[PATCH] D126535: [WebAssembly][NFC] Update reftype and table tests to use opaque pointers

Fri May 27 07:44:12 PDT 2022

asb added inline comments.

================
Comment at: llvm/test/CodeGen/WebAssembly/externref-globalset.ll:3

-%extern = type opaque
-%externref = type %extern addrspace(10)* ;; addrspace 10 is nonintegral
+%externref = type ptr addrspace(10) ;; addrspace 10 is nonintegral

----------------
sbc100 wrote:
> asb wrote:
> > sbc100 wrote:
> > > So the old code used `type opaque` and the new code just uses `ptr`?     But I take that the `ptr` is also somehow implicitly "opaque".. otherwise the PR description would not make sense?   (Sorry i'm not familiar the opaque pointer stuff).
> > See [here](https://llvm.org/docs/OpaquePointers.html) for an overview. Basically, pointers can still have an address space, but no longer have type information associated with them. Instead, you make use of a type argument for instructions that need it. As the page mentions, you can look at treatment of integer signedness as analogous - IR doesn't have signed and unsigned types, but has different operations that treat a value as signed or unsigned.
> > 
> > Clang now emits IR using opaque pointers by default, but the majority of .ll tests in llvm/test/ haven't been transitioned.
> I see..  the new `ptr` type is also opaque, so that the fact that you are removing the keyword `opaque` is not actually removing the opaqueness, since its now baked in.
Yep that's right, previously we had a typed pointer to an opaque struct type. Moving to `ptr`, all pointers are opaque.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D126535/new/

https://reviews.llvm.org/D126535