Have GPUTracer process traces during CommandBuffer Idle time.
[chromium-blink-merge.git] / third_party / libwebp / dec / vp8i.h
bloba02d9ff73de4463f2762c01f8c8ea85c6311df77
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.
12 // Author: Skal (pascal.massimino@gmail.com)
14 #ifndef WEBP_DEC_VP8I_H_
15 #define WEBP_DEC_VP8I_H_
17 #include <string.h> // for memcpy()
18 #include "./vp8li.h"
19 #include "../utils/bit_reader.h"
20 #include "../utils/random.h"
21 #include "../utils/thread.h"
22 #include "../dsp/dsp.h"
24 #ifdef __cplusplus
25 extern "C" {
26 #endif
28 //------------------------------------------------------------------------------
29 // Various defines and enums
31 // version numbers
32 #define DEC_MAJ_VERSION 0
33 #define DEC_MIN_VERSION 4
34 #define DEC_REV_VERSION 3
36 // intra prediction modes
37 enum { B_DC_PRED = 0, // 4x4 modes
38 B_TM_PRED,
39 B_VE_PRED,
40 B_HE_PRED,
41 B_RD_PRED,
42 B_VR_PRED,
43 B_LD_PRED,
44 B_VL_PRED,
45 B_HD_PRED,
46 B_HU_PRED,
47 NUM_BMODES = B_HU_PRED + 1 - B_DC_PRED, // = 10
49 // Luma16 or UV modes
50 DC_PRED = B_DC_PRED, V_PRED = B_VE_PRED,
51 H_PRED = B_HE_PRED, TM_PRED = B_TM_PRED,
52 B_PRED = NUM_BMODES, // refined I4x4 mode
54 // special modes
55 B_DC_PRED_NOTOP = 4,
56 B_DC_PRED_NOLEFT = 5,
57 B_DC_PRED_NOTOPLEFT = 6,
58 NUM_B_DC_MODES = 7 };
60 enum { MB_FEATURE_TREE_PROBS = 3,
61 NUM_MB_SEGMENTS = 4,
62 NUM_REF_LF_DELTAS = 4,
63 NUM_MODE_LF_DELTAS = 4, // I4x4, ZERO, *, SPLIT
64 MAX_NUM_PARTITIONS = 8,
65 // Probabilities
66 NUM_TYPES = 4,
67 NUM_BANDS = 8,
68 NUM_CTX = 3,
69 NUM_PROBAS = 11,
70 NUM_MV_PROBAS = 19 };
72 // YUV-cache parameters.
73 // Constraints are: We need to store one 16x16 block of luma samples (y),
74 // and two 8x8 chroma blocks (u/v). These are better be 16-bytes aligned,
75 // in order to be SIMD-friendly. We also need to store the top, left and
76 // top-left samples (from previously decoded blocks), along with four
77 // extra top-right samples for luma (intra4x4 prediction only).
78 // One possible layout is, using 32 * (17 + 9) bytes:
80 // .+------ <- only 1 pixel high
81 // .|yyyyt.
82 // .|yyyyt.
83 // .|yyyyt.
84 // .|yyyy..
85 // .+--.+-- <- only 1 pixel high
86 // .|uu.|vv
87 // .|uu.|vv
89 // Every character is a 4x4 block, with legend:
90 // '.' = unused
91 // 'y' = y-samples 'u' = u-samples 'v' = u-samples
92 // '|' = left sample, '-' = top sample, '+' = top-left sample
93 // 't' = extra top-right sample for 4x4 modes
94 // With this layout, BPS (=Bytes Per Scan-line) is one cacheline size.
95 #define BPS 32 // this is the common stride used by yuv[]
96 #define YUV_SIZE (BPS * 17 + BPS * 9)
97 #define Y_SIZE (BPS * 17)
98 #define Y_OFF (BPS * 1 + 8)
99 #define U_OFF (Y_OFF + BPS * 16 + BPS)
100 #define V_OFF (U_OFF + 16)
102 // minimal width under which lossy multi-threading is always disabled
103 #define MIN_WIDTH_FOR_THREADS 512
105 //------------------------------------------------------------------------------
106 // Headers
108 typedef struct {
109 uint8_t key_frame_;
110 uint8_t profile_;
111 uint8_t show_;
112 uint32_t partition_length_;
113 } VP8FrameHeader;
115 typedef struct {
116 uint16_t width_;
117 uint16_t height_;
118 uint8_t xscale_;
119 uint8_t yscale_;
120 uint8_t colorspace_; // 0 = YCbCr
121 uint8_t clamp_type_;
122 } VP8PictureHeader;
124 // segment features
125 typedef struct {
126 int use_segment_;
127 int update_map_; // whether to update the segment map or not
128 int absolute_delta_; // absolute or delta values for quantizer and filter
129 int8_t quantizer_[NUM_MB_SEGMENTS]; // quantization changes
130 int8_t filter_strength_[NUM_MB_SEGMENTS]; // filter strength for segments
131 } VP8SegmentHeader;
134 // probas associated to one of the contexts
135 typedef uint8_t VP8ProbaArray[NUM_PROBAS];
137 typedef struct { // all the probas associated to one band
138 VP8ProbaArray probas_[NUM_CTX];
139 } VP8BandProbas;
141 // Struct collecting all frame-persistent probabilities.
142 typedef struct {
143 uint8_t segments_[MB_FEATURE_TREE_PROBS];
144 // Type: 0:Intra16-AC 1:Intra16-DC 2:Chroma 3:Intra4
145 VP8BandProbas bands_[NUM_TYPES][NUM_BANDS];
146 } VP8Proba;
148 // Filter parameters
149 typedef struct {
150 int simple_; // 0=complex, 1=simple
151 int level_; // [0..63]
152 int sharpness_; // [0..7]
153 int use_lf_delta_;
154 int ref_lf_delta_[NUM_REF_LF_DELTAS];
155 int mode_lf_delta_[NUM_MODE_LF_DELTAS];
156 } VP8FilterHeader;
158 //------------------------------------------------------------------------------
159 // Informations about the macroblocks.
161 typedef struct { // filter specs
162 uint8_t f_limit_; // filter limit in [3..189], or 0 if no filtering
163 uint8_t f_ilevel_; // inner limit in [1..63]
164 uint8_t f_inner_; // do inner filtering?
165 uint8_t hev_thresh_; // high edge variance threshold in [0..2]
166 } VP8FInfo;
168 typedef struct { // Top/Left Contexts used for syntax-parsing
169 uint8_t nz_; // non-zero AC/DC coeffs (4bit for luma + 4bit for chroma)
170 uint8_t nz_dc_; // non-zero DC coeff (1bit)
171 } VP8MB;
173 // Dequantization matrices
174 typedef int quant_t[2]; // [DC / AC]. Can be 'uint16_t[2]' too (~slower).
175 typedef struct {
176 quant_t y1_mat_, y2_mat_, uv_mat_;
178 int uv_quant_; // U/V quantizer value
179 int dither_; // dithering amplitude (0 = off, max=255)
180 } VP8QuantMatrix;
182 // Data needed to reconstruct a macroblock
183 typedef struct {
184 int16_t coeffs_[384]; // 384 coeffs = (16+4+4) * 4*4
185 uint8_t is_i4x4_; // true if intra4x4
186 uint8_t imodes_[16]; // one 16x16 mode (#0) or sixteen 4x4 modes
187 uint8_t uvmode_; // chroma prediction mode
188 // bit-wise info about the content of each sub-4x4 blocks (in decoding order).
189 // Each of the 4x4 blocks for y/u/v is associated with a 2b code according to:
190 // code=0 -> no coefficient
191 // code=1 -> only DC
192 // code=2 -> first three coefficients are non-zero
193 // code=3 -> more than three coefficients are non-zero
194 // This allows to call specialized transform functions.
195 uint32_t non_zero_y_;
196 uint32_t non_zero_uv_;
197 uint8_t dither_; // local dithering strength (deduced from non_zero_*)
198 uint8_t skip_;
199 uint8_t segment_;
200 } VP8MBData;
202 // Persistent information needed by the parallel processing
203 typedef struct {
204 int id_; // cache row to process (in [0..2])
205 int mb_y_; // macroblock position of the row
206 int filter_row_; // true if row-filtering is needed
207 VP8FInfo* f_info_; // filter strengths (swapped with dec->f_info_)
208 VP8MBData* mb_data_; // reconstruction data (swapped with dec->mb_data_)
209 VP8Io io_; // copy of the VP8Io to pass to put()
210 } VP8ThreadContext;
212 // Saved top samples, per macroblock. Fits into a cache-line.
213 typedef struct {
214 uint8_t y[16], u[8], v[8];
215 } VP8TopSamples;
217 //------------------------------------------------------------------------------
218 // VP8Decoder: the main opaque structure handed over to user
220 struct VP8Decoder {
221 VP8StatusCode status_;
222 int ready_; // true if ready to decode a picture with VP8Decode()
223 const char* error_msg_; // set when status_ is not OK.
225 // Main data source
226 VP8BitReader br_;
228 // headers
229 VP8FrameHeader frm_hdr_;
230 VP8PictureHeader pic_hdr_;
231 VP8FilterHeader filter_hdr_;
232 VP8SegmentHeader segment_hdr_;
234 // Worker
235 WebPWorker worker_;
236 int mt_method_; // multi-thread method: 0=off, 1=[parse+recon][filter]
237 // 2=[parse][recon+filter]
238 int cache_id_; // current cache row
239 int num_caches_; // number of cached rows of 16 pixels (1, 2 or 3)
240 VP8ThreadContext thread_ctx_; // Thread context
242 // dimension, in macroblock units.
