Class TensorShapeExtensions

Namespace

AiDotNet.JitCompiler.IR

Assembly

AiDotNet.dll

Provides extension methods and utilities for working with tensor shapes in the IR.

public static class TensorShapeExtensions

Inheritance

object

TensorShapeExtensions

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

This class provides helper methods for working with tensor shapes (represented as int[] arrays). It integrates with the existing Tensor<T> infrastructure which already uses int[] for shapes.

For Beginners: In AiDotNet, tensor shapes are represented as integer arrays.

For example:

• [5] is a vector with 5 elements
• [3, 4] is a 3×4 matrix
• [2, 3, 4] is a 3D tensor

This class provides utilities to work with these shapes:

• Check if two shapes are compatible for operations
• Compute the result shape when broadcasting
• Validate shapes
• Compare shapes

These utilities are used by the JIT compiler to understand tensor dimensions and generate optimized code.

Methods

BroadcastWith(int[], int[])

Computes the broadcast shape resulting from combining two shapes.

public static int[] BroadcastWith(this int[] shape1, int[] shape2)

Parameters

shape1 int[]

The first shape.

shape2 int[]

The second shape.

Returns

int[]

The broadcast result shape.

Remarks

The broadcast shape is computed by taking the maximum dimension at each position when comparing from right to left.

For Beginners: This calculates what shape results when broadcasting two tensors.

Examples:

• [3, 4] + [3, 4] → [3, 4] (same shape)
• [3, 4] + [1, 4] → [3, 4] (first dimension expands from 1 to 3)
• [3, 4] + [4] → [3, 4] (vector broadcasts to match all rows)
• [5, 3, 4] + [4] → [5, 3, 4] (vector broadcasts across all 5×3 positions)

The result tells us what shape the output will have after the operation.

Exceptions

InvalidOperationException

Thrown if shapes are not compatible.

GetElementCount(int[])

Computes the total number of elements in a tensor with the given shape.

public static int GetElementCount(this int[] shape)

Parameters

shape int[]

The tensor shape.

Returns

int

The total number of elements, or -1 if any dimension is dynamic.

Remarks

For Beginners: This calculates how many total values a tensor holds.

For example:

• [] has 1 element (scalar - a single number)
• [5] has 5 elements
• [3, 4] has 3 × 4 = 12 elements
• [2, 3, 4] has 2 × 3 × 4 = 24 elements

If any dimension is -1 (meaning "dynamic" or "unknown"), returns -1.

GetRank(int[])

Gets the rank (number of dimensions) of a tensor shape.

public static int GetRank(this int[] shape)

Parameters

shape int[]

The tensor shape.

Returns

int

The number of dimensions.

Remarks

For Beginners: The rank is how many dimensions the tensor has.

• [5] has rank 1 (a vector)
• [3, 4] has rank 2 (a matrix)
• [2, 3, 4] has rank 3 (a 3D tensor)
• [] has rank 0 (a scalar - single number)

GetShapeHashCode(int[])

Computes a hash code for a tensor shape.

public static int GetShapeHashCode(this int[] shape)

Parameters

shape int[]

The shape to hash.

Returns

int

A hash code.

Remarks

This hash code can be used to cache compiled graphs based on shape. Shapes with the same dimensions will have the same hash.

For Beginners: This creates a unique number that represents the shape.

It's like a fingerprint for the shape - two identical shapes will have the same hash code. This is used to quickly check if we've already compiled code for a tensor of this shape, so we can reuse it instead of recompiling.

GetShape<T>(Tensor<T>)

Extracts the shape from a Tensor.

public static int[] GetShape<T>(this Tensor<T> tensor)

Parameters

tensor Tensor<T>

The tensor.

Returns

int[]

The shape as an int array.

Type Parameters

T

The numeric type of the tensor.

Remarks

For Beginners: This gets the shape from an existing Tensor object.

Since Tensor already has a Shape property, this just returns it. It's provided for consistency with the IR infrastructure.

IsCompatibleWith(int[], int[])

Checks if this shape is compatible with another shape for broadcasting.

public static bool IsCompatibleWith(this int[] shape1, int[] shape2)

Parameters

shape1 int[]

The first shape.

shape2 int[]

The second shape.

Returns

bool

True if the shapes are compatible for broadcasting.

Remarks

Broadcasting allows operations between tensors of different shapes by automatically expanding dimensions. Two shapes are compatible if: - They have the same rank and all dimensions match, OR - One dimension is 1 (can be broadcast), OR - One tensor has fewer dimensions (will be expanded)

For Beginners: Broadcasting lets you do operations on tensors of different sizes.

For example:

• [3, 4] and [3, 4] are compatible (same shape)
• [3, 4] and [1, 4] are compatible (first dimension broadcasts)
• [3, 4] and [4] are compatible (vector broadcasts across all rows)
• [3, 4] and [3, 5] are NOT compatible (incompatible dimensions)

This is very useful in neural networks where you often add a bias vector to every row of a matrix - broadcasting handles this automatically.

IsValidShape(int[])

Validates that a shape is well-formed.

public static bool IsValidShape(this int[] shape)

Parameters

shape int[]

The shape to validate.

Returns

bool

True if valid.

Remarks

A shape is valid if all dimensions are either positive or -1 (dynamic). Zero dimensions are not allowed.

For Beginners: This checks that a shape makes sense.

Valid shapes:

• [] (scalar)
• [5] (vector with 5 elements)
• [3, 4] (3×4 matrix)
• [-1, 4] (dynamic first dimension, 4 columns)

Invalid shapes:

• [0, 4] (can't have zero dimension)
• [3, -2] (only -1 is allowed for dynamic)

ShapeToString(int[])

Creates a string representation of a shape.

public static string ShapeToString(this int[] shape)

Parameters

shape int[]

The shape to represent.

Returns

string

A string representation.

Remarks

For Beginners: This converts a shape to a readable string for debugging.

Examples:

• [] → "scalar"
• [5] → "[5]"
• [3, 4] → "[3, 4]"
• [2, -1, 4] → "[2, ?, 4]" (? means dynamic)

ShapesEqual(int[]?, int[]?)

Checks if two shapes are exactly equal.

public static bool ShapesEqual(int[]? shape1, int[]? shape2)

Parameters

shape1 int[]

The first shape.

shape2 int[]

The second shape.

Returns

bool

True if shapes are equal.

Remarks

For Beginners: This checks if two shapes are identical.

Examples:

• [3, 4] equals [3, 4] → true
• [3, 4] equals [4, 3] → false (different order!)
• [3, 4] equals [1, 4] → false (different dimensions)