[REDMINE1D-189] [RM-8156] [linemodel templatefitting fft] study the minimal number of redshift points needed to activate fftprocessing

Wed, 5 Jul 2023 02:28:56 +0000

Created on 2023-06-29 09:00:52 by Didier Vibert. % Done: 0

The cpu time gain of fftprocessing for template fitting is depending on the number of redshfit samples: when this number is too small the gain may actually be a loss !

Thus a minimal number of z-points have been hard-coded for the templatefitting in the linemodel, (which is mostly relevant for the linemodel second pass, where the z-points are only those of the second pass redshift window).
The hard-coding is here, in the code source:
https://gitlab.lam.fr/CPF/cpf-redshift/-/blob/0.42.0/RedshiftLibrary/src/lib/operator/linemodel.cpp#L236

The trick is that this minimal z-points number is also depending on the number of samples in the spectrum, thus it could be expressed as a % of the number of spectrum samples.

TODO:

set this minimal number as a parameter, expressed as a % of the number of spectral samples (use the log-sampled spectra, ie without holes)
study the performance impact for several values of the parameter and spectrum size, to deduce a typical value for this new param

PFS-JIRA

[REDMINE1D-189] [RM-8156] [linemodel templatefitting fft] study the minimal number of redshift points needed to activate fftprocessing