[patch][pr22947] Use sext in fast isel

[Eric Christopher]

Definitely more of what I had in mind on the IR. I'd suggest a comment
saying what the code generation in the testcases means (at least over the
first one).

Might want to cc one of the mips guys as a follow up to the commit to make
sure that they're happy with it. I believe this is what they want to be
actually testing, but there's a chance I'm wrong :)

-eric

On Mon, Apr 6, 2015 at 3:11 PM Rafael Espíndola <rafael.espindola at gmail.com>
wrote:

> New patch attached changing the IR instead.
>
> On 6 April 2015 at 17:23, Eric Christopher <echristo at gmail.com> wrote:
> > Ah right. That's... both obvious and non-obvious without some thought. :)
> >
> > That said, can you fix the testcases to do the right thing here? I think
> > instead of truncating and zero extending the value it should and with i8.
> >
> > -eric
> >
> > On Mon, Apr 6, 2015 at 1:28 PM Rafael Espíndola <
> rafael.espindola at gmail.com>
> > wrote:
> >>
> >> On 6 April 2015 at 16:23, Eric Christopher <echristo at gmail.com> wrote:
> >> > Hi Rafael,
> >> >
> >> > The changes to the mips fast isel test are worrying. From looking at
> the
> >> > testcase it's materializing a constant 1 and you have it now
> >> > materializing a
> >> > constant -1 (extended) into the register.
> >> >
> >>
> >> I think that is a bug fix, no? "i1 1" is -1.
> >>
> >> Cheers,
> >> Rafael
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150406/91856162/attachment.html>