2-D Convolution: Bug in FFT?

I am trying to convolute one of the dimensions of a 2-dimensional pdf in RooFit (as per page 41 of the RooFit users’ manual).

It works fine (although very slow) when using numerical convolution through the RooNumConvPdf class.

However, when I try it using the FFT convolution scheme via the RooFFTConvPdf class, I get strange results: The convolution variable (x in the macro) is handled correctly, but the non-convoluted variable (y in the macro) ends up with a flat, uniform distribution.

Is there any way to preserve the distribution in the non-convoluted variable?
TwoDimConv.C (1.43 KB)