TIDL (TI Deep Learning) Kernel APIs

Detailed Description

TIDL Network Kernels.

Data Structures
struct	tivxTIDLJ7Params
	TIDL params structure. More...

Functions
static void	tivx_tidl_j7_params_init (tivxTIDLJ7Params *tidlParams)
	TIDL params initialization.
VX_API_ENTRY vx_node VX_API_CALL	tivxTIDLNode (vx_graph graph, vx_kernel kernel, vx_reference appParams[], vx_tensor input_tensors[], vx_tensor output_tensors[])
	[Graph] Creates a TIDL Node. More...
vx_kernel	tivxAddKernelTIDL (vx_context context, uint32_t num_input_tensors, uint32_t num_output_tensors)
	Used for the Application to create the tidl kernel from the context. More...

Function Documentation

tivxTIDLNode()

VX_API_ENTRY vx_node VX_API_CALL tivxTIDLNode	(	vx_graph	graph,
		vx_kernel	kernel,
		vx_reference	appParams[],
		vx_tensor	input_tensors[],
		vx_tensor	output_tensors[]
	)

[Graph] Creates a TIDL Node.

Parameters

[in]	graph	Reference to vx_graph.
[in]	kernel	Reference to vx_kernel.
[in,out]	appParams	is an array of 5 parameters: config vx_user_data_object type corresponding to the configuration (named string: sTIDL_IOBufDesc_t) network vx_user_data_object type corresponding to the network (named string: TIDL_network) createParams: vx_user_data_object type corresponding to the structure TIDL_CreateParams. This structure contains create-time parameters and are used only one time to initialize the node. Currently members quantHistoryParam1, quantHistoryParam2, quantMargin need to be initialized by the application. inArgs: vx_user_data_object type corresponding to the structure TIDL_InArgs. Used to pass input parameters that goes in effect during the node's process call when graph is executed. These parameters can be updated before each frame execution of a graph. Currently the list of parameters is empty. So inArgs is just a placeholder for future extension. outArgs: vx_user_data_object type corresponding to the structure TIDL_outArgs. Used to collect output parameters that are returned by the node's process call when graph is executed. These parameters are refreshed after each frame execution of a graph. Currently the list of parameters is empty. So outArgs is just a placeholder for future extension.
[in]	input_tensors	Array of input tensors This parameter is ignored when the first layer of the network is a data layer, which is most of the time. Only networks that are dependent on the output of a previous networks have first layer that are not data layer.
[out]	output_tensors	Array of output tensors

Returns

vx_node.

Return values

vx_node

A node reference. Any possible errors preventing a successful creation should be checked using vxGetStatus

tivxAddKernelTIDL()

vx_kernel tivxAddKernelTIDL	(	vx_context	context,
		uint32_t	num_input_tensors,
		uint32_t	num_output_tensors
	)

Used for the Application to create the tidl kernel from the context.

Parameters

[in]	context	OpenVX context which must be created using vxCreateContext
[in]	num_input_tensors	Number of input vx_tensor objects to be created
[in]	num_output_tensors	Number of output vx_tensor objects to be created

Returns

Handle to vx_kernel object if successful, NULL otherwise

Note

The caller of this function should check status of the return vx_kernel handle and if found to be NULL must handle or propagate error appropriately.