Re: ffmpeg 2.8.6-1 test failures on hppa
Hi,
On 07.02.2016 17:09, Helge Deller wrote:
> On 06.02.2016 20:55, Helge Deller wrote:
>> On 06.02.2016 14:45, Andreas Cadhalpun wrote:
>>> It would be great if someone with the necessary hardware could investigate
>>> what causes this problem.
>>
>> For now I've installed the latest gcc version on the buildds and
>> did a give-back on ffmpeg. If it fails again, I'll take a closer look.
>
> Yes, there seems to be a gcc bug/problem.
> I've manually rebuilt ffmpeg-2.8.5-1 with gcc-5_5.3.1-8 and it fails the same way (but it built earlier with gcc-5_5.3.1-6).
> So, gcc-5_5.3.1-7 (and higher) currently have the problem.
Thanks for confirming that it is a regression in gcc-5.
On 07.02.2016 19:57, John David Anglin wrote:
> On 2016-02-07, at 1:29 PM, John David Anglin wrote:
>> On 2016-02-07, at 12:00 PM, John David Anglin wrote:
>>> We really need a testcase to isolate the problematic gcc change.
>> I've extracted preprocessed source from libavutil/crc.c. The crc-test has no dependencies on
>> any ffmpeg libraries, so it's a good test to use to find gcc problem.
>>
>> Program seems to compile okay with gcc-4.9.
>
> Program compiles okay with my latest build of gcc-6 but gcc-5_5.3.1-8 fails as expected.
>
> Gcc-5 was bootstrapped using itself. Suspect it was mis-compiled in stage1 leading
> to current bug. This type of bug can be hard to find.
>
> Maybe problem will go away if gcc-5 is bootstrapped with gcc-4.9.
I don't know. Maybe this problem is related to the regression seen on mips [1].
Does using '-fno-expensive-optimizations' work around the problem?
Best regards,
Andreas
1: https://bugs.debian.org/813858
Reply to: