Mercurial > libavcodec.hg
annotate h264.h @ 11303:b0cb1f7efcbc libavcodec
Do not read beyond end of input in EA-TGV. This should avoid FATE test #362
result depending on uninitialized data.
FATE result may change for this test.
author | vitor |
---|---|
date | Fri, 26 Feb 2010 18:12:27 +0000 |
parents | 090fcf7610e8 |
children | c3d8a8eb1a45 |
rev | line source |
---|---|
4975 | 1 /* |
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder | |
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> | |
4 * | |
5 * This file is part of FFmpeg. | |
6 * | |
7 * FFmpeg is free software; you can redistribute it and/or | |
8 * modify it under the terms of the GNU Lesser General Public | |
9 * License as published by the Free Software Foundation; either | |
10 * version 2.1 of the License, or (at your option) any later version. | |
11 * | |
12 * FFmpeg is distributed in the hope that it will be useful, | |
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
18 * License along with FFmpeg; if not, write to the Free Software | |
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
20 */ | |
21 | |
22 /** | |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
8596
diff
changeset
|
23 * @file libavcodec/h264.h |
4975 | 24 * H.264 / AVC / MPEG4 part10 codec. |
25 * @author Michael Niedermayer <michaelni@gmx.at> | |
26 */ | |
27 | |
7760 | 28 #ifndef AVCODEC_H264_H |
29 #define AVCODEC_H264_H | |
4975 | 30 |
11203 | 31 #include "libavutil/intreadwrite.h" |
4975 | 32 #include "dsputil.h" |
33 #include "cabac.h" | |
34 #include "mpegvideo.h" | |
5638
4a26dc4ca11d
Move H.264 intra prediction functions into their own context
kostya
parents:
5231
diff
changeset
|
35 #include "h264pred.h" |
10866 | 36 #include "rectangle.h" |
4975 | 37 |
38 #define interlaced_dct interlaced_dct_is_a_bad_name | |
5129 | 39 #define mb_intra mb_intra_is_not_initialized_see_mb_type |
4975 | 40 |
41 #define LUMA_DC_BLOCK_INDEX 25 | |
42 #define CHROMA_DC_BLOCK_INDEX 26 | |
43 | |
44 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8 | |
45 #define COEFF_TOKEN_VLC_BITS 8 | |
46 #define TOTAL_ZEROS_VLC_BITS 9 | |
47 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3 | |
48 #define RUN_VLC_BITS 3 | |
49 #define RUN7_VLC_BITS 6 | |
50 | |
51 #define MAX_SPS_COUNT 32 | |
52 #define MAX_PPS_COUNT 256 | |
53 | |
54 #define MAX_MMCO_COUNT 66 | |
55 | |
7283
39e81fdd8d23
Move the size of the H264Context delayed_pic array into a #define.
astrange
parents:
7178
diff
changeset
|
56 #define MAX_DELAYED_PIC_COUNT 16 |
39e81fdd8d23
Move the size of the H264Context delayed_pic array into a #define.
astrange
parents:
7178
diff
changeset
|
57 |
4975 | 58 /* Compiling in interlaced support reduces the speed |
59 * of progressive decoding by about 2%. */ | |
60 #define ALLOW_INTERLACE | |
61 | |
7407 | 62 #define ALLOW_NOCHROMA |
63 | |
11282 | 64 #define FMO 0 |
65 | |
8079
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
66 /** |
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
67 * The maximum number of slices supported by the decoder. |
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
68 * must be a power of 2 |
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
69 */ |
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
70 #define MAX_SLICES 16 |
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
71 |
4975 | 72 #ifdef ALLOW_INTERLACE |
73 #define MB_MBAFF h->mb_mbaff | |
74 #define MB_FIELD h->mb_field_decoding_flag | |
75 #define FRAME_MBAFF h->mb_aff_frame | |
5801 | 76 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME) |
4975 | 77 #else |
78 #define MB_MBAFF 0 | |
79 #define MB_FIELD 0 | |
80 #define FRAME_MBAFF 0 | |
5767 | 81 #define FIELD_PICTURE 0 |
4975 | 82 #undef IS_INTERLACED |
83 #define IS_INTERLACED(mb_type) 0 | |
84 #endif | |
5781
0b3aa6f4c313
Modifies macroblock addressing and current macroblock y-position for field decoding.
andoma
parents:
5772
diff
changeset
|
85 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE) |
4975 | 86 |
7407 | 87 #ifdef ALLOW_NOCHROMA |
88 #define CHROMA h->sps.chroma_format_idc | |
89 #else | |
90 #define CHROMA 1 | |
91 #endif | |
92 | |
10893
2aafcafbe1f0
Replace cabac checks in inline functions from h264.h with constants.
michael
parents:
10883
diff
changeset
|
93 #ifndef CABAC |
2aafcafbe1f0
Replace cabac checks in inline functions from h264.h with constants.
michael
parents:
10883
diff
changeset
|
94 #define CABAC h->pps.cabac |
2aafcafbe1f0
Replace cabac checks in inline functions from h264.h with constants.
michael
parents:
10883
diff
changeset
|
95 #endif |
2aafcafbe1f0
Replace cabac checks in inline functions from h264.h with constants.
michael
parents:
10883
diff
changeset
|
96 |
7988
a7dfe657968d
Move nal unit types enum and EXTENDED_SAR #define from h264data.h to h264.h.
michael
parents:
7906
diff
changeset
|
97 #define EXTENDED_SAR 255 |
a7dfe657968d
Move nal unit types enum and EXTENDED_SAR #define from h264data.h to h264.h.
michael
parents:
7906
diff
changeset
|
98 |
8207
2f35f9781c31
Move #defines that are mostly used in h264.c out of h264data.h and into h264.h.
diego
parents:
8107
diff
changeset
|
99 #define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit |
2f35f9781c31
Move #defines that are mostly used in h264.c out of h264data.h and into h264.h.
diego
parents:
8107
diff
changeset
|
100 #define MB_TYPE_8x8DCT 0x01000000 |
2f35f9781c31
Move #defines that are mostly used in h264.c out of h264data.h and into h264.h.
diego
parents:
8107
diff
changeset
|
101 #define IS_REF0(a) ((a) & MB_TYPE_REF0) |
2f35f9781c31
Move #defines that are mostly used in h264.c out of h264data.h and into h264.h.
diego
parents:
8107
diff
changeset
|
102 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT) |
2f35f9781c31
Move #defines that are mostly used in h264.c out of h264data.h and into h264.h.
diego
parents:
8107
diff
changeset
|
103 |
10862
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
104 /** |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
105 * Value of Picture.reference when Picture is not a reference picture, but |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
106 * is held for delayed output. |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
107 */ |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
108 #define DELAYED_PIC_REF 4 |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
109 |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
110 |
7988
a7dfe657968d
Move nal unit types enum and EXTENDED_SAR #define from h264data.h to h264.h.
michael
parents:
7906
diff
changeset
|
111 /* NAL unit types */ |
a7dfe657968d
Move nal unit types enum and EXTENDED_SAR #define from h264data.h to h264.h.
michael
parents:
7906
diff
changeset
|
112 enum { |
7990 | 113 NAL_SLICE=1, |
114 NAL_DPA, | |
115 NAL_DPB, | |
116 NAL_DPC, | |
117 NAL_IDR_SLICE, | |
118 NAL_SEI, | |
119 NAL_SPS, | |
120 NAL_PPS, | |
121 NAL_AUD, | |
122 NAL_END_SEQUENCE, | |
123 NAL_END_STREAM, | |
124 NAL_FILLER_DATA, | |
125 NAL_SPS_EXT, | |
126 NAL_AUXILIARY_SLICE=19 | |
7988
a7dfe657968d
Move nal unit types enum and EXTENDED_SAR #define from h264data.h to h264.h.
michael
parents:
7906
diff
changeset
|
127 }; |
a7dfe657968d
Move nal unit types enum and EXTENDED_SAR #define from h264data.h to h264.h.
michael
parents:
7906
diff
changeset
|
128 |
4975 | 129 /** |
8762
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
130 * SEI message types |
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
131 */ |
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
132 typedef enum { |
8972 | 133 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1) |
8762
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
134 SEI_TYPE_PIC_TIMING = 1, ///< picture timing |
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
135 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data |
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
136 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync) |
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
137 } SEI_Type; |
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
138 |
4f1567ce75c4
Replace hard-coded SEI type constants with symbolic names
superdump
parents:
8740
diff
changeset
|
139 /** |
8107
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
140 * pic_struct in picture timing SEI message |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
141 */ |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
142 typedef enum { |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
143 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
144 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
145 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
146 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
147 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
148 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
149 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
150 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
151 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
152 } SEI_PicStructType; |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
153 |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
154 /** |
4975 | 155 * Sequence parameter set |
156 */ | |
157 typedef struct SPS{ | |
158 | |
159 int profile_idc; | |
160 int level_idc; | |
7407 | 161 int chroma_format_idc; |
4975 | 162 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag |
163 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4 | |
164 int poc_type; ///< pic_order_cnt_type | |
165 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4 | |
166 int delta_pic_order_always_zero_flag; | |
167 int offset_for_non_ref_pic; | |
168 int offset_for_top_to_bottom_field; | |
169 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle | |
170 int ref_frame_count; ///< num_ref_frames | |
171 int gaps_in_frame_num_allowed_flag; | |
5755
b45894d869da
Cosmetic preparations for h264/PAFF implementation.
