Function: set_tensor_format

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product / Atlas A3 inference product	√
Atlas A2 training product / Atlas A2 inference product	√
Atlas training product	x
Atlas inference product	√
Atlas 200I/500 A2 inference product	√

Function Usage

Call acl.set_tensor_format to set the format of a tensor when the tensor is processed in an Ascend AI Processor after a tensor description is created by the acl.create_tensor_desc call. This API is mainly used in the PyTorch scenarios.

For example, if the original operator shape is 4D with format NHWC or 5D with format NDHWC, the format is automatically converted to NC1HWC0 or NDC1HWC0 during model conversion or online network build to facilitate data access efficiency on the AI processor. Along the process, the operator shape is also inferred to facilitate internal computation.

Prototype

C Prototype

        
             aclError aclSetTensorFormat(aclTensorDesc *desc, aclFormat format)

Python Function

        
             ret = acl.set_tensor_format(desc, format)

Parameters

Parameter	Description
desc	Int, pointer address of the tensor description data. Call acl.create_tensor_desc to create data of the aclTensorDesc type in advance.
format	Int, format of the specified tensor description.

Parameter

Description

desc

Int, pointer address of the tensor description data.

Call acl.create_tensor_desc to create data of the aclTensorDesc type in advance.

format

Int, format of the specified tensor description.

Returns

Return Value	Description
ret	Int, error code. 0 indicates success. Other values indicate failure.

Parent topic: aclmdlTensorDesc