DirectML API Reference
Explore the comprehensive set of APIs available for DirectML, enabling you to build high-performance machine learning models and applications on Windows.
Operators
DirectML provides a rich set of operators for common deep learning tasks. Below is a list of key operators, with more details available in the full API documentation.
-
DML_GEMM_OPERATORPerforms general matrix multiplication (GEMM).
-
DML_CONVOLUTION_OPERATORApplies a convolution operation.
-
DML_ACTIVATION_RELU_OPERATORApplies the Rectified Linear Unit (ReLU) activation function.
-
DML_MAX_POOLING_OPERATORPerforms max pooling.
-
DML_ELEMENT_WISE_ADD_OPERATORPerforms element-wise addition.
-
DML_ELEMENT_WISE_MULTIPLY_OPERATORPerforms element-wise multiplication.
-
DML_REDUCE_SUM_OPERATORComputes the sum of elements along specified axes.
Enumerations
Key enumerations used throughout the DirectML API.
-
DML_STATUSDefines the possible return values for DirectML functions, indicating success or failure.
-
DML_TENSOR_DATA_TYPESpecifies the data types supported for tensors, such as
FLOAT32,FLOAT16, andUINT8.
Structures
Core data structures used for defining tensors, operators, and device configurations.
-
DML_TENSOR_DESCDescribes a DirectML tensor, including its dimensions, data type, and format.
-
DML_OPERATOR_DESCA generic structure to describe various types of DirectML operators.
-
DML_BINDING_DESCDescribes the bindings between operator inputs, outputs, and persistent resources.
Interfaces
Key DirectML interfaces for managing devices, operators, and execution.
-
IDMLDeviceRepresents a DirectML device, used for creating operators and managing resources.
-
IDMLOperatorRepresents a compiled DirectML operator ready for execution.
-
IDMLCommandRecorderRecords DirectML commands for later execution on a command queue.