vpx/vp9/common/vp9_onyxc_int.h
hkuang be6aeadaf4 Try again to merge branch 'frame-parallel' into master branch.
In frame parallel decode, libvpx decoder decodes several frames on all
cpus in parallel fashion. If not being flushed, it will only return frame
when all the cpus are busy. If getting flushed, it will return all the
frames in the decoder. Compare with current serial decode mode in which
libvpx decoder is idle between decode calls, libvpx decoder is busy
between decode calls.

Current frame parallel decode will only speed up the decoding for frame
parallel encoded videos. For non frame parallel encoded videos, frame
parallel decode is slower than serial decode due to lack of loopfilter
worker thread.

There are still some known issues that need to be addressed. For example:
decode frame parallel videos with segmentation enabled is not right sometimes.

* frame-parallel:
  Add error handling for frame parallel decode and unit test for that.
  Fix a bug in frame parallel decode and add a unit test for that.
  Add two test vectors to test frame parallel decode.
  Add key frame seeking to webmdec and webm_video_source.
  Implement frame parallel decode for VP9.
  Increase the thread test range to cover 5, 6, 7, 8 threads.
  Fix a bug in adding frame parallel unit test.
  Add VP9 frame-parallel unit test.
  Manually pick "Make the api behavior conform to api spec." from master branch.
  Move vp9_dec_build_inter_predictors_* to decoder folder.
  Add segmentation map array for current and last frame segmentation.
  Include the right header for VP9 worker thread.
  Move vp9_thread.* to common.
  ctrl_get_reference does not need user_priv.
  Seperate the frame buffers from VP9 encoder/decoder structure.
  Revert "Revert "Revert "Revert 3 patches from Hangyu to get Chrome to build:"""
 Conflicts:
       test/codec_factory.h
       test/decode_test_driver.cc
       test/decode_test_driver.h
       test/invalid_file_test.cc
       test/test-data.sha1
       test/test.mk
       test/test_vectors.cc
       vp8/vp8_dx_iface.c
       vp9/common/vp9_alloccommon.c
       vp9/common/vp9_entropymode.c
       vp9/common/vp9_loopfilter_thread.c
       vp9/common/vp9_loopfilter_thread.h
       vp9/common/vp9_mvref_common.c
       vp9/common/vp9_onyxc_int.h
       vp9/common/vp9_reconinter.c
       vp9/decoder/vp9_decodeframe.c
       vp9/decoder/vp9_decodeframe.h
       vp9/decoder/vp9_decodemv.c
       vp9/decoder/vp9_decoder.c
       vp9/decoder/vp9_decoder.h
       vp9/encoder/vp9_encoder.c
       vp9/encoder/vp9_pickmode.c
       vp9/encoder/vp9_rdopt.c
       vp9/vp9_cx_iface.c
       vp9/vp9_dx_iface.c

This reverts commit a18da9760a.

Change-Id: I361442ffec1586d036ea2e0ee97ce4f077585f02
2015-01-30 21:00:13 -08:00

440 lines
13 KiB
C

/*
* 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 VP9_COMMON_VP9_ONYXC_INT_H_
#define VP9_COMMON_VP9_ONYXC_INT_H_
#include "./vpx_config.h"
#include "vpx/internal/vpx_codec_internal.h"
#include "./vp9_rtcd.h"
#include "vp9/common/vp9_loopfilter.h"
#include "vp9/common/vp9_entropymv.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_frame_buffers.h"
#include "vp9/common/vp9_quant_common.h"
#include "vp9/common/vp9_thread.h"
#include "vp9/common/vp9_tile_common.h"
#if CONFIG_VP9_POSTPROC
#include "vp9/common/vp9_postproc.h"
#endif
#ifdef __cplusplus
extern "C" {
#endif
#define REFS_PER_FRAME 3
#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
extern const struct {
PARTITION_CONTEXT above;
PARTITION_CONTEXT left;
} partition_context_lookup[BLOCK_SIZES];
typedef enum {
SINGLE_REFERENCE = 0,
COMPOUND_REFERENCE = 1,
REFERENCE_MODE_SELECT = 2,
REFERENCE_MODES = 3,
} REFERENCE_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.
VP9Worker *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 {
// 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 VP9Common {
struct vpx_internal_error_info error;
DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]);
vpx_color_space_t color_space;
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.
int intra_only;
int allow_high_precision_mv;
// Flag signaling that the frame context should be reset to default values.
// 0 or 1 implies don't reset, 2 reset just the context specified in the
// frame header, 3 reset all contexts.
int 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;
/* 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 VP9Common *cm, int mi_size);
void (*free_mi)(struct VP9Common *cm);
void (*setup_mi)(struct VP9Common *cm);
// 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;
INTERP_FILTER interp_filter;
loop_filter_info_n lf_info;
int refresh_frame_context; /* Two state 0 = NO, 1 = YES */
int ref_frame_sign_bias[MAX_REF_FRAMES]; /* Two state 0, 1 */
struct loopfilter lf;
struct segmentation seg;
// TODO(hkuang): Remove this as it is the same as frame_parallel_decode
// in pbi.
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 frame_parallel_decoding_mode;
int log2_tile_cols, log2_tile_rows;
int byte_alignment;
// 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;
} VP9_COMMON;
// TODO(hkuang): Don't need to lock the whole pool after implementing atomic
// frame reference count.
void lock_buffer_pool(BufferPool *const pool);
void unlock_buffer_pool(BufferPool *const pool);
static INLINE YV12_BUFFER_CONFIG *get_ref_frame(VP9_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(VP9_COMMON *cm) {
return &cm->buffer_pool->frame_bufs[cm->new_fb_idx].buf;
}
static INLINE int get_free_fb(VP9_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;
assert(i < FRAME_BUFFERS);
frame_bufs[i].ref_count = 1;
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 void init_macroblockd(VP9_COMMON *cm, MACROBLOCKD *xd) {
int i;
for (i = 0; i < MAX_MB_PLANE; ++i) {
xd->plane[i].dqcoeff = xd->dqcoeff;
xd->above_context[i] = cm->above_context +
i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols);
}
xd->above_seg_context = cm->above_seg_context;
xd->mi_stride = cm->mi_stride;
xd->error_info = &cm->error;
}
static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) {
return cm->frame_type == KEY_FRAME || cm->intra_only;
}
static INLINE const vp9_prob* get_partition_probs(const VP9_COMMON *cm,
int ctx) {
return frame_is_intra_only(cm) ? vp9_kf_partition_probs[ctx]
: cm->fc->partition_prob[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].src_mi;
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].src_mi;
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.
vpx_memset(above_ctx, partition_context_lookup[subsize].above, bs);
vpx_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];
const int bs = 1 << bsl;
int above = 0, left = 0, i;
assert(b_width_log2_lookup[bsize] == b_height_log2_lookup[bsize]);
assert(bsl >= 0);
for (i = 0; i < bs; i++) {
above |= above_ctx[i];
left |= left_ctx[i];
}
above = (above & bs) > 0;
left = (left & bs) > 0;
return (left * 2 + above) + bsl * PARTITION_PLOFFSET;
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_ONYXC_INT_H_