isContiguous

Determines if this tensor view represents a contiguous memory layout.

A contiguous view is one where the elements can be accessed in a linear fashion without gaps or stride adjustments. Contiguous views offer significant performance advantages for:

• Bulk memory operations (copy, fill, etc.)

• Vectorized computations

• Cache-friendly access patterns

• Direct memory mapping to external libraries

Contiguity Analysis

A view is considered contiguous when:

• Sequential element access follows natural memory layout

• No gaps exist between accessed elements

• Stride patterns match expected row-major or column-major order

• View boundaries align with natural tensor structure

Performance Implications

Contiguous Views:

• Enable bulk memory operations

• Support vectorized SIMD instructions

• Provide optimal cache locality

• Allow direct interfacing with BLAS/LAPACK libraries

Non-Contiguous Views:

• Require element-by-element access

• May have poor cache performance

• Need coordinate transformation for each access

• Cannot use optimized bulk operations

Usage Examples

val tensor = tensorOf(Shape(4, 4)) { /* data */}

// Contiguous slice (full rows)
val contiguousView = tensor.sliceView { range(1, 3) }
println(contiguousView.isContiguous()) // true

// Non-contiguous slice (columns)
val nonContiguousView = tensor.sliceView { at(0); range(1, 3) }
println(nonContiguousView.isContiguous()) // false

Parameters