[Bug] performZeropadding leads to incorrect results in 2-D

[Vandermode]

As the title suggested, I compare the results of vkFFT and numpy/scipy FFT under C2C convolution with zero padding setting (Note I used the normal FFT + elementwise multiplication + IFFT, rather than performConvolution mode)

The results are shown below, it can be seen vkFFT works as expectation in 1D but fails completely in 2-D case.

Likely a potential bug, since it works in 1-D case, though I am not sure if my application in 2-D is set correctly.

My test code is available in
https://github.com/Vandermode/pyvkfft/blob/3d054f27272c2529a8d45798709a2ae544bdfb8c/examples/test_fft_conv.py

(need to install my adapted version of pyvkfft to support performZeropadding mode in python

[DTolm]

Hello,

I have identified the issue with the 2D zeropadding, your test now passes on my local machine.

Thank you for reporting it!

Best regards,
Dmitrii

[Vandermode]

Hi, I tested it again, and it works, thanks!