[PATCH] D79416: [SVE] Fix wrong usage of getNumElements() in matchIntrinsicType

[Eli Friedman via Phabricator via llvm-commits]

efriedma added inline comments.


================
Comment at: llvm/test/CodeGen/AArch64/sve-bad-intrinsics.ll:13
+
+declare <vscale x  2 x i64> @llvm.aarch64.sve.add.nxv2i64(<vscale x  2 x i1>, <vscale x  2 x i64>, <vscale x  2 x i64>)
----------------
david-arm wrote:
> efriedma wrote:
> > This isn't the testcase I asked for; this isn't affected by the patch. Please take a look at my suggested testcase again.
> OK I see. So I wrote a test exactly as you described, but I'm not sure how you want to verify it is rejected? When I run "./bin/opt -verify -S < ../llvm/test/CodeGen/AArch64/sve-bad-intrinsics.ll" on your test I get:
> 
> define <2 x i64> @add_i64_invalid() {
>   %out = call <2 x i64> @llvm.aarch64.sve.add.v2i64(<2 x i1> zeroinitializer, <2 x i64> zeroinitializer, <2 x i64> zeroinitializer)
>   ret <2 x i64> %out
> }
> 
> Were you expecting to see an error? Or were you simply hoping to see a remangled name for the intrinsic? I'm just not exactly sure what I should be checking.
Sorry, I messed up the testcase.  Should be something like this:

```
declare <vscale x 4 x i16> @llvm.arm.neon.vcvtfp2hf(<vscale x 4 x float>)
define <vscale x 4 x i16> @f() {
  %r = call <vscale x 4 x i16> @llvm.arm.neon.vcvtfp2hf(<vscale x 4 x float> zeroinitializer)
  ret <vscale x 4 x i16> %r
}
```


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D79416/new/

https://reviews.llvm.org/D79416