Nmodl 1247 #20
Nmodl 1247 #20
Conversation
| static inline void* mem_alloc(size_t num, size_t size, size_t alignment = 16) { | ||
| void* ptr; | ||
| posix_memalign(&ptr, alignment, num*size); | ||
| memset(ptr, 0, size); |
There was a problem hiding this comment.
We allocated num*size bytes and are zeroing size bytes. Omar might have a PR for this.
table/neuron/table.cpp
Outdated
| auto* const _ml = &_lmr; | ||
| auto* _thread = _ml_arg->_thread; | ||
| for (int id = 0; id < nodecount; id++) { | ||
|
|
There was a problem hiding this comment.
We'll need to find what printing this empty line.
table/neuron/table.cpp
Outdated
| } | ||
|
|
||
|
|
||
| void check_sigmoid1_tbl(_nrn_mechanism_cache_range* _ml, tbl_Instance& inst, size_t id, Datum* _ppvar, Datum* _thread, NrnThread* _nt) { |
There was a problem hiding this comment.
This function performs a lazy update, not a check. Are we free to change the name?
There was a problem hiding this comment.
I think it can be whatever, but due to merging BlueBrain/nmodl#1251 the changes will also have a minor impact on coreNEURON codegen.
There was a problem hiding this comment.
Note that in NEURON there's the nrn_thread_table_check_t object (link), so I just went with that convention as coreNEURON codegen uses the same naming.
There was a problem hiding this comment.
Okay, however, please create a ticket. It can include all the other naming issues pointed out on nmodl-references. This is all extremely simple (linear interpolation) but not easy to read.
table/neuron/table.cpp
Outdated
| make_table = false; | ||
| inst.global->tmin_sigmoid1 = -100.0; | ||
| double tmax = 100.0; | ||
| double dx = (tmax-inst.global->tmin_sigmoid1) / 100.; |
There was a problem hiding this comment.
This one place where one can quite nicely see that the interpolation variable (a voltage) doesn't have a name. I've seen, t, x, \xi and theta. While also using t_ as a prefix signifying "tabulated".
| inst.sig[id] = inst.global->t_sig[index]; | ||
| return 0; | ||
| } | ||
| int i = int(xi); |
There was a problem hiding this comment.
Is size_t the preferred integer type for this use?
| return 0; | ||
| } | ||
| double xi = inst.global->mfac_sigmoid1 * (v - inst.global->tmin_sigmoid1); | ||
| if (isnan(xi)) { |
There was a problem hiding this comment.
This is an odd check, because we can know if mfac or tmin are nan much earlier (during codegen) and I can't remember us asserting that v is valid in any other context.
|
|
||
|
|
||
| inline int sigmoid1_tbl(_nrn_mechanism_cache_range* _ml, tbl_Instance& inst, size_t id, Datum* _ppvar, Datum* _thread, NrnThread* _nt, double v){ | ||
| if (inst.global->usetable == 0) { |
There was a problem hiding this comment.
This puts a lot of faith in the optimizer. If it can't see that inst.global->usetable is likely a constant, it can't even attempt to lift this branch outside of a loop. Which I suspect will cause it to fail to vectorize.
|
Many of the comments can be addressed by creating issues in BlueBrain/nmodl. |
No description provided.