vpx/vp10/common/onyxc_int.h

/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#ifndef VP10_COMMON_ONYXC_INT_H_
#define VP10_COMMON_ONYXC_INT_H_

#include "./vpx_config.h"
#include "vpx/internal/vpx_codec_internal.h"
#include "vpx_util/vpx_thread.h"
#include "./vp10_rtcd.h"
#include "vp10/common/alloccommon.h"
#include "vp10/common/loopfilter.h"
#include "vp10/common/entropymv.h"
#include "vp10/common/entropy.h"
#include "vp10/common/entropymode.h"
#include "vp10/common/frame_buffers.h"
#include "vp10/common/quant_common.h"
#include "vp10/common/tile_common.h"

#if CONFIG_VP9_POSTPROC
#include "vp10/common/postproc.h"
#endif

#ifdef __cplusplus
extern "C" {
#endif

#define REFS_PER_FRAME (ALTREF_FRAME - LAST_FRAME + 1)

#define REF_FRAMES_LOG2 3
#define REF_FRAMES (1 << REF_FRAMES_LOG2)

// 4 scratch frames for the new frames to support a maximum of 4 cores decoding
// in parallel, 3 for scaled references on the encoder.
// TODO(hkuang): Add ondemand frame buffers instead of hardcoding the number
// of framebuffers.
// TODO(jkoleszar): These 3 extra references could probably come from the
// normal reference pool.
#define FRAME_BUFFERS (REF_FRAMES + 7)

#define FRAME_CONTEXTS_LOG2 2
#define FRAME_CONTEXTS (1 << FRAME_CONTEXTS_LOG2)

#define NUM_PING_PONG_BUFFERS 2

typedef enum {
  SINGLE_REFERENCE      = 0,
  COMPOUND_REFERENCE    = 1,
  REFERENCE_MODE_SELECT = 2,
  REFERENCE_MODES       = 3,
} REFERENCE_MODE;

typedef enum {
  RESET_FRAME_CONTEXT_NONE = 0,
  RESET_FRAME_CONTEXT_CURRENT = 1,
  RESET_FRAME_CONTEXT_ALL = 2,
} RESET_FRAME_CONTEXT_MODE;

typedef enum {
  /**
   * Don't update frame context
   */
  REFRESH_FRAME_CONTEXT_OFF,
  /**
   * Update frame context to values resulting from forward probability
   * updates signaled in the frame header
   */
  REFRESH_FRAME_CONTEXT_FORWARD,
  /**
   * Update frame context to values resulting from backward probability
   * updates based on entropy/counts in the decoded frame
   */
  REFRESH_FRAME_CONTEXT_BACKWARD,
} REFRESH_FRAME_CONTEXT_MODE;

typedef struct {
  int_mv mv[2];
  MV_REFERENCE_FRAME ref_frame[2];
} MV_REF;

typedef struct {
  int ref_count;
  MV_REF *mvs;
  int mi_rows;
  int mi_cols;
  vpx_codec_frame_buffer_t raw_frame_buffer;
  YV12_BUFFER_CONFIG buf;

  // The Following variables will only be used in frame parallel decode.

  // frame_worker_owner indicates which FrameWorker owns this buffer. NULL means
  // that no FrameWorker owns, or is decoding, this buffer.
  VPxWorker *frame_worker_owner;

  // row and col indicate which position frame has been decoded to in real
  // pixel unit. They are reset to -1 when decoding begins and set to INT_MAX
  // when the frame is fully decoded.
  int row;
  int col;
} RefCntBuffer;

typedef struct BufferPool {
  // Protect BufferPool from being accessed by several FrameWorkers at
  // the same time during frame parallel decode.
  // TODO(hkuang): Try to use atomic variable instead of locking the whole pool.
#if CONFIG_MULTITHREAD
  pthread_mutex_t pool_mutex;
#endif

  // Private data associated with the frame buffer callbacks.
  void *cb_priv;

  vpx_get_frame_buffer_cb_fn_t get_fb_cb;
  vpx_release_frame_buffer_cb_fn_t release_fb_cb;

  RefCntBuffer frame_bufs[FRAME_BUFFERS];

  // Frame buffers allocated internally by the codec.
  InternalFrameBufferList int_frame_buffers;
} BufferPool;

typedef struct VP10Common {
  struct vpx_internal_error_info  error;
  vpx_color_space_t color_space;
  int color_range;
  int width;
  int height;
  int display_width;
  int display_height;
  int last_width;
  int last_height;

