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 14 #ifndef WEBP_DEC_VP8I_H_ 15 #define WEBP_DEC_VP8I_H_ 16 17 #include <string.h> // for memcpy() 18 #include "./vp8li.h" 19 #include "../utils/bit_reader.h" 20 #include "../utils/thread.h" 21 #include "../dsp/dsp.h" 22 23 #if defined(__cplusplus) || defined(c_plusplus) 24 extern "C" { 25 #endif 26 27 //------------------------------------------------------------------------------ 28 // Various defines and enums 29 30 // version numbers 31 #define DEC_MAJ_VERSION 0 32 #define DEC_MIN_VERSION 3 33 #define DEC_REV_VERSION 1 34 35 #define ONLY_KEYFRAME_CODE // to remove any code related to P-Frames 36 37 // intra prediction modes 38 enum { B_DC_PRED = 0, // 4x4 modes 39 B_TM_PRED, 40 B_VE_PRED, 41 B_HE_PRED, 42 B_RD_PRED, 43 B_VR_PRED, 44 B_LD_PRED, 45 B_VL_PRED, 46 B_HD_PRED, 47 B_HU_PRED, 48 NUM_BMODES = B_HU_PRED + 1 - B_DC_PRED, // = 10 49 50 // Luma16 or UV modes 51 DC_PRED = B_DC_PRED, V_PRED = B_VE_PRED, 52 H_PRED = B_HE_PRED, TM_PRED = B_TM_PRED, 53 B_PRED = NUM_BMODES, // refined I4x4 mode 54 55 // special modes 56 B_DC_PRED_NOTOP = 4, 57 B_DC_PRED_NOLEFT = 5, 58 B_DC_PRED_NOTOPLEFT = 6, 59 NUM_B_DC_MODES = 7 }; 60 61 enum { MB_FEATURE_TREE_PROBS = 3, 62 NUM_MB_SEGMENTS = 4, 63 NUM_REF_LF_DELTAS = 4, 64 NUM_MODE_LF_DELTAS = 4, // I4x4, ZERO, *, SPLIT 65 MAX_NUM_PARTITIONS = 8, 66 // Probabilities 67 NUM_TYPES = 4, 68 NUM_BANDS = 8, 69 NUM_CTX = 3, 70 NUM_PROBAS = 11, 71 NUM_MV_PROBAS = 19 }; 72 73 // YUV-cache parameters. 74 // Constraints are: We need to store one 16x16 block of luma samples (y), 75 // and two 8x8 chroma blocks (u/v). These are better be 16-bytes aligned, 76 // in order to be SIMD-friendly. We also need to store the top, left and 77 // top-left samples (from previously decoded blocks), along with four 78 // extra top-right samples for luma (intra4x4 prediction only). 79 // One possible layout is, using 32 * (17 + 9) bytes: 80 // 81 // .+------ <- only 1 pixel high 82 // .|yyyyt. 83 // .|yyyyt. 84 // .|yyyyt. 85 // .|yyyy.. 86 // .+--.+-- <- only 1 pixel high 87 // .|uu.|vv 88 // .|uu.|vv 89 // 90 // Every character is a 4x4 block, with legend: 91 // '.' = unused 92 // 'y' = y-samples 'u' = u-samples 'v' = u-samples 93 // '|' = left sample, '-' = top sample, '+' = top-left sample 94 // 't' = extra top-right sample for 4x4 modes 95 // With this layout, BPS (=Bytes Per Scan-line) is one cacheline size. 96 #define BPS 32 // this is the common stride used by yuv[] 97 #define YUV_SIZE (BPS * 17 + BPS * 9) 98 #define Y_SIZE (BPS * 17) 99 #define Y_OFF (BPS * 1 + 8) 100 #define U_OFF (Y_OFF + BPS * 16 + BPS) 101 #define V_OFF (U_OFF + 16) 102 103 //------------------------------------------------------------------------------ 104 // Headers 105 106 typedef struct { 107 uint8_t key_frame_; 108 uint8_t profile_; 109 uint8_t show_; 110 uint32_t partition_length_; 111 } VP8FrameHeader; 112 113 typedef struct { 114 uint16_t width_; 115 uint16_t height_; 116 uint8_t xscale_; 117 uint8_t yscale_; 118 uint8_t colorspace_; // 0 = YCbCr 119 uint8_t clamp_type_; 120 } VP8PictureHeader; 121 122 // segment features 123 typedef struct { 124 int use_segment_; 125 int update_map_; // whether to update the segment map or not 126 int absolute_delta_; // absolute or delta values for quantizer and filter 127 int8_t quantizer_[NUM_MB_SEGMENTS]; // quantization changes 128 int8_t filter_strength_[NUM_MB_SEGMENTS]; // filter strength for segments 129 } VP8SegmentHeader; 130 131 // Struct collecting all frame-persistent probabilities. 