It does! Both platform-specific compute shaders as well as cross-compilation to CUDA. The authors even provide some basic PyTorch bindings to help use existing

It does! Both platform-specific compute shaders as well as cross-compilation to CUDA. The authors even provide some basic PyTorch bindings to help use existing shader code for gradient computation and backpropagation in ML and differentiable programming of graphics-adjacent tasks: https://github.com/shader-slang/slang-torch https://github.com/shader-slang/slang-torch (Disclaimer: this is the work of my colleagues, and I helped test-drive differentiable Slang and wrote one of the example applications/use-cases)