On profiling some specific Neo4j workloads, some cases where a sequence of load (li) and compare (cmpld) can be
replaced by a single cmpldi instructions were identified. On replacing that sequence by a single instruction it's
possible to reduce the path length so the program is shorter, and therefore is faster.
Additionally, the 2 instruction sequence requires an additional register that with enough register pressure a
spill would be needed to accommodate that approach.
It's related to the discussion started here:
http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2016-October/024664.html