multithreaded data_prep
caseyjlaw opened this issue · comments
Currently the data_prep function takes state and data and then applies calibration and flagging in place. This uses only one thread, which slows it considerably.
Calibration has been made into a functional form, so it can be scheduled separately now. Perhaps with separate threads working on separate subbands.
Flagging is also still dependent on rtpipe, single-threaded, and quite slow.
Flagging and calibration are now much faster via numba.
Also added flagging test suite with simulated and FRB data.