[FFmpeg-devel] tsan warning about data race in h264 decoder

Wan-Teh Chang wtc at google.com
Wed Jul 19 02:43:10 EEST 2017

Hi Ronald,

On Tue, Jul 18, 2017 at 11:41 AM, Ronald S. Bultje <rsbultje at gmail.com> wrote:
> Reminds me a little of
> http://git.videolan.org/?p=ffmpeg.git;a=commit;h=e72690b18da064f6c0f04f09ccde72b6636e3159.
> From a quick look, it appears that default_ref[] is unconditionally
> initialized in initialise_ref_list() (called from ff_h264_build_ref_list(),
> called from h264_slice_init()), so I think your suggested patch is correct.

Thanks a lot for taking a look and pointing me at the relevant code. I
reviewed those functions, and I think you are right.

I am curious: why do we need to sync short_ref[], long_ref[],
short_ref_count, and long_ref_count between threads?

> Can you send it as a git-formatted patch?

Done. Please see http://ffmpeg.org/pipermail/ffmpeg-devel/2017-July/213777.html.

Wan-Teh Chang

More information about the ffmpeg-devel mailing list