sliceView

Creates a zero-copy view of this tensor using the provided slicing DSL.

This method enables fluent syntax for tensor slicing that creates efficient views without copying data. The resulting TensorView shares memory with the parent tensor and applies coordinate transformations for element access.

Usage Example

val view = tensor.sliceView {
    segment { range(0, 10) }  // First dimension: indices 0-9
    segment { at(5) }         // Second dimension: index 5 (reduces rank)
    segment { all() }         // Third dimension: all indices
    segment { step(0, 20, 2) } // Fourth dimension: every 2nd index
}

Performance Characteristics

• Zero-copy: No data duplication, constant memory overhead

• Lazy: Coordinate mapping applied during element access

• Efficient: Optimized for NCHW layout and contiguous patterns

Parameters

Throws

Creates a view of an existing tensor view, enabling view chaining.

This extension allows creating views of views (view composition) while optimizing the composition by flattening slice operations to prevent deep nesting and maintain performance.

Usage Example

val baseView = tensor.sliceView {
    segment { range(0, 20) }    // First 20 batches
    segment { all() }           // All channels
    segment { all() }           // All height
    segment { all() }           // All width
}

val refinedView = baseView.sliceView {
    segment { range(5, 15) }    // Batches 5-14 from the base view (actual: 5-14 from original)
    segment { range(0, 64) }    // First 64 channels
    segment { all() }           // All height
    segment { all() }           // All width
}

Optimization Features

• Slice flattening: Combines multiple slice operations into a single optimized slice

• Bounds validation: Ensures composed slices remain within valid ranges

• Contiguity preservation: Maintains memory access optimizations where possible

• Chain optimization: Prevents excessive view nesting

Parameters

Throws