DML_ELEMENT_WISE_CLIP_GRAD_OPERATOR_DESC structure (directml.h)

Computes backpropagation gradients for element-wise clip.

f(x, gradient) = if x <= Min then 0
                 if x >= Max then 0
                 else        then gradient

This operator supports in-place execution, meaning OutputTensor is permitted to alias InputTensor during binding.

Syntax

struct DML_ELEMENT_WISE_CLIP_GRAD_OPERATOR_DESC {
  const DML_TENSOR_DESC *InputTensor;
  const DML_TENSOR_DESC *InputGradientTensor;
  const DML_TENSOR_DESC *OutputGradientTensor;
  FLOAT                 Min;
  FLOAT                 Max;
};

Members

InputTensor

Type: const DML_TENSOR_DESC*

The input feature tensor. This is typically the same tensor that was provided as the InputTensor to DML_ELEMENT_WISE_CLIP_OPERATOR_DESC in the forward pass.

InputGradientTensor

Type: const DML_TENSOR_DESC*

The incoming gradient tensor. This is typically obtained from the output of backpropagation of a preceding layer. Typically this tensor would have the same sizes as the output of the corresponding DML_OPERATOR_ELEMENT_WISE_CLIP in the forward pass.

OutputGradientTensor

Type: const DML_TENSOR_DESC*

An output tensor containing the backpropagated gradients. Typically this tensor would have the same sizes as the input of the corresponding DML_OPERATOR_ELEMENT_WISE_CLIP in the forward pass.

Min

Type: FLOAT

The minimum value. If x is at or below this value, then the gradient result is 0.

Max

Type: FLOAT

The maximum value. If x is at or above this value, then the gradient result is 0.

Remarks

Availability

This operator was introduced in DML_FEATURE_LEVEL_3_1.

Tensor constraints

InputGradientTensor, InputTensor, and OutputGradientTensor must have the same DataType, DimensionCount, and Sizes.

Tensor support

DML_FEATURE_LEVEL_5_0 and above

DML_FEATURE_LEVEL_3_1 and above

Requirements