[object Object]

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

• Description:
  • When fake_quant_enabled is greater than or equal to 1: Performs fake quantization on the input self with scale and zero_point, limits the value range of the fake-quantized output by quant_min and quant_max, and finally returns the output out as well as the corresponding position mask mask.
  • When fake_quant_enabled is less than 1: Returns out as a clone of self, and mask with all values set to True.
• Formula: When fake_quant_enabled is greater than or equal to 1, computes the temporary variable qval, then calculates out and mask.$$qval = Round(std::nearby\_int(self / scale) + zero\_point)$$ $$out = (Min(quant\_max, Max(quant\_min, qval)) - zero\_point) * scale$$ $$mask = (qval >= quant\_min) \& (qval <= quant\_max)$$

[object Object]

Each operator has calls. First, aclnnFakeQuantPerTensorAffineCachemaskGetWorkspaceSize is called to obtain the input parameters and compute the required workspace size based on the process. Then, aclnnFakeQuantPerTensorAffineCachemask is called to perform computation.

• [object Object]
• [object Object]

[object Object]

• Parameters

  • self (aclTensor*, input): input tensor, corresponding to [object Object] in the formula. This tensor is a device-side aclTensor. The data type can be FLOAT16 or FLOAT32. are supported. The can be ND.
  • scale (aclTensor*, input): scaling coefficient for input fake quantization, corresponding to [object Object] in the formula. This tensor is a device-side aclTensor. The data type can be FLOAT16 or FLOAT32, and the shape must be one-dimensional with a size of 1.
  • zeroPoint (aclTensor*, input): zero-point parameter for input fake quantization, corresponding to [object Object] in the formula. This tensor is a device-side aclTensor. The data type must be INT32, and the shape must be one-dimensional with a size of 1.
  • fakeQuantEnabled (float, input): host-side floating-point value that indicates whether to perform fake quantization.
  • quantMin (int64_t, input): host-side integer that represents the minimum value after fake quantization of the input data. It must be less than or equal to quantMax.
  • quantMax (int64_t, input): host-side integer that represents the maximum value after fake quantization of the input data. It must be greater than or equal to quantMin.
  • out (aclTensor*, output): output tensor. This tensor is a device-side aclTensor. The data type can be FLOAT16 or FLOAT32, and the shape must be the same as that of [object Object]. are supported. The can be ND.
  • mask (aclTensor*, output): mask tensor. This tensor is a device-side aclTensor. The data type must be BOOL, and the shape must be the same as that of [object Object]. are supported. The can be ND.
  • workspaceSize (uint64_t*, output): size of the workspace to be allocated on the device.
  • executor (aclOpExecutor**, output): operator executor, containing the operator computation flow.

• Returns

  aclnnStatus: status code. For details, see .

[object Object]

• Parameters

  • workspace (void*, input): address of the workspace to be allocated on the device.
  • workspaceSize (uint64_t, input): size of the workspace to be allocated on the device, which is obtained via the first-phase API aclnnFakeQuantPerTensorAffineCachemaskGetWorkspaceSize.
  • executor (aclOpExecutor*, input): operator executor, containing the operator computation flow.
  • stream (aclrtStream, input): stream for executing the task.

• Returns

  aclnnStatus: status code. For details, see .

[object Object]

• Deterministic computation:
  • aclnnFakeQuantPerTensorAffineCachemask defaults to a deterministic implementation.

[object Object]

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