132 typedef struct { 133 uint8_t segments_[MB_FEATURE_TREE_PROBS]; 134 // Type: 0:Intra16-AC 1:Intra16-DC 2:Chroma 3:Intra4 135 uint8_t coeffs_[NUM_TYPES][NUM_BANDS][NUM_CTX][NUM_PROBAS]; 136 #ifndef ONLY_KEYFRAME_CODE 137 uint8_t ymode_[4], uvmode_[3]; 138 uint8_t mv_[2][NUM_MV_PROBAS]; 139 #endif 140 } VP8Proba; 141 142 // Filter parameters 143 typedef struct { 144 int simple_; // 0=complex, 1=simple 145 int level_; // [0..63] 146 int sharpness_; // [0..7] 147 int use_lf_delta_; 148 int ref_lf_delta_[NUM_REF_LF_DELTAS]; 149 int mode_lf_delta_[NUM_MODE_LF_DELTAS]; 150 } VP8FilterHeader; 151 152 //------------------------------------------------------------------------------ 153 // Informations about the macroblocks. 154 155 typedef struct { // filter specs 156 unsigned int f_level_:6; // filter strength: 0..63 157 unsigned int f_ilevel_:6; // inner limit: 1..63 158 unsigned int f_inner_:1; // do inner filtering? 159 } VP8FInfo; 160 161 typedef struct { // used for syntax-parsing 162 unsigned int nz_:24; // non-zero AC/DC coeffs (24bit) 163 unsigned int dc_nz_:1; // non-zero DC coeffs 164 unsigned int skip_:1; // block type 165 } VP8MB; 166 167 // Dequantization matrices 168 typedef int quant_t[2]; // [DC / AC]. Can be 'uint16_t[2]' too (~slower). 169 typedef struct { 170 quant_t y1_mat_, y2_mat_, uv_mat_; 171 } VP8QuantMatrix; 172 173 // Persistent information needed by the parallel processing 174 typedef struct { 175 int id_; // cache row to process (in [0..2]) 176 int mb_y_; // macroblock position of the row 177 int filter_row_; // true if row-filtering is needed 178 VP8FInfo* f_info_; // filter strengths 179 VP8Io io_; // copy of the VP8Io to pass to put() 180 } VP8ThreadContext; 181 182 //------------------------------------------------------------------------------ 183 // VP8Decoder: the main opaque structure handed over to user 184 185 struct VP8Decoder { 186 VP8StatusCode status_; 187 int ready_; // true if ready to decode a picture with VP8Decode() 188 const char* error_msg_; // set when status_ is not OK. 189 190 // Main data source 191 VP8BitReader br_; 192 193 // headers 194 VP8FrameHeader frm_hdr_; 195 VP8PictureHeader pic_hdr_; 196 VP8FilterHeader filter_hdr_; 197 VP8SegmentHeader segment_hdr_; 198 199 // Worker 200 WebPWorker worker_; 201 int use_threads_; // use multi-thread 202 int cache_id_; // current cache row 203 int num_caches_; // number of cached rows of 16 pixels (1, 2 or 3) 204 VP8ThreadContext thread_ctx_; // Thread context 205 206 // dimension, in macroblock units. 207 int mb_w_, mb_h_; 208 209 // Macroblock to process/filter, depending on cropping and filter_type. 210 int tl_mb_x_, tl_mb_y_; // top-left MB that must be in-loop filtered 211 int br_mb_x_, br_mb_y_; // last bottom-right MB that must be decoded 212 213 // number of partitions. 214 int num_parts_; 215 // per-partition boolean decoders. 216 VP8BitReader parts_[MAX_NUM_PARTITIONS]; 217 218 // buffer refresh flags 219 // bit 0: refresh Gold, bit 1: refresh Alt 220 // bit 2-3: copy to Gold, bit 4-5: copy to Alt 221 // bit 6: Gold sign bias, bit 7: Alt sign bias 222 // bit 8: refresh last frame 223 uint32_t buffer_flags_; 224 225 // dequantization (one set of DC/AC dequant factor per segment) 226 VP8QuantMatrix dqm_[NUM_MB_SEGMENTS]; 227 228 // probabilities 229 VP8Proba proba_; 230 int use_skip_proba_; 231 uint8_t skip_p_; 232 #ifndef ONLY_KEYFRAME_CODE 233 uint8_t intra_p_, last_p_, golden_p_; 234 VP8Proba proba_saved_; 235 int update_proba_; 236 #endif 237 238 // Boundary data cache and persistent buffers. 