[MachineRegisterInfo] Avoid having dbg.values affect code generation

[uabelho]

Summary

Use use_nodbg_empty() rather than use_empty() in
MachineRegisterInfo::EmitLiveInCopies() when determining if a livein
register has any uses or not. Otherwise a single dbg.value can make us
generate different code, meaning -g would affect code generation.

Found when compiling code for my out-of-tree target. Unfortunately I
haven't been able to reproduce the problem on X86 or any of the other
in-tree targets that I tried, so no test case.

[uabelho]

Not sure who should review this. Added Matthias who seems to have reviewed other patches in MRI.

Does this make sense even without a test case?

When I've tried to reproduce on X86 there always seems to be another COPY making use_empty() return
false, or the dbg.value has been dropped for some reason during ISel.

[MatzeB]

LGTM, thanks.

[uabelho]

Thank you!

llvm/trunk/lib/CodeGen/MachineRegisterInfo.cpp

 /// into the given entry block.
 void
 MachineRegisterInfo::EmitLiveInCopies(MachineBasicBlock *EntryMBB,
                                       const TargetRegisterInfo &TRI,
                                       const TargetInstrInfo &TII) {
   // Emit the copies into the top of the block.
   for (unsigned i = 0, e = LiveIns.size(); i != e; ++i)
     if (LiveIns[i].second) {
-      if (use_empty(LiveIns[i].second)) {
-        // The livein has no uses. Drop it.
+      if (use_nodbg_empty(LiveIns[i].second)) {
+        // The livein has no non-dbg uses. Drop it.
         //
         // It would be preferable to have isel avoid creating live-in
         // records for unused arguments in the first place, but it's
         // complicated by the debug info code for arguments.
         LiveIns.erase(LiveIns.begin() + i);
         --i; --e;
       } else {
         // Emit a copy.