Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Implement remaining copying APIs in pylibcudf along with required helper functions #14640

Merged
merged 9 commits into from
Dec 15, 2023
Merged

Implement remaining copying APIs in pylibcudf along with required helper functions #14640

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
3707364
Implement Column.mutable_view and use it to implement copying.copy_ra…
vyasr Nov 28, 2023
5e6c0c1
Implement Column.from_column_view
vyasr Nov 28, 2023
0a48a38
Implement column split
vyasr Nov 28, 2023
02f7480
Implement Table.from_table_view
vyasr Nov 28, 2023
a8d0766
Implement table split
vyasr Nov 28, 2023
e181fbd
Implement slice, has one bug
vyasr Nov 28, 2023
f3273f5
Implement DeviceScalar.from_pylibcudf
vyasr Nov 28, 2023
b3ee1e7
Implement get_element
vyasr Nov 28, 2023
8c8b30e
Some cleanup
vyasr Dec 15, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
246 changes: 38 additions & 208 deletions python/cudf/cudf/_lib/copying.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,30 +19,23 @@ from cudf._lib.column cimport Column
from cudf._lib.scalar import as_device_scalar

from cudf._lib.scalar cimport DeviceScalar
from cudf._lib.utils cimport table_view_from_columns, table_view_from_table
from cudf._lib.utils cimport table_view_from_table

from cudf._lib.reduce import minmax
from cudf.core.abc import Serializable

from libcpp.memory cimport make_unique

cimport cudf._lib.cpp.contiguous_split as cpp_contiguous_split
cimport cudf._lib.cpp.copying as cpp_copying
from cudf._lib.cpp.column.column cimport column
from cudf._lib.cpp.column.column_view cimport column_view, mutable_column_view
from cudf._lib.cpp.column.column_view cimport column_view
from cudf._lib.cpp.lists.gather cimport (
segmented_gather as cpp_segmented_gather,
)
from cudf._lib.cpp.lists.lists_column_view cimport lists_column_view
from cudf._lib.cpp.scalar.scalar cimport scalar
from cudf._lib.cpp.table.table_view cimport table_view
from cudf._lib.cpp.types cimport size_type
from cudf._lib.utils cimport (
columns_from_pylibcudf_table,
columns_from_table_view,
data_from_table_view,
table_view_from_columns,
)
from cudf._lib.utils cimport columns_from_pylibcudf_table, data_from_table_view

# workaround for https://github.com/cython/cython/issues/3885
ctypedef const scalar constscalar
Expand Down Expand Up @@ -92,20 +85,13 @@ def _copy_range_in_place(Column input_column,
size_type input_begin,
size_type input_end,
size_type target_begin):

cdef column_view input_column_view = input_column.view()
cdef mutable_column_view target_column_view = target_column.mutable_view()
cdef size_type c_input_begin = input_begin
cdef size_type c_input_end = input_end
cdef size_type c_target_begin = target_begin

with nogil:
cpp_copying.copy_range_in_place(
input_column_view,
target_column_view,
c_input_begin,
c_input_end,
c_target_begin)
pylibcudf.copying.copy_range(
input_column.to_pylibcudf(mode="write"),
target_column.to_pylibcudf(mode="write"),
input_begin,
input_end,
target_begin
)


def _copy_range(Column input_column,
Expand Down Expand Up @@ -244,198 +230,46 @@ def columns_empty_like(list input_columns):

@acquire_spill_lock()
def column_slice(Column input_column, object indices):

cdef column_view input_column_view = input_column.view()
cdef vector[size_type] c_indices
c_indices.reserve(len(indices))

cdef vector[column_view] c_result

cdef int index

for index in indices:
c_indices.push_back(index)

with nogil:
c_result = move(
cpp_copying.slice(
input_column_view,
c_indices)
return [
Column.from_pylibcudf(c)
for c in pylibcudf.copying.column_slice(
input_column.to_pylibcudf(mode="read"),
list(indices),
)

num_of_result_cols = c_result.size()
result = [
Column.from_column_view(
c_result[i],
input_column) for i in range(num_of_result_cols)]

return result
]


@acquire_spill_lock()
def columns_slice(list input_columns, list indices):
"""
Given a list of input columns, return columns sliced by ``indices``.

Returns a list of list of columns. The length of return is
`len(indices) / 2`. The `i`th item in return is a list of columns sliced
from ``input_columns`` with `slice(indices[i*2], indices[i*2 + 1])`.
"""
cdef table_view input_table_view = table_view_from_columns(input_columns)
cdef vector[size_type] c_indices = indices
cdef vector[table_view] c_result

with nogil:
c_result = move(
cpp_copying.slice(
input_table_view,
c_indices)
)

def columns_slice(list input_columns, object indices):
return [
columns_from_table_view(
c_result[i], input_columns
) for i in range(c_result.size())
columns_from_pylibcudf_table(tbl)
for tbl in pylibcudf.copying.table_slice(
pylibcudf.Table([col.to_pylibcudf(mode="read") for col in input_columns]),
list(indices),
)
]


