Struct TensorOpFusion 

pub struct TensorOpFusion { /* private fields */ }

Stateful engine that fuses primitive tensor-op sequences into compound ops.

Example

use ipfrs_tensorlogic::op_fusion::{TensorOp as FusionTensorOp, TensorOpFusion};

let mut engine = TensorOpFusion::new();
let ops = vec![
    FusionTensorOp::Scale { factor: 2.0 },
    FusionTensorOp::Bias  { offset: 1.0 },
];
let plan = engine.fuse(ops);
assert_eq!(plan.fused_op_count, 1);
assert!((plan.reduction_ratio() - 0.5).abs() < 1e-10);

Implementations

impl TensorOpFusion

pub fn new() -> Self

Create a new, zero-stats fusion engine.

pub fn fuse(&mut self, ops: Vec<TensorOp>) -> FusionPlan

Fuse ops left-to-right using greedy matching and return a FusionPlan.

pub fn stats(&self) -> &FusionStats

Return a reference to the accumulated statistics.

pub fn can_fuse(a: &TensorOp, b: &TensorOp) -> bool

Return true when a and b form the start of a fuseable pattern.

This covers all two-op pairs that appear in any fusion rule:

• Scale → Bias (ScaleBias)
• Scale → Relu (start of ScaleReluBias)
• Relu → Bias (tail of ScaleReluBias)
• Clamp → Normalize (ClampNormalize)

Trait Implementations

impl Debug for TensorOpFusion

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

Formats the value using the given formatter.

impl Default for TensorOpFusion

fn default() -> TensorOpFusion

Returns the “default value” for a type.