Merge "Reduce mem copies in encoder loopfilter level picking"
[profile/ivi/libvpx.git] / vp8 / encoder / onyx_int.h
index c460b9d..68ec3ab 100644 (file)
@@ -13,7 +13,7 @@
 #define __INC_VP8_INT_H
 
 #include <stdio.h>
-#include "vpx_ports/config.h"
+#include "vpx_config.h"
 #include "vp8/common/onyx.h"
 #include "treewriter.h"
 #include "tokenize.h"
 #define VP8_TEMPORAL_ALT_REF 1
 #endif
 
+#define MAX_PERIODICITY 16
+
+#define MAX(x,y) (((x)>(y))?(x):(y))
+#define MIN(x,y) (((x)<(y))?(x):(y))
+
 typedef struct
 {
     int kf_indicated;
@@ -108,6 +113,7 @@ typedef struct
     double MVrv;
     double MVcv;
     double mv_in_out_count;
+    double new_mv_count;
     double duration;
     double count;
 }
@@ -130,32 +136,32 @@ typedef struct
 
 typedef enum
 {
-    THR_ZEROMV         = 0,
+    THR_ZERO         = 0,
     THR_DC             = 1,
 
-    THR_NEARESTMV      = 2,
-    THR_NEARMV         = 3,
+    THR_NEAREST      = 2,
+    THR_NEAR         = 3,
 
-    THR_ZEROG          = 4,
-    THR_NEARESTG       = 5,
+    THR_ZERO2          = 4,
+    THR_NEAREST2       = 5,
 
-    THR_ZEROA          = 6,
-    THR_NEARESTA       = 7,
+    THR_ZERO3          = 6,
+    THR_NEAREST3       = 7,
 
-    THR_NEARG          = 8,
-    THR_NEARA          = 9,
+    THR_NEAR2          = 8,
+    THR_NEAR3          = 9,
 
     THR_V_PRED         = 10,
     THR_H_PRED         = 11,
     THR_TM             = 12,
 
-    THR_NEWMV          = 13,
-    THR_NEWG           = 14,
-    THR_NEWA           = 15,
+    THR_NEW          = 13,
+    THR_NEW2           = 14,
+    THR_NEW3           = 15,
 
-    THR_SPLITMV        = 16,
-    THR_SPLITG         = 17,
-    THR_SPLITA         = 18,
+    THR_SPLIT        = 16,
+    THR_SPLIT2         = 17,
+    THR_SPLIT3         = 18,
 
     THR_B_PRED         = 19,
 }
@@ -180,9 +186,6 @@ typedef struct
     int half_pixel_search;
     int quarter_pixel_search;
     int thresh_mult[MAX_MODES];
-    int full_freq[2];
-    int min_fs_radius;
-    int max_fs_radius;
     int max_step_search_steps;
     int first_step;
     int optimize_coefficients;
@@ -240,6 +243,52 @@ enum
     BLOCK_MAX_SEGMENTS
 };
 
+typedef struct
+{
+    // Layer configuration
+    double frame_rate;
+    int target_bandwidth;
+
+    // Layer specific coding parameters
+    int starting_buffer_level;
+    int optimal_buffer_level;
+    int maximum_buffer_size;
+
+    int avg_frame_size_for_layer;
+
+    int buffer_level;
+    int bits_off_target;
+
+    long long total_actual_bits;
+    int total_target_vs_actual;
+
+    int worst_quality;
+    int active_worst_quality;
+    int best_quality;
+    int active_best_quality;
+
+    int ni_av_qi;
+    int ni_tot_qi;
+    int ni_frames;
+    int avg_frame_qindex;
+
+    double rate_correction_factor;
+    double key_frame_rate_correction_factor;
+    double gf_rate_correction_factor;
+
+    int zbin_over_quant;
+
+    int inter_frame_target;
+    INT64 total_byte_count;
+
+    int filter_level;
+
+    int last_frame_percent_intra;
+
+    int count_mb_ref_frame_usage[MAX_REF_FRAMES];
+
+} LAYER_CONTEXT;
+
 typedef struct VP8_COMP
 {
 
@@ -268,8 +317,7 @@ typedef struct VP8_COMP
 
     MACROBLOCK mb;
     VP8_COMMON common;
-    vp8_writer bc, bc2;
-    // bool_writer *bc2;
+    vp8_writer bc[9]; // one boolcoder for each partition
 
     VP8_CONFIG oxcf;
 
@@ -291,7 +339,7 @@ typedef struct VP8_COMP
     int gold_is_alt;  // don't do both alt and gold search ( just do gold).
 
