[llvm-commits] [llvm] r45623 - /llvm/trunk/lib/Target/X86/X86InstrInfo.cpp

[Evan Cheng]

Hrm. ReMat of load from argument area is being handled as a special  
case in LiveIntervals::isReMaterializable(). Having  
isReallyTrivallyReMaterializable() returns true for it shouldn't make  
any difference. I'll look at this.

Evan

On Jan 4, 2008, at 10:10 PM, Chris Lattner wrote:

> Author: lattner
> Date: Sat Jan  5 00:10:42 2008
> New Revision: 45623
>
> URL: http://llvm.org/viewvc/llvm-project?rev=45623&view=rev
> Log:
> enable sinking and licm of loads from the argument area.  I'd like  
> to enable this
> for remat, but can't due to an RA bug.
>
> Modified:
>     llvm/trunk/lib/Target/X86/X86InstrInfo.cpp
>
> Modified: llvm/trunk/lib/Target/X86/X86InstrInfo.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/ 
> X86InstrInfo.cpp?rev=45623&r1=45622&r2=45623&view=diff
>
> ====================================================================== 
> ========
> --- llvm/trunk/lib/Target/X86/X86InstrInfo.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86InstrInfo.cpp Sat Jan  5 00:10:42  
> 2008
> @@ -141,6 +141,18 @@
>          MI->getOperand(2).getImm() == 1 &&
>          MI->getOperand(3).getReg() == 0)
>        return true;
> +
> +    // If this is a load from a fixed argument slot, we know the  
> value is
> +    // invariant across the whole function, because we don't  
> redefine argument
> +    // values.
> +#if 0
> +    // FIXME: This is disabled due to a remat bug. rdar://5671644
> +    MachineFunction *MF = MI->getParent()->getParent();
> +    if (MI->getOperand(1).isFI() &&
> +        MF->getFrameInfo()->isFixedObjectIndex(MI->getOperand 
> (1).getIndex()))
> +      return true;
> +#endif
> +
>      return false;
>    }
>    // All other instructions marked M_REMATERIALIZABLE are always  
> trivially
> @@ -188,6 +200,15 @@
>          MI->getOperand(2).getImm() == 1 &&
>          MI->getOperand(3).getReg() == 0)
>        return true;
> +
> +    // If this is a load from a fixed argument slot, we know the  
> value is
> +    // invariant across the whole function, because we don't  
> redefine argument
> +    // values.
> +    MachineFunction *MF = MI->getParent()->getParent();
> +    if (MI->getOperand(1).isFI() &&
> +        MF->getFrameInfo()->isFixedObjectIndex(MI->getOperand 
> (1).getIndex()))
> +      return true;
> +
>      return false;
>    }
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits