RFR: Eliminate write-barrier assembly stub (part 1)

[Aleksey Shipilev]

On 03/05/2018 01:59 PM, Roman Kennke wrote:
> The fast-path is already handled completely in C2 IR.

AFAIU, C2 WB still calls shenandoah_wb stub for doing the actual evac. What is probably going
against having writing the evac path in assembly is very rare writes to the cset regions. We have
never seen shenandoah_wb stub to be hot.

If we ever do see RT call to be hot, the way out would be to ask the mutator to evacuate more
objects at once in the RT slowpath, thus amortizing the cost of RT transition, instead of doing this
mess in arch-specific code.

> http://cr.openjdk.java.net/~rkennke/eliminate_wb_stub/webrev.01/

This makes sense to me. Looks good!

-Aleksey