[llvm-commits] [llvm] r47043 - in /llvm/trunk/lib: CodeGen/TargetInstrInfoImpl.cpp Target/PowerPC/PPCInstrInfo.cpp Target/X86/X86InstrInfo.cpp

[Chris Lattner]

On Feb 12, 2008, at 6:46 PM, Evan Cheng wrote:

> Author: evancheng
> Date: Tue Feb 12 20:46:49 2008
> New Revision: 47043
>
> URL: http://llvm.org/viewvc/llvm-project?rev=47043&view=rev
> Log:
> commuteInstr() can now commute non-ssa machine instrs.
>
> Modified:
>    llvm/trunk/lib/CodeGen/TargetInstrInfoImpl.cpp
>    llvm/trunk/lib/Target/PowerPC/PPCInstrInfo.cpp
>    llvm/trunk/lib/Target/X86/X86InstrInfo.cpp
>
> Modified: llvm/trunk/lib/CodeGen/TargetInstrInfoImpl.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/TargetInstrInfoImpl.cpp?rev=47043&r1=47042&r2=47043&view=diff
>
> =
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
> --- llvm/trunk/lib/CodeGen/TargetInstrInfoImpl.cpp (original)
> +++ llvm/trunk/lib/CodeGen/TargetInstrInfoImpl.cpp Tue Feb 12  
> 20:46:49 2008
> @@ -23,8 +23,17 @@
>          "This only knows how to commute register operands so far");
>   unsigned Reg1 = MI->getOperand(1).getReg();
>   unsigned Reg2 = MI->getOperand(2).getReg();
> +  MachineOperand &MO = MI->getOperand(0);
> +  bool UpdateReg0 = MO.isReg() && MO.getReg() == Reg1;
>   bool Reg1IsKill = MI->getOperand(1).isKill();
>   bool Reg2IsKill = MI->getOperand(2).isKill();
> +  if (UpdateReg0) {

Why not just use:

if (MO.isReg() && MO.getReg() == Reg1)

Further, does it even make sense for instructions to be marked  
commutable if the first operand isn't a register?  Why not just use:

if (MO.getReg() == Reg1)

Bonus points for making tblgen reject instructions marked commutable  
that don't start with three register operands.

-Chris