[PATCH] D131562: [AArch64][SME] Document SME ABI implementation in LLVM

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

> For the purpose of LLVM IR it is sufficient to support only a single vscale as long as the input IR is restricted to not pass/return (pointers to) scalable vectors between such functions and any streaming mode changes happen only on function-call boundaries.

Sure, that makes sense: supporting multiple different scales within a function would be complicated.

But you still need a LangRef change.  Interprocedural optimizations need to know that "vscale" computed in the callee might be different from "vscale" computed in the caller.  As a practical matter, it probably affects very few optimizations, but certain optimizations might need to be aware of it.  (At the moment, this is most likely to be relevant for GEP constant expressions, since we don't really do interprocedural hoisting or sinking.  But in theory it could also be relevant for calls to llvm.vscale.)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D131562/new/

https://reviews.llvm.org/D131562