Module shape_inference 

Static shape inference for tensor operation graphs.

Provides TensorShapeInference which computes output shapes for common tensor operations (broadcast, matmul, reshape, transpose, concat, slice) following NumPy-compatible broadcasting rules.

Examples

use ipfrs_tensorlogic::shape_inference::{TensorShapeInference, TensorShape, ShapeOp, InferenceRule};
use std::collections::HashMap;

let mut engine = TensorShapeInference::new();

// Matrix multiply (3,4) x (4,5) -> (3,5)
let rule = InferenceRule {
    op: ShapeOp::MatMul,
    input_shapes: vec![
        TensorShape { dims: vec![3, 4] },
        TensorShape { dims: vec![4, 5] },
    ],
    params: HashMap::new(),
};
let result = engine.infer(&rule).expect("example: should succeed in docs");
assert_eq!(result.dims, vec![3, 5]);

Structs

InferenceRule

An inference rule: an operation together with its input shapes and optional parameters.

ShapeInferenceStats

Statistics collected by TensorShapeInference.

TensorShape

Shape descriptor for a tensor.

TensorShapeInference

Static shape inference engine for tensor operation graphs.

Enums

ShapeOp

Supported tensor operations for shape inference.