@acquire_spill_lock()
def column_split(Column input_column, object splits):

cdef column_view input_column_view = input_column.view()
cdef vector[size_type] c_splits
c_splits.reserve(len(splits))

cdef vector[column_view] c_result

cdef int split

for split in splits:
c_splits.push_back(split)

with nogil:
c_result = move(
cpp_copying.split(
input_column_view,
c_splits)
return [
Column.from_pylibcudf(c)
for c in pylibcudf.copying.column_split(
input_column.to_pylibcudf(mode="read"),
list(splits),
)

num_of_result_cols = c_result.size()
result = [
Column.from_column_view(
c_result[i],
input_column
) for i in range(num_of_result_cols)
]

return result


@acquire_spill_lock()
def columns_split(list input_columns, object splits):

cdef table_view input_table_view = table_view_from_columns(input_columns)
cdef vector[size_type] c_splits = splits
cdef vector[table_view] c_result

with nogil:
c_result = move(
cpp_copying.split(
input_table_view,
c_splits)
)

return [
columns_from_table_view(
c_result[i], input_columns
) for i in range(c_result.size())
]


def _copy_if_else_column_column(Column lhs, Column rhs, Column boolean_mask):

cdef column_view lhs_view = lhs.view()
cdef column_view rhs_view = rhs.view()
cdef column_view boolean_mask_view = boolean_mask.view()

cdef unique_ptr[column] c_result

with nogil:
c_result = move(
cpp_copying.copy_if_else(
lhs_view,
rhs_view,
boolean_mask_view
)
)

return Column.from_unique_ptr(move(c_result))


def _copy_if_else_scalar_column(DeviceScalar lhs,
Column rhs,
Column boolean_mask):

cdef const scalar* lhs_scalar = lhs.get_raw_ptr()
cdef column_view rhs_view = rhs.view()
cdef column_view boolean_mask_view = boolean_mask.view()

cdef unique_ptr[column] c_result

with nogil:
c_result = move(
cpp_copying.copy_if_else(
lhs_scalar[0],
rhs_view,
boolean_mask_view
)
columns_from_pylibcudf_table(tbl)
for tbl in pylibcudf.copying.table_split(
pylibcudf.Table([col.to_pylibcudf(mode="read") for col in input_columns]),
list(splits),
)

return Column.from_unique_ptr(move(c_result))


def _copy_if_else_column_scalar(Column lhs,
DeviceScalar rhs,
Column boolean_mask):

cdef column_view lhs_view = lhs.view()
cdef const scalar* rhs_scalar = rhs.get_raw_ptr()
cdef column_view boolean_mask_view = boolean_mask.view()

cdef unique_ptr[column] c_result

with nogil:
c_result = move(
cpp_copying.copy_if_else(
lhs_view,
rhs_scalar[0],
boolean_mask_view
)
)

return Column.from_unique_ptr(move(c_result))


def _copy_if_else_scalar_scalar(DeviceScalar lhs,
DeviceScalar rhs,
Column boolean_mask):

cdef const scalar* lhs_scalar = lhs.get_raw_ptr()
cdef const scalar* rhs_scalar = rhs.get_raw_ptr()
cdef column_view boolean_mask_view = boolean_mask.view()

cdef unique_ptr[column] c_result

with nogil:
c_result = move(
cpp_copying.copy_if_else(
lhs_scalar[0],
rhs_scalar[0],
boolean_mask_view
)
)

return Column.from_unique_ptr(move(c_result))
]


@acquire_spill_lock()
Expand Down Expand Up @@ -504,16 +338,12 @@ def shift(Column input, int offset, object fill_value=None):

@acquire_spill_lock()
def get_element(Column input_column, size_type index):
cdef column_view col_view = input_column.view()

cdef unique_ptr[scalar] c_output
with nogil:
c_output = move(
cpp_copying.get_element(col_view, index)
)

return DeviceScalar.from_unique_ptr(
move(c_output), dtype=input_column.dtype
return DeviceScalar.from_pylibcudf(
pylibcudf.copying.get_element(
input_column.to_pylibcudf(mode="read"),
index,
),
dtype=input_column.dtype,
)


Expand Down
6 changes: 5 additions & 1 deletion python/cudf/cudf/_lib/pylibcudf/column.pxd
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@ from libcpp.memory cimport unique_ptr
from libcpp.vector cimport vector

from cudf._lib.cpp.column.column cimport column
from cudf._lib.cpp.column.column_view cimport column_view
from cudf._lib.cpp.column.column_view cimport column_view, mutable_column_view
from cudf._lib.cpp.types cimport bitmask_type, size_type

from .gpumemoryview cimport gpumemoryview
Expand All @@ -26,10 +26,14 @@ cdef class Column:
size_type _num_children

cdef column_view view(self) nogil
cdef mutable_column_view mutable_view(self) nogil

@staticmethod
cdef Column from_libcudf(unique_ptr[column] libcudf_col)

@staticmethod
cdef Column from_column_view(const column_view& libcudf_col, Column owner)

cpdef DataType type(self)
cpdef Column child(self, size_type index)
cpdef size_type num_children(self)
Expand Down
Loading