[PATCH] D139551: [OpenMP][AMDGPU] Use CAS loop for min/max atomics at system scope

[Gheorghe-Teodor Bercea via Phabricator via llvm-commits]

doru1004 created this revision.
doru1004 added reviewers: ABataev, carlo.bertolli, jdoerfert, tianshilei1992, ronl, jhuber6.
doru1004 added a project: OpenMP.
Herald added subscribers: kosarev, foad, kerbowa, guansong, hiraditya, tpr, dstuttard, yaxunl, jvesely, kzhuravl, arsenm.
Herald added a project: All.
doru1004 requested review of this revision.
Herald added subscribers: llvm-commits, sstefan1, wdng.
Herald added a project: LLVM.

Use CAS loop for min/max atomics at system scope. Currently a hardware atomic instruction is being used which does not work at system scope.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D139551

Files:
  llvm/lib/Target/AMDGPU/SIISelLowering.cpp
  llvm/test/CodeGen/AMDGPU/dag-divergence-atomic.ll
  llvm/test/CodeGen/AMDGPU/flat_atomics.ll
  llvm/test/CodeGen/AMDGPU/flat_atomics_i64.ll
  llvm/test/CodeGen/AMDGPU/global-saddr-atomics.ll
  llvm/test/CodeGen/AMDGPU/global_atomics.ll
  llvm/test/CodeGen/AMDGPU/global_atomics_i64.ll
  llvm/test/CodeGen/AMDGPU/move-to-valu-atomicrmw.ll