ARM64 support

[LouisLeNezet]

Hi,

I would like to be able to run GLIMPSE2 on ARM64 architecture.
Unfortunately, the phase/src/models/imputation_hmm.h and phase/src/models/phasing_hmm.h include immintrin.h which is not compatible with arm64.
Would it be possible to find a fix ?

Best regards,
Louis

[tfenne]

@LouisLeNezet I have a branch where I've added support for mac/aarch64. It's working in my hands, but I have a little more work to do before I'm ready to submit a PR - I'm not affiliated with GLIMPSE or odelaneau, just someone who also wants this.

You're welcome to check it out: https://github.com/tfenne/GLIMPSE/tree/tf_support_aarch64

There are simple docs in the top-level README, and expanded docs on how to build in the docs/docs/installation subtree.

I'd be very open to any feedback!

[LouisLeNezet]

Thanks for adding support for this architecture, unfortunately I'm working with nf-core and therefore I would need GLIMPSE to have an official release that would support this change.
It seems there is already 2 PR proposing changes that would add support for it, so I will just wait for the authors to answer 😄

[srubinacci]

Hi,

Yes, this was something we had planned, but it got delayed due to relocations and academic life being not very supportive in allocating time on this. @rwk-unil has proposed a PR for this, and I’ve asked for an update on its status. We mainly just need to coordinate, as this should be relatively straightforward to add.

Thanks @tfenne for your work on this, it definitively helps to have an additional implementation. Let’s make sure everything integrates well, and we should be able to get it approved soon.

Best,
Simone

[rwk-unil]

Hello, in #237 I propose replacing the current Intel specific vector instructions with the SIMDe library. The PR is tested for x86 and ready to go (I don't have access to ARM hardware). This should allow to compile with ARM NEON vector instructions easily. However, I am not sure that SIMDe provides support for ARM64 SVE instructions.

The PR from @tfenne also relies on SIMDe for ARM64 support so both PR's rely on the same library.

Best,
Rick

[tfenne]

Yes - #286 follows the same approach of using SIMDe (as a submodule vs. include the headers directly). It also makes non-trivial changes to the makefiles to enable building on a variety of platforms (most linuxes and macos). I've tested the builds there on:

• x86_64 on ubuntu and fedora based linuxes
• aarch64 on ubuntu and fedora based linuxes
• aarch64 on macos

All produce working binaries; my motivation for doing this was to be able to build and debug on macos/aarch64 so I could more easily/interactively debug the problem I was having in #282, which I was then able to do, resulting in #285.

I would be more than happy to accept review feedback on #286, update that PR, and re-test across platforms.

[srubinacci]

The pull request has been merged. It might take a few weeks to put together a release, but hopefully it will be available soon.