Struct GpuAcceleratedOps 

pub struct GpuAcceleratedOps { /* private fields */ }

GPU-accelerated operations manager

This provides a high-level interface for GPU-accelerated tensor operations, automatically selecting the best implementation based on available hardware.

Implementations

impl GpuAcceleratedOps

pub fn new(config: GpuOpsConfig) -> Result<Self>

Create new GPU-accelerated operations manager

pub fn matmul(&self, a: &Tensor, b: &Tensor) -> Result<Tensor>

Matrix multiplication with GPU acceleration

pub fn batch_matmul(&self, a: &Tensor, b: &Tensor) -> Result<Tensor>

Batch matrix multiplication

pub fn flash_attention( &self, query: &Tensor, key: &Tensor, value: &Tensor, scale: f32, mask: Option<&Tensor>, ) -> Result<Tensor>

Flash attention implementation

pub fn layer_norm( &self, input: &Tensor, gamma: &Tensor, beta: &Tensor, epsilon: f32, ) -> Result<Tensor>

Layer normalization with GPU acceleration

pub fn gelu(&self, input: &Tensor) -> Result<Tensor>

GELU activation with GPU acceleration

pub fn reduce_sum(&self, input: &Tensor, dim: usize) -> Result<Tensor>

Reduce sum with GPU acceleration

pub fn softmax(&self, input: &Tensor, dim: usize) -> Result<Tensor>

Softmax with GPU acceleration

pub fn is_gpu_available(&self) -> bool

Check if GPU acceleration is available

pub fn get_backend(&self) -> GpuBackend

Get current GPU backend

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

Get GPU memory usage

pub fn synchronize(&self) -> Result<()>

Synchronize GPU operations