search

cgarling / StarFormationHistories.jl

Fitting astrophysical star formation histories via CMD modelling.
MIT License
4 stars 0 forks source link

@turbo threading #10

Closed cgarling closed 1 year ago

cgarling commented 1 year ago

It seems we get reasonable scaling out of setting thread=true for LoopVectorization.@turbo in the basic functions loglikelihood, ∇loglikelihood, and ∇loglikelihood! but it currently breaks the parallel hmc_sample method when used with matplotlib due to some GIL error on the python side.

cgarling commented 1 year ago

This is annoying enough that I'm disabling threading on these functions. Keeping the issue open in case we want to revisit in the future.

cgarling commented 1 year ago

Now looking like we probably don't want to thread these functions anyway. We can thread at a higher level (e.g., by using sampling methods).