1 /******************************************************************** 2 * * 3 * THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. * 4 * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS * 5 * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * 6 * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * 7 * * 8 * THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2009 * 9 * by the Xiph.Org Foundation http://www.xiph.org/ * 10 * * 11 ******************************************************************** 12 13 function: 14 last mod: $Id$ 15 16 ********************************************************************/ 17 #if !defined(_encint_H) 18 # define _encint_H (1) 19 # if defined(HAVE_CONFIG_H) 20 # include "config.h" 21 # endif 22 # include "theora/theoraenc.h" 23 # include "internal.h" 24 # include "ocintrin.h" 25 # include "mathops.h" 26 # include "enquant.h" 27 # include "huffenc.h" 28 /*# define OC_COLLECT_METRICS*/ 29 30 31 32 typedef oc_mv oc_mv2[2]; 33 34 typedef struct oc_enc_opt_vtable oc_enc_opt_vtable; 35 typedef struct oc_mb_enc_info oc_mb_enc_info; 36 typedef struct oc_mode_scheme_chooser oc_mode_scheme_chooser; 37 typedef struct oc_iir_filter oc_iir_filter; 38 typedef struct oc_frame_metrics oc_frame_metrics; 39 typedef struct oc_rc_state oc_rc_state; 40 typedef struct th_enc_ctx oc_enc_ctx; 41 typedef struct oc_token_checkpoint oc_token_checkpoint; 42 43 44 45 /*Constants for the packet-out state machine specific to the encoder.*/ 46 47 /*Next packet to emit: Data packet, but none are ready yet.*/ 48 #define OC_PACKET_EMPTY (0) 49 /*Next packet to emit: Data packet, and one is ready.*/ 50 #define OC_PACKET_READY (1) 51 52 /*All features enabled.*/ 53 #define OC_SP_LEVEL_SLOW (0) 54 /*Enable early skip.*/ 55 #define OC_SP_LEVEL_EARLY_SKIP (1) 56 /*Disable motion compensation.*/ 57 #define OC_SP_LEVEL_NOMC (2) 58 /*Maximum valid speed level.*/ 59 #define OC_SP_LEVEL_MAX (2) 60 61 62 /*The bits used for each of the MB mode codebooks.*/ 63 extern const unsigned char OC_MODE_BITS[2][OC_NMODES]; 64 65 /*The bits used for each of the MV codebooks.*/ 66 extern const unsigned char OC_MV_BITS[2][64]; 67 68 /*The minimum value that can be stored in a SB run for each codeword. 69 The last entry is the upper bound on the length of a single SB run.*/ 70 extern const ogg_uint16_t OC_SB_RUN_VAL_MIN[8]; 71 /*The bits used for each SB run codeword.*/ 72 extern const unsigned char OC_SB_RUN_CODE_NBITS[7]; 73 74 /*The bits used for each block run length (starting with 1).*/ 75 extern const unsigned char OC_BLOCK_RUN_CODE_NBITS[30]; 76 77 78 79 /*Encoder specific functions with accelerated variants.*/ 80 struct oc_enc_opt_vtable{ 81 unsigned (*frag_sad)(const unsigned char *_src, 82 const unsigned char *_ref,int _ystride); 83 unsigned (*frag_sad_thresh)(const unsigned char *_src, 84 const unsigned char *_ref,int _ystride,unsigned _thresh); 85 unsigned (*frag_sad2_thresh)(const unsigned char *_src, 86 const unsigned char *_ref1,const unsigned char *_ref2,int _ystride, 87 unsigned _thresh); 88 unsigned (*frag_satd_thresh)(const unsigned char *_src, 89 const unsigned char *_ref,int _ystride,unsigned _thresh); 90 unsigned (*frag_satd2_thresh)(const unsigned char *_src, 91 const unsigned char *_ref1,const unsigned char *_ref2,int _ystride, 92 unsigned _thresh); 93 unsigned (*frag_intra_satd)(const unsigned char *_src,int _ystride); 94 void (*frag_sub)(ogg_int16_t _diff[64],const unsigned char *_src, 95 const unsigned char *_ref,int _ystride); 96 void (*frag_sub_128)(ogg_int16_t _diff[64], 97 const unsigned char *_src,int _ystride); 98 void (*frag_copy2)(unsigned char *_dst, 99 const unsigned char *_src1,const unsigned char *_src2,int _ystride); 100 void (*frag_recon_intra)(unsigned char *_dst,int _ystride, 101 const ogg_int16_t _residue[64]); 102 void (*frag_recon_inter)(unsigned char *_dst, 103 const unsigned char *_src,int _ystride,const ogg_int16_t _residue[64]); 104 void (*fdct8x8)(ogg_int16_t _y[64],const ogg_int16_t _x[64]); 105 }; 106 107 108 void oc_enc_vtable_init(oc_enc_ctx *_enc); 109 110 111 112 /*Encoder-specific macroblock information.*/ 113 struct oc_mb_enc_info{ 114 /*Neighboring macro blocks that have MVs available from the current frame.*/ 115 unsigned cneighbors[4]; 116 /*Neighboring macro blocks to use for MVs from the previous frame.*/ 117 unsigned pneighbors[4]; 118 /*The number of current-frame neighbors.*/ 119 unsigned char ncneighbors; 120 /*The number of previous-frame neighbors.*/ 121 unsigned char npneighbors; 122 /*Flags indicating which MB modes have been refined.*/ 123 unsigned char refined; 124 /*Motion vectors for a macro block for the current frame and the 125 previous two frames. 126 Each is a set of 2 vectors against OC_FRAME_GOLD and OC_FRAME_PREV, which 127 can be used to estimate constant velocity and constant acceleration 128 predictors. 129 Uninitialized MVs are (0,0).*/ 130 oc_mv2 analysis_mv[3]; 131 /*Current unrefined analysis MVs.*/ 132 oc_mv unref_mv[2]; 133 /*Unrefined block MVs.*/ 134 oc_mv block_mv[4]; 135 /*Refined block MVs.*/ 136 oc_mv ref_mv[4]; 137 /*Minimum motion estimation error from the analysis stage.*/ 138 ogg_uint16_t error[2]; 139 /*MB error for half-pel refinement for each frame type.*/ 140 unsigned satd[2]; 141 /*Block error for half-pel refinement.*/ 142 unsigned block_satd[4]; 143 }; 144 145 146 147 /*State machine to estimate the opportunity cost of coding a MB mode.*/ 148 struct oc_mode_scheme_chooser{ 149 /*Pointers to the a list containing the index of each mode in the mode 150 alphabet used by each scheme. 151 The first entry points to the dynamic scheme0_ranks, while the remaining 7 152 point to the constant entries stored in OC_MODE_SCHEMES.*/ 153 const unsigned char *mode_ranks[8]; 154 /*The ranks for each mode when coded with scheme 0. 155 These are optimized so that the more frequent modes have lower ranks.*/ 156 unsigned char scheme0_ranks[OC_NMODES]; 157 /*The list of modes, sorted in descending order of frequency, that 158 corresponds to the ranks above.*/ 159 unsigned char scheme0_list[OC_NMODES]; 160 /*The number of times each mode has been chosen so far.*/ 161 int mode_counts[OC_NMODES]; 162 /*The list of mode coding schemes, sorted in ascending order of bit cost.*/ 163 unsigned char scheme_list[8]; 164 /*The number of bits used by each mode coding scheme.*/ 165 ptrdiff_t scheme_bits[8]; 166 }; 167 168 169 void oc_mode_scheme_chooser_init(oc_mode_scheme_chooser *_chooser); 170 171 172 173 /*A 2nd order low-pass Bessel follower. 174 We use this for rate control because it has fast reaction time, but is 175 critically damped.*/ 176 struct oc_iir_filter{ 177 ogg_int32_t c[2]; 178 ogg_int64_t g; 179 ogg_int32_t x[2]; 180 ogg_int32_t y[2]; 181 }; 182 183 184 185 /*The 2-pass metrics associated with a single frame.*/ 186 struct oc_frame_metrics{ 187 /*The log base 2 of the scale factor for this frame in Q24 format.*/ 188 ogg_int32_t log_scale; 189 /*The number of application-requested duplicates of this frame.*/ 190 unsigned dup_count:31; 191 /*The frame type from pass 1.*/ 192 unsigned frame_type:1; 193 }; 194 195 196 197 /*Rate control state information.*/ 198 struct oc_rc_state{ 199 /*The target average bits per frame.*/ 200 ogg_int64_t bits_per_frame; 201 /*The current buffer fullness (bits available to be used).*/ 202 ogg_int64_t fullness; 203 /*The target buffer fullness. 204 This is where we'd like to be by the last keyframe the appears in the next 205 buf_delay frames.*/ 206 ogg_int64_t target; 207 /*The maximum buffer fullness (total size of the buffer).*/ 208 ogg_int64_t max; 209 /*The log of the number of pixels in a frame in Q57 format.*/ 210 ogg_int64_t log_npixels; 211 /*The exponent used in the rate model in Q8 format.*/ 212 unsigned exp[2]; 213 /*The number of frames to distribute the buffer usage over.*/ 214 int buf_delay; 215 /*The total drop count from the previous frame. 216 This includes duplicates explicitly requested via the 217 TH_ENCCTL_SET_DUP_COUNT API as well as frames we chose to drop ourselves.*/ 218 ogg_uint32_t prev_drop_count; 219 /*The log of an estimated scale factor used to obtain the real framerate, for 220 VFR sources or, e.g., 12 fps content doubled to 24 fps, etc.*/ 221 ogg_int64_t log_drop_scale; 222 /*The log of estimated scale factor for the rate model in Q57 format.*/ 223 ogg_int64_t log_scale[2]; 224 /*The log of the target quantizer level in Q57 format.*/ 225 ogg_int64_t log_qtarget; 226 /*Will we drop frames to meet bitrate target?*/ 227 unsigned char drop_frames; 228 /*Do we respect the maximum buffer fullness?*/ 229 unsigned char cap_overflow; 230 /*Can the reservoir go negative?*/ 231 unsigned char cap_underflow; 232 /*Second-order lowpass filters to track scale and VFR.*/ 233 oc_iir_filter scalefilter[2]; 234 int inter_count; 235 int inter_delay; 236 int inter_delay_target; 237 oc_iir_filter vfrfilter; 238 /*Two-pass mode state. 239 0 => 1-pass encoding. 240 1 => 1st pass of 2-pass encoding. 241 2 => 2nd pass of 2-pass encoding.*/ 242 int twopass; 243 /*Buffer for current frame metrics.*/ 244 unsigned char twopass_buffer[48]; 245 /*The number of bytes in the frame metrics buffer. 246 When 2-pass encoding is enabled, this is set to 0 after each frame is 247 submitted, and must be non-zero before the next frame will be accepted.*/ 248 int twopass_buffer_bytes; 249 int twopass_buffer_fill; 250 /*Whether or not to force the next frame to be a keyframe.*/ 251 unsigned char twopass_force_kf; 252 /*The metrics for the previous frame.*/ 253 oc_frame_metrics prev_metrics; 254 /*The metrics for the current frame.*/ 255 oc_frame_metrics cur_metrics; 256 /*The buffered metrics for future frames.*/ 257 oc_frame_metrics *frame_metrics; 258 int nframe_metrics; 259 int cframe_metrics; 260 /*The index of the current frame in the circular metric buffer.*/ 261 int frame_metrics_head; 262 /*The frame count of each type (keyframes, delta frames, and dup frames); 263 32 bits limits us to 2.268 years at 60 fps.*/ 264 ogg_uint32_t frames_total[3]; 265 /*The number of frames of each type yet to be processed.