Clarify HowFarToZero computation when the step is a positive power of two. Functionally this should be identical to the existing code except for the case where Step is maximally negative (eg, INT_MIN). We now punt in that one corner case to make reasoning about the code easier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224274 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Mark Heffernan 2014-12-15 21:19:53 +00:00
parent 1eb5ad1ecd
commit b10837d169

View File

@ -6094,14 +6094,19 @@ ScalarEvolution::HowFarToZero(const SCEV *V, const Loop *L, bool ControlsExit) {
return ExitLimit(Distance, MaxBECount); return ExitLimit(Distance, MaxBECount);
} }
// If Step is a power of two that evenly divides Start we know that the loop // As a special case, handle the instance where Step is a positive power of
// will always terminate. Start may not be a constant so we just have the // two. In this case, determining whether Step divides Distance evenly can be
// number of trailing zeros available. This is safe even in presence of // done by counting and comparing the number of trailing zeros of Step and
// overflow as the recurrence will overflow to exactly 0. // Distance.
const APInt &StepV = StepC->getValue()->getValue(); if (!CountDown) {
if (StepV.isPowerOf2() && const APInt &StepV = StepC->getValue()->getValue();
GetMinTrailingZeros(getNegativeSCEV(Start)) >= StepV.countTrailingZeros()) // StepV.isPowerOf2() returns true if StepV is an positive power of two. It
return getUDivExactExpr(Distance, CountDown ? getNegativeSCEV(Step) : Step); // also returns true if StepV is maximally negative (eg, INT_MIN), but that
// case is not handled as this code is guarded by !CountDown.
if (StepV.isPowerOf2() &&
GetMinTrailingZeros(Distance) >= StepV.countTrailingZeros())
return getUDivExactExpr(Distance, Step);
}
// If the condition controls loop exit (the loop exits only if the expression // If the condition controls loop exit (the loop exits only if the expression
// is true) and the addition is no-wrap we can use unsigned divide to // is true) and the addition is no-wrap we can use unsigned divide to