[PATCH] D93307: [SVE][CodeGen] Add bfloat16 support to scalable masked gather

[Cullen Rhodes via Phabricator via llvm-commits]

c-rhodes added inline comments.


================
Comment at: llvm/test/CodeGen/AArch64/sve-masked-scatter-legalise.ll:28-36
+define void @masked_scatter_nxv8bf16(<vscale x 8 x bfloat> %data, bfloat* %base, <vscale x 8 x i16> %offsets, <vscale x 8 x i1> %mask) #0 {
+; CHECK-LABEL: masked_scatter_nxv8bf16
+; CHECK-DAG: st1h { {{z[0-9]+}}.s }, {{p[0-9]+}}, [x0, {{z[0-9]+}}.s, sxtw #1]
+; CHECK-DAG: st1h { {{z[0-9]+}}.s }, {{p[0-9]+}}, [x0, {{z[0-9]+}}.s, sxtw #1]
+; CHECK: ret
+  %ptrs = getelementptr bfloat, bfloat* %base, <vscale x 8 x i16> %offsets
+  call void @llvm.masked.scatter.nxv8bf16(<vscale x 8 x bfloat> %data, <vscale x 8 x bfloat*> %ptrs, i32 1, <vscale x 8 x i1> %mask)
----------------
david-arm wrote:
> david-arm wrote:
> > c-rhodes wrote:
> > > Do we need a similar test for gather in `test/CodeGen/AArch64/sve-masked-gather-legalize.ll`? I also noticed these test files differ between American/British spelling of legalize/legalise in the filename, can this be renamed to use `legalize`?
> > +1 for the gather-legalize.ll tests.
> Hi @c-rhodes, I wonder if it's better to do the renaming in a different patch because @kmclaughlin has made changes in this file too and it makes the diff a bit harder to read? Any new files created can be with the American spelling.
> Hi @c-rhodes, I wonder if it's better to do the renaming in a different patch because @kmclaughlin has made changes in this file too and it makes the diff a bit harder to read? Any new files created can be with the American spelling.

Sure, I think it's fine to push an NFC patch with that single change straight to main.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D93307/new/

https://reviews.llvm.org/D93307