The original post also linked to some very interesting documentation on the PTX IR that they will be targeting. I haven't personally seen this before but it helps understand a lot of fuzzy concepts that only seemed to exist in the docs of higher-level C++ API:
I hope we can end up with something more generic that just a good CUDA frontend (ie: ROCm, SIMD), but obviously that's a step even further.
I'd recommend skipping this somewhat fluffy blog post and just going to the source post: https://rust-gpu.github.io/blog/2025/01/27/rust-cuda-reboot/...
The original post also linked to some very interesting documentation on the PTX IR that they will be targeting. I haven't personally seen this before but it helps understand a lot of fuzzy concepts that only seemed to exist in the docs of higher-level C++ API:
https://docs.nvidia.com/cuda/parallel-thread-execution/