Re: [openpgp] OpenPGP encryption block modes

[Bruce Walzer <[e-mail]>]

On Thu, Aug 11, 2022 at 12:16:04PM +0200, Werner Koch wrote:

[...]

> I can throw in some other numbers from Libgcrypt 1.10.1 [0]

[...]

> On a AMD Ryzen 7 4800U NUC:
> 
>  AES256         |  nanosecs/byte   mebibytes/sec   cycles/byte  auto Mhz
>         CFB enc |     0.819 ns/B      1164 MiB/s      3.50 c/B      4269
>         CFB dec |     0.113 ns/B      8431 MiB/s     0.481 c/B      4256
>         GCM enc |     0.242 ns/B      3938 MiB/s      1.03 c/B      4260±1
>         GCM dec |     0.243 ns/B      3931 MiB/s      1.04 c/B      4268±3
>        GCM auth |     0.120 ns/B      7931 MiB/s     0.513 c/B      4264±1
>         OCB enc |     0.134 ns/B      7142 MiB/s     0.569 c/B      4259
>         OCB dec |     0.131 ns/B      7297 MiB/s     0.556 c/B      4258±1
>        OCB auth |     0.132 ns/B      7210 MiB/s     0.563 c/B      4259
>  SHA1           |     0.443 ns/B      2151 MiB/s      1.89 c/B      4262±1
>  SHA256         |     0.475 ns/B      2010 MiB/s      2.02 c/B      4265
>  SHA512         |      1.22 ns/B     783.6 MiB/s      5.17 c/B      4245±4
> 
> 

What am I doing wrong here? All in cycles per byte (c/B).

> I can't find our GnuPG benchmarks comparing CFB and OCB right now, but
> the above numbers should give some hints: For the Ryzen OCB encryption
> is roughly 15 times faster than CFB+SHA1.

(3.50 + 1.89) / 0.569 = 9.5 (different result)

> Decryption is 4 times faster.

(0.481 + 1.89) / 0.556 = 4.3 (same result)

Bruce