[object Object]

[object Object][object Object]undefined[object Object]

• Description: Implements the backpropagation of the grouped bias addition ([object Object]). The extended API of this API is .
• Formula:[object Object] (1) If [object Object] is available:

$$out(G,H) = \begin{cases} \sum_{i=groupIdxOptional(j-1)}^{groupIdxOptional(j)} gradY(i, H), & 1 \leq j \leq G-1 \\ \sum_{i=0}^{groupIdxOptional(j)} gradY(i, H), & j = 0 \end{cases}$$

  [object Object] has two dimensions. [object Object] indicates the size of the last dimension of [object Object], while [object Object] indicates the size of dimension 0 of [object Object]. That is, [object Object] has [object Object] numbers, and [object Object] indicates the size of the jth dimension. After computation, [object Object] is 2-dimensional, with the shape [object Object].[object Object]   (2) If [object Object] is unavailable:

$$out(G, H) = \sum_{i=0}^{C} gradY(G, i, H)$$

  [object Object] has three dimensions. [object Object], [object Object], and [object Object] indicate the sizes of dimensions 0 to 2 of [object Object]. After computation, [object Object] is 2-dimensional, with the shape [object Object].

• Example:[object Object] (1) If [object Object] is available:[object Object] The shape of [object Object] is (1000, 30), and the shape of [object Object] is (400, 600, 1000). [object Object] is divided into three groups, and the accumulated number of rows in each group is 400, 200, and 400 respectively. After computation, the shape of [object Object] is (3, 30).[object Object] (2) If [object Object] is unavailable:[object Object] The shape of [object Object] is (10, 100, 30). [object Object] is divided into 10 groups. The accumulated number of rows in each group is 100. After computation, the shape of [object Object] is (10, 30).

[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*, computation input): required parameter, backpropagation gradient, [object Object] in the formula, aclTensor on the device. The data type can be FLOAT, FLOAT16, or BFLOAT16. If [object Object] is available, the shape can only be 2-dimensional. If [object Object] is not available, the shape can only be 3-dimensional. are supported. The can be ND.
  • [object Object] (aclTensor*, computation input): optional parameter, end position of each group, [object Object] in the formula, aclTensor on the device. The data type can be INT32 or INT64. The shape supports only one dimension. are supported. The can be ND.
  • [object Object] (aclTensor*, output): gradient of bias, out in the formula, aclTensor on the device. The data type can be FLOAT, FLOAT16, or BFLOAT16. The data type must be the same as that of [object Object]. The shape supports only two dimensions. are supported. The supports ND.
  • [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.
• • [object Object] supports a maximum of 2048 numbers.
• • If [object Object] is available, ensure that the tensor data is sorted in ascending order and the last value is equal to the size of dimension 0 of [object Object].
• • If [object Object] is available, ensure that the tensor values are not greater than the maximum INT32 value and are not negative numbers.

[object Object]

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