RFR(XS): JDK-8207746: C2: Lucene crashes on AVX512 instruction: Fix

[Vladimir Kozlov]

Good. I will restart testing.

Thanks,
Vladimir

On 8/14/18 2:25 PM, Deshpande, Vivek R wrote:
> Hi Vladimir
> 
> Thanks Vladimir for reviewing it and suggesting for other places.
> We looked at other places and we have an updated webrev which fixes similar problem at 2 more locations.
> The updated webrev is here:
> http://cr.openjdk.java.net/~vdeshpande/Lucene_avx512/webrev.01/
> I have also updated the bug.
> 
> Regards,
> Vivek
> 
> -----Original Message-----
> From: Vladimir Kozlov [mailto:vladimir.kozlov at oracle.com]
> Sent: Tuesday, August 14, 2018 11:18 AM
> To: Deshpande, Vivek R <vivek.r.deshpande at intel.com>; hotspot-compiler-dev at openjdk.java.net compiler <hotspot-compiler-dev at openjdk.java.net>
> Cc: Viswanathan, Sandhya <sandhya.viswanathan at intel.com>
> Subject: Re: RFR(XS): JDK-8207746: C2: Lucene crashes on AVX512 instruction: Fix
> 
> Thank you very much, Vivek, for your hard work to find the issue and fixing it.
> 
> The fix totally makes sense. I will run all our test with it.
> 
> Meanwhile can you check other places (intrinsics) where we operate on stack pointer to make sure that they don't have the same issue? It is not urgent and separate issue but I am now concern about how we use it in code.
> 
> Thanks,
> Vladimir
> 
> On 8/14/18 10:37 AM, Deshpande, Vivek R wrote:
>> Hi All
>>
>> I have a patch for the fix for Lucene crash using AVX512.
>>
>> You can find the webrev at this location:
>>
>> http://cr.openjdk.java.net/~vdeshpande/Lucene_avx512/webrev.00/
>>
>> Could you please review the change.
>>
>> The EFLAGS register was getting clobbered, causing the result of
>> vptest instruction getting clobbered, leading to crash.
>>
>> This fix avoids the clobbering of the result of the vptest instruction.
>>
>> I tested the fix with elastic-search docker container on skylake
>> server. The container is running for more than 18 hours now.
>>
>> Regards,
>>
>> Vivek
>>