Edit

DML_FILL_VALUE_CONSTANT_OPERATOR_DESC structure (directml.h)

  • Article

Fills a tensor with the given constant Value. This operator performs the following pseudocode.

for each coordinate in OutputTensor
    OutputTensor[coordinate] = Value
endfor

Syntax

struct DML_FILL_VALUE_CONSTANT_OPERATOR_DESC {
  const DML_TENSOR_DESC *OutputTensor;
  DML_TENSOR_DATA_TYPE  ValueDataType;
  DML_SCALAR_UNION      Value;
};

Members

OutputTensor

Type: const DML_TENSOR_DESC*

The tensor to write the results to. This tensor may have any size.

ValueDataType

Type: DML_TENSOR_DATA_TYPE

The data type of the Value field, which must match OutputTensor.DataType.

Value

Type: DML_SCALAR_UNION

A constant value to fill the output, with ValueDataType determining how to interpret the field.

Examples

Value = 13.0

OutputTensor: (Sizes:{1,1,2,4}, DataType:FLOAT32)
    [[[[13.0f, 13.0f, 13.0f, 13.0f],
       [13.0f, 13.0f, 13.0f, 13.0f]]]]

Availability

This operator was introduced in DML_FEATURE_LEVEL_2_1.

Tensor support

DML_FEATURE_LEVEL_4_1 and above

Tensor Kind Supported dimension counts Supported data types
OutputTensor Output 1 to 8 FLOAT64, FLOAT32, FLOAT16, INT64, INT32, INT16, INT8, UINT64, UINT32, UINT16, UINT8

DML_FEATURE_LEVEL_4_0 and above

Tensor Kind Supported dimension counts Supported data types
OutputTensor Output 1 to 8 FLOAT32, FLOAT16, INT32, INT16, INT8, UINT32, UINT16, UINT8

DML_FEATURE_LEVEL_2_1 and above

Tensor Kind Supported dimension counts Supported data types
OutputTensor Output 4 FLOAT32, FLOAT16, INT32, INT16, INT8, UINT32, UINT16, UINT8

Requirements

Requirement Value
Minimum supported client Windows 10 Build 20348
Minimum supported server Windows 10 Build 20348
Header directml.h