[1]ZLUDA as the open-source project working on CUDA for non-NVIDIA GPUs like AMD Radeon graphics cards now has implemented support for some of NVIDIA's core cuFFT APIs.

Some of the core cuFFT APIs are now implemented by ZLUDA for this NVIDIA GPU-accelerated Fast Fourier Transform library. Open-source developer "AmosKito1" worked on the core cuFFT APIs where there is a direct equivalent in hipFFT, AMD's alternative FFT library as part of their ROCm/HIP ecosystem.

Per [2]the pull request , this dynamic translation from NVIDIA cuFFT to AMD hipFFT was enough to get the stock Folding@Home Core24 and Core27 cores for CUDA working on AMD hardware. Not all of the cuFFT APIs are supported by ZLUDA but only where there is a direct hipFFT equivalent interface.

This support is found in today's [3]ZLUDA 7-preview.11 release .



[1] https://www.phoronix.com/search/ZLUDA

[2] https://github.com/vosen/ZLUDA/pull/672

[3] https://github.com/vosen/ZLUDA/releases/tag/v7-preview.11