interop.hpp

查看此文件的文档。

/*
 * 版权所有 (c) 2020-2025, NVIDIA CORPORATION。
 *
 * 根据 Apache 许可，版本 2.0（“许可”）获得许可；
 * 除非遵守许可，否则您不得使用此文件。
 * 您可以在以下位置获取许可副本
 *
 * https://apache.ac.cn/licenses/LICENSE-2.0
 *
 * 除非适用法律要求或书面同意，否则软件
 * 根据许可分发，基础是“按原样”分发，
 * 不附带任何明示或暗示的保证或条件。
 * 请参阅许可，了解特定语言的权限和
 * 限制。
 */

#pragma once

#include <cudf/column/column.hpp>
#include <cudf/detail/transform.hpp>
#include <cudf/table/table.hpp>
#include <cudf/table/table_view.hpp>
#include <cudf/types.hpp>
#include <cudf/utilities/export.hpp>
#include <cudf/utilities/memory_resource.hpp>
#include <cudf/utilities/span.hpp>

#include <rmm/resource_ref.hpp>

#include <utility>

struct DLManagedTensor;

struct ArrowDeviceArray;

struct ArrowSchema;

struct ArrowArray;

struct ArrowArrayStream;

// 这些是来自 Arrow 的类型，我们在此前置声明是为了避免 API 需要包含 nanoarrow 头文件。
// avoid needing to include nanoarrow headers.
typedef int32_t ArrowDeviceType; // NOLINT

#define ARROW_DEVICE_CUDA 2 // NOLINT

namespace CUDF_EXPORT cudf {
std::unique_ptr<table> from_dlpack(
 DLManagedTensor const* managed_tensor,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

DLManagedTensor* to_dlpack(
 table_view const& input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 // end of group

struct column_metadata {
 std::string name;
 std::vector<column_metadata> children_meta;

 column_metadata(std::string _name) : name(std::move(_name)) {}
 column_metadata() = default;
};

using unique_schema_t = std::unique_ptr<ArrowSchema, void (*)(ArrowSchema*)>;

using unique_device_array_t = std::unique_ptr<ArrowDeviceArray, void (*)(ArrowDeviceArray*)>;

using owned_columns_t = std::vector<std::unique_ptr<cudf::column>>;

template <typename ViewType>
struct custom_view_deleter {
 explicit custom_view_deleter(owned_columns_t&& owned) : owned_mem_{std::move(owned)} {}

 void operator()(ViewType* ptr) const { delete ptr; }

 owned_columns_t owned_mem_;
};

using unique_table_view_t =
 std::unique_ptr<cudf::table_view, custom_view_deleter<cudf::table_view>>;

using unique_column_view_t =
 std::unique_ptr<cudf::column_view, custom_view_deleter<cudf::column_view>>;

namespace interop {

struct arrow_array_container;

cudf::column_metadata get_column_metadata(cudf::column_view const& input);

std::vector<cudf::column_metadata> get_table_metadata(cudf::table_view const& input);

class arrow_column {
 public
 arrow_column(cudf::column&& input,
 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());

 arrow_column(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());

 arrow_column(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());

 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;

 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;

 [[nodiscard]] column_view view() const;

 private
 std::shared_ptr<arrow_array_container>
 container;
 owned_columns_t view_columns;
 column_view cached_view;
};

class arrow_table {
 public
 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());

 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());

 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());

 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());

 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;

 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;

 [[nodiscard]] table_view view() const;

 private
 std::shared_ptr<arrow_array_container>
 container;
 owned_columns_t view_columns;
 table_view cached_view;
};

} // namespace interop

unique_schema_t to_arrow_schema(cudf::table_view const& input,
 cudf::host_span<column_metadata const> metadata);

unique_device_array_t to_arrow_device(
 cudf::table&& table,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

unique_device_array_t to_arrow_device(
 cudf::column&& col,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

unique_device_array_t to_arrow_device(
 cudf::table_view const& table,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

unique_device_array_t to_arrow_device(
 cudf::column_view const& col,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

unique_device_array_t to_arrow_host(
 cudf::table_view const& table,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

unique_device_array_t to_arrow_host(
 cudf::column_view const& col,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

std::unique_ptr<cudf::table> from_arrow(
 ArrowSchema const* schema,
 ArrowArray const* input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

std::unique_ptr<cudf::column> from_arrow_column(
 ArrowSchema const* schema,
 ArrowArray const* input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

std::unique_ptr<table> from_arrow_host(
 ArrowSchema const* schema,
 ArrowDeviceArray const* input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

std::unique_ptr<table> from_arrow_stream(
 ArrowArrayStream* input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

std::unique_ptr<column> from_arrow_host_column(
 ArrowSchema const* schema,
 ArrowDeviceArray const* input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

unique_table_view_t from_arrow_device(
 ArrowSchema const* schema,
 ArrowDeviceArray const* input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());

unique_column_view_t from_arrow_device_column(
 ArrowSchema const* schema,
 ArrowDeviceArray const* input,
 rmm::cuda_stream_view stream = cudf::get_default_stream(),
 rmm::device_async_resource_ref mr = cudf::get_current_device_resource_ref());
 // end of group
} // namespace CUDF_EXPORT cudf