This was added in e3e3c82555, probably
as a workaround for the fact that the quant table was not permutated
and the IDCT coefficients are, meaning that you'd only get correct
reconstruction if the IDCT permutation was an identity matrix, which
happens to be the case when you use the simple IDCT. The quant table
permutation bug was fixed in 42dd1434bf,
meaning this workaround is no longer necessary.
In practical terms, before 42dd1434bf,
the PSNR between decodes of the fate-mdec using simple (C) or simplemmx
IDCTs was 35. After 42dd1434bf, it's 90.
These variables are shared between frame threads, but they are updated
post-setup_finished() if a EOB/EOS slice type occurs. Moving the EOB/EOS
slices to the next frame thread instance (by parsing them leading into
the next picture instead of trailing behind the last picture) effectively
prevents this race condition.
This fixes tsan failures on hevc-conformance-NoOutPrior_A_Qualcomm_1.
The av_log() is done outside the lock, but this way the accesses to the
field (reads and writes) are always protected by a mutex. The av_log()
is not run inside the lock context because it may involve user callbacks
and doing that in performance-sensitive code is probably not a good idea.
This should fix occasional tsan warnings when running fate-h264, like:
WARNING: ThreadSanitizer: data race (pid=10916)
Write of size 4 at 0x7d64000174fc by main thread (mutexes: write M2313):
#0 update_context_from_user src/libavcodec/pthread_frame.c:335 (ffmpeg+0x000000df7b06)
[..]
Previous read of size 4 at 0x7d64000174fc by thread T1 (mutexes: write M2311):
#0 ff_thread_await_progress src/libavcodec/pthread_frame.c:592 (ffmpeg+0x000000df8b3e)
I'm hoping that this will address the remaining tsan fate-h264 issues:
WARNING: ThreadSanitizer: data race (pid=24478)
Read of size 8 at 0x7dbc0001c828 by main thread (mutexes: write M3243):
#0 ff_h264_ref_picture src/libavcodec/h264_picture.c:107 (ffmpeg+0x0000013b78d8)
[..]
Previous write of size 1 at 0x7dbc0001c82e by thread T2 (mutexes: write M3245):
#0 ff_h264_direct_ref_list_init src/libavcodec/h264_direct.c:137 (ffmpeg+0x000001382c93)
But I'm not sure because I haven't been able to reproduce locally.
Fixes tsan warnings like this in fate-vp8-test-vector-007:
WARNING: ThreadSanitizer: data race (pid=65909)
Write of size 4 at 0x7d8c0000e088 by thread T1:
#0 vp8_decode_mb_row_sliced vp8.c:2519 (ffmpeg:x86_64+0x100995ede)
[..]
Previous write of size 4 at 0x7d8c0000e088 by thread T2:
#0 vp8_decode_mb_row_sliced vp8.c:2519 (ffmpeg:x86_64+0x100995ede)
Fixes tsan warnings like this in fate-vp8-test-vector-007:
WARNING: ThreadSanitizer: data race (pid=3590)
Write of size 4 at 0x7d8c0000e07c by thread T2:
#0 decode_mb_row_no_filter src/libavcodec/vp8.c:2330 (ffmpeg+0x000000ffb59e)
[..]
Previous write of size 4 at 0x7d8c0000e07c by thread T1:
#0 decode_mb_row_no_filter src/libavcodec/vp8.c:2330 (ffmpeg+0x000000ffb59e)
This removes the last use of the ff_put/add_pixels_clamped global
function pointers, and as such they are removed.
This patch has a negative effect on performance on MIPS, since there's
a SIMD-optimized put/add_pixels_clamped, but no xvid or jrev. From a
code maintenance point of view, that is probably acceptable.
Because the global function pointers are removed, this fixes the following
tsan warnings when running e.g. fate-dnxhr-parse:
WARNING: ThreadSanitizer: data race (pid=29917)
Write of size 8 at 0x0000025b12d8 by thread T2 (mutexes: write M1543):
#0 ff_idctdsp_init src/libavcodec/idctdsp.c:313 (ffmpeg+0x00000044b68e)
[..]
Previous write of size 8 at 0x0000025b12d8 by thread T1 (mutexes: write M1541):
#0 ff_idctdsp_init src/libavcodec/idctdsp.c:313 (ffmpeg+0x00000044b68e)
These use the mmx IDCT, but sse2 put/add_pixels_clamped implementations.
This way we don't need to use the ff_put/add_pixels_clamped function
pointers.
Instead, hardcode the use of the _arm implementation of add_pixels,
and use the C version for put_pixels (as no arm-optimized version
exists). Since there's separate implementations of idct{,_put,_add}
for neon, this has no practical impact on performance.
