Open3D (C++ API)  0.18.0+3975044
Public Member Functions | Protected Member Functions | Static Protected Member Functions | Protected Attributes
open3d::core::Indexer Class Reference

#include <Indexer.h>

Public Member Functions

 Indexer ()
 
 Indexer (const Indexer &)=default
 
Indexeroperator= (const Indexer &)=default
 
 Indexer (const std::vector< Tensor > &input_tensors, const Tensor &output_tensor, DtypePolicy dtype_policy=DtypePolicy::ALL_SAME, const SizeVector &reduction_dims={})
 
 Indexer (const std::vector< Tensor > &input_tensors, const std::vector< Tensor > &output_tensors, DtypePolicy dtype_policy=DtypePolicy::ALL_SAME, const SizeVector &reduction_dims={})
 
bool CanUse32BitIndexing () const
 Returns true iff the maximum_offsets in bytes are smaller than 2^31 - 1. More...
 
IndexerIterator SplitTo32BitIndexing () const
 
std::unique_ptr< IndexerSplitLargestDim ()
 
Indexer GetPerOutputIndexer (int64_t output_idx) const
 
bool ShouldAccumulate () const
 
bool IsFinalOutput () const
 
void ShrinkDim (int64_t dim, int64_t start, int64_t size)
 
int64_t NumReductionDims () const
 Returns the number of reduction dimensions. More...
 
int64_t NumDims () const
 Returns number of dimensions of the Indexer. More...
 
const int64_t * GetPrimaryShape () const
 
int64_t * GetPrimaryShape ()
 
const int64_t * GetPrimaryStrides () const
 
int64_t NumWorkloads () const
 
int64_t NumOutputElements () const
 Returns the number of output elements. More...
 
int64_t NumInputs () const
 Number of input Tensors. More...
 
int64_t NumOutputs () const
 Number of output Tensors. More...
 
TensorRefGetInput (int64_t i)
 Returns input TensorRef. More...
 
const TensorRefGetInput (int64_t i) const
 
TensorRefGetOutput (int64_t i)
 Returns output TensorRef. More...
 
const TensorRefGetOutput (int64_t i) const
 
TensorRefGetOutput ()
 
const TensorRefGetOutput () const
 
bool IsReductionDim (int64_t dim) const
 Returns true if the dim -th dimension is reduced. More...
 
OPEN3D_HOST_DEVICE char * GetInputPtr (int64_t input_idx, int64_t workload_idx) const
 
template<typename T >
OPEN3D_HOST_DEVICE T * GetInputPtr (int64_t input_idx, int64_t workload_idx) const
 
OPEN3D_HOST_DEVICE char * GetOutputPtr (int64_t workload_idx) const
 
template<typename T >
OPEN3D_HOST_DEVICE T * GetOutputPtr (int64_t workload_idx) const
 
OPEN3D_HOST_DEVICE char * GetOutputPtr (int64_t output_idx, int64_t workload_idx) const
 
template<typename T >
OPEN3D_HOST_DEVICE T * GetOutputPtr (int64_t output_idx, int64_t workload_idx) const
 

Protected Member Functions

void CoalesceDimensions ()
 
void ReorderDimensions (const SizeVector &reduction_dims)
 
void UpdatePrimaryStrides ()
 Update primary_strides_ based on primary_shape_. More...
 
void UpdateContiguousFlags ()
 Update input_contiguous_ and output_contiguous_. More...
 
OPEN3D_HOST_DEVICE char * GetWorkloadDataPtr (const TensorRef &tr, bool tr_contiguous, int64_t workload_idx) const
 
template<typename T >
OPEN3D_HOST_DEVICE T * GetWorkloadDataPtr (const TensorRef &tr, bool tr_contiguous, int64_t workload_idx) const
 

Static Protected Member Functions

static void BroadcastRestride (TensorRef &src, int64_t dst_ndims, const int64_t *dst_shape)
 
static void ReductionRestride (TensorRef &dst, int64_t src_ndims, const int64_t *src_shape, const SizeVector &reduction_dims)
 

Protected Attributes

int64_t num_inputs_ = 0
 Number of input and output Tensors. More...
 
int64_t num_outputs_ = 0
 
TensorRef inputs_ [MAX_INPUTS]
 Array of input TensorRefs. More...
 
TensorRef outputs_ [MAX_OUTPUTS]
 Array of output TensorRefs. More...
 
bool inputs_contiguous_ [MAX_INPUTS]
 Array of contiguous flags for all input TensorRefs. More...
 
bool outputs_contiguous_ [MAX_OUTPUTS]
 Array of contiguous flags for all output TensorRefs. More...
 
int64_t primary_shape_ [MAX_DIMS]
 
int64_t primary_strides_ [MAX_DIMS]
 
int64_t ndims_ = 0
 Indexer's global number of dimensions. More...
 
bool final_output_ = true
 
bool accumulate_ = false
 

Detailed Description

Indexing engine for elementwise ops with broadcasting support.

Fancy indexing is supported by restriding input tensor and treating the operation as elementwise op.

After constructing Indexer on the host, the indexing methods can be used from both host and device.

Constructor & Destructor Documentation

◆ Indexer() [1/4]

open3d::core::Indexer::Indexer ( )
inline

◆ Indexer() [2/4]

open3d::core::Indexer::Indexer ( const Indexer )
default

◆ Indexer() [3/4]

open3d::core::Indexer::Indexer ( const std::vector< Tensor > &  input_tensors,
const Tensor output_tensor,
DtypePolicy  dtype_policy = DtypePolicy::ALL_SAME,
const SizeVector reduction_dims = {} 
)

Only single output is supported for simplicity. To extend this function to support multiple outputs, one may check for shape compatibility of all outputs.

◆ Indexer() [4/4]

open3d::core::Indexer::Indexer ( const std::vector< Tensor > &  input_tensors,
const std::vector< Tensor > &  output_tensors,
DtypePolicy  dtype_policy = DtypePolicy::ALL_SAME,
const SizeVector reduction_dims = {} 
)

Member Function Documentation

◆ BroadcastRestride()

void open3d::core::Indexer::BroadcastRestride ( TensorRef src,
int64_t  dst_ndims,
const int64_t *  dst_shape 
)
staticprotected

Broadcast src to dst by setting shape 1 to omitted dimensions and setting stride 0 to brocasted dimensions.

Note that other approaches may also work. E.g. one could set src's shape to exactly the same as dst's shape. In general, if a dimension is of size 1, the stride have no effect in computing offsets; or likewise if a dimension has stride 0, the shape have no effect in computing offsets.

[Before] Omitted | Broadcast | | No broadcast | | | V V V src.shape_: [ 2, 1, 1, 3] src.strides_: [ 3, 3, 3, 1] dst.shape_: [ 2, 2, 2, 1, 3] dst.strides_: [12, 6, 3, 3, 1]

[After] src.shape_: [ 1, 2, 1, 1, 3] src.strides_: [ 0, 3, 0, 3, 1]

Parameters
srcThe source TensorRef to be broadcasted.
dst_ndimsNumber of dimensions to be broadcasted to.
dst_shapeShape to be broadcasted to.

◆ CanUse32BitIndexing()

bool open3d::core::Indexer::CanUse32BitIndexing ( ) const

Returns true iff the maximum_offsets in bytes are smaller than 2^31 - 1.

◆ CoalesceDimensions()

void open3d::core::Indexer::CoalesceDimensions ( )
protected

Merge adjacent dimensions if either dim is 1 or if: shape[n] * stride[n] == shape[n + 1]

◆ GetInput() [1/2]

TensorRef& open3d::core::Indexer::GetInput ( int64_t  i)
inline

Returns input TensorRef.

◆ GetInput() [2/2]

const TensorRef& open3d::core::Indexer::GetInput ( int64_t  i) const
inline

◆ GetInputPtr() [1/2]

OPEN3D_HOST_DEVICE char* open3d::core::Indexer::GetInputPtr ( int64_t  input_idx,
int64_t  workload_idx 
) const
inline

Get input Tensor data pointer based on workload_idx.

Parameters
input_idxInput tensor index.
workload_idxThe index of the compute workload, similar to thread_id, if a thread only processes one workload.

◆ GetInputPtr() [2/2]

template<typename T >
OPEN3D_HOST_DEVICE T* open3d::core::Indexer::GetInputPtr ( int64_t  input_idx,
int64_t  workload_idx 
) const
inline

Get input Tensor data pointer based on workload_idx.

Parameters
input_idxInput tensor index.
workload_idxThe index of the compute workload, similar to thread_id, if a thread only processes one workload.

Note: Assumes that sizeof(T) matches the input's dtype size, but does not check this constraint for performance reasons.

◆ GetOutput() [1/4]

TensorRef& open3d::core::Indexer::GetOutput ( )
inline

Returns output TensorRef. Only works if there's only one output. Equivalent to GetOutput(0).

◆ GetOutput() [2/4]

const TensorRef& open3d::core::Indexer::GetOutput ( ) const
inline

◆ GetOutput() [3/4]

TensorRef& open3d::core::Indexer::GetOutput ( int64_t  i)
inline

Returns output TensorRef.

◆ GetOutput() [4/4]

const TensorRef& open3d::core::Indexer::GetOutput ( int64_t  i) const
inline

◆ GetOutputPtr() [1/4]

OPEN3D_HOST_DEVICE char* open3d::core::Indexer::GetOutputPtr ( int64_t  output_idx,
int64_t  workload_idx 
) const
inline

Get output Tensor data pointer based on workload_idx.

Parameters
output_idxOutput tensor index.
workload_idxThe index of the compute workload, similar to thread_id, if a thread only processes one workload.

◆ GetOutputPtr() [2/4]

template<typename T >
OPEN3D_HOST_DEVICE T* open3d::core::Indexer::GetOutputPtr ( int64_t  output_idx,
int64_t  workload_idx 
) const
inline

Get output Tensor data pointer based on workload_idx.

Parameters
output_idxOutput tensor index.
workload_idxThe index of the compute workload, similar to thread_id, if a thread only processes one workload.

◆ GetOutputPtr() [3/4]

OPEN3D_HOST_DEVICE char* open3d::core::Indexer::GetOutputPtr ( int64_t  workload_idx) const
inline

Get output Tensor data pointer based on workload_idx.

Parameters
workload_idxThe index of the compute workload, similar to thread_id, if a thread only processes one workload.

◆ GetOutputPtr() [4/4]

template<typename T >
OPEN3D_HOST_DEVICE T* open3d::core::Indexer::GetOutputPtr ( int64_t  workload_idx) const
inline

Get output Tensor data pointer based on workload_idx.

Parameters
workload_idxThe index of the compute workload, similar to thread_id, if a thread only processes one workload.

Note: Assumes that sizeof(T) matches the output's dtype size, but does not check this constraint for performance reasons.

◆ GetPerOutputIndexer()

Indexer open3d::core::Indexer::GetPerOutputIndexer ( int64_t  output_idx) const

Get a sub-indexer that loops through all inputs corresponding to a single output.

◆ GetPrimaryShape() [1/2]

int64_t* open3d::core::Indexer::GetPrimaryShape ( )
inline

◆ GetPrimaryShape() [2/2]

const int64_t* open3d::core::Indexer::GetPrimaryShape ( ) const
inline

Returns Indexer's primary shape, one can iterate the Indexer with this shape.

◆ GetPrimaryStrides()

const int64_t* open3d::core::Indexer::GetPrimaryStrides ( ) const
inline

Returns Indexer's primary strides, one can iterate the Indexer with this strides. It is always set to be the default strides from primary_shape_.

◆ GetWorkloadDataPtr() [1/2]

OPEN3D_HOST_DEVICE char* open3d::core::Indexer::GetWorkloadDataPtr ( const TensorRef tr,
bool  tr_contiguous,
int64_t  workload_idx 
) const
inlineprotected

Get data pointer from a TensorRef with workload_idx. Note: can be optimized by computing all input ptrs and output ptr together.

◆ GetWorkloadDataPtr() [2/2]

template<typename T >
OPEN3D_HOST_DEVICE T* open3d::core::Indexer::GetWorkloadDataPtr ( const TensorRef tr,
bool  tr_contiguous,
int64_t  workload_idx 
) const
inlineprotected

Get data pointer from a TensorRef with workload_idx. Note: can be optimized by computing all input ptrs and output ptr together.

Note: Assumes that sizeof(T) matches the data's dtype size, but does not check this constraint for performance reasons.

◆ IsFinalOutput()

bool open3d::core::Indexer::IsFinalOutput ( ) const
inline

◆ IsReductionDim()

bool open3d::core::Indexer::IsReductionDim ( int64_t  dim) const
inline

Returns true if the dim -th dimension is reduced.

◆ NumDims()

int64_t open3d::core::Indexer::NumDims ( ) const
inline

Returns number of dimensions of the Indexer.

◆ NumInputs()

int64_t open3d::core::Indexer::NumInputs ( ) const
inline

Number of input Tensors.

◆ NumOutputElements()

int64_t open3d::core::Indexer::NumOutputElements ( ) const

Returns the number of output elements.

◆ NumOutputs()

int64_t open3d::core::Indexer::NumOutputs ( ) const
inline

Number of output Tensors.

◆ NumReductionDims()

int64_t open3d::core::Indexer::NumReductionDims ( ) const

Returns the number of reduction dimensions.

◆ NumWorkloads()

int64_t open3d::core::Indexer::NumWorkloads ( ) const

Returns the total number of workloads (e.g. computations) needed for the op. The scheduler schedules these workloads to run on parallel threads.

For non-reduction ops, NumWorkloads() is the same as number of output elements (e.g. for broadcasting ops).

For reduction ops, NumWorkLoads() is the same as the number of input elements. Currently we don't allow mixing broadcasting and reduction in one op kernel.

◆ operator=()

Indexer& open3d::core::Indexer::operator= ( const Indexer )
default

◆ ReductionRestride()

void open3d::core::Indexer::ReductionRestride ( TensorRef dst,
int64_t  src_ndims,
const int64_t *  src_shape,
const SizeVector reduction_dims 
)
staticprotected

Symmetrical to BroadcastRestride. Set the reduced dimensions' stride to 0 at output. Currently only support the keepdim=true case.

◆ ReorderDimensions()

void open3d::core::Indexer::ReorderDimensions ( const SizeVector reduction_dims)
protected

◆ ShouldAccumulate()

bool open3d::core::Indexer::ShouldAccumulate ( ) const
inline

◆ ShrinkDim()

void open3d::core::Indexer::ShrinkDim ( int64_t  dim,
int64_t  start,
int64_t  size 
)

Shrink iteration to a specific range in a specific dimension.

Parameters
dimThe dimension to be shrunken to.
startStarting index (inclusive) for dimension dim. No dimension wrapping is available.
sizeThe size to iterate in dimension dim.

◆ SplitLargestDim()

std::unique_ptr< Indexer > open3d::core::Indexer::SplitLargestDim ( )

Split the indexer such that the largest-span-dimension is split into two halves. The returned new indexer iterates the first half while the current indexer iterates the second half.

◆ SplitTo32BitIndexing()

IndexerIterator open3d::core::Indexer::SplitTo32BitIndexing ( ) const

Returns an iterator of Indexers, each of which can be indexed in 32 bits.

◆ UpdateContiguousFlags()

void open3d::core::Indexer::UpdateContiguousFlags ( )
protected

Update input_contiguous_ and output_contiguous_.

◆ UpdatePrimaryStrides()

void open3d::core::Indexer::UpdatePrimaryStrides ( )
protected

Update primary_strides_ based on primary_shape_.

Field Documentation

◆ accumulate_

bool open3d::core::Indexer::accumulate_ = false
protected

If the kernel should accumulate into the output. Only relevant for CUDA reductions.

◆ final_output_

bool open3d::core::Indexer::final_output_ = true
protected

Whether this iterator produces the actual output, as opposed to something that will be accumulated further. Only relevant for CUDA reductions.

◆ inputs_

TensorRef open3d::core::Indexer::inputs_[MAX_INPUTS]
protected

Array of input TensorRefs.

◆ inputs_contiguous_

bool open3d::core::Indexer::inputs_contiguous_[MAX_INPUTS]
protected

Array of contiguous flags for all input TensorRefs.

◆ ndims_

int64_t open3d::core::Indexer::ndims_ = 0
protected

Indexer's global number of dimensions.

◆ num_inputs_

int64_t open3d::core::Indexer::num_inputs_ = 0
protected

Number of input and output Tensors.

◆ num_outputs_

int64_t open3d::core::Indexer::num_outputs_ = 0
protected

◆ outputs_

TensorRef open3d::core::Indexer::outputs_[MAX_OUTPUTS]
protected

Array of output TensorRefs.

◆ outputs_contiguous_

bool open3d::core::Indexer::outputs_contiguous_[MAX_OUTPUTS]
protected

Array of contiguous flags for all output TensorRefs.

◆ primary_shape_

int64_t open3d::core::Indexer::primary_shape_[MAX_DIMS]
protected

Indexer's global shape. The shape's number of elements is the same as GetNumWorkloads() for the Indexer.

  • For broadcasting, primary_shape_ is the same as the output shape.
  • For reduction, primary_shape_ is the same as the input shape.
  • Currently we don't allow broadcasting mixed with reduction. But if broadcasting mixed with reduction is allowed, primary_shape_ is a mix of input shape and output shape. First, fill in all omitted dimensions (in inputs for broadcasting) and reduction dimensions (as if keepdim=true always) with size 1. For each axis, the primary dimension is the non-1 dimension (if both are 1, then the primary dimension is 1 in that axis).

◆ primary_strides_

int64_t open3d::core::Indexer::primary_strides_[MAX_DIMS]
protected

The default strides for primary_shape_ for internal use only. Used to compute the actual strides and ultimately the index offsets.


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