On Sat, May 09, 2020 at 05:30:18PM +0200, Joerg Sonnenberger wrote: > On Sat, May 09, 2020 at 03:25:25PM +0000, nia wrote: > > On Sat, May 09, 2020 at 01:13:44PM +0000, nia wrote: > > > Right, yes, the CPUID stuff wasn't being defined when builing EVP which > > > meant the dumb asm implementation was being used instead of HW/VP/BSAES. > > > > > > That's fixed now. Everything else appears to still be slightly slower, > > > but that's less critical than the AES implementation. > > > > ... and SHA2 is slower because we're patching OpenSSL to use the libc > > implementation, which is pure C. Any reason for this? > > ABI differences.
We could -- probably should -- arrange for OpenSSL to use its own implementation when called via EVP. Thor
