CUDNN Frontend API  8.3.0
cudnn_frontend::Tensor_v8 Class Reference

#include <cudnn_frontend_Tensor.h>

Inheritance diagram for cudnn_frontend::Tensor_v8:
Inheritance graph
Collaboration diagram for cudnn_frontend::Tensor_v8:
Collaboration graph

Public Member Functions

std::string describe () const override
 Return a string describing the backend Descriptor. More...
 
int64_t getPackedElementCount () const
 
int64_t getDimensionCount () const
 
int64_t const * getDimArray () const
 
int64_t const * getStrideArray () const
 
int64_t getDataType () const
 
 Tensor_v8 (Tensor_v8 &&from)=default
 
Tensor_v8operator= (Tensor_v8 &&)=default
 
 ~Tensor_v8 ()=default
 
- Public Member Functions inherited from cudnn_frontend::BackendDescriptor
cudnnBackendDescriptor_t get_raw_desc () const
 
cudnnStatus_t get_status () const
 Current status of the descriptor. More...
 
void set_status (cudnnStatus_t const status_) const
 Set status of the descriptor. More...
 
void set_error (const char *message) const
 Set Diagonistic error message. More...
 
const char * get_error () const
 Diagonistic error message if any. More...
 
ManagedOpaqueDescriptor get_desc () const
 Returns a copy of underlying managed descriptor. More...
 
cudnnStatus_t initialize_managed_backend_pointer (cudnnBackendDescriptorType_t type)
 Initializes the underlying managed descriptor. More...
 

Private Member Functions

 Tensor_v8 ()=default
 
 Tensor_v8 (Tensor_v8 const &)=delete
 
Tensor_v8operator= (Tensor_v8 const &)=delete
 

Private Attributes

cudnnDataType_t data_type = CUDNN_DATA_FLOAT
 
int64_t btensor_dimA [CUDNN_DIM_MAX+1] = {-1}
 Datatype of the elements. More...
 
int64_t btensor_strA [CUDNN_DIM_MAX+1] = {-1}
 n, g, c, d, h, w More...
 
int64_t id = -1
 n, g, c, d, h, w More...
 
int64_t alignment = -1
 Unique id of the tensor. More...
 
int64_t nDims = -1
 Certain engine config expect minimum alignment of 16B. More...
 
int64_t vectorDimension = -1
 Number of Dimensions of the tensor. More...
 
int64_t vectorCount = 1
 Which dimension of the tensor is vectorized (Generally the c dim) More...
 
bool isVirtual = false
 What is the vectorization count (4 or 32) More...
 
bool isByValue = false
 Whether it is an intermediate tensor of an op graph. More...
 

Friends

class TensorBuilder_v8
 

Additional Inherited Members

- Protected Member Functions inherited from cudnn_frontend::BackendDescriptor
 BackendDescriptor (ManagedOpaqueDescriptor pointer_, cudnnStatus_t status_, std::string err_msg_)
 
 BackendDescriptor ()=default
 
- Protected Attributes inherited from cudnn_frontend::BackendDescriptor
ManagedOpaqueDescriptor pointer
 
cudnnStatus_t status = CUDNN_STATUS_SUCCESS
 Shared pointer of the OpaqueBackendPointer. More...
 
std::string err_msg
 Error message if any being set. More...
 

Detailed Description

Tensor_v8 Class This class tells the properties of the Tensor_v8 on which the operation will be performed Properties:

  • dataType
  • alignment
  • unique identifier
  • tensor dimensions
  • tensor strides
  • isVirtual
  • isByValue

Use TensorBuilder_v8 to build this class. Describe returns a string describing the tensor class

Definition at line 52 of file cudnn_frontend_Tensor.h.

Constructor & Destructor Documentation

◆ Tensor_v8() [1/3]

cudnn_frontend::Tensor_v8::Tensor_v8 ( Tensor_v8 &&  from)
default

◆ ~Tensor_v8()

cudnn_frontend::Tensor_v8::~Tensor_v8 ( )
default

Referenced by getDataType().

◆ Tensor_v8() [2/3]

cudnn_frontend::Tensor_v8::Tensor_v8 ( )
privatedefault

Referenced by getDataType().

◆ Tensor_v8() [3/3]

cudnn_frontend::Tensor_v8::Tensor_v8 ( Tensor_v8 const &  )
privatedelete

Member Function Documentation

◆ describe()

std::string cudnn_frontend::Tensor_v8::describe ( ) const
inlineoverridevirtual

Return a string describing the backend Descriptor.

Implements cudnn_frontend::BackendDescriptor.

Definition at line 56 of file cudnn_frontend_Tensor.h.

