DML_DIAGONAL_MATRIX1_OPERATOR_DESC structure (directml.h)

Generates an identity-like matrix with ones (or other explicit value) along the given diagonal span, with other elements being filled with either the input values or zeros (if no InputTensor is passed). The diagonal values may be shifted anywhere between DiagonalFillBegin and DiagonalFillEnd, where a value greater than zero shifts all values to the right, and less than zero shifts them to the left. This generator operator is useful for models to avoid storing a large constant tensor. Any leading dimensions before the last two are treated as a batch count, meaning that the tensor is treated as stack of 2D matrices. This operator performs the following pseudocode:

// Given each current coordinate's x, figure out the corresponding x on the top (0th) row.
// Then fill it with Value if that x coordinate is either within the fill bounds, or outside
// the fill bounds when inverted. Otherwise, use the original input value or zero.

for each coordinate in OutputTensor
    topX = coordinate.x - coordinate.y
    useValue = (DiagonalFillEnd >= DiagonalFillBegin) ^ (topX >= DiagonalFillBegin) ^ (topX < DiagonalFillEnd)
    OutputTensor[coordinate] = if useValue then Value
                               else if InputTensor not null then InputTensor[coordinate]
                               else 0
endfor

Syntax

struct DML_DIAGONAL_MATRIX1_OPERATOR_DESC
{
    _Maybenull_ const DML_TENSOR_DESC* InputTensor;
    const DML_TENSOR_DESC* OutputTensor;
    DML_TENSOR_DATA_TYPE ValueDataType;
    DML_SCALAR_UNION Value;
    INT DiagonalFillBegin;
    INT DiagonalFillEnd;
};

Members

InputTensor

Type: _Maybenull_ const DML_TENSOR_DESC*

An optional input tensor.

OutputTensor

Type: const DML_TENSOR_DESC*

The tensor to write the results to. The dimensions are { Batch1, Batch2, OutputHeight, OutputWidth }. The height and width don't need to be square.

ValueDataType

Type: DML_TENSOR_DATA_TYPE

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

Value

Type: DML_SCALAR_UNION

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

DiagonalFillBegin

Type: INT

The lower limit (beginning inclusive) of the range to fill with Value. If the beginning and ending are inverted (begin > end), then the fill will be reversed.

DiagonalFillEnd

Type: INT

The upper limit (ending exclusive) of the range to fill with Value. If the beginning and ending are inverted (begin > end), then the fill will be reversed.

Examples

Default identity matrix:

InputTensor: none
ValueDataType: FLOAT32
Value: 7
DiagonalFillBegin: 0
DiagonalFillEnd: 1
OutputTensor: (Sizes:{4,5} DataType:FLOAT32)
    [[1, 0, 0, 0, 0],
     [0, 1, 0, 0, 0],
     [0, 0, 1, 0, 0],
     [0, 0, 0, 1, 0]]

Fill a 3-element wide diagonal strip:

InputTensor: none
ValueDataType: FLOAT32
Value: 7
DiagonalFillBegin: 0
DiagonalFillEnd: 3
OutputTensor: (Sizes:{4,5} DataType:FLOAT32)
    [[7, 7, 7, 0, 0],
     [0, 7, 7, 7, 0],
     [0, 0, 7, 7, 7],
     [0, 0, 0, 7, 7]]

Keep top-right diagonal half (the strictly upper triangle), zeroing the bottom left.

InputTensor: (Sizes:{4,5} DataType:FLOAT32)
    [[4, 7, 3, 7, 9],
     [1, 2, 8, 6, 9],
     [9, 4, 1, 8, 7],
     [4, 3, 4, 2, 4]]
ValueDataType: FLOAT32
Value: 0
DiagonalFillBegin: INT32_MIN
DiagonalFillEnd: 1
OutputTensor: (Sizes:{4,5} DataType:FLOAT32)
    [[0, 7, 3, 7, 9],
     [0, 0, 8, 6, 9],
     [0, 0, 0, 8, 7],
     [0, 0, 0, 0, 4]]

Keep only the values along the diagonal, zeroing all others.

InputTensor: (Sizes:{4,5} DataType:FLOAT32)
    [[4, 7, 3, 7, 9],
     [1, 2, 8, 6, 9],
     [9, 4, 1, 8, 7],
     [4, 3, 4, 2, 4]]
ValueDataType: FLOAT32
Value: 0
DiagonalFillBegin: 1
DiagonalFillEnd: 0
OutputTensor: (Sizes:{4,5} DataType:FLOAT32)
    [[4, 0, 0, 0, 0],
     [0, 2, 0, 0, 0],
     [0, 0, 1, 0, 0],
     [0, 0, 0, 2, 0]]

Remarks

This operator is equivalent to DML_DIAGONAL_MATRIX_OPERATOR_DESC when DiagonalFillBegin is set to DML_OPERATOR_DIAGONAL_MATRIX::Offset, and DiagonalFillEnd is set to DML_OPERATOR_DIAGONAL_MATRIX::Offset + 1.

Availability

This operator was introduced in DML_FEATURE_LEVEL_5_1.

Tensor constraints

InputTensor and OutputTensor must have the same DataType and DimensionCount.

Tensor support

Requirements