On Monday, January 20, 2014, Sean McGovern <[email protected]> wrote:
> On 1/20/14, Janne Grunau <[email protected]> wrote:
>> On 2014-01-20 15:27:46 +0100, Anton Khirnov wrote:
>>> The default get_buffer2() implementation (and possibly some
>>> user ones) does not allocate edges when this flag is set, which may
>>> expose bugs in some decoders. Until the 10 release is out, it is safer
>>> to remove this part.
>>> ---
>>>  libavcodec/utils.c |    5 -----
>>>  1 file changed, 5 deletions(-)
>>>
>>> diff --git a/libavcodec/utils.c b/libavcodec/utils.c
>>> index 2e418ec..044413a 100644
>>> --- a/libavcodec/utils.c
>>> +++ b/libavcodec/utils.c
>>> @@ -1091,11 +1091,6 @@ int attribute_align_arg
>>> avcodec_open2(AVCodecContext *avctx, const AVCodec *code
>>>              ret = AVERROR(EINVAL);
>>>              goto free_and_end;
>>>          }
>>> -
>>> -#if FF_API_EMU_EDGE
>>> -        /* force the emu edge flag on, since it's now always active */
>>> -        avctx->flags |= CODEC_FLAG_EMU_EDGE;
>>> -#endif
>>>      }
>>>  end:
>>>      entangled_thread_counter--;
>>
>> Fix for the h264 fate failure sent. Do you still want this committed as
>> protection against possible other bugs? fate-h264 seems to pass here now
>> with AddressSanitizer. One thing I've noticed though is that the default
>> get_buffer2() doesn't even allocate a little padding at the end of a
>> picture plane. I think at least some NEON code can read past the image.
>>
>> Janne
>
> It would not surprise me if AltiVec is similarly affected.
>
> This test has been failing on all the valgrind FATE stations (ppc64
> and x86_{32,64}) for quite some time now -- but I suspect for
> different reasons.
>
> -- Sean McG.
>

I should really check my facts better before clicking Send -- the failure
I've been seeing on this test is not via valgrind. Its the one I documented
in Bugzilla #552.

-- Sean McG.
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to