cudf::interop::arrow_table Class Reference

A standard interchange medium for ArrowDeviceArray data in cudf. More...

#include <interop.hpp>

Public Member Functions
	arrow_table (cudf::table &&input, cudf::host_span< column_metadata const > metadata, rmm::cuda_stream_view stream=cudf::get_default_stream(), rmm::device_async_resource_ref mr=cudf::get_current_device_resource_ref())
	Construct a new arrow table object. More...
	arrow_table (ArrowSchema &&schema, ArrowDeviceArray &&input, rmm::cuda_stream_view stream=cudf::get_default_stream(), rmm::device_async_resource_ref mr=cudf::get_current_device_resource_ref())
	Construct a new arrow table object. More...
	arrow_table (ArrowSchema &&schema, ArrowArray &&input, rmm::cuda_stream_view stream=cudf::get_default_stream(), rmm::device_async_resource_ref mr=cudf::get_current_device_resource_ref())
	Construct a new arrow table object. More...
	arrow_table (ArrowArrayStream &&input, rmm::cuda_stream_view stream=cudf::get_default_stream(), rmm::device_async_resource_ref mr=cudf::get_current_device_resource_ref())
	Construct a new arrow table object. More...
void	to_arrow_schema (ArrowSchema *output, rmm::cuda_stream_view stream=cudf::get_default_stream(), rmm::device_async_resource_ref mr=cudf::get_current_device_resource_ref()) const
	Convert the table to an ArrowSchema. More...
void	to_arrow (ArrowDeviceArray *output, ArrowDeviceType device_type=ARROW_DEVICE_CUDA, rmm::cuda_stream_view stream=cudf::get_default_stream(), rmm::device_async_resource_ref mr=cudf::get_current_device_resource_ref()) const
	Convert the table to an ArrowDeviceArray. More...
table_view	view () const
	Get a view of the table data. More...

Detailed Description

A standard interchange medium for ArrowDeviceArray data in cudf.

This class provides a way to work with ArrowDeviceArray data in cudf without sacrificing the APIs expected of a cudf table. On the other end, it provides the shared lifetime management expected by arrow consumers rather than the single-owner mechanism of cudf::table.

Definition at line 337 of file interop.hpp.

Constructor & Destructor Documentation

arrow_table() [1/4]

cudf::interop::arrow_table::arrow_table	(	cudf::table &&	input,
		cudf::host_span< column_metadata const >	metadata,
		rmm::cuda_stream_view	stream = cudf::get_default_stream(),
		rmm::device_async_resource_ref	mr = cudf::get_current_device_resource_ref()
	)

Construct a new arrow table object.

The input table will be moved into the arrow_table, so it is no longer suitable for use afterwards.

Parameters

input	cudf table to convert to arrow
metadata	The hierarchy of names of columns and children
stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource used for any allocations during conversion

arrow_table() [2/4]

cudf::interop::arrow_table::arrow_table	(	ArrowSchema &&	schema,
		ArrowDeviceArray &&	input,
		rmm::cuda_stream_view	stream = cudf::get_default_stream(),
		rmm::device_async_resource_ref	mr = cudf::get_current_device_resource_ref()
	)

Construct a new arrow table object.

The input array will be moved into the arrow_table, so it is no longer suitable for use afterwards. For consistency, this is done even if the source array points to host data.

Parameters

schema	Arrow schema for the table
input	ArrowDeviceArray data for the table
stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource used for any allocations during conversion

arrow_table() [3/4]

cudf::interop::arrow_table::arrow_table	(	ArrowSchema &&	schema,
		ArrowArray &&	input,
		rmm::cuda_stream_view	stream = cudf::get_default_stream(),
		rmm::device_async_resource_ref	mr = cudf::get_current_device_resource_ref()
	)

Construct a new arrow table object.

The input array will be released, so it is no longer suitable for use afterwards. This is done for consistency with other constructors of arrow_table even though the source data is always host data.

Parameters

schema	Arrow schema for the table
input	ArrowArray data for the table
stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource used for any allocations during conversion

arrow_table() [4/4]

cudf::interop::arrow_table::arrow_table	(	ArrowArrayStream &&	input,
		rmm::cuda_stream_view	stream = cudf::get_default_stream(),
		rmm::device_async_resource_ref	mr = cudf::get_current_device_resource_ref()
	)

Construct a new arrow table object.

The stream will be released after the table is created, so it is no longer suitable for use afterwards. This is done for consistency with other constructors of arrow_table even though the source data is always host data.

Parameters

input	ArrowArrayStream data for the table
stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource used for any allocations during conversion

Member Function Documentation

to_arrow()

void cudf::interop::arrow_table::to_arrow	(	ArrowDeviceArray *	output,
		ArrowDeviceType	device_type = ARROW_DEVICE_CUDA,
		rmm::cuda_stream_view	stream = cudf::get_default_stream(),
		rmm::device_async_resource_ref	mr = cudf::get_current_device_resource_ref()
	)		const

Convert the table to an ArrowDeviceArray.

Parameters

output	ArrowDeviceArray to populate with the table's data
device_type	ArrowDeviceType to set on the output
stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource used for any allocations during conversion

to_arrow_schema()

void cudf::interop::arrow_table::to_arrow_schema	(	ArrowSchema *	output,
		rmm::cuda_stream_view	stream = cudf::get_default_stream(),
		rmm::device_async_resource_ref	mr = cudf::get_current_device_resource_ref()
	)		const

Convert the table to an ArrowSchema.

The resulting schema is a deep copy of the arrow_column's schema and is not tied to its lifetime.

Parameters

output	ArrowSchema to populate with the table's schema
stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource used for any allocations during conversion

view()

table_view cudf::interop::arrow_table::view

(

)

const

Get a view of the table data.

Parameters

stream	CUDA stream used for device memory operations and kernel launches
mr	Device memory resource used for any allocations during conversion

Returns

unique_table_view_t containing a view of the table data

---

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

• interop.hpp