libwebp: update snapshot to v0.3.0-rc6

third_party/libwebp/enc/vp8enci.h

 // Copyright 2011 Google Inc. All Rights Reserved.
 //
 // This code is licensed under the same terms as WebM:
 //  Software License Agreement:  http://www.webmproject.org/license/software/
 //  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
 // -----------------------------------------------------------------------------
 //
 //   WebP encoder: internal header.
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
 #ifndef WEBP_ENC_VP8ENCI_H_
 #define WEBP_ENC_VP8ENCI_H_
 
 #include <string.h>     // for memcpy()
 #include "../webp/encode.h"
 #include "../dsp/dsp.h"
 #include "../utils/bit_writer.h"
+#include "../utils/thread.h"
 
 #if defined(__cplusplus) || defined(c_plusplus)
 extern "C" {
 #endif
 
 //------------------------------------------------------------------------------
 // Various defines and enums
 
 // version numbers
 #define ENC_MAJ_VERSION 0
-#define ENC_MIN_VERSION 2
+#define ENC_MIN_VERSION 3
 #define ENC_REV_VERSION 0
 
-// size of histogram used by CollectHistogram.
-#define MAX_COEFF_THRESH   64
-
 // intra prediction modes
 enum { B_DC_PRED = 0,   // 4x4 modes
        B_TM_PRED = 1,
        B_VE_PRED = 2,
        B_HE_PRED = 3,
        B_RD_PRED = 4,
        B_VR_PRED = 5,
        B_LD_PRED = 6,
        B_VL_PRED = 7,
        B_HD_PRED = 8,
        B_HU_PRED = 9,
        NUM_BMODES = B_HU_PRED + 1 - B_DC_PRED,  // = 10
 
        // Luma16 or UV modes
        DC_PRED = B_DC_PRED, V_PRED = B_VE_PRED,
-       H_PRED = B_HE_PRED, TM_PRED = B_TM_PRED
+       H_PRED = B_HE_PRED, TM_PRED = B_TM_PRED,
+       NUM_PRED_MODES = 4
      };
 
 enum { NUM_MB_SEGMENTS = 4,
        MAX_NUM_PARTITIONS = 8,
        NUM_TYPES = 4,   // 0: i16-AC,  1: i16-DC,  2:chroma-AC,  3:i4-AC
        NUM_BANDS = 8,
        NUM_CTX = 3,
        NUM_PROBAS = 11,
        MAX_LF_LEVELS = 64,      // Maximum loop filter level
        MAX_VARIABLE_LEVEL = 67  // last (inclusive) level with variable cost
      };
 
+typedef enum {   // Rate-distortion optimization levels
+  RD_OPT_NONE        = 0,  // no rd-opt
+  RD_OPT_BASIC       = 1,  // basic scoring (no trellis)
+  RD_OPT_TRELLIS     = 2,  // perform trellis-quant on the final decision only
+  RD_OPT_TRELLIS_ALL = 3   // trellis-quant for every scoring (much slower)
+} VP8RDLevel;
+
 // YUV-cache parameters. Cache is 16-pixels wide.
 // The original or reconstructed samples can be accessed using VP8Scan[]
 // The predicted blocks can be accessed using offsets to yuv_p_ and
 // the arrays VP8*ModeOffsets[];
 //         +----+      YUV Samples area. See VP8Scan[] for accessing the blocks.
 //  Y_OFF  |YYYY| <- original samples  (enc->yuv_in_)
 //         |YYYY|
 //         |YYYY|
 //         |YYYY|
 //  U_OFF  |UUVV| V_OFF  (=U_OFF + 8)
@@ skipping 80 matching lines @@
 typedef int64_t score_t;     // type used for scores, rate, distortion
 #define MAX_COST ((score_t)0x7fffffffffffffLL)
 
 #define QFIX 17
 #define BIAS(b)  ((b) << (QFIX - 8))
 // Fun fact: this is the _only_ line where we're actually being lossy and
 // discarding bits.
 static WEBP_INLINE int QUANTDIV(int n, int iQ, int B) {
   return (n * iQ + B) >> QFIX;
 }
-extern const uint8_t VP8Zigzag[16];
+
+// size of histogram used by CollectHistogram.
+#define MAX_COEFF_THRESH   31
+typedef struct VP8Histogram VP8Histogram;
+struct VP8Histogram {
+  // TODO(skal): we only need to store the max_value and last_non_zero actually.
+  int distribution[MAX_COEFF_THRESH + 1];
+};
+
+// Uncomment the following to remove token-buffer code:
+// #define DISABLE_TOKEN_BUFFER
 
