[llvm-commits] [llvm] r98007 - /llvm/trunk/lib/CodeGen/MachineCSE.cpp

[Evan Cheng]

On Mar 8, 2010, at 4:57 PM, Jakob Stoklund Olesen wrote:

> 
> On Mar 8, 2010, at 3:28 PM, Evan Cheng wrote:
> 
>> Author: evancheng
>> Date: Mon Mar  8 17:28:08 2010
>> New Revision: 98007
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=98007&view=rev
>> Log:
>> Restrict machine cse to really trivial coalescing. Leave the heavy lifting to a real coalescer.
>> 
>> Modified:
>>   llvm/trunk/lib/CodeGen/MachineCSE.cpp
>> 
>> Modified: llvm/trunk/lib/CodeGen/MachineCSE.cpp
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/MachineCSE.cpp?rev=98007&r1=98006&r2=98007&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/CodeGen/MachineCSE.cpp (original)
>> +++ llvm/trunk/lib/CodeGen/MachineCSE.cpp Mon Mar  8 17:28:08 2010
>> @@ -91,7 +91,10 @@
>>    unsigned SrcReg, DstReg, SrcSubIdx, DstSubIdx;
>>    if (TII->isMoveInstr(*DefMI, SrcReg, DstReg, SrcSubIdx, DstSubIdx) &&
>>        TargetRegisterInfo::isVirtualRegister(SrcReg) &&
>> +        MRI->getRegClass(SrcReg) == MRI->getRegClass(Reg) &&
>>        !SrcSubIdx && !DstSubIdx) {
>> +      DEBUG(dbgs() << "Coalescing: " << *DefMI);
>> +      DEBUG(dbgs() << "*** to: " << *MI);
>>      MO.setReg(SrcReg);
>>      DefMI->eraseFromParent();
>>      ++NumCoalesces;
> 
> You could add "|| MRI->getRegClass(Reg)->hasSubClass(MRI->getRegClass(SrcReg))" without getting too complicated. That could help a bit with the few x86_64 subclasses as well.
> 

I don't think it's necessary here. It's doing local coalescing here to avoid missing some obvious cse opportunities. If sub-reg class is involved, then it's already too complicated.

Evan