Versioned name: Reorder-1

Category: Movement

Short description: Reorder converts the input tensor to output tensor with different layouts. It supports the conversion between 1) two different opaque layouts 2) two different public layouts 3) one opaque layout and another public layout.

Detailed description: Reorder also requires that the input tensor and output tensor should have the same data type (f32/bf16/f16) and shape

For example, if the input tensor has public layout with strides, and users want to convert it to an output tensor with opaque layout (specified by layout id), then reorder can be used for this case.

Currently, reorder operator doesn’t support layout conversion cross backends or cross engines.


  • 1: input - input tensor. Required.

    • Type: T


  • 1: output - the output tensor with different layout from input tensor.

    • Type: T


  • T: f32, f16, bf16.

  • Note: Inputs and outputs have the same data type denoted by T. For example, if input is f32 tensor, then all other tensors have f32 data type.