 //------------------------------------------------------------------------------
 // Headers
 
+typedef uint32_t proba_t;   // 16b + 16b
 typedef uint8_t ProbaArray[NUM_CTX][NUM_PROBAS];
-typedef uint64_t StatsArray[NUM_CTX][NUM_PROBAS][2];
+typedef proba_t StatsArray[NUM_CTX][NUM_PROBAS];
 typedef uint16_t CostArray[NUM_CTX][MAX_VARIABLE_LEVEL + 1];
 typedef double LFStats[NUM_MB_SEGMENTS][MAX_LF_LEVELS];  // filter stats
 
 typedef struct VP8Encoder VP8Encoder;
 
 // segment features
 typedef struct {
   int num_segments_;      // Actual number of segments. 1 segment only = unused.
   int update_map_;        // whether to update the segment map or not.
                           // must be 0 if there's only 1 segment.
   int size_;              // bit-cost for transmitting the segment map
 } VP8SegmentHeader;
 
 // Struct collecting all frame-persistent probabilities.
 typedef struct {
   uint8_t segments_[3];     // probabilities for segment tree
   uint8_t skip_proba_;      // final probability of being skipped.
   ProbaArray coeffs_[NUM_TYPES][NUM_BANDS];      // 924 bytes
-  StatsArray stats_[NUM_TYPES][NUM_BANDS];       // 7.4k
+  StatsArray stats_[NUM_TYPES][NUM_BANDS];       // 4224 bytes
   CostArray level_cost_[NUM_TYPES][NUM_BANDS];   // 11.4k
+  int dirty_;               // if true, need to call VP8CalculateLevelCosts()
   int use_skip_proba_;      // Note: we always use skip_proba for now.
   int nb_skip_;             // number of skipped blocks
 } VP8Proba;
 
 // Filter parameters. Not actually used in the code (we don't perform
 // the in-loop filtering), but filled from user's config
 typedef struct {
   int simple_;             // filtering type: 0=complex, 1=simple
   int level_;              // base filter level [0..63]
   int sharpness_;          // [0..7]
@@ skipping 105 matching lines @@
 // Helper functions to set mode properties
 void VP8SetIntra16Mode(const VP8EncIterator* const it, int mode);
 void VP8SetIntra4Mode(const VP8EncIterator* const it, const uint8_t* modes);
 void VP8SetIntraUVMode(const VP8EncIterator* const it, int mode);
 void VP8SetSkip(const VP8EncIterator* const it, int skip);
 void VP8SetSegment(const VP8EncIterator* const it, int segment);
 
 //------------------------------------------------------------------------------
 // Paginated token buffer
 
-// WIP: #define USE_TOKEN_BUFFER
-
-#ifdef USE_TOKEN_BUFFER
-
-#define MAX_NUM_TOKEN 2048
-
-typedef struct VP8Tokens VP8Tokens;
-struct VP8Tokens {
-  uint16_t tokens_[MAX_NUM_TOKEN];  // bit#15: bit, bits 0..14: slot
-  int left_;
-  VP8Tokens* next_;
-};
+typedef struct VP8Tokens VP8Tokens;  // struct details in token.c
 
 typedef struct {
-  VP8Tokens* rows_;
-  uint16_t* tokens_;    // set to (*last_)->tokens_
-  VP8Tokens** last_;
-  int left_;
-  int error_;  // true in case of malloc error
+#if !defined(DISABLE_TOKEN_BUFFER)
+  VP8Tokens* pages_;        // first page
+  VP8Tokens** last_page_;   // last page
+  uint16_t* tokens_;        // set to (*last_page_)->tokens_
+  int left_;          // how many free tokens left before the page is full.
+#endif
+  int error_;         // true in case of malloc error
 } VP8TBuffer;
 
 void VP8TBufferInit(VP8TBuffer* const b);    // initialize an empty buffer
-int VP8TBufferNewPage(VP8TBuffer* const b);  // allocate a new page
-void VP8TBufferClear(VP8TBuffer* const b);   // de-allocate memory
+void VP8TBufferClear(VP8TBuffer* const b);   // de-allocate pages memory
 
-int VP8EmitTokens(const VP8TBuffer* const b, VP8BitWriter* const bw,
-                  const uint8_t* const probas);
+#if !defined(DISABLE_TOKEN_BUFFER)
 
-static WEBP_INLINE int VP8AddToken(VP8TBuffer* const b,
-                                   int bit, int proba_idx) {
-  if (b->left_ > 0 || VP8TBufferNewPage(b)) {
-    const int slot = --b->left_;
-    b->tokens_[slot] = (bit << 15) | proba_idx;
-  }
-  return bit;
-}
+// Finalizes bitstream when probabilities are known.
+// Deletes the allocated token memory if final_pass is true.
+int VP8EmitTokens(VP8TBuffer* const b, VP8BitWriter* const bw,
+                  const uint8_t* const probas, int final_pass);
 
-#endif  // USE_TOKEN_BUFFER
+// record the coding of coefficients without knowing the probabilities yet
+int VP8RecordCoeffTokens(int ctx, int coeff_type, int first, int last,
+                         const int16_t* const coeffs,
+                         VP8TBuffer* const tokens);
+
+// unused for now
+void VP8TokenToStats(const VP8TBuffer* const b, proba_t* const stats);
+
+#endif  // !DISABLE_TOKEN_BUFFER
 
 //------------------------------------------------------------------------------
 // VP8Encoder
 
 struct VP8Encoder {
   const WebPConfig* config_;    // user configuration and parameters
   WebPPicture* pic_;            // input / output picture
 
   // headers
   VP8FilterHeader   filter_hdr_;     // filtering information
   VP8SegmentHeader  segment_hdr_;    // segment information
 
   int profile_;                      // VP8's profile, deduced from Config.
 
   // dimension, in macroblock units.
   int mb_w_, mb_h_;
   int preds_w_;   // stride of the *preds_ prediction plane (=4*mb_w + 1)
 
   // number of partitions (1, 2, 4 or 8 = MAX_NUM_PARTITIONS)
   int num_parts_;
 
   // per-partition boolean decoders.
   VP8BitWriter bw_;                         // part0
   VP8BitWriter parts_[MAX_NUM_PARTITIONS];  // token partitions
+  VP8TBuffer tokens_;                       // token buffer
 
   int percent_;                             // for progress
 
   // transparency blob
   int has_alpha_;
   uint8_t* alpha_data_;       // non-NULL if transparency is present
   uint32_t alpha_data_size_;
+  WebPWorker alpha_worker_;
 
   // enhancement layer
   int use_layer_;
   VP8BitWriter layer_bw_;
   uint8_t* layer_data_;
   size_t layer_data_size_;
 
   // quantization info (one set of DC/AC dequant factor per segment)
   VP8SegmentInfo dqm_[NUM_MB_SEGMENTS];
   int base_quant_;                 // nominal quantizer value. Only used
                                    // for relative coding of segments' quant.
+  int alpha_;                      // global susceptibility (<=> complexity)
   int uv_alpha_;                   // U/V quantization susceptibility
   // global offset of quantizers, shared by all segments
   int dq_y1_dc_;
   int dq_y2_dc_, dq_y2_ac_;
   int dq_uv_dc_, dq_uv_ac_;
 
   // probabilities and statistics
   VP8Proba proba_;
   uint64_t sse_[4];        // sum of Y/U/V/A squared errors for all macroblocks
   uint64_t sse_count_;     // pixel count for the sse_[] stats
   int      coded_size_;
   int      residual_bytes_[3][4];
   int      block_count_[3];
 
   // quality/speed settings
-  int method_;              // 0=fastest, 6=best/slowest.
-  int rd_opt_level_;        // Deduced from method_.
-  int max_i4_header_bits_;  // partition #0 safeness factor
+  int method_;               // 0=fastest, 6=best/slowest.
+  VP8RDLevel rd_opt_level_;  // Deduced from method_.
+  int max_i4_header_bits_;   // partition #0 safeness factor
+  int thread_level_;         // derived from config->thread_level
+  int do_search_;            // derived from config->target_XXX
+  int use_tokens_;           // if true, use token buffer
 
