Elastic constants about dftk.jl HOT 7 OPEN

Actually I'm not sure if it's the precompilation or just the computation, should investigate (profile).

from dftk.jl.

Thanks for the basic setup ! I will play a bit with it. I'll let you know as soon as I've been able to do the profiling.

from dftk.jl.

From the profiling a lot of the time seems to be devoted to the computation of the Hessian involved in the ForwardDiff rule applying to the self_consistent_field function. But I'm not sure that I'm reading the graph correctly.

from dftk.jl.

Can you maybe post a screenshot? If that's the case it's actually computation and not compilation and we should probably optimize it.

from dftk.jl.

Cool, looks similar to what I did for the lattice constant sensitivities.

We can probably simplify a bit with proper update constructors or set lattice functions for planewavebasis, which is needed for lattice optimisations as well.

Regarding performance: The tolerances in solve omega+k are not yet tuned very well and parallelisation could be improved, but maybe there are more obvious things, too.

from dftk.jl.

Sorry for the delay ! I'm not sure a screen of the flame graph is readable but here is the file generated by ProfileView. Apparently you can just do "ProfileView.view(nothing)" and then open the file for an interactive flame graph !
HF_elastic_constant_pofiling.zip

from dftk.jl.

Nothing weird here, just lots of ffts and matmuls, so yeah looks like it needs some finetuning

from dftk.jl.