ffmpeg/libavcodec/h264dec.h
Andreas Rheinhardt fa77cb258b avcodec/h264dec: Fix data race when updating decode_error_flags
When using multi-threaded decoding, every decoding thread
has its own DBP consisting of H264Pictures and each of these
points to its own AVFrames. They are synced during
update_thread_context via av_frame_ref() and therefore
the threads actually decoding (as well as all the others)
must not modify any field that is copied by av_frame_ref()
after ff_thread_finish_setup().

Yet this is exactly what happens when an error occurs
during decoding and the AVFrame's decode_error_flags are updated.
Given that these errors only become apparent during decoding,
this can't be set before ff_thread_finish_setup() without
defeating the point of frame-threading; in practice,
this meant that the decoder did not set these flags correctly
in case frame-threading was in use. (This means that e.g.
the ffmpeg cli tool fails to output its "corrupt decoded frame"
message in a nondeterministic fashion.)

This commit fixes this by adding a new H264Picture field
that is actually propagated across threads; the field
is an AVBufferRef* whose data is an atomic_int; it is
atomic in order to allow multiple threads to update it
concurrently and not to provide synchronization
between the threads setting the field and the thread
ultimately returning the AVFrame.

This unfortunately has the overhead of one allocation
per H264Picture (both the original one as well as
creating a reference to an existing one), even in case
of no errors. In order to mitigate this, an AVBufferPool
has been used and only if frame-threading is actually
in use. This expense will be removed as soon as
a proper API for refcounted objects (not based upon
AVBuffer) is in place.

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
2023-09-15 13:08:37 +02:00

684 lines
19 KiB
C

/*
* H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* H.264 / AVC / MPEG-4 part10 codec.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
#ifndef AVCODEC_H264DEC_H
#define AVCODEC_H264DEC_H
#include "libavutil/buffer.h"
#include "libavutil/mem_internal.h"
#include "cabac.h"
#include "error_resilience.h"
#include "h264_parse.h"
#include "h264_ps.h"
#include "h264_sei.h"
#include "h2645_parse.h"
#include "h264chroma.h"
#include "h264dsp.h"
#include "h264pred.h"
#include "h264qpel.h"
#include "h274.h"
#include "mpegutils.h"
#include "videodsp.h"
#define H264_MAX_PICTURE_COUNT 36
/* Compiling in interlaced support reduces the speed
* of progressive decoding by about 2%. */
#define ALLOW_INTERLACE
#define FMO 0
/**
* The maximum number of slices supported by the decoder.
* must be a power of 2
*/
#define MAX_SLICES 32
#ifdef ALLOW_INTERLACE
#define MB_MBAFF(h) (h)->mb_mbaff
#define MB_FIELD(sl) (sl)->mb_field_decoding_flag
#define FRAME_MBAFF(h) (h)->mb_aff_frame
#define FIELD_PICTURE(h) ((h)->picture_structure != PICT_FRAME)
#define LEFT_MBS 2
#define LTOP 0
#define LBOT 1
#define LEFT(i) (i)
#else
#define MB_MBAFF(h) 0
#define MB_FIELD(sl) 0
#define FRAME_MBAFF(h) 0
#define FIELD_PICTURE(h) 0
#undef IS_INTERLACED
#define IS_INTERLACED(mb_type) 0
#define LEFT_MBS 1
#define LTOP 0
#define LBOT 0
#define LEFT(i) 0
#endif
#define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
#ifndef CABAC
#define CABAC(h) (h)->ps.pps->cabac
#endif
#define CHROMA(h) ((h)->ps.sps->chroma_format_idc)
#define CHROMA422(h) ((h)->ps.sps->chroma_format_idc == 2)
#define CHROMA444(h) ((h)->ps.sps->chroma_format_idc == 3)
#define IS_REF0(a) ((a) & MB_TYPE_REF0)
#define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
/**
* Memory management control operation.
*/
typedef struct MMCO {
MMCOOpcode opcode;
int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
int long_arg; ///< index, pic_num, or num long refs depending on opcode
} MMCO;
typedef struct H264Picture {
AVFrame *f;
ThreadFrame tf;
AVFrame *f_grain;
AVBufferRef *qscale_table_buf;
int8_t *qscale_table;
AVBufferRef *motion_val_buf[2];
int16_t (*motion_val[2])[2];
AVBufferRef *mb_type_buf;
uint32_t *mb_type;
AVBufferRef *hwaccel_priv_buf;
void *hwaccel_picture_private; ///< hardware accelerator private data
AVBufferRef *ref_index_buf[2];
int8_t *ref_index[2];
int field_poc[2]; ///< top/bottom POC
int poc; ///< frame POC
int frame_num; ///< frame_num (raw frame_num from slice header)
int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
not mix pictures before and after MMCO_RESET. */
int pic_id; /**< pic_num (short -> no wrap version of pic_num,
pic_num & max_pic_num; long -> long_pic_num) */
int long_ref; ///< 1->long term reference 0->short term reference
int ref_poc[2][2][32]; ///< POCs of the frames/fields used as reference (FIXME need per slice)
int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
int field_picture; ///< whether or not picture was encoded in separate fields
/**
* H264Picture.reference has this flag set,
* when the picture is held for delayed output.