andoma
parents:
5642
diff
changeset
|
172 int mb_width; ///< pic_width_in_mbs_minus1 + 1 |
b45894d869da
Cosmetic preparations for h264/PAFF implementation.
andoma
parents:
5642
diff
changeset
|
173 int mb_height; ///< pic_height_in_map_units_minus1 + 1 |
4975 | 174 int frame_mbs_only_flag; |
175 int mb_aff; ///<mb_adaptive_frame_field_flag | |
176 int direct_8x8_inference_flag; | |
177 int crop; ///< frame_cropping_flag | |
6582
7d40da88f0c7
Crop parameters are unsigned, having them negative could be bad and lead
michael
parents:
6488
diff
changeset
|
178 unsigned int crop_left; ///< frame_cropping_rect_left_offset |
7d40da88f0c7
Crop parameters are unsigned, having them negative could be bad and lead
michael
parents:
6488
diff
changeset
|
179 unsigned int crop_right; ///< frame_cropping_rect_right_offset |
7d40da88f0c7
Crop parameters are unsigned, having them negative could be bad and lead
michael
parents:
6488
diff
changeset
|
180 unsigned int crop_top; ///< frame_cropping_rect_top_offset |
7d40da88f0c7
Crop parameters are unsigned, having them negative could be bad and lead
michael
parents:
6488
diff
changeset
|
181 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset |
4975 | 182 int vui_parameters_present_flag; |
183 AVRational sar; | |
10837
e5905bfa625d
Export fullrange flag and color information for h.264
conrad
parents:
10829
diff
changeset
|
184 int video_signal_type_present_flag; |
e5905bfa625d
Export fullrange flag and color information for h.264
conrad
parents:
10829
diff
changeset
|
185 int full_range; |
e5905bfa625d
Export fullrange flag and color information for h.264
conrad
parents:
10829
diff
changeset
|
186 int colour_description_present_flag; |
e5905bfa625d
Export fullrange flag and color information for h.264
conrad
parents:
10829
diff
changeset
|
187 enum AVColorPrimaries color_primaries; |
e5905bfa625d
Export fullrange flag and color information for h.264
conrad
parents:
10829
diff
changeset
|
188 enum AVColorTransferCharacteristic color_trc; |
e5905bfa625d
Export fullrange flag and color information for h.264
conrad
parents:
10829
diff
changeset
|
189 enum AVColorSpace colorspace; |
4975 | 190 int timing_info_present_flag; |
191 uint32_t num_units_in_tick; | |
192 uint32_t time_scale; | |
193 int fixed_frame_rate_flag; | |
194 short offset_for_ref_frame[256]; //FIXME dyn aloc? | |
195 int bitstream_restriction_flag; | |
196 int num_reorder_frames; | |
197 int scaling_matrix_present; | |
198 uint8_t scaling_matrix4[6][16]; | |
199 uint8_t scaling_matrix8[2][64]; | |
8107
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
200 int nal_hrd_parameters_present_flag; |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
201 int vcl_hrd_parameters_present_flag; |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
202 int pic_struct_present_flag; |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
203 int time_offset_length; |
8970 | 204 int cpb_cnt; ///< See H.264 E.1.2 |
8965 | 205 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1 |
8107
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
206 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1 |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
207 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1 |
8735
5d7ebbb7e91b
Add fields to H264Context and SPS for upcoming VA API support.
cehoyos
parents:
8718
diff
changeset
|
208 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8 |
5d7ebbb7e91b
Add fields to H264Context and SPS for upcoming VA API support.
cehoyos
parents:
8718
diff
changeset
|
209 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8 |
5d7ebbb7e91b
Add fields to H264Context and SPS for upcoming VA API support.
cehoyos
parents:
8718
diff
changeset
|
210 int residual_color_transform_flag; ///< residual_colour_transform_flag |
4975 | 211 }SPS; |
212 | |
213 /** | |
214 * Picture parameter set | |
215 */ | |
216 typedef struct PPS{ | |
217 unsigned int sps_id; | |
218 int cabac; ///< entropy_coding_mode_flag | |
219 int pic_order_present; ///< pic_order_present_flag | |
220 int slice_group_count; ///< num_slice_groups_minus1 + 1 | |
221 int mb_slice_group_map_type; | |
222 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1 | |
223 int weighted_pred; ///< weighted_pred_flag | |
224 int weighted_bipred_idc; | |
225 int init_qp; ///< pic_init_qp_minus26 + 26 | |
226 int init_qs; ///< pic_init_qs_minus26 + 26 | |
5231
07a97575d0c4
Add support for streams with different chroma_qp_index_offset
gpoirier
parents:
5226
diff
changeset
|
227 int chroma_qp_index_offset[2]; |
4975 | 228 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag |
229 int constrained_intra_pred; ///< constrained_intra_pred_flag | |
230 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag | |
231 int transform_8x8_mode; ///< transform_8x8_mode_flag | |
232 uint8_t scaling_matrix4[6][16]; | |
233 uint8_t scaling_matrix8[2][64]; | |
7347
612a78c3b128
qscale has a range of 0..51 we thus do not need a 256 entry table and neither need
michael
parents:
7343
diff
changeset
|
234 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table |
5231
07a97575d0c4
Add support for streams with different chroma_qp_index_offset
gpoirier
parents:
5226
diff
changeset
|
235 int chroma_qp_diff; |
4975 | 236 }PPS; |
237 | |
238 /** | |
239 * Memory management control operation opcode. | |
240 */ | |
241 typedef enum MMCOOpcode{ | |
242 MMCO_END=0, | |
243 MMCO_SHORT2UNUSED, | |
244 MMCO_LONG2UNUSED, | |
245 MMCO_SHORT2LONG, | |
246 MMCO_SET_MAX_LONG, | |
247 MMCO_RESET, | |
248 MMCO_LONG, | |
249 } MMCOOpcode; | |
250 | |
251 /** | |
252 * Memory management control operation. | |
253 */ | |
254 typedef struct MMCO{ | |
255 MMCOOpcode opcode; | |
5756
db5a041fd77c
Rename MMCO stuff to prepare for h264/PAFF implementation.
andoma
parents:
5755
diff
changeset
|
256 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num) |
db5a041fd77c
Rename MMCO stuff to prepare for h264/PAFF implementation.
andoma
parents:
5755
diff
changeset
|
257 int long_arg; ///< index, pic_num, or num long refs depending on opcode |
4975 | 258 } MMCO; |
259 | |
260 /** | |
261 * H264Context | |
262 */ | |
263 typedef struct H264Context{ | |
264 MpegEncContext s; | |
265 int nal_ref_idc; | |
266 int nal_unit_type; | |
5174 | 267 uint8_t *rbsp_buffer[2]; |
268 unsigned int rbsp_buffer_size[2]; | |
4975 | 269 |
270 /** | |
271 * Used to parse AVC variant of h264 | |
272 */ | |
273 int is_avc; ///< this flag is != 0 if codec is avc1 | |
274 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4) | |
275 | |
5231
07a97575d0c4
Add support for streams with different chroma_qp_index_offset
gpoirier
parents:
5226
diff
changeset
|
276 int chroma_qp[2]; //QPc |
4975 | 277 |
10896
f112b4d030fa
Avoid wasting 4 cpu cycles per MB in redundantly calculating qp_thresh.
michael
parents:
10893
diff
changeset
|
278 int qp_thresh; ///< QP threshold to skip loopfilter |
f112b4d030fa
Avoid wasting 4 cpu cycles per MB in redundantly calculating qp_thresh.
michael
parents:
10893
diff
changeset
|
279 |
4975 | 280 int prev_mb_skipped; |
281 int next_mb_skipped; | |
282 | |
283 //prediction stuff | |
284 int chroma_pred_mode; | |
285 int intra16x16_pred_mode; | |
286 | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
287 int topleft_mb_xy; |
4975 | 288 int top_mb_xy; |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
289 int topright_mb_xy; |
4975 | 290 int left_mb_xy[2]; |
291 | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
292 int topleft_type; |
11012
b2fd83b26dd9
Set top & left types for deblock in fill_caches().
michael
parents:
11010
diff
changeset
|
293 int top_type; |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
294 int topright_type; |
11012
b2fd83b26dd9
Set top & left types for deblock in fill_caches().
michael
parents:
11010
diff
changeset
|
295 int left_type[2]; |
b2fd83b26dd9
Set top & left types for deblock in fill_caches().
