search

cgarling / StarFormationHistories.jl

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

BLAS threads for Optim.jl #5

Closed cgarling closed 1 year ago

cgarling commented 1 year ago

Optimizations using Optim.jl's BFGS implementation seem to be spending a lot of time in BLAS calls. By default BLAS has multi-threading but for problems of our sizes this threading is not efficient. Worth profiling and examining more closely.

cgarling commented 1 year ago

Not having much luck with the profiler at the moment; looks like some bug with graphics over wsl. Tried with MKL and timing results were the same; performance is slightly better when the number of BLAS threads is 1 as set via LinearAlgebra.BLAS.set_num_threads(1) than when we have 8 BLAS threads. I don't imagine it's a big issue for many people (my solve goes from 2.8 -> 2.2 seconds) but worth remembering.

cgarling commented 1 year ago

I made a note in the documentation and in the example Jupyter notebook; think that's all I'll do for now.