genericLayer

template<typename StateTypes> // e.g. thrust::tuple<float,float,float> class genericLayer

Generic layer wrapper holding neuron states and inputs.

StateTypes

Thrust tuple describing the per-neuron state variables.

Functions

Name	Description
make_state_tuple	Helper to construct the tuple of state vectors on the device.
genericLayer	Constructs a layer with the given neuron count.
size	Returns the number of neurons in the layer.
state_vec	Access a specific state vector by index.
state_vec	Access a specific state vector by index (const overload).
make_zip_begin	Create a zip iterator over all state vectors (mutable).
make_zip_end	Create a zip iterator over all state vectors (mutable) - end.
make_zip_begin	Create a zip iterator over all state vectors (const).
make_zip_end	Create a zip iterator over all state vectors (const) - end.
full_zip_begin	Create a zip iterator over state and input (mutable).
full_zip_end	Create a zip iterator over state and input (mutable) - end.
step	Advance the layer state by one integration step using RK4.
input	Access the input buffer for modification.
input	Access the input buffer (read-only).
make_begin_tuple	Build a tuple of begin iterators for all state vectors.
make_end_tuple	Build a tuple of end iterators for all state vectors.
make_cbegin_tuple	Build a tuple of const begin iterators for all state vectors.
make_cend_tuple	Build a tuple of const end iterators for all state vectors.

Function Details

full_zip_begin

auto full_zip_begin()

Create a zip iterator over state and input (mutable).

The iterator dereferences to a tuple of (state tuple iterator, input value).

Return

Iterator to the beginning of the zipped state+input range.

full_zip_end

auto full_zip_end()

Create a zip iterator over state and input (mutable) - end.

Return

Iterator to the end of the zipped state+input range.

genericLayer

explicit genericLayer(size_t N)

Constructs a layer with the given neuron count.

N

Number of neurons/states to allocate.

input

thrust::device_vector<float>& input() noexcept

Access the input buffer for modification.

Return

Reference to the device vector storing input currents.

const thrust::device_vector<float>& input() const noexcept

Access the input buffer (read-only).

Return

Const reference to the device vector storing input currents.

make_begin_tuple

template<size_t... I> auto make_begin_tuple(std::index_sequence<I...>)

Build a tuple of begin iterators for all state vectors.

I

Index pack over the elements of StateTypes.

Return

Thrust tuple of begin iterators.

make_cbegin_tuple

template<size_t... I> auto make_cbegin_tuple(std::index_sequence<I...>) const

Build a tuple of const begin iterators for all state vectors.

I

Index pack over the elements of StateTypes.

Return

Thrust tuple of const begin iterators.

make_cend_tuple

template<size_t... I> auto make_cend_tuple(std::index_sequence<I...>) const

Build a tuple of const end iterators for all state vectors.

I

Index pack over the elements of StateTypes.

Return

Thrust tuple of const end iterators.

make_end_tuple

template<size_t... I> auto make_end_tuple(std::index_sequence<I...>)

Build a tuple of end iterators for all state vectors.

I

Index pack over the elements of StateTypes.

Return

Thrust tuple of end iterators.

make_state_tuple

template<size_t... I> static auto make_state_tuple(size_t N, std::index_sequence<I...>)

Helper to construct the tuple of state vectors on the device.

For each element type in StateTypes, creates a corresponding thrust::device_vector of size N initialized to zero.

I

Index pack over the elements of StateTypes.

N

Number of elements in each state vector.

Return

Tuple of device vectors matching StateTypes.

make_zip_begin

auto make_zip_begin()

Create a zip iterator over all state vectors (mutable).

Each element of the iterator is a thrust::tuple of references to the components of the neuron state.

Return

Iterator to the beginning of the zipped state range.

auto make_zip_begin() const

Create a zip iterator over all state vectors (const).

Return

Const iterator to the beginning of the zipped state range.

make_zip_end

auto make_zip_end()

Create a zip iterator over all state vectors (mutable) - end.

Return

Iterator to the end of the zipped state range.

auto make_zip_end() const

Create a zip iterator over all state vectors (const) - end.

Return

Const iterator to the end of the zipped state range.

size

size_t size() const noexcept

Returns the number of neurons in the layer.

state_vec

template<size_t I> auto& state_vec() noexcept

Access a specific state vector by index.

I

Index of the state component in the StateTypes tuple.

Return

Reference to the corresponding device vector.

template<size_t I> const auto& state_vec() const noexcept

Access a specific state vector by index (const overload).

I

Index of the state component in the StateTypes tuple.

Return

Const reference to the corresponding device vector.

step

template<typename RHS_Functor> void step(const RHS_Functor& rhs, float t, float dt, cudaStream_t stream = 0)

Advance the layer state by one integration step using RK4.

Applies the provided RHS functor on all neurons in parallel using Thrust on the given CUDA stream.

RHS_Functor

Functor type providing the right-hand side of the ODE. It must be callable from device code.

See

Solvers

rhs

Right-hand side functor.

t

Current time.

dt

Time step.

stream

CUDA stream to execute the operation on (default 0).