This tries to handle cases where separate invocations of decode_frame()
(each running in separate threads) write to respective fields in the
same AVFrame->data[]. Having per-field owners makes interaction between
readers (the referencing thread) and writers (the decoding thread)
slightly more optimal if both accesses are field-based, since they will
use the respective producer's thread objects (mutex/cond) instead of
sharing the thread objects of the first field's producer.
In practice, this fixes the following tsan-warning in fate-h264:
WARNING: ThreadSanitizer: data race (pid=21615)
Read of size 4 at 0x7d640000d9fc by thread T2 (mutexes: write M1006):
#0 ff_thread_report_progress pthread_frame.c:569 (ffmpeg:x86_64+0x100f7cf54)
[..]
Previous write of size 4 at 0x7d640000d9fc by main thread (mutexes: write M1004):
#0 update_context_from_user pthread_frame.c:335 (ffmpeg:x86_64+0x100f81abb)
Fixes the following tsan warning when running fate-vsynth_lena-ffvhuff:
WARNING: ThreadSanitizer: data race (pid=6484)
Write of size 8 at 0x7d64000154b8 by main thread (mutexes: write M1331):
#0 update_context_from_user src/libavcodec/pthread_frame.c:331 (ffmpeg+0x000000dca887)
[..]
Previous read of size 8 at 0x7d64000154b8 by thread T2 (mutexes: write M1334):
#0 draw_slice src/libavcodec/huffyuvdec.c:857 (ffmpeg+0x000000bcc86f)
Fixes tsan warnings in fate-apng:
WARNING: ThreadSanitizer: data race (pid=51230)
Read of size 4 at 0x7d50000042fc by main thread (mutexes: write M1000):
#0 frame_copy_props frame.c:302 (ffmpeg:x86_64+0x1019a35d6)
[..]
Previous write of size 4 at 0x7d50000042fc by thread T1 (mutexes: write M997):
#0 decode_idat_chunk pngdec.c:708 (ffmpeg:x86_64+0x100f5562a)
Fixes a reported (but false) race condition in tsan for fate-apng:
WARNING: ThreadSanitizer: data race (pid=6274)
Read of size 4 at 0x7d680001ec78 by main thread (mutexes: write M1338):
#0 update_thread_context src/libavcodec/pngdec.c:1456 (ffmpeg+0x000000dacf0c)
[..]
Previous write of size 4 at 0x7d680001ec78 by thread T1 (mutexes: write M1335):
#0 decode_idat_chunk src/libavcodec/pngdec.c:737 (ffmpeg+0x000000dae951)
Values from subsequent values are guaranteed to be identical (since
poc and nal_unit_type are checked to be the same between slices), so
this doesn't affect output in any way, but does resolve the remaining
reported race conditions (by tsan) in fate-hevc.
In practice, this fixes tsan warnings like this:
WARNING: ThreadSanitizer: data race (pid=25334)
Read of size 4 at 0x7d9c0001adcc by main thread (mutexes: write M1386):
#0 hevc_update_thread_context src/libavcodec/hevcdec.c:3310 (ffmpeg+0x000000b41c7c)
[..]
Previous write of size 4 at 0x7d9c0001adcc by thread T1 (mutexes: write M1383):
#0 hls_slice_header src/libavcodec/hevcdec.c:596 (ffmpeg+0x000000b43a22)
Otherwise the thread may still be in the middle of decoding a previous
frame, which would effectively trigger a race condition on any field
concurrently read and written.
In practice, this fixes tsan warnings like the following:
WARNING: ThreadSanitizer: data race (pid=17380)
Write of size 4 at 0x7d64000160fc by main thread:
#0 update_context_from_user src/libavcodec/pthread_frame.c:335 (ffmpeg+0x000000dca515)
[..]
Previous read of size 4 at 0x7d64000160fc by thread T2 (mutexes: write M1821):
#0 ff_thread_report_progress src/libavcodec/pthread_frame.c:565 (ffmpeg+0x000000dcb08a)
Should fix tsan warnings in fate-fifo-muxer-h264/wav:
WARNING: ThreadSanitizer: data race (pid=26552)
Write of size 4 at 0x000001e0d7c0 by main thread:
#0 transcode_init src/ffmpeg.c:3761 (ffmpeg+0x00000050ca1c)
[..]
