#include <torch/extension.h>

void state_step_cuda(torch::Tensor v, torch::Tensor g, torch::Tensor refractory,
                     torch::Tensor arrival, torch::Tensor spikes,
                     torch::Tensor spike_counts, double g_decay, double alpha,
                     double rest, double reset, double threshold,
                     int64_t refractory_steps);
void force_step_cuda(torch::Tensor v, torch::Tensor refractory,
                     torch::Tensor spikes, torch::Tensor spike_counts,
                     torch::Tensor targets, torch::Tensor forced,
                     double reset, int64_t refractory_steps);
void fanout_cuda(torch::Tensor spikes, torch::Tensor indptr,
                 torch::Tensor indices, torch::Tensor weights,
                 torch::Tensor destination);
void fill_edges_cuda(torch::Tensor indptr, torch::Tensor sources,
                     torch::Tensor output_offsets, torch::Tensor output);

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
  m.def("state_step", &state_step_cuda, "Fused exact LIF state update (CUDA)");
  m.def("force_step", &force_step_cuda, "Apply external forced spikes (CUDA)");
  m.def("fanout", &fanout_cuda, "Event-driven CSR fan-out (CUDA)");
  m.def("fill_edges", &fill_edges_cuda, "Exact source-ordered CSR edge expansion (CUDA)");
}