CR for RFR 8170039

[Vladimir Kozlov]

Looks good. I will sponsor it.

Thanks,
Vladimir

On 11/22/16 11:18 AM, Berg, Michael C wrote:
> Vladimir, please see the updated webrev which addresses the requested changes.
>
> webrev: http://cr.openjdk.java.net/~mcberg/8170039/webrev.02
>
> Thanks,
> Michael
>
> -----Original Message-----
> From: Vladimir Kozlov [mailto:vladimir.kozlov at oracle.com]
> Sent: Monday, November 21, 2016 11:59 AM
> To: Berg, Michael C <michael.c.berg at intel.com>; 'hotspot-compiler-dev at openjdk.java.net' <hotspot-compiler-dev at openjdk.java.net>
> Subject: Re: CR for RFR 8170039
>
> Hi Michael,
>
> Please, use FLAG_IS_DEFAULT(UseAVX) to check if flag is not changed on command line.
>
> Also I think is better to set a local boolean at the beginning of
> generate_get_cpu_info() method instead of using saved_useavx.
>
> And modify code at lines 595-608 to set default values based on cpu FLAG_IS_DEFAULT(UseAVX, use_avx) and using an other local. Guard it with if (FLAG_IS_DEFAULT(UseAVX)) and issue warning if flag set on command line bud conflict with supported by CPU value (as we do for other flags).
>
> Thanks,
> Vladimir
>
> On 11/21/16 10:59 AM, Berg, Michael C wrote:
>> Hi Folks,
>>
>> Glibc uses a single convention for save/restore, so when the user asks
>> for UseAVX={0,1,2} that excludes EVEX (3), we should offer non EVEX
>> based save/restore for that path by not emitting the EVEX vector check code.
>>
>>
>>
>> This code was tested as follows: hotspot jreg, SPECjvm2008, SPECjvm98
>> on hsw, skx and knl targets complete with no issues on 64-bit.
>>
>>
>> Bug-id: https://bugs.openjdk.java.net/browse/JDK-8170039
>>
>>
>> webrev: http://cr.openjdk.java.net/~mcberg/8170039/webrev.01
>>
>>
>>
>> Regards,
>>
>> Michael
>>