Should we also optimize ECB mode for parallel processing?
I've noticed that users might make uneducated decision to use ECB for benchmarking different libraries against each other, even though ECB may be left intentionally unoptimized for parallel processing (like in libgcrypt now). Optimization in ECB would allow it to better mirror performance expected from modern AEAD modes (eg OCB).
| rC libgcrypt | |||
| rC8a1fe5f78f9f serpent: accelerate XTS and ECB modes | |||
| rCd078a928f5c6 twofish: accelerate XTS and ECB modes | |||
| rC14f39993d632 sm4: accelerate ECB (for benchmarking) | |||
| rC6475d0915ffe camellia: accelerate ECB (for benchmarking) | |||
| rC84f3d41acb23 rijndael: add ECB acceleration (for benchmarking purposes) | |||
It's not that needed, in my opinion, as nobody actually uses ECB itself (in real use case). But I understand the point of (possibly, students') benchmarking.
So, I don't have any objection to optimize ECB. Please go ahead, when/if you have time.