Index: lib/Analysis/ScalarEvolution.cpp =================================================================== --- lib/Analysis/ScalarEvolution.cpp +++ lib/Analysis/ScalarEvolution.cpp @@ -7207,6 +7207,25 @@ // N = Distance (as unsigned) if (StepC->getValue()->equalsInt(1) || StepC->getValue()->isAllOnesValue()) { APInt MaxBECount = getUnsignedRange(Distance).getUnsignedMax(); + + // When a loop like "for (int i = 0; i != n; ++i) { /* body */ }" is rotated, + // we end up with a loop whose backedge-taken count is n - 1. Detect this + // case, and see if we can improve the bound. + // + // Explicitly handling this here is necessary because getUnsignedRange + // isn't context-sensitive; it doesn't know that we only care about the + // range inside the loop. + const SCEV *Zero = getZero(Distance->getType()); + const SCEV *One = getOne(Distance->getType()); + const SCEV *DistancePlusOne = getAddExpr(Distance, One); + if (isLoopEntryGuardedByCond(L, ICmpInst::ICMP_NE, DistancePlusOne, Zero)) { + // If Distance + 1 doesn't overflow, we can compute the maximum distance + // as "unsigned_max(Distance + 1) - 1". + ConstantRange CR = getUnsignedRange(DistancePlusOne); + APInt NewMaxBECount = CR.getUnsignedMax() - 1; + if (NewMaxBECount.ult(MaxBECount)) + MaxBECount = NewMaxBECount; + } return ExitLimit(Distance, getConstant(MaxBECount), false, Predicates); } Index: test/Analysis/ScalarEvolution/max-trip-count.ll =================================================================== --- test/Analysis/ScalarEvolution/max-trip-count.ll +++ test/Analysis/ScalarEvolution/max-trip-count.ll @@ -242,9 +242,8 @@ ret i32 0 } -; TODO: Improve count based on guard. ; CHECK-LABEL: @ne_max_trip_count_3 -; CHECK: Loop %for.body: max backedge-taken count is -1 +; CHECK: Loop %for.body: max backedge-taken count is 6 define i32 @ne_max_trip_count_3(i32 %n) { entry: %masked = and i32 %n, 7 @@ -267,9 +266,8 @@ ret i32 0 } -; TODO: Improve count based on guard. ; CHECK-LABEL: @ne_max_trip_count_4 -; CHECK: Loop %for.body: max backedge-taken count is -1 +; CHECK: Loop %for.body: max backedge-taken count is -2 define i32 @ne_max_trip_count_4(i32 %n) { entry: %guard = icmp eq i32 %n, 0