1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-11-21 10:55:51 +02:00
FFmpeg/libavcodec/ffv1dec.c

1107 lines
41 KiB
C
Raw Normal View History

/*
* FFV1 decoder
*
* Copyright (c) 2003-2013 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
* FF Video Codec 1 (a lossless codec) decoder
*/
#include "libavutil/avassert.h"
#include "libavutil/crc.h"
#include "libavutil/mem.h"
#include "libavutil/imgutils.h"
#include "libavutil/pixdesc.h"
#include "avcodec.h"
#include "codec_internal.h"
#include "get_bits.h"
#include "rangecoder.h"
#include "golomb.h"
#include "mathops.h"
#include "ffv1.h"
#include "progressframe.h"
lavc/ffv1: change FFV1SliceContext.plane into a RefStruct object Frame threading in the FFV1 decoder works in a very unusual way - the state that needs to be propagated from the previous frame is not decoded pixels(¹), but each slice's entropy coder state after decoding the slice. For that purpose, the decoder's update_thread_context() callback stores a pointer to the previous frame thread's private data. Then, when decoding each slice, the frame thread uses the standard progress mechanism to wait for the corresponding slice in the previous frame to be completed, then copies the entropy coder state from the previously-stored pointer. This approach is highly dubious, as update_thread_context() should be the only point where frame-thread contexts come into direct contact. There are no guarantees that the stored pointer will be valid at all, or will contain any particular data after update_thread_context() finishes. More specifically, this code can break due to the fact that keyframes reset entropy coder state and thus do not need to wait for the previous frame. As an example, consider a decoder process with 2 frame threads - thread 0 with its context 0, and thread 1 with context 1 - decoding a previous frame P, current frame F, followed by a keyframe K. Then consider concurrent execution consistent with the following sequence of events: * thread 0 starts decoding P * thread 0 reads P's slice header, then calls ff_thread_finish_setup() allowing next frame thread to start * main thread calls update_thread_context() to transfer state from context 0 to context 1; context 1 stores a pointer to context 0's private data * thread 1 starts decoding F * thread 1 reads F's slice header, then calls ff_thread_finish_setup() allowing the next frame thread to start decoding * thread 0 finishes decoding P * thread 0 starts decoding K; since K is a keyframe, it does not wait for F and reallocates the arrays holding entropy coder state * thread 0 finishes decoding K * thread 1 reads entropy coder state from its stored pointer to context 0, however it finds state from K rather than from P This execution is currently prevented by special-casing FFV1 in the generic frame threading code, however that is supremely ugly. It also involves unnecessary copies of the state arrays, when in fact they can only be used by one thread at a time. This commit addresses these deficiencies by changing the array of PlaneContext (each of which contains the allocated state arrays) embedded in FFV1SliceContext into a RefStruct object. This object can then be propagated across frame threads in standard manner. Since the code structure guarantees only one thread accesses it at a time, no copies are necessary. It is also re-created for keyframes, solving the above issue cleanly. Special-casing of FFV1 in the generic frame threading code will be removed in a later commit. (¹) except in the case of a damaged slice, when previous frame's pixels are used directly
2024-07-11 11:08:55 +02:00
#include "refstruct.h"
#include "thread.h"
static inline av_flatten int get_symbol_inline(RangeCoder *c, uint8_t *state,
int is_signed)
{
if (get_rac(c, state + 0))
return 0;
else {
int e;
unsigned a;
e = 0;
while (get_rac(c, state + 1 + FFMIN(e, 9))) { // 1..10
e++;
if (e > 31)
return AVERROR_INVALIDDATA;
}
a = 1;
for (int i = e - 1; i >= 0; i--)
a += a + get_rac(c, state + 22 + FFMIN(i, 9)); // 22..31
e = -(is_signed && get_rac(c, state + 11 + FFMIN(e, 10))); // 11..21
return (a ^ e) - e;
}
}
static av_noinline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed)
{
return get_symbol_inline(c, state, is_signed);
}
static inline int get_vlc_symbol(GetBitContext *gb, VlcState *const state,
int bits)
{
int k, i, v, ret;
i = state->count;
k = 0;
while (i < state->error_sum) { // FIXME: optimize
k++;
i += i;
}
v = get_sr_golomb(gb, k, 12, bits);
ff_dlog(NULL, "v:%d bias:%d error:%d drift:%d count:%d k:%d",
v, state->bias, state->error_sum, state->drift, state->count, k);
v ^= ((2 * state->drift + state->count) >> 31);
ret = fold(v + state->bias, bits);
update_vlc_state(state, v);
return ret;
}
static int is_input_end(RangeCoder *c, GetBitContext *gb, int ac)
{
if (ac != AC_GOLOMB_RICE) {
if (c->overread > MAX_OVERREAD)
return AVERROR_INVALIDDATA;
} else {
if (get_bits_left(gb) < 1)
return AVERROR_INVALIDDATA;
}
return 0;
}
#define TYPE int16_t
#define RENAME(name) name
#include "ffv1dec_template.c"
#undef TYPE
#undef RENAME
#define TYPE int32_t
#define RENAME(name) name ## 32
#include "ffv1dec_template.c"
static int decode_plane(FFV1Context *f, FFV1SliceContext *sc,
