DML_ELEMENT_WISE_IF_OPERATOR_DESC structure (directml.h)

Selects elements either from ATensor or BTensor, depending on the value of the corresponding element in ConditionTensor. Non-zero elements of ConditionTensor select from ATensor, while zero-valued elements select from BTensor.

f(cond, a, b) = a, if cond != 0
                b, otherwise

Example:
    [[1, 0], [1, 1]] // ConditionTensor
    [[1, 2], [3, 4]] // ATensor
    [[9, 8], [7, 6]] // BTensor

    [[1, 8], [3, 4]] // Output

Syntax

struct DML_ELEMENT_WISE_IF_OPERATOR_DESC {
  const DML_TENSOR_DESC *ConditionTensor;
  const DML_TENSOR_DESC *ATensor;
  const DML_TENSOR_DESC *BTensor;
  const DML_TENSOR_DESC *OutputTensor;
};

Members

ConditionTensor

Type: const DML_TENSOR_DESC*

The condition tensor to read from.

ATensor

Type: const DML_TENSOR_DESC*

A tensor containing the left-hand side inputs.

BTensor

Type: const DML_TENSOR_DESC*

A tensor containing the right-hand side inputs.

OutputTensor

Type: const DML_TENSOR_DESC*

The output tensor to write the results to.

Remarks

Can be used to functionally build up other aggregate operators, such as LeakyRelu. Here's an illustration in pseudo-code (not the most efficient way, but possible): LeakyRelu(x) = If(Less(x, 0), Mul(x, alpha), x).

Availability

This operator was introduced in DML_FEATURE_LEVEL_2_0.

Tensor constraints

• ATensor, BTensor, ConditionTensor, and OutputTensor must have the same DimensionCount and Sizes.
• ATensor, BTensor, and OutputTensor must have the same DataType.

Tensor support

DML_FEATURE_LEVEL_5_0 and above

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

DML_FEATURE_LEVEL_3_0 and above

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

DML_FEATURE_LEVEL_2_0 and above

Tensor	Kind	Supported dimension counts	Supported data types
ConditionTensor	Input	4	UINT8
ATensor	Input	4	FLOAT16
BTensor	Input	4	FLOAT16
OutputTensor	Output	4	FLOAT16

Requirements

Requirement	Value
Minimum supported client	Windows 10, version 2004 (10.0; Build 19041)
Minimum supported server	Windows Server, version 2004 (10.0; Build 19041)
Header	directml.h