# EESSI community meeting (shared notes)

## Wed 14 Sept 2022

### Roundtable discussion

- virtualized instances: impact on host CPU detection
- push responsibility of software installation partially to users

### glibc discussion after Caspar's talk
- Bart: I did some tests with glibc a few years ago. Clear Linux has a seperate branch for avx512, but in general the default is well optimized these days (since 2.29, olders lag in some transcendental functions such as exp/sin/etc)
(I couldn't find any benefit with the avx512 clear linux glibc binary)
Note that Clear Linux uses the hwcaps mechanism, where the dynamic loader looks in hwcaps/x86-64-v[432] subdirectories of the runtime library path entries.
You can also consider moving the Gentoo prefix baseline to x86-64-v3, systems without avx2 are getting rare (https://gitlab.archlinux.org/archlinux/rfcs/-/merge_requests/2/diffs)

## Thu 15 Sept 2022

### Compat layer
- we could potentially offer a thin and fat compat layer for the same version, where sites/users can choose which one to use (using a variable symlink)
- this will be much work to maintain, so let's keep it minimal for now
- no objections for removing ppc64le support
- Hugo is working on a bash script as replacement for the archspec CPU detection: https://github.com/EESSI/software-layer/pull/187
- Samuel: small remark: software that relies on osdependencies could benefit from those deps being available in the compat layer

### Software layer
- document how modules can be hidden/overridden, how different views can be generated
- be concise and clear about which CPUs we support and why