RFR: 8234688 [lworld] Supporting volatile semantic for flattenable fields

Frederic Parain frederic.parain at oracle.com
Mon Nov 25 14:39:50 UTC 2019


David,

Thank you for the review.

The main problem with flattening of small volatile inline fields is not
the layout or the runtime, but the support in the JIT that is not
trivial (it impacts scalarization). After discussion with Tobias,
this is not something likely to be implemented in the short term.

Fred

On 11/25/19 4:27 AM, David Simms wrote:
> 
> I'm good with the change, nice to add a test, thanks for that.
> 
> Just a small comment: we could probably allow for the flattening for 
> sizes that are naturally atomic for the H/W, i.e. x86_64 allows 64 bits 
> ? (Assuming correct alignment for "Copy::conjoint_memory_atomic" to 
> consume).
> 
> /D
> 
> 
> On 22/11/19 10:31 PM, Frederic Parain wrote:
>> Please review this small changeset to support volatile flattenable 
>> fields.
>> The short term solution is to not flatten these fields.
>>
>> CR:
>> https://bugs.openjdk.java.net/browse/JDK-8234688
>>
>> Webrev:
>> http://cr.openjdk.java.net/~fparain/volatile/webrev.00/
>>
>> Thank you,
>>
>> Fred
> 
> 


More information about the valhalla-dev mailing list