On Thu, 01 Aug 2019 09:04:11 +0300, Ard Biesheuvel said:

> The fact that crypto_aegis128_have_simd() does get optimized away, but
> crypto_aegis128_update_simd() doesn't (which is only called directly
> and not via a function pointer like the other two routines) makes me
> suspicious that this is some pathology in the compiler. Is this a
> distro build of gcc? Also, which architecture are you compiling for?

It's the Fedora Rawhide build on x86_64.

 [/usr/src/linux-next] gcc --version
gcc (GCC) 9.1.1 20190605 (Red Hat 9.1.1-2)
Copyright (C) 2019 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.