Open oerc0122 opened 7 months ago
What I've seen in the code was reminiscent of selection between split_vector_fixed_sum
and split_vector_max_sum || split_vector_max_sum_or_numel
. I think thes covers all reasonable cases but may be something more is there too and another procedure should be introduced. If it is the case, it should be separately tested procedure with clear scope.
Distribute currently uses its own implementation of
split_vector_fixed_sum
to account for the additional metadata, however, it may be possible to replace this, though it may require some re-working of the parallel merge procedures too.Places which may be affected: ./horace_core/sqw/@DnDBase/distribute.m ./horace_core/sqw/@sqw/distribute.m ./horace_core/sqw/PixelData/@PixelDataMemory/distribute.m ./horace_core/sqw/PixelData/@PixelDataFileBacked/distribute.m ./herbert_core/utilities/classes/@IX_dataset/distribute.m ./herbert_core/utilities/misc/distribute.m ./herbert_core/classes/MPIFramework/parallel_call/ParallelSQWEval.m ./herbert_core/applications/multifit/MFParallel_Job.m