Struct GPUAccelerator

pub struct GPUAccelerator { /* private fields */ }

GPU acceleration backend with intelligent device detection and CPU fallback

Implementations

impl GPUAccelerator

pub fn global() -> &'static GPUAccelerator

Get the global GPU accelerator instance (singleton pattern for efficiency)

pub fn new() -> CandleResult<Self>

Create a new GPU accelerator with intelligent device detection

pub fn device_type(&self) -> &DeviceType

Get the device type being used

pub fn device(&self) -> &Device

Get the underlying Candle device

pub fn is_gpu_enabled(&self) -> bool

Check if GPU acceleration is available

pub fn device_count(&self) -> usize

Get number of available GPU devices

pub fn is_multi_gpu_available(&self) -> bool

Check if multiple GPU devices are available

pub fn all_devices(&self) -> &[Device]

Get all available devices for multi-GPU operations

pub fn switch_device(&mut self, device_index: usize) -> Result<(), String>

Switch to a specific device (for multi-GPU operations)

pub fn current_device_index(&self) -> usize

Get current device index

pub fn array_to_tensor(&self, array: &Array1<f32>) -> CandleResult<Tensor>

Convert ndarray to Candle tensor on the appropriate device

pub fn array2_to_tensor(&self, array: &Array2<f32>) -> CandleResult<Tensor>

Convert 2D ndarray to Candle tensor on the appropriate device

pub fn tensor_to_array(&self, tensor: &Tensor) -> CandleResult<Array1<f32>>

Convert Candle tensor back to ndarray

pub fn tensor_to_array2(&self, tensor: &Tensor) -> CandleResult<Array2<f32>>

Convert 2D Candle tensor back to ndarray

pub fn cosine_similarity_batch( &self, query: &Array1<f32>, vectors: &Array2<f32>, ) -> CandleResult<Array1<f32>>

Accelerated cosine similarity computation

pub fn matmul( &self, a: &Array2<f32>, b: &Array2<f32>, ) -> CandleResult<Array2<f32>>

Accelerated matrix multiplication

pub fn add_vectors(&self, vectors: &[Array1<f32>]) -> CandleResult<Array1<f32>>

Accelerated vector addition

pub fn memory_info(&self) -> String

Get memory usage information

pub fn benchmark(&self) -> CandleResult<f64>

Benchmark the device performance

pub fn multi_gpu_similarity_search( &self, query: &Array1<f32>, vectors: &Array2<f32>, ) -> CandleResult<Array1<f32>>

Multi-GPU parallel similarity search (when multiple GPUs available)

pub fn multi_gpu_batch_process<T, F>( &self, data: &[T], process_fn: F, ) -> Result<Vec<T>, String>

where T: Clone + Send + Sync, F: Fn(&[T], usize) -> Result<Vec<T>, String> + Send + Sync,

Multi-GPU batch processing for large operations

Trait Implementations

impl Clone for GPUAccelerator

fn clone(&self) -> GPUAccelerator

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for GPUAccelerator

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for GPUAccelerator

fn default() -> Self

Returns the “default value” for a type.