[llvm-dev] [RFC] Improving integer divide optimization (related to D12082)

[Steve King via llvm-dev]

On Thu, Aug 20, 2015 at 9:59 AM, Mehdi Amini <mehdi.amini at apple.com> wrote:
> If you want the attributes, I think you should pass the attributes and not the whole Function.

OK.

> Did you consider what I suggested, i.e. inverting the order of the two tests?

I don't see this suggestion in our thread.


> Thinking about it more, I even think than removing the first test entirely would even be nicer here.
> The default implementation for BuildSDIVPow2() would be:
>
> virtual SDValue BuildSDIVPow2(SDValue N) {
>    bool MinSize = …. // choose a default, target can override anyway
>    if (TLI.isIntDivCheap(N->getValueType(0), MinSize))
>      return N;
>    return SDValue();
> }
>
> I believe this would preserve the current default behavior and leave targets the ability to override as much as they want.
>

Nice.  visitSDIV() and visitUDIV() still call isIntDivCheap() for
non-power-of-2 cases.  Passing attributes to isIntDivCheap() to enable
a smart check is still desirable, but you eliminated the need for a
power-of-2 flag.

Today's TargetLowering default:

virtual SDValue BuildSDIVPow2(SDNode *N, const APInt &Divisor,
                              SelectionDAG &DAG,
                              std::vector<SDNode *> *Created) const {
  return SDValue();
}

New default with your approach and Fn attributes:

virtual SDValue BuildSDIVPow2(SDNode *N, const APInt &Divisor,
                              SelectionDAG &DAG,
                              std::vector<SDNode *> *Created) const {

   AttributeSet Attr = DAG.getMachineFunction().getFunction()->getAttributes();
   if (TLI.isIntDivCheap(N->getValueType(0), Attr))
     return N;
   return SDValue();
}

virtual bool isIntDivCheap(EVT VT, AttributeSet Attr) const {
  return false;
}

X86 is the only in-tree user, so to preserve today's behavior:

bool X86TargetLowering::isIntDivCheap(EVT VT, AttributeSet Attr) const {
  bool OptSize = Attr.hasAttribute(AttributeSet::FunctionIndex,
Attribute::MinSize)
  return OptSize && !VT.isVector();
}