[vectorIntrinsics+compress] RFR: 8274975: Add micro benchmark: ALIBABA selective store use case

[Joshua Zhu]

On Sat, 9 Oct 2021 23:56:31 GMT, Jatin Bhateja <jbhateja at openjdk.org> wrote:

>> I separate my implementation of "compress" API into several patches for easy review.
>> This micro benchmark is extracted from the real use case inside in-memory databases: selectiveStore.
>> Test scenario: UseAVX=3; thread number = 8; conflict data percentage: 20% (that means 20% of mask bits are true)
>
> test/micro/org/openjdk/bench/jdk/incubator/vector/bigdata/SelectiveStore.java line 109:
> 
>> 107:       IntVector bv = IntVector.fromArray(INT_128_SPECIES, tld.input2, i);
>> 108:       IntVector cv = IntVector.fromArray(INT_128_SPECIES, tld.index, i);
>> 109:       VectorMask<Integer> mask = av.compare(VectorOperators.NE, bv);
> 
> Can we move vector compare out of benchmarking loop to reduce the noise.

Thanks Jatin for your review.
This test I added is placed together with other benchmarks extracted from user cases.
The mask computation inside the loop is the essential logic in real scenario.
We can add micro benchmarks in test/micro/org/openjdk/bench/jdk/incubator/vector/operation and then evaluate performance for compression / expanding in single API level.

-------------

PR: https://git.openjdk.java.net/panama-vector/pull/149