     //int refresh_alt_ref_frame;
-    YV12_BUFFER_CONFIG last_frame_uf;
+    YV12_BUFFER_CONFIG pick_lf_lvl_frame;
 
     TOKENEXTRA *tok;
     unsigned int tok_count;
@@ -320,10 +368,10 @@ typedef struct VP8_COMP
     CODING_CONTEXT coding_context;
 
     // Rate targetting variables
-    long long prediction_error;
-    long long last_prediction_error;
-    long long intra_error;
-    long long last_intra_error;
+    int64_t prediction_error;
+    int64_t last_prediction_error;
+    int64_t intra_error;
+    int64_t last_intra_error;
 
     int this_frame_target;
     int projected_frame_size;
@@ -346,20 +394,16 @@ typedef struct VP8_COMP
     int baseline_gf_interval;
     int active_arnr_frames;           // <= cpi->oxcf.arnr_max_frames
 
-    INT64 key_frame_count;
+    int64_t key_frame_count;
     int prior_key_frame_distance[KEY_FRAME_CONTEXT];
     int per_frame_bandwidth;          // Current section per frame bandwidth target
     int av_per_frame_bandwidth;        // Average frame size target for clip
     int min_frame_bandwidth;          // Minimum allocation that should be used for any frame
-    int buffered_av_per_frame_bandwidth; // Average bitrate over the last buffer
-    int buffered_av_per_frame_bandwidth_rem; // Average bitrate remainder
-    int accumulated_overshoot;           // Accumulated # of bits spent > target
-
     int inter_frame_target;
     double output_frame_rate;
-    long long last_time_stamp_seen;
-    long long last_end_time_stamp_seen;
-    long long first_time_stamp_ever;
+    int64_t last_time_stamp_seen;
+    int64_t last_end_time_stamp_seen;
+    int64_t first_time_stamp_ever;
 
     int ni_av_qi;
     int ni_tot_qi;
@@ -369,12 +413,15 @@ typedef struct VP8_COMP
     int zbin_over_quant;
     int zbin_mode_boost;
     int zbin_mode_boost_enabled;
+    int last_zbin_over_quant;
+    int last_zbin_mode_boost;
 
-    INT64 total_byte_count;
+    int64_t total_byte_count;
 
     int buffered_mode;
 
-    int buffer_level;
+    double frame_rate;
+    int64_t buffer_level;
     int bits_off_target;
 
     int rolling_target_bits;
@@ -383,7 +430,7 @@ typedef struct VP8_COMP
     int long_rolling_target_bits;
     int long_rolling_actual_bits;
 
-    long long total_actual_bits;
+    int64_t total_actual_bits;
     int total_target_vs_actual;        // debug stats
 
     int worst_quality;
@@ -405,11 +452,11 @@ typedef struct VP8_COMP
 
     unsigned int MVcount [2] [MVvals];  /* (row,col) MV cts this frame */
 
-    unsigned int coef_counts [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens];  /* for this frame */
-    //DECLARE_ALIGNED(16, int, coef_counts_backup [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens]);   //not used any more
+    unsigned int coef_counts [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS];  /* for this frame */
+    //DECLARE_ALIGNED(16, int, coef_counts_backup [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [MAX_ENTROPY_TOKENS]);   //not used any more
     //save vp8_tree_probs_from_distribution result for each frame to avoid repeat calculation
-    vp8_prob frame_coef_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens-1];
-    unsigned int frame_branch_ct [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens-1][2];
+    vp8_prob frame_coef_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES];
+    unsigned int frame_branch_ct [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES][2];
 
     int gfu_boost;
     int kf_boost;
@@ -505,6 +552,7 @@ typedef struct VP8_COMP
 #endif
 
     TOKENLIST *tplist;
+    unsigned int partition_sz[MAX_PARTITIONS];
     // end of multithread data
 
 
@@ -528,18 +576,15 @@ typedef struct VP8_COMP
         unsigned int this_iiratio;
         FIRSTPASS_STATS *total_stats;
         FIRSTPASS_STATS *this_frame_stats;
-        FIRSTPASS_STATS *stats_in, *stats_in_end;
+        FIRSTPASS_STATS *stats_in, *stats_in_end, *stats_in_start;
+        FIRSTPASS_STATS *total_left_stats;
         int first_pass_done;
-        long long bits_left;
-        long long clip_bits_total;
+        int64_t bits_left;
+        int64_t clip_bits_total;
         double avg_iiratio;
         double modified_error_total;
         double modified_error_used;
         double modified_error_left;
-        double total_error_left;
-        double total_intra_error_left;
-        double total_coded_error_left;
-        double start_tot_err_left;
         double kf_intra_err_min;
         double gf_intra_err_min;
         int frames_to_key;
@@ -551,14 +596,14 @@ typedef struct VP8_COMP
         int gf_group_error_left;           // Remaining error from uncoded frames in a gf group. Two pass use only
 
         // Projected total bits available for a key frame group of frames
-        long long kf_group_bits;
+        int64_t kf_group_bits;
 
         // Error score of frames still to be coded in kf group
-        long long kf_group_error_left;
+        int64_t kf_group_error_left;
 
         int gf_group_bits;                // Projected Bits available for a group of frames including 1 GF or ARF
         int gf_bits;                     // Bits for the golden frame or ARF - 2 pass only
-        int mid_gf_extra_bits;             // A few extra bits for the frame half way between two gfs.
+        int alt_extra_bits;
         double est_max_qcorrection_factor;
     } twopass;
 
@@ -608,21 +653,47 @@ typedef struct VP8_COMP
     unsigned char *gf_active_flags;
     int gf_active_count;
 
+    int output_partition;
+
     //Store last frame's MV info for next frame MV prediction
     int_mv *lfmv;
     int *lf_ref_frame_sign_bias;
     int *lf_ref_frame;
 
-#if CONFIG_REALTIME_ONLY
     int force_next_frame_intra; /* force next frame to intra when kf_auto says so */
+
+    int droppable;
+
+    // Coding layer state variables
+    unsigned int current_layer;
+    LAYER_CONTEXT layer_context[MAX_LAYERS];
+
+    long long frames_in_layer[MAX_LAYERS];
+    long long bytes_in_layer[MAX_LAYERS];
+    double sum_psnr[MAX_LAYERS];
+    double sum_psnr_p[MAX_LAYERS];
+    double total_error2[MAX_LAYERS];
+    double total_error2_p[MAX_LAYERS];
+    double sum_ssim[MAX_LAYERS];
+    double sum_weights[MAX_LAYERS];
+
+    double total_ssimg_y_in_layer[MAX_LAYERS];
+    double total_ssimg_u_in_layer[MAX_LAYERS];
+    double total_ssimg_v_in_layer[MAX_LAYERS];
+    double total_ssimg_all_in_layer[MAX_LAYERS];
+
+#if CONFIG_MULTI_RES_ENCODING
+    /* Number of MBs per row at lower-resolution level */
+    int    mr_low_res_mb_cols;
 #endif
+
 } VP8_COMP;
 
 void control_data_rate(VP8_COMP *cpi);
 
 void vp8_encode_frame(VP8_COMP *cpi);
 
-void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest, unsigned long *size);
+void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest, unsigned char *dest_end, unsigned long *size);
 
 void vp8_activity_masking(VP8_COMP *cpi, MACROBLOCK *x);