sources/libs/gst-nvdstensor/nvdstensor.h File Reference

Detailed Description

NvDsTensor Interface

This file specifies the NvDsTensor management API.

The NvDsTensor API provides methods to allocate / deallocate, map / unmap and copy batched buffers.

Definition in file sources/libs/gst-nvdstensor/nvdstensor.h.

Go to the source code of this file.

Data Structures
struct	NvDsTensorCreateParams
	Holds parameters required to allocate an NvDsTensor. More...
struct	NvDsTensor
	Holds information about a single buffer in a batch. More...
struct	NvDsTensorArray
	Holds information about batched buffers. More...

NvDsTensor types and functions.
This section describes types and functions of NvDsTensor application programming interface.
#define	STRUCTURE_PADDING   4
	Defines the default padding length for reserved fields of structures. More...
enum	NvDsTensorMemMapFlags {   NVDSTENSOR_MAP_READ,   NVDSTENSOR_MAP_WRITE,   NVDSTENSOR_MAP_READ_WRITE,   NVDSTENSOR_MAP_READ,   NVDSTENSOR_MAP_WRITE,   NVDSTENSOR_MAP_READ_WRITE }
	Specifies mapping types for NvDsTensor. More...
enum	NvDsTensorFormat {   NVDSTENSOR_NCHW,   NVDSTENSOR_NCHW }
enum	NvDsTensorMemType {   NVDSTENSOR_MEM_DEFAULT,   NVDSTENSOR_MEM_CUDA_PINNED,   NVDSTENSOR_MEM_CUDA_DEVICE,   NVDSTENSOR_MEM_CUDA_UNIFIED,   NVDSTENSOR_MEM_SYSTEM,   NVDSTENSOR_MEM_DEFAULT,   NVDSTENSOR_MEM_CUDA_PINNED,   NVDSTENSOR_MEM_CUDA_DEVICE,   NVDSTENSOR_MEM_CUDA_UNIFIED,   NVDSTENSOR_MEM_SYSTEM }
	Specifies memory types for NvDsTensor. More...
enum	NvDsTensorPrecision {   NVDSTENSOR_FP32,   NVDSTENSOR_FP16,   NVDSTENSOR_INT32,   NVDSTENSOR_INT16,   NVDSTENSOR_INT8,   NVDSTENSOR_FP32,   NVDSTENSOR_FP16,   NVDSTENSOR_INT32,   NVDSTENSOR_INT16,   NVDSTENSOR_INT8 }
typedef struct NvDsTensorCreateParams	NvDsTensorCreateParams
	Holds parameters required to allocate an NvDsTensor. More...
typedef struct NvDsTensor	NvDsTensor
	Holds information about a single buffer in a batch. More...
typedef struct NvDsTensorArray	NvDsTensorArray
	Holds information about batched buffers. More...
int	NvDsTensorCreate (NvDsTensorArray **tensor, uint32_t arraySize, NvDsTensorCreateParams *params)
	Allocates a batch of tensor. More...
int	NvDsTensorDestroy (NvDsTensorArray *tensor)
	Frees batched buffers previously allocated by NvDsTensorCreate(). More...
int	NvDsTensorMap (NvDsTensorArray *tensors, int index, NvDsTensorMemMapFlags type)
	Maps hardware batched buffers to the HOST or CPU address space. More...
int	NvDsTensorUnMap (NvDsTensorArray *tensors, int index)
	Unmaps previously mapped buffer(s). More...

Macro Definition Documentation

STRUCTURE_PADDING

#define STRUCTURE_PADDING   4

Defines the default padding length for reserved fields of structures.

Definition at line 47 of file sources/libs/gst-nvdstensor/nvdstensor.h.

Typedef Documentation

NvDsTensor

typedef struct NvDsTensor NvDsTensor

Holds information about a single buffer in a batch.

NvDsTensorArray

typedef struct NvDsTensorArray NvDsTensorArray

Holds information about batched buffers.

NvDsTensorCreateParams

typedef struct NvDsTensorCreateParams NvDsTensorCreateParams

Holds parameters required to allocate an NvDsTensor.

Enumeration Type Documentation

NvDsTensorFormat

enum NvDsTensorFormat

Enumerator
NVDSTENSOR_NCHW
NVDSTENSOR_NCHW

Definition at line 59 of file sources/libs/gst-nvdstensor/nvdstensor.h.

NvDsTensorMemMapFlags

enum NvDsTensorMemMapFlags

Specifies mapping types for NvDsTensor.

