uint64_t error= 0, max_error=0; | |
8510 | |
8511 for(i=0; i<COUNT; i++){ | |
8512 int j; | |
8513 // printf("%d %d %d\n", r1, r2, (r2-r1)*16); | |
8514 for(j=0; j<16; j++){ | |
8515 ref[j]= random()%255; | |
8516 src[j]= random()%255; | |
8517 } | |
8518 | |
8519 h264_diff_dct_c(block, src, ref, 4); | |
2967 | 8520 |
1168 | 8521 //normalize |
8522 for(j=0; j<16; j++){ | |
8523 // printf("%d ", block[j]); | |
8524 block[j]= block[j]*4; | |
8525 if(j&1) block[j]= (block[j]*4 + 2)/5; | |
8526 if(j&4) block[j]= (block[j]*4 + 2)/5; | |
8527 } | |
8528 // printf("\n"); | |
2967 | 8529 |
2272
cd43603c46f9
move h264 idct to its own file and call via function pointer in DspContext
michael
parents:
2255
diff
changeset
|
8530 s->dsp.h264_idct_add(ref, block, 4); |
1168 | 8531 /* for(j=0; j<16; j++){ |
8532 printf("%d ", ref[j]); | |
8533 } | |
8534 printf("\n");*/ | |
2967 | 8535 |
1168 | 8536 for(j=0; j<16; j++){ |
4001 | 8537 int diff= FFABS(src[j] - ref[j]); |
2967 | 8538 |
1168 | 8539 error+= diff*diff; |
8540 max_error= FFMAX(max_error, diff); | |
8541 } | |
8542 } | |
8543 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error ); | |
8544 #if 0 | |
8545 printf("testing quantizer\n"); | |
8546 for(qp=0; qp<52; qp++){ | |
8547 for(i=0; i<16; i++) | |
8548 src1_block[i]= src2_block[i]= random()%255; | |
2967 | 8549 |
1168 | 8550 } |
8551 #endif | |
8552 printf("Testing NAL layer\n"); | |
2967 | 8553 |
1168 | 8554 uint8_t bitstream[COUNT]; |
8555 uint8_t nal[COUNT*2]; | |
8556 H264Context h; | |
8557 memset(&h, 0, sizeof(H264Context)); | |
2967 | 8558 |
1168 | 8559 for(i=0; i<COUNT; i++){ |
8560 int zeros= i; | |
8561 int nal_length; | |
8562 int consumed; | |
8563 int out_length; | |
8564 uint8_t *out; | |
8565 int j; | |
2967 | 8566 |
1168 | 8567 for(j=0; j<COUNT; j++){ |
8568 bitstream[j]= (random() % 255) + 1; | |
8569 } | |
2967 | 8570 |
1168 | 8571 for(j=0; j<zeros; j++){ |
8572 int pos= random() % COUNT; | |
8573 while(bitstream[pos] == 0){ | |
8574 pos++; | |
8575 pos %= COUNT; | |
8576 } | |
8577 bitstream[pos]=0; | |
8578 } | |
2967 | 8579 |
1168 | 8580 START_TIMER |
2967 | 8581 |
1168 | 8582 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2); |
8583 if(nal_length<0){ | |
8584 printf("encoding failed\n"); | |
8585 return -1; | |
8586 } | |
2967 | 8587 |
1168 | 8588 out= decode_nal(&h, nal, &out_length, &consumed, nal_length); |
8589 | |
8590 STOP_TIMER("NAL") | |
2967 | 8591 |
1168 | 8592 if(out_length != COUNT){ |
8593 printf("incorrect length %d %d\n", out_length, COUNT); | |
8594 return -1; | |
8595 } | |
2967 | 8596 |
1168 | 8597 if(consumed != nal_length){ |
8598 printf("incorrect consumed length %d %d\n", nal_length, consumed); | |
8599 return -1; | |
8600 } | |
2967 | 8601 |
1168 | 8602 if(memcmp(bitstream, out, COUNT)){ |
8603 printf("missmatch\n"); | |
8604 return -1; | |
8605 } | |
8606 } | |
2967 | 8607 |
1168 | 8608 printf("Testing RBSP\n"); |
2967 | 8609 |
8610 | |
1168 | 8611 return 0; |
8612 } | |
8613 #endif | |
8614 | |
8615 | |
8616 static int decode_end(AVCodecContext *avctx) | |
8617 { | |
8618 H264Context *h = avctx->priv_data; | |
8619 MpegEncContext *s = &h->s; | |
2967 | 8620 |
2891
4c6eb826e9cb
Just noticed there is a memory leak in h264.c with the usage of rbsp_buffer.
michael
parents:
2883
diff
changeset
|
8621 av_freep(&h->rbsp_buffer); |
1168 | 8622 free_tables(h); //FIXME cleanup init stuff perhaps |
8623 MPV_common_end(s); | |
8624 | |
8625 // memset(h, 0, sizeof(H264Context)); | |
2967 | 8626 |
1168 | 8627 return 0; |
8628 } | |
8629 | |
8630 | |
8631 AVCodec h264_decoder = { | |
8632 "h264", | |
8633 CODEC_TYPE_VIDEO, | |
8634 CODEC_ID_H264, | |
8635 sizeof(H264Context), | |
8636 decode_init, | |
8637 NULL, | |
8638 decode_end, | |
8639 decode_frame, | |
2453 | 8640 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY, |
2640 | 8641 .flush= flush_dpb, |
1168 | 8642 }; |
8643 | |
3455
cc4b4ea83e29
--enable/disable parsers. Warning: some combinations are broken.
mru
parents:
3409
diff
changeset
|
8644 #ifdef CONFIG_H264_PARSER |
1988 | 8645 AVCodecParser h264_parser = { |
8646 { CODEC_ID_H264 }, | |
2392 | 8647 sizeof(H264Context), |
1988 | 8648 NULL, |
8649 h264_parse, | |
8650 ff_parse_close, | |
2774
3eddfe241bb9
support extracting sps/pps from bitstream and putting it in extradata
michael
parents:
2773
diff
changeset
|
8651 h264_split, |
1988 | 8652 }; |
3455
cc4b4ea83e29
--enable/disable parsers. Warning: some combinations are broken.
mru
parents:
3409
diff
changeset
|
8653 #endif |
1988 | 8654 |
1234 | 8655 #include "svq3.c" |