The problem of the execution of the mod file(the meaning of the Hoc exp & understanding execution workflow/call graph)

[Kim1230]

Hello, I have a question of execution of mod file. I used neuron with coreneuron to transform the mod file to the C++ file. And I can execute it and try to generate the dependency graph of the execution for static analysis. And when I look into the nrn_current_xxx function. such as :

And the first instruction will be the exp() which is defined as:
#undef exp #define exp hoc_Exp
What is the meaning and execution of this hoc_Exp?

What I think is that it will call nrn_state_xxx/exp：

And it still has the exp().

My question is that ：

1. the first function(nrn_current_xxx) that uses the exp which is not defined. So it will call the second function and uses the exp() in second function ?
2. the second function(nrn_state_xxx) that uses the exp which has been defined. So it will directly implement the naive exp() function?

Hope for your answer. Thanks!

[nrnhines]

double hoc_Exp(double x) is a wrapper for double exp(double x) defined in nrn/src/oc/math.cpp that returns 0 if x < 700 and raises an out of range error if x > 700 (instead of returning Inf)

In both your functions, hoc_Exp is the function actually being called.

[pramodk]

and I can execute it and try to generate the dependency graph of the execution for static analysis.

@Kim1230: many years ago we/I looked into execution workflow and generating call graphs. I attached some examples here. (note that these are from old neuron code)

In case it's helpful, I am happy to have a call and discuss your goals and what would be possible approaches.

[Kim1230]

@pramodk Hello, my main work is try to use mixed precision for this application. Because I found the spike deliver is based on the threshold with v and presyn. I think that this is a classification problem and use float and half-precision may be optimized.
I try to use float and half-precision to replace double variables. And I got some results but I don't know how to evaluate this result(the spike result under float or half-precision situation.) Do you know the benchmark to evaluate the spike result? Hope for your help. Thanks!

[pramodk]

And I got some results but I don't know how to evaluate this result(the spike result under float or half-precision situation.) Do you know the benchmark to evaluate the spike result?

Difficult for me to answer. But you can see people using different methods like peri-stimulus time histogram (PSTH), raster plot, inter-spike interval histogram (ISIH) etc (section 2.1. Comparing Common Tools in this paper)