[PATCH] D130170: [AMDGPU] Stop running IR code sinking pass

Thu Jul 21 06:34:19 PDT 2022

foad added inline comments.

================
Comment at: llvm/test/CodeGen/AMDGPU/cgp-addressing-modes.ll:40

-; OPT-LABEL: @test_sink_global_small_max_i32_ds_offset(
-; OPT: %in.gep = getelementptr i8, i8 addrspace(1)* %in, i64 65535
----------------
Most of the tests in this file seem to be testing that we sink a getelementptr instruction into the same block as the load/store that uses it. I simply removed all the cases that fail.

================
Comment at: llvm/test/CodeGen/AMDGPU/sink-image-sample.ll:4
-
-; Test that image.sample instruction is sunk across the branch and not left in the first block. Since the kill may terminate the shader there might be no need for sampling the image.
-
----------------
This is no longer done by the codegen pipeline, by design.

Instead of removing this test I suppose we could change the RUN line to do "opt -sink" or "opt -sink | llc"?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D130170/new/

https://reviews.llvm.org/D130170