GetBitContext *gb,
uint8_t *src, int w, int h, int stride, int plane_index,
int pixel_stride, int ac)
{
int x, y;
int16_t *sample[2];
sample[0] = sc->sample_buffer + 3;
sample[1] = sc->sample_buffer + w + 6 + 3;
sc->run_index = 0;
memset(sc->sample_buffer, 0, 2 * (w + 6) * sizeof(*sc->sample_buffer));
for (y = 0; y < h; y++) {
int16_t *temp = sample[0]; // FIXME: try a normal buffer
sample[0] = sample[1];
sample[1] = temp;
sample[1][-1] = sample[0][0];
sample[0][w] = sample[0][w - 1];
if (f->avctx->bits_per_raw_sample <= 8) {
int ret = decode_line(f, sc, gb, w, sample, plane_index, 8, ac);
if (ret < 0)
return ret;
for (x = 0; x < w; x++)
src[x*pixel_stride + stride * y] = sample[1][x];
} else {
int ret = decode_line(f, sc, gb, w, sample, plane_index, f->avctx->bits_per_raw_sample, ac);
if (ret < 0)
return ret;
if (f->packed_at_lsb) {
for (x = 0; x < w; x++) {
((uint16_t*)(src + stride*y))[x*pixel_stride] = sample[1][x];
}
} else {
for (x = 0; x < w; x++) {
((uint16_t*)(src + stride*y))[x*pixel_stride] = sample[1][x] << (16 - f->avctx->bits_per_raw_sample) | ((uint16_t **)sample)[1][x] >> (2 * f->avctx->bits_per_raw_sample - 16);
}
}
}
}
return 0;
}
static int decode_slice_header(const FFV1Context *f,
FFV1SliceContext *sc, AVFrame *frame)
{
RangeCoder *c = &sc->c;
uint8_t state[CONTEXT_SIZE];
unsigned ps, context_count;
int sx, sy, sw, sh;
memset(state, 128, sizeof(state));
sx = get_symbol(c, state, 0);
sy = get_symbol(c, state, 0);
sw = get_symbol(c, state, 0) + 1U;
sh = get_symbol(c, state, 0) + 1U;
av_assert0(f->version > 2);
if (sx < 0 || sy < 0 || sw <= 0 || sh <= 0)
return AVERROR_INVALIDDATA;
if (sx > f->num_h_slices - sw || sy > f->num_v_slices - sh)
return AVERROR_INVALIDDATA;
sc->slice_x = ff_slice_coord(f, f->width , sx , f->num_h_slices, f->chroma_h_shift);
sc->slice_y = ff_slice_coord(f, f->height, sy , f->num_v_slices, f->chroma_v_shift);
sc->slice_width = ff_slice_coord(f, f->width , sx + sw, f->num_h_slices, f->chroma_h_shift) - sc->slice_x;
sc->slice_height = ff_slice_coord(f, f->height, sy + sh, f->num_v_slices, f->chroma_v_shift) - sc->slice_y;
av_assert0((unsigned)sc->slice_width <= f->width &&
(unsigned)sc->slice_height <= f->height);
av_assert0 ( (unsigned)sc->slice_x + (uint64_t)sc->slice_width <= f->width
&& (unsigned)sc->slice_y + (uint64_t)sc->slice_height <= f->height);
if (f->ac == AC_GOLOMB_RICE && sc->slice_width >= (1<<23))
return AVERROR_INVALIDDATA;
for (unsigned i = 0; i < f->plane_count; i++) {
PlaneContext * const p = &sc->plane[i];
int idx = get_symbol(c, state, 0);
if (idx >= (unsigned)f->quant_table_count) {
av_log(f->avctx, AV_LOG_ERROR, "quant_table_index out of range\n");
return -1;
}
p->quant_table_index = idx;
context_count = f->context_count[idx];
if (p->context_count < context_count) {
av_freep(&p->state);
av_freep(&p->vlc_state);
}
p->context_count = context_count;
}
ps = get_symbol(c, state, 0);
if (ps == 1) {
frame->flags |= AV_FRAME_FLAG_INTERLACED;
frame->flags |= AV_FRAME_FLAG_TOP_FIELD_FIRST;
} else if (ps == 2) {
frame->flags |= AV_FRAME_FLAG_INTERLACED;
frame->flags &= ~AV_FRAME_FLAG_TOP_FIELD_FIRST;
} else if (ps == 3) {
frame->flags &= ~AV_FRAME_FLAG_INTERLACED;
}
frame->sample_aspect_ratio.num = get_symbol(c, state, 0);
frame->sample_aspect_ratio.den = get_symbol(c, state, 0);
if (av_image_check_sar(f->width, f->height,
frame->sample_aspect_ratio) < 0) {
av_log(f->avctx, AV_LOG_WARNING, "ignoring invalid SAR: %u/%u\n",
frame->sample_aspect_ratio.num,
frame->sample_aspect_ratio.den);
frame->sample_aspect_ratio = (AVRational){ 0, 1 };
}
if (f->version > 3) {
sc->slice_reset_contexts = get_rac(c, state);
sc->slice_coding_mode = get_symbol(c, state, 0);
if (sc->slice_coding_mode != 1 && f->colorspace == 1) {
sc->slice_rct_by_coef = get_symbol(c, state, 0);
sc->slice_rct_ry_coef = get_symbol(c, state, 0);
if ((uint64_t)sc->slice_rct_by_coef + (uint64_t)sc->slice_rct_ry_coef > 4) {
av_log(f->avctx, AV_LOG_ERROR, "slice_rct_y_coef out of range\n");
return AVERROR_INVALIDDATA;
}
}
}
return 0;
}
static void slice_set_damaged(FFV1Context *f, FFV1SliceContext *sc)
{
sc->slice_damaged = 1;
// only set this for frame threading, as for slice threading its value is
// not used and setting it would be a race
if (f->avctx->active_thread_type & FF_THREAD_FRAME)
f->frame_damaged = 1;
}
static int decode_slice(AVCodecContext *c, void *arg)
{
FFV1Context *f = c->priv_data;
FFV1SliceContext *sc = arg;
int width, height, x, y, ret;
const int ps = av_pix_fmt_desc_get(c->pix_fmt)->comp[0].step;
AVFrame * const p = f->picture.f;
const int si = sc - f->slices;
GetBitContext gb;
int ac = f->ac || sc->slice_coding_mode == 1;
if (!(p->flags & AV_FRAME_FLAG_KEY) && f->last_picture.f)
ff_progress_frame_await(&f->last_picture, si);
if (f->slice_damaged[si])
slice_set_damaged(f, sc);
sc->slice_rct_by_coef = 1;
sc->slice_rct_ry_coef = 1;
if (f->version > 2) {
if (ff_ffv1_init_slice_state(f, sc) < 0)
return AVERROR(ENOMEM);
if (decode_slice_header(f, sc, p) < 0) {
sc->slice_x = sc->slice_y = sc->slice_height = sc->slice_width = 0;
slice_set_damaged(f, sc);
return AVERROR_INVALIDDATA;
}
}
if ((ret = ff_ffv1_init_slice_state(f, sc)) < 0)
return ret;
