[LoopUnroll] Keep UP.Count when considering complete unroll

[zzheng]

Summary

r286389 updates UP.Count unconditionally when considering complete unroll of loop with compile-time known TripCount.
This can overwrite UP.Count retrieved by getUnrollingPreferences().

[evstupac]

I agree that r286389 have changed behavior.
However to prevent this in future would you mind to add a test case?

[zzheng]

In D27911#628167, @evstupac wrote:

However to prevent this in future would you mind to add a test case?

This issue was found in our in-house branch with custom TTI::getUnrollingPreferences()... not easy to produce a test case.

[evstupac]

lib/Transforms/Scalar/LoopUnrollPass.cpp

785

Will it work if we put UP.Count change under FullUnrollTripCount condition?
If we leave the change as is it could happen that for full unroll UP.Count is less than MaxTripCount. Which is not ok.
Also could you please add context to the patch:
svn diff --diff-cmd=diff -x -U999999

[zzheng]

Moved UP.Count = FullUnrollTripCount under condition as suggested by Evgeny.

[haicheng]

lib/Transforms/Scalar/LoopUnrollPass.cpp

761

I think you need to set UP.Count in this situation, too.

[haicheng]

lib/Transforms/Scalar/LoopUnrollPass.cpp

761

Sorry, never mind.

[evstupac]

This issue was found in our in-house branch with custom TTI::getUnrollingPreferences()... not easy to produce a test case.

It looks like no one from trunk targets has UP.Count set in Unroll Preferences. So to add a test we'll need to add an option setting UP.Count default.

LGTM if all tests are passed.

lib/Transforms/Scalar/LoopUnrollPass.cpp

static bool computeUnrollCount(
   // If using the upper bound to unroll, TripMultiple should be set to 1 because
   // we do not know when loop may exit.
   // MaxTripCount and ExactTripCount cannot both be non zero since we only
   // compute the former when the latter is zero.
   unsigned ExactTripCount = TripCount;
   assert((ExactTripCount == 0 || MaxTripCount == 0) &&
          "ExtractTripCound and MaxTripCount cannot both be non zero.");
   unsigned FullUnrollTripCount = ExactTripCount ? ExactTripCount : MaxTripCount;
-  UP.Count = FullUnrollTripCount;
   if (FullUnrollTripCount && FullUnrollTripCount <= UP.FullUnrollMaxCount) {
+    UP.Count = FullUnrollTripCount;
     // When computing the unrolled size, note that BEInsns are not replicated
     // like the rest of the loop body.
     if (getUnrolledLoopSize(LoopSize, UP) < UP.Threshold) {
       UseUpperBound = (MaxTripCount == FullUnrollTripCount);
       TripCount = FullUnrollTripCount;
       TripMultiple = UP.UpperBound ? 1 : TripMultiple;
       return ExplicitUnroll;
     } else {
       // The loop isn't that small, but we still can fully unroll it if that
       // helps to remove a significant number of instructions.
       // To check that, run additional analysis on the loop.
       if (Optional<EstimatedUnrollCost> Cost = analyzeLoopUnrollCost(
               L, FullUnrollTripCount, DT, *SE, TTI,
               UP.Threshold * UP.MaxPercentThresholdBoost / 100)) {
         unsigned Boost =
             getFullUnrollBoostingFactor(*Cost, UP.MaxPercentThresholdBoost);
         if (Cost->UnrolledCost < UP.Threshold * Boost / 100) {
           UseUpperBound = (MaxTripCount == FullUnrollTripCount);
           TripCount = FullUnrollTripCount;
           TripMultiple = UP.UpperBound ? 1 : TripMultiple;
           return ExplicitUnroll;
         }
       }
     }
   }
 
   // 4rd priority is partial unrolling.
   // Try partial unroll only when TripCount could be staticaly calculated.
   if (TripCount) {
     UP.Partial |= ExplicitUnroll;
     if (!UP.Partial) {
       DEBUG(dbgs() << "  will not try to unroll partially because "
                    << "-unroll-allow-partial not given\n");
       UP.Count = 0;
       return false;
     }
     if (UP.Count == 0)
       UP.Count = TripCount;
     if (UP.PartialThreshold != NoThreshold) {
       // Reduce unroll count to be modulo of TripCount for partial unrolling.
       if (getUnrolledLoopSize(LoopSize, UP) > UP.PartialThreshold)
         UP.Count =
             (std::max(UP.PartialThreshold, UP.BEInsns + 1) - UP.BEInsns) /
             (LoopSize - UP.BEInsns);
       if (UP.Count > UP.MaxCount)
         UP.Count = UP.MaxCount;
       while (UP.Count != 0 && TripCount % UP.Count != 0)