michael
parents:
11010
diff
changeset
|
296 |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
297 const uint8_t * left_block; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
298 int topleft_partition; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
299 |
4975 | 300 int8_t intra4x4_pred_mode_cache[5*8]; |
11285 | 301 int8_t (*intra4x4_pred_mode); |
5638
4a26dc4ca11d
Move H.264 intra prediction functions into their own context
kostya
parents:
5231
diff
changeset
|
302 H264PredContext hpc; |
4975 | 303 unsigned int topleft_samples_available; |
304 unsigned int top_samples_available; | |
305 unsigned int topright_samples_available; | |
306 unsigned int left_samples_available; | |
307 uint8_t (*top_borders[2])[16+2*8]; | |
308 uint8_t left_border[2*(17+2*9)]; | |
309 | |
310 /** | |
311 * non zero coeff count cache. | |
312 * is 64 if not available. | |
313 */ | |
10961
34a65026fa06
Move array specifiers outside DECLARE_ALIGNED() invocations
mru
parents:
10934
diff
changeset
|
314 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8]; |
10909
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
315 |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
316 /* |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
317 .UU.YYYY |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
318 .UU.YYYY |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
319 .vv.YYYY |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
320 .VV.YYYY |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
321 */ |
10906 | 322 uint8_t (*non_zero_count)[32]; |
4975 | 323 |
324 /** | |
325 * Motion vector cache. | |
326 */ | |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
327 DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2]; |
10961
34a65026fa06
Move array specifiers outside DECLARE_ALIGNED() invocations
mru
parents:
10934
diff
changeset
|
328 DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8]; |
4975 | 329 #define LIST_NOT_USED -1 //FIXME rename? |
330 #define PART_NOT_AVAILABLE -2 | |
331 | |
332 /** | |
333 * is 1 if the specific list MV&references are set to 0,0,-2. | |
334 */ | |
335 int mv_cache_clean[2]; | |
336 | |
337 /** | |
338 * number of neighbors (top and/or left) that used 8x8 dct | |
339 */ | |
340 int neighbor_transform_size; | |
341 | |
342 /** | |
343 * block_offset[ 0..23] for frame macroblocks | |
344 * block_offset[24..47] for field macroblocks | |
345 */ | |
346 int block_offset[2*(16+8)]; | |
347 | |
348 uint32_t *mb2b_xy; //FIXME are these 4 a good idea? | |
11282 | 349 uint32_t *mb2br_xy; |
4975 | 350 int b_stride; //FIXME use s->b4_stride |
351 | |
352 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff | |
353 int mb_uvlinesize; | |
354 | |
355 int emu_edge_width; | |
356 int emu_edge_height; | |
357 | |
358 int halfpel_flag; | |
359 int thirdpel_flag; | |
360 | |
361 int unknown_svq3_flag; | |
362 int next_slice_index; | |
363 | |
5079 | 364 SPS *sps_buffers[MAX_SPS_COUNT]; |
4975 | 365 SPS sps; ///< current sps |
366 | |
5079 | 367 PPS *pps_buffers[MAX_PPS_COUNT]; |
4975 | 368 /** |
369 * current pps | |
370 */ | |
371 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that? | |
372 | |
373 uint32_t dequant4_buffer[6][52][16]; | |
374 uint32_t dequant8_buffer[2][52][64]; | |
375 uint32_t (*dequant4_coeff[6])[16]; | |
376 uint32_t (*dequant8_coeff[2])[64]; | |
377 int dequant_coeff_pps; ///< reinit tables when pps changes | |
378 | |
379 int slice_num; | |
8079
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
380 uint16_t *slice_table_base; |
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
381 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1 |
4975 | 382 int slice_type; |
7338 | 383 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P) |
4975 | 384 int slice_type_fixed; |
385 | |
386 //interlacing specific flags | |
387 int mb_aff_frame; | |
388 int mb_field_decoding_flag; | |
389 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag | |
390 | |
10961
34a65026fa06
Move array specifiers outside DECLARE_ALIGNED() invocations
mru
parents:
10934
diff
changeset
|
391 DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4]; |
4975 | 392 |
393 //POC stuff | |
394 int poc_lsb; | |
395 int poc_msb; | |
396 int delta_poc_bottom; | |
397 int delta_poc[2]; | |
398 int frame_num; | |
399 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0 | |
400 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0 | |
401 int frame_num_offset; ///< for POC type 2 | |
402 int prev_frame_num_offset; ///< for POC type 2 | |
403 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2 | |
404 | |
405 /** | |
5772
65b71bd21a4d
Fix h->curr_pic_num for field pictures. Necessary for proper PAFF support.
andoma
parents:
5767
diff
changeset
|
406 * frame_num for frames or 2*frame_num+1 for field pics. |
4975 | 407 */ |
408 int curr_pic_num; | |
409 | |
410 /** | |
411 * max_frame_num or 2*max_frame_num for field pics. | |
412 */ | |
413 int max_pic_num; | |
414 | |
415 //Weighted pred stuff | |
416 int use_weight; | |
417 int use_weight_chroma; | |
418 int luma_log2_weight_denom; | |
419 int chroma_log2_weight_denom; | |
420 int luma_weight[2][48]; | |
421 int luma_offset[2][48]; | |
422 int chroma_weight[2][48][2]; | |
423 int chroma_offset[2][48][2]; | |
424 int implicit_weight[48][48]; | |
425 | |
426 //deblock | |
427 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0 | |
428 int slice_alpha_c0_offset; | |
429 int slice_beta_offset; | |
430 | |
431 int redundant_pic_count; | |
432 | |
433 int direct_spatial_mv_pred; | |
11096 | 434 int col_parity; |
435 int col_fieldoff; | |
4975 | 436 int dist_scale_factor[16]; |
7898
a33287a39a55
Make MBAFF temporal direct mode closer to the spec.
michael
parents:
7760
diff
changeset
|
437 int dist_scale_factor_field[2][32]; |
7906
5be944626072
Another try to fix temporal direct mode references.
michael
parents:
7898
diff
changeset
|
438 int map_col_to_list0[2][16+32]; |
5be944626072
Another try to fix temporal direct mode references.
michael
parents:
7898
diff
changeset
|
439 int map_col_to_list0_field[2][2][16+32]; |
4975 | 440 |
441 /** | |
442 * num_ref_idx_l0/1_active_minus1 + 1 | |
443 */ | |
444 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode | |
445 unsigned int list_count; | |
10906 | 446 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type |
4975 | 447 Picture *short_ref[32]; |
448 Picture *long_ref[32]; | |
5755
b45894d869da
Cosmetic preparations for h264/PAFF implementation.
andoma
parents:
5642
diff
changeset
|
449 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture |
b45894d869da
Cosmetic preparations for h264/PAFF implementation.
andoma
parents:
5642
diff
changeset
|
450 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs. |
b45894d869da
Cosmetic preparations for h264/PAFF implementation.
andoma
parents:
5642
diff
changeset
|
451 Reordered version of default_ref_list |
b45894d869da
Cosmetic preparations for h264/PAFF implementation.
andoma
parents:
5642
diff
changeset
|
452 according to picture reordering in slice header */ |
8079
2d3c7cd7d143
Replace literally hardcoded max slice number by named constant.
michael
parents:
8025
diff
changeset
|
453 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1 |
7283
39e81fdd8d23
Move the size of the H264Context delayed_pic array into a #define.
astrange
parents:
7178
diff
changeset
|
454 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size? |
7310
033d10927d45
Remove delayed_output_pic, I do not understand what this variable was good for.
michael
parents:
7283
diff
changeset
|
455 int outputed_poc; |
4975 | 456 |
457 /** | |
458 * memory management control operations buffer. | |
459 */ | |
460 MMCO mmco[MAX_MMCO_COUNT]; | |
461 int mmco_index; | |
462 | |
463 int long_ref_count; ///< number of actual long term references | |
464 int short_ref_count; ///< number of actual short term references | |
465 | |
466 //data partitioning | |
467 GetBitContext intra_gb; | |
468 GetBitContext inter_gb; | |
469 GetBitContext *intra_gb_ptr; | |
470 GetBitContext *inter_gb_ptr; | |
471 | |
10961
34a65026fa06
Move array specifiers outside DECLARE_ALIGNED() invocations
mru
parents:
10934
diff
changeset
|
472 DECLARE_ALIGNED_16(DCTELEM, mb)[16*24]; |
6488 | 473 DCTELEM mb_padding[256]; ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb |
4975 | 474 |
475 /** | |
476 * Cabac | |
477 */ | |
478 CABACContext cabac; | |
479 uint8_t cabac_state[460]; | |
480 int cabac_init_idc; | |
481 | |
482 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */ | |
483 uint16_t *cbp_table; | |
484 int cbp; | |
485 int top_cbp; | |
486 int left_cbp; | |
487 /* chroma_pred_mode for i4x4 or i16x16, else 0 */ | |
488 uint8_t *chroma_pred_mode_table; | |
489 int last_qscale_diff; | |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
490 uint8_t (*mvd_table[2])[2]; |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
491 DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2]; |
4975 | 492 uint8_t *direct_table; |
493 uint8_t direct_cache[5*8]; | |
494 | |
495 uint8_t zigzag_scan[16]; | |
496 uint8_t zigzag_scan8x8[64]; | |
497 uint8_t zigzag_scan8x8_cavlc[64]; | |
498 uint8_t field_scan[16]; | |
499 uint8_t field_scan8x8[64]; | |
500 uint8_t field_scan8x8_cavlc[64]; | |
501 const uint8_t *zigzag_scan_q0; | |
502 const uint8_t *zigzag_scan8x8_q0; | |
503 const uint8_t *zigzag_scan8x8_cavlc_q0; | |
504 const uint8_t *field_scan_q0; | |
505 const uint8_t *field_scan8x8_q0; | |
506 const uint8_t *field_scan8x8_cavlc_q0; | |
507 | |
508 int x264_build; | |
5642 | 509 |
510 /** | |
511 * @defgroup multithreading Members for slice based multithreading | |
512 * @{ | |
513 */ | |
514 struct H264Context *thread_context[MAX_THREADS]; | |
515 | |
516 /** | |
517 * current slice number, used to initalize slice_num of each thread/context | |
518 */ | |
519 int current_slice; | |
520 | |
521 /** | |
522 * Max number of threads / contexts. | |
523 * This is equal to AVCodecContext.thread_count unless | |
524 * multithreaded decoding is impossible, in which case it is | |
525 * reduced to 1. | |
526 */ | |
527 int max_contexts; | |
528 | |
529 /** | |
530 * 1 if the single thread fallback warning has already been | |
531 * displayed, 0 otherwise. | |
532 */ | |
533 int single_decode_warning; | |
534 | |
535 int last_slice_type; | |
536 /** @} */ | |
537 | |
6783
df0893f4fd86
Store mb_xy in H264Context and only calculate it once per MB.
