SHA-3 Hash algorithm performance improvements (~12x speedup)

Performance of SHA3 (SUN Provider) Hash algorithm was checked.

As result we got poor performance. It's 50% slower than alternative BouncyCastle java SHA3 implementation.
Moreover, SHA3 implementation has shown very huge memory consumption. In average allocation rate is 400 bytes per byte of hashed mesage for SHA3-512 and 200 bytes per per byte of hashed mesage for SHA3-256.
So calculating digest for 1Mb original data we got:
Allocaton rate (SHA3-512) = 380 Mb/op (per means per a single digest invocation),
Allocaton rate (SHA3-256) = 202 Mb/op.

Here is performance results:

Benchmark (algorithm) (dataSize) Score Error Units
MessageDigestBench.digest SHA3-224 1048576 11.537 ± 0.059 ops/s
MessageDigestBench.digest SHA3-256 1048576 11.088 ± 0.085 ops/s
MessageDigestBench.digest SHA3-384 1048576 8.459 ± 0.079 ops/s
MessageDigestBench.digest SHA3-512 1048576 5.851 ± 0.059 ops/s

Suggested patch fully eliminates memory consumption. Allocation rate is ~871 bytes per digest invocation and doesn't depend on input data size.
SHA3 performance was impoved by ~6x times:

Benchmark (algorithm) (dataSize) Score Error Units
MessageDigestBench.digest SHA3-224 1048576 62.418 ± 0.319 ops/s
MessageDigestBench.digest SHA3-256 1048576 58.933 ± 0.575 ops/s
MessageDigestBench.digest SHA3-384 1048576 44.358 ± 4.114 ops/s
MessageDigestBench.digest SHA3-512 1048576 30.029 ± 3.492 ops/s