RFR (M) #5 CR 8003985: Support @Contended annotation

[Vitaly Davidovich]

Sorry, typo - I meant to say it's not interesting if CPU treats two
adjacent lines as a coherence unit, not for prefetch.

Sent from my phone
On Jan 11, 2013 5:31 PM, "Vitaly Davidovich" <vitalyd at gmail.com> wrote:

> Yes I understand the issue, and I need to check this, but I'd hope the OS
> and/or CPU would report the proper coherence unit.  Most use cases for
> querying this info actually cares about the coherence unit size, so if they
> under report the size then those query capabilities are of almost no value
> (knowing that a line is 64 bytes isn't that interesting if CPU always
> fetches two of them).  So my point was that I'd think it's possible to get
> this info without hard coding and erring on the high side.
>
> Thanks
>
> Sent from my phone
> On Jan 11, 2013 4:50 PM, "Doug Lea" <dl at cs.oswego.edu> wrote:
>
>>
>>  On 01/11/2013 05:14 PM, Vitaly Davidovich wrote:
>>>
>>
>>  I still think it would be better to get line size from OS (or CPU)
>>>> rather than hardcoding a default - this seems orthogonal to the
>>>> prefetcher issue since the default you picked is probably going to be
>>>> either the same size or larger than what os/CPU would've reported.
>>>>
>>>
>> In case it helps resolve this a bit:
>>
>> The logistic problem is that we do not know a way to
>> reliably compute the effective cacheline length given
>> the possibility (and these days, likelihood) of adjacent
>> prefetching or other memory system quirks. And unreliable
>> detection is worse than no detection. So, in the near
>> term, the only course of action is to set this up using
>> reasonable defaults, and improve them over time.
>> Which is what Aleksey did.
>>
>> -Doug
>>
>>