RFR: 8333867: SHA3 performance can be improved

[Vladimir Kozlov]

On Tue, 11 Jun 2024 09:55:11 GMT, Ferenc Rakoczi <duke at openjdk.org> wrote:

>> This PR removes some unnecessary conversions between byte arrays and long arrays during SHA3 digest computations.
>
> Some microbenchmark data (the percentages are improvements in ops/sec):
> Benchmark                                                                                                                                                                       Linux aarch64  Linux x64  MacOSX aarch64  MacOSX x64
> openjdk.bench.java.security.MessageDigests.digest-digesterName:SHA3_256-length:16384-provider:DEFAULT         27.67%     20.60%          -0.00%      24.75%
> openjdk.bench.java.security.MessageDigests.digest-digesterName:SHA3_256-length:64-provider:DEFAULT            11.95%      6.27%          -3.97%      12.72%
> openjdk.bench.java.security.MessageDigests.digest-digesterName:SHA3_512-length:16384-provider:DEFAULT         18.00%     14.99%           0.01%      15.35%
> openjdk.bench.java.security.MessageDigests.digest-digesterName:SHA3_512-length:64-provider:DEFAULT            11.91%      5.87%          -3.39%      10.04%

@ferakocz what testing in Mach5 you ran? You need at least to run first 3 tiers.

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

PR Comment: https://git.openjdk.org/jdk/pull/19632#issuecomment-2161300019