[aarch64-port-dev ] RFR(S): 8214922: Add vectorization support for fmin/fmax

[Andrew Haley]

On 12/21/18 3:37 PM, Andrew Haley wrote:
> On 12/21/18 2:25 PM, Roland Westrelin wrote:
>>
>>>> Why is the Matcher::max_vector_size(T_FLOAT) > 2 check needed?
>>>
>>> Code quality is worse, apparently.
>>
>> Isn't that a piece of aarch64 specific logic in shared code?
> 
> Yes. Worse than that, it's a piece of AArch64-microarchitecture-specific
> logic in shared code. So it doesn't even apply to all AArch64 implementations.

And, as it happens, I don't get a huge win when I take it out.

-XX:+UseSuperWord:

Benchmark                                Mode  Cnt    Score   Error  Units
TestSIMDFpMinMax.testVectMaxDouble       avgt    5  997.213 ± 7.660  ns/op



-XX:-UseSuperWord:


Benchmark                                Mode  Cnt    Score    Error  Units
TestSIMDFpMinMax.testVectMaxDouble       avgt    5  970.649 ± 11.908  ns/op


The difference is in the noise, I think.

-- 
Andrew Haley
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671