if ((p->flags & AV_FRAME_FLAG_KEY) || sc->slice_reset_contexts) {
ff_ffv1_clear_slice_state(f, sc);
} else if (sc->slice_damaged) {
return AVERROR_INVALIDDATA;
}
width = sc->slice_width;
height = sc->slice_height;
x = sc->slice_x;
y = sc->slice_y;
if (ac == AC_GOLOMB_RICE) {
if (f->version == 3 && f->micro_version > 1 || f->version > 3)
get_rac(&sc->c, (uint8_t[]) { 129 });
sc->ac_byte_count = f->version > 2 || (!x && !y) ? sc->c.bytestream - sc->c.bytestream_start - 1 : 0;
init_get_bits(&gb,
sc->c.bytestream_start + sc->ac_byte_count,
(sc->c.bytestream_end - sc->c.bytestream_start - sc->ac_byte_count) * 8);
}
av_assert1(width && height);
if (f->colorspace == 0 && (f->chroma_planes || !f->transparency)) {
const int chroma_width = AV_CEIL_RSHIFT(width, f->chroma_h_shift);
const int chroma_height = AV_CEIL_RSHIFT(height, f->chroma_v_shift);
const int cx = x >> f->chroma_h_shift;
const int cy = y >> f->chroma_v_shift;
decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1, ac);
if (f->chroma_planes) {
decode_plane(f, sc, &gb, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1, ac);
decode_plane(f, sc, &gb, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 1, ac);
}
if (f->transparency)
decode_plane(f, sc, &gb, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], (f->version >= 4 && !f->chroma_planes) ? 1 : 2, 1, ac);
} else if (f->colorspace == 0) {
decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0] , width, height, p->linesize[0], 0, 2, ac);
decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0] + 1, width, height, p->linesize[0], 1, 2, ac);
} else if (f->use32bit) {
uint8_t *planes[4] = { p->data[0] + ps * x + y * p->linesize[0],
p->data[1] + ps * x + y * p->linesize[1],
p->data[2] + ps * x + y * p->linesize[2],
p->data[3] + ps * x + y * p->linesize[3] };
decode_rgb_frame32(f, sc, &gb, planes, width, height, p->linesize);
} else {
uint8_t *planes[4] = { p->data[0] + ps * x + y * p->linesize[0],
p->data[1] + ps * x + y * p->linesize[1],
p->data[2] + ps * x + y * p->linesize[2],
p->data[3] + ps * x + y * p->linesize[3] };
decode_rgb_frame(f, sc, &gb, planes, width, height, p->linesize);
}
if (ac != AC_GOLOMB_RICE && f->version > 2) {
int v;
get_rac(&sc->c, (uint8_t[]) { 129 });
v = sc->c.bytestream_end - sc->c.bytestream - 2 - 5*!!f->ec;
if (v) {
av_log(f->avctx, AV_LOG_ERROR, "bytestream end mismatching by %d\n", v);
slice_set_damaged(f, sc);
}
}
if (sc->slice_damaged && (f->avctx->err_recognition & AV_EF_EXPLODE))
return AVERROR_INVALIDDATA;
if ((c->active_thread_type & FF_THREAD_FRAME) && !f->frame_damaged)
ff_progress_frame_report(&f->picture, si);
return 0;
}
static int read_quant_table(RangeCoder *c, int16_t *quant_table, int scale)
{
int v;
int i = 0;
uint8_t state[CONTEXT_SIZE];
memset(state, 128, sizeof(state));
for (v = 0; i < 128; v++) {
unsigned len = get_symbol(c, state, 0) + 1U;
if (len > 128 - i || !len)
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
while (len--) {
quant_table[i] = scale * v;
i++;
}
}
for (i = 1; i < 128; i++)
quant_table[256 - i] = -quant_table[i];
quant_table[128] = -quant_table[127];
return 2 * v - 1;
}
static int read_quant_tables(RangeCoder *c,
int16_t quant_table[MAX_CONTEXT_INPUTS][256])
{
int i;
int context_count = 1;
for (i = 0; i < 5; i++) {
int ret = read_quant_table(c, quant_table[i], context_count);
if (ret < 0)
return ret;
context_count *= ret;
if (context_count > 32768U) {
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
}
}
return (context_count + 1) / 2;
}
static int read_extra_header(FFV1Context *f)
{
RangeCoder c;
uint8_t state[CONTEXT_SIZE];
int ret;
uint8_t state2[32][CONTEXT_SIZE];
unsigned crc = 0;
memset(state2, 128, sizeof(state2));
memset(state, 128, sizeof(state));
ff_init_range_decoder(&c, f->avctx->extradata, f->avctx->extradata_size);
ff_build_rac_states(&c, 0.05 * (1LL << 32), 256 - 8);
f->version = get_symbol(&c, state, 0);
if (f->version < 2) {
av_log(f->avctx, AV_LOG_ERROR, "Invalid version in global header\n");
return AVERROR_INVALIDDATA;
}
if (f->version > 4) {
av_log(f->avctx, AV_LOG_ERROR, "unsupported version %d\n",
f->version);
return AVERROR_PATCHWELCOME;
}
if (f->version > 2) {
c.bytestream_end -= 4;
f->micro_version = get_symbol(&c, state, 0);
if (f->micro_version < 0)
return AVERROR_INVALIDDATA;
}
f->ac = get_symbol(&c, state, 0);
if (f->ac == AC_RANGE_CUSTOM_TAB) {
for (int i = 1; i < 256; i++)
f->state_transition[i] = get_symbol(&c, state, 1) + c.one_state[i];
}
f->colorspace = get_symbol(&c, state, 0); //YUV cs type
f->avctx->bits_per_raw_sample = get_symbol(&c, state, 0);
f->chroma_planes = get_rac(&c, state);
f->chroma_h_shift = get_symbol(&c, state, 0);
f->chroma_v_shift = get_symbol(&c, state, 0);
f->transparency = get_rac(&c, state);
f->plane_count = 1 + (f->chroma_planes || f->version<4) + f->transparency;
f->num_h_slices = 1 + get_symbol(&c, state, 0);
f->num_v_slices = 1 + get_symbol(&c, state, 0);
if (f->chroma_h_shift > 4U || f->chroma_v_shift > 4U) {
av_log(f->avctx, AV_LOG_ERROR, "chroma shift parameters %d %d are invalid\n",
f->chroma_h_shift, f->chroma_v_shift);
return AVERROR_INVALIDDATA;
}
if (f->num_h_slices > (unsigned)f->width || !f->num_h_slices ||
f->num_v_slices > (unsigned)f->height || !f->num_v_slices
