[FFmpeg-devel] ICC version checks

[Måns Rullgård]

Carl Eugen Hoyos <cehoyos at ag.or.at> writes:

> M?ns Rullg?rd <mans <at> mansr.com> writes:
>
>> Currently 10.1.022 passes all FATE tests on x86_64, while 11.0.083 and
>> 11.1.059 pass the tests on x86_32.  All other combinations fail one or
>> more tests.
>
> 10.1 32 bit works, Reimar just broke one float test while FATE
> wasn't working, and I am too lazy to regression test.

That's an odd thing to say, seeing as it failed the mp2 test less than
an hour ago.  No floats involved there either, so I don't know what
you're getting at.

> 11.1 64 bit mostly works (with -parallel which I recommend anyway),

There is no mostly.  It either works or it does not.

> 11.0 64 bit is quite broken. I think it makes sense to keep the FATE
> test, so everybody sees that it is broken.

Agree.  This does however mean that we don't need to bother tuning
ifdefs optimally for this version.

> Even if we agree that 11.1 is also broken, I fear we have to support
> it since it is the only version Intel supports (according to their
> answer on bugreports) ...

Do they respond well at all to bug reports?

>> If the gcc compat mapping isn't too complicated for the versions we
>> want choose to support, we could put that logic into the
>> AV_GCC_VERSION_AT_LEAST macro.
>
> I don't think this would work reliably, no matter how few versions
> we decide to support.

Why not?

>> That would keep all the complexity in one place, and we wouldn't
>> need to worry when using that macro.
>
> I continue to volunteer to take care of these worries.

But I don't volunteer to look at the resulting mess.

-- 
M?ns Rullg?rd
mans at mansr.com