[llvm] r175122 - Fix some issues with rematerialization in RegisterCoalescer when the destination

[Cameron Zwarich]

Author: zwarich
Date: Wed Feb 13 20:51:03 2013
New Revision: 175122

URL: http://llvm.org/viewvc/llvm-project?rev=175122&view=rev
Log:
Fix some issues with rematerialization in RegisterCoalescer when the destination
of the copy is a subregister def. The current code assumes that it can do a full
def of the destination register, but it is not checking that the def operand is
read-undef. It also doesn't clear the subregister index of the destination in
the new instruction to reflect the full subregister def.

These issues were found running 'make check' with my next commit that enables
rematerialization in more cases.

Modified:
    llvm/trunk/lib/CodeGen/RegisterCoalescer.cpp

Modified: llvm/trunk/lib/CodeGen/RegisterCoalescer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/RegisterCoalescer.cpp?rev=175122&r1=175121&r2=175122&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/RegisterCoalescer.cpp (original)
+++ llvm/trunk/lib/CodeGen/RegisterCoalescer.cpp Wed Feb 13 20:51:03 2013
@@ -755,6 +755,10 @@ bool RegisterCoalescer::reMaterializeTri
   const MCInstrDesc &MCID = DefMI->getDesc();
   if (MCID.getNumDefs() != 1)
     return false;
+  // Only support subregister destinations when the def is read-undef.
+  MachineOperand &DstOperand = CopyMI->getOperand(0);
+  if (DstOperand.getSubReg() && !DstOperand.isUndef())
+    return false;
   if (!DefMI->isImplicitDef()) {
     // Make sure the copy destination register class fits the instruction
     // definition register class. The mismatch can happen as a result of earlier
@@ -773,6 +777,12 @@ bool RegisterCoalescer::reMaterializeTri
   TII->reMaterialize(*MBB, MII, DstReg, 0, DefMI, *TRI);
   MachineInstr *NewMI = prior(MII);
 
+  // The original DefMI may have been a subregister def, but the full register
+  // class of its destination matches the destination of CopyMI, and CopyMI is
+  // either a full register def or is read-undef. Therefore we can clear the
+  // subregister index on the rematerialized instruction.
+  NewMI->getOperand(0).setSubReg(0);
+
   // NewMI may have dead implicit defs (E.g. EFLAGS for MOV<bits>r0 on X86).
   // We need to remember these so we can add intervals once we insert
   // NewMI into SlotIndexes.