*/ 266 ogg_uint32_t frames_left[3]; 267 /*The sum of the scale values for each frame type.*/ 268 ogg_int64_t scale_sum[2]; 269 /*The start of the window over which the current scale sums are taken.*/ 270 int scale_window0; 271 /*The end of the window over which the current scale sums are taken.*/ 272 int scale_window_end; 273 /*The frame count of each type in the current 2-pass window; this does not 274 include dup frames.*/ 275 int nframes[3]; 276 /*The total accumulated estimation bias.*/ 277 ogg_int64_t rate_bias; 278 }; 279 280 281 void oc_rc_state_init(oc_rc_state *_rc,oc_enc_ctx *_enc); 282 void oc_rc_state_clear(oc_rc_state *_rc); 283 284 void oc_enc_rc_resize(oc_enc_ctx *_enc); 285 int oc_enc_select_qi(oc_enc_ctx *_enc,int _qti,int _clamp); 286 void oc_enc_calc_lambda(oc_enc_ctx *_enc,int _frame_type); 287 int oc_enc_update_rc_state(oc_enc_ctx *_enc, 288 long _bits,int _qti,int _qi,int _trial,int _droppable); 289 int oc_enc_rc_2pass_out(oc_enc_ctx *_enc,unsigned char **_buf); 290 int oc_enc_rc_2pass_in(oc_enc_ctx *_enc,unsigned char *_buf,size_t _bytes); 291 292 293 294 /*The internal encoder state.*/ 295 struct th_enc_ctx{ 296 /*Shared encoder/decoder state.*/ 297 oc_theora_state state; 298 /*Buffer in which to assemble packets.*/ 299 oggpack_buffer opb; 300 /*Encoder-specific macroblock information.*/ 301 oc_mb_enc_info *mb_info; 302 /*DC coefficients after prediction.*/ 303 ogg_int16_t *frag_dc; 304 /*The list of coded macro blocks, in coded order.*/ 305 unsigned *coded_mbis; 306 /*The number of coded macro blocks.*/ 307 size_t ncoded_mbis; 308 /*Whether or not packets are ready to be emitted. 309 This takes on negative values while there are remaining header packets to 310 be emitted, reaches 0 when the codec is ready for input, and becomes 311 positive when a frame has been processed and data packets are ready.*/ 312 int packet_state; 313 /*The maximum distance between keyframes.*/ 314 ogg_uint32_t keyframe_frequency_force; 315 /*The number of duplicates to produce for the next frame.*/ 316 ogg_uint32_t dup_count; 317 /*The number of duplicates remaining to be emitted for the current frame.*/ 318 ogg_uint32_t nqueued_dups; 319 /*The number of duplicates emitted for the last frame.*/ 320 ogg_uint32_t prev_dup_count; 321 /*The current speed level.*/ 322 int sp_level; 323 /*Whether or not VP3 compatibility mode has been enabled.*/ 324 unsigned char vp3_compatible; 325 /*Whether or not any INTER frames have been coded.*/ 326 unsigned char coded_inter_frame; 327 /*Whether or not previous frame was dropped.*/ 328 unsigned char prevframe_dropped; 329 /*Stores most recently chosen Huffman tables for each frame type, DC and AC 330 coefficients, and luma and chroma tokens. 331 The actual Huffman table used for a given coefficient depends not only on 332 the choice made here, but also its index in the zig-zag ordering.*/ 333 unsigned char huff_idxs[2][2][2]; 334 /*Current count of bits used by each MV coding mode.*/ 335 size_t mv_bits[2]; 336 /*The mode scheme chooser for estimating mode coding costs.*/ 337 oc_mode_scheme_chooser chooser; 338 /*The number of vertical super blocks in an MCU.*/ 339 int mcu_nvsbs; 340 /*The SSD error for skipping each fragment in the current MCU.*/ 341 unsigned *mcu_skip_ssd; 342 /*The DCT token lists for each coefficient and each plane.*/ 343 unsigned char **dct_tokens[3]; 344 /*The extra bits associated with each DCT token.