DML_REDUCE_OPERATOR_DESC structure (directml.h)

Outputs the reduction of elements (sum, product, minimum, and so on) within one or more dimensions of the input tensor.

Each output element is the result of applying a reduction function on a subset of the input tensor. A reduction function, such as sum, maps N input elements to a single output element. The input elements involved in each reduction are determined by the provided input axes: N is equal to the product of the sizes of the reduced axes. If all input axes are specified, then the operator performs a reduction on the entire input tensor and produces a single output element.

Syntax

struct DML_REDUCE_OPERATOR_DESC {
  DML_REDUCE_FUNCTION   Function;
  const DML_TENSOR_DESC *InputTensor;
  const DML_TENSOR_DESC *OutputTensor;
  UINT                  AxisCount;
  const UINT            *Axes;
};

Members

Function

Type: DML_REDUCE_FUNCTION

Specifies the reduction function to apply to the input.

InputTensor

Type: const DML_TENSOR_DESC*

The tensor to read from.

OutputTensor

Type: const DML_TENSOR_DESC*

The tensor to write the results to. Each output element is the result of a reduction on a subset of elements from the InputTensor.

• DimensionCount must match InputTensor.DimensionCount (the rank of the input tensor is preserved).
• Sizes must match InputTensor.Sizes, except for dimensions included in the reduced Axes, which must be size 1.

AxisCount

Type: UINT

The number of axes to reduce. This field determines the size of the Axes array.

Axes

Type: _Field_size_(AxisCount) const UINT*

The axes along which to reduce. Values must be in the range [0, InputTensor.DimensionCount - 1].

Examples

The following examples all use this same two-dimensional input tensor.

InputTensor: (Sizes:{3, 3}, DataType:FLOAT32)
[[1, 2, 3],
 [3, 0, 4],
 [2, 4, 2]]

Example 1. Applying sum to columns

Function: DML_REDUCE_FUNCTION_SUM
AxisCount: 1
Axes: {0}
OutputTensor: (Sizes:{1, 3}, DataType:FLOAT32)
[[6,  // sum({1, 3, 2})
  6,  // sum({2, 0, 4})
  9]] // sum({3, 4, 2})

Example 2. Applying sum to rows

Function: DML_REDUCE_FUNCTION_SUM
AxisCount: 1
Axes: {1}
OutputTensor: (Sizes:{3, 1}, DataType:FLOAT32)
[[6], // sum({1, 2, 3})
 [7], // sum({3, 0, 4})
 [8]] // sum({2, 4, 2})

Example 3.Applying sum to all axes (the entire tensor)

Function: DML_REDUCE_FUNCTION_SUM
AxisCount: 2
Axes: {0, 1}
OutputTensor: (Sizes:{1, 1}, DataType:FLOAT32)
[[21]]  // sum({1, 2, 3, 3, 0, 4, 2, 5, 2})

Availability

This operator was introduced in DML_FEATURE_LEVEL_1_0.

Tensor constraints

• The input and output tensor data types must match except when using the ARGMAX and ARGMIN functions, which always output an integral data type.
• The output sizes must be the same as the input sizes except for the reduced axes, which must be 1.

Tensor support according to function

ARGMIN and ARGMAX

DML_FEATURE_LEVEL_4_1 and above

DML_FEATURE_LEVEL_3_0 and above

DML_FEATURE_LEVEL_2_1 and above

DML_FEATURE_LEVEL_1_0 and above

AVERAGE, L2, LOG_SUM, and LOG_SUM_EXP

DML_FEATURE_LEVEL_3_0 and above

DML_FEATURE_LEVEL_1_0 and above

L1 and SUM_SQUARE

DML_FEATURE_LEVEL_5_0 and above

DML_FEATURE_LEVEL_3_0 and above

DML_FEATURE_LEVEL_1_0 and above

MIN and MAX

DML_FEATURE_LEVEL_5_0 and above

DML_FEATURE_LEVEL_3_0 and above

DML_FEATURE_LEVEL_2_1 and above

DML_FEATURE_LEVEL_1_0 and above

MULTIPLY and SUM

DML_FEATURE_LEVEL_5_0 and above

DML_FEATURE_LEVEL_3_0 and above

DML_FEATURE_LEVEL_2_1 and above

DML_FEATURE_LEVEL_1_0 and above

Requirements

See also

Feature level 3_0 introduced these stand-alone operators that supersede the functionality available with DML_REDUCE_FUNCTION_ARGMAX and DML_REDUCE_FUNCTION_ARGMIN.

• DML_ARGMAX_OPERATOR_DESC
• DML_ARGMIN_OPERATOR_DESC