RFR: Bulk backports sh/jdk11 -> sh/jdk8u

[Roman Kennke]

After some more testing, we decided to cut out the C2 barrier 
reshapings. This requires more work, and we don't want to hold back the 
other stuff.

Changes summary:
http://cr.openjdk.java.net/~rkennke/backports-sh-jdk8-2019-03-15/changes.txt

Full webrev:
http://cr.openjdk.java.net/~rkennke/backports-sh-jdk8-2019-03-15/webrev.02/

Mercurial patch queue:
http://cr.openjdk.java.net/~rkennke/backports-sh-jdk8-2019-03-15/jdk8u-backports.tar.gz

WDYT now?

Roman

> On 3/16/19 11:36 AM, Roman Kennke wrote:
>>> *) Asserts seem excessive here, already under UseShenandoahGC branch:
>>>
>>> 2349 #if INCLUDE_ALL_GCS
>>> 2350   if (UseShenandoahGC) {
>>> 2351     if (mode == LoopOptsShenandoahExpand) {
>>> 2352       assert(UseShenandoahGC, "only for shenandoah");
>>> 2353       ShenandoahWriteBarrierNode::pin_and_expand(this);
>>> 2354     } else if (mode == LoopOptsShenandoahPostExpand) {
>>> 2355       assert(UseShenandoahGC, "only for shenandoah");
>>> 2356       visited.Clear();
>>> 2357       ShenandoahWriteBarrierNode::optimize_after_expansion(visited, nstack, worklist, this);
>>> 2358     }
>>
>> Newer versions don't have 'UseShenandoahGC' around the part, but should only ever executed (as
>> asserted). I suggest we either remove the if (UseShenandoahGC) in 8, or add it to later versions. Or
>> maybe just remove the assert in 8 and live with the difference? What do you think?
> 
> Nah, leave the assert as is. It is more important that the block contents are visually the same,
> even if it is doing excessive work for debug builds.
> 
> -Aleksey
>