*/ 345 ogg_uint16_t **extra_bits[3]; 346 /*The number of DCT tokens for each coefficient for each plane.*/ 347 ptrdiff_t ndct_tokens[3][64]; 348 /*Pending EOB runs for each coefficient for each plane.*/ 349 ogg_uint16_t eob_run[3][64]; 350 /*The offset of the first DCT token for each coefficient for each plane.*/ 351 unsigned char dct_token_offs[3][64]; 352 /*The last DC coefficient for each plane and reference frame.*/ 353 int dc_pred_last[3][3]; 354 #if defined(OC_COLLECT_METRICS) 355 /*Fragment SATD statistics for MB mode estimation metrics.*/ 356 unsigned *frag_satd; 357 /*Fragment SSD statistics for MB mode estimation metrics.*/ 358 unsigned *frag_ssd; 359 #endif 360 /*The R-D optimization parameter.*/ 361 int lambda; 362 /*The huffman tables in use.*/ 363 th_huff_code huff_codes[TH_NHUFFMAN_TABLES][TH_NDCT_TOKENS]; 364 /*The quantization parameters in use.*/ 365 th_quant_info qinfo; 366 oc_iquant *enquant_tables[64][3][2]; 367 oc_iquant_table enquant_table_data[64][3][2]; 368 /*An "average" quantizer for each quantizer type (INTRA or INTER) and qi 369 value. 370 This is used to paramterize the rate control decisions. 371 They are kept in the log domain to simplify later processing. 372 Keep in mind these are DCT domain quantizers, and so are scaled by an 373 additional factor of 4 from the pixel domain.*/ 374 ogg_int64_t log_qavg[2][64]; 375 /*The buffer state used to drive rate control.*/ 376 oc_rc_state rc; 377 /*Table for encoder acceleration functions.*/ 378 oc_enc_opt_vtable opt_vtable; 379 }; 380 381 382 void oc_enc_analyze_intra(oc_enc_ctx *_enc,int _recode); 383 int oc_enc_analyze_inter(oc_enc_ctx *_enc,int _allow_keyframe,int _recode); 384 #if defined(OC_COLLECT_METRICS) 385 void oc_enc_mode_metrics_collect(oc_enc_ctx *_enc); 386 void oc_enc_mode_metrics_dump(oc_enc_ctx *_enc); 387 #endif 388 389 390 391 /*Perform fullpel motion search for a single MB against both reference frames.*/ 392 void oc_mcenc_search(oc_enc_ctx *_enc,int _mbi); 393 /*Refine a MB MV for one frame.*/ 394 void oc_mcenc_refine1mv(oc_enc_ctx *_enc,int _mbi,int _frame); 395 /*Refine the block MVs.*/ 396 void oc_mcenc_refine4mv(oc_enc_ctx *_enc,int _mbi); 397 398 399 400 /*Used to rollback a tokenlog transaction when we retroactively decide to skip 401 a fragment. 402 A checkpoint is taken right before each token is added.*/ 403 struct oc_token_checkpoint{ 404 /*The color plane the token was added to.*/ 405 unsigned char pli; 406 /*The zig-zag index the token was added to.*/ 407 unsigned char zzi; 408 /*The outstanding EOB run count before the token was added.*/ 409 ogg_uint16_t eob_run; 410 /*The token count before the token was added.*/ 411 ptrdiff_t ndct_tokens; 412 }; 413 414 415 416 void oc_enc_tokenize_start(oc_enc_ctx *_enc); 417 int oc_enc_tokenize_ac(oc_enc_ctx *_enc,int _pli,ptrdiff_t _fragi, 418 ogg_int16_t *_qdct,const ogg_uint16_t *_dequant,const ogg_int16_t *_dct, 419 int _zzi,oc_token_checkpoint **_stack,int _acmin); 420 void oc_enc_tokenlog_rollback(oc_enc_ctx *_enc, 421 const oc_token_checkpoint *_stack,int _n); 422 void oc_enc_pred_dc_frag_rows(oc_enc_ctx *_enc, 423 int _pli,int _fragy0,int _frag_yend); 424 void oc_enc_tokenize_dc_frag_list(oc_enc_ctx *_enc,int _pli, 425 const ptrdiff_t *_coded_fragis,ptrdiff_t _ncoded_fragis, 426 int _prev_ndct_tokens1,int _prev_eob_run1); 427 void oc_enc_tokenize_finish(oc_enc_ctx *_enc); 428 429 430 431 /*Utility routine to encode one of the header packets.