[llvm-branch-commits] [X86][NewPM] Port X86FlagsCopyLowering to NPM (PR #173752)

[Alexis Engelke via llvm-branch-commits]

================
@@ -945,3 +951,18 @@ void X86FlagsCopyLoweringPass::rewriteMI(MachineBasicBlock &MBB,
   MI.findRegisterUseOperand(X86::EFLAGS, /*TRI=*/nullptr)->setIsKill(true);
   LLVM_DEBUG(dbgs() << "    fixed instruction: "; MI.dump());
 }
+
+bool X86FlagsCopyLoweringLegacy::runOnMachineFunction(MachineFunction &MF) {
+  auto *MDTWrapper = getAnalysisIfAvailable<MachineDominatorTreeWrapperPass>();
+  MachineDominatorTree *MDT = MDTWrapper ? &MDTWrapper->getDomTree() : nullptr;
+  return X86FlagsCopyLoweringImpl(MDT).runOnMachineFunction(MF);
+}
+
+PreservedAnalyses
+X86FlagsCopyLoweringPass::run(MachineFunction &MF,
+                              MachineFunctionAnalysisManager &MFAM) {
+  MachineDominatorTree *MDT = &MFAM.getResult<MachineDominatorTreeAnalysis>(MF);
----------------
aengelke wrote:

Shouldn't this be getCachedResult? Most of the time the pass won't use the MDT, so computing it is often unnecessary.

https://github.com/llvm/llvm-project/pull/173752