split part of the slave kernel into a device function
introduced tunable parameter 'use_kernel' to the master coherencies kernel
if use_kernel == 1 the master kernel calls the slave kernel
if use_kernel == 0 the master kernel uses a for-loop over K instead
the input data in the tuning script is random and as such there may still be small errors that we can't detect right now, testing with better data is needed.
the tuning script also assumes that all sources are point sources, if more complex sources are used we may run into diverge issues for which we can use more sophisticated methods (gpu-mulitisplit for example). Whether anything like that is needed depends on the diversity among sources and how they are sorted.
split part of the slave kernel into a device function introduced tunable parameter 'use_kernel' to the master coherencies kernel if use_kernel == 1 the master kernel calls the slave kernel if use_kernel == 0 the master kernel uses a for-loop over K instead
the input data in the tuning script is random and as such there may still be small errors that we can't detect right now, testing with better data is needed.
the tuning script also assumes that all sources are point sources, if more complex sources are used we may run into diverge issues for which we can use more sophisticated methods (gpu-mulitisplit for example). Whether anything like that is needed depends on the diversity among sources and how they are sorted.