   // Memory
   VP8MBInfo* mb_info_;   // contextual macroblock infos (mb_w_ + 1)
   uint8_t*   preds_;     // predictions modes: (4*mb_w+1) * (4*mb_h+1)
   uint32_t*  nz_;        // non-zero bit context: mb_w+1
   uint8_t*   yuv_in_;    // input samples
   uint8_t*   yuv_out_;   // output samples
   uint8_t*   yuv_out2_;  // secondary scratch out-buffer. swapped with yuv_out_.
   uint8_t*   yuv_p_;     // scratch buffer for prediction
   uint8_t   *y_top_;     // top luma samples.
@@ skipping 23 matching lines @@
   // in syntax.c
 // Generates the final bitstream by coding the partition0 and headers,
 // and appending an assembly of all the pre-coded token partitions.
 // Return true if everything is ok.
 int VP8EncWrite(VP8Encoder* const enc);
 // Release memory allocated for bit-writing in VP8EncLoop & seq.
 void VP8EncFreeBitWriters(VP8Encoder* const enc);
 
   // in frame.c
 extern const uint8_t VP8EncBands[16 + 1];
+extern const uint8_t VP8Cat3[];
+extern const uint8_t VP8Cat4[];
+extern const uint8_t VP8Cat5[];
+extern const uint8_t VP8Cat6[];
+
 // Form all the four Intra16x16 predictions in the yuv_p_ cache
 void VP8MakeLuma16Preds(const VP8EncIterator* const it);
 // Form all the four Chroma8x8 predictions in the yuv_p_ cache
 void VP8MakeChroma8Preds(const VP8EncIterator* const it);
 // Form all the ten Intra4x4 predictions in the yuv_p_ cache
 // for the 4x4 block it->i4_
 void VP8MakeIntra4Preds(const VP8EncIterator* const it);
 // Rate calculation
 int VP8GetCostLuma16(VP8EncIterator* const it, const VP8ModeScore* const rd);
 int VP8GetCostLuma4(VP8EncIterator* const it, const int16_t levels[16]);
 int VP8GetCostUV(VP8EncIterator* const it, const VP8ModeScore* const rd);
-// Main stat / coding passes
+// Main coding calls
 int VP8EncLoop(VP8Encoder* const enc);
-int VP8StatLoop(VP8Encoder* const enc);
+int VP8EncTokenLoop(VP8Encoder* const enc);
 
   // in webpenc.c
 // Assign an error code to a picture. Return false for convenience.
 int WebPEncodingSetError(const WebPPicture* const pic, WebPEncodingError error);
 int WebPReportProgress(const WebPPicture* const pic,
                        int percent, int* const percent_store);
 
   // in analysis.c
 // Main analysis loop. Decides the segmentations and complexity.
 // Assigns a first guess for Intra16 and uvmode_ prediction modes.
 int VP8EncAnalyze(VP8Encoder* const enc);
 
   // in quant.c
 // Sets up segment's quantization values, base_quant_ and filter strengths.
 void VP8SetSegmentParams(VP8Encoder* const enc, float quality);
 // Pick best modes and fills the levels. Returns true if skipped.
-int VP8Decimate(VP8EncIterator* const it, VP8ModeScore* const rd, int rd_opt);
+int VP8Decimate(VP8EncIterator* const it, VP8ModeScore* const rd,
+                VP8RDLevel rd_opt);
 
   // in alpha.c
 void VP8EncInitAlpha(VP8Encoder* const enc);    // initialize alpha compression
+int VP8EncStartAlpha(VP8Encoder* const enc);    // start alpha coding process
 int VP8EncFinishAlpha(VP8Encoder* const enc);   // finalize compressed data
-void VP8EncDeleteAlpha(VP8Encoder* const enc);  // delete compressed data
+int VP8EncDeleteAlpha(VP8Encoder* const enc);   // delete compressed data
 
   // in layer.c
 void VP8EncInitLayer(VP8Encoder* const enc);     // init everything
 void VP8EncCodeLayerBlock(VP8EncIterator* it);   // code one more macroblock
 int VP8EncFinishLayer(VP8Encoder* const enc);    // finalize coding
 void VP8EncDeleteLayer(VP8Encoder* enc);         // reclaim memory
 
   // in filter.c
 
 // SSIM utils
@@ skipping 12 matching lines @@
 void VP8StoreFilterStats(VP8EncIterator* const it);
 void VP8AdjustFilterStrength(VP8EncIterator* const it);
 
 //------------------------------------------------------------------------------
 
 #if defined(__cplusplus) || defined(c_plusplus)
 }    // extern "C"
 #endif
 
 #endif  /* WEBP_ENC_VP8ENCI_H_ */