239 uint8_t* intra_t_; // top intra modes values: 4 * mb_w_ 240 uint8_t intra_l_[4]; // left intra modes values 241 uint8_t* y_t_; // top luma samples: 16 * mb_w_ 242 uint8_t* u_t_, *v_t_; // top u/v samples: 8 * mb_w_ each 243 244 VP8MB* mb_info_; // contextual macroblock info (mb_w_ + 1) 245 VP8FInfo* f_info_; // filter strength info 246 uint8_t* yuv_b_; // main block for Y/U/V (size = YUV_SIZE) 247 int16_t* coeffs_; // 384 coeffs = (16+8+8) * 4*4 248 249 uint8_t* cache_y_; // macroblock row for storing unfiltered samples 250 uint8_t* cache_u_; 251 uint8_t* cache_v_; 252 int cache_y_stride_; 253 int cache_uv_stride_; 254 255 // main memory chunk for the above data. Persistent. 256 void* mem_; 257 size_t mem_size_; 258 259 // Per macroblock non-persistent infos. 260 int mb_x_, mb_y_; // current position, in macroblock units 261 uint8_t is_i4x4_; // true if intra4x4 262 uint8_t imodes_[16]; // one 16x16 mode (#0) or sixteen 4x4 modes 263 uint8_t uvmode_; // chroma prediction mode 264 uint8_t segment_; // block's segment 265 266 // bit-wise info about the content of each sub-4x4 blocks: there are 16 bits 267 // for luma (bits #0->#15), then 4 bits for chroma-u (#16->#19) and 4 bits for 268 // chroma-v (#20->#23), each corresponding to one 4x4 block in decoding order. 269 // If the bit is set, the 4x4 block contains some non-zero coefficients. 270 uint32_t non_zero_; 271 uint32_t non_zero_ac_; 272 273 // Filtering side-info 274 int filter_type_; // 0=off, 1=simple, 2=complex 275 int filter_row_; // per-row flag 276 VP8FInfo fstrengths_[NUM_MB_SEGMENTS][2]; // precalculated per-segment/type 277 278 // extensions 279 const uint8_t* alpha_data_; // compressed alpha data (if present) 280 size_t alpha_data_size_; 281 int is_alpha_decoded_; // true if alpha_data_ is decoded in alpha_plane_ 282 uint8_t* alpha_plane_; // output. Persistent, contains the whole data. 283 284 int layer_colorspace_; 285 const uint8_t* layer_data_; // compressed layer data (if present) 286 size_t layer_data_size_; 287 }; 288 289 //------------------------------------------------------------------------------ 290 // internal functions. Not public. 291 292 // in vp8.c 293 int VP8SetError(VP8Decoder* const dec, 294 VP8StatusCode error, const char* const msg); 295 296 // in tree.c 297 void VP8ResetProba(VP8Proba* const proba); 298 void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec); 299 void VP8ParseIntraMode(VP8BitReader* const br, VP8Decoder* const dec); 300 301 // in quant.c 302 void VP8ParseQuant(VP8Decoder* const dec); 303 304 // in frame.c 305 int VP8InitFrame(VP8Decoder* const dec, VP8Io* io); 306 // Predict a block and add residual 307 void VP8ReconstructBlock(VP8Decoder* const dec); 308 // Call io->setup() and finish setting up scan parameters. 309 // After this call returns, one must always call VP8ExitCritical() with the 310 // same parameters. Both functions should be used in pair. Returns VP8_STATUS_OK 311 // if ok, otherwise sets and returns the error status on *dec. 312 VP8StatusCode VP8EnterCritical(VP8Decoder* const dec, VP8Io* const io); 313 // Must always be called in pair with VP8EnterCritical(). 314 // Returns false in case of error. 315 int VP8ExitCritical(VP8Decoder* const dec, VP8Io* const io); 316 // Process the last decoded row (filtering + output) 317 int VP8ProcessRow(VP8Decoder* const dec, VP8Io* const io); 318 // To be called at the start of a new scanline, to initialize predictors. 319 void VP8InitScanline(VP8Decoder* const dec); 320 // Decode one macroblock. Returns false if there is not enough data. 321 int VP8DecodeMB(VP8Decoder* const dec, VP8BitReader* const token_br); 322 323 // in alpha.c 324 const uint8_t* VP8DecompressAlphaRows(VP8Decoder* const dec, 325 int row, int num_rows); 326 327 // in layer.c 328 int VP8DecodeLayer(VP8Decoder* const dec); 329 330 //------------------------------------------------------------------------------ 331 332 #if defined(__cplusplus) || defined(c_plusplus) 333 } // extern "C" 334 #endif 335 336 #endif /* WEBP_DEC_VP8I_H_ */ 337