vp8i.h 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354
  1. // Copyright 2010 Google Inc. All Rights Reserved.
  2. //
  3. // Use of this source code is governed by a BSD-style license
  4. // that can be found in the COPYING file in the root of the source
  5. // tree. An additional intellectual property rights grant can be found
  6. // in the file PATENTS. All contributing project authors may
  7. // be found in the AUTHORS file in the root of the source tree.
  8. // -----------------------------------------------------------------------------
  9. //
  10. // VP8 decoder: internal header.
  11. //
  12. // Author: Skal (pascal.massimino@gmail.com)
  13. #ifndef WEBP_DEC_VP8I_H_
  14. #define WEBP_DEC_VP8I_H_
  15. #include <string.h> // for memcpy()
  16. #include "./vp8li.h"
  17. #include "../utils/bit_reader.h"
  18. #include "../utils/random.h"
  19. #include "../utils/thread.h"
  20. #include "../dsp/dsp.h"
  21. #ifdef __cplusplus
  22. extern "C" {
  23. #endif
  24. //------------------------------------------------------------------------------
  25. // Various defines and enums
  26. // version numbers
  27. #define DEC_MAJ_VERSION 0
  28. #define DEC_MIN_VERSION 4
  29. #define DEC_REV_VERSION 2
  30. // intra prediction modes
  31. enum { B_DC_PRED = 0, // 4x4 modes
  32. B_TM_PRED,
  33. B_VE_PRED,
  34. B_HE_PRED,
  35. B_RD_PRED,
  36. B_VR_PRED,
  37. B_LD_PRED,
  38. B_VL_PRED,
  39. B_HD_PRED,
  40. B_HU_PRED,
  41. NUM_BMODES = B_HU_PRED + 1 - B_DC_PRED, // = 10
  42. // Luma16 or UV modes
  43. DC_PRED = B_DC_PRED, V_PRED = B_VE_PRED,
  44. H_PRED = B_HE_PRED, TM_PRED = B_TM_PRED,
  45. B_PRED = NUM_BMODES, // refined I4x4 mode
  46. // special modes
  47. B_DC_PRED_NOTOP = 4,
  48. B_DC_PRED_NOLEFT = 5,
  49. B_DC_PRED_NOTOPLEFT = 6,
  50. NUM_B_DC_MODES = 7 };
  51. enum { MB_FEATURE_TREE_PROBS = 3,
  52. NUM_MB_SEGMENTS = 4,
  53. NUM_REF_LF_DELTAS = 4,
  54. NUM_MODE_LF_DELTAS = 4, // I4x4, ZERO, *, SPLIT
  55. MAX_NUM_PARTITIONS = 8,
  56. // Probabilities
  57. NUM_TYPES = 4,
  58. NUM_BANDS = 8,
  59. NUM_CTX = 3,
  60. NUM_PROBAS = 11,
  61. NUM_MV_PROBAS = 19 };
  62. // YUV-cache parameters.
  63. // Constraints are: We need to store one 16x16 block of luma samples (y),
  64. // and two 8x8 chroma blocks (u/v). These are better be 16-bytes aligned,
  65. // in order to be SIMD-friendly. We also need to store the top, left and
  66. // top-left samples (from previously decoded blocks), along with four
  67. // extra top-right samples for luma (intra4x4 prediction only).
  68. // One possible layout is, using 32 * (17 + 9) bytes:
  69. //
  70. // .+------ <- only 1 pixel high
  71. // .|yyyyt.
  72. // .|yyyyt.
  73. // .|yyyyt.
  74. // .|yyyy..
  75. // .+--.+-- <- only 1 pixel high
  76. // .|uu.|vv
  77. // .|uu.|vv
  78. //
  79. // Every character is a 4x4 block, with legend:
  80. // '.' = unused
  81. // 'y' = y-samples 'u' = u-samples 'v' = u-samples
  82. // '|' = left sample, '-' = top sample, '+' = top-left sample
  83. // 't' = extra top-right sample for 4x4 modes
  84. // With this layout, BPS (=Bytes Per Scan-line) is one cacheline size.
  85. #define BPS 32 // this is the common stride used by yuv[]
  86. #define YUV_SIZE (BPS * 17 + BPS * 9)
  87. #define Y_SIZE (BPS * 17)
  88. #define Y_OFF (BPS * 1 + 8)
  89. #define U_OFF (Y_OFF + BPS * 16 + BPS)
  90. #define V_OFF (U_OFF + 16)
  91. // minimal width under which lossy multi-threading is always disabled
  92. #define MIN_WIDTH_FOR_THREADS 512
  93. //------------------------------------------------------------------------------
  94. // Headers
  95. typedef struct {
  96. uint8_t key_frame_;
  97. uint8_t profile_;
  98. uint8_t show_;
  99. uint32_t partition_length_;
  100. } VP8FrameHeader;
  101. typedef struct {
  102. uint16_t width_;
  103. uint16_t height_;
  104. uint8_t xscale_;
  105. uint8_t yscale_;
  106. uint8_t colorspace_; // 0 = YCbCr
  107. uint8_t clamp_type_;
  108. } VP8PictureHeader;
  109. // segment features
  110. typedef struct {
  111. int use_segment_;
  112. int update_map_; // whether to update the segment map or not
  113. int absolute_delta_; // absolute or delta values for quantizer and filter
  114. int8_t quantizer_[NUM_MB_SEGMENTS]; // quantization changes
  115. int8_t filter_strength_[NUM_MB_SEGMENTS]; // filter strength for segments
  116. } VP8SegmentHeader;
  117. // probas associated to one of the contexts
  118. typedef uint8_t VP8ProbaArray[NUM_PROBAS];
  119. typedef struct { // all the probas associated to one band
  120. VP8ProbaArray probas_[NUM_CTX];
  121. } VP8BandProbas;
  122. // Struct collecting all frame-persistent probabilities.