*/
#define DELAYED_PIC_REF (1 << 2)
int reference;
int recovered; ///< picture at IDR or recovery point + recovery count
int invalid_gap;
int sei_recovery_frame_cnt;
int needs_fg; ///< whether picture needs film grain synthesis (see `f_grain`)
AVBufferRef *pps_buf;
const PPS *pps;
int mb_width, mb_height;
int mb_stride;
/* data points to an atomic_int */
AVBufferRef *decode_error_flags;
} H264Picture;
typedef struct H264Ref {
uint8_t *data[3];
int linesize[3];
int reference;
int poc;
int pic_id;
H264Picture *parent;
} H264Ref;
typedef struct H264SliceContext {
const struct H264Context *h264;
GetBitContext gb;
ERContext *er;
int slice_num;
int slice_type;
int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
int slice_type_fixed;
int qscale;
int chroma_qp[2]; // QPc
int qp_thresh; ///< QP threshold to skip loopfilter
int last_qscale_diff;
// deblock
int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
int slice_alpha_c0_offset;
int slice_beta_offset;
H264PredWeightTable pwt;
int prev_mb_skipped;
int next_mb_skipped;
int chroma_pred_mode;
int intra16x16_pred_mode;
int8_t intra4x4_pred_mode_cache[5 * 8];
int8_t(*intra4x4_pred_mode);
int topleft_mb_xy;
int top_mb_xy;
int topright_mb_xy;
int left_mb_xy[LEFT_MBS];
int topleft_type;
int top_type;
int topright_type;
int left_type[LEFT_MBS];
const uint8_t *left_block;
int topleft_partition;
unsigned int topleft_samples_available;
unsigned int top_samples_available;
unsigned int topright_samples_available;
unsigned int left_samples_available;
ptrdiff_t linesize, uvlinesize;
ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
ptrdiff_t mb_uvlinesize;
int mb_x, mb_y;
int mb_xy;
int resync_mb_x;
int resync_mb_y;
unsigned int first_mb_addr;
// index of the first MB of the next slice
int next_slice_idx;
int mb_skip_run;
int is_complex;
int picture_structure;
int mb_field_decoding_flag;
int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
int redundant_pic_count;
/**
* number of neighbors (top and/or left) that used 8x8 dct
*/
int neighbor_transform_size;
int direct_spatial_mv_pred;
int col_parity;
int col_fieldoff;
int cbp;
int top_cbp;
int left_cbp;
int dist_scale_factor[32];
int dist_scale_factor_field[2][32];
int map_col_to_list0[2][16 + 32];
int map_col_to_list0_field[2][2][16 + 32];
/**
* num_ref_idx_l0/1_active_minus1 + 1
*/
unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
unsigned int list_count;
H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
* Reordered version of default_ref_list
* according to picture reordering in slice header */
struct {
uint8_t op;
uint32_t val;
} ref_modifications[2][32];
int nb_ref_modifications[2];
unsigned int pps_id;
const uint8_t *intra_pcm_ptr;
uint8_t *bipred_scratchpad;
uint8_t *edge_emu_buffer;
uint8_t (*top_borders[2])[(16 * 3) * 2];
int bipred_scratchpad_allocated;
int edge_emu_buffer_allocated;
int top_borders_allocated[2];
/**
* non zero coeff count cache.
* is 64 if not available.
*/
DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
/**
* Motion vector cache.
*/
DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
uint8_t direct_cache[5 * 8];
DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
///< as a DCT coefficient is int32_t in high depth, we need to reserve twice the space.
DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
///< check that i is not too large or ensure that there is some unused stuff after mb
int16_t mb_padding[256 * 2];
uint8_t (*mvd_table[2])[2];
/**
* Cabac
*/
CABACContext cabac;
uint8_t cabac_state[1024];
int cabac_init_idc;
MMCO mmco[H264_MAX_MMCO_COUNT];
int nb_mmco;
int explicit_ref_marking;
int frame_num;
int idr_pic_id;
int poc_lsb;
int delta_poc_bottom;
int delta_poc[2];
int curr_pic_num;
int max_pic_num;
} H264SliceContext;
/**
* H264Context
*/
typedef struct H264Context {
const AVClass *class;
AVCodecContext *avctx;
VideoDSPContext vdsp;
H264DSPContext h264dsp;
H264ChromaContext h264chroma;
H264QpelContext h264qpel;
H274FilmGrainDatabase h274db;
H264Picture DPB[H264_MAX_PICTURE_COUNT];
H264Picture *cur_pic_ptr;
H264Picture cur_pic;
H264Picture last_pic_for_ec;
H264SliceContext *slice_ctx;
int nb_slice_ctx;
int nb_slice_ctx_queued;
H2645Packet pkt;
int pixel_shift; ///< 0 for 8-bit H.264, 1 for high-bit-depth H.264
/* coded dimensions -- 16 * mb w/h */
int width, height;
int chroma_x_shift, chroma_y_shift;
int droppable;
int coded_picture_number;
int context_initialized;
int flags;
int workaround_bugs;
int x264_build;
/* Set when slice threading is used and at least one slice uses deblocking
* mode 1 (i.e. across slice boundaries). Then we disable the loop filter
* during normal MB decoding and execute it serially at the end.
*/
int postpone_filter;
/*
* Set to 1 when the current picture is IDR, 0 otherwise.
*/
int picture_idr;
/*
* Set to 1 when the current picture contains only I slices, 0 otherwise.
*/
int picture_intra_only;
int crop_left;
int crop_right;
int crop_top;
int crop_bottom;
int8_t(*intra4x4_pred_mode);
H264PredContext hpc;
uint8_t (*non_zero_count)[48];
#define LIST_NOT_USED -1 // FIXME rename?
/**
* block_offset[ 0..23] for frame macroblocks
* block_offset[24..47] for field macroblocks
*/
int block_offset[2 * (16 * 3)];
uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
uint32_t *mb2br_xy;
int b_stride; // FIXME use s->b4_stride
uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
// interlacing specific flags
int mb_aff_frame;
int picture_structure;
int first_field;
uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
/* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
uint16_t *cbp_table;
/* chroma_pred_mode for i4x4 or i16x16, else 0 */
uint8_t *chroma_pred_mode_table;
uint8_t (*mvd_table[2])[2];
uint8_t *direct_table;
uint8_t scan_padding[16];
uint8_t zigzag_scan[16];
uint8_t zigzag_scan8x8[64];
uint8_t zigzag_scan8x8_cavlc[64];
uint8_t field_scan[16];
uint8_t field_scan8x8[64];
uint8_t field_scan8x8_cavlc[64];
uint8_t zigzag_scan_q0[16];
uint8_t zigzag_scan8x8_q0[64];
uint8_t zigzag_scan8x8_cavlc_q0[64];
uint8_t field_scan_q0[16];
uint8_t field_scan8x8_q0[64];
uint8_t field_scan8x8_cavlc_q0[64];
int mb_y;
int mb_height, mb_width;
int mb_stride;
int mb_num;
// =============================================================
// Things below are not used in the MB or more inner code
int nal_ref_idc;
int nal_unit_type;
int has_slice; ///< slice NAL is found in the packet, set by decode_nal_units, its state does not need to be preserved outside h264_decode_frame()
/**
* Used to parse AVC variant of H.264
*/
int is_avc; ///< this flag is != 0 if codec is avc1
int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
int bit_depth_luma; ///< luma bit depth from sps to detect changes
int chroma_format_idc; ///< chroma format from sps to detect changes
H264ParamSets ps;
uint16_t *slice_table_base;
H264POCContext poc;
H264Ref default_ref[2];
H264Picture *short_ref[32];
H264Picture *long_ref[32];
H264Picture *delayed_pic[H264_MAX_DPB_FRAMES + 2]; // FIXME size?
int last_pocs[H264_MAX_DPB_FRAMES];
H264Picture *next_output_pic;
int next_outputed_poc;
int poc_offset; ///< PicOrderCnt_offset from SMPTE RDD-2006
/**
* memory management control operations buffer.
*/
MMCO mmco[H264_MAX_MMCO_COUNT];
int nb_mmco;
int mmco_reset;
int explicit_ref_marking;
int long_ref_count; ///< number of actual long term references
int short_ref_count; ///< number of actual short term references
/**
* @name Members for slice based multithreading
* @{
*/
/**
* current slice number, used to initialize slice_num of each thread/context
*/
int current_slice;
/** @} */
/**
* Complement sei_pic_struct
* SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
* However, soft telecined frames may have these values.
* This is used in an attempt to flag soft telecine progressive.
*/
int prev_interlaced_frame;
/**
* Are the SEI recovery points looking valid.
*/
int valid_recovery_point;
/**
* recovery_frame is the frame_num at which the next frame should
* be fully constructed.
*
* Set to -1 when not expecting a recovery point.
*/
int recovery_frame;
/**
* We have seen an IDR, so all the following frames in coded order are correctly
* decodable.
*/
#define FRAME_RECOVERED_IDR (1 << 0)
/**
* Sufficient number of frames have been decoded since a SEI recovery point,
* so all the following frames in presentation order are correct.
*/
#define FRAME_RECOVERED_SEI (1 << 1)
int frame_recovered; ///< Initial frame has been completely recovered
int has_recovery_point;
int missing_fields;
/* for frame threading, this is set to 1
* after finish_setup() has been called, so we cannot modify
* some context properties (which are supposed to stay constant between
* slices) anymore */
int setup_finished;
int cur_chroma_format_idc;
int cur_bit_depth_luma;
int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
/* original AVCodecContext dimensions, used to handle container
* cropping */
int width_from_caller;
int height_from_caller;
int enable_er;
ERContext er;
int16_t *dc_val_base;
H264SEIContext sei;
AVBufferPool *qscale_table_pool;
AVBufferPool *mb_type_pool;
AVBufferPool *motion_val_pool;
AVBufferPool *ref_index_pool;
AVBufferPool *decode_error_flags_pool;
int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
} H264Context;
extern const uint16_t ff_h264_mb_sizes[4];
/**
* Reconstruct bitstream slice_type.
*/
int ff_h264_get_slice_type(const H264SliceContext *sl);
/**
* Allocate tables.
* needs width/height
*/
int ff_h264_alloc_tables(H264Context *h);
int ff_h264_decode_ref_pic_list_reordering(H264SliceContext *sl, void *logctx);
int ff_h264_build_ref_list(H264Context *h, H264SliceContext *sl);
void ff_h264_remove_all_refs(H264Context *h);
/**
* Execute the reference picture marking (memory management control operations).
*/
int ff_h264_execute_ref_pic_marking(H264Context *h);
int ff_h264_decode_ref_pic_marking(H264SliceContext *sl, GetBitContext *gb,
const H2645NAL *nal, void *logctx);
void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
void ff_h264_decode_init_vlc(void);
/**
* Decode a macroblock
* @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
*/
int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
/**
* Decode a CABAC coded macroblock
* @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
*/
int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
int *mb_type);
void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
unsigned int linesize, unsigned int uvlinesize);
void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
unsigned int linesize, unsigned int uvlinesize);
/*
* o-o o-o
* / / /
* o-o o-o
* ,---'
* o-o o-o
* / / /
* o-o o-o
*/
/* Scan8 organization:
* 0 1 2 3 4 5 6 7
* 0 DY y y y y y
* 1 y Y Y Y Y
* 2 y Y Y Y Y
* 3 y Y Y Y Y
* 4 y Y Y Y Y
* 5 DU u u u u u
* 6 u U U U U
* 7 u U U U U
* 8 u U U U U
* 9 u U U U U
* 10 DV v v v v v
* 11 v V V V V
* 12 v V V V V
* 13 v V V V V
* 14 v V V V V
* DY/DU/DV are for luma/chroma DC.
*/
#define LUMA_DC_BLOCK_INDEX 48
#define CHROMA_DC_BLOCK_INDEX 49
/**
* Get the chroma qp.
*/
static av_always_inline int get_chroma_qp(const PPS *pps, int t, int qscale)
{
return pps->chroma_qp_table[t][qscale];
}
int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
int ff_h264_replace_picture(H264Context *h, H264Picture *dst, const H264Picture *src);
void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
void ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
/**
* Submit a slice for decoding.
*
* Parse the slice header, starting a new field/frame if necessary. If any
* slices are queued for the previous field, they are decoded.
*/
int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal);
int ff_h264_execute_decode_slices(H264Context *h);
int ff_h264_update_thread_context(AVCodecContext *dst,
const AVCodecContext *src);
int ff_h264_update_thread_context_for_user(AVCodecContext *dst,
const AVCodecContext *src);
void ff_h264_flush_change(H264Context *h);
void ff_h264_free_tables(H264Context *h);
void ff_h264_set_erpic(ERPicture *dst, H264Picture *src);
#endif /* AVCODEC_H264DEC_H */