[PATCH] D135462: [SelectionDAG] Do not second-guess alignment for alloca

Martin Storsjö via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Jul 31 14:31:33 PDT 2023


mstorsjo added a comment.

In D135462#4548436 <https://reviews.llvm.org/D135462#4548436>, @efriedma wrote:

>> I've bisected a broken test in libcxx for the mingw/x86_64 target down to this commit - see https://github.com/llvm/llvm-project/issues/64253 for a full bug report including a reduced reproducer.
>
> Most likely exposing an issue elsewhere.  All this patch does is change the alignment of allocas, so most likely there's some unrelated problem related to or exposed by the stack layout.  (Or maybe some other optimization is making bad assumptions about alloca alignment, but that's unlikely given the way the APIs in question work.)  Probably using opt-bisect-limit to bisect the exact optimization in question would help narrow down the issue.

Thanks, I'll try to look into that and see if I can pinpoint anything. The surprising thing here though, is that the error occurs in unoptimized builds, while when building with optimizations enabled, the code turns out to work just right.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D135462/new/

https://reviews.llvm.org/D135462



More information about the llvm-commits mailing list