astrange
parents:
6582
diff
changeset
|
538 int mb_xy; |
df0893f4fd86
Store mb_xy in H264Context and only calculate it once per MB.
astrange
parents:
6582
diff
changeset
|
539 |
8025 | 540 uint32_t svq3_watermark_key; |
8107
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
541 |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
542 /** |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
543 * pic_struct in picture timing SEI message |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
544 */ |
e61f76efc9f3
h264: Implement decoding of picture timing SEI message.
andoma
parents:
8079
diff
changeset
|
545 SEI_PicStructType sei_pic_struct; |
8353
add1a0d65370
Do not recalculate constant part of is_complex for every MB.
michael
parents:
8207
diff
changeset
|
546 |
8783
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
547 /** |
9831
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
548 * Complement sei_pic_struct |
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
549 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames. |
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
550 * However, soft telecined frames may have these values. |
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
551 * This is used in an attempt to flag soft telecine progressive. |
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
552 */ |
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
553 int prev_interlaced_frame; |
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
554 |
febe1855b64a
Add field prev_interlaced_frame to H264Context to be able to flag soft telecine
cehoyos
parents:
9384
diff
changeset
|
555 /** |
9128
793cf8c68c4f
Add support for ct_type to correctly detect interlaced flag
schreter
parents:
9071
diff
changeset
|
556 * Bit set of clock types for fields/frames in picture timing SEI message. |
793cf8c68c4f
Add support for ct_type to correctly detect interlaced flag
schreter
parents:
9071
diff
changeset
|
557 * For each found ct_type, appropriate bit is set (e.g., bit 1 for |
793cf8c68c4f
Add support for ct_type to correctly detect interlaced flag
schreter
parents:
9071
diff
changeset
|
558 * interlaced). |
793cf8c68c4f
Add support for ct_type to correctly detect interlaced flag
schreter
parents:
9071
diff
changeset
|
559 */ |
793cf8c68c4f
Add support for ct_type to correctly detect interlaced flag
schreter
parents:
9071
diff
changeset
|
560 int sei_ct_type; |
793cf8c68c4f
Add support for ct_type to correctly detect interlaced flag
schreter
parents:
9071
diff
changeset
|
561 |
793cf8c68c4f
Add support for ct_type to correctly detect interlaced flag
schreter
parents:
9071
diff
changeset
|
562 /** |
8966 | 563 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2 |
564 */ | |
565 int sei_dpb_output_delay; | |
566 | |
567 /** | |
8967 | 568 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2 |
569 */ | |
570 int sei_cpb_removal_delay; | |
571 | |
572 /** | |
8783
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
573 * recovery_frame_cnt from SEI message |
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
574 * |
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
575 * Set to -1 if no recovery point SEI message found or to number of frames |
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
576 * before playback synchronizes. Frames having recovery point are key |
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
577 * frames. |
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
578 */ |
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
579 int sei_recovery_frame_cnt; |
e91ea98d868a
Add SEI recovery point frame counter to H264Context and use it when
cehoyos
parents:
8762
diff
changeset
|
580 |
8353
add1a0d65370
Do not recalculate constant part of is_complex for every MB.
michael
parents:
8207
diff
changeset
|
581 int is_complex; |
8735
5d7ebbb7e91b
Add fields to H264Context and SPS for upcoming VA API support.
cehoyos
parents:
8718
diff
changeset
|
582 |
5d7ebbb7e91b
Add fields to H264Context and SPS for upcoming VA API support.
cehoyos
parents:
8718
diff
changeset
|
583 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag |
5d7ebbb7e91b
Add fields to H264Context and SPS for upcoming VA API support.
cehoyos
parents:
8718
diff
changeset
|
584 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag |
8972 | 585 |
586 // Timestamp stuff | |
587 int sei_buffering_period_present; ///< Buffering period SEI flag | |
588 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs | |
4975 | 589 }H264Context; |
590 | |
10859
762e6bb0ba40
Split (picture|seq) parameter set decoding out of h264.c.
michael
parents:
10858
diff
changeset
|
591 |
762e6bb0ba40
Split (picture|seq) parameter set decoding out of h264.c.
michael
parents:
10858
diff
changeset
|
592 extern const uint8_t ff_h264_chroma_qp[52]; |
762e6bb0ba40
Split (picture|seq) parameter set decoding out of h264.c.
michael
parents:
10858
diff
changeset
|
593 |
11031 | 594 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp); |
595 | |
596 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc); | |
10859
762e6bb0ba40
Split (picture|seq) parameter set decoding out of h264.c.
michael
parents:
10858
diff
changeset
|
597 |
8996
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
598 /** |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
599 * Decode SEI |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
600 */ |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
601 int ff_h264_decode_sei(H264Context *h); |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
602 |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
603 /** |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
604 * Decode SPS |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
605 */ |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
606 int ff_h264_decode_seq_parameter_set(H264Context *h); |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
607 |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
608 /** |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
609 * Decode PPS |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
610 */ |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
611 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length); |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
612 |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
613 /** |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
614 * Decodes a network abstraction layer unit. |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
615 * @param consumed is the number of bytes used as input |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
616 * @param length is the length of the array |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
617 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing? |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
618 * @returns decoded bytes, might be src+1 if no escapes |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
619 */ |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
620 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length); |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
621 |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
622 /** |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
623 * identifies the exact end of the bitstream |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
624 * @return the length of the trailing, or 0 if damaged |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
625 */ |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
626 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src); |
e65778184ded
Make the following H264 functions available to the parser:
cehoyos
parents:
8972
diff
changeset
|
627 |
9380
54e650136c87
Add a ff_h264_free_context function and call it from the H.264 parser.
reimar
parents:
9128
diff
changeset
|
628 /** |
54e650136c87
Add a ff_h264_free_context function and call it from the H.264 parser.
reimar
parents:
9128
diff
changeset
|
629 * frees any data that may have been allocated in the H264 context like SPS, PPS etc. |
54e650136c87
Add a ff_h264_free_context function and call it from the H.264 parser.
reimar
parents:
9128
diff
changeset
|
630 */ |
9384
828d15d322ed
Add forgotten void return type to ff_h264_free_context
reimar
parents:
9380
diff
changeset
|
631 av_cold void ff_h264_free_context(H264Context *h); |
9380
54e650136c87
Add a ff_h264_free_context function and call it from the H.264 parser.
reimar
parents:
9128
diff
changeset
|
632 |
10829
46a247e54d6e
Export get_slice_type() as ff_h264_get_slice_type().
ramiro
parents:
9831
diff
changeset
|
633 /** |
46a247e54d6e
Export get_slice_type() as ff_h264_get_slice_type().
ramiro
parents:
9831
diff
changeset
|
634 * reconstructs bitstream slice_type. |
46a247e54d6e
Export get_slice_type() as ff_h264_get_slice_type().
ramiro
parents:
9831
diff
changeset
|
635 */ |
10987
85de0c8a19b7
Added a missing const to ff_h264_get_slice_type().
fenrir
parents:
10985
diff
changeset
|
636 int ff_h264_get_slice_type(const H264Context *h); |
10829
46a247e54d6e
Export get_slice_type() as ff_h264_get_slice_type().
ramiro
parents:
9831
diff
changeset
|
637 |
10852 | 638 /** |
639 * allocates tables. | |
640 * needs width/height | |
641 */ | |
642 int ff_h264_alloc_tables(H264Context *h); | |
643 | |
644 /** | |
10862
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
645 * fills the default_ref_list. |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
646 */ |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
647 int ff_h264_fill_default_ref_list(H264Context *h); |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
648 |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
649 int ff_h264_decode_ref_pic_list_reordering(H264Context *h); |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
650 void ff_h264_fill_mbaff_ref_list(H264Context *h); |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
651 void ff_h264_remove_all_refs(H264Context *h); |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
652 |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
653 /** |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
654 * Executes the reference picture marking (memory management control operations). |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
655 */ |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
656 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count); |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
657 |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
658 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb); |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
659 |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
660 |
d9c084a0c22b
Split all the reference picture handling off h264.c.
michael
parents:
10859
diff
changeset
|
661 /** |
10852 | 662 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. |
663 */ | |
10863
974ac220c93a
Move check_intra4x4_pred_mode() back from h264.h to h264.c, the function is just
michael
parents:
10862
diff
changeset
|
664 int ff_h264_check_intra4x4_pred_mode(H264Context *h); |
974ac220c93a
Move check_intra4x4_pred_mode() back from h264.h to h264.c, the function is just
michael
parents:
10862
diff
changeset
|
665 |
974ac220c93a
Move check_intra4x4_pred_mode() back from h264.h to h264.c, the function is just
michael
parents:
10862
diff
changeset
|
666 /** |
974ac220c93a
Move check_intra4x4_pred_mode() back from h264.h to h264.c, the function is just
michael
parents:
10862
diff
changeset
|
667 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks. |
974ac220c93a
Move check_intra4x4_pred_mode() back from h264.h to h264.c, the function is just
michael
parents:
10862
diff
changeset
|
668 */ |
10852 | 669 int ff_h264_check_intra_pred_mode(H264Context *h, int mode); |
670 | |
671 void ff_h264_write_back_intra_pred_mode(H264Context *h); | |
672 void ff_h264_hl_decode_mb(H264Context *h); | |
673 int ff_h264_frame_start(H264Context *h); | |
674 av_cold int ff_h264_decode_init(AVCodecContext *avctx); | |
675 av_cold int ff_h264_decode_end(AVCodecContext *avctx); | |
10866 | 676 av_cold void ff_h264_decode_init_vlc(void); |
677 | |
678 /** | |
679 * decodes a macroblock | |
680 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed | |
681 */ | |
682 int ff_h264_decode_mb_cavlc(H264Context *h); | |
10852 | 683 |
10869 | 684 /** |
685 * decodes a CABAC coded macroblock | |
686 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed | |
687 */ | |
688 int ff_h264_decode_mb_cabac(H264Context *h); | |
689 | |
690 void ff_h264_init_cabac_states(H264Context *h); | |
691 | |
10857
b20434143fd5
Split direct mode (macro)block decoding off h264.c.
michael
parents:
10854
diff
changeset
|
692 void ff_h264_direct_dist_scale_factor(H264Context * const h); |
b20434143fd5
Split direct mode (macro)block decoding off h264.c.
michael
parents:
10854
diff
changeset
|
693 void ff_h264_direct_ref_list_init(H264Context * const h); |
b20434143fd5
Split direct mode (macro)block decoding off h264.c.
michael
parents:
10854
diff
changeset
|
694 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type); |
b20434143fd5
Split direct mode (macro)block decoding off h264.c.
michael
parents:
10854
diff
changeset
|
695 |
10854 | 696 void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize); |
697 void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize); | |
698 | |
10858 | 699 /** |
700 * Reset SEI values at the beginning of the frame. | |
701 * | |
702 * @param h H.264 context. | |
703 */ | |
704 void ff_h264_reset_sei(H264Context *h); | |
705 | |
706 | |
10852 | 707 /* |
708 o-o o-o | |
709 / / / | |
710 o-o o-o | |
711 ,---' | |
712 o-o o-o | |
713 / / / | |
714 o-o o-o | |
715 */ | |
716 //This table must be here because scan8[constant] must be known at compiletime | |
717 static const uint8_t scan8[16 + 2*4]={ | |
718 4+1*8, 5+1*8, 4+2*8, 5+2*8, | |
719 6+1*8, 7+1*8, 6+2*8, 7+2*8, | |
720 4+3*8, 5+3*8, 4+4*8, 5+4*8, | |
721 6+3*8, 7+3*8, 6+4*8, 7+4*8, | |
722 1+1*8, 2+1*8, | |
723 1+2*8, 2+2*8, | |
724 1+4*8, 2+4*8, | |
725 1+5*8, 2+5*8, | |
726 }; | |
727 | |
728 static av_always_inline uint32_t pack16to32(int a, int b){ | |
729 #if HAVE_BIGENDIAN | |
730 return (b&0xFFFF) + (a<<16); | |
731 #else | |
732 return (a&0xFFFF) + (b<<16); | |
733 #endif | |
734 } | |
735 | |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
736 static av_always_inline uint16_t pack8to16(int a, int b){ |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
737 #if HAVE_BIGENDIAN |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
738 return (b&0xFF) + (a<<8); |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
739 #else |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
740 return (a&0xFF) + (b<<8); |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
741 #endif |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
742 } |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
743 |
10852 | 744 /** |
10854 | 745 * gets the chroma qp. |
746 */ | |
747 static inline int get_chroma_qp(H264Context *h, int t, int qscale){ | |
748 return h->pps.chroma_qp_table[t][qscale]; | |
749 } | |
750 | |
10866 | 751 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my); |
752 | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
753 static void fill_decode_neighbors(H264Context *h, int mb_type){ |
10866 | 754 MpegEncContext * const s = &h->s; |
755 const int mb_xy= h->mb_xy; | |
756 int topleft_xy, top_xy, topright_xy, left_xy[2]; | |
10909
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
757 static const uint8_t left_block_options[4][16]={ |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
758 {0,1,2,3,7,10,8,11,7+0*8, 7+1*8, 7+2*8, 7+3*8, 2+0*8, 2+3*8, 2+1*8, 2+2*8}, |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
759 {2,2,3,3,8,11,8,11,7+2*8, 7+2*8, 7+3*8, 7+3*8, 2+1*8, 2+2*8, 2+1*8, 2+2*8}, |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
760 {0,0,1,1,7,10,7,10,7+0*8, 7+0*8, 7+1*8, 7+1*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8}, |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
761 {0,2,0,2,7,10,7,10,7+0*8, 7+2*8, 7+0*8, 7+2*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8} |
10866 | 762 }; |
763 | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
764 h->topleft_partition= -1; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
765 |
11008
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
766 top_xy = mb_xy - (s->mb_stride << MB_FIELD); |
10866 | 767 |
768 /* Wow, what a mess, why didn't they simplify the interlacing & intra | |
769 * stuff, I can't imagine that these complex rules are worth it. */ | |
770 | |
771 topleft_xy = top_xy - 1; | |
772 topright_xy= top_xy + 1; | |
773 left_xy[1] = left_xy[0] = mb_xy-1; | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
774 h->left_block = left_block_options[0]; |
10866 | 775 if(FRAME_MBAFF){ |
11008
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
776 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]); |
10866 | 777 const int curr_mb_field_flag = IS_INTERLACED(mb_type); |
11008
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
778 if(s->mb_y&1){ |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
779 if (left_mb_field_flag != curr_mb_field_flag) { |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
780 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1; |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
781 if (curr_mb_field_flag) { |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
782 left_xy[1] += s->mb_stride; |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
783 h->left_block = left_block_options[3]; |
11008
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
784 } else { |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
785 topleft_xy += s->mb_stride; |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
786 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
787 h->topleft_partition = 0; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
788 h->left_block = left_block_options[1]; |
11008
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
789 } |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
790 } |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
791 }else{ |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
792 if(curr_mb_field_flag){ |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
793 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1); |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
794 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1); |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
795 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1); |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
796 } |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
797 if (left_mb_field_flag != curr_mb_field_flag) { |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
798 if (curr_mb_field_flag) { |
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
799 left_xy[1] += s->mb_stride; |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
800 h->left_block = left_block_options[3]; |
11008
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
801 } else { |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
802 h->left_block = left_block_options[2]; |
11008
ec192d9ebac5
Optimize mb neighbor initialization for MBAFF in fill_caches().
michael
parents:
11002
diff
changeset
|
803 } |
10866 | 804 } |
805 } | |
806 } | |
807 | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
808 h->topleft_mb_xy = topleft_xy; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
809 h->top_mb_xy = top_xy; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
810 h->topright_mb_xy= topright_xy; |
10866 | 811 h->left_mb_xy[0] = left_xy[0]; |
812 h->left_mb_xy[1] = left_xy[1]; | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
813 //FIXME do we need all in the context? |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
814 h->topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
815 h->top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
816 h->topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
817 h->left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
818 h->left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
819 } |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
820 |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
821 static void fill_decode_caches(H264Context *h, int mb_type){ |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
822 MpegEncContext * const s = &h->s; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
823 int topleft_xy, top_xy, topright_xy, left_xy[2]; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
824 int topleft_type, top_type, topright_type, left_type[2]; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
825 const uint8_t * left_block= h->left_block; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
826 int i; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
827 |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
828 topleft_xy = h->topleft_mb_xy ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
829 top_xy = h->top_mb_xy ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
830 topright_xy = h->topright_mb_xy; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
831 left_xy[0] = h->left_mb_xy[0] ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
832 left_xy[1] = h->left_mb_xy[1] ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
833 topleft_type = h->topleft_type ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
834 top_type = h->top_type ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
835 topright_type= h->topright_type ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
836 left_type[0] = h->left_type[0] ; |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
837 left_type[1] = h->left_type[1] ; |
10866 | 838 |
11158
81c3f88f460f
Skiped MBs dont need the cbp stuff so skip initing that.
