[vector] RFR 8219878: Vectorized Poly1305 benchmark

[Viswanathan, Sandhya]

Hi Adam,

Thank a lot for the crypto benchmarks.

Just curious, what performance gain do you see for Poly1305 with the multiplication optimization?

Best Regards,
Sandhya


-----Original Message-----
From: panama-dev [mailto:panama-dev-bounces at openjdk.java.net] On Behalf Of Adam Petcher
Sent: Tuesday, March 05, 2019 12:06 PM
To: panama-dev at openjdk.java.net
Subject: [vector] RFR 8219878: Vectorized Poly1305 benchmark

webrev: https://cr.openjdk.java.net/~apetcher/8219878/webrev.00/
JBS: https://bugs.openjdk.java.net/browse/JDK-8219878

I have another crypto benchmark ready for review. This time it's Poly1305, which is a symmetric authenticator that is typically combined with ChaCha20 to produce authenticated encryption. I tested this benchmark with a slightly modified JVM containing a multiplication optimization[1]. Without this optimization, the performance of this vectorized implementation on AVX2 is not much better than the scalar implementation of Poly1305 in the JDK.

The webrev also includes some changes to the ChaCha20 benchmark which were necessary to get it to compile after some Vector API changes.


[1] https://bugs.openjdk.java.net/browse/JDK-8219881