Stream

The main entry-point to the API is the VPIStream object. More...

Typedefs
typedef struct VPIStreamImpl *	VPIStream
	A handle to a stream.

Enumerations
enum	VPIDeviceType {   VPI_DEVICE_TYPE_INVALID,   VPI_DEVICE_TYPE_CPU,   VPI_DEVICE_TYPE_CUDA,   VPI_DEVICE_TYPE_PVA }
	Device types. More...

Functions
VPIStatus	vpiStreamCreate (VPIDeviceType devType, VPIStream *stream)
	Create a stream instance. More...
VPIStatus	vpiStreamWrapCuda (CUstream cudaStream, VPIStream *stream)
	Wraps an existing CUDA stream in a VPI stream for CUDA backend. More...
void	vpiStreamDestroy (VPIStream stream)
	Destroy a stream instance and deallocate all HW resources. More...
VPIStatus	vpiStreamGetDeviceType (VPIStream stream, VPIDeviceType *deviceType)
	Returns the backend type associated with the given stream. More...
VPIStatus	vpiStreamSync (VPIStream stream)
	Blocks the calling thread until all submitted commands in this stream queue are done (queue is empty). More...
VPIStatus	vpiStreamWaitFor (VPIStream stream, VPIEvent event)
	Pushes a command that blocks the processing of all future commands by the backend until the event is completed. More...
VPIStatus	vpiStreamGetThreadHandle (VPIStream stream, VPINativeThreadHandle *handle)
	Returns OS-specific handle of the background stream processing thread. More...

Detailed Description

The main entry-point to the API is the VPIStream object.

This object represents:

• A backend-specific resource handle - creating a VPIStream instance for a specific backend allocates system resources and sets up the hardware to run API functionality.
• A command queue - FIFO queue storing a list of commands to execute. Command might comprise running a particular CV function or signaling an event. This allows the API functions to be executed asynchronously with respect to the calling thread. Invoking any CV function simply pushes a corresponding command to the VPIStream instance command queue and immediately returns. The queued commands get consumed by the underlying backend implementation.
• An owner of device-specific scratch-memory or, so called, function payloads - VPIPayload.

Enumeration Type Documentation

VPIDeviceType

enum VPIDeviceType

#include <vpi/Types.h>

Device types.

Enumerator
VPI_DEVICE_TYPE_INVALID	Invalid backend.
VPI_DEVICE_TYPE_CPU	CPU backend.
VPI_DEVICE_TYPE_CUDA	CUDA backend.
VPI_DEVICE_TYPE_PVA	PVA backend.

Definition at line 553 of file Types.h.

Function Documentation

vpiStreamCreate()

VPIStatus vpiStreamCreate	(	VPIDeviceType	devType,
		VPIStream *	stream
	)

#include <vpi/Stream.h>

Create a stream instance.

Parameters

devType	[in] device type
stream	[out] pointer to stream handle

Returns

an error code on failure else VPI_SUCCESS

vpiStreamDestroy()

void vpiStreamDestroy

(

VPIStream

stream

)

#include <vpi/Stream.h>

Destroy a stream instance and deallocate all HW resources.

Parameters

stream

[in] stream handle

vpiStreamGetDeviceType()

VPIStatus vpiStreamGetDeviceType	(	VPIStream	stream,
		VPIDeviceType *	deviceType
	)

#include <vpi/Stream.h>

Returns the backend type associated with the given stream.

Parameters

stream	[in] stream handle
deviceType	[out] pointer to where the device type will be stored. It can't be NULL.

Returns

an error code on failure else VPI_SUCCESS

vpiStreamGetThreadHandle()

VPIStatus vpiStreamGetThreadHandle	(	VPIStream	stream,
		VPINativeThreadHandle *	handle
	)

#include <vpi/Stream.h>

Returns OS-specific handle of the background stream processing thread.

Parameters

stream	[in] stream handle
handle	[out] stream processing thread handle

Returns

an error code on failure else VPI_SUCCESS

vpiStreamSync()

VPIStatus vpiStreamSync

(

VPIStream

stream

)

#include <vpi/Stream.h>

Blocks the calling thread until all submitted commands in this stream queue are done (queue is empty).

This function call is equivalent to atomically calling vpiEventRecord followed by vpiStreamWaitFor.

Parameters

stream

[in] stream handle

Returns

an error code on failure else VPI_SUCCESS

vpiStreamWaitFor()

VPIStatus vpiStreamWaitFor	(	VPIStream	stream,
		VPIEvent	event
	)

#include <vpi/Stream.h>

Pushes a command that blocks the processing of all future commands by the backend until the event is completed.

Parameters

stream	[in] Stream handle
event	[in] event handle

Returns

an error code on failure else VPI_SUCCESS

vpiStreamWrapCuda()

VPIStatus vpiStreamWrapCuda	(	CUstream	cudaStream,
		VPIStream *	stream
	)

#include <vpi/Stream.h>

Wraps an existing CUDA stream in a VPI stream for CUDA backend.

This allows interleaving user-defined and vpi processing in the same CUDA stream.

Parameters

cudaStream	[in] The CUDA stream handle to be wrapped
stream	[out] Pointer that will receive the newly created VPIStream.