Releases: vosen/ZLUDA
Releases · vosen/ZLUDA
Version 3
Nobody expects the Red Team
Too many changes to list, but broadly:
- Remove Intel GPU support from the compiler
- Add AMD GPU support to the compiler
- Remove Intel GPU host code
- Add AMD GPU host code
- More device instructions. From 40 to 68
- More host functions. From 48 to 184
- Add proof of concept implementation of OptiX framework
- Add minimal support of cuDNN, cuBLAS, cuSPARSE, cuFFT, NCCL, NVML
- Improve ZLUDA launcher for Windows
Version 2
The goal of version 2 has been to fix end to end execution of GeekBench and improve Windows support:
- Several new host-side functions are supported now (e.g. cuModuleLoadDataEx)
- Several bugs have been fixed on the kernel side (e.g. threadIdx/blockIdx is now handled correctly)
- Minor improvement in generated code brought better I/O performance when reading/writing vector objects. This improved performance by several percentage points in select GeekBench benchmarks
- ZLUDA now ships its own injector (with_zluda.exe) which should make running ZLUDA on Windows much easier
- Additionally, we have gained ability to easily create traces of CUDA kernel execution, making enabling new workloads much easier
- ZLUDA now has a CI, which produces binaries on every pull request and commit
Special thanks to @take-cheeze, @nilsmartel and @ritschwumm for contributing to this release