michael
parents:
11156
diff
changeset
|
839 if(!IS_SKIP(mb_type)){ |
10928 | 840 if(IS_INTRA(mb_type)){ |
841 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1; | |
842 h->topleft_samples_available= | |
843 h->top_samples_available= | |
844 h->left_samples_available= 0xFFFF; | |
845 h->topright_samples_available= 0xEEEA; | |
10866 | 846 |
10928 | 847 if(!(top_type & type_mask)){ |
848 h->topleft_samples_available= 0xB3FF; | |
849 h->top_samples_available= 0x33FF; | |
850 h->topright_samples_available= 0x26EA; | |
851 } | |
852 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){ | |
853 if(IS_INTERLACED(mb_type)){ | |
854 if(!(left_type[0] & type_mask)){ | |
855 h->topleft_samples_available&= 0xDFFF; | |
856 h->left_samples_available&= 0x5FFF; | |
857 } | |
858 if(!(left_type[1] & type_mask)){ | |
859 h->topleft_samples_available&= 0xFF5F; | |
860 h->left_samples_available&= 0xFF5F; | |
861 } | |
862 }else{ | |
11296
4709312b48d5
Remove useless check of the 2 left MBs of a pair being in the same slice.
michael
parents:
11294
diff
changeset
|
863 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride]; |
4709312b48d5
Remove useless check of the 2 left MBs of a pair being in the same slice.
michael
parents:
11294
diff
changeset
|
864 |
10928 | 865 assert(left_xy[0] == left_xy[1]); |
866 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){ | |
867 h->topleft_samples_available&= 0xDF5F; | |
868 h->left_samples_available&= 0x5F5F; | |
869 } | |
10866 | 870 } |
871 }else{ | |
10928 | 872 if(!(left_type[0] & type_mask)){ |
10866 | 873 h->topleft_samples_available&= 0xDF5F; |
874 h->left_samples_available&= 0x5F5F; | |
875 } | |
876 } | |
877 | |
10928 | 878 if(!(topleft_type & type_mask)) |
879 h->topleft_samples_available&= 0x7FFF; | |
10866 | 880 |
10928 | 881 if(!(topright_type & type_mask)) |
882 h->topright_samples_available&= 0xFBFF; | |
10866 | 883 |
10928 | 884 if(IS_INTRA4x4(mb_type)){ |
885 if(IS_INTRA4x4(top_type)){ | |
11286
db94c9bc5694
Reorder intra4x4_pred_mode so that we can read/write 4 values at once.
michael
parents:
11285
diff
changeset
|
886 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]); |
10866 | 887 }else{ |
10928 | 888 h->intra4x4_pred_mode_cache[4+8*0]= |
889 h->intra4x4_pred_mode_cache[5+8*0]= | |
890 h->intra4x4_pred_mode_cache[6+8*0]= | |
11287 | 891 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask); |
10928 | 892 } |
893 for(i=0; i<2; i++){ | |
894 if(IS_INTRA4x4(left_type[i])){ | |
11285 | 895 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]]; |
11286
db94c9bc5694
Reorder intra4x4_pred_mode so that we can read/write 4 values at once.
michael
parents:
11285
diff
changeset
|
896 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]]; |
db94c9bc5694
Reorder intra4x4_pred_mode so that we can read/write 4 values at once.
michael
parents:
11285
diff
changeset
|
897 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]]; |
10928 | 898 }else{ |
899 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= | |
11287 | 900 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask); |
10928 | 901 } |
10866 | 902 } |
903 } | |
904 } | |
905 | |
906 | |
907 /* | |
908 0 . T T. T T T T | |
909 1 L . .L . . . . | |
910 2 L . .L . . . . | |
911 3 . T TL . . . . | |
912 4 L . .L . . . . | |
913 5 L . .. . . . . | |
914 */ | |
915 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec) | |
916 if(top_type){ | |
11203 | 917 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]); |
10928 | 918 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8]; |
919 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8]; | |
10866 | 920 |
10928 | 921 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8]; |
922 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8]; | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
923 }else { |
10928 | 924 h->non_zero_count_cache[1+8*0]= |
925 h->non_zero_count_cache[2+8*0]= | |
10866 | 926 |
10928 | 927 h->non_zero_count_cache[1+8*3]= |
928 h->non_zero_count_cache[2+8*3]= | |
11203 | 929 AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040); |
10866 | 930 } |
931 | |
932 for (i=0; i<2; i++) { | |
933 if(left_type[i]){ | |
10909
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
934 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]]; |
f4cf3960b8c6
Reorganize how values are stored in h->non_zero_count.
michael
parents:
10908
diff
changeset
|
935 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]]; |
10928 | 936 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]]; |
937 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]]; | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
938 }else{ |
10928 | 939 h->non_zero_count_cache[3+8*1 + 2*8*i]= |
940 h->non_zero_count_cache[3+8*2 + 2*8*i]= | |
941 h->non_zero_count_cache[0+8*1 + 8*i]= | |
942 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64; | |
10866 | 943 } |
944 } | |
945 | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
946 if( CABAC ) { |
10866 | 947 // top_cbp |
948 if(top_type) { | |
949 h->top_cbp = h->cbp_table[top_xy]; | |
950 } else { | |
11300 | 951 h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F; |
10866 | 952 } |
953 // left_cbp | |
954 if (left_type[0]) { | |
11301
5843fc07bda0
Remove some useless operations from the code setting left_cbp.
michael
parents:
11300
diff
changeset
|
955 h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0) |
5843fc07bda0
Remove some useless operations from the code setting left_cbp.
michael
parents:
11300
diff
changeset
|
956 | ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2) |
5843fc07bda0
Remove some useless operations from the code setting left_cbp.
michael
parents:
11300
diff
changeset
|
957 | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2); |
10866 | 958 } else { |
11300 | 959 h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F; |
10866 | 960 } |
961 } | |
11158
81c3f88f460f
Skiped MBs dont need the cbp stuff so skip initing that.
michael
parents:
11156
diff
changeset
|
962 } |
10866 | 963 |
964 #if 1 | |
11144
0f7dcfca7b2e
Dont calculate any surrounding MVs for temporal MBs
michael
parents:
11102
diff
changeset
|
965 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){ |
10866 | 966 int list; |
967 for(list=0; list<h->list_count; list++){ | |
11145
8805efb4c3fd
Move setting MB_TYPE_L0L1 for direct MBs up, this is simpler.
michael
parents:
11144
diff
changeset
|
968 if(!USES_LIST(mb_type, list)){ |
10866 | 969 /*if(!h->mv_cache_clean[list]){ |
970 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all? | |
971 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t)); | |
972 h->mv_cache_clean[list]= 1; | |
973 }*/ | |
974 continue; | |
975 } | |
11144
0f7dcfca7b2e
Dont calculate any surrounding MVs for temporal MBs
michael
parents:
11102
diff
changeset
|
976 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)); |
0f7dcfca7b2e
Dont calculate any surrounding MVs for temporal MBs
michael
parents:
11102
diff
changeset
|
977 |
10866 | 978 h->mv_cache_clean[list]= 0; |
979 | |
980 if(USES_LIST(top_type, list)){ | |
981 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; | |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
982 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]); |
10928 | 983 h->ref_cache[list][scan8[0] + 0 - 1*8]= |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
984 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2]; |
10928 | 985 h->ref_cache[list][scan8[0] + 2 - 1*8]= |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
986 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3]; |
10866 | 987 }else{ |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
988 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]); |
11203 | 989 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101); |
10866 | 990 } |
991 | |
992 for(i=0; i<2; i++){ | |
993 int cache_idx = scan8[0] - 1 + i*2*8; | |
994 if(USES_LIST(left_type[i], list)){ | |
995 const int b_xy= h->mb2b_xy[left_xy[i]] + 3; | |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
996 const int b8_xy= 4*left_xy[i] + 1; |
11203 | 997 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]); |
998 AV_COPY32(h->mv_cache[list][cache_idx+8], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]]); | |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
999 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)]; |
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1000 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)]; |
10866 | 1001 }else{ |
11203 | 1002 AV_ZERO32(h->mv_cache [list][cache_idx ]); |
1003 AV_ZERO32(h->mv_cache [list][cache_idx+8]); | |
10866 | 1004 h->ref_cache[list][cache_idx ]= |
11019 | 1005 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE; |
10866 | 1006 } |
1007 } | |
1008 | |
1009 if(USES_LIST(topright_type, list)){ | |
1010 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride; | |
11203 | 1011 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]); |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1012 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2]; |
10866 | 1013 }else{ |
11203 | 1014 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]); |
10866 | 1015 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; |
1016 } | |
11302
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1017 if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){ |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1018 if(USES_LIST(topleft_type, list)){ |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1019 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride); |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1020 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2); |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1021 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]); |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1022 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy]; |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1023 }else{ |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1024 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]); |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1025 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE; |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1026 } |
090fcf7610e8
Only load the topleft mv/ref when the topright is unavailable.
