After #148, unnecessary reset work is done for the SLIL LB if it has not been computed before.

[Quincunx271]

#148 requires some work to reset the SLIL LB, because graph transformations can change the LB and thus the LB must be recomputed. Without a reset, the recomputation will fail. However, for the first computation of the SLIL LB, this is unnecessary reset work.

---

Seems like we may want to use a state variable / different control flow to only reset when necessary. In other words, if we haven't computed the SLILLB before, then this is just wasted cycles.

Originally posted by @jrbyrnes in #148 (comment)

[jrbyrnes]

Responding to your PR reply --

This issue seems at least somewhat related to #174 and can probably be handled in the same PR (after you've merged in).