single vs. double precision in runtime
@jpblaschke What is the default precision? Everything in extern seems to be set to double precision.
Specifically, when I run the legion code with -f enabled, it's printing out Using v2 CUDA to solve the NUFFT 3D T2 on device 0, and I get TypeError: cufinufft plan.dtype and kx dtypes do not match. so I change everything in the legion code to double precision manually.
Perhaps this why in the toml file the default use_single_prec = false?
I wonder if this is already being taken care of in settings so that the precision can be controlled in a consistent way without doing manually? I don't seem to see it in runtime though.