michael
parents:
11301
diff
changeset
|
1027 } |
10866 | 1028 |
11159 | 1029 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF) |
10866 | 1030 continue; |
1031 | |
11156
81ef4df2c774
Also skip direct/mvd_cache init for skiped blocks.
michael
parents:
11155
diff
changeset
|
1032 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) { |
10866 | 1033 h->ref_cache[list][scan8[4 ]] = |
1034 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE; | |
11203 | 1035 AV_ZERO32(h->mv_cache [list][scan8[4 ]]); |
1036 AV_ZERO32(h->mv_cache [list][scan8[12]]); | |
10866 | 1037 |
11155 | 1038 if( CABAC ) { |
10866 | 1039 /* XXX beurk, Load mvd */ |
1040 if(USES_LIST(top_type, list)){ | |
11283
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1041 const int b_xy= h->mb2br_xy[top_xy]; |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1042 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]); |
10866 | 1043 }else{ |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1044 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]); |
10866 | 1045 } |
1046 if(USES_LIST(left_type[0], list)){ | |
11283
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1047 const int b_xy= h->mb2br_xy[left_xy[0]] + 6; |
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1048 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]); |
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1049 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]); |
10866 | 1050 }else{ |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1051 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]); |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1052 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]); |
10866 | 1053 } |
1054 if(USES_LIST(left_type[1], list)){ | |
11283
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1055 const int b_xy= h->mb2br_xy[left_xy[1]] + 6; |
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1056 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]); |
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1057 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]); |
10866 | 1058 }else{ |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1059 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]); |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1060 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]); |
10866 | 1061 } |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1062 AV_ZERO16(h->mvd_cache [list][scan8[5 ]+1]); |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1063 AV_ZERO16(h->mvd_cache [list][scan8[7 ]+1]); |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1064 AV_ZERO16(h->mvd_cache [list][scan8[13]+1]); //FIXME remove past 3 (init somewhere else) |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1065 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]); |
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1066 AV_ZERO16(h->mvd_cache [list][scan8[12]]); |
10866 | 1067 if(h->slice_type_nos == FF_B_TYPE){ |
11167 | 1068 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1); |
10866 | 1069 |
1070 if(IS_DIRECT(top_type)){ | |
11203 | 1071 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1)); |
10866 | 1072 }else if(IS_8X8(top_type)){ |
11288 | 1073 int b8_xy = 4*top_xy; |
1074 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2]; | |
1075 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3]; | |
10866 | 1076 }else{ |
11203 | 1077 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1)); |
10866 | 1078 } |
1079 | |
1080 if(IS_DIRECT(left_type[0])) | |
11167 | 1081 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1; |
10866 | 1082 else if(IS_8X8(left_type[0])) |
11288 | 1083 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)]; |
10866 | 1084 else |
11167 | 1085 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1; |
10866 | 1086 |
1087 if(IS_DIRECT(left_type[1])) | |
11167 | 1088 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1; |
10866 | 1089 else if(IS_8X8(left_type[1])) |
11288 | 1090 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)]; |
10866 | 1091 else |
11167 | 1092 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1; |
10866 | 1093 } |
1094 } | |
11155 | 1095 } |
10866 | 1096 if(FRAME_MBAFF){ |
1097 #define MAP_MVS\ | |
1098 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\ | |
1099 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\ | |
1100 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\ | |
1101 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\ | |
1102 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\ | |
1103 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\ | |
1104 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\ | |
1105 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\ | |
1106 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\ | |
1107 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1]) | |
1108 if(MB_FIELD){ | |
1109 #define MAP_F2F(idx, mb_type)\ | |
1110 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ | |
1111 h->ref_cache[list][idx] <<= 1;\ | |
1112 h->mv_cache[list][idx][1] /= 2;\ | |
11268
e817a3c2ec2e
Replace /2 by faster >>1 as the mvd values are now all positive.
michael
parents:
11267
diff
changeset
|
1113 h->mvd_cache[list][idx][1] >>=1;\ |
10866 | 1114 } |
1115 MAP_MVS | |
1116 #undef MAP_F2F | |
1117 }else{ | |
1118 #define MAP_F2F(idx, mb_type)\ | |
1119 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\ | |
1120 h->ref_cache[list][idx] >>= 1;\ | |
1121 h->mv_cache[list][idx][1] <<= 1;\ | |
1122 h->mvd_cache[list][idx][1] <<= 1;\ | |
1123 } | |
1124 MAP_MVS | |
1125 #undef MAP_F2F | |
1126 } | |
1127 } | |
1128 } | |
1129 } | |
1130 #endif | |
1131 | |
10928 | 1132 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]); |
10907 | 1133 } |
1134 | |
10910 | 1135 /** |
1136 * | |
1137 * @returns non zero if the loop filter can be skiped | |
1138 */ | |
1139 static int fill_filter_caches(H264Context *h, int mb_type){ | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1140 MpegEncContext * const s = &h->s; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1141 const int mb_xy= h->mb_xy; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1142 int top_xy, left_xy[2]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1143 int top_type, left_type[2]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1144 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1145 top_xy = mb_xy - (s->mb_stride << MB_FIELD); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1146 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1147 //FIXME deblocking could skip the intra and nnz parts. |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1148 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1149 /* Wow, what a mess, why didn't they simplify the interlacing & intra |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1150 * stuff, I can't imagine that these complex rules are worth it. */ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1151 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1152 left_xy[1] = left_xy[0] = mb_xy-1; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1153 if(FRAME_MBAFF){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1154 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1155 const int curr_mb_field_flag = IS_INTERLACED(mb_type); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1156 if(s->mb_y&1){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1157 if (left_mb_field_flag != curr_mb_field_flag) { |
11023 | 1158 left_xy[0] -= s->mb_stride; |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1159 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1160 }else{ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1161 if(curr_mb_field_flag){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1162 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1163 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1164 if (left_mb_field_flag != curr_mb_field_flag) { |
11023 | 1165 left_xy[1] += s->mb_stride; |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1166 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1167 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1168 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1169 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1170 h->top_mb_xy = top_xy; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1171 h->left_mb_xy[0] = left_xy[0]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1172 h->left_mb_xy[1] = left_xy[1]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1173 { |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1174 //for sufficiently low qp, filtering wouldn't do anything |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1175 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1176 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1177 int qp = s->current_picture.qscale_table[mb_xy]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1178 if(qp <= qp_thresh |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1179 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh) |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1180 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1181 if(!FRAME_MBAFF) |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1182 return 1; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1183 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh) |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1184 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh)) |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1185 return 1; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1186 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1187 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1188 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1189 if(h->deblocking_filter == 2){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1190 h->top_type = top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1191 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1192 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1193 }else{ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1194 h->top_type = top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1195 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1196 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1197 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1198 if(IS_INTRA(mb_type)) |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1199 return 0; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1200 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1201 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1202 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]); |
11203 | 1203 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]); |
1204 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]); | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1205 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1206 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1207 h->cbp= h->cbp_table[mb_xy]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1208 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1209 { |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1210 int list; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1211 for(list=0; list<h->list_count; list++){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1212 int8_t *ref; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1213 int y, b_stride; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1214 int16_t (*mv_dst)[2]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1215 int16_t (*mv_src)[2]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1216 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1217 if(!USES_LIST(mb_type, list)){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1218 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4); |
11203 | 1219 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u); |
1220 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u); | |
1221 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u); | |
1222 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u); | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1223 continue; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1224 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1225 |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1226 ref = &s->current_picture.ref_index[list][4*mb_xy]; |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1227 { |
11165 | 1228 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2); |
11203 | 1229 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101); |
1230 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101); | |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1231 ref += 2; |
11203 | 1232 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101); |
1233 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101); | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1234 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1235 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1236 b_stride = h->b_stride; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1237 mv_dst = &h->mv_cache[list][scan8[0]]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1238 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1239 for(y=0; y<4; y++){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1240 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1241 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1242 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1243 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1244 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1245 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1246 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1247 /* |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1248 0 . T T. T T T T |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1249 1 L . .L . . . . |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1250 2 L . .L . . . . |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1251 3 . T TL . . . . |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1252 4 L . .L . . . . |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1253 5 L . .. . . . . |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1254 */ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1255 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec) |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1256 if(top_type){ |
