Add stream parameters in pylibcudf IO APIs

python/pylibcudf/pylibcudf/io/avro.pxd

@@ -2,7 +2,7 @@
 from pylibcudf.io.types cimport SourceInfo, TableWithMetadata
 from pylibcudf.libcudf.io.avro cimport avro_reader_options, avro_reader_options_builder
 from pylibcudf.libcudf.types cimport size_type
-
+from rmm._cuda.stream cimport Stream
 
 from pylibcudf.libcudf.types cimport size_type
 
@@ -20,4 +20,4 @@ cdef class AvroReaderOptionsBuilder:
     cpdef AvroReaderOptionsBuilder num_rows(self, size_type num_rows)
     cpdef AvroReaderOptions build(self)
 
-cpdef TableWithMetadata read_avro(AvroReaderOptions options)
+cpdef TableWithMetadata read_avro(AvroReaderOptions options, Stream stream = *)

python/pylibcudf/pylibcudf/io/avro.pyi

@@ -1,4 +1,7 @@
 # Copyright (c) 2024, NVIDIA CORPORATION.
+
+from rmm._cuda.stream import Stream
+
 from pylibcudf.io.types import SourceInfo, TableWithMetadata
 
 __all__ = ["AvroReaderOptions", "AvroReaderOptionsBuilder", "read_avro"]
@@ -13,4 +16,6 @@ class AvroReaderOptionsBuilder:
     def num_rows(num_rows: int) -> AvroReaderOptionsBuilder: ...
     def build(self) -> AvroReaderOptions: ...
 
-def read_avro(options: AvroReaderOptions) -> TableWithMetadata: ...
+def read_avro(
+    options: AvroReaderOptions, stream: Stream = None
+) -> TableWithMetadata: ...

python/pylibcudf/pylibcudf/io/avro.pyx

@@ -9,6 +9,7 @@ from pylibcudf.libcudf.io.avro cimport (
     read_avro as cpp_read_avro,
 )
 from pylibcudf.libcudf.types cimport size_type
+from rmm._cuda.stream cimport Stream
 
 __all__ = ["read_avro", "AvroReaderOptions", "AvroReaderOptionsBuilder"]
 
@@ -126,7 +127,8 @@ cdef class AvroReaderOptionsBuilder:
 
 
 cpdef TableWithMetadata read_avro(
-    AvroReaderOptions options
+    AvroReaderOptions options,
+    Stream stream = None,
 ):
     """
     Read from Avro format.
@@ -141,7 +143,9 @@ cpdef TableWithMetadata read_avro(
     options: AvroReaderOptions
         Settings for controlling reading behavior
     """
+    if stream is None:
+        stream = Stream()
     with nogil:
-        c_result = move(cpp_read_avro(options.c_obj))
+        c_result = move(cpp_read_avro(options.c_obj, stream.view()))
 
     return TableWithMetadata.from_libcudf(c_result)

python/pylibcudf/pylibcudf/io/csv.pxd

@@ -18,6 +18,7 @@ from pylibcudf.libcudf.io.types cimport (
     table_with_metadata,
 )
 from pylibcudf.libcudf.types cimport size_type
+from rmm._cuda.stream cimport Stream
 
 cdef class CsvReaderOptions:
     cdef csv_reader_options c_obj
@@ -61,7 +62,7 @@ cdef class CsvReaderOptionsBuilder:
     cpdef CsvReaderOptionsBuilder dayfirst(self, bool dayfirst)
     cpdef CsvReaderOptions build(self)
 
-cpdef TableWithMetadata read_csv(CsvReaderOptions options)
+cpdef TableWithMetadata read_csv(CsvReaderOptions options, Stream stream = *)
 
 cdef class CsvWriterOptions:
     cdef csv_writer_options c_obj
@@ -84,4 +85,4 @@ cdef class CsvWriterOptionsBuilder:
     cpdef CsvWriterOptions build(self)
 
 
-cpdef void write_csv(CsvWriterOptions options)
+cpdef void write_csv(CsvWriterOptions options, Stream stream = *)

python/pylibcudf/pylibcudf/io/csv.pyi

@@ -1,9 +1,9 @@
 # Copyright (c) 2024, NVIDIA CORPORATION.
 
