[PATCH] D156398: [AMDGPU] Remove post-PromoteAlloca SROA run

Pierre van Houtryve via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Jul 28 00:33:38 PDT 2023


Pierre-vh added a comment.

@arsenm There is a crash in `amdgpu_isel.test` from `check-llvm-tools-updatetestchecks`
This function fails ISel:

  ; RUN: llc -mtriple=amdgcn-amd-amdhsa -stop-after=finalize-isel -debug-only=isel -o /dev/null %s 2>&1 | FileCheck %s
  
  define i64 @i64_test(i64 %i) nounwind readnone {
    %loc = alloca i64
    %j = load i64, i64 * %loc
    %r = add i64 %i, %j
    ret i64 %r
  }

The `FrameIndex` fails:

  t25: i32,ch = load<(dereferenceable load (s32) from %ir.loc, align 8)> # D:1 t0, FrameIndex:i64<0>, undef:i64



  LLVM ERROR: Cannot select: t6: i64 = FrameIndex<0>
  In function: i64_test

Likewise there is a regression in `amdgpu_generated_funcs.ll` tests.

I think SROA was doing some (good) stuff for these tests. 
A simple fix seems to be to enable optimizations (-O3 works) for the crash.

I would suggest to just use O3 <https://reviews.llvm.org/owners/package/3/> there and open a separate (internal or external) ticket for that ISel failure. What do you think?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D156398/new/

https://reviews.llvm.org/D156398



More information about the llvm-commits mailing list