platform support for arm64 on Linux (expand to surface as well)

[cjy1708]

Feature Description

I have a Jetson Orin Nano 8GB version (it is an aarch64 architecture), and I want to run the entire FastSurfer workflow on it. Could you please provide an installation tutorial?

Motivation

I am frustrated with FreeSurfer, which can only perform calculations using the CPU, because the Orin Nano can provide 67 TOPS of AI computing power, and the Nano only has six cores, with weak CPU performance. If it had powerful GPU support, I think the Nano's computing power would also be formidable.

Additional Context

This is the result of lscpu.

Architecture:            aarch64
  CPU op-mode(s):        32-bit, 64-bit
  Byte Order:            Little Endian
CPU(s):                  6
  On-line CPU(s) list:   0-5
Vendor ID:               ARM
  Model name:            Cortex-A78AE
    Model:               1
    Thread(s) per core:  1
    Core(s) per cluster: 3
    Socket(s):           -
    Cluster(s):          2
    Stepping:            r0p1
    CPU max MHz:         1728.0000
    CPU min MHz:         115.2000
    BogoMIPS:            62.50
    Flags:               fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm lrcpc dcpop asimddp uscat ilrcpc flagm paca pacg
Caches (sum of all):     
  L1d:                   384 KiB (6 instances)
  L1i:                   384 KiB (6 instances)
  L2:                    1.5 MiB (6 instances)
  L3:                    4 MiB (2 instances)
NUMA:                    
  NUMA node(s):          1
  NUMA node0 CPU(s):     0-5
Vulnerabilities:         
  Gather data sampling:  Not affected
  Itlb multihit:         Not affected
  L1tf:                  Not affected
  Mds:                   Not affected
  Meltdown:              Not affected
  Mmio stale data:       Not affected
  Retbleed:              Not affected
  Spec rstack overflow:  Not affected
  Spec store bypass:     Mitigation; Speculative Store Bypass disabled via prctl
  Spectre v1:            Mitigation; __user pointer sanitization
  Spectre v2:            Mitigation; CSV2, but not BHB
  Srbds:                 Not affected
  Tsx async abort:       Not affected

[m-reuter]

Our docker/singlarity images are not yet arm (mainly due to the freesurfer components for the surface stream).
You could try a native install, download all python dependencies and only running the segmentation component (although I am not sure any longer if that also uses some of FreeSurfer for eTIV estimation, maybe that part needs to be explicitly skipped).
The various Segmentation Networks should work. We only tested a similar setup on the Mac M-chips. Take a look at our install documentation and see how far you get. You are definitely outside our support range. But happy to help with concrete questions here.

[dkuegler]

To clarify the ARM support in the segmentation component.
FastSurfer only uses FreeSurfer components (and thus components with non-ARM binaries) under very specific circumstances. By default, no x86 binaries are needed in --seg_only mode.

Exceptions (features not supported for segmentation)

• you provide a T2 image that needs to be registered (via --t2 <file> flag, only affects hypothalamus subsegmentation as of FastSurfer up to 2.4.2)
• you need eTIV estimates (via --tag_reg flag, only required to include the etiv estimates and talairach registration files)

Future plans

This is the first time we have had a request for support for ARM on non-Mac platforms. It is therefore unclear whether there is a bigger community demand for this. You would be very welcome to 1) provide a PR to build a docker container for arm, 2) provide documentation on how to install on ARM (probably add here ) or 3) even just add your experience in either native install, ARM docker image and both segmentation and surface pipeline usage.

We would love to support ARM more broadly as well, but our development capacity is limited and (like mentioned) this is the first request for support -- and support is non-trivial.