Skip to main content

Crate axonml_fusion

Crate axonml_fusion 

Source
Expand description

Kernel fusion optimization for AxonML.

Automatic fusion pattern detection (detect_patterns), FusedLinear (matmul + bias + activation in one kernel), FusedElementwise (chain multiple elementwise ops into a single pass), FusionOptimizer (graph- level optimizer with conservative/aggressive modes, OptimizationStats), and the FusedOp trait for custom fusion implementations.

§File

crates/axonml-fusion/src/lib.rs

§Author

Andrew Jewell Sr. — AutomataNexus LLC ORCID: 0009-0005-2158-7060

§Updated

April 14, 2026 11:15 PM EST

§Disclaimer

Use at own risk. This software is provided “as is”, without warranty of any kind, express or implied. The author and AutomataNexus shall not be held liable for any damages arising from the use of this software.

Re-exports§

pub use elementwise::ElementwiseOp;
pub use elementwise::FusedElementwise;
pub use error::FusionError;
pub use error::FusionResult;
pub use linear::Activation;
pub use linear::FusedLinear;
pub use optimizer::FusionConfig;
pub use optimizer::FusionOptimizer;
pub use optimizer::OptimizationStats;
pub use optimizer::optimize_graph;
pub use patterns::FusionPattern;
pub use patterns::OpType;
pub use patterns::detect_patterns;

Modules§

elementwise
Fused Elementwise Operations
error
Fusion Error Types — Kernel Fusion Failure Modes
linear
Fused Linear Operations
optimizer
Graph Fusion Optimizer
patterns
Fusion Pattern Detection

Traits§

FusedOp
Trait for fused operations that combine multiple ops into one.