kernfel
commented
2 years ago So, the issue I was trying to address is the following. I have a moderately complex model where, by the inscrutable magic of brian2's naming schemes, in the second run, the NeuronGroup would be named as requested (e.g. neurons), but its subordinate objects (Thresholder, StateUpdater etc.) would have a numeric suffix (e.g. neurons_thresholder_1). Since process_neuron_groups is looking specifically for {NeuronGroup.name}_thresholder, it would not pick these suffixed objects up, and therefore not process them, nor add them to the code passed to GeNN.
So far, so good: I changed all remaining hardcoded names that I could find, reflected in the two commits above. Tests pass, job done, right?
Not so. Now that the models in the first and second run are almost completely identical, it seems that GeNN is not picking up the subtle differences in naming. To wit:
from brian2 import *
import brian2genn
import os
set_device('genn')
for i in range(2):
# os.system('rm -rf GeNNworkspace')
if i > 0:
device.reinit()
device.activate()
G = NeuronGroup(1, model='dv/dt = -v/(1*ms) : 1',
threshold='rand()>.5', method='exact', name='neurons')
S = Synapses(G, G, '', on_pre='v += 1')
S.connect()
spikemon = SpikeMonitor(G)
statemon = StateMonitor(G, variables='v', record=[0])
net = Network(G, S, spikemon, statemon)
net.run(10*ms)
print(spikemon.count, statemon.v[0, -1])Running this code as-is, the first run (usually) passes just fine, but the second one raises during the final model compilation step:
~ ...snip: make output... ~
g++ -std=c++11 -Wno-write-strings -I. -Ibrianlib/randomkit -w -O3 -ffast-math -fno-finite-math-only -march=native -std=c++11 main.cpp code_objects/statemonitor_1_codeobject.cpp objects.cpp code_objects/synapses_1_synapses_create_generator_codeobject.cpp code_objects/synapses_1_pre_push_spikes.cpp code_objects/spikemonitor_1_codeobject.cpp brianlib/randomkit/randomkit.cc -o main -Lmagicnetwork_model_CODE -lrunner -Wl,-rpath magicnetwork_model_CODE
~ ...snip: warnings... ~
In file included from main.cpp:19:
engine.cpp: In member function ‘void engine::run(double)’:
engine.cpp:45:19: error: ‘vneurongroup_1’ was not declared in this scope; did you mean ‘vneurongroup’?
45 | std::copy_n(vneurongroup_1, 1, brian::_array_neurongroup_1_v);
| ^~~~~~~~~~~~~~
| vneurongroup
engine.cpp:51:7: error: ‘pullneurongroup_1CurrentSpikesFromDevice’ was not declared in this scope; did you mean ‘pullneurongroupCurrentSpikesFromDevice’?
51 | pullneurongroup_1CurrentSpikesFromDevice();
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| pullneurongroupCurrentSpikesFromDevice
engine.cpp:52:7: error: ‘pullneurongroup_1StateFromDevice’ was not declared in this scope; did you mean ‘pullneurongroupStateFromDevice’?
52 | pullneurongroup_1StateFromDevice();
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| pullneurongroupStateFromDevice
main.cpp: In function ‘int main(int, char**)’:
main.cpp:73:30: error: ‘rowLengthsynapses_1’ was not declared in this scope; did you mean ‘rowLengthsynapses’?
73 | rowLengthsynapses_1, indsynapses_1, maxRowLengthsynapses_1,
| ^~~~~~~~~~~~~~~~~~~
| rowLengthsynapses
main.cpp:73:51: error: ‘indsynapses_1’ was not declared in this scope; did you mean ‘indsynapses’?
73 | rowLengthsynapses_1, indsynapses_1, maxRowLengthsynapses_1,
| ^~~~~~~~~~~~~
| indsynapses
main.cpp:73:66: error: ‘maxRowLengthsynapses_1’ was not declared in this scope; did you mean ‘maxRowLengthsynapses’?
73 | rowLengthsynapses_1, indsynapses_1, maxRowLengthsynapses_1,
| ^~~~~~~~~~~~~~~~~~~~~~
| maxRowLengthsynapses
main.cpp:79:49: error: ‘ineurongroup_1’ was not declared in this scope; did you mean ‘ineurongroup’?
79 | std::copy_n(brian::_array_neurongroup_1_i, 1, ineurongroup_1);
| ^~~~~~~~~~~~~~
| ineurongroup
main.cpp:80:49: error: ‘vneurongroup_1’ was not declared in this scope; did you mean ‘vneurongroup’?
