On Wed, Aug 12, 2020 at 08:25:20PM -0700, Richard Henderson wrote: > Mostly this is intended to cleanup the class hierarchy > used for the ciphers. We currently have multiple levels > of dispatch, and multiple separate allocations. The final > patches rearrange this to one level of indirect call, and > all memory allocated contiguously. > > But on the way there are a number of other misc cleanups. > > I know those final patches are somewhat big, but I don't > immediately see how to split them apart.
Yeah, I can't see a better way off hand. > I noticed this while profiling patches to make ARM PAUTH > use the crypto subsystem. The qcrypto_cipher_* dispatch > routines were consuming a noticeable portion of the runtime, > and with these changes they were down below 1% where they > ought to be. > > While I did not continue with PAUTH using AES, I still think > these are good cleanups. They'll probably improve the LUKS block driver performance too. What were you measuring performance with ? Did you use the benchmark-crypto-cipher program or something else ? Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
