How good is code sharing between CPU and GPU?

[Telokis]

Hello!

I'm very interested in rust-gpu and was wondering if it could be used to make Rust code that can either run on GPU or CPU.
The concrete use-case I'm thinking of is a basic raytracer that defines a universal render function that compiles as a shader and as a CPU-callable function.
It could be used to generate code that can run on non-GPU devices like a simple headless server or on-GPU like a modern battlestation.

Is it something that can already be done? Is there any docs/examples regarding this? Does it even make sense in practice? I'm a beginner to GPU computing but I was mainly considering using compute shaders which don't sound too different from normal CPU computations except they're heavy parallel.

Thanks in advance!

[LegNeato]

You can see this section and the code for this blog post:

https://rust-gpu.github.io/blog/optimizing-matmul/#running-and-debugging-shaders-on-the-cpu

As it runs the kernels on CPU for testing and GPU for "production".

You can also check out these projects, all of which run on both GPU and CPU. Sometimes just partly for debugging, sometimes the full thing does:

• https://github.com/schell/renderling
• https://github.com/charles-r-earp/krnl
• https://github.com/Firestar99/nanite-at-home

[Telokis]

Thank you very much, the word I couldn't manage to remember was "isomorphic" haha!
This is perfect and the examples are great!