DML_PADDING_OPERATOR_DESC structure (directml.h)

Inflates the input tensor with constant or mirrored values on the edges, and writes the result to the output.

Syntax

struct DML_PADDING_OPERATOR_DESC {
  const DML_TENSOR_DESC *InputTensor;
  const DML_TENSOR_DESC *OutputTensor;
  DML_PADDING_MODE      PaddingMode;
  FLOAT                 PaddingValue;
  UINT                  DimensionCount;
  const UINT            *StartPadding;
  const UINT            *EndPadding;
};

Members

InputTensor

Type: const DML_TENSOR_DESC*

A tensor containing the input data.

OutputTensor

Type: const DML_TENSOR_DESC*

A tensor containing the output data. For each dimension i, OutputTensor.Sizes[i] = InputTensor.Sizes[i] + StartPadding[i] + EndPadding[i].

PaddingMode

Type: DML_PADDING_MODE

The padding mode to use when filling the padding regions.

• DML_PADDING_MODE_CONSTANT. Uses a single constant value defined by PaddingValue for all padding values (see Example 1).
• DML_PADDING_MODE_EDGE. For each dimension, use the edge values of that dimension for all padding values (see Example 2).
• DML_PADDING_MODE_REFLECTION. Mirror the values of the tensor as if we folded it right on the edges, which means that edges are not mirrored. Note that StartPadding[i] >= InputTensor.Sizes[i], and EndPadding[i] >= InputTensor.Sizes[i] is valid, which means that we can mirror new padding regions periodically by folding them over previous padding regions (see Example 3).
• DML_PADDING_MODE_SYMMETRIC. Similar to DML_PADDING_MODE_REFLECTION, but edges are also mirrored. Note that StartPadding[i] > InputTensor.Sizes[i], and EndPadding[i] > InputTensor.Sizes[i] is valid, which means that we can mirror new padding regions periodically by folding them over previous padding regions (see Example 4). This mode was introduced in feature level DML_FEATURE_LEVEL_3_0.

PaddingValue

Type: FLOAT

The padding value to use when PaddingMode == DML_PADDING_MODE_CONSTANT. This value is ignored for other padding modes. Note that if the DataType of the tensors is not DML_TENSOR_DATA_TYPE_FLOAT16 or DML_TENSOR_DATA_TYPE_FLOAT32, then the value might be truncated (for example, 10.6 will become 10).

DimensionCount

Type: UINT

The size of the arrays pointed to by StartPadding and EndPadding. This value must be the same value as the dimension count of InputTensor and OutputTensor.

StartPadding

Type: _Field_size_(DimensionCount) const UINT*

The sizes of the padding regions to add at the beginning of each dimension. For each dimension i, StartPadding[i] = OutputTensor.Sizes[i] - InputTensor.Sizes[i] - EndPadding[i].

EndPadding

Type: _Field_size_(DimensionCount) const UINT*

The sizes of the padding regions to add at the end of each dimension. For each dimension i, EndPadding[i] = OutputTensor.Sizes[i] - InputTensor.Sizes[i] - StartPadding[i].

Examples

Example 1

PaddingMode: DML_PADDING_MODE_CONSTANT
PaddingValue: 9
StartPadding: {0, 0, 1, 2}
EndPadding: {0, 0, 3, 4}

InputTensor: (Sizes:{1, 1, 4, 4}, DataType:FLOAT32)
[[[[1, 2, 3, 4],
   [5, 6, 7, 8],
   [1, 2, 3, 4],
   [5, 6, 7, 8]]]]

OutputTensor: (Sizes:{1, 1, 8, 10}, DataType:FLOAT32)
[[[[9, 9, 9, 9, 9, 9, 9, 9, 9, 9]
   [9, 9, 1, 2, 3, 4, 9, 9, 9, 9],
   [9, 9, 5, 6, 7, 8, 9, 9, 9, 9],
   [9, 9, 1, 2, 3, 4, 9, 9, 9, 9],
   [9, 9, 5, 6, 7, 8, 9, 9, 9, 9],
   [9, 9, 9, 9, 9, 9, 9, 9, 9, 9],
   [9, 9, 9, 9, 9, 9, 9, 9, 9, 9],
   [9, 9, 9, 9, 9, 9, 9, 9, 9, 9]]]]

Example 2

PaddingMode: DML_PADDING_MODE_EDGE
StartPadding: {0, 0, 1, 2}
EndPadding: {0, 0, 3, 4}

InputTensor: (Sizes:{1, 1, 4, 4}, DataType:FLOAT32)
[[[[1, 2, 3, 4],
   [5, 6, 7, 8],
   [1, 2, 3, 4],
   [5, 6, 7, 8]]]]

OutputTensor: (Sizes:{1, 1, 8, 10}, DataType:FLOAT32)
[[[[1, 1, 1, 2, 3, 4, 4, 4, 4, 4]
   [1, 1, 1, 2, 3, 4, 4, 4, 4, 4],
   [5, 5, 5, 6, 7, 8, 8, 8, 8, 8],
   [1, 1, 1, 2, 3, 4, 4, 4, 4, 4],
   [5, 5, 5, 6, 7, 8, 8, 8, 8, 8],
   [5, 5, 5, 6, 7, 8, 8, 8, 8, 8],
   [5, 5, 5, 6, 7, 8, 8, 8, 8, 8],
   [5, 5, 5, 6, 7, 8, 8, 8, 8, 8]]]]

Example 3

PaddingMode: DML_PADDING_MODE_REFLECTION
StartPadding: {0, 0, 1, 2}
EndPadding: {0, 0, 3, 4}

InputTensor: (Sizes:{1, 1, 4, 4}, DataType:FLOAT32)
[[[[1, 2, 3, 4],
   [5, 6, 7, 8],
   [1, 2, 3, 4],
   [5, 6, 7, 8]]]]

OutputTensor: (Sizes:{1, 1, 8, 10}, DataType:FLOAT32)
[[[[7, 6, 5, 6, 7, 8, 7, 6, 5, 6]
   [3, 2, 1, 2, 3, 4, 3, 2, 1, 2],
   [7, 6, 5, 6, 7, 8, 7, 6, 5, 6],
   [3, 2, 1, 2, 3, 4, 3, 2, 1, 2],
   [7, 6, 5, 6, 7, 8, 7, 6, 5, 6],
   [3, 2, 1, 2, 3, 4, 3, 2, 1, 2],
   [7, 6, 5, 6, 7, 8, 7, 6, 5, 6],
   [3, 2, 1, 2, 3, 4, 3, 2, 1, 2]]]]

Example 4 (starting from DML_FEATURE_LEVEL_3_0)

PaddingMode: DML_PADDING_MODE_SYMMETRIC
StartPadding: {0, 0, 1, 2}
EndPadding: {0, 0, 3, 4}

InputTensor: (Sizes:{1, 1, 4, 4}, DataType:FLOAT32)
[[[[1, 2, 3, 4],
   [5, 6, 7, 8],
   [1, 2, 3, 4],
   [5, 6, 7, 8]]]]

OutputTensor: (Sizes:{1, 1, 8, 10}, DataType:FLOAT32)
[[[[2, 1, 1, 2, 3, 4, 4, 3, 2, 1]
   [2, 1, 1, 2, 3, 4, 4, 3, 2, 1],
   [6, 5, 5, 6, 7, 8, 8, 7, 6, 5],
   [2, 1, 1, 2, 3, 4, 4, 3, 2, 1],
   [6, 5, 5, 6, 7, 8, 8, 7, 6, 5],
   [6, 5, 5, 6, 7, 8, 8, 7, 6, 5],
   [2, 1, 1, 2, 3, 4, 4, 3, 2, 1],
   [6, 5, 5, 6, 7, 8, 8, 7, 6, 5]]]]

Availability

This operator was introduced in DML_FEATURE_LEVEL_1_0.

Tensor constraints

InputTensor and OutputTensor must have the same DataType and DimensionCount.

Tensor support

DML_FEATURE_LEVEL_5_0 and above

DML_FEATURE_LEVEL_3_1 and above

DML_FEATURE_LEVEL_2_1 and above

DML_FEATURE_LEVEL_1_0 and above

Requirements