  // TODO(jkoleszar): this implies chroma ss right now, but could vary per
  // plane. Revisit as part of the future change to YV12_BUFFER_CONFIG to
  // support additional planes.
  int subsampling_x;
  int subsampling_y;

#if CONFIG_VP9_HIGHBITDEPTH
  int use_highbitdepth;  // Marks if we need to use 16bit frame buffers.
#endif

  YV12_BUFFER_CONFIG *frame_to_show;
  RefCntBuffer *prev_frame;

  // TODO(hkuang): Combine this with cur_buf in macroblockd.
  RefCntBuffer *cur_frame;

  int ref_frame_map[REF_FRAMES]; /* maps fb_idx to reference slot */

  // Prepare ref_frame_map for the next frame.
  // Only used in frame parallel decode.
  int next_ref_frame_map[REF_FRAMES];

  // TODO(jkoleszar): could expand active_ref_idx to 4, with 0 as intra, and
  // roll new_fb_idx into it.

  // Each frame can reference REFS_PER_FRAME buffers
  RefBuffer frame_refs[REFS_PER_FRAME];

  int new_fb_idx;

#if CONFIG_VP9_POSTPROC
  YV12_BUFFER_CONFIG post_proc_buffer;
  YV12_BUFFER_CONFIG post_proc_buffer_int;
#endif

  FRAME_TYPE last_frame_type;  /* last frame's frame type for motion search.*/
  FRAME_TYPE frame_type;

  int show_frame;
  int last_show_frame;
  int show_existing_frame;

  // Flag signaling that the frame is encoded using only INTRA modes.
  uint8_t intra_only;
  uint8_t last_intra_only;

  int allow_high_precision_mv;

  // Flag signaling which frame contexts should be reset to default values.
  RESET_FRAME_CONTEXT_MODE reset_frame_context;

  // MBs, mb_rows/cols is in 16-pixel units; mi_rows/cols is in
  // MODE_INFO (8-pixel) units.
  int MBs;
  int mb_rows, mi_rows;
  int mb_cols, mi_cols;
  int mi_stride;

  /* profile settings */
  TX_MODE tx_mode;

  int base_qindex;
  int y_dc_delta_q;
  int uv_dc_delta_q;
  int uv_ac_delta_q;
  int16_t y_dequant[MAX_SEGMENTS][2];
  int16_t uv_dequant[MAX_SEGMENTS][2];

  /* We allocate a MODE_INFO struct for each macroblock, together with
     an extra row on top and column on the left to simplify prediction. */
  int mi_alloc_size;
  MODE_INFO *mip; /* Base of allocated array */
  MODE_INFO *mi;  /* Corresponds to upper left visible macroblock */

  // TODO(agrange): Move prev_mi into encoder structure.
  // prev_mip and prev_mi will only be allocated in VP9 encoder.
  MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */
  MODE_INFO *prev_mi;  /* 'mi' from last frame (points into prev_mip) */

  // Separate mi functions between encoder and decoder.
  int (*alloc_mi)(struct VP10Common *cm, int mi_size);
  void (*free_mi)(struct VP10Common *cm);
  void (*setup_mi)(struct VP10Common *cm);

  // Grid of pointers to 8x8 MODE_INFO structs.  Any 8x8 not in the visible
  // area will be NULL.
  MODE_INFO **mi_grid_base;
  MODE_INFO **mi_grid_visible;
  MODE_INFO **prev_mi_grid_base;
  MODE_INFO **prev_mi_grid_visible;

  // Whether to use previous frame's motion vectors for prediction.
  int use_prev_frame_mvs;

  // Persistent mb segment id map used in prediction.
  int seg_map_idx;
  int prev_seg_map_idx;

  uint8_t *seg_map_array[NUM_PING_PONG_BUFFERS];
  uint8_t *last_frame_seg_map;
  uint8_t *current_frame_seg_map;
  int seg_map_alloc_size;

  INTERP_FILTER interp_filter;

  loop_filter_info_n lf_info;

  // Flag signaling how frame contexts should be updated at the end of
  // a frame decode
  REFRESH_FRAME_CONTEXT_MODE refresh_frame_context;

  int ref_frame_sign_bias[MAX_REF_FRAMES];    /* Two state 0, 1 */

  struct loopfilter lf;
  struct segmentation seg;

  int frame_parallel_decode;  // frame-based threading.

