[12] RFR(XXS): JDK-8196950: AARCH64 - Add VM flags presets for Cavium Thunder X2 CPU
Dmitry Chuyko
dmitry.chuyko at bell-sw.com
Fri Nov 2 14:01:18 UTC 2018
Hi Andrew,
On 11/1/18 12:47 PM, Andrew Dinn wrote:
> On 31/10/18 17:40, Dmitry Chuyko wrote:
>> Please review small defaults correction.
>>
>> rfe: https://bugs.openjdk.java.net/browse/JDK-8196950
>> code:
>>
>> diff -r 16950b2eaebf src/hotspot/cpu/aarch64/vm_version_aarch64.cpp
>> --- a/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp Tue Oct 30
>> 09:13:00 2018 -0400
>> +++ b/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp Wed Oct 31
>> 20:33:41 2018 +0300
>> @@ -219,7 +219,7 @@
>> }
>> #ifdef COMPILER2
>> if (FLAG_IS_DEFAULT(UseFPUForSpilling)) {
>> - FLAG_SET_DEFAULT(UseFPUForSpilling, true);
>> + FLAG_SET_DEFAULT(UseFPUForSpilling, false);
>> }
>> #endif
>> }
>>
>> Having "-XX:+AvoidUnalignedAccesses -XX:+UseSIMDForMemoryOps
>> -XX:-UseFPUForSpilling" as a default works (in general) better for
>> SPECjvm and SPECjbb.
> That's fine as a goal. However, I don't think this is the right patch.
>
> Flag UseFPUForSpilling is defined in
> src/hotspot/share/opto/c2_globals.hpp as follows:
>
> product(bool, UseFPUForSpilling, false,
> \
> "Spill integer registers to FPU instead of stack when
> possible") \
>
> In other words the global default is already false. Indeed, that is the
> the default used for the other AArch64 variants.
>
> So, the correct patch is to remove the whole if clause. Of course, you
> should also remove the surrounding ifdef COMPILER2 endif that brackets it.
That's an option. Though "default" (false) and "false" are not exactly
the same, in the sense of explicitly expressing the difference in
benchmarks score, we'll have history for that.
Webrev with updated patch:
http://cr.openjdk.java.net/~dchuyko/8198294/webrev.00/
>
> regards,
>
>
> Andrew Dinn
> -----------
> Senior Principal Software Engineer
> Red Hat UK Ltd
> Registered in England and Wales under Company Registration No. 03798903
> Directors: Michael Cunningham, Michael ("Mike") O'Neill, Eric Shander
More information about the hotspot-compiler-dev
mailing list