[PATCH 3/3] ARM cost model: Cost for scalar integer casts and floating point conversions

[Renato Golin]

Hi Arnold,

Looks a lot better. The x86 function has multiple tables first, than the IF
blocks, which is slightly different style, if anything. I don't mind either
way, maybe Nadav has some preference.

If not, LGTM.

Thanks!

--renato


On 1 February 2013 19:05, Arnold Schwaighofer <aschwaighofer at apple.com>wrote:

> I created four tables:
>
> - NEONVectorConversionTbl for vector
> - NEONFloatConversionTbl for scalar floating point
> - NEONIntegerConversionTbl for scalar int to floating point
> - ARMIntegerConversionTbl for scalar int
>
> Thanks,
> Arnold
>
> On Feb 1, 2013, at 10:52 AM, Arnold Schwaighofer <aschwaighofer at apple.com>
> wrote:
> > On Feb 1, 2013, at 10:03 AM, Renato Golin <renato.golin at linaro.org>
> wrote:
> >>
> >> However, the table is called NEONConversionTbl and there's a test below
> (if ST->hasNEON), so it'd be better if the non-vector instructions could be
> added to a separate table.
> >>
> >> Possibly adding an Ty->isVector() && ST->hasNEON() or some combination
> that allows for both vectorized and scalar codes to execute correctly on
> all possible combinations, but doesn't iterate through the wrong table when
> the type is obviously not there.
> >
> > Yes, you are absolutely right. Will fix this. I was so focused on the
> table that I forgot about the surrounding code. Also better for performance
> (makes for shorter linear searches).
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130201/f218c453/attachment.html>