GeNNDevice class

(Shortest import: from brian2genn.device import GeNNDevice)

class brian2genn.device.GeNNDevice

Bases: brian2.devices.cpp_standalone.device.CPPStandaloneDevice

The main “genn” device. This does most of the translation work from Brian 2 generated code to functional GeNN code, assisted by the “GeNN language”.

Attributes

kernel_timings	Use GeNN’s kernel timings?
source_files	Set of all source and header files (to be included in runner)

Methods

activate([build_on_run])	Called when this device is set as the current device.
add_array_variable(model, varname, variable)
add_array_variables(model, owner)
add_parameter(model, varname, variable)
build([directory, compile, run, use_GPU, …])	This function does the main post-translation work for the genn device.
code_object(owner, name, abstract_code, …)	Processes abstract code into code objects and stores them in different arrays for GeNNCodeObjects and GeNNUserCodeObjects.
code_object_class([codeobj_class])	Return CodeObject class (either CPPStandaloneCodeObject class or input)
collect_synapses_variables(synapse_model, …)
compile_source(debug, directory, use_GPU)
consolidate_pull_operations(…)
copy_source_files(writer, directory)
fill_with_array(var, arr)	Fill an array with the values given in another array.
fix_random_generators(model, code)	Translates cpp_standalone style random number generator calls into GeNN- compatible calls by replacing the cpp_standalone _vectorisation_idx argument with the GeNN _seed argument.
fix_synapses_code(synapse_model, pathway, …)
generate_code_objects(writer)
generate_engine_source(writer, objects)
generate_main_source(writer, main_lines)
generate_makefile(directory, use_GPU)
generate_max_row_length_code_objects(writer)
generate_model_source(writer, main_lines, …)
generate_objects_source(arange_arrays, net, …)
insert_code(slot, code)	Insert custom C++ code directly into main.cpp.
make_main_lines()	Generates the code lines that handle initialisation of Brian 2 cpp_standalone type arrays.
network_get_profiling_info(net)
network_run(net, duration[, report, …])
process_neuron_groups(neuron_groups, objects)
process_poisson_groups(objects, poisson_groups)
process_rate_monitors(rate_monitors)
process_spike_monitors(spike_monitors)
process_spikegenerators(spikegenerator_groups)
process_state_monitors(directory, …)
process_synapses(synapse_groups, objects)
run(directory, use_GPU, with_output)
variableview_set_with_expression(…[, …])
variableview_set_with_expression_conditional(…)
variableview_set_with_index_array(…)

Details

kernel_timings

Use GeNN’s kernel timings?

source_files

Set of all source and header files (to be included in runner)

activate(build_on_run=True, **kwargs)

Called when this device is set as the current device.

add_array_variable(model, varname, variable)

add_array_variables(model, owner)

add_parameter(model, varname, variable)

build(directory='GeNNworkspace', compile=True, run=True, use_GPU=True, debug=False, with_output=True, direct_call=True)

This function does the main post-translation work for the genn device. It uses the code generated during/before run() and extracts information about neuron groups, synapse groups, monitors, etc. that is then formatted for use in GeNN-specific templates. The overarching strategy of the brian2genn interface is to use cpp_standalone code generation and templates for most of the “user-side code” (in the meaning defined in GeNN) and have GeNN-specific templates for the model definition and the main code for the executable that pulls everything together (in main.cpp and engine.cpp templates). The handling of input/output arrays for everything is lent from cpp_standalone and the cpp_standalone arrays are then translated into GeNN-suitable data structures using the static (not code-generated) b2glib library functions. This means that the GeNN specific cod only has to be concerned about executing the correct model and feeding back results into the appropriate cpp_standalone data structures.

code_object(owner, name, abstract_code, variables, template_name, variable_indices, codeobj_class=None, template_kwds=None, override_conditional_write=None, **kwds)

Processes abstract code into code objects and stores them in different arrays for GeNNCodeObjects and GeNNUserCodeObjects.

code_object_class(codeobj_class=None, *args, **kwds)

Return CodeObject class (either CPPStandaloneCodeObject class or input)

collect_synapses_variables(synapse_model, pathway, codeobj)

compile_source(debug, directory, use_GPU)

consolidate_pull_operations(run_regularly_operations)

copy_source_files(writer, directory)

fill_with_array(var, arr)

Fill an array with the values given in another array.

fix_random_generators(model, code)

Translates cpp_standalone style random number generator calls into GeNN- compatible calls by replacing the cpp_standalone _vectorisation_idx argument with the GeNN _seed argument.

fix_synapses_code(synapse_model, pathway, codeobj, code)

generate_code_objects(writer)

generate_engine_source(writer, objects)

generate_main_source(writer, main_lines)

generate_makefile(directory, use_GPU)

generate_max_row_length_code_objects(writer)

generate_model_source(writer, main_lines, use_GPU)

generate_objects_source(arange_arrays, net, static_array_specs, synapses, writer)

insert_code(slot, code)

Insert custom C++ code directly into main.cpp. The available slots are:

before_start / after_start

Before/after allocating memory for the arrays and loading arrays from disk.

before_network_run / after_network_run

Before/after calling GeNN’s run function.

before_end / after_end

Before/after writing results to disk and deallocating memory.

make_main_lines()

Generates the code lines that handle initialisation of Brian 2 cpp_standalone type arrays. These are then translated into the appropriate GeNN data structures in separately generated code.

network_get_profiling_info(net)

network_run(net, duration, report=None, report_period=10. * second, namespace=None, profile=None, level=0, **kwds)

process_neuron_groups(neuron_groups, objects)

process_poisson_groups(objects, poisson_groups)

process_rate_monitors(rate_monitors)

process_spike_monitors(spike_monitors)

process_spikegenerators(spikegenerator_groups)

process_state_monitors(directory, state_monitors, writer)

process_synapses(synapse_groups, objects)

run(directory, use_GPU, with_output)

variableview_set_with_expression(variableview, item, code, run_namespace, check_units=True)

variableview_set_with_expression_conditional(variableview, cond, code, run_namespace, check_units=True)

variableview_set_with_index_array(variableview, item, value, check_units)