[AVX512] Add 512b integer shift by immediate intrinsics

[Robert Khasanov]

Hi Cameron,

LGTM, thank you!
Probably, you will modify instructions in TD file to support masking
patterns, I recommend to do these using AVX512_maskable multiclass.



2014-11-10 17:39 GMT+03:00 Cameron McInally <cameron.mcinally at nyu.edu>:

> Hey Robert,
>
> Please find an updated patch attached.
>
> Also, note that patterns will need to be added to match the masking
> shift operations. I will have to address these in a separate patch.
>
> -Cameron
>
> On Fri, Nov 7, 2014 at 12:19 PM, Robert Khasanov <rob.khasanov at gmail.com>
> wrote:
> > Hi Cameron,
> >
> > Your AVX512 builtins don't support merge-masking and zero-masking
> versions.
> > General, merge-masked and zero-masked Intel intrinsics will return one
> > builtin with the different mask and passthrough args.
> > Please add these arguments to your intrinsics, then you will need to
> create
> > another intrinsic type (VSHIFT_MASK), and handle it using
> > getVectorMaskingNode() function.
> >
> >
> > 2014-11-07 2:24 GMT+03:00 Cameron McInally <cameron.mcinally at nyu.edu>:
> >>
> >> Hey guys,
> >>
> >> Here is a small patch to add 512b shift by immediate intrinsics.
> >>
> >> If everything is cool, I'll send a patch for shift by vector next.
> >>
> >> Thanks,
> >> Cam
> >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141111/8e2d083a/attachment.html>