How to compile Julia on cluster with slightly different login / compute nodes?

I’m on a cluster where I think the login / compute nodes are sufficiently different architectures that compiling Julia (1.3.0-rc4) on one and running on the either always yields a ERROR: Unable to find compatible target in system image. The prebuilt Julia executables from the website work fine, however. If I do want to compile my own versions though, can someone suggest what compile flags I might need? For reference, I attach a diff of the /proc/cpuinfo between the login/compute nodes. Thanks.

Not an expert but have you looked at https://github.com/JuliaLang/julia/blob/master/doc/build/linux.md#architecture-customization ?

1 Like

Are all of the compute nodes the same architecture?

If so, why not just compile Julia once on the compute nodes, and save the results. And then use that Julia every time you are running on the compute nodes.

Thanks. Based on that plus blindly taking a hint from Use PackageCompiler to produce EXE on non-VM, execute on VM gives "ERROR: Unable to find compatible target in system image" - #2 by nodrygo it seems to be working with MARCH=core2 in my Make.user. Not sure if there’s a better choice, will see if I see any noticeable slowdowns…