Is i686 the minimum required for HotSpot on x86?

[Dmitry Samersoff]

Erik,

Can't say for i686 superset but lots of x86 compatible CPUs (AMD K6, VIA 
Eden etc) primary used in embedded space doesn't have cmov instruction.

So IMHO, we should care about it.

-Dmitry

On 2011-03-30 05:15, Erik Trimble wrote:
> On 3/29/2011 4:00 PM, Dr Andrew John Hughes wrote:
>> On 08:23 Tue 29 Mar , Vladimir Kozlov wrote:
>>> Andrew,
>>>
>>> It would be nice to have call stack trace. I looked on attached
>>> hs_er* files and non of them has it. JIT (Client VM)
>>> should not emit cmov instruction if it is not supported. Also there
>>> was suggestion in the report about running with
>>> -Xint, did it help? I surprise someone still try to run Hotspot on
>>> i586 ;) I can't even find P4 around here. And we
>>> definitely do not test on such machine (i586) so it is not supported.
>>>
>> I've prompted on the bug about the -Xint check. It sounds like twisti
>> has found the issue anyway.
>>
>> I don't even have an x86 box I'd want to try compiling OpenJDK on. The
>> only x86 box I still have is a 233MHz one
>> I use for basic stuff like DNS, proxy, etc. and even that's i686.
>> Presumably, someone else thought of this and
>> that's why there's the switch twisti mentioned that allows you to only
>> use certain instructions :-)
>>
>
> Actually, that brings up a good point. I can't think of any current x86
> CPU that isn't a superset of i686.
>
> Are there any such beasts? I ask, simply because at some point, we (i.e.
> the community, not Oracle) should probably be making some
> statement/decision about minimum supported configuration.
>
>
>


-- 
Dmitry Samersoff
Java Hotspot development team, SPB04
* There will come soft rains ...