[FFmpeg-devel] [PATCH] remove SIGILL detection hacks

[Rich Felker]

On Tue, Oct 02, 2007 at 08:48:40PM +0200, Reimar D?ffinger wrote:
> Hello,
> On Tue, Oct 02, 2007 at 08:47:18PM +0200, Balatoni Denes wrote:
> > Tuesday 02 October 2007 15:42-kor Michael Niedermayer ezt ?rta:
> > > > The sparc patch would break on all non-vis2 sparcs, esp. since I
> > > > did not find any configure flag for it, but it seems the vis2 test flag
> > > > was never used?? So it should only break VIS-enabled builds on non-VIS
> > > > machines and nothing more.
> > > > Again, even compilation is untested for both.
> > >
> > > patch ok
> > 
> > If only I saw the point in removing VIS2 detection - it could easily be used 
> > later, and then it will have to be added back (I don't know any other way to 
> > detect VIS2 for Solaris SPARC).
> 
> There is a trivial way to detect: the user specified --disable-vis2 or
> he did not.
> You simply can't play with signal handlers in a library, that is not
> acceptable.
> And lastly: isn't that problem the designers of those extensions should
> have thought of? Maybe we should try filing a bug against the hardware
> guys once in a while :-P (though there are ways to work around it in
> software like adding a syscall to check - though that is uncomfortably slow -
> or a hard mapped cpuinfo page (you can fit a hell of a lot of flags in 4k
> ;-) )).

Are you sure the hardware doesn?t have a way to test for it? On the
i386 arch, 586 and up have a cpuid instruction that returns a good bit
of info about the specific cpu model, probably enough to determine
which extensions are available. I would be surprised if ppc doesn?t
have something similar..

Rich