RFR 8003426: Remove UseFastAccessors and UseFastEmptyMethods except for zero
Volker Simonis
volker.simonis at gmail.com
Fri Aug 8 16:49:20 UTC 2014
Hi Coleen,
thank you very much for doing the ppc64 changes. I've checked the
ppc64 port on Aix/Linux with both opt/dbg and template/cpp
interpreter. The only two issues I found can be easily fixed (see
attached patch).
So for the ppc64 part of your change thumbs up from me!
Regards,
Volker
On Thu, Aug 7, 2014 at 6:17 PM, Coleen Phillimore
<coleen.phillimore at oracle.com> wrote:
> Summary: These options have been long disabled in Xmixed mode because they
> prevent these small methods from being inlined and are subject to bit rot,
> and we don't need more macro assembler code to maintain and change if the
> constant pool cache format changes.
>
> I ran specjvm98 -Xint and there wasn't a performance improvement with these
> flags. Zero reportedly needs these flags for performance, so I left them in
> for zero. I did build zero with some other changes to make zero build (not
> included in this patch) though. There are PPC64 changes which Volker
> offered to test. Also built c++ interpreter on x86 and sparc and got same
> runtime error as already reported.
>
> generate_method_entry was at the wrong level of inheritance so I fixed that
> and made it common. There was also some duplicate comment about frame
> layout and calling conventions (not sure it it's correct) that I either
> removed because it's in frame_x86.hpp or moved to before
> generate_fixed_frame() where it makes more sense.
>
> Tested with JPRT, nsk, jck and hotspot jtreg tests.
>
> open webrev at http://cr.openjdk.java.net/~coleenp/8003426/
> bug link https://bugs.openjdk.java.net/browse/JDK-8003426
>
> Thanks,
> Coleen
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 8003426_ppc64_addon.patch
Type: text/x-patch
Size: 1141 bytes
Desc: not available
URL: <http://mail.openjdk.java.net/pipermail/hotspot-dev/attachments/20140808/37cf580e/8003426_ppc64_addon.patch>
More information about the hotspot-dev
mailing list