Valhalla/MVT microbenchmarks (first benchmarks and first results)

[Sergey Kuksenko]

I've checked that.

I didn't use fastdebug build, because of fastdebug is not acceptable for 
performance estimations. I did my own build where I disabled these options.
What I've got:

SumIN.mhie_derivedLoop - time to performance is increased from 45 
seconds to 2 seconds
matrix.MHIE_Derived.*TotalLoop - time to performance is increased from 
~220 seconds to 3 seconds.


On 08/01/2017 09:23 AM, Roland Westrelin wrote:
>> Also such slow warmup can't be explained simply by slower
>> reachingcompilation thresholds.
>> Derived value types (with counted loops) are ~3x-4x times slower than
>> boxed types analogues in the interpreter, but time to performance is
>> ~20x-40x times worse.
> You're right. I found the slow down is related to changes to the calling
> convention.
> With -XX:-ValueTypePassFieldsAsArgs -XX:-ValueTypeReturnedAsFields, the
> warmup is a lot faster. -XX:-ValueTypeReturnedAsFields is actually the
> one that makes the biggest difference. I'm experimenting with ways to
> speed things up.
>
> Roland.

-- 
Best regards,
Sergey Kuksenko