RFR(S): 8173196: [REDO] C2 does not optimize redundant memory operations with G1
Vladimir Kozlov
vladimir.kozlov at oracle.com
Wed May 29 19:58:00 UTC 2019
Looks good to me too.
Thanks,
Vladimir
On 5/27/19 2:17 AM, Tobias Hartmann wrote:
> Hi Roland,
>
> this looks good to me. I'll run some extended testing and let you know once it's done.
>
> Thanks,
> Tobias
>
> On 22.05.19 11:25, Roland Westrelin wrote:
>>
>> http://cr.openjdk.java.net/~roland/8173196/webrev.00/
>>
>> Previous attempt at this was discussed here:
>> http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2016-January/021014.html
>>
>> And the follow up bugs with some comments on a possible fix:
>>
>> https://bugs.openjdk.java.net/browse/JDK-8172850
>>
>> The new fix is very similar to the previous one. The 2 differences are:
>>
>> - aarch64 code shouldn't need any change because of 8209420 ("Track
>> membars for volatile accesses so they can be properly optimized")
>>
>> - The membar only affects the raw memory slice which is now properly
>> handled by MachNodes thanks to 8209691 ("Allow MemBar on single memory
>> slice")
>>
>> Roland.
>>
More information about the hotspot-compiler-dev
mailing list