batchMaterialize

Materializes a list of tensor views into a list of standalone tensors efficiently.

This method provides batch materialization of multiple tensor views, which can offer performance benefits over materializing views individually. The batch operation can optimize for:

• Memory allocation patterns

• CPU cache efficiency

• Parallel processing opportunities

• Reduced overhead from repeated strategy initialization

Batch Processing Benefits

Memory Optimization:

• Pre-allocates memory for all materialized tensors

• Reduces memory fragmentation

• Enables memory pool optimizations

Performance Optimization:

• Amortizes strategy setup costs across multiple views

• Enables vectorized operations when possible

• Reduces system call overhead

Resource Management:

• Coordinates garbage collection timing

• Manages memory pressure more effectively

• Provides better resource utilization monitoring

Usage Examples

val parentTensor = tensorOf(Shape(100, 100)) { /* data */}
val views = listOf(
    parentTensor.sliceView { range(0, 25) },
    parentTensor.sliceView { range(25, 50) },
    parentTensor.sliceView { range(50, 75) },
    parentTensor.sliceView { range(75, 100) }
)

// Batch materialize all views efficiently
val materialized = views.batchMaterialize()

Performance Considerations

Batch materialization is most beneficial when:

• Processing multiple views from the same parent tensor

• Views have similar complexity and size characteristics

• Memory allocation overhead is significant compared to data copying

• The system benefits from predictable memory allocation patterns

Parameters

Throws

Materializes a list of tensor views using a specific materialization strategy.

This overload allows explicit control over the materialization strategy used for the entire batch operation, enabling optimization for specific use cases or resource constraints.

Parameters

Throws