References alignment, btensor_dimA, btensor_strA, data_type, isByValue, isVirtual, nDims, cudnn_frontend::to_string(), nlohmann::to_string(), vectorCount, and vectorDimension.

◆ getDataType()

int64_t cudnn_frontend::Tensor_v8::getDataType ( ) const
inline

Definition at line 105 of file cudnn_frontend_Tensor.h.

References data_type, operator=(), Tensor_v8(), and ~Tensor_v8().

◆ getDimArray()

int64_t const* cudnn_frontend::Tensor_v8::getDimArray ( ) const
inline

Definition at line 95 of file cudnn_frontend_Tensor.h.

References btensor_dimA.

Referenced by cudnn_frontend::cudnnReorderFilterAndBiasInt8x32().

◆ getDimensionCount()

int64_t cudnn_frontend::Tensor_v8::getDimensionCount ( ) const
inline

Definition at line 90 of file cudnn_frontend_Tensor.h.

References nDims.

Referenced by cudnn_frontend::cudnnReorderFilterAndBiasInt8x32().

◆ getPackedElementCount()

int64_t cudnn_frontend::Tensor_v8::getPackedElementCount ( ) const
inline

Definition at line 81 of file cudnn_frontend_Tensor.h.

References btensor_dimA, nDims, and vectorCount.

◆ getStrideArray()

int64_t const* cudnn_frontend::Tensor_v8::getStrideArray ( ) const
inline

Definition at line 100 of file cudnn_frontend_Tensor.h.

References btensor_strA.

◆ operator=() [1/2]

Tensor_v8& cudnn_frontend::Tensor_v8::operator= ( Tensor_v8 &&  )
default

◆ operator=() [2/2]

Tensor_v8& cudnn_frontend::Tensor_v8::operator= ( Tensor_v8 const &  )
privatedelete

Friends And Related Function Documentation

◆ TensorBuilder_v8

friend class TensorBuilder_v8
friend

Definition at line 54 of file cudnn_frontend_Tensor.h.

Referenced by cudnn_frontend::TensorBuilder_v8::build().

Member Data Documentation

◆ alignment

int64_t cudnn_frontend::Tensor_v8::alignment = -1
private

Unique id of the tensor.

Definition at line 125 of file cudnn_frontend_Tensor.h.

Referenced by describe().

◆ btensor_dimA

int64_t cudnn_frontend::Tensor_v8::btensor_dimA[CUDNN_DIM_MAX+1] = {-1}
private

Datatype of the elements.

Definition at line 122 of file cudnn_frontend_Tensor.h.

Referenced by describe(), getDimArray(), and getPackedElementCount().

◆ btensor_strA

int64_t cudnn_frontend::Tensor_v8::btensor_strA[CUDNN_DIM_MAX+1] = {-1}
private

n, g, c, d, h, w

Definition at line 123 of file cudnn_frontend_Tensor.h.

Referenced by describe(), and getStrideArray().

◆ data_type

cudnnDataType_t cudnn_frontend::Tensor_v8::data_type = CUDNN_DATA_FLOAT
private

Definition at line 121 of file cudnn_frontend_Tensor.h.

Referenced by describe(), and getDataType().

◆ id

int64_t cudnn_frontend::Tensor_v8::id = -1
private

n, g, c, d, h, w

Definition at line 124 of file cudnn_frontend_Tensor.h.

◆ isByValue

bool cudnn_frontend::Tensor_v8::isByValue = false
private

Whether it is an intermediate tensor of an op graph.

Definition at line 131 of file cudnn_frontend_Tensor.h.

Referenced by describe().

◆ isVirtual

bool cudnn_frontend::Tensor_v8::isVirtual = false
private

What is the vectorization count (4 or 32)

Definition at line 130 of file cudnn_frontend_Tensor.h.

Referenced by describe().

◆ nDims

int64_t cudnn_frontend::Tensor_v8::nDims = -1
private

Certain engine config expect minimum alignment of 16B.

Alignment of the tensor.

Definition at line 127 of file cudnn_frontend_Tensor.h.

Referenced by describe(), getDimensionCount(), and getPackedElementCount().

◆ vectorCount

int64_t cudnn_frontend::Tensor_v8::vectorCount = 1
private

Which dimension of the tensor is vectorized (Generally the c dim)

Definition at line 129 of file cudnn_frontend_Tensor.h.

Referenced by describe(), and getPackedElementCount().

◆ vectorDimension

int64_t cudnn_frontend::Tensor_v8::vectorDimension = -1
private

Number of Dimensions of the tensor.

Definition at line 128 of file cudnn_frontend_Tensor.h.

Referenced by describe().


The documentation for this class was generated from the following file: