`mod types`¶

module types¶

C-compatible core data types for force/energy calculations.

These two #[repr(C)] structs are the fundamental data exchange types at the FFI boundary. Each field that previously held a raw pointer to a flat array now holds a *mut DLManagedTensorVersioned, making the types device-agnostic (CPU, CUDA, ROCm, etc.) via the DLPack tensor exchange protocol.

Memory Model

Input tensors are borrowed — the caller retains ownership.
Output forces are callee-allocated — the callback sets output.forces to a DLPack tensor it creates. After the call, the caller takes ownership and must free it via rgpot_tensor_free.
Energy and variance are plain f64 scalars, always on the host.

DLPack Tensor Shapes

Field	dtype	ndim	shape
`positions`	f64	2	`[n_atoms, 3]`
`atomic_numbers`	i32	1	`[n_atoms]`
`box_matrix`	f64	2	`[3, 3]`
`forces`	f64	2	`[n_atoms, 3]`

Structs and Unions

struct rgpot_force_input_t¶

Input configuration for a potential energy evaluation.

All tensor fields are borrowed — the caller retains ownership and must keep them alive for the lifetime of this struct.

Example (from C)

double positions[] = {0.0, 0.0, 0.0,  1.0, 0.0, 0.0};
int    types[]     = {1, 1};
double cell[]      = {10.0,0,0, 0,10.0,0, 0,0,10.0};

rgpot_force_input_t input = rgpot_force_input_create(2, positions, types, cell);
// ... use input ...
rgpot_force_input_free(&input);

positions: *mut DLManagedTensorVersioned¶: Positions tensor: [n_atoms, 3], dtype f64, any device.

atomic_numbers: *mut DLManagedTensorVersioned¶: Atomic numbers tensor: [n_atoms], dtype i32, any device.

box_matrix: *mut DLManagedTensorVersioned¶: Box matrix tensor: [3, 3], dtype f64, any device.

Implementations

impl rgpot_force_input_t¶

Functions

unsafe fn n_atoms(&self) -> Option<usize>¶

Extract n_atoms from the positions tensor’s shape[0].

Returns None if positions is null or has no shape.

Safety

The positions tensor must be a valid DLPack tensor with ndim >= 1.

struct rgpot_force_out_t¶

Results from a potential energy evaluation.

The forces field starts as NULL and is set by the potential callback to a callee-allocated DLPack tensor. After the call, the caller owns the tensor and must free it via rgpot_tensor_free.

Fields

forces: output force tensor, same shape/dtype as positions.
energy: the calculated potential energy.
variance: uncertainty estimate; zero when not applicable.

forces: *mut DLManagedTensorVersioned¶: Forces tensor [n_atoms, 3], f64 — set by the callback.

energy: f64¶: Calculated potential energy.

variance: f64¶: Variance / uncertainty of the calculation (0.0 when unused).

mod types¶

`mod types`¶