Struct GpuEmbeddingAccelerator 

pub struct GpuEmbeddingAccelerator { /* private fields */ }

Enhanced GPU-accelerated embedding computations with memory pooling and adaptive batching

Implementations

impl GpuEmbeddingAccelerator

pub fn new(device_id: u32) -> Result<Self, GpuError>

Create a new enhanced GPU accelerator with memory pooling and adaptive batching Note: Currently using placeholder until scirs2_linalg::gpu is available

pub async fn get_optimal_batch_size(&self, data_size: usize) -> usize

Get optimal batch size based on recent performance

pub async fn update_batch_size_feedback( &self, _batch_size: usize, performance_score: f32, )

Update optimal batch size based on performance feedback

pub fn batch_l2_distances_gpu( &self, vectors_a: &[Array1<f64>], vectors_b: &[Array1<f64>], ) -> Result<Vec<f64>, GpuError>

GPU-accelerated batch distance computation

pub fn cosine_similarity_matrix_gpu( &self, vectors: &[Array1<f64>], ) -> Result<Array2<f64>, GpuError>

GPU-accelerated cosine similarity matrix Note: Currently using CPU fallback until scirs2_linalg::gpu is available

pub fn batch_gradient_update_gpu( &self, embeddings: &mut [Array2<f64>], gradients: &[Array2<f64>], learning_rate: f64, l2_reg: f64, ) -> Result<(), GpuError>

GPU-accelerated gradient updates for large embedding matrices Note: Currently using CPU fallback until scirs2_linalg::gpu is available

pub async fn adaptive_batch_processing<T, R>( &self, data: &[T], process_fn: impl FnMut(&[T]) -> Result<Vec<R>, GpuError>, ) -> Result<Vec<R>, GpuError>

Advanced GPU-accelerated adaptive batch processing with memory pooling

pub async fn optimized_matrix_multiply( &self, a: &Array2<f32>, b: &Array2<f32>, ) -> Result<Array2<f32>, GpuError>

GPU-accelerated matrix multiplication with memory reuse Note: Currently using CPU fallback until scirs2_linalg::gpu is available

pub async fn gpu_embedding_search( &self, query_embedding: &Array1<f32>, database_embeddings: &[Array1<f32>], top_k: usize, ) -> Result<Vec<(usize, f32)>, GpuError>

High-performance embedding search with GPU acceleration

pub fn xavier_init_gpu( &self, shapes: &[(usize, usize)], fan_in: usize, fan_out: usize, seed: u64, ) -> Result<Vec<Array2<f64>>, GpuError>

GPU-accelerated Xavier initialization for large embedding matrices Note: Currently using CPU fallback until scirs2_linalg::gpu is available

pub fn contrastive_learning_gpu( &self, _entity_embeddings: &mut [Array1<f32>], _similarity_pairs: &[(usize, usize)], _negative_samples: &[(usize, usize)], _temperature: f32, _learning_rate: f32, ) -> Result<f32, GpuError>

GPU-accelerated contrastive learning updates Note: Currently using CPU fallback until scirs2_linalg::gpu is available

pub fn device_info(&self) -> String

Get GPU device info

pub fn available_memory(&self) -> Result<u64, GpuError>

Get available GPU memory Note: Currently using placeholder until scirs2_linalg::gpu is available

pub async fn get_performance_report(&self) -> GpuPerformanceReport

GPU memory and performance monitoring

pub fn reset_performance_stats(&self)

Reset performance statistics

pub fn get_memory_pool_status(&self) -> (usize, u64, u64)

Get current memory pool status