Figure 5

Running simulations with Brian2GeNN. Top: Excerpt from an example Brian script that will execute in a hybrid Brian/GeNN simulation due to the import of the brian2genn library (line 2) and the selection of the “GeNN device” (line 3). Bottom left: Typical workflow of a Brian2GeNN simulation: the run call triggers the code generation, compilation and execution. After the successful run, results are stored to disk and made available to the Python script. Numbers refer to the corresponding lines in the example code on top. Bottom right: Structure of generated code. Parts of the code result from Brian’s standard code generation process (blue), while the main run step is implemented by GeNN (red) and everything is connected together by Brian2GeNN (green). The preparation of the simulation and actions such as variable monitoring are executed on the CPU (left), while the core of the simulation is executed on the GPU (right). The numbers in the boxes refer to the elements of the example code (top) and general schematic (bottom left) which are the base for the code of the corresponding block. For example, the “run step”, i.e. the advancement of the state variables of neurons and synapses at every time step, is based on the definitions of the neuron and synapse models in (1) and (2).