) {
av_log(f->avctx, AV_LOG_ERROR, "slice count invalid\n");
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
}
if (f->num_h_slices > MAX_SLICES / f->num_v_slices) {
av_log(f->avctx, AV_LOG_ERROR, "slice count unsupported\n");
return AVERROR_PATCHWELCOME;
}
f->quant_table_count = get_symbol(&c, state, 0);
if (f->quant_table_count > (unsigned)MAX_QUANT_TABLES || !f->quant_table_count) {
av_log(f->avctx, AV_LOG_ERROR, "quant table count %d is invalid\n", f->quant_table_count);
f->quant_table_count = 0;
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
}
for (int i = 0; i < f->quant_table_count; i++) {
f->context_count[i] = read_quant_tables(&c, f->quant_tables[i]);
if (f->context_count[i] < 0) {
av_log(f->avctx, AV_LOG_ERROR, "read_quant_table error\n");
return AVERROR_INVALIDDATA;
}
}
2015-08-22 03:14:12 +02:00
if ((ret = ff_ffv1_allocate_initial_states(f)) < 0)
return ret;
for (int i = 0; i < f->quant_table_count; i++)
if (get_rac(&c, state)) {
for (int j = 0; j < f->context_count[i]; j++)
for (int k = 0; k < CONTEXT_SIZE; k++) {
int pred = j ? f->initial_states[i][j - 1][k] : 128;
f->initial_states[i][j][k] =
(pred + get_symbol(&c, state2[k], 1)) & 0xFF;
}
}
if (f->version > 2) {
f->ec = get_symbol(&c, state, 0);
if (f->ec >= 2)
f->crcref = 0x7a8c4079;
if (f->micro_version > 2)
f->intra = get_symbol(&c, state, 0);
}
if (f->version > 2) {
unsigned v;
v = av_crc(av_crc_get_table(AV_CRC_32_IEEE), f->crcref,
f->avctx->extradata, f->avctx->extradata_size);
if (v != f->crcref || f->avctx->extradata_size < 4) {
av_log(f->avctx, AV_LOG_ERROR, "CRC mismatch %X!\n", v);
return AVERROR_INVALIDDATA;
}
crc = AV_RB32(f->avctx->extradata + f->avctx->extradata_size - 4);
}
if (f->avctx->debug & FF_DEBUG_PICT_INFO)
av_log(f->avctx, AV_LOG_DEBUG,
"global: ver:%d.%d, coder:%d, colorspace: %d bpr:%d chroma:%d(%d:%d), alpha:%d slices:%dx%d qtabs:%d ec:%d intra:%d CRC:0x%08X\n",
f->version, f->micro_version,
f->ac,
f->colorspace,
f->avctx->bits_per_raw_sample,
f->chroma_planes, f->chroma_h_shift, f->chroma_v_shift,
f->transparency,
f->num_h_slices, f->num_v_slices,
f->quant_table_count,
f->ec,
f->intra,
crc
);
return 0;
}
static int read_header(FFV1Context *f)
{
uint8_t state[CONTEXT_SIZE];
int context_count = -1; //-1 to avoid warning
RangeCoder *const c = &f->slices[0].c;
memset(state, 128, sizeof(state));
if (f->version < 2) {
int chroma_planes, chroma_h_shift, chroma_v_shift, transparency, colorspace, bits_per_raw_sample;
unsigned v= get_symbol(c, state, 0);
if (v >= 2) {
av_log(f->avctx, AV_LOG_ERROR, "invalid version %d in ver01 header\n", v);
return AVERROR_INVALIDDATA;
}
f->version = v;
f->ac = get_symbol(c, state, 0);
if (f->ac == AC_RANGE_CUSTOM_TAB) {
for (int i = 1; i < 256; i++) {
int st = get_symbol(c, state, 1) + c->one_state[i];
if (st < 1 || st > 255) {
av_log(f->avctx, AV_LOG_ERROR, "invalid state transition %d\n", st);
return AVERROR_INVALIDDATA;
}
f->state_transition[i] = st;
}
}
colorspace = get_symbol(c, state, 0); //YUV cs type
bits_per_raw_sample = f->version > 0 ? get_symbol(c, state, 0) : f->avctx->bits_per_raw_sample;
chroma_planes = get_rac(c, state);
chroma_h_shift = get_symbol(c, state, 0);
chroma_v_shift = get_symbol(c, state, 0);
transparency = get_rac(c, state);
if (colorspace == 0 && f->avctx->skip_alpha)
transparency = 0;
if (f->plane_count) {
if (colorspace != f->colorspace ||
bits_per_raw_sample != f->avctx->bits_per_raw_sample ||
chroma_planes != f->chroma_planes ||
chroma_h_shift != f->chroma_h_shift ||
chroma_v_shift != f->chroma_v_shift ||
transparency != f->transparency) {
av_log(f->avctx, AV_LOG_ERROR, "Invalid change of global parameters\n");
return AVERROR_INVALIDDATA;
}
}
if (chroma_h_shift > 4U || chroma_v_shift > 4U) {
av_log(f->avctx, AV_LOG_ERROR, "chroma shift parameters %d %d are invalid\n",
chroma_h_shift, chroma_v_shift);
return AVERROR_INVALIDDATA;
}
f->colorspace = colorspace;
f->avctx->bits_per_raw_sample = bits_per_raw_sample;
f->chroma_planes = chroma_planes;
f->chroma_h_shift = chroma_h_shift;
f->chroma_v_shift = chroma_v_shift;
f->transparency = transparency;
f->plane_count = 2 + f->transparency;
}
if (f->colorspace == 0) {
if (!f->transparency && !f->chroma_planes) {
if (f->avctx->bits_per_raw_sample <= 8)
f->avctx->pix_fmt = AV_PIX_FMT_GRAY8;
else if (f->avctx->bits_per_raw_sample == 9) {
f->packed_at_lsb = 1;
f->avctx->pix_fmt = AV_PIX_FMT_GRAY9;
} else if (f->avctx->bits_per_raw_sample == 10) {
f->packed_at_lsb = 1;
f->avctx->pix_fmt = AV_PIX_FMT_GRAY10;
} else if (f->avctx->bits_per_raw_sample == 12) {
f->packed_at_lsb = 1;
f->avctx->pix_fmt = AV_PIX_FMT_GRAY12;
} else if (f->avctx->bits_per_raw_sample == 14) {
f->packed_at_lsb = 1;
f->avctx->pix_fmt = AV_PIX_FMT_GRAY14;
} else if (f->avctx->bits_per_raw_sample == 16) {
f->packed_at_lsb = 1;
f->avctx->pix_fmt = AV_PIX_FMT_GRAY16;
} else if (f->avctx->bits_per_raw_sample < 16) {
f->avctx->pix_fmt = AV_PIX_FMT_GRAY16;
} else
return AVERROR(ENOSYS);
} else if (f->transparency && !f->chroma_planes) {
if (f->avctx->bits_per_raw_sample <= 8)
f->avctx->pix_fmt = AV_PIX_FMT_YA8;
else
return AVERROR(ENOSYS);
} else if (f->avctx->bits_per_raw_sample<=8 && !f->transparency) {
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUV444P; break;