  123. typedef struct {
  124. uint8_t segments_[MB_FEATURE_TREE_PROBS];
  125. // Type: 0:Intra16-AC 1:Intra16-DC 2:Chroma 3:Intra4
  126. VP8BandProbas bands_[NUM_TYPES][NUM_BANDS];
  127. } VP8Proba;
  128. // Filter parameters
  129. typedef struct {
  130. int simple_; // 0=complex, 1=simple
  131. int level_; // [0..63]
  132. int sharpness_; // [0..7]
  133. int use_lf_delta_;
  134. int ref_lf_delta_[NUM_REF_LF_DELTAS];
  135. int mode_lf_delta_[NUM_MODE_LF_DELTAS];
  136. } VP8FilterHeader;
  137. //------------------------------------------------------------------------------
  138. // Informations about the macroblocks.
  139. typedef struct { // filter specs
  140. uint8_t f_limit_; // filter limit in [3..189], or 0 if no filtering
  141. uint8_t f_ilevel_; // inner limit in [1..63]
  142. uint8_t f_inner_; // do inner filtering?
  143. uint8_t hev_thresh_; // high edge variance threshold in [0..2]
  144. } VP8FInfo;
  145. typedef struct { // Top/Left Contexts used for syntax-parsing
  146. uint8_t nz_; // non-zero AC/DC coeffs (4bit for luma + 4bit for chroma)
  147. uint8_t nz_dc_; // non-zero DC coeff (1bit)
  148. } VP8MB;
  149. // Dequantization matrices
  150. typedef int quant_t[2]; // [DC / AC]. Can be 'uint16_t[2]' too (~slower).
  151. typedef struct {
  152. quant_t y1_mat_, y2_mat_, uv_mat_;
  153. int uv_quant_; // U/V quantizer value
  154. int dither_; // dithering amplitude (0 = off, max=255)
  155. } VP8QuantMatrix;
  156. // Data needed to reconstruct a macroblock
  157. typedef struct {
  158. int16_t coeffs_[384]; // 384 coeffs = (16+4+4) * 4*4
  159. uint8_t is_i4x4_; // true if intra4x4
  160. uint8_t imodes_[16]; // one 16x16 mode (#0) or sixteen 4x4 modes
  161. uint8_t uvmode_; // chroma prediction mode
  162. // bit-wise info about the content of each sub-4x4 blocks (in decoding order).
  163. // Each of the 4x4 blocks for y/u/v is associated with a 2b code according to:
  164. // code=0 -> no coefficient
  165. // code=1 -> only DC
  166. // code=2 -> first three coefficients are non-zero
  167. // code=3 -> more than three coefficients are non-zero
  168. // This allows to call specialized transform functions.
  169. uint32_t non_zero_y_;
  170. uint32_t non_zero_uv_;
  171. uint8_t dither_; // local dithering strength (deduced from non_zero_*)
  172. uint8_t skip_;
  173. uint8_t segment_;
  174. } VP8MBData;
  175. // Persistent information needed by the parallel processing
  176. typedef struct {
  177. int id_; // cache row to process (in [0..2])
  178. int mb_y_; // macroblock position of the row
  179. int filter_row_; // true if row-filtering is needed
  180. VP8FInfo* f_info_; // filter strengths (swapped with dec->f_info_)
  181. VP8MBData* mb_data_; // reconstruction data (swapped with dec->mb_data_)
  182. VP8Io io_; // copy of the VP8Io to pass to put()
  183. } VP8ThreadContext;
  184. // Saved top samples, per macroblock. Fits into a cache-line.
  185. typedef struct {
  186. uint8_t y[16], u[8], v[8];
  187. } VP8TopSamples;
  188. //------------------------------------------------------------------------------
  189. // VP8Decoder: the main opaque structure handed over to user
  190. struct VP8Decoder {
  191. VP8StatusCode status_;
  192. int ready_; // true if ready to decode a picture with VP8Decode()
  193. const char* error_msg_; // set when status_ is not OK.
  194. // Main data source
  195. VP8BitReader br_;
  196. // headers
  197. VP8FrameHeader frm_hdr_;
  198. VP8PictureHeader pic_hdr_;
  199. VP8FilterHeader filter_hdr_;
  200. VP8SegmentHeader segment_hdr_;
  201. // Worker
  202. WebPWorker worker_;
  203. int mt_method_; // multi-thread method: 0=off, 1=[parse+recon][filter]
  204. // 2=[parse][recon+filter]
  205. int cache_id_; // current cache row
  206. int num_caches_; // number of cached rows of 16 pixels (1, 2 or 3)
  207. VP8ThreadContext thread_ctx_; // Thread context
  208. // dimension, in macroblock units.
