Open franziskuskiefer opened 1 week ago
Can @mamonet investigate? Thanks for pinpointing this!
I addressed this issue in PR #955
For some reason, there is a regression in the generation of nice C types, and there is a question on the tradeoff between code size and performance, but the immediate perf issue is addressed.
The new SHA3 code from #897 (avx2 and scalar) lost about 30% performance just before merge. We have been using the version before merge for a while and noticed this performance regression on updating hacl-packages.