Re: [Aften-devel] gnu extensions

[Prakash P. <pr...@pu...>]

Hi,


> I noticed that _GNU_SOURCE was added for exp10.  I know exp10 is a gnu
> extension, but that's why I put in the emulation using exp(x * M_LN10).
>  It seems to me that using _GNU_SOURCE only quiets a gcc warning and
> nothing else, so is it really necessary?

Well, if we don't set _GNU_SOURCE, it is treated like it isn't really there=
,=20
ie, the header hides it, which is not very clean. So the compiler has to=20
guess how its prototype looks like. This is fragile (eg. if you cast=20
implicitly on calling exp10).

> I guess an easier way to get=20
> rid of all the mess would be to change the code to always use the
> alternate version.  What do you think?

It depends on how optimized the exp10 version is. If they do the same thing=
,=20
we could get rid of it, yes.


> Something similar applies to posix_memalign, but it's a little trickier.
>  At least it's only the backup.  A 2nd fallback for systems without
> posix_memalign might be to do something like FFmpeg's MEMALIGN_HACK.

I haven't taken a look into this. Aren't x86(_64) platforms (and those are =
the=20
only ones we have to care about because of SSE) w/o posix-memalign ancient?=
=20
Or do we need something like this on other OSes? I want to see someone=20
complaining first, before trying to think of a fallback. Using gcc 4.x , th=
e=20
compiler should provide _mm_malloc and all should be fine.

bye,

Prakash