-from collections.abc import Mapping
-
 from typing_extensions import Self
 
+from rmm._cuda.stream import Stream
+
 from pylibcudf.io.types import (
     CompressionType,
     QuoteStyle,
@@ -56,48 +56,10 @@ class CsvReaderOptionsBuilder:
     def build(self) -> CsvReaderOptions: ...
 
 def read_csv(
-    source_info: SourceInfo,
-    *,
-    compression: CompressionType = CompressionType.AUTO,
-    byte_range_offset: int = 0,
-    byte_range_size: int = 0,
-    col_names: list[str] | None = None,
-    prefix: str = "",
-    mangle_dupe_cols: bool = True,
-    usecols: list[int] | list[str] | None = None,
-    nrows: int = -1,
-    skiprows: int = 0,
-    skipfooter: int = 0,
-    header: int = 0,
-    lineterminator: str = "\n",
-    delimiter: str | None = None,
-    thousands: str | None = None,
-    decimal: str = ".",
-    comment: str | None = None,
-    delim_whitespace: bool = False,
-    skipinitialspace: bool = False,
-    skip_blank_lines: bool = True,
-    quoting: QuoteStyle = QuoteStyle.MINIMAL,
-    quotechar: str = '"',
-    doublequote: bool = True,
-    parse_dates: list[str] | list[int] | None = None,
-    parse_hex: list[str] | list[int] | None = None,
-    # Technically this should be dict/list
-    # but using a fused type prevents using None as default
-    dtypes: Mapping[str, DataType] | list[DataType] | None = None,
-    true_values: list[str] | None = None,
-    false_values: list[str] | None = None,
-    na_values: list[str] | None = None,
-    keep_default_na: bool = True,
-    na_filter: bool = True,
-    dayfirst: bool = False,
-    # Note: These options are supported by the libcudf reader
-    # but are not exposed here since there is no demand for them
-    # on the Python side yet.
-    # detect_whitespace_around_quotes: bool = False,
-    # timestamp_type: DataType = DataType(type_id.EMPTY),
+    options: CsvReaderOptions,
+    stream: Stream = None,
 ) -> TableWithMetadata: ...
-def write_csv(options: CsvWriterOptionsBuilder): ...
+def write_csv(options: CsvWriterOptionsBuilder, stream: Stream = None): ...
 
 class CsvWriterOptions:
     def __init__(self): ...

python/pylibcudf/pylibcudf/io/csv.pyx

@@ -22,6 +22,7 @@ from pylibcudf.libcudf.io.types cimport (
 from pylibcudf.libcudf.types cimport data_type, size_type
 from pylibcudf.types cimport DataType
 from pylibcudf.table cimport Table
+from rmm._cuda.stream cimport Stream
 
 __all__ = [
     "read_csv",
@@ -629,7 +630,8 @@ cdef class CsvReaderOptionsBuilder:
 
 
 cpdef TableWithMetadata read_csv(
-    CsvReaderOptions options
+    CsvReaderOptions options,
+    Stream stream = None,
 ):
     """
     Read from CSV format.
@@ -644,9 +646,11 @@ cpdef TableWithMetadata read_csv(
     options: CsvReaderOptions
         Settings for controlling reading behavior
     """
+    if stream is None:
+        stream = Stream()
     cdef table_with_metadata c_result
     with nogil:
-        c_result = move(cpp_read_csv(options.c_obj))
+        c_result = move(cpp_read_csv(options.c_obj, stream.view()))
 
     cdef TableWithMetadata tbl_meta = TableWithMetadata.from_libcudf(c_result)
     return tbl_meta
@@ -831,7 +835,8 @@ cdef class CsvWriterOptionsBuilder:
 
 
 cpdef void write_csv(
-    CsvWriterOptions options
+    CsvWriterOptions options,
+    Stream stream = None,
 ):
     """
     Write to CSV format.
@@ -846,6 +851,7 @@ cpdef void write_csv(
     options: CsvWriterOptions
         Settings for controlling writing behavior
     """
-
+    if stream is None:
+        stream = Stream()
     with nogil:
-        cpp_write_csv(move(options.c_obj))
+        cpp_write_csv(move(options.c_obj), stream.view())

python/pylibcudf/pylibcudf/io/json.pxd

@@ -14,6 +14,7 @@ from pylibcudf.libcudf.io.json cimport (
     json_writer_options_builder,
 )
 from pylibcudf.libcudf.types cimport size_type
+from rmm._cuda.stream cimport Stream
 from pylibcudf.table cimport Table
 
 
@@ -73,9 +74,10 @@ cdef class JsonWriterOptionsBuilder:
     cpdef JsonWriterOptionsBuilder lines(self, bool val)
     cpdef JsonWriterOptions build(self)
 
-cpdef void write_json(JsonWriterOptions options)
+cpdef void write_json(JsonWriterOptions options, Stream stream = *)
 
 cpdef tuple chunked_read_json(
     JsonReaderOptions options,
     int chunk_size= *,
+    Stream stream = *,
 )

python/pylibcudf/pylibcudf/io/json.pyi

@@ -4,6 +4,8 @@ from typing import TypeAlias
 
 from typing_extensions import Self
 
+from rmm._cuda.stream import Stream
+
 from pylibcudf.column import Column
 from pylibcudf.io.types import (
     CompressionType,
@@ -68,8 +70,9 @@ class JsonWriterOptionsBuilder:
     def lines(self, val: bool) -> Self: ...
     def build(self) -> JsonWriterOptions: ...
 
-def write_json(options: JsonWriterOptions) -> None: ...
+def write_json(options: JsonWriterOptions, stream: Stream = None) -> None: ...
 def chunked_read_json(
     options: JsonReaderOptions,
     chunk_size: int = 100_000_000,
+    stream: Stream = None,
 ) -> tuple[list[Column], list[str], ChildNameToTypeMap]: ...

python/pylibcudf/pylibcudf/io/json.pyx

@@ -20,6 +20,8 @@ from pylibcudf.libcudf.io.types cimport (
 )
 from pylibcudf.libcudf.types cimport data_type, size_type
 from pylibcudf.types cimport DataType
+from rmm._cuda.stream cimport Stream
+
 
 __all__ = [
     "chunked_read_json",
@@ -422,6 +424,7 @@ cdef class JsonReaderOptionsBuilder:
 cpdef tuple chunked_read_json(
     JsonReaderOptions options,
     int chunk_size=100_000_000,
+    Stream stream = None,
 ):
     """
     Reads chunks of a JSON file into a :py:class:`~.types.TableWithMetadata`.
@@ -442,6 +445,9 @@ cpdef tuple chunked_read_json(
     cdef size_type c_range_size = (
         chunk_size if chunk_size is not None else 0
     )
+    if stream is None:
+        stream = Stream()
+
     cdef table_with_metadata c_result
 
     final_columns = []
@@ -455,7 +461,7 @@ cpdef tuple chunked_read_json(
 
         try:
             with nogil:
-                c_result = move(cpp_read_json(options.c_obj))
+                c_result = move(cpp_read_json(opts, stream.view()))
         except (ValueError, OverflowError):
             break
         if meta_names is None:
@@ -483,7 +489,8 @@ cpdef tuple chunked_read_json(
 
 
 cpdef TableWithMetadata read_json(
-    JsonReaderOptions options
+    JsonReaderOptions options,
+    Stream stream = None,
 ):
     """
     Read from JSON format.
@@ -503,10 +510,13 @@ cpdef TableWithMetadata read_json(
     TableWithMetadata
         The Table and its corresponding metadata (column names) that were read in.
     """
+    if stream is None:
+        stream = Stream()
+
     cdef table_with_metadata c_result
 
     with nogil:
-        c_result = move(cpp_read_json(options.c_obj))
+        c_result = move(cpp_read_json(opts, stream.view()))
 
     return TableWithMetadata.from_libcudf(c_result)
 
@@ -664,7 +674,7 @@ cdef class JsonWriterOptionsBuilder:
         return json_options
 
 
-cpdef void write_json(JsonWriterOptions options):
+cpdef void write_json(JsonWriterOptions options, Stream stream = None):
     """
     Writes a set of columns to JSON format.
 
@@ -677,5 +687,7 @@ cpdef void write_json(JsonWriterOptions options):
     -------
     None
     """
+    if stream is None:
+        stream = Stream()
     with nogil:
-        cpp_write_json(options.c_obj)
+        cpp_write_json(options.c_obj, stream.view())

python/pylibcudf/pylibcudf/io/orc.pxd

@@ -33,6 +33,8 @@ from pylibcudf.libcudf.io.types cimport (
     compression_type,
     statistics_freq,
 )
+from rmm._cuda.stream cimport Stream
+
 
 cdef class OrcReaderOptions:
     cdef orc_reader_options c_obj
@@ -93,7 +95,7 @@ cdef class OrcWriterOptionsBuilder:
     cpdef OrcWriterOptionsBuilder metadata(self, TableInputMetadata meta)
     cpdef OrcWriterOptions build(self)
 
-cpdef void write_orc(OrcWriterOptions options)
+cpdef void write_orc(OrcWriterOptions options, Stream stream = *)
 
 cdef class OrcChunkedWriter:
     cdef unique_ptr[orc_chunked_writer] c_obj