Hello, I am the creator of VkFFT library(
https://github.com/DTolm/VkFFT). I was recently invited to be a speaker at the EUROfusion Webinar #5 about Vulkan Compute. I would like to share the recording of it with the community:
https://www.youtube.com/watch?v=YD_QO5Ilz2U. I have discussed the following topics there:
-Compute and memory-bound problems. Importance of proper memory management and why it is important to have coalesced global memory accesses and avoid shared memory bank conflicts.
-VkFFT, Vulkan Spirit and memory layout optimizations implemented there
-There was a coding session in the end where I showed a simple out-of-place transposition routine in Vulkan. It was launched on Nvidia GTX 1660Ti, Intel UHD 610 and AMD Radeon Vega RX 10 to illustrate the differences in bandwidth and memory coalescing of the different architectures. Full code with comments can be found here: https://github.com/DTolm/VulkanComputeSamples-Transposition
Hope this can be interesting to people willing to try Vulkan for computations!
P.S. I have added double and half precision support (with precision verification) to VkFFT and a choice to perform FFTs using lookup tables. VkFFT now also has a command line interface and it is possible to build cuFFT benchmark and launch it right after VkFFT one.
