user managed staging memory buffer in transferDataFromCPU() and transferDataToCPU()

[anarkiwi]

Thanks so much for VkFFT!

I observe an overhead in these two functions, calling allocateBuffer() for staging memory and then free/destroying it with each call.

Would it be possible to have a variation, where the caller supplies their own staging buffer (that is then remapped as necessary within transferData())? I have experimented with this on Pi4 specifically, and it makes a significant difference allocating the staging memory buffer once on application startup.

Thanks,

[DTolm]

Hello, thanks for using VkFFT!

Sure, I will add an option to provide a user-defined staging buffer in VkGPU and in VkFFTApplication. However, transferDataFromCPU and transferDataToCPU are not part of the VkFFT - they are simple tools in the benchmark suite that were not meant to be used in any production codes, just as a reference to how simply transfer data.

Best regards, Dmitrii

[DTolm]

Hello,

I have added an option to provide the staging buffer. I haven't tested it thoroughly yet and haven't updated documentation about it, but it should be pretty straightforward (just provide the staging buffer and memory pointers to the configuration).

Best regards, Dmitrii

[anarkiwi]

That's really nice - thanks!

[DTolm]

It should be working now, if you have any other improvement ideas about the staging buffer - feel free to reopen the issue!