[clang-tools-extra] [CodeGen] Avoid potential sideeffects from XOR (PR #67193)

[Nick Desaulniers via cfe-commits]

================
@@ -9796,27 +9796,34 @@ X86InstrInfo::insertOutlinedCall(Module &M, 
MachineBasicBlock &MBB,
   return It;
 }
 
-void X86InstrInfo::buildClearRegister(Register Reg,
-                                      MachineBasicBlock &MBB,
+void X86InstrInfo::buildClearRegister(Register Reg, MachineBasicBlock &MBB,
                                       MachineBasicBlock::iterator Iter,
-                                      DebugLoc &DL) const {
+                                      DebugLoc &DL, bool NoSideEffects) const {
   const MachineFunction &MF = *MBB.getParent();
   const X86Subtarget &ST = MF.getSubtarget<X86Subtarget>();
   const TargetRegisterInfo &TRI = getRegisterInfo();
 
   if (ST.hasMMX() && X86::VR64RegClass.contains(Reg))
-    // FIXME: Ignore MMX registers?
+    // FIXME: Should we ignore MMX registers?
     return;
 
   if (TRI.isGeneralPurposeRegister(MF, Reg)) {
-    BuildMI(MBB, Iter, DL, get(X86::XOR32rr), Reg)
-      .addReg(Reg, RegState::Undef)
-      .addReg(Reg, RegState::Undef);
+    // Convert register to the 32-bit version.
----------------
nickdesaulniers wrote:

Maybe note that xor32rr and mov32ri will clear the upper 32b (if that's the 
case?)

https://github.com/llvm/llvm-project/pull/67193
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits