Re: [PATCH v2] [RFC] Proposal for implementing AArch64 port of libmvec

[Szabolcs Nagy <szabolcs.nagy@arm.com>]

The 02/13/2023 16:46, Joe Ramsay via Libc-alpha wrote:
> Any comments/thoughts much appreciated! In particular, the patch
> raises the minimum GCC to 10, in order to be able to submit routines
> written using ACLE instead of assembly. This is clearly a big jump,
> but we have options if this is not acceptable. One option would be to
> submit compiler-generated assembly, similar to the equivalent routines
> under sysdeps/x86_64. If GCC 9 is an acceptable compromise then this
> would only have to be for SVE routines.

1) making the min required gcc version gcc-10 is a big jump, but
probably the cleanest solution.

2) including asm code for old gcc adds complications (e.g. issues
with configuration dependent instrumentation such as pac/bti) for
a rarely tested case.

3) optionally disabling libmvec for old gcc is doable but then the
declarations in math.h should be avoided, so this affects API.

we want users to build with >= gcc-10, but i think this can be
disruptive for some glibc developers (builds running in older
distro environments), so i think 3) may be a reasonable trade-off:
require >= gcc-10 with --enable-mathvec, allow old gcc otherwise.

--disable-mathvec is not expected to be a deployed configuration
that we support in the long term so if we run into issues with it
then we can drop it moving to 1).