r197224 - Enabling thumb2 mode used to force support for armv6t2. Replace this

Bernie Ogden bogden at arm.com
Mon Dec 16 09:09:47 PST 2013


Ah yes, I see now.

Perhaps it would have been worth using the subtarget feature combinations,
if only so that it is clear exactly what is being tested. If 1156 is being
used as a shorthand for the subtarget features, then all of that CPU's
subtarget features are set for every test. But I'm nitpicking.

Thanks for explaining.

Bernie

> -----Original Message-----
> From: Tim Northover [mailto:t.p.northover at gmail.com]
> Sent: 16 December 2013 16:33
> To: Bernard Ogden
> Cc: Joerg Sonnenberger; LLVM Commits
> Subject: Re: r197224 - Enabling thumb2 mode used to force support for
> armv6t2. Replace this
> 
> Hi Bernie,
> 
> > Perhaps I'm just failing to understand, but I don't understand why
> you need
> > to add an explicit CPU to all of these tests. If you're testing that
> CPUs
> > with Thumb2 also have V6T2Ops then wouldn't it be better to add a
> single
> > test that explicitly checks for this?
> 
> I haven't experimented, but I think the problem is more that after
> Joerg's change, just giving "-mattr=+thumb2" to llc would leave it in
> an inconsistent (or at least very odd) state. What's really desired in
> most cases is "a v6T2 processor", so he just went and added it.
> 
> An alternative would have been "-mattr=+thumb2,+v6t2". I don't know
> why he picked the one he did; perhaps even those two attributes aren't
> a sensible choice alone.
> 
> Cheers.
> 
> Tim.








More information about the llvm-commits mailing list