Function: get_input_dynamic_dims

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	√
Atlas inference product	√
Atlas 200I/500 A2 inference product	√

Function Usage

Obtains the dynamic shape profiles supported by a model based on the model description.

Prototype

C Prototype

        
             aclError aclmdlGetInputDynamicDims(const aclmdlDesc *modelDesc, size_t index, aclmdlIODims *dims, size_t gearCount)

Python Function

        
             dims_out, ret = acl.mdl.get_input_dynamic_dims(model_desc, index, gear_count)

Parameters

Parameter	Description
model_desc	Int, pointer address of the data of the aclmdlDesc type. Call acl.mdl.create_desc to create data of the aclmdlDesc type in advance.
index	Int, reserved. The value is fixed to –1.
gear_count	Int, number of dynamic shape profiles supported by the model, which needs to be obtained through the acl.mdl.get_input_dynamic_gear_count call in advance.

Parameter

Description

model_desc

Int, pointer address of the data of the aclmdlDesc type.

Call acl.mdl.create_desc to create data of the aclmdlDesc type in advance.

index

Int, reserved. The value is fixed to –1.

gear_count

Int, number of dynamic shape profiles supported by the model, which needs to be obtained through the acl.mdl.get_input_dynamic_gear_count call in advance.

Returns

Return Value	Description
dims_out	List, an array for the dynamic shape. The dims_out parameter is a list. The number of elements in the list is the same as the number of inputs of the model. Each element in the list is a aclmdlIODims dictionary. dims in the aclmdlIODims dictionary is a list. Each element in the list corresponds to a specific value in each dimension size profile. For example: dims_out, ret = acl.mdl.get_input_dynamic_dims(model_desc, -1, gear_count)
ret	Int, error code. 0 indicates success. Other values indicate failure.

Return Value

Description

dims_out

List, an array for the dynamic shape.

The dims_out parameter is a list. The number of elements in the list is the same as the number of inputs of the model. Each element in the list is a aclmdlIODims dictionary. dims in the aclmdlIODims dictionary is a list. Each element in the list corresponds to a specific value in each dimension size profile.

For example:

dims_out, ret = acl.mdl.get_input_dynamic_dims(model_desc, -1, gear_count)

ret

Int, error code.

0 indicates success.
Other values indicate failure.

Restrictions

This API can be called only after the --dynamic_dims parameter is specified for model conversion.

Assume that a model has three inputs: data (1, 1, 40, -1), label (1, -1), and mask (-1, -1), where -1 indicates a dynamic dimension. If dynamic_dims is set to --dynamic_dims="20,20,1,1; 40,40,2,2; 80,60,4,4" for model conversion, then the dynamic shape returned by this API call is as follows (Note that the name parameter in the aclmdlIODims dictionary is reserved currently):

Profile 0:
- In aclmdlIODims, dimCount: 8 indicates that the inputs have eight dimensions altogether.
- In aclmdlIODims, dims: 1,1,40,20,1,20,1,1 indicates shape profile 0 of each input: data (1, 1, 40, 20), label (1, 20), and mask (1, 1).
Profile 1:
- In aclmdlIODims, dimCount: 8 indicates that the inputs have eight dimensions altogether.
- In aclmdlIODims, dims: 1,1,40,40,1,40,2,2 indicates shape profile 1 of each input: data (1, 1, 40, 40), label (1, 40), and mask (2, 2).
Profile 2:
- In aclmdlIODims, dimCount: 8 indicates that the inputs have eight dimensions altogether.
- In aclmdlIODims, dims: 1,1,40,80,1,60,4,4 indicates shape profile 2 of each input: data (1, 1, 40, 80), label (1, 60), and mask (4, 4).

Parent topic: aclmdlDesc