[8u-dev] RFA 8130150: Implement BigInteger.montgomeryMultiply intrinsic

[Seán Coffey]

Apologies for the delay. This is now approved for jdk8u-dev but with a 
caveat. The quality team are concerned with putting such a change into 
the update releases. For the short term, let's put the new features 
behind a flag and let's have the new feature off by default. Such 
intrinsics flags already exist in the hotspot code.

I'd suggest that you push the code as is and follow up with the default 
flag change in a separate bug ID. I would ask that both 8081778 and 
8130150 enhancements be off by default. We can document how to switch 
them on. Once enough testing is done, they should be switched on by 
default (in a later update release)

Regards,
Sean.

On 03/03/16 17:03, Vladimir Kozlov wrote:
> Sean, do you have other concerns?
>
> Thanks,
> Vladimir
>
> On 2/29/16 11:37 AM, Vladimir Kozlov wrote:
>> I filed:
>>
>> https://bugs.openjdk.java.net/browse/JDK-8150878
>>
>> Vladimir
>>
>> On 2/26/16 8:04 AM, Vladimir Kempik wrote:
>>> Hello Sean
>>>
>>> I have filled such bug yet because I haven't pushed the changeset yet,
>>> so the "bug" isn't present at the moment in any of java repos.
>>>
>>> Vladimir.
>>>
>>> On 26.02.2016 18:31, Seán Coffey wrote:
>>>> Vladimir Kozlov, Vladimir Kempik,
>>>>
>>>>> Separate bug is filed to resolve this issue later.
>>>>
>>>> can one of you give the bug ID capturing this ? Please link it to
>>>> JDK-8130150
>>>>
>>>> Regards,
>>>> Sean.
>>>>
>>>> On 19/02/16 18:24, Vladimir Kozlov wrote:
>>>>> Changes are reviewed. There is difference from jdk9 because of issue
>>>>> with code generated by SunStudio C++ version used to build jdk 8u and
>>>>> 7u. As result next changes where made:
>>>>>
>>>>> +  //montgomery_square fails to pass BigIntegerTest on solaris amd64
>>>>> +  //on jdk7 and jdk8.
>>>>> +#ifndef SOLARIS
>>>>>    if (len >= MONTGOMERY_SQUARING_THRESHOLD) {
>>>>> +#else
>>>>> +  if (0) {
>>>>> +#endif
>>>>>      ::montgomery_square(a, n, m, (unsigned long)inv, longwords);
>>>>>
>>>>> Note, this change is fine since it affects only performance for some
>>>>> range of values.
>>>>> Separate bug is filed to resolve this issue later.
>>>>>
>>>>> Thanks,
>>>>> Vladimir
>>>>>
>>>>> On 2/19/16 9:29 AM, Vladimir Kempik wrote:
>>>>>> Hello
>>>>>>
>>>>>> I'd like to port this fix into 8u-dev.
>>>>>>
>>>>>> Implement BigInteger.montgomeryMultiply intrinsic
>>>>>>
>>>>>> The patch has two parts, for jdk and hotspot.
>>>>>>
>>>>>> Testing: jprt, testcase.
>>>>>>
>>>>>> Bug:https://bugs.openjdk.java.net/browse/JDK-8130150
>>>>>> Webrev for
>>>>>> hotspot:http://cr.openjdk.java.net/~vkempik/8130150/webrev_hs.01/
>>>>>> Webrev for jdk:
>>>>>> http://cr.openjdk.java.net/~vkempik/8130150/webrev_jdk.00/
>>>>>> Review thread:
>>>>>> http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2016-February/021321.html 
>>>>>>
>>>>>>
>>>>>>
>>>>>> Thanks
>>>>>> -Vladimir
>>>>>>
>>>>
>>>