[Request] to Bring zluda project (github) into Mesa Project for CUDA-On-Iris support
ZLUDA exists to run CUDA code on Intel GPUs
Last release was in 2021. Project owner (Intel employee) had to step down, and made a plea for a new maintainer around then. Project has not had much attention with the pandemic and then noise re: HIP/ROCM/OpenSYCL, ComputeCPP, SyCL, etc.
Requesting from user perspective:
- Someone fork the project, then move it into Mesa
- Ideally CUDA on Gen9+, Xe and Arc will have all features
- Gen7.5, Gen 8 GPUs limited support, or older CUDA API support.
Mesa has OpenCL support of varying degrees, with rusticl bringing a new OpenCL compute method. Open-sourced CUDA-->iris (future: amdgpu? What abut even nouveau in the future for accessing CUDA compute features without having CUDA? Or on the Apple end, running CUDA on Asahi Linux via the Apple GPU or GPU+Tensor Cores/Neural Engine.
There are a lot of possibilities in terms of bringing this in as a mesa module or mesa driver: more so than just running Geekbench.