[PATCH]: finish .cpu support for ARM

[Roman Divacky]

On Wed, Dec 03, 2014 at 12:51:19PM +0000, Renato Golin wrote:
> Hi Roman,
> 
> First, the test is very weak. It'd be good to show support for all our
> supported CPUs, and error on small variants or garbage. Please split
> in two files, one for what should work and another for what it
> shouldn't.
 
There are 56 different cpus to test according to your suggestion. I dont
even know what instruction to use to test that many variants.

Besides, the test is meant to test two things

1) That switching the cpu changes anything at all.

2) That switching to a invalid cpu produces an error.

I think the test does that. What you're requesting equals to testing
that the ARM*.td definitions are correct, which I think belongs somewhere
else.

Can you explain why you think testing all variants is useful? Note that
the switching depends on already existing (and I think already tested)
components.

Thanks, Roman

> Second, setting the feature bits is the wrong thing to do here, see
> http://llvm.org/PR20757 and http://llvm.org/PR20787.
> 
> This has been done for fpu and arch, so having it for cpu won't be
> terribly horrible, but the logic is broken. So, while having this
> patch (with a more thorough test) would be ok, you should put a big
> FIXME pointing to those two bugs.
> 
> I'll clear all of them up when I have the chance to work on this
> again, so adding the cpu hack won't be worse than it already is.
> 
> thanks,
> --renato
> 
> On 2 December 2014 at 19:23, Tim Northover <t.p.northover at gmail.com> wrote:
> > On 2 December 2014 at 11:18, Roman Divacky <rdivacky at vlakno.cz> wrote:
> >> Like this?
> >
> > Thanks. Yep, I think that looks reasonable now.
> >
> > Tim.
> > _______________________________________________
> > llvm-commits mailing list
> > llvm-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits