2 * Generic DCT based hybrid video encoder
3 * Copyright (c) 2000,2001 Gerard Lantau.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 /* Macros for picture code type. */
24 #define S_TYPE 4 //S(GMC)-VOP MPEG4
26 #define FRAME_SKIPED 100 // return value for header parsers if frame is not coded
34 #define MPEG_BUF_SIZE (16 * 1024)
36 #define QMAT_SHIFT_MMX 19
41 #define REORDER_BUFFER_SIZE (FF_MAX_B_FRAMES+2)
43 typedef struct Predictor{
49 typedef struct RateControlEntry{
56 uint64_t expected_bits;
61 typedef struct RateControlContext{
64 RateControlEntry *entry;
67 typedef struct ReorderBuffer{
73 int picture_in_gop_number;
76 typedef struct MpegEncContext {
77 struct AVCodecContext *avctx;
78 /* the following parameters must be initialized before encoding */
79 int width, height; /* picture size. must be a multiple of 16 */
81 int frame_rate; /* number of frames per second */
82 int intra_only; /* if true, only intra pictures are generated */
83 int bit_rate; /* wanted bit rate */
84 int bit_rate_tolerance; /* amount of +- bits (>0)*/
85 enum OutputFormat out_format; /* output format */
86 int h263_plus; /* h263 plus headers */
87 int h263_rv10; /* use RV10 variation for H263 */
88 int h263_pred; /* use mpeg4/h263 ac/dc predictions */
89 int h263_msmpeg4; /* generate MSMPEG4 compatible stream */
90 int h263_intel; /* use I263 intel h263 header */
91 int fixed_qscale; /* fixed qscale if non zero */
92 float qcompress; /* amount of qscale change between easy & hard scenes (0.0-1.0) */
93 float qblur; /* amount of qscale smoothing over time (0.0-1.0) */
94 int qmin; /* min qscale */
95 int qmax; /* max qscale */
96 int max_qdiff; /* max qscale difference between frames */
97 int encoding; /* true if we are encoding (vs decoding) */
98 int flags; /* AVCodecContext.flags (HQ, MV4, ...) */
99 int force_input_type;/* 0= no force, otherwise I_TYPE, P_TYPE, ... */
100 int max_b_frames; /* max number of b-frames for encoding */
101 float b_quant_factor;/* qscale factor between ips and b frames */
103 int b_frame_strategy;
104 /* the following fields are managed internally by the encoder */
109 /* sequence parameters */
110 int context_initialized;
111 int input_picture_number;
112 int input_picture_in_gop_number; /* 0-> first pic in gop, ... */
114 int fake_picture_number; /* picture number at the bitstream frame rate */
115 int gop_picture_number; /* index of the first picture of a GOP based on fake_pic_num & mpeg1 specific */
116 int picture_in_gop_number; /* 0-> first pic in gop, ... */
117 int b_frames_since_non_b; /* used for encoding, relative to not yet reordered input */
118 int mb_width, mb_height; /* number of MBs horizontally & vertically */
119 int mb_num; /* number of MBs of a picture */
120 int linesize; /* line size, in bytes, may be different from width */
121 UINT8 *new_picture[3]; /* picture to be compressed */
122 UINT8 *picture_buffer[REORDER_BUFFER_SIZE][3]; /* internal buffers used for reordering of input pictures */
123 int picture_buffer_index;
124 ReorderBuffer coded_order[REORDER_BUFFER_SIZE];
125 UINT8 *last_picture[3]; /* previous picture */
126 UINT8 *last_picture_base[3]; /* real start of the picture */
127 UINT8 *next_picture[3]; /* previous picture (for bidir pred) */
128 UINT8 *next_picture_base[3]; /* real start of the picture */
129 UINT8 *aux_picture[3]; /* aux picture (for B frames only) */
130 UINT8 *aux_picture_base[3]; /* real start of the picture */
131 UINT8 *current_picture[3]; /* buffer to store the decompressed current picture */
132 int last_dc[3]; /* last DC values for MPEG1 */
133 INT16 *dc_val[3]; /* used for mpeg4 DC prediction, all 3 arrays must be continuous */
134 int y_dc_scale, c_dc_scale;
135 UINT8 *coded_block; /* used for coded block pattern prediction (msmpeg4v3, wmv1)*/
136 INT16 (*ac_val[3])[16]; /* used for for mpeg4 AC prediction, all 3 arrays must be continuous */
138 int mb_skiped; /* MUST BE SET only during DECODING */
139 UINT8 *mbskip_table; /* used to avoid copy if macroblock skipped (for black regions for example)
