Changes in directory llvm/lib/Transforms/Scalar:
InstructionCombining.cpp updated: 1.392 -> 1.393 --- Log message: fold nested and's early to avoid inefficiencies in MaskedValueIsZero. This fixes a very slow compile in PR639: http://llvm.cs.uiuc.edu/PR639 . --- Diffs of the changes: (+9 -0) InstructionCombining.cpp | 9 +++++++++ 1 files changed, 9 insertions(+) Index: llvm/lib/Transforms/Scalar/InstructionCombining.cpp diff -u llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.392 llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.393 --- llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.392 Mon Oct 24 01:35:18 2005 +++ llvm/lib/Transforms/Scalar/InstructionCombining.cpp Wed Oct 26 12:18:16 2005 @@ -1725,6 +1725,15 @@ // and X, -1 == X if (AndRHS->isAllOnesValue()) return ReplaceInstUsesWith(I, Op0); + + // and (and X, c1), c2 -> and (x, c1&c2). Handle this case here, before + // calling MaskedValueIsZero, to avoid inefficient cases where we traipse + // through many levels of ands. + { + Value *X; ConstantInt *C1; + if (match(Op0, m_And(m_Value(X), m_ConstantInt(C1)))) + return BinaryOperator::createAnd(X, ConstantExpr::getAnd(C1, AndRHS)); + } if (MaskedValueIsZero(Op0, AndRHS)) // LHS & RHS == 0 return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType())); _______________________________________________ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