80 | std::copy_n(brian::_array_neurongroup_1_v, 1, vneurongroup_1);
| ^~~~~~~~~~~~~~
| vneurongroup
main.cpp:87:7: error: ‘_seedneurongroup_1’ was not declared in this scope; did you mean ‘_seedneurongroup’?
87 | _seedneurongroup_1[i]= (uint64_t) (rand()*0x0000FFFFFFFFFFFFLL);
| ^~~~~~~~~~~~~~~~~~
| _seedneurongroup
code_objects/spikemonitor_1_codeobject.cpp: In function ‘void _run_spikemonitor_1_codeobject()’:
code_objects/spikemonitor_1_codeobject.cpp:108:24: error: ‘spikeCount_neurongroup_1’ was not declared in this scope; did you mean ‘spikeCount_neurongroup’?
108 | int32_t _num_events = spikeCount_neurongroup_1;
| ^~~~~~~~~~~~~~~~~~~~~~~~
| spikeCount_neurongroup
code_objects/spikemonitor_1_codeobject.cpp:117:23: error: ‘spike_neurongroup_1’ was not declared in this scope; did you mean ‘spike_neurongroup’?
117 | const int _idx = spike_neurongroup_1[_j];
| ^~~~~~~~~~~~~~~~~~~
| spike_neurongroup
make: *** [Makefile:10: main] Error 1
ERROR Brian 2 encountered an unexpected error. If you think this is a bug in Brian 2, please report this issue either to the discourse forum at <http://brian.discourse.group/>, or to the issue tracker at <https://github.com/brian-team/brian2/issues>. Please include this file with debug information in your report: /tmp/brian_debug_e2sgnrl0.log Additionally, you can also include a copy of the script that was run, available at: /tmp/brian_script_rfh3a_al.py You can also include a copy of the redirected std stream outputs, available at '/tmp/brian_stdout_l9pglhv0.log' and '/tmp/brian_stderr_yv5s018d.log'. Thanks! [brian2]
Traceback (most recent call last):
File "/home/felix/projects/lib/brian2genn/brian2genn/device.py", line 877, in build
self.compile_source(debug, directory, use_GPU)
File "/home/felix/projects/lib/brian2genn/brian2genn/device.py", line 1166, in compile_source
check_call(["make"], cwd=directory, env=env)
File "/usr/lib/python3.9/subprocess.py", line 373, in check_call
raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['make']' returned non-zero exit status 2.
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/home/felix/projects/ikeda/scratch.py", line 24, in <module>
net.run(10*ms)
File "/home/felix/projects/lib/brian2/brian2/core/base.py", line 291, in device_override_decorated_function
return getattr(curdev, name)(*args, **kwds)
File "/home/felix/projects/lib/brian2genn/brian2genn/device.py", line 1844, in network_run
super(GeNNDevice, self).network_run(net=net, duration=duration,
File "/home/felix/projects/lib/brian2/brian2/devices/cpp_standalone/device.py", line 1502, in network_run
self.build(direct_call=False, **self.build_options)
File "/home/felix/projects/lib/brian2genn/brian2genn/device.py", line 879, in build
raise RuntimeError(('Project compilation failed (Command {cmd} '
RuntimeError: Project compilation failed (Command ['make'] failed with error code 2).
See the output above (if any) for more details.There seems to be a disconnect between what's provided to GeNN (code objects, magicnetwork_model.cpp etc. with suffixed neurongroup_1 etc.) and what comes out the other end (magicnetwork_model_CODE/* with suffix-less neurongroup). I've found that only the Makefile is actually touched, whereas all other files in the _CODE directory are left untouched in the second iteration.
Of course, removing the GeNNworkspace directory fixes this, but that's hardly an ideal workaround.
Since this may well be a GeNN issue, I'm tagging @neworderofjamie here.
Unrelatedly, and without wanting to add to everyone's workload: Is it just me, or is brian2 master + brian2genn master + GeNN 4.6.0 completely broken?
neworderofjamie
tnowotny
mstimberg
This pull request attempts to fix an issue where the names of StateUpdaters, Thresholders etc. are suffixed with numeric indices, and are therefore not picked up by the hardcoded name checks in process_*_groups.
However, since this fix raises other hidden issues (see below), I can't recommend merging it just yet.