Edit

DML_ELEMENT_WISE_QUANTIZED_LINEAR_ADD_OPERATOR_DESC structure (directml.h)

  • Article

Adds every element in ATensor to its corresponding element in BTensor, placing the result into the corresponding element of OutputTensor. Values contained in ATensor and BTensor are dequantized using the following equation, and then added and requantized.

AValue = (A - AZeroPoint) * AScale
BValue = (B - BZeroPoint) * BScale

OutputValue = AValue + BValue

// For uint8 output, Min = 0, Max = 255
// For int8 output, Min = -128, Max = 127
OutputTensor = clamp(round(OutputValue / OutputScale) + OutputZeroPoint, Min, Max)

Syntax

struct DML_ELEMENT_WISE_QUANTIZED_LINEAR_ADD_OPERATOR_DESC {
  const DML_TENSOR_DESC *ATensor;
  const DML_TENSOR_DESC *AScaleTensor;
  const DML_TENSOR_DESC *AZeroPointTensor;
  const DML_TENSOR_DESC *BTensor;
  const DML_TENSOR_DESC *BScaleTensor;
  const DML_TENSOR_DESC *BZeroPointTensor;
  const DML_TENSOR_DESC *OutputScaleTensor;
  const DML_TENSOR_DESC *OutputZeroPointTensor;
  const DML_TENSOR_DESC *OutputTensor;
};

Members

ATensor

Type: const DML_TENSOR_DESC*

A tensor containing the left-hand-side inputs.

AScaleTensor

Type: const DML_TENSOR_DESC*

The tensor containing the desired scale factor for ATensor. The expected number of elements in AScaleTensor is 1.

AZeroPointTensor

Type: _Maybenull_ const DML_TENSOR_DESC*

The tensor containing the desired zero point for ATensor. The expected number of elements in AZeroPointTensor is 1. AZeroPointTensor is an optional tensor that defaults to 0 if not provided.

BTensor

Type: const DML_TENSOR_DESC*

A tensor containing the right-hand-side inputs.

BScaleTensor

Type: const DML_TENSOR_DESC*

The tensor containing the desired scale factor for BTensor. The expected number of elements in BScaleTensor is 1.

BZeroPointTensor

Type: _Maybenull_ const DML_TENSOR_DESC*

The tensor containing the desired zero point for BTensor. The expected number of elements in BZeroPointTensor is 1. BZeroPointTensor is an optional tensor that defaults to 0 if not provided.

OutputScaleTensor

Type: const DML_TENSOR_DESC*

The tensor containing the desired scale factor for OutputTensor. This is an input tensor defining the output quantization scale factor to use while quantizing the output values. The expected number of elements in OutputScaleTensor is 1.

OutputZeroPointTensor

Type: _Maybenull_ const DML_TENSOR_DESC*

The tensor containing the desired zero point for OutputTensor. This is an input tensor defining the output quantization zero point to use while quantizing the output values. The expected number of elements in OutputZeroPointTensor is 1. OutputZeroPointTensor is an optional tensor that defaults to 0 if not provided.

OutputTensor

Type: const DML_TENSOR_DESC*

The output tensor to write the results to.

Remarks

Availability

This operator was introduced in DML_FEATURE_LEVEL_4_0.

Tensor constraints

  • AScaleTensor, ATensor, AZeroPointTensor, BScaleTensor, BTensor, BZeroPointTensor, OutputScaleTensor, OutputTensor, and OutputZeroPointTensor must have the same DimensionCount.
  • BTensor and BZeroPointTensor must have the same DataType.
  • OutputTensor and OutputZeroPointTensor must have the same DataType.
  • ATensor and AZeroPointTensor must have the same DataType.

Tensor support

Tensor Kind Supported dimension counts Supported data types
ATensor Input 1 to 8 INT8, UINT8
AScaleTensor Input 1 to 8 FLOAT32
AZeroPointTensor Optional input 1 to 8 INT8, UINT8
BTensor Input 1 to 8 INT8, UINT8
BScaleTensor Input 1 to 8 FLOAT32
BZeroPointTensor Optional input 1 to 8 INT8, UINT8
OutputScaleTensor Input 1 to 8 FLOAT32
OutputZeroPointTensor Optional input 1 to 8 INT8, UINT8
OutputTensor Output 1 to 8 INT8, UINT8

Requirements

Requirement Value
Minimum supported client Windows Build 22000
Minimum supported server Windows Build 22000
Header directml.h

See also