spikingjelly.activation_based.cuda_kernel.neuron_kernel package

neuron_kernel owns concrete CuPy neuron kernels and the public multi-step custom-kernel extension interface. Generic CUDA code-generation primitives remain in spikingjelly.activation_based.cuda_kernel.auto_cuda.

Implementation overview

The built-in kernels use two implementation styles:

• AutoCUDA-generated: single-step IF/LIF and multi-step IF/LIF/PLIF kernels.

• Hand-written CUDA: multi-step EIF/QIF/Izhikevich kernels, which build fixed CUDA sources directly with cupy.RawKernel.

auto_cuda provides code-generation primitives only; neuron-specific kernels remain under neuron_kernel.