case 0x01: f->avctx->pix_fmt = AV_PIX_FMT_YUV440P; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUV422P; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUV420P; break;
case 0x20: f->avctx->pix_fmt = AV_PIX_FMT_YUV411P; break;
case 0x22: f->avctx->pix_fmt = AV_PIX_FMT_YUV410P; break;
}
} else if (f->avctx->bits_per_raw_sample <= 8 && f->transparency) {
switch(16*f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUVA444P; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUVA422P; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUVA420P; break;
}
} else if (f->avctx->bits_per_raw_sample == 9 && !f->transparency) {
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUV444P9; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUV422P9; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUV420P9; break;
}
} else if (f->avctx->bits_per_raw_sample == 9 && f->transparency) {
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUVA444P9; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUVA422P9; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUVA420P9; break;
}
} else if (f->avctx->bits_per_raw_sample == 10 && !f->transparency) {
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUV444P10; break;
case 0x01: f->avctx->pix_fmt = AV_PIX_FMT_YUV440P10; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUV422P10; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUV420P10; break;
}
} else if (f->avctx->bits_per_raw_sample == 10 && f->transparency) {
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUVA444P10; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUVA422P10; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUVA420P10; break;
}
} else if (f->avctx->bits_per_raw_sample == 12 && !f->transparency) {
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUV444P12; break;
case 0x01: f->avctx->pix_fmt = AV_PIX_FMT_YUV440P12; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUV422P12; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUV420P12; break;
}
} else if (f->avctx->bits_per_raw_sample == 12 && f->transparency) {
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUVA444P12; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUVA422P12; break;
}
} else if (f->avctx->bits_per_raw_sample == 14 && !f->transparency) {
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUV444P14; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUV422P14; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUV420P14; break;
}
} else if (f->avctx->bits_per_raw_sample == 16 && !f->transparency){
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUV444P16; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUV422P16; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUV420P16; break;
}
} else if (f->avctx->bits_per_raw_sample == 16 && f->transparency){
f->packed_at_lsb = 1;
switch(16 * f->chroma_h_shift + f->chroma_v_shift) {
case 0x00: f->avctx->pix_fmt = AV_PIX_FMT_YUVA444P16; break;
case 0x10: f->avctx->pix_fmt = AV_PIX_FMT_YUVA422P16; break;
case 0x11: f->avctx->pix_fmt = AV_PIX_FMT_YUVA420P16; break;
}
}
} else if (f->colorspace == 1) {
if (f->chroma_h_shift || f->chroma_v_shift) {
av_log(f->avctx, AV_LOG_ERROR,
"chroma subsampling not supported in this colorspace\n");
2012-10-19 11:33:43 +03:00
return AVERROR(ENOSYS);
}
if ( f->avctx->bits_per_raw_sample <= 8 && !f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_0RGB32;
else if (f->avctx->bits_per_raw_sample <= 8 && f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_RGB32;
else if (f->avctx->bits_per_raw_sample == 9 && !f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_GBRP9;
else if (f->avctx->bits_per_raw_sample == 10 && !f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_GBRP10;
else if (f->avctx->bits_per_raw_sample == 10 && f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_GBRAP10;
else if (f->avctx->bits_per_raw_sample == 12 && !f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_GBRP12;
else if (f->avctx->bits_per_raw_sample == 12 && f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_GBRAP12;
else if (f->avctx->bits_per_raw_sample == 14 && !f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_GBRP14;
else if (f->avctx->bits_per_raw_sample == 14 && f->transparency)
f->avctx->pix_fmt = AV_PIX_FMT_GBRAP14;
else if (f->avctx->bits_per_raw_sample == 16 && !f->transparency) {
f->avctx->pix_fmt = AV_PIX_FMT_GBRP16;
f->use32bit = 1;
}
else if (f->avctx->bits_per_raw_sample == 16 && f->transparency) {
f->avctx->pix_fmt = AV_PIX_FMT_GBRAP16;
f->use32bit = 1;
}
} else {
av_log(f->avctx, AV_LOG_ERROR, "colorspace not supported\n");
2012-10-19 11:33:43 +03:00
return AVERROR(ENOSYS);
}
if (f->avctx->pix_fmt == AV_PIX_FMT_NONE) {
av_log(f->avctx, AV_LOG_ERROR, "format not supported\n");
return AVERROR(ENOSYS);
}
ff_dlog(f->avctx, "%d %d %d\n",
f->chroma_h_shift, f->chroma_v_shift, f->avctx->pix_fmt);
if (f->version < 2) {
context_count = read_quant_tables(c, f->quant_tables[0]);
if (context_count < 0) {
av_log(f->avctx, AV_LOG_ERROR, "read_quant_table error\n");
return AVERROR_INVALIDDATA;
}
f->slice_count = f->max_slice_count;
} else if (f->version < 3) {
f->slice_count = get_symbol(c, state, 0);
