Currently we are seeing square residuals compared to base GalSim images when rednering with no transforms applied. Additionally, perfromance seems to be suffering a lot compared to GalSim even when no transforms are applied.
The issue is likely related to how convolution is being done in the C++ layer and possibly how downsampling is being handled in Fourier space post-convolution also.
I am investigating potential fixes in a new branch debug-render, I will push an in-progress investigation of this branch soon.
Currently we are seeing square residuals compared to base GalSim images when rednering with no transforms applied. Additionally, perfromance seems to be suffering a lot compared to GalSim even when no transforms are applied.
The issue is likely related to how convolution is being done in the C++ layer and possibly how downsampling is being handled in Fourier space post-convolution also.
I am investigating potential fixes in a new branch debug-render, I will push an in-progress investigation of this branch soon.