Skip to content

Reshape Sync Task

The Reshape sync task changes the logical shape of a tensor on the GPU while preserving element order and dtype. It writes the input buffer into a new tensor whose extents are provided by the shape setting, effectively reinterpreting the linear data with different dimensions.

Note

Reshape only remaps contiguous data; it neither transposes nor reorders samples. The linear memory layout remains identical to the source tensor.

Warning

The configured shape must contain only positive dimensions and its product must exactly match the number of elements in the input tensor. If either check fails, task creation is rejected during inference.

Inputs

This task consumes a single input tensor of arbitrary rank, shape, dtype, and memory location.

Outputs

This task produces one output tensor located in the same memory space as the input tensor. The output tensor has the same dtype as the input tensor but a shape defined by the configured shape setting.

Inplace

This task does not support inplace operation.

Ownership

This task does not own any inputs or outputs.

Settings

Example:

1
2
3
4
5
6
7
{
    "shape": [
        1,
        512,
        512
    ]
}
Property Pattern Type Deprecated Definition Title/Description
+ shape No array of integer No - Desired output shape expressed as a list of dimensions. Must satisfy:
- shape has at least one dimension
- no dimension is zero
- product(shape) equals the number of input elements