} else {
const uint8_t *p = c->bytestream_end;
for (f->slice_count = 0;
f->slice_count < MAX_SLICES && 3 + 5*!!f->ec < p - c->bytestream_start;
f->slice_count++) {
int trailer = 3 + 5*!!f->ec;
int size = AV_RB24(p-trailer);
if (size + trailer > p - c->bytestream_start)
break;
p -= size + trailer;
}
}
if (f->slice_count > (unsigned)MAX_SLICES || f->slice_count <= 0 || f->slice_count > f->max_slice_count) {
av_log(f->avctx, AV_LOG_ERROR, "slice count %d is invalid (max=%d)\n", f->slice_count, f->max_slice_count);
return AVERROR_INVALIDDATA;
}
ff_refstruct_unref(&f->slice_damaged);
f->slice_damaged = ff_refstruct_allocz(f->slice_count * sizeof(*f->slice_damaged));
if (!f->slice_damaged)
return AVERROR(ENOMEM);
for (int j = 0; j < f->slice_count; j++) {
FFV1SliceContext *sc = &f->slices[j];
if (f->version == 2) {
int sx = get_symbol(c, state, 0);
int sy = get_symbol(c, state, 0);
int sw = get_symbol(c, state, 0) + 1U;
int sh = get_symbol(c, state, 0) + 1U;
if (sx < 0 || sy < 0 || sw <= 0 || sh <= 0)
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
if (sx > f->num_h_slices - sw || sy > f->num_v_slices - sh)
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
sc->slice_x = sx * (int64_t)f->width / f->num_h_slices;
sc->slice_y = sy * (int64_t)f->height / f->num_v_slices;
sc->slice_width = (sx + sw) * (int64_t)f->width / f->num_h_slices - sc->slice_x;
sc->slice_height = (sy + sh) * (int64_t)f->height / f->num_v_slices - sc->slice_y;
av_assert0((unsigned)sc->slice_width <= f->width &&
(unsigned)sc->slice_height <= f->height);
av_assert0 ( (unsigned)sc->slice_x + (uint64_t)sc->slice_width <= f->width
&& (unsigned)sc->slice_y + (uint64_t)sc->slice_height <= f->height);
}
lavc/ffv1: change FFV1SliceContext.plane into a RefStruct object Frame threading in the FFV1 decoder works in a very unusual way - the state that needs to be propagated from the previous frame is not decoded pixels(¹), but each slice's entropy coder state after decoding the slice. For that purpose, the decoder's update_thread_context() callback stores a pointer to the previous frame thread's private data. Then, when decoding each slice, the frame thread uses the standard progress mechanism to wait for the corresponding slice in the previous frame to be completed, then copies the entropy coder state from the previously-stored pointer. This approach is highly dubious, as update_thread_context() should be the only point where frame-thread contexts come into direct contact. There are no guarantees that the stored pointer will be valid at all, or will contain any particular data after update_thread_context() finishes. More specifically, this code can break due to the fact that keyframes reset entropy coder state and thus do not need to wait for the previous frame. As an example, consider a decoder process with 2 frame threads - thread 0 with its context 0, and thread 1 with context 1 - decoding a previous frame P, current frame F, followed by a keyframe K. Then consider concurrent execution consistent with the following sequence of events: * thread 0 starts decoding P * thread 0 reads P's slice header, then calls ff_thread_finish_setup() allowing next frame thread to start * main thread calls update_thread_context() to transfer state from context 0 to context 1; context 1 stores a pointer to context 0's private data * thread 1 starts decoding F * thread 1 reads F's slice header, then calls ff_thread_finish_setup() allowing the next frame thread to start decoding * thread 0 finishes decoding P * thread 0 starts decoding K; since K is a keyframe, it does not wait for F and reallocates the arrays holding entropy coder state * thread 0 finishes decoding K * thread 1 reads entropy coder state from its stored pointer to context 0, however it finds state from K rather than from P This execution is currently prevented by special-casing FFV1 in the generic frame threading code, however that is supremely ugly. It also involves unnecessary copies of the state arrays, when in fact they can only be used by one thread at a time. This commit addresses these deficiencies by changing the array of PlaneContext (each of which contains the allocated state arrays) embedded in FFV1SliceContext into a RefStruct object. This object can then be propagated across frame threads in standard manner. Since the code structure guarantees only one thread accesses it at a time, no copies are necessary. It is also re-created for keyframes, solving the above issue cleanly. Special-casing of FFV1 in the generic frame threading code will be removed in a later commit. (¹) except in the case of a damaged slice, when previous frame's pixels are used directly
2024-07-11 11:08:55 +02:00
ff_refstruct_unref(&sc->plane);
sc->plane = ff_ffv1_planes_alloc();
if (!sc->plane)
return AVERROR(ENOMEM);
for (int i = 0; i < f->plane_count; i++) {
PlaneContext *const p = &sc->plane[i];
if (f->version == 2) {
int idx = get_symbol(c, state, 0);
if (idx >= (unsigned)f->quant_table_count) {
av_log(f->avctx, AV_LOG_ERROR,
"quant_table_index out of range\n");
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
}
p->quant_table_index = idx;
context_count = f->context_count[idx];
}
if (f->version <= 2) {
av_assert0(context_count >= 0);
p->context_count = context_count;
}
}
}
return 0;
}
static av_cold int decode_init(AVCodecContext *avctx)
{
FFV1Context *f = avctx->priv_data;
2012-10-19 11:33:43 +03:00
int ret;
2015-08-22 03:14:12 +02:00
if ((ret = ff_ffv1_common_init(avctx)) < 0)
return ret;
if (avctx->extradata_size > 0 && (ret = read_extra_header(f)) < 0)
2012-10-19 11:33:43 +03:00
return ret;
2015-08-22 03:14:12 +02:00
if ((ret = ff_ffv1_init_slice_contexts(f)) < 0)
2012-10-19 11:33:43 +03:00
return ret;
return 0;
}
static int decode_frame(AVCodecContext *avctx, AVFrame *rframe,
int *got_frame, AVPacket *avpkt)
{
uint8_t *buf = avpkt->data;
int buf_size = avpkt->size;
FFV1Context *f = avctx->priv_data;
RangeCoder *const c = &f->slices[0].c;
int ret, key_frame;
uint8_t keystate = 128;
uint8_t *buf_p;
AVFrame *p;
ff_progress_frame_unref(&f->last_picture);
FFSWAP(ProgressFrame, f->picture, f->last_picture);
f->avctx = avctx;
f->frame_damaged = 0;
ff_init_range_decoder(c, buf, buf_size);
ff_build_rac_states(c, 0.05 * (1LL << 32), 256 - 8);
if (get_rac(c, &keystate)) {
key_frame = AV_FRAME_FLAG_KEY;
f->key_frame_ok = 0;
2012-10-19 11:33:43 +03:00
if ((ret = read_header(f)) < 0)
return ret;
f->key_frame_ok = 1;
} else {
if (!f->key_frame_ok) {
av_log(avctx, AV_LOG_ERROR,
2012-12-19 20:48:21 +03:00
"Cannot decode non-keyframe without valid keyframe\n");
return AVERROR_INVALIDDATA;
}
key_frame = 0;
}
if (f->ac != AC_GOLOMB_RICE) {
if (buf_size < avctx->width * avctx->height / (128*8))
return AVERROR_INVALIDDATA;
} else {
int w = avctx->width;
int s = 1 + w / (1<<23);
int i;
w /= s;
for (i = 0; w > (1<<ff_log2_run[i]); i++)
w -= ff_log2_run[i];
if (buf_size < (avctx->height + i + 6) / 8 * s)
return AVERROR_INVALIDDATA;
}
ret = ff_progress_frame_get_buffer(avctx, &f->picture,
AV_GET_BUFFER_FLAG_REF);
if (ret < 0)
2012-10-19 11:33:43 +03:00
return ret;
p = f->picture.f;
p->pict_type = AV_PICTURE_TYPE_I; //FIXME I vs. P
p->flags = (p->flags & ~AV_FRAME_FLAG_KEY) | key_frame;
if (f->version < 3 && avctx->field_order > AV_FIELD_PROGRESSIVE) {
/* we have interlaced material flagged in container */
p->flags |= AV_FRAME_FLAG_INTERLACED;
if (avctx->field_order == AV_FIELD_TT || avctx->field_order == AV_FIELD_TB)
p->flags |= AV_FRAME_FLAG_TOP_FIELD_FIRST;
}
2012-10-19 11:19:36 +03:00
if (avctx->debug & FF_DEBUG_PICT_INFO)
av_log(avctx, AV_LOG_DEBUG, "ver:%d keyframe:%d coder:%d ec:%d slices:%d bps:%d\n",
f->version, !!(p->flags & AV_FRAME_FLAG_KEY), f->ac, f->ec, f->slice_count, f->avctx->bits_per_raw_sample);
ff_thread_finish_setup(avctx);
buf_p = buf + buf_size;
for (int i = f->slice_count - 1; i >= 0; i--) {
FFV1SliceContext *sc = &f->slices[i];
int trailer = 3 + 5*!!f->ec;
int v;
sc->slice_damaged = 0;
if (i || f->version > 2) {
if (trailer > buf_p - buf) v = INT_MAX;
else v = AV_RB24(buf_p-trailer) + trailer;
} else v = buf_p - c->bytestream_start;
if (buf_p - c->bytestream_start < v) {
av_log(avctx, AV_LOG_ERROR, "Slice pointer chain broken\n");
ff_progress_frame_report(&f->picture, INT_MAX);
2012-10-19 11:33:43 +03:00
return AVERROR_INVALIDDATA;
}
buf_p -= v;
if (f->ec) {
unsigned crc = av_crc(av_crc_get_table(AV_CRC_32_IEEE), f->crcref, buf_p, v);
if (crc != f->crcref) {
int64_t ts = avpkt->pts != AV_NOPTS_VALUE ? avpkt->pts : avpkt->dts;
av_log(f->avctx, AV_LOG_ERROR, "slice CRC mismatch %X!", crc);
if (ts != AV_NOPTS_VALUE && avctx->pkt_timebase.num) {
av_log(f->avctx, AV_LOG_ERROR, "at %f seconds\n", ts*av_q2d(avctx->pkt_timebase));
} else if (ts != AV_NOPTS_VALUE) {
av_log(f->avctx, AV_LOG_ERROR, "at %"PRId64"\n", ts);
} else {
av_log(f->avctx, AV_LOG_ERROR, "\n");
}
slice_set_damaged(f, sc);
}
if (avctx->debug & FF_DEBUG_PICT_INFO) {
av_log(avctx, AV_LOG_DEBUG, "slice %d, CRC: 0x%08"PRIX32"\n", i, AV_RB32(buf_p + v - 4));
}
}
if (i) {
ff_init_range_decoder(&sc->c, buf_p, v);
ff_build_rac_states(&sc->c, 0.05 * (1LL << 32), 256 - 8);
} else
sc->c.bytestream_end = buf_p + v;
}
avctx->execute(avctx,
decode_slice,
f->slices,
NULL,
f->slice_count,
sizeof(*f->slices));
for (int i = f->slice_count - 1; i >= 0; i--) {
FFV1SliceContext *sc = &f->slices[i];
if (sc->slice_damaged && f->last_picture.f) {
const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
const uint8_t *src[4];
uint8_t *dst[4];
ff_progress_frame_await(&f->last_picture, INT_MAX);
for (int j = 0; j < desc->nb_components; j++) {
int pixshift = desc->comp[j].depth > 8;
int sh = (j == 1 || j == 2) ? f->chroma_h_shift : 0;
int sv = (j == 1 || j == 2) ? f->chroma_v_shift : 0;
dst[j] = p->data[j] + p->linesize[j] *
(sc->slice_y >> sv) + ((sc->slice_x >> sh) << pixshift);
src[j] = f->last_picture.f->data[j] + f->last_picture.f->linesize[j] *
(sc->slice_y >> sv) + ((sc->slice_x >> sh) << pixshift);
}
av_image_copy(dst, p->linesize, src,
f->last_picture.f->linesize,
avctx->pix_fmt,
sc->slice_width,
sc->slice_height);
f->slice_damaged[i] = 1;
}
}
ff_progress_frame_report(&f->picture, INT_MAX);
ff_progress_frame_unref(&f->last_picture);
if ((ret = av_frame_ref(rframe, f->picture.f)) < 0)
return ret;
*got_frame = 1;
return buf_size;
}
#if HAVE_THREADS
static int update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
{
FFV1Context *fsrc = src->priv_data;
