Developer Guide for Intel® Data Analytics Acceleration Library 2019 Update 1

Essential Interfaces for Algorithms

In addition to Generic Interfaces, more methods enable interfacing tensors with algorithms.

The key methods that Intel DAAL algorithms use for per-block data access are getSubtensor()/releaseSubtensor(). These methods access a part of the tensor assuming the row-major order. The efficiency of data access highly depends on the data layout and on whether the data type is natively supported by the algorithm without type conversions. Refer to the Performance Considerations section in the description of a particular algorithm for a discussion of the data layout and natively supported data types.

Use the getSize() method to get the number of elements stored in a Tensor object.

Special Interfaces for the HomogenTensor Class

Use the getArray() method to get direct access to the memory block associated with the HomogenTensor object.

Use the setArray() method to set the number of dimensions, dimension values and associate the memory buffer with the HomogenTensor object.