Previous read of size 4 at 0x000001e0d7c0 by thread T1:
#0 decode_interrupt_cb src/ffmpeg.c:460 (ffmpeg+0x0000004fde19)
Fixes tsan warnings in several audio codecs (flac, alac, wavpack, tta
and tak) that look like this:
WARNING: ThreadSanitizer: data race (pid=14340)
Read of size 4 at 0x7d64000169d8 by main thread (mutexes: write M1335):
#0 update_context_from_thread src/libavcodec/pthread_frame.c:284 (ffmpeg+0x000000dc795f)
[..]
Previous write of size 4 at 0x7d64000169d8 by thread T1 (mutexes: write M1333):
#0 wavpack_decode_block src/libavcodec/wavpack.c:1012 (ffmpeg+0x00000112b175)
This is how the ref list manager links bitstream IDs to H264Picture/Ref
objects, and is local to the producer thread. There is no need for the
consumer thread to know the bitstream IDs of its references in their
respective producer threads.
In practice, this fixes tsan warnings when running fate-h264:
WARNING: ThreadSanitizer: data race (pid=19295)
Read of size 4 at 0x7dbc0000e614 by main thread (mutexes: write M1914):
#0 ff_h264_ref_picture src/libavcodec/h264_picture.c:112 (ffmpeg+0x0000013b3709)
[..]
Previous write of size 4 at 0x7dbc0000e614 by thread T2 (mutexes: write M1917):
#0 build_def_list src/libavcodec/h264_refs.c:91 (ffmpeg+0x0000013b46cf)
The value must be identical between slices, since mbaff depends on
picture_structure and sps, both of which are checked to be identical
to the first slice before this point.
In practice, this silences some tsan warnings in fate-h264.
This fixes race conditions reported by tsan in fate-lagarith. The races
were because each thread's LagarithContext::avctx was set to the first
thread's AVCodecContext.
Otherwise all thread's private contexts have the avctx pointer set to
the AVCodecContext of the first thread, which means all writes to
ctx->avctx->* (in e.g. read_header) are effectively race conditions.
Fixes fate-dnxhd under tsan.
Intra-only codecs should either be able to read these items from the
bitstream, or they should be set upon codec initialization. In both
cases, syncing these items at runtime is unnecessary.
In practice, this fixes race conditions for decoders that read these
values from the bitstream.
The checked bitstream reader does that already. To allow parsing of
superframes split over a packet boundary, we always decode the last
superframe in each packet at the start of the next packet, even if
theoretically we could have decoded it. The last superframe in the
last packet is decoded using AV_CODEC_CAP_DELAY.
Also a small cosmetic change to the avx2 idct16 version to make it
explicit that one of the arguments to the write-out macros is unused
for >=avx2 (it uses pmovzxbw instead of punpcklbw).
This is intended to workaround bug "665 Integer Divide Instruction May
Cause Unpredictable Behavior" on some early AMD CPUs, which causes a
div-by-zero in this codepath, such as reported in Mozilla bug #1293996.
Note that this isn't guaranteed to fix the bug, since a compiler is free
to reorder instructions that don't depend on each other. However, it
appears to fix the bug in Firefox, and a similar patch was applied to
libvpx also (see Chrome bug #599899).
filter16 goes from 508 to 482 (h) or 346 to 314 (v) cycles; filter88
goes from 240 to 238 (h) or 174 to 165 (v) cycles, measured on TOS.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
The randomize_buffer() implementation assures that "most of the time",
we'll do a good mix of wide16/wide8/hev/regular/no filters for complete
code coverage. However, this is not mathematically assured because that
would make the code either much more complex, or much less random.
Some fixes and improvements by Rodger Combs <rodger.combs@gmail.com>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
We need more information from last/cur_frame than from reference
buffers, so we can use a simplified structure for reference buffers,
and then store mvs and segmentation map information in last/cur.
This prepares the decoder for frame threading support.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
The frame dimensions are 16bit, so the mv bounds can easily overflow
int16 for large videos.
Bug-Id: Handbrake/46
CC: libav-stable@libav.org
Signed-off-by: Anton Khirnov <anton@khirnov.net>
Also check return value of av_frame_copy() in passthrough mode, so that
if a copy fails (as it did here, because bitdepth didn't match), the filter
doesn't return success, which would mean sending an uninitialized framebuffer
further down the filtergraph.
This tests automatic insertion of the vp9_superframe BSF as well as
ensuring that the colorspace properties in the video header can be
modified when remuxing (-c:v copy).
I have no idea why the first hunk uses ost->enc_ctx, because as far as
I understand, that is never used in case of -c:v copy, but this code
block is only entered if encoding_needed=0, which means stream_copy=1.
My point being: review from someone that knows this really well would
be appreciated.
Ronald S. Bultje