140 and used for b-frame encoding & decoding (contains skip table of next P Frame) */
141 UINT8 *mbintra_table; /* used to avoid setting {ac, dc, cbp}-pred stuff to zero on inter MB decoding */
143 int input_qscale; /* qscale prior to reordering of frames */
144 int input_pict_type; /* pict_type prior to reordering of frames */
145 int force_type; /* 0= no force, otherwise I_TYPE, P_TYPE, ... */
147 int last_non_b_qscale; /* QP of last non b frame used for b frame qscale*/
148 int pict_type; /* I_TYPE, P_TYPE, B_TYPE, ... */
149 int last_non_b_pict_type; /* used for mpeg4 gmc b-frames & ratecontrol */
150 int frame_rate_index;
151 /* motion compensation */
153 int h263_long_vectors; /* use horrible h263v1 long vector mode */
155 int f_code; /* forward MV resolution */
156 int b_code; /* backward MV resolution for B Frames (mpeg4) */
157 INT16 (*motion_val)[2]; /* used for MV prediction (4MV per MB) */
158 INT16 (*p_mv_table)[2]; /* MV table (1MV per MB) p-frame encoding */
159 INT16 (*last_p_mv_table)[2]; /* MV table (1MV per MB) p-frame encoding */
160 INT16 (*b_forw_mv_table)[2]; /* MV table (1MV per MB) forward mode b-frame encoding */
161 INT16 (*b_back_mv_table)[2]; /* MV table (1MV per MB) backward mode b-frame encoding */
162 INT16 (*b_bidir_forw_mv_table)[2]; /* MV table (1MV per MB) bidir mode b-frame encoding */
163 INT16 (*b_bidir_back_mv_table)[2]; /* MV table (1MV per MB) bidir mode b-frame encoding */
164 INT16 (*b_direct_forw_mv_table)[2];/* MV table (1MV per MB) direct mode b-frame encoding */
165 INT16 (*b_direct_back_mv_table)[2];/* MV table (1MV per MB) direct mode b-frame encoding */
166 INT16 (*b_direct_mv_table)[2]; /* MV table (1MV per MB) direct mode b-frame encoding */
167 int me_method; /* ME algorithm */
168 uint8_t *me_scratchpad; /* data area for the me algo, so that the ME doesnt need to malloc/free */
170 #define MV_DIR_BACKWARD 1
171 #define MV_DIR_FORWARD 2
172 #define MV_DIRECT 4 // bidirectional mode where the difference equals the MV of the last P/S/I-Frame (mpeg4)
174 #define MV_TYPE_16X16 0 /* 1 vector for the whole mb */
175 #define MV_TYPE_8X8 1 /* 4 vectors (h263, mpeg4 4MV) */
176 #define MV_TYPE_16X8 2 /* 2 vectors, one per 16x8 block */
177 #define MV_TYPE_FIELD 3 /* 2 vectors, one per field */
178 #define MV_TYPE_DMV 4 /* 2 vectors, special mpeg2 Dual Prime Vectors */
179 /* motion vectors for a macroblock
180 first coordinate : 0 = forward 1 = backward
181 second " : depend on type
182 third " : 0 = x, 1 = y
185 int field_select[2][2];
186 int last_mv[2][2][2]; /* last MV, used for MV prediction in MPEG1 & B-frame MPEG4 */
187 UINT16 (*mv_penalty)[MAX_MV*2+1]; /* amount of bits needed to encode a MV, used for ME */
188 UINT8 *fcode_tab; /* smallest fcode needed for each MV */
191 int no_rounding; /* apply no rounding to motion compensation (MPEG4, msmpeg4, ...)
192 for b-frames rounding mode is allways 0 */
194 /* macroblock layer */
198 UINT16 *mb_var; /* Table for MB variances */
199 UINT8 *mb_type; /* Table for MB type */
200 #define MB_TYPE_INTRA 0x01
201 #define MB_TYPE_INTER 0x02
202 #define MB_TYPE_INTER4V 0x04
203 #define MB_TYPE_SKIPED 0x08
204 #define MB_TYPE_DIRECT 0x10
205 #define MB_TYPE_FORWARD 0x20
206 #define MB_TYPE_BACKWARD 0x40
207 #define MB_TYPE_BIDIR 0x80
209 int block_index[6]; /* index to current MB in block based arrays with edges*/
212 /* matrix transmitted in the bitstream */
213 UINT16 intra_matrix[64];
214 UINT16 chroma_intra_matrix[64];
215 UINT16 non_intra_matrix[64];
216 UINT16 chroma_non_intra_matrix[64];
217 /* precomputed matrix (combine qscale and DCT renorm) */
218 int q_intra_matrix[64];
219 int q_non_intra_matrix[64];
220 /* identical to the above but for MMX & these are not permutated */
221 UINT16 __align8 q_intra_matrix16[64];
222 UINT16 __align8 q_non_intra_matrix16[64];
223 int block_last_index[6]; /* last non zero coefficient in block */
225 void *opaque; /* private data for the user */
227 /* bit rate control */
228 int I_frame_bits; //FIXME used in mpeg12 ...
