Benchmark on Spacemit X60

[smuellerDD]

I integrated your code into [1]. I have not pushed it yet, because I see interesting performance numbers.

I am trying basically to squeeze about 1GB of data from a SHAKE instance. For that I am using a system with a RISC-V Spacemit X60. This has the vector extensions, but not the Zvbb extensions.

I see the following numbers:

plain C implementation: 26s
Your keccak-vector.c after static constify the applicable variables: 2m17s
Same, but with -funroll-loops: 2m20s
Same, but with -funroll-all-loops: 2m30s
Your keccak-vector-multi-round.c: 3m40s

I have to admit, I am a bit surprised by the numbers.

[1] https://github.com/smuellerDD/leancrypto