Skip to content

TensorIterator Implementation

Jump to bottom
Sameer Deshmukh edited this page Oct 24, 2019 · 2 revisions

TensorIterator

The TensorIterator is mainly useful for creating iterator abstractions that can be executed in parallel on CPU or GPU. It has two modes: reduction and point-wise.

The point-wise mode makes full use of the fact that the entries are independent and uses this for parallelization of the operation using thread blocks and arbitrary execution of the operation on each entry.

Dimension Reduction

Clone this wiki locally