  // Context probabilities for reference frame prediction
  MV_REFERENCE_FRAME comp_fixed_ref;
  MV_REFERENCE_FRAME comp_var_ref[2];
  REFERENCE_MODE reference_mode;

  FRAME_CONTEXT *fc;  /* this frame entropy */
  FRAME_CONTEXT *frame_contexts;   // FRAME_CONTEXTS
  unsigned int  frame_context_idx; /* Context to use/update */
  FRAME_COUNTS counts;

  unsigned int current_video_frame;
  BITSTREAM_PROFILE profile;

  // VPX_BITS_8 in profile 0 or 1, VPX_BITS_10 or VPX_BITS_12 in profile 2 or 3.
  vpx_bit_depth_t bit_depth;
  vpx_bit_depth_t dequant_bit_depth;  // bit_depth of current dequantizer

#if CONFIG_VP9_POSTPROC
  struct postproc_state  postproc_state;
#endif

  int error_resilient_mode;

  int log2_tile_cols, log2_tile_rows;
  int byte_alignment;
  int skip_loop_filter;

  // Private data associated with the frame buffer callbacks.
  void *cb_priv;
  vpx_get_frame_buffer_cb_fn_t get_fb_cb;
  vpx_release_frame_buffer_cb_fn_t release_fb_cb;

  // Handles memory for the codec.
  InternalFrameBufferList int_frame_buffers;

  // External BufferPool passed from outside.
  BufferPool *buffer_pool;

  PARTITION_CONTEXT *above_seg_context;
  ENTROPY_CONTEXT *above_context;
  int above_context_alloc_cols;
} VP10_COMMON;

// TODO(hkuang): Don't need to lock the whole pool after implementing atomic
// frame reference count.
static void lock_buffer_pool(BufferPool *const pool) {
#if CONFIG_MULTITHREAD
  pthread_mutex_lock(&pool->pool_mutex);
#else
  (void)pool;
#endif
}

static void unlock_buffer_pool(BufferPool *const pool) {
#if CONFIG_MULTITHREAD
  pthread_mutex_unlock(&pool->pool_mutex);
#else
  (void)pool;
#endif
}

static INLINE YV12_BUFFER_CONFIG *get_ref_frame(VP10_COMMON *cm, int index) {
  if (index < 0 || index >= REF_FRAMES)
    return NULL;
  if (cm->ref_frame_map[index] < 0)
    return NULL;
  assert(cm->ref_frame_map[index] < FRAME_BUFFERS);
  return &cm->buffer_pool->frame_bufs[cm->ref_frame_map[index]].buf;
}

static INLINE YV12_BUFFER_CONFIG *get_frame_new_buffer(VP10_COMMON *cm) {
  return &cm->buffer_pool->frame_bufs[cm->new_fb_idx].buf;
}

static INLINE int get_free_fb(VP10_COMMON *cm) {
  RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
  int i;

  lock_buffer_pool(cm->buffer_pool);
  for (i = 0; i < FRAME_BUFFERS; ++i)
    if (frame_bufs[i].ref_count == 0)
      break;

  if (i != FRAME_BUFFERS) {
    frame_bufs[i].ref_count = 1;
  } else {
    // Reset i to be INVALID_IDX to indicate no free buffer found.
    i = INVALID_IDX;
  }

  unlock_buffer_pool(cm->buffer_pool);
  return i;
}

static INLINE void ref_cnt_fb(RefCntBuffer *bufs, int *idx, int new_idx) {
  const int ref_index = *idx;

  if (ref_index >= 0 && bufs[ref_index].ref_count > 0)
    bufs[ref_index].ref_count--;

  *idx = new_idx;

  bufs[new_idx].ref_count++;
}

static INLINE int mi_cols_aligned_to_sb(int n_mis) {
  return ALIGN_POWER_OF_TWO(n_mis, MI_BLOCK_SIZE_LOG2);
}

static INLINE int frame_is_intra_only(const VP10_COMMON *const cm) {
  return cm->frame_type == KEY_FRAME || cm->intra_only;
}

static INLINE void set_partition_probs(const VP10_COMMON *const cm,
                                       MACROBLOCKD *const xd) {
  xd->partition_probs =
      frame_is_intra_only(cm) ?
          &vp10_kf_partition_probs[0] :
          (const vpx_prob (*)[PARTITION_TYPES - 1])cm->fc->partition_prob;
}

static INLINE void vp10_init_macroblockd(VP10_COMMON *cm, MACROBLOCKD *xd,
                                        tran_low_t *dqcoeff) {
  int i;

  for (i = 0; i < MAX_MB_PLANE; ++i) {
    xd->plane[i].dqcoeff = dqcoeff;
    xd->above_context[i] = cm->above_context +
        i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols);

