[PATCH] D139817: [AMDGPU] Legalize soffset of buffer instruction

[Jay Foad via Phabricator via llvm-commits]

foad added a comment.

Looks good.



================
Comment at: llvm/test/CodeGen/AMDGPU/legalize-soffset-mbuf.ll:3
+
+define float @llvm_amdgcn_raw_buffer_load_f32(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: llvm_amdgcn_raw_buffer_load_f32
----------------
Remove `inreg` throughout? I don't think it does anything unless you use one of the graphics calling conventionsl like `amdgpu_ps`.

Also, you are using a non-uniform %rsrc so the compiler is (correctly) generating a waterfall loop, but that's not what you're testing for here. So could you change the test to use a uniform %rsrc, and then the generated code would be much simpler, and you could use `update_mir_test_checks.py`.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D139817/new/

https://reviews.llvm.org/D139817