[PATCH] [AArch64 NEON]Replace fpimm with fpz32 for floating compare with zero.

Kevin Qin kevinqindev at gmail.com
Sun Dec 8 23:54:49 PST 2013


Hi Tim,

The reason changing fpz32short to fpz32 is that we think it's a bug in
ARMARM: all floating point zero should use unique syntax, and obviously #0.0
is more appropriate than #0.  ARMARM will fix this in future.


2013/12/6 Tim Northover <t.p.northover at gmail.com>

>
>   Hi Kevin,
>
>   I'm glad you're in the area! I was looking at these just yesterday and
> noticed that the syntax they use for printing is non-canonical. The zero
> should apparently print as "#0" rather than "#0.0"
>
>   Looks like there was some miscommunication in the code-review when they
> were committed. You originally suggested an "fpz32short" operand class for
> them, which printed as "#0" but was otherwise identical to fpz32. That was
> a good idea, and as I recall the only problem was that it had an unneeded
> "PrintMethod", but it somehow got dropped from the final patch.
>
>   Cheers.
>
>   Tim.
>
>   (I'm also horrified by that Neon_cmpz: it really shouldn't exist  as a
> separate AArch64ISD node from Neon_cmp. But we can probably leave that
> until the two backends are merged).
>
> http://llvm-reviews.chandlerc.com/D2349
>



-- 
Best Regards,

Kevin Qin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131209/b2dd16a4/attachment.html>


More information about the llvm-commits mailing list