[PATCH] D103342: [AtomicExpand] Merge cmpxchg success and failure ordering when appropriate.

[Eli Friedman via Phabricator via llvm-commits]

efriedma created this revision.
efriedma added reviewers: nemanjai, jyknight, t.p.northover, reames, craig.topper, asb.
Herald added subscribers: frasercrmck, dexonsmith, luismarques, apazos, sameer.abuasal, steven.zhang, s.egerton, Jim, jocewei, PkmX, jfb, the_o, brucehoult, MartinMosbeck, rogfer01, edward-jones, zzheng, jrtc27, niosHD, sabuasal, simoncook, johnrusso, rbar, hiraditya.
efriedma requested review of this revision.
Herald added a subscriber: MaskRay.
Herald added a project: LLVM.

If we're not emitting separate fences for the success/failure cases, we need to pass the merged ordering to the target so it can emit the correct instructions.

For the PowerPC testcase, we end up with extra fences, but that seems like an improvement over missing fences.  If someone wants to improve that, the PowerPC backed could be taught to emit the fences after isel, instead of depending on fences emitted by AtomicExpand.

Fixes https://bugs.llvm.org/show_bug.cgi?id=33332 .


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D103342

Files:
  llvm/include/llvm/IR/Instructions.h
  llvm/lib/CodeGen/AtomicExpandPass.cpp
  llvm/test/CodeGen/AArch64/GlobalISel/arm64-atomic-128.ll
  llvm/test/CodeGen/AArch64/GlobalISel/arm64-atomic.ll
  llvm/test/CodeGen/PowerPC/atomics-regression.ll
  llvm/test/CodeGen/RISCV/atomic-cmpxchg.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D103342.348584.patch
Type: text/x-patch
Size: 12954 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210528/c20fbf31/attachment.bin>