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

Fix conditional joins with empty left table #9146

Merged
merged 6 commits into from
Sep 8, 2021
Merged

Fix conditional joins with empty left table #9146

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5067f9b
Fix empty left table behavior for all inner and left joins.
vyasr Aug 30, 2021
1908bb4
Ensure that complementary indices are added for full joins as well.
vyasr Aug 30, 2021
0312adc
Skip kernel execution entirely for empty left tables.
vyasr Aug 30, 2021
9947608
Add back missed return.
vyasr Sep 3, 2021
70daa5d
Add default cases for all switches to avoid warnings.
vyasr Sep 3, 2021
d47dd81
Fix and properly comment early return.
vyasr Sep 6, 2021
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
55 changes: 44 additions & 11 deletions cpp/src/join/conditional_join.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,9 +50,8 @@ conditional_join(table_view const& left,
// null index for the right table; in others, we return an empty output.
if (right.num_rows() == 0) {
switch (join_type) {
// Left, left anti, and full (which are effectively left because we are
// guaranteed that left has more rows than right) all return a all the
// row indices from left with a corresponding NULL from the right.
// Left, left anti, and full all return all the row indices from left
// with a corresponding NULL from the right.
case join_kind::LEFT_JOIN:
case join_kind::LEFT_ANTI_JOIN:
case join_kind::FULL_JOIN: return get_trivial_left_join_indices(left, stream);
Expand All @@ -61,6 +60,23 @@ conditional_join(table_view const& left,
case join_kind::LEFT_SEMI_JOIN:
return std::make_pair(std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr),
std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr));
default: CUDF_FAIL("Invalid join kind."); break;
}
} else if (left.num_rows() == 0) {
switch (join_type) {
// Left, left anti, left semi, and inner joins all return empty sets.
case join_kind::LEFT_JOIN:
case join_kind::LEFT_ANTI_JOIN:
case join_kind::INