Was having another issue with sum_phase_closures which was related to the numpy-array-in-a-dictionary design pattern we're using for inter-process communication - this simply doesn't work with big stacks of interferograms because if the limit on maximum size of a pickle, so I replaced it with a preallocated buffer and a call to Gatherv, along with the associated calculations of buffer sizes and stuff for each process.
Was having another issue with
sum_phase_closureswhich was related to the numpy-array-in-a-dictionary design pattern we're using for inter-process communication - this simply doesn't work with big stacks of interferograms because if the limit on maximum size of a pickle, so I replaced it with a preallocated buffer and a call toGatherv, along with the associated calculations of buffer sizes and stuff for each process.