229 int avg_mb_var; /* average MB variance for current frame */
230 int mc_mb_var; /* motion compensated MB variance for current frame */
231 int last_non_b_mc_mb_var;/* motion compensated MB variance for last non b frame */
234 int frame_bits; /* bits used for the current frame */
235 int pb_frame_bits; /* bits of the last b...bp group */
238 double qsum; /* sum of qscales */
239 double qcount; /* count of qscales */
240 double short_term_qsum; /* sum of recent qscales */
241 double short_term_qcount; /* count of recent qscales */
242 RateControlContext rc_context;
244 /* statistics, used for 2-pass encoding */
252 int misc_bits; // cbp, mb_type
253 int last_bits; //temp var used for calculating the above vars
260 /* H.263+ specific */
263 int h263_aic; /* Advanded INTRA Coding (AIC) */
264 int h263_aic_dir; /* AIC direction: 0 = left, 1 = top */
267 int time_increment_resolution;
268 int time_increment_bits; /* number of bits to represent the fractional part of time */
270 int time_base; /* time in seconds of last I,P,S Frame */
271 int64_t time; /* time of current frame */
272 int64_t last_non_b_time;
273 uint16_t pp_time; /* time distance between the last 2 p,s,i frames */
274 uint16_t bp_time; /* time distance between the last b and p,s,i frame */
276 int vol_sprite_usage;
281 int sprite_brightness_change;
282 int num_sprite_warping_points;
283 int real_sprite_warping_points;
284 int sprite_offset[2][2];
285 int sprite_delta[2][2][2];
286 int sprite_shift[2][2];
289 int quarter_sample; /* 1->qpel, 0->half pel ME/MC */
293 int aspect_ratio_info;
294 int sprite_warping_accuracy;
295 int low_latency_sprite;
299 int low_delay; /* no reordering needed / has no b-frames */
302 /* divx specific, used to workaround (many) bugs in divx5 */
305 #define BITSTREAM_BUFFER_SIZE 1024*256
306 uint8_t *bitstream_buffer; //Divx 5.01 puts several frames in a single one, this is used to reorder them
307 int bitstream_buffer_size;
310 int rv10_version; /* RV10 version: 0 or 3 */
311 int rv10_first_dc_coded[3];
314 struct MJpegContext *mjpeg_ctx;
315 int mjpeg_vsample[3]; /* vertical sampling factors, default = {2, 1, 1} */
316 int mjpeg_hsample[3]; /* horizontal sampling factors, default = {2, 1, 1} */
317 int mjpeg_write_tables; /* do we want to have quantisation- and
318 huffmantables in the jpeg file ? */
320 /* MSMPEG4 specific */
323 int rl_chroma_table_index;
325 int use_skip_mb_code;
326 int slice_height; /* in macroblocks */
327 int first_slice_line; /* used in mpeg4 too to handle resync markers */
328 int flipflop_rounding;
330 int msmpeg4_version; /* 1=mp41, 2=mp42, 3=mp43/divx3 */
331 /* decompression specific */
334 /* MPEG2 specific - I wish I had not to support this mess. */
335 int progressive_sequence;
336 int mpeg_f_code[2][2];
337 int picture_structure;
339 #define PICT_TOP_FIELD 1
340 #define PICT_BOTTOM_FIELD 2
343 int intra_dc_precision;
344 int frame_pred_frame_dct;
346 int concealment_motion_vectors;
348 int intra_vlc_format;
350 int repeat_first_field;
352 int progressive_frame;
360 /* These are explained on avcodec.h */
362 int rtp_payload_size;
363 void (*rtp_callback)(void *data, int size, int packet_number);
365 UINT8 *ptr_last_mb_line;
366 UINT32 mb_line_avgsize;
368 DCTELEM (*block)[64]; /* points to one of the following blocks */
369 DCTELEM blocks[2][6][64] __align8; // for HQ mode we need to keep the best block
370 void (*dct_unquantize_mpeg1)(struct MpegEncContext *s,
371 DCTELEM *block, int n, int qscale);
372 void (*dct_unquantize_mpeg2)(struct MpegEncContext *s,
373 DCTELEM *block, int n, int qscale);
374 void (*dct_unquantize_h263)(struct MpegEncContext *s,
375 DCTELEM *block, int n, int qscale);
376 void (*dct_unquantize)(struct MpegEncContext *s, // unquantizer to use (mpeg4 can use both)
377 DCTELEM *block, int n, int qscale);
380 int MPV_common_init(MpegEncContext *s);
381 void MPV_common_end(MpegEncContext *s);
382 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]);
383 void MPV_frame_start(MpegEncContext *s);
384 void MPV_frame_end(MpegEncContext *s);
386 void MPV_common_init_mmx(MpegEncContext *s);
390 void ff_estimate_p_frame_motion(MpegEncContext * s,
392 void ff_estimate_b_frame_motion(MpegEncContext * s,
394 int ff_get_best_fcode(MpegEncContext * s, int16_t (*mv_table)[2], int type);
395 void ff_fix_long_p_mvs(MpegEncContext * s);
396 void ff_fix_long_b_mvs(MpegEncContext * s, int16_t (*mv_table)[2], int f_code, int type);
399 extern INT16 default_intra_matrix[64];
400 extern INT16 default_non_intra_matrix[64];
402 void mpeg1_encode_picture_header(MpegEncContext *s, int picture_number);
403 void mpeg1_encode_mb(MpegEncContext *s,
404 DCTELEM block[6][64],
405 int motion_x, int motion_y);
406 void mpeg1_encode_init(MpegEncContext *s);
410 /* run length table */
414 typedef struct RLTable {
415 int n; /* number of entries of table_vlc minus 1 */
416 int last; /* number of values for last = 0 */
417 const UINT16 (*table_vlc)[2];
418 const INT8 *table_run;
419 const INT8 *table_level;
420 UINT8 *index_run[2]; /* encoding only */
421 INT8 *max_level[2]; /* encoding & decoding */
422 INT8 *max_run[2]; /* encoding & decoding */
423 VLC vlc; /* decoding only */
426 void init_rl(RLTable *rl);
427 void init_vlc_rl(RLTable *rl);
429 static inline int get_rl_index(const RLTable *rl, int last, int run, int level)
432 index = rl->index_run[last][run];
435 if (level > rl->max_level[last][run])
437 return index + level - 1;
440 void h263_encode_mb(MpegEncContext *s,
441 DCTELEM block[6][64],
442 int motion_x, int motion_y);
443 void mpeg4_encode_mb(MpegEncContext *s,
444 DCTELEM block[6][64],
445 int motion_x, int motion_y);
446 void h263_encode_picture_header(MpegEncContext *s, int picture_number);
447 int h263_encode_gob_header(MpegEncContext * s, int mb_line);
448 void h263_dc_scale(MpegEncContext *s);
449 INT16 *h263_pred_motion(MpegEncContext * s, int block,
451 void mpeg4_pred_ac(MpegEncContext * s, INT16 *block, int n,
453 void ff_set_mpeg4_time(MpegEncContext * s, int picture_number);
454 void mpeg4_encode_picture_header(MpegEncContext *s, int picture_number);
455 void h263_encode_init(MpegEncContext *s);
457 void h263_decode_init_vlc(MpegEncContext *s);
458 int h263_decode_picture_header(MpegEncContext *s);
459 int h263_decode_gob_header(MpegEncContext *s);
460 int mpeg4_decode_picture_header(MpegEncContext * s);
461 int intel_h263_decode_picture_header(MpegEncContext *s);
462 int h263_decode_mb(MpegEncContext *s,
463 DCTELEM block[6][64]);
464 int h263_get_picture_format(int width, int height);
467 void rv10_encode_picture_header(MpegEncContext *s, int picture_number);
468 int rv_decode_dc(MpegEncContext *s, int n);
471 void msmpeg4_encode_picture_header(MpegEncContext * s, int picture_number);
472 void msmpeg4_encode_ext_header(MpegEncContext * s);
473 void msmpeg4_encode_mb(MpegEncContext * s,
474 DCTELEM block[6][64],
475 int motion_x, int motion_y);
476 void msmpeg4_dc_scale(MpegEncContext * s);
477 int msmpeg4_decode_picture_header(MpegEncContext * s);
478 int msmpeg4_decode_ext_header(MpegEncContext * s, int buf_size);
479 int msmpeg4_decode_mb(MpegEncContext *s,
480 DCTELEM block[6][64]);
481 int msmpeg4_decode_init_vlc(MpegEncContext *s);
485 int mjpeg_init(MpegEncContext *s);
486 void mjpeg_close(MpegEncContext *s);
487 void mjpeg_encode_mb(MpegEncContext *s,
488 DCTELEM block[6][64]);
489 void mjpeg_picture_header(MpegEncContext *s);
490 void mjpeg_picture_trailer(MpegEncContext *s);
493 int ff_rate_control_init(MpegEncContext *s);
494 int ff_rate_estimate_qscale(MpegEncContext *s);
495 int ff_rate_estimate_qscale_pass2(MpegEncContext *s);
496 void ff_write_pass1_stats(MpegEncContext *s);
497 void ff_rate_control_uninit(MpegEncContext *s);