[PATCHES] AMDGPU/SI: Small ISel improvements

[Marek Olšák via llvm-commits]

On Wed, Oct 14, 2015 at 8:32 PM, Matt Arsenault
<Matthew.Arsenault at amd.com> wrote:
> diff --git a/lib/Target/AMDGPU/SIISelLowering.cpp
> b/lib/Target/AMDGPU/SIISelLowering.cpp
> index 804b5e6..8d7a4d1 100644
> --- a/lib/Target/AMDGPU/SIISelLowering.cpp
> +++ b/lib/Target/AMDGPU/SIISelLowering.cpp
> @@ -1091,6 +1091,11 @@ SDValue
> SITargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
>                         DAG.getConstant(2, DL, MVT::i32), // P0
>                         Op.getOperand(1), Op.getOperand(2), Glue);
>    }
> +  case AMDGPUIntrinsic::SI_packf16:
> +    if (Op.getOperand(1).isUndef() && Op.getOperand(2).isUndef())
>
> Should this be ||?
>
> +      return DAG.getUNDEF(MVT::i32);
> +    else
> +      return Op;
>
> No return after else
>

What do you mean? Can you be more specific? It's the default behavior.

Marek