FFV1Context *fdst = dst->priv_data;
if (dst == src)
return 0;
fdst->version = fsrc->version;
fdst->micro_version = fsrc->micro_version;
fdst->chroma_planes = fsrc->chroma_planes;
fdst->chroma_h_shift = fsrc->chroma_h_shift;
fdst->chroma_v_shift = fsrc->chroma_v_shift;
fdst->transparency = fsrc->transparency;
fdst->plane_count = fsrc->plane_count;
fdst->ac = fsrc->ac;
fdst->colorspace = fsrc->colorspace;
fdst->ec = fsrc->ec;
fdst->intra = fsrc->intra;
fdst->key_frame_ok = fsrc->key_frame_ok;
fdst->packed_at_lsb = fsrc->packed_at_lsb;
fdst->slice_count = fsrc->slice_count;
fdst->use32bit = fsrc->use32bit;
memcpy(fdst->state_transition, fsrc->state_transition,
sizeof(fdst->state_transition));
// in version 1 there is a single per-keyframe quant table, so
// we need to propagate it between threads
if (fsrc->version < 2)
memcpy(fdst->quant_tables[0], fsrc->quant_tables[0], sizeof(fsrc->quant_tables[0]));
for (int i = 0; i < fdst->num_h_slices * fdst->num_v_slices; i++) {
FFV1SliceContext *sc = &fdst->slices[i];
const FFV1SliceContext *sc0 = &fsrc->slices[i];
lavc/ffv1: change FFV1SliceContext.plane into a RefStruct object Frame threading in the FFV1 decoder works in a very unusual way - the state that needs to be propagated from the previous frame is not decoded pixels(¹), but each slice's entropy coder state after decoding the slice. For that purpose, the decoder's update_thread_context() callback stores a pointer to the previous frame thread's private data. Then, when decoding each slice, the frame thread uses the standard progress mechanism to wait for the corresponding slice in the previous frame to be completed, then copies the entropy coder state from the previously-stored pointer. This approach is highly dubious, as update_thread_context() should be the only point where frame-thread contexts come into direct contact. There are no guarantees that the stored pointer will be valid at all, or will contain any particular data after update_thread_context() finishes. More specifically, this code can break due to the fact that keyframes reset entropy coder state and thus do not need to wait for the previous frame. As an example, consider a decoder process with 2 frame threads - thread 0 with its context 0, and thread 1 with context 1 - decoding a previous frame P, current frame F, followed by a keyframe K. Then consider concurrent execution consistent with the following sequence of events: * thread 0 starts decoding P * thread 0 reads P's slice header, then calls ff_thread_finish_setup() allowing next frame thread to start * main thread calls update_thread_context() to transfer state from context 0 to context 1; context 1 stores a pointer to context 0's private data * thread 1 starts decoding F * thread 1 reads F's slice header, then calls ff_thread_finish_setup() allowing the next frame thread to start decoding * thread 0 finishes decoding P * thread 0 starts decoding K; since K is a keyframe, it does not wait for F and reallocates the arrays holding entropy coder state * thread 0 finishes decoding K * thread 1 reads entropy coder state from its stored pointer to context 0, however it finds state from K rather than from P This execution is currently prevented by special-casing FFV1 in the generic frame threading code, however that is supremely ugly. It also involves unnecessary copies of the state arrays, when in fact they can only be used by one thread at a time. This commit addresses these deficiencies by changing the array of PlaneContext (each of which contains the allocated state arrays) embedded in FFV1SliceContext into a RefStruct object. This object can then be propagated across frame threads in standard manner. Since the code structure guarantees only one thread accesses it at a time, no copies are necessary. It is also re-created for keyframes, solving the above issue cleanly. Special-casing of FFV1 in the generic frame threading code will be removed in a later commit. (¹) except in the case of a damaged slice, when previous frame's pixels are used directly
2024-07-11 11:08:55 +02:00
ff_refstruct_replace(&sc->plane, sc0->plane);
if (fsrc->version < 3) {
sc->slice_x = sc0->slice_x;
sc->slice_y = sc0->slice_y;
sc->slice_width = sc0->slice_width;
sc->slice_height = sc0->slice_height;
}
}
ff_refstruct_replace(&fdst->slice_damaged, fsrc->slice_damaged);
av_assert1(fdst->max_slice_count == fsrc->max_slice_count);
ff_progress_frame_replace(&fdst->picture, &fsrc->picture);
return 0;
}
#endif
static av_cold int ffv1_decode_close(AVCodecContext *avctx)
{
FFV1Context *const s = avctx->priv_data;
ff_progress_frame_unref(&s->picture);
ff_progress_frame_unref(&s->last_picture);
return ff_ffv1_close(avctx);
}
const FFCodec ff_ffv1_decoder = {
.p.name = "ffv1",
CODEC_LONG_NAME("FFmpeg video codec #1"),
.p.type = AVMEDIA_TYPE_VIDEO,
.p.id = AV_CODEC_ID_FFV1,
.priv_data_size = sizeof(FFV1Context),
.init = decode_init,
.close = ffv1_decode_close,
FF_CODEC_DECODE_CB(decode_frame),
UPDATE_THREAD_CONTEXT(update_thread_context),
.p.capabilities = AV_CODEC_CAP_DR1 |
AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_SLICE_THREADS,
.caps_internal = FF_CODEC_CAP_INIT_CLEANUP |
FF_CODEC_CAP_USES_PROGRESSFRAMES,
};