[PATCH] D147245: [AMDGPU] Make amdgcn.s.buffer.load a memory-reading intrinsic

[Krzysztof Drewniak via Phabricator via llvm-commits]

krzysz00 added a comment.

>From what I can tell, if `!invariant.load` is supported for load-like intrinsics, that's not documented anywhere. It might be neat to make that work, but we can cross that bridge when we get there.

On the other hand, I'm imaging the following sequence of rewrites:

  %v = load T, ptr addrspace(7) %const.buffer, !invariant.load

=> (late in the frontend)

  %buf, %offset = ...(%const.buffer)
  %v = call T @llvm.amdgcn.raw.buffer.load(%buf, i32 %off. [...]) !invariant.load

=> (in IR translation)

  INTRINSIC_W_SIDE_EFFECT(@llvm.amdgcn.raw.buffer.load, ...) :: (dereferencable invariant load from BufferAddr(%buf, %off) ....)

=>(because the load was marked invariant, assuming `%off` is uniform)

  AMDGCN_S_BUFFER_LOAD %buf, %off, ... :: (...)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147245/new/

https://reviews.llvm.org/D147245