Is i686 the minimum required for HotSpot on x86?

[Christian Thalinger]

On Mar 30, 2011, at 10:33 AM, Christian Thalinger wrote:
> On Mar 30, 2011, at 12:44 AM, Dr Andrew John Hughes wrote:
>> On 17:31 Tue 29 Mar     , Christian Thalinger wrote:
>>> On Mar 29, 2011, at 4:35 PM, Christian Thalinger wrote:
>>>> On Mar 29, 2011, at 4:26 PM, Dr Andrew John Hughes wrote:
>>>>> We've just had a bug report for IcedTea which suggests that
>>>>> HotSpot fails on any x86 platform below i686.
>>>>> 
>>>>> The report is:
>>>>> 
>>>>> http://icedtea.classpath.org/bugzilla/show_bug.cgi?id=682
>>>>> 
>>>>> and shows a failure caused by an attempt by the JIT to emit
>>>>> an instruction which isn't present on i586.
>>>>> 
>>>>> Does HotSpot require i686 as a minimum? Or is there something
>>>>> else at fault here?
>>>> 
>>>> There is a switch -XX:+Use486InstrsOnly but I have an open bug for that one:
>>>> 
>>>> 6808665: Use486InstrsOnly aborts 32-bit VM
>>> 
>>> Looking a little closer this could actually be my fault.  I added unguarded cmov instructions to C1 with:
>>> 
>>> 6919934: JSR 292 needs to support x86 C1
>>> 
>>> http://hg.openjdk.java.net/jdk7/hotspot-comp/hotspot/rev/3cf667df43ef
>>> 
>>> And that was delivered in HS18b01.  I will fix that.
>>> 
>>> -- Christian
>> 
>> Ah thanks for finding the issue!  I await the fix.
> 
> 7032388: guarantee(VM_Version::supports_cmov()) failed: illegal instruction on i586 after 6919934

I case you haven't seen it:

http://hg.openjdk.java.net/jdk7/hotspot-comp/hotspot/rev/09f96c3ff1ad

Can you apply that patch to IcedTea and ask the bug submitter to try again?

-- Christian