Enumerator
NVDSTENSOR_MAP_READ	Specifies NvDsTensor mapping type "read.".
NVDSTENSOR_MAP_WRITE	Specifies NvDsTensor mapping type "write.".
NVDSTENSOR_MAP_READ_WRITE	Specifies NvDsTensor mapping type "read/write.".
NVDSTENSOR_MAP_READ	Specifies NvDsTensor mapping type "read.".
NVDSTENSOR_MAP_WRITE	Specifies NvDsTensor mapping type "write.".
NVDSTENSOR_MAP_READ_WRITE	Specifies NvDsTensor mapping type "read/write.".

Definition at line 52 of file sources/libs/gst-nvdstensor/nvdstensor.h.

NvDsTensorMemType

enum NvDsTensorMemType

Specifies memory types for NvDsTensor.

Enumerator
NVDSTENSOR_MEM_DEFAULT
NVDSTENSOR_MEM_CUDA_PINNED	Specifies CUDA Host memory type.
NVDSTENSOR_MEM_CUDA_DEVICE	Specifies CUDA Device memory type.
NVDSTENSOR_MEM_CUDA_UNIFIED	Specifies CUDA Unified memory type.
NVDSTENSOR_MEM_SYSTEM	Specifies memory allocated by malloc().
NVDSTENSOR_MEM_DEFAULT
NVDSTENSOR_MEM_CUDA_PINNED	Specifies CUDA Host memory type.
NVDSTENSOR_MEM_CUDA_DEVICE	Specifies CUDA Device memory type.
NVDSTENSOR_MEM_CUDA_UNIFIED	Specifies CUDA Unified memory type.
NVDSTENSOR_MEM_SYSTEM	Specifies memory allocated by malloc().

Definition at line 67 of file sources/libs/gst-nvdstensor/nvdstensor.h.

NvDsTensorPrecision

enum NvDsTensorPrecision

Enumerator
NVDSTENSOR_FP32
NVDSTENSOR_FP16
NVDSTENSOR_INT32
NVDSTENSOR_INT16
NVDSTENSOR_INT8
NVDSTENSOR_FP32
NVDSTENSOR_FP16
NVDSTENSOR_INT32
NVDSTENSOR_INT16
NVDSTENSOR_INT8

Definition at line 80 of file sources/libs/gst-nvdstensor/nvdstensor.h.

Function Documentation

NvDsTensorCreate()

int NvDsTensorCreate	(	NvDsTensorArray **	tensor,
		uint32_t	arraySize,
		NvDsTensorCreateParams *	params
	)

Allocates a batch of tensor.

Allocates memory for batchSize tensors and returns a pointer to an allocated NvDsTensorArray. The params structure must have the allocation parameters of a single buffer. If params.size is set, a buffer of that size is allocated, and all other parameters (width, height, tensor format, etc.) are ignored.

Call NvDsTensorDestroy() to free resources allocated by this function.

Parameters

[out]	tensor	array An indirect pointer to the allocated batched buffers.
[in]	arraySize	Batch size of buffers.
[in]	params	A pointer to an NvDsTensorCreateParams structure.

Returns

0 if successful, or -1 otherwise.

NvDsTensorDestroy()

int NvDsTensorDestroy

(

NvDsTensorArray *

tensor

)

Frees batched buffers previously allocated by NvDsTensorCreate().

Parameters

[in]

tensors

A pointer to an NvDsTensorArray to be freed.

Returns

0 if successful, or -1 otherwise.

NvDsTensorMap()

int NvDsTensorMap	(	NvDsTensorArray *	tensors,
		int	index,
		NvDsTensorMemMapFlags	type
	)

Maps hardware batched buffers to the HOST or CPU address space.

This function fills an array of pointers at tensors->tensorArray->mappedAddr tensors is a pointer to an NvDsTensorArray. tensorArray is a pointer to an NvDsTensor. mappedAddr holds pointer to buffer.

Use NvDsTensorUnMap() to unmap buffer(s) and release any resource.

Parameters

[in,out]	tensors	A pointer to an NvDsTensorArray structure. The function stores pointers to the buffers in a descendant of this structure; see the notes above.
[in]	index	Index of a buffer in the batch. -1 refers to all buffers in the batch.
[in]	type	A flag for mapping type.

Returns

0 if successful, or -1 otherwise.

NvDsTensorUnMap()

int NvDsTensorUnMap	(	NvDsTensorArray *	tensors,
		int	index
	)

Unmaps previously mapped buffer(s).

Parameters

[in]	tensors	A pointer to an NvDsTensorArray structure.
[in]	index	Index of a buffer in the batch. -1 indicates all buffers in the batch.

Returns

0 if successful, or -1 otherwise.