11203 | 1257 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]); |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1258 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1259 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1260 if(left_type[0]){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1261 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1262 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1263 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1264 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1265 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1266 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1267 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1268 if(!CABAC && h->pps.transform_8x8_mode){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1269 if(IS_8x8DCT(top_type)){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1270 h->non_zero_count_cache[4+8*0]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1271 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1272 h->non_zero_count_cache[6+8*0]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1273 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1274 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1275 if(IS_8x8DCT(left_type[0])){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1276 h->non_zero_count_cache[3+8*1]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1277 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1278 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1279 if(IS_8x8DCT(left_type[1])){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1280 h->non_zero_count_cache[3+8*3]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1281 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1282 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1283 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1284 if(IS_8x8DCT(mb_type)){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1285 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1286 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1287 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1288 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1289 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1290 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1291 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1292 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1293 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1294 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1295 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1296 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1297 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1298 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1299 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1300 int list; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1301 for(list=0; list<h->list_count; list++){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1302 if(USES_LIST(top_type, list)){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1303 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride; |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1304 const int b8_xy= 4*top_xy + 2; |
11165 | 1305 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2); |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1306 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]); |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1307 h->ref_cache[list][scan8[0] + 0 - 1*8]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1308 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1309 h->ref_cache[list][scan8[0] + 2 - 1*8]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1310 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]]; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1311 }else{ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1312 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]); |
11203 | 1313 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u); |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1314 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1315 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1316 if(!IS_INTERLACED(mb_type^left_type[0])){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1317 if(USES_LIST(left_type[0], list)){ |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1318 const int b_xy= h->mb2b_xy[left_xy[0]] + 3; |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1319 const int b8_xy= 4*left_xy[0] + 1; |
11165 | 1320 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2); |
11203 | 1321 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]); |
1322 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]); | |
1323 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]); | |
1324 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]); | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1325 h->ref_cache[list][scan8[0] - 1 + 0 ]= |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1326 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]]; |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1327 h->ref_cache[list][scan8[0] - 1 +16 ]= |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1328 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]]; |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1329 }else{ |
11203 | 1330 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]); |
1331 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]); | |
1332 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]); | |
1333 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]); | |
11022
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1334 h->ref_cache[list][scan8[0] - 1 + 0 ]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1335 h->ref_cache[list][scan8[0] - 1 + 8 ]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1336 h->ref_cache[list][scan8[0] - 1 + 16 ]= |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1337 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED; |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1338 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1339 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1340 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1341 } |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1342 |
44529d4292ce
Split fill_caches() between loopfilter & decode, the 2 no longer where common
michael
parents:
11020
diff
changeset
|
1343 return 0; |
10907 | 1344 } |
1345 | |
10866 | 1346 /** |
1347 * gets the predicted intra4x4 prediction mode. | |
1348 */ | |
1349 static inline int pred_intra_mode(H264Context *h, int n){ | |
1350 const int index8= scan8[n]; | |
1351 const int left= h->intra4x4_pred_mode_cache[index8 - 1]; | |
1352 const int top = h->intra4x4_pred_mode_cache[index8 - 8]; | |
1353 const int min= FFMIN(left, top); | |
1354 | |
1355 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min); | |
1356 | |
1357 if(min<0) return DC_PRED; | |
1358 else return min; | |
1359 } | |
1360 | |
1361 static inline void write_back_non_zero_count(H264Context *h){ | |
1362 const int mb_xy= h->mb_xy; | |
1363 | |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1364 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]); |
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1365 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]); |
11203 | 1366 AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]); |
1367 AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]); | |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1368 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]); |
10866 | 1369 } |
1370 | |
1371 static inline void write_back_motion(H264Context *h, int mb_type){ | |
1372 MpegEncContext * const s = &h->s; | |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1373 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy |
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1374 const int b8_xy= 4*h->mb_xy; |
10866 | 1375 int list; |
1376 | |
1377 if(!USES_LIST(mb_type, 0)) | |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1378 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1); |
10866 | 1379 |
1380 for(list=0; list<h->list_count; list++){ | |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1381 int y, b_stride; |
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1382 int16_t (*mv_dst)[2]; |
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1383 int16_t (*mv_src)[2]; |
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1384 |
10866 | 1385 if(!USES_LIST(mb_type, list)) |
1386 continue; | |
1387 | |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1388 b_stride = h->b_stride; |
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1389 mv_dst = &s->current_picture.motion_val[list][b_xy]; |
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1390 mv_src = &h->mv_cache[list][scan8[0]]; |
10866 | 1391 for(y=0; y<4; y++){ |
11002
1c8892d7a090
H.264: Use 64-/128-bit write-combining macros for copies
astrange
parents:
10987
diff
changeset
|
1392 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y); |
10866 | 1393 } |
10893
2aafcafbe1f0
Replace cabac checks in inline functions from h264.h with constants.
michael
parents:
10883
diff
changeset
|
1394 if( CABAC ) { |
11283
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1395 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]]; |
11277
c12d6c6c027e
Change mvd_cache & mvd_table to 8bit, this is overall a bit faster
michael
parents:
11268
diff
changeset
|
1396 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]]; |
10866 | 1397 if(IS_SKIP(mb_type)) |
11283
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1398 AV_ZERO128(mvd_dst); |
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1399 else{ |
853e93a50fe5
Cut the size of mvd_table by yet another factor of 2.
michael
parents:
11282
diff
changeset
|
1400 AV_COPY64(mvd_dst, mvd_src + 8*3); |
11284
aaca4b58880f
unroll tiny and trivial loop. Same speed but clearer.
michael
parents:
11283
diff
changeset
|
1401 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0); |
aaca4b58880f
unroll tiny and trivial loop. Same speed but clearer.
michael
parents:
11283
diff
changeset
|
1402 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1); |
aaca4b58880f
unroll tiny and trivial loop. Same speed but clearer.
michael
parents:
11283
diff
changeset
|
1403 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2); |
10866 | 1404 } |
1405 } | |
1406 | |
1407 { | |
1408 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy]; | |
11292
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1409 ref_index[0+0*2]= h->ref_cache[list][scan8[0]]; |
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1410 ref_index[1+0*2]= h->ref_cache[list][scan8[4]]; |
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1411 ref_index[0+1*2]= h->ref_cache[list][scan8[8]]; |
411ab09ada91
Get rid of mb2b8_xy and b8_stride, change arrays organized based on b8_stride to
michael
parents:
11288
diff
changeset
|
1412 ref_index[1+1*2]= h->ref_cache[list][scan8[12]]; |
10866 | 1413 } |
1414 } | |
1415 | |
10893
2aafcafbe1f0
Replace cabac checks in inline functions from h264.h with constants.
michael
parents:
10883
diff
changeset
|
1416 if(h->slice_type_nos == FF_B_TYPE && CABAC){ |
10866 | 1417 if(IS_8X8(mb_type)){ |
11288 | 1418 uint8_t *direct_table = &h->direct_table[4*h->mb_xy]; |
1419 direct_table[1] = h->sub_mb_type[1]>>1; | |
1420 direct_table[2] = h->sub_mb_type[2]>>1; | |
1421 direct_table[3] = h->sub_mb_type[3]>>1; | |
10866 | 1422 } |
1423 } | |
1424 } | |
1425 | |
1426 static inline int get_dct8x8_allowed(H264Context *h){ | |
1427 if(h->sps.direct_8x8_inference_flag) | |
11203 | 1428 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL)); |
10866 | 1429 else |
11203 | 1430 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL)); |
10866 | 1431 } |
1432 | |
1433 /** | |
1434 * decodes a P_SKIP or B_SKIP macroblock | |
1435 */ | |
1436 static void decode_mb_skip(H264Context *h){ | |
1437 MpegEncContext * const s = &h->s; | |
1438 const int mb_xy= h->mb_xy; | |
1439 int mb_type=0; | |
1440 | |
10906 | 1441 memset(h->non_zero_count[mb_xy], 0, 32); |
10866 | 1442 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui |
1443 | |
1444 if(MB_FIELD) | |
1445 mb_type|= MB_TYPE_INTERLACED; | |
1446 | |
1447 if( h->slice_type_nos == FF_B_TYPE ) | |
1448 { | |
1449 // just for fill_caches. pred_direct_motion will set the real mb_type | |
11145
8805efb4c3fd
Move setting MB_TYPE_L0L1 for direct MBs up, this is simpler.
michael
parents:
11144
diff
changeset
|
1450 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP; |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
1451 if(h->direct_spatial_mv_pred){ |
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
1452 fill_decode_neighbors(h, mb_type); |
10907 | 1453 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ... |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
1454 } |
10866 | 1455 ff_h264_pred_direct_motion(h, &mb_type); |
1456 mb_type|= MB_TYPE_SKIP; | |
1457 } | |
1458 else | |
1459 { | |
1460 int mx, my; | |
1461 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP; | |
1462 | |
11183
d1a855cb0a0c
Split setting neighboring MBs from fill_decode_caches()
michael
parents:
11167
diff
changeset
|
1463 fill_decode_neighbors(h, mb_type); |
10907 | 1464 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ... |
10866 | 1465 pred_pskip_motion(h, &mx, &my); |
1466 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1); | |
1467 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4); | |
1468 } | |
1469 | |
1470 write_back_motion(h, mb_type); | |
1471 s->current_picture.mb_type[mb_xy]= mb_type; | |
1472 s->current_picture.qscale_table[mb_xy]= s->qscale; | |
1473 h->slice_table[ mb_xy ]= h->slice_num; | |
1474 h->prev_mb_skipped= 1; | |
1475 } | |
1476 | |
10883
4701b9355d93
Add forgotten include of h264_mvpred.h to h264.h.
michael
parents:
10869
diff
changeset
|
1477 #include "h264_mvpred.h" //For pred_pskip_motion() |
4701b9355d93
Add forgotten include of h264_mvpred.h to h264.h.
michael
parents:
10869
diff
changeset
|
1478 |
7760 | 1479 #endif /* AVCODEC_H264_H */ |