[aarch64-port-dev ] [8u] RFR: JDK-8257192: Integrate AArch64 JIT port into 8u

[Andrew Haley]

On 11/27/20 8:43 AM, Aleksey Shipilev wrote:
> On 11/27/20 8:33 AM, Andrew Hughes wrote:
>> On 07:21 Fri 27 Nov     , Andrew Hughes wrote:
>>> Umbrella Bug: https://bugs.openjdk.java.net/browse/JDK-8257192
>>> Webrevs:
>>>    jdk: https://cr.openjdk.java.net/~andrew/openjdk8/8257192/jdk/webrev.01/
> 
> *) src/share/vm/adlc/formssel.cpp:
>    - equivalent_predicates addition changes the shared code path. Is this a bugfix?

It's necessary because only on AArch64 do we need to distinguish matcher
rules for loads and stores based on their predicates. Unfortunately this
caused performance regressions on SPARC (on Unsafe, IIRC.)

Therefore it should be AArch64-only.

> *) src/share/vm/opto/graphKit.cpp:
>    - I am a bit nervous about changing the default MemOrd for all GCs. I understand that is the part 
> of a large "8064611: AARCH64: Changes to HotSpot shared code" integration. Shouldn't it be protected 
> with AARCH64_ONLY as well, as other shared VM changes are in this patch?

Yes.

> 
> *) src/share/vm/opto/library_call.cpp
>     src/share/vm/opto/memnode.cpp:
>    - Ugh. This touches a rather delicate piece of code. I'd like Andrews (Haley and/or Dinn) 
> explicitly sign off on this.

We already did:
https://mail.openjdk.java.net/pipermail/aarch64-port-dev/2019-August/007870.html

Is this discussion adequate? I believe it is, and we'd only come to the same
conclusion if we had it again.

-- 
Andrew Haley  (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671