243 int mb_w_, mb_h_;
245 // Macroblock to process/filter, depending on cropping and filter_type.
246 int tl_mb_x_, tl_mb_y_; // top-left MB that must be in-loop filtered
247 int br_mb_x_, br_mb_y_; // last bottom-right MB that must be decoded
249 // number of partitions.
250 int num_parts_;
251 // per-partition boolean decoders.
252 VP8BitReader parts_[MAX_NUM_PARTITIONS];
254 // Dithering strength, deduced from decoding options
255 int dither_; // whether to use dithering or not
256 VP8Random dithering_rg_; // random generator for dithering
258 // dequantization (one set of DC/AC dequant factor per segment)
259 VP8QuantMatrix dqm_[NUM_MB_SEGMENTS];
261 // probabilities
262 VP8Proba proba_;
263 int use_skip_proba_;
264 uint8_t skip_p_;
266 // Boundary data cache and persistent buffers.
267 uint8_t* intra_t_; // top intra modes values: 4 * mb_w_
268 uint8_t intra_l_[4]; // left intra modes values
270 VP8TopSamples* yuv_t_; // top y/u/v samples
272 VP8MB* mb_info_; // contextual macroblock info (mb_w_ + 1)
273 VP8FInfo* f_info_; // filter strength info
274 uint8_t* yuv_b_; // main block for Y/U/V (size = YUV_SIZE)
276 uint8_t* cache_y_; // macroblock row for storing unfiltered samples
277 uint8_t* cache_u_;
278 uint8_t* cache_v_;
279 int cache_y_stride_;
280 int cache_uv_stride_;
282 // main memory chunk for the above data. Persistent.
283 void* mem_;
284 size_t mem_size_;
286 // Per macroblock non-persistent infos.
287 int mb_x_, mb_y_; // current position, in macroblock units
288 VP8MBData* mb_data_; // parsed reconstruction data
290 // Filtering side-info
291 int filter_type_; // 0=off, 1=simple, 2=complex
292 VP8FInfo fstrengths_[NUM_MB_SEGMENTS][2]; // precalculated per-segment/type
294 // Alpha
295 struct ALPHDecoder* alph_dec_; // alpha-plane decoder object
296 const uint8_t* alpha_data_; // compressed alpha data (if present)
297 size_t alpha_data_size_;
298 int is_alpha_decoded_; // true if alpha_data_ is decoded in alpha_plane_
299 uint8_t* alpha_plane_; // output. Persistent, contains the whole data.
300 int alpha_dithering_; // derived from decoding options (0=off, 100=full).
303 //------------------------------------------------------------------------------
304 // internal functions. Not public.
306 // in vp8.c
307 int VP8SetError(VP8Decoder* const dec,
308 VP8StatusCode error, const char* const msg);
310 // in tree.c
311 void VP8ResetProba(VP8Proba* const proba);
312 void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec);
313 // parses one row of intra mode data in partition 0, returns !eof
314 int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec);
316 // in quant.c
317 void VP8ParseQuant(VP8Decoder* const dec);
319 // in frame.c
320 int VP8InitFrame(VP8Decoder* const dec, VP8Io* io);
321 // Call io->setup() and finish setting up scan parameters.
322 // After this call returns, one must always call VP8ExitCritical() with the
323 // same parameters. Both functions should be used in pair. Returns VP8_STATUS_OK
324 // if ok, otherwise sets and returns the error status on *dec.
325 VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io);
326 // Must always be called in pair with VP8EnterCritical().
327 // Returns false in case of error.
328 int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io);
329 // Return the multi-threading method to use (0=off), depending
330 // on options and bitstream size. Only for lossy decoding.
331 int VP8GetThreadMethod(const WebPDecoderOptions* const options,
332 const WebPHeaderStructure* const headers,
333 int width, int height);
334 // Initialize dithering post-process if needed.
335 void VP8InitDithering(const WebPDecoderOptions* const options,
336 VP8Decoder* const dec);
337 // Process the last decoded row (filtering + output).
338 int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io);
339 // To be called at the start of a new scanline, to initialize predictors.
340 void VP8InitScanline(VP8Decoder* const dec);
341 // Decode one macroblock. Returns false if there is not enough data.
342 int VP8DecodeMB(VP8Decoder* const dec, VP8BitReader* const token_br);
344 // in alpha.c
345 const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec,
346 int row, int num_rows);
348 //------------------------------------------------------------------------------
350 #ifdef __cplusplus
351 } // extern "C"
352 #endif
354 #endif /* WEBP_DEC_VP8I_H_ */