[object Object][object Object]

[object Object]undefined[object Object]

Description: Performs backpropagation of the Unfold operator and computes its gradient.

The Unfold operator computes all slices whose size is [object Object] in dimension [object Object] based on the input parameter [object Object]. The step between two slices is given by [object Object]. If [object Object] is the size of dimension [object Object] of the input parameter self, the size of dimension [object Object] in the returned tensor is $(sizedim – size)/step + 1$. An additional dimension whose size is [object Object] is added to the returned tensor.

The shape of the input [object Object] of the UnfoldGrad operator is the shape of the forward output of the Unfold operator. The shape of the input [object Object] is the shape of the forward input [object Object] of the Unfold operator. The shape of the output [object Object] of the UnfoldGrad operator is the shape of the forward input [object Object] of the Unfold operator.

Examples:

[object Object]

Each operator has calls. First, [object Object] is called to obtain the workspace size required for computation and the executor that contains the operator computation process. Then, [object Object] is called to perform computation.

• [object Object]
• [object Object]

[object Object]

• Parameters

  • [object Object] (aclTensor*, compute input): [object Object] on the device, indicating the gradient update coefficient. The shape is (..., (sizedim – size)/step + 1, size). The [object Object] dimension of [object Object] must be equal to $(inputSizes[dim] – size)/step + 1$, and the size of [object Object] must be equal to size of inputSizes plus 1. The data type can be FLOAT, FLOAT16, or BFLOAT16. The can be ND.
  • [object Object] (aclIntArray*, compute input): [object Object] on the host, indicating the shape of the output tensor. The value is (..., sizedim). The size of [object Object] is less than or equal to 8. The data type can be INT64. The can be ND.
  • [object Object] (int64_t, compute input): [object Object] in the formula. Dimension. [object Object] must be greater than or equal to 0 and less than the size of [object Object].
  • [object Object] (int64_t, compute input): [object Object] in the formula. Size of each slice. [object Object] must be greater than 0 and less than or equal to the [object Object] dimension of [object Object].
  • [object Object] (int64_t, compute input): [object Object] in the formula. Indicates the step between slices. The value of [object Object] must be greater than 0.
  • [object Object] (aclTensor*, computation output): gradient of Unfold. It is [object Object] on the device. The shape is [object Object]. The data type can be FLOAT, FLOAT16, or BFLOAT16, and must be the same as that of gradOut.
  • [object Object] (uint64_t*, output): size of the workspace to be allocated on the device.
  • [object Object] (aclOpExecutor**, output): operator executor, containing the operator computation process.

• Returns:

  [object Object]: status code. For details, see .

[object Object]

• Parameters

  • [object Object] (void*, input): address of the workspace to be allocated on the device.
  • [object Object] (uint64_t, input): size of the workspace to be allocated on the device, which is obtained by the first-phase API [object Object].
  • [object Object] (aclOpExecutor*, input): operator executor, containing the operator computation process.
  • [object Object] (aclrtStream, input): stream for executing the task.

• Returns:

  [object Object]: status code. For details, see .

[object Object]

• Deterministic computing:
  • [object Object] defaults to a deterministic implementation.

1. The shape of [object Object] must meet the following constraints: (1) The [object Object] dimension of [object Object] is equal to (inputSizes[dim] – size)/step + 1. (2) The [object Object] of [object Object] is equal to the size of [object Object] plus 1.
2. Requirements for [object Object], [object Object], and [object Object]: (1) [object Object] must be greater than or equal to 0 and less than the [object Object] of [object Object]. (2) [object Object] must be greater than 0 and less than or equal to the [object Object] dimension of [object Object]. (3) [object Object] is greater than 0. (4) [object Object] and [object Object] are less than or equal to 100.

[object Object]

The following example is for reference only. For details, see .