[RISCV] Use Zacas for AtomicRMWInst::Nand i32 and XLen. (llvm#80119)

We don't have an AMO instruction for Nand, so with the A extension we use an LR/SC loop. If we have Zacas we can use a CAS loop instead. According to the Zacas spec, a CAS loop scales to highly parallel systems better than LR/SC.
ROCm · Jan 31, 2024 · cf401f7 · cf401f7
1 parent 40ebe52
commit cf401f7
Show file tree

Hide file tree

Showing 3 changed files with 832 additions and 352 deletions.
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -19514,6 +19514,11 @@ RISCVTargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
  unsigned Size = AI->getType()->getPrimitiveSizeInBits();
  if (Size == 8 || Size == 16)
  return AtomicExpansionKind::MaskedIntrinsic;
+
+ if (Subtarget.hasStdExtZacas() && AI->getOperation() == AtomicRMWInst::Nand &&
+ (Size == Subtarget.getXLen() || Size == 32))
+ return AtomicExpansionKind::CmpXChg;
+
  return AtomicExpansionKind::None;
 }