[PATCH] D156358: [AArch64] Do not unnecessarily spill LR because of @llvm.returnaddress

[Anatoly Trosinenko via Phabricator via llvm-commits]

atrosinenko created this revision.
atrosinenko added reviewers: ab, kristof.beyls, apazos, pcc, psmith, t.p.northover.
Herald added a subscriber: hiraditya.
Herald added a project: All.
atrosinenko published this revision for review.
atrosinenko added a comment.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

While most features of https://github.com/apple/llvm-project/commit/7924c7d75ae0015a9fd9786a580b10b2190bccc6 are already implemented in mainline, `@llvm.returnaddress` still tends to be lowered like

  xpaci x30
  mov dest, x30

instead of

  mov dest, x30
  xpaci dest

This patch forces the return address to be copied out of LR first, if FEAT_PAuth is avaialble (at the instruction selection phase).


When lowering @llvm.returnaddress with FEAT_PAuth, move possibly signed
return address out of LR first, to prevent XPACI from clobbering the LR
when it would not be spilled otherwise.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D156358

Files:
  llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
  llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
  llvm/test/CodeGen/AArch64/aarch64-signedreturnaddress.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D156358.544467.patch
Type: text/x-patch
Size: 6158 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230726/860bce73/attachment.bin>