search for: quadrad

...oblem, so it > may be desirable to give the list-burr scheduler as much > flexibility as possible. OK. > I think it's worthwhile to investigate alternatives before going > too far down the path of splitting TokenFactors. I totally agree and I tried it. > Do we know where the quadradic FoldingSetNodeID operations are coming from? Re-computation of the FoldingSetNodeID hashes takes a lot of time for a node with 3200 operands, since the Bits array contains 6400 elements. Moreover, it looks like a temporary data structure is allocated for this computation, which also takes time...

Hi Chris, Your were totally right with your suggestion. I have implemented the code that : a) does not merge multiple TokenFactor nodes in the DAGCombiner::visitTokenFactor(), if the resulting TF node would contain more than 64 operands. b) produces a bunch of TokenFactor nodes with at most 64 operands, instead of one huge TokenFactor in the SelectionDAGLowering::getRoot(). If we have n