CopyMaterializationStrategy

A materialization strategy that immediately copies all data from a tensor view into a new, standalone tensor with contiguous memory layout.

This strategy provides immediate materialization by iterating through all elements in the view and copying them to a new tensor with the view's shape. The resulting tensor is completely independent of the parent tensor and can be used even after the parent tensor is garbage collected.

Characteristics

• Immediate Execution: Materialization happens synchronously when called

• Memory Independent: Result tensor has no dependencies on parent tensor

• Contiguous Layout: Output data is stored in standard row-major order

• Type Preservation: Maintains the same data type and value type as the view

Trade-offs

Benefits:

• Predictable memory usage and performance

• No ongoing computational overhead for element access

• Enables garbage collection of parent tensors

• Compatible with all downstream operations

Costs:

• Immediate memory allocation for full tensor size

• Computational cost of copying all elements

• Temporary memory pressure during materialization

Usage Scenarios

This strategy is optimal when:

• The materialized tensor will be accessed frequently

• Memory usage is predictable and acceptable

• The parent tensor can be released after materialization

• Compatibility with external libraries is required

Parameters