If I understand parts of #2787 correctly, one problem is related to the use of diam. I've looked at the generated code and see:
#define diam *_ppvar[0].get<double*>()
which would cause each access to dereference the generic_data_handle, which we've seen in #2886 to cause performance issues.
My understanding is that diam is a dparam, and therefore semantically one double* per instance. Slight quirk, I think it's into the MORPHOLOGY "mechanism" which is special. However, it's still a double* per instance and therefore it should be part of neuron::cache::Model.
The purpose of this issue is simply to extract a solvable chunk of #2787.
If I understand parts of #2787 correctly, one problem is related to the use of
diam. I've looked at the generated code and see:which would cause each access to dereference the
generic_data_handle, which we've seen in #2886 to cause performance issues.My understanding is that
diamis a dparam, and therefore semantically onedouble*per instance. Slight quirk, I think it's into the MORPHOLOGY "mechanism" which is special. However, it's still adouble*per instance and therefore it should be part ofneuron::cache::Model.The purpose of this issue is simply to extract a solvable chunk of #2787.