Does Mojo support cross-compilation for GPU targets on non-GPU platforms?

Mojo
,
1

My current local machine doesn’t meet Mojo’s GPU runtime requirements, and I’m currently unable to compile on it. Given that most modern programming languages and compilers support cross-compilation, I was wondering: does the Mojo compiler currently support this feature? If not, is it on the roadmap for the future?

Right now, I’m relying on cloud GPUs to learn Mojo. However, keeping a cloud instance running continuously is quite expensive, and having to repeatedly spin up and reconfigure the environment from scratch every time. If I could compile the executable locally and only use the cloud instance for execution, it would significantly improve the workflow and save costs.

2

Hey! Yes there are flags for specifying the target you’re building for. Hopefully this page helps: Compilation targets | Mojo

3

thx! I will try it!!!

4

Yes, it is very useful! I can really compile the mojo file on my computer. But when I scp the executable to the remote GPU target and run it, it will complain about can't load shared libraries: libKGENCompilerRTShared.so. I guess I need to compile locally by using -Xlinker -static, but I don’t have static libraries like libKGENCompilerRTShare.a.

So the question has changed interestingly. How to statically link a mojo executable?

5

I can run it by copying all needed shared libraries to the remote GPU target, and exporting LD_LIBRARY_PATH before running it.

6

If you had mojo or max installed on your remote machine you should also have a copy of the library there too? But glad to see you got it working.