Thanks to Rob for pointing to the adequate process documentation.
On behalf of Igor, I'm submitting an enhancement request that conforms to the process:
This changeset adds a new matching rule to the instruction selection phase (adlc) of the ppc vm.
It consists of the exchange of the pair li/cmpld with cmpldi.
Jtreg hotspot tests didn't regress.
Got an 1% performance improvement when benchmarking neo4j.
Since the patch is rather small and affects only the ppc vm, the risks are minimal.
Please, let us if we are breaking any jdk8u ground rules.