[PATCH] D158603: [AMDGPU][TargetMachine] Handle case when +extended-image-insts is set, and the user forces +wave64

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D158603#4613311 <https://reviews.llvm.org/D158603#4613311>, @jmmartinez wrote:

> I have one question though. Why device_libs has that target specific attribute? Shouldn't it be deduced later in the optimization pipeline since the function has calls to image intrinsics?

This isn't an optimization. It cannot be inferred. This is supposed to be a restriction. Programs trying to use this on targets without the underlying instructions should be rejected


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D158603/new/

https://reviews.llvm.org/D158603