Statistics
| Branch: | Revision:

ffmpeg / mt-work / todo.txt @ 55ce3c67

History | View | Annotate | Download (3.35 KB)

1
Todo
2

    
3
-- For other people
4
- Multithread vp8 or vc1.
5
- Multithread an intra codec like mjpeg (trivial).
6
- Fix mpeg1 (see below).
7
- Try the first three items under Optimization.
8
- Fix h264 (see below).
9
- Try mpeg4 (see below).
10

    
11
-- Bug fixes
12

    
13
General critical:
14
- Error resilience has to run before ff_report_frame_progress()
15
is called. Otherwise there will be race conditions. (This might already
16
work.) In general testing error paths should be done more.
17
- 'make fate THREADS=2' doesn't pass. Most failures are due to
18
bugs in vsync in ffmpeg.c, which are currently obscuring real failures.
19

    
20
h264:
21
- Files split at the wrong NAL unit don't (and can't)
22
be decoded with threads (e.g. TS split so PPS is after
23
the frame, PAFF with two fields in a packet). Scan the
24
packet at the start of decode and don't finish setup
25
until all PPS/SPS have been encountered.
26

    
27
mpeg4:
28
- Packed B-frames need to be explicitly split up
29
when frame threading is on. It's not very fast
30
without this.
31
- The buffer age optimization is disabled due to
32
the way buffers are allocated across threads. The
33
branch 'fix_buffer_age' has an attempt to fix it
34
which breaks ffplay.
35
- Support interlaced.
36

    
37
mpeg1/2:
38
- Seeking always prints "first frame not a keyframe"
39
with threads on. Currently disabled for this reason.
40

    
41
-- Prove correct
42

    
43
- decode_update_progress() in h264.c
44
race_checking branch has some work on h264,
45
but not that function. It might be worth putting
46
the branch under #ifdef DEBUG in mainline, but
47
the code would have to be cleaner.
48
- MPV_lowest_referenced_row() and co in mpegvideo.c
49
- Same in vp3.
50

    
51
-- Optimization
52

    
53
- Merge h264 decode_update_progress() with loop_filter().
54
Add CODEC_CAP_DRAW_HORIZ_BAND as a side effect.
55
- EMU_EDGE is always set for h264 PAFF+MT
56
because draw_edges() writes into the other field's
57
thread's pixels. Needs an option to skip T/B fields.
58
- Check update_thread_context() functions and make
59
sure they only copy what they need to.
60
- Try some more optimization of the "ref < 48; ref++"
61
loop in h264.c await_references(), try turning the list0/list1 check
62
above into a loop without being slower.
63
- Support frame+slice threading at the same time
64
by assigning slice_count threads for frame threads
65
to use with execute(). This is simpler but unbalanced
66
if only one frame thread uses any.
67

    
68
-- Features
69

    
70
- Support streams with width/height changing. This
71
requires flushing all current frames (and buffering
72
the input in the meantime), closing the codec and
73
reopening it. Or don't support it.
74
- Support encoding. Might need more threading primitives
75
for good ratecontrol; would be nice for audio and libavfilter too.
76
- Async decoding part 1: instead of trying to
77
start every thread at the beginning, return a picture
78
if the earliest thread is already done, but don't wait
79
for it. Not sure what effect this would have.
80
- Part 2: have an API that doesn't wait for the decoding
81
thread, only returns EAGAIN if it's not ready. What will
82
it do with the next input packet if it returns that?
83
- Have an API that returns finished pictures but doesn't
84
require sending new ones. Maybe allow NULL avpkt when
85
not at the end of the stream.
86

    
87
-- Samples
88

    
89
http://astrange.ithinksw.net/ffmpeg/mt-samples/
90

    
91
See yuvcmp.c in this directory to compare decoded samples.
92

    
93
For debugging, try commenting out ff_thread_finish_setup calls so
94
that only one thread runs at once, and then binary search+
95
scatter printfs to look for differences in codec contexts.