[8u] RFR: JDK-8253036: Support building the Zero assembler port on AArch64

[Volker Simonis]

Hi Andrew Hughes,

in your first email in this thread you mentioned that this change is a
prerequisite for the integration of the aarch64 port in 8u-dev.

In another private mail you were mentioning that the aarch64 port
which is intended for integration into 8u-dev will be based on
IcedTea's HotSpot at 8u272 and is targeted for 8u282 in January. The
Corretto team is very interested in having  the aarch64 port in 8u-dev
and we want to offer some assistance if you think that could help? Do
you already have a JBS issue to track this work?

We are also interested in which compiler you will be using for
building the aarch64 in 8u? We are currently building with GCC 7 but
that still produces quite some warnings in shared code so we are
interested in downporting changes from higher jdk versions which fix
these warnings. This is even more important as we plan to move to gcc
10 eventually because that has support for the new ARM outline atomics
[1] but that requires even more warning fixes. Using outline atomics
in compiled code showed nice performance improvements in our
benchmark results for some GC-heavy workloads.

Would be great if you could share the current state of the project and
let us know how we can help.

Thank you and best regards,
Volker

[1] https://community.arm.com/developer/tools-software/tools/b/tools-software-ides-blog/posts/making-the-most-of-the-arm-architecture-in-gcc-10

On Fri, Sep 18, 2020 at 10:31 AM Andrew Haley <aph at redhat.com> wrote:
>
> On 18/09/2020 02:58, Andrew Hughes wrote:
>  > On 15:04 Thu 17 Sep     , Aleksey Shipilev wrote:
>  >> On 9/17/20 1:41 PM, Andrew Haley wrote:
>  >>> On 17/09/2020 12:35, Aleksey Shipilev wrote:
>  >>>    > On 9/17/20 12:49 PM, Aleksey Shipilev wrote:
>  >>>    >> But hey, those definitions are missing even in the head JDK!
>  >>>    >
>  >>>    > Ha-ha. Oops. I have somewhat of the fix here:
>  >>>    >     https://bugs.openjdk.java.net/browse/JDK-8253284
>  >>>    >
>  >>>    > ...that would have to trickle in separately.
>  >>>
>  >>> OK. It doesn't seem right to me to enable AArch64/Zero in a update
>  >>> version if we know it's incorrect, and quite nastily so. But I don't
>  >>> mind if we can make sure that it gets fixed before it gets released.
>  >>> GCC has all the atomic builtins we need to get this right; I don't
>  >>> believe we need any platform-specific code at all.
>  >
>  > I don't see this as a blocker to this patch. It is an improvement on
>  > AArch64/Zero not building at all,
>
> I disagree. But in practice it doesn't matter because 8253284 is in progress
> and will be backported soon.
>
> --
> 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
>