  209. int mb_w_, mb_h_;
  210. // Macroblock to process/filter, depending on cropping and filter_type.
  211. int tl_mb_x_, tl_mb_y_; // top-left MB that must be in-loop filtered
  212. int br_mb_x_, br_mb_y_; // last bottom-right MB that must be decoded
  213. // number of partitions.
  214. int num_parts_;
  215. // per-partition boolean decoders.
  216. VP8BitReader parts_[MAX_NUM_PARTITIONS];
  217. // Dithering strength, deduced from decoding options
  218. int dither_; // whether to use dithering or not
  219. VP8Random dithering_rg_; // random generator for dithering
  220. // dequantization (one set of DC/AC dequant factor per segment)
  221. VP8QuantMatrix dqm_[NUM_MB_SEGMENTS];
  222. // probabilities
  223. VP8Proba proba_;
  224. int use_skip_proba_;
  225. uint8_t skip_p_;
  226. // Boundary data cache and persistent buffers.
  227. uint8_t* intra_t_; // top intra modes values: 4 * mb_w_
  228. uint8_t intra_l_[4]; // left intra modes values
  229. VP8TopSamples* yuv_t_; // top y/u/v samples
  230. VP8MB* mb_info_; // contextual macroblock info (mb_w_ + 1)
  231. VP8FInfo* f_info_; // filter strength info
  232. uint8_t* yuv_b_; // main block for Y/U/V (size = YUV_SIZE)
  233. uint8_t* cache_y_; // macroblock row for storing unfiltered samples
  234. uint8_t* cache_u_;
  235. uint8_t* cache_v_;
  236. int cache_y_stride_;
  237. int cache_uv_stride_;
  238. // main memory chunk for the above data. Persistent.
  239. void* mem_;
  240. size_t mem_size_;
  241. // Per macroblock non-persistent infos.
  242. int mb_x_, mb_y_; // current position, in macroblock units
  243. VP8MBData* mb_data_; // parsed reconstruction data
  244. // Filtering side-info
  245. int filter_type_; // 0=off, 1=simple, 2=complex
  246. VP8FInfo fstrengths_[NUM_MB_SEGMENTS][2]; // precalculated per-segment/type
  247. // Alpha
  248. struct ALPHDecoder* alph_dec_; // alpha-plane decoder object
  249. const uint8_t* alpha_data_; // compressed alpha data (if present)
  250. size_t alpha_data_size_;
  251. int is_alpha_decoded_; // true if alpha_data_ is decoded in alpha_plane_
  252. uint8_t* alpha_plane_; // output. Persistent, contains the whole data.
  253. int alpha_dithering_; // derived from decoding options (0=off, 100=full).
  254. };
  255. //------------------------------------------------------------------------------
  256. // internal functions. Not public.
  257. // in vp8.c
  258. int VP8SetError(VP8Decoder* const dec,
  259. VP8StatusCode error, const char* const msg);
  260. // in tree.c
  261. void VP8ResetProba(VP8Proba* const proba);
  262. void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec);
  263. // parses one row of intra mode data in partition 0, returns !eof
  264. int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec);
  265. // in quant.c
  266. void VP8ParseQuant(VP8Decoder* const dec);
  267. // in frame.c
  268. int VP8InitFrame(VP8Decoder* const dec, VP8Io* io);
  269. // Call io->setup() and finish setting up scan parameters.
  270. // After this call returns, one must always call VP8ExitCritical() with the
  271. // same parameters. Both functions should be used in pair. Returns VP8_STATUS_OK
  272. // if ok, otherwise sets and returns the error status on *dec.
  273. VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io);
  274. // Must always be called in pair with VP8EnterCritical().
  275. // Returns false in case of error.
  276. int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io);
  277. // Return the multi-threading method to use (0=off), depending
  278. // on options and bitstream size. Only for lossy decoding.
  279. int VP8GetThreadMethod(const WebPDecoderOptions* const options,
  280. const WebPHeaderStructure* const headers,
  281. int width, int height);
  282. // Initialize dithering post-process if needed.
  283. void VP8InitDithering(const WebPDecoderOptions* const options,
  284. VP8Decoder* const dec);
  285. // Process the last decoded row (filtering + output).
  286. int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io);
  287. // To be called at the start of a new scanline, to initialize predictors.
  288. void VP8InitScanline(VP8Decoder* const dec);
  289. // Decode one macroblock. Returns false if there is not enough data.
  290. int VP8DecodeMB(VP8Decoder* const dec, VP8BitReader* const token_br);
  291. // in alpha.c
  292. const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
  293. int row, int num_rows);
  294. //------------------------------------------------------------------------------
  295. #ifdef __cplusplus
  296. } // extern "C"
  297. #endif
  298. #endif /* WEBP_DEC_VP8I_H_ */