[llvm-commits] [llvm] r48673 - in /llvm/trunk: include/llvm/CodeGen/SelectionDAG.h lib/CodeGen/SelectionDAG/DAGCombiner.cpp lib/CodeGen/SelectionDAG/SelectionDAG.cpp test/CodeGen/X86/dagcombine-cse.ll

[Chris Lattner]

On Mar 21, 2008, at 6:55 PM, Evan Cheng wrote:
> URL: http://llvm.org/viewvc/llvm-project?rev=48673&view=rev
> Log:
> Teach DAG combiner to commute commutable binary nodes in order to  
> achieve sdisel CSE.

Nice!

> +  // If N is a commutative binary node, try commuting it to enable  
> more
> +  // sdisel CSE.
> +  if (RV.Val == 0 &&
> +      SelectionDAG::isCommutativeBinOp(N->getOpcode()) &&
> +      N->getNumValues() == 1) {
> +    SDOperand N0 = N->getOperand(0);
> +    SDOperand N1 = N->getOperand(1);
> +    // Constant operands are canonicalized to RHS.
> +    if (isa<ConstantSDNode>(N0) || !isa<ConstantSDNode>(N1)) {
> +      SDOperand Ops[] = { N1, N0 };
> +      SDNode *CSENode = DAG.getNodeIfExists(N->getOpcode(), N- 
> >getVTList(),
> +                                            Ops, 2);

Ok, all this seems pretty reasonable.

> +      if (CSENode && CSENode->use_size() <= N->use_size())

Why are you comparing use_size?  use_size is soon to be a linear time  
operation, but in any case, I don't see why the # of uses should  
affect which one "wins".

-Chris