[PATCH][AVX512] Add 512b masked integer shift by immediate patterns

[Robert Khasanov]

LGTM. Thanks!

2014-11-14 8:31 GMT+03:00 Cameron McInally <cameron.mcinally at nyu.edu>:

> Tests updated. Also updated the CHECK statements to resemble the
> valign patterns.
>
>
> On Thu, Nov 13, 2014 at 11:21 PM, Cameron McInally
> <cameron.mcinally at nyu.edu> wrote:
> > Ugh, sorry. I was 1/2 asleep. I caught this just after sending. Here
> > is the updated patch...
> >
> > I'll also fix up the tests.
> >
> > -Cam
> >
> > On Thu, Nov 13, 2014 at 11:17 PM, Robert Khasanov
> > <rob.khasanov at gmail.com> wrote:
> >> In addition to previous:
> >>
> >> 3) In zero-masking tests I see that masks are not passed through args.
> >> Please rewrite them.
> >> +define <16 x i32> @test_x86_avx512_maskz_psrai_d(<16 x i32> %a0) {
> >> +  ; CHECK-LABEL: test_x86_avx512_maskz_psrai_d
> >> +  ; CHECK: vpsrad  $7, %zmm{{[0-9]+}}, %zmm{{[0-9]+}} {%k{{[0-9]+}}}
> {z}
> >> +  %res = call <16 x i32> @llvm.x86.avx512.mask.psrai.d(<16 x i32> %a0,
> i32
> >> 7, <16 x i32> zeroinitializer, i16 2)
> >> +  ret <16 x i32> %res
> >> +}
> >>
> >> For merge-masking tests please pass also pass-through value through
> args.
> >>
> >> 2014-11-14 7:07 GMT+03:00 Robert Khasanov <rob.khasanov at gmail.com>:
> >>>
> >>> Hi Cameron,
> >>>
> >>> 1) Please remove arguments from avx512_shift_rmi that would be not
> needed:
> >>> RC. You can get vt, x86memop and mem_frag from X86VectorVTInfo, they
> could
> >>> be also removed from args.
> >>> 2) I see you only added zero-masking tests. Please include also
> >>> merge-masking tests.
> >>>
> >>> 2014-11-14 6:58 GMT+03:00 Cameron McInally <cameron.mcinally at nyu.edu>:
> >>>>
> >>>> Hi guys,
> >>>>
> >>>> Here is a patch to add masked patterns for 512b integer shift by
> >>>> immediate.
> >>>>
> >>>> Thanks,
> >>>> Cam
> >>>
> >>>
> >>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141114/82f87c61/attachment.html>