Struct BackendRegistry 

pub struct BackendRegistry { /* private fields */ }

Registry of available compute backends.

Current behavior: Only the CPU backend is available. load_cutensor and load_hiptensor always return errors. GPU backend loading is not yet implemented.

When GPU support is implemented, this registry will hold the CPU backend (always available) and optional GPU backends loaded at runtime.

Examples

// Aspirational API — GPU loading not yet functional.
use tenferro_prims::BackendRegistry;

let mut registry = BackendRegistry::new(); // CPU only
registry.load_cutensor("/usr/lib/libcutensor.so").unwrap();
assert!(registry.cuda().is_some());

Implementations

impl BackendRegistry

pub fn new() -> Self

Create a registry with CPU backend only.

pub fn load_cutensor(&mut self, _path: &str) -> Result<()>

Load the cuTENSOR library from the given path.

Status: Not available. The cuda feature is not enabled. Rebuild with --features cuda to enable cuTENSOR support.

pub fn load_hiptensor(&mut self, _path: &str) -> Result<()>

Load the hipTENSOR library from the given path.

Status: Not yet implemented. Always returns Err(DeviceError).

When implemented, the caller (Julia, Python, or standalone Rust) will provide the path to the shared library. No auto-search.

pub fn cpu(&self) -> &CpuBackend

Returns a reference to the CPU backend.

pub fn cuda(&self) -> Option<&CudaBackend>

Returns a reference to the CUDA backend, if loaded.

pub fn rocm(&self) -> Option<&RocmBackend>

Returns a reference to the ROCm backend, if loaded.

Trait Implementations

impl Default for BackendRegistry

fn default() -> Self

Returns the “default value” for a type.