DTolm/VkFFT

Allow complex transforms to use non-interleaved memory layouts (separate I and Q buffers)

Opened this issue · 3 comments

Some of the tooling I'm working with expects separate i[] and q[] buffers for complex valued data. It would be nice to support this natively in vkFFT rather than having to add a transpose shader before/after calling vkFFT.

DTolm commented

Hello,

this should not be difficult to add, I will try to implement this during the next develop cycle.

Best regards,
Dmitrii

Hello,

I'm also interested in this feature.

Hello,

I have added initial support for this feature - need to update the documentation, add more samples and test HIP, Level Zero and Metal APIs before the release, but Vulkan, CUDA and OpenCL should be working. You can see how it works in tests 0, 11 and 50.

Best regards,
Dmitrii