    if (xd->plane[i].plane_type == PLANE_TYPE_Y) {
      memcpy(xd->plane[i].seg_dequant, cm->y_dequant, sizeof(cm->y_dequant));
    } else {
      memcpy(xd->plane[i].seg_dequant, cm->uv_dequant, sizeof(cm->uv_dequant));
    }
    xd->fc = cm->fc;
  }

  xd->above_seg_context = cm->above_seg_context;
  xd->mi_stride = cm->mi_stride;
  xd->error_info = &cm->error;

  set_partition_probs(cm, xd);
}

static INLINE const vpx_prob* get_partition_probs(const MACROBLOCKD *xd,
                                                  int ctx) {
  return xd->partition_probs[ctx];
}

static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col) {
  const int above_idx = mi_col * 2;
  const int left_idx = (mi_row * 2) & 15;
  int i;
  for (i = 0; i < MAX_MB_PLANE; ++i) {
    struct macroblockd_plane *const pd = &xd->plane[i];
    pd->above_context = &xd->above_context[i][above_idx >> pd->subsampling_x];
    pd->left_context = &xd->left_context[i][left_idx >> pd->subsampling_y];
  }
}

static INLINE int calc_mi_size(int len) {
  // len is in mi units.
  return len + MI_BLOCK_SIZE;
}

static INLINE void set_mi_row_col(MACROBLOCKD *xd, const TileInfo *const tile,
                                  int mi_row, int bh,
                                  int mi_col, int bw,
                                  int mi_rows, int mi_cols) {
  xd->mb_to_top_edge    = -((mi_row * MI_SIZE) * 8);
  xd->mb_to_bottom_edge = ((mi_rows - bh - mi_row) * MI_SIZE) * 8;
  xd->mb_to_left_edge   = -((mi_col * MI_SIZE) * 8);
  xd->mb_to_right_edge  = ((mi_cols - bw - mi_col) * MI_SIZE) * 8;

  // Are edges available for intra prediction?
  xd->up_available    = (mi_row != 0);
  xd->left_available  = (mi_col > tile->mi_col_start);
  if (xd->up_available) {
    xd->above_mi = xd->mi[-xd->mi_stride];
    // above_mi may be NULL in VP9 encoder's first pass.
    xd->above_mbmi = xd->above_mi ? &xd->above_mi->mbmi : NULL;
  } else {
    xd->above_mi = NULL;
    xd->above_mbmi = NULL;
  }

  if (xd->left_available) {
    xd->left_mi = xd->mi[-1];
    // left_mi may be NULL in VP9 encoder's first pass.
    xd->left_mbmi = xd->left_mi ? &xd->left_mi->mbmi : NULL;
  } else {
    xd->left_mi = NULL;
    xd->left_mbmi = NULL;
  }
}

static INLINE void update_partition_context(MACROBLOCKD *xd,
                                            int mi_row, int mi_col,
                                            BLOCK_SIZE subsize,
                                            BLOCK_SIZE bsize) {
  PARTITION_CONTEXT *const above_ctx = xd->above_seg_context + mi_col;
  PARTITION_CONTEXT *const left_ctx = xd->left_seg_context + (mi_row & MI_MASK);

  // num_4x4_blocks_wide_lookup[bsize] / 2
  const int bs = num_8x8_blocks_wide_lookup[bsize];

  // update the partition context at the end notes. set partition bits
  // of block sizes larger than the current one to be one, and partition
  // bits of smaller block sizes to be zero.
  memset(above_ctx, partition_context_lookup[subsize].above, bs);
  memset(left_ctx, partition_context_lookup[subsize].left, bs);
}

static INLINE int partition_plane_context(const MACROBLOCKD *xd,
                                          int mi_row, int mi_col,
                                          BLOCK_SIZE bsize) {
  const PARTITION_CONTEXT *above_ctx = xd->above_seg_context + mi_col;
  const PARTITION_CONTEXT *left_ctx = xd->left_seg_context + (mi_row & MI_MASK);
  const int bsl = mi_width_log2_lookup[bsize];
  int above = (*above_ctx >> bsl) & 1 , left = (*left_ctx >> bsl) & 1;

  assert(b_width_log2_lookup[bsize] == b_height_log2_lookup[bsize]);
  assert(bsl >= 0);

  return (left * 2 + above) + bsl * PARTITION_PLOFFSET;
}

#ifdef __cplusplus
}  // extern "C"
#endif

#endif  // VP10_COMMON_ONYXC_INT_H_