*/ 432 int oc_state_flushheader(oc_theora_state *_state,int *_packet_state, 433 oggpack_buffer *_opb,const th_quant_info *_qinfo, 434 const th_huff_code _codes[TH_NHUFFMAN_TABLES][TH_NDCT_TOKENS], 435 const char *_vendor,th_comment *_tc,ogg_packet *_op); 436 437 438 439 /*Encoder-specific accelerated functions.*/ 440 void oc_enc_frag_sub(const oc_enc_ctx *_enc,ogg_int16_t _diff[64], 441 const unsigned char *_src,const unsigned char *_ref,int _ystride); 442 void oc_enc_frag_sub_128(const oc_enc_ctx *_enc,ogg_int16_t _diff[64], 443 const unsigned char *_src,int _ystride); 444 unsigned oc_enc_frag_sad(const oc_enc_ctx *_enc,const unsigned char *_src, 445 const unsigned char *_ref,int _ystride); 446 unsigned oc_enc_frag_sad_thresh(const oc_enc_ctx *_enc, 447 const unsigned char *_src,const unsigned char *_ref,int _ystride, 448 unsigned _thresh); 449 unsigned oc_enc_frag_sad2_thresh(const oc_enc_ctx *_enc, 450 const unsigned char *_src,const unsigned char *_ref1, 451 const unsigned char *_ref2,int _ystride,unsigned _thresh); 452 unsigned oc_enc_frag_satd_thresh(const oc_enc_ctx *_enc, 453 const unsigned char *_src,const unsigned char *_ref,int _ystride, 454 unsigned _thresh); 455 unsigned oc_enc_frag_satd2_thresh(const oc_enc_ctx *_enc, 456 const unsigned char *_src,const unsigned char *_ref1, 457 const unsigned char *_ref2,int _ystride,unsigned _thresh); 458 unsigned oc_enc_frag_intra_satd(const oc_enc_ctx *_enc, 459 const unsigned char *_src,int _ystride); 460 void oc_enc_frag_copy2(const oc_enc_ctx *_enc,unsigned char *_dst, 461 const unsigned char *_src1,const unsigned char *_src2,int _ystride); 462 void oc_enc_frag_recon_intra(const oc_enc_ctx *_enc, 463 unsigned char *_dst,int _ystride,const ogg_int16_t _residue[64]); 464 void oc_enc_frag_recon_inter(const oc_enc_ctx *_enc,unsigned char *_dst, 465 const unsigned char *_src,int _ystride,const ogg_int16_t _residue[64]); 466 void oc_enc_fdct8x8(const oc_enc_ctx *_enc,ogg_int16_t _y[64], 467 const ogg_int16_t _x[64]); 468 469 /*Default pure-C implementations.*/ 470 void oc_enc_vtable_init_c(oc_enc_ctx *_enc); 471 472 void oc_enc_frag_sub_c(ogg_int16_t _diff[64], 473 const unsigned char *_src,const unsigned char *_ref,int _ystride); 474 void oc_enc_frag_sub_128_c(ogg_int16_t _diff[64], 475 const unsigned char *_src,int _ystride); 476 void oc_enc_frag_copy2_c(unsigned char *_dst, 477 const unsigned char *_src1,const unsigned char *_src2,int _ystride); 478 unsigned oc_enc_frag_sad_c(const unsigned char *_src, 479 const unsigned char *_ref,int _ystride); 480 unsigned oc_enc_frag_sad_thresh_c(const unsigned char *_src, 481 const unsigned char *_ref,int _ystride,unsigned _thresh); 482 unsigned oc_enc_frag_sad2_thresh_c(const unsigned char *_src, 483 const unsigned char *_ref1,const unsigned char *_ref2,int _ystride, 484 unsigned _thresh); 485 unsigned oc_enc_frag_satd_thresh_c(const unsigned char *_src, 486 const unsigned char *_ref,int _ystride,unsigned _thresh); 487 unsigned oc_enc_frag_satd2_thresh_c(const unsigned char *_src, 488 const unsigned char *_ref1,const unsigned char *_ref2,int _ystride, 489 unsigned _thresh); 490 unsigned oc_enc_frag_intra_satd_c(const unsigned char *_src,int _ystride); 491 void oc_enc_fdct8x8_c(ogg_int16_t _y[64],const ogg_int16_t _x[64]); 492 493 #endif 494