cudf::roaring_bitmap Class Reference

A type-erased wrapper around cuco's roaring bitmap supporting both 32-bit and 64-bit keys. More...

#include <cudf/utilities/roaring_bitmap.hpp>

Public Member Functions
	roaring_bitmap (roaring_bitmap_type type, std::span< cuda::std::byte const > serialized_bitmap_data)
	Constructs a roaring_bitmap from serialized bitmap data (payload) More...
	~roaring_bitmap ()
	Destructor for the roaring bitmap class.
	roaring_bitmap (roaring_bitmap &&other) noexcept
	Move constructor for the roaring bitmap class. More...
roaring_bitmap &	operator= (roaring_bitmap &&other) noexcept
	Move assignment operator for the roaring bitmap class. More...
	roaring_bitmap (roaring_bitmap const &)=delete
roaring_bitmap &	operator= (roaring_bitmap const &)=delete
void	materialize (cuda::stream_ref stream=cudf::get_default_stream()) const
	Materialize the underlying cuco roaring bitmap. More...
roaring_bitmap_type	type () const
	Returns the roaring bitmap type. More...
bool	empty () const
	Checks whether the bitmap contains no keys. More...
cuda::std::size_t	size () const
	Returns the number of keys stored in the bitmap. More...
cuda::std::size_t	size_bytes () const
	Returns the size of the serialized bitmap storage in bytes. More...
std::unique_ptr< cudf::column >	contains_async (cudf::column_view const &keys, cuda::stream_ref stream=cudf::get_default_stream(), rmm::device_async_resource_ref mr=cudf::get_current_device_resource_ref()) const
	Asynchronously queries the bitmap for membership of each key in the input column. More...
void	contains_async (cudf::column_view const &keys, cudf::mutable_column_view const &output, cuda::stream_ref stream) const
	Asynchronously queries the bitmap for membership of each key in a column and stores the result in the output column. More...

Detailed Description

A type-erased wrapper around cuco's roaring bitmap supporting both 32-bit and 64-bit keys.

This class holds a span of serialized roaring bitmap data (specified 32-bit or 64-bit) on the host and lazily materializes the corresponding underlying cuco roaring bitmap when materialize() is called or implicitly when contains_async() is first called.

Example usage (64-bit version):

auto bitmap = cudf::roaring_bitmap(cudf::roaring_bitmap_type::BITS_64, serialized_bitmap_data);
bitmap.materialize(stream);
auto result = bitmap.contains_async(keys_column, stream, mr);

Definition at line 57 of file roaring_bitmap.hpp.

Constructor & Destructor Documentation

roaring_bitmap() [1/2]

cudf::roaring_bitmap::roaring_bitmap ( roaring_bitmap_type  type, std::span< cuda::std::byte const >  serialized_bitmap_data  )

explicit

Constructs a roaring_bitmap from serialized bitmap data (payload)

The serialized bitmap data must remain valid until the underlying cuco roaring bitmap is materialized via materialize().

Parameters

type	The bitmap key type (BITS_32 or BITS_64)
serialized_bitmap_data	Host span of bytes containing a roaring bitmap serialized in portable format

Exceptions

std::invalid_argument

if the serialized bitmap data is empty

roaring_bitmap() [2/2]

cudf::roaring_bitmap::roaring_bitmap ( roaring_bitmap &&  other )

noexcept

Move constructor for the roaring bitmap class.

Parameters

other

Roaring bitmap to move from

Member Function Documentation

contains_async() [1/2]

std::unique_ptr<cudf::column> cudf::roaring_bitmap::contains_async	(	cudf::column_view const &	keys,
		cuda::stream_ref	stream = cudf::get_default_stream(),
		rmm::device_async_resource_ref	mr = cudf::get_current_device_resource_ref()
	)		const

Asynchronously queries the bitmap for membership of each key in the input column.

The input column must have dtype UINT32 (for BITS_32) or UINT64 (for BITS_64).

Parameters

keys	Key column to query
stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource for the output column allocation

Returns

A BOOL8 column indicating positions of the present keys

Exceptions

std::invalid_argument

if the key column dtype is invalid

contains_async() [2/2]

void cudf::roaring_bitmap::contains_async	(	cudf::column_view const &	keys,
		cudf::mutable_column_view const &	output,
		cuda::stream_ref	stream
	)		const

Asynchronously queries the bitmap for membership of each key in a column and stores the result in the output column.

The input column must have dtype UINT32 (for BITS_32) or UINT64 (for BITS_64). The output column must have BOOL8 dtype.

Parameters

keys	Key column to query
output	Output column to store the result
stream	CUDA stream used for device memory operations and kernel launches

Exceptions

std::invalid_argument

if the key or output column dtypes are invalid

empty()

bool cudf::roaring_bitmap::empty

(

)

const

Checks whether the bitmap contains no keys.

Returns

Whether the roaring bitmap contains no keys

materialize()

void cudf::roaring_bitmap::materialize

(

cuda::stream_ref

stream = cudf::get_default_stream()

)

const

Materialize the underlying cuco roaring bitmap.

The serialized bitmap data span is cleared after this call.

Parameters

stream

CUDA stream used for device memory operations and kernel launches

operator=()

roaring_bitmap& cudf::roaring_bitmap::operator= ( roaring_bitmap &&  other )

noexcept

Move assignment operator for the roaring bitmap class.

Parameters

other

Roaring bitmap to move from

Returns

Reference to the moved-from roaring bitmap

size()

cuda::std::size_t cudf::roaring_bitmap::size

(

)

const

Returns the number of keys stored in the bitmap.

Returns

Number of keys stored in the bitmap

size_bytes()

cuda::std::size_t cudf::roaring_bitmap::size_bytes

(

)

const

Returns the size of the serialized bitmap storage in bytes.

Returns

Size of the serialized bitmap storage in bytes

type()

roaring_bitmap_type cudf::roaring_bitmap::type

(

)

const

Returns the roaring bitmap type.

Returns

Roaring bitmap type

---

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

• cudf/utilities/roaring_bitmap.hpp