[llvm-commits] CVS: llvm/lib/Target/PowerPC/PPC32ISelDAGToDAG.cpp

[Chris Lattner]

On Thu, 18 Aug 2005, Jim Laskey wrote:
> Is there a is implemented on target test for this somewhere?

Yup, in PPC32ISelLowering.cpp, we have this:

   // If we're enabling GP optimizations, use hardware square root
   if (!TM.getSubtarget<PPCSubtarget>().isGigaProcessor()) {
     setOperationAction(ISD::FSQRT, MVT::f64, Expand);
     setOperationAction(ISD::FSQRT, MVT::f32, Expand);
   }

So the legalizer will take care of the processors that don't have it 
(expanding to a sqrt[f] call).

-Chris

>
> On Aug 17, 2005, at 9:53 PM, Nate Begeman wrote:
>
>> +  case ISD::FSQRT: {
>> +    MVT::ValueType Ty = N->getValueType(0);
>> +    CurDAG->SelectNodeTo(N, Ty, Ty == MVT::f64 ? PPC::FSQRT : PPC::FSQRTS,
>> +                         Select(N->getOperand(0)));
>> +    break;
>> +  }
>> 
>
>

-Chris

-- 
http://nondot.org/sabre/
http://llvm.org/