Crate axonml_fusion 

Axonml Fusion - Kernel Fusion Library

Provides kernel fusion support for combining multiple operations into single optimized kernels. Common fusion patterns include:

• MatMul + Bias + Activation: Fused dense layer
• Conv + BatchNorm + ReLU: Fused convolution block
• Elementwise chains: Multiple elementwise ops in one pass
• Reduction + Transform: Softmax, LayerNorm patterns

Example

use axonml_fusion::{FusedOp, fuse_matmul_bias_relu};

let fused = fuse_matmul_bias_relu(&weight, &bias);
let output = fused.execute(&input);

@version 0.1.0 @author AutomataNexus Development Team

Re-exports

pub use error::FusionError;

pub use error::FusionResult;

pub use patterns::FusionPattern;

pub use patterns::detect_patterns;

pub use elementwise::FusedElementwise;

pub use elementwise::fuse_elementwise;

pub use linear::FusedLinear;

pub use linear::fuse_matmul_bias_relu;

pub use optimizer::FusionOptimizer;

pub use optimizer::optimize_graph;

Modules

elementwise

Fused Elementwise Operations

error

Fusion Error Types

linear

Fused Linear Operations

optimizer

Graph Fusion Optimizer

patterns

Fusion Pattern Detection

Traits

FusedOp

Trait for fused operations.