On Friday, 29 October 2021 at 15:33:57 UTC, dd wrote:
On Friday, 29 October 2021 at 15:18:39 UTC, WebFreak001 wrote:
Would be cool having benchmarks comparing this to other
implementations if great performance is a goal, as you already
mentioned in your README it's much faster than keccack-tiny.
From memory, I remember getting ~10MiB/s when I first
implemented keccak-tiny from C to D. When I rewrote the package
to my liking, and I got ~130MiB/s. But these tests were in a VM
on my i7-3770 so it's not really a conclusive thing.
The readme mentions compilation with `-b release-nobounds`.
1. These are flags for DUB (?), can you add/replace it with the
conversion to actual compiler flags?
2. Did you test without removing bounds checks? If there is a
noticable performance drop, I think it'd be very worthwhile to
modify the code such that there is no penalty of boundschecks (at
-O3); I think in theory only very few bounds checks are needed.
Enabling good performance _with_ boundschecks on means that the
user is not forced to build his whole project without
boundschecks, or move to separate compilation.
cheers,
Johan
