On Wed, Feb 23, 2022 at 11:11 AM Amos Jeffries squid3@treenet.co.nz wrote:
On 23/02/22 13:28, Maamoun TK wrote:
Benchmark the implementations on POWER9 and Z15
*---------------------------------------------------------------------------------------------*
| Arch | Nettle patch (radix 2^44) | OpenSSL (radix 2^26) |
|-------------------|--------------------------------------|-----------------------------------|
| PowerPC | 0.972 cycles/byte | 1.453 cycles/byte |
|-------------------|--------------------------------------|-----------------------------------|
| IBMz | 0.840 cycles/byte | 0.936 cycles/byte |
*---------------------------------------------------------------------------------------------*
What would be more useful here is a column against the non-patched Nettle speeds. It is sure nice to see nettle being faster than OpenSSL, but what if this proposed change is actually a degrade in current nettle speeds on those machines?
I've attached a benchmark numbers of these patches against nettle mainstream that reference C implementation in the merge request descriptions. There's no drawback here whatsoever.
regards, Mamone
Amos _______________________________________________ nettle-bugs mailing list nettle-bugs@lists.lysator.liu.se http://lists.lysator.liu.se/mailman/listinfo/nettle-bugs