vosen/ZLUDA ? reverse-engineered prompt

I want to build a tool that can run applications designed for NVIDIA GPUs on other graphics cards, like those from AMD or Intel. A lot of scientific computing and machine learning software is locked into NVIDIA's CUDA platform, and I want to change that.

This project, let's call it ZLUDA, should work as a drop-in compatibility layer. The whole point is that users shouldn't have to change or recompile the original applications they want to run. They'd just install this tool, and the CUDA programs would work on their non-NVIDIA hardware. Performance is key, so it should be fast enough that it feels like running on a native NVIDIA card. It's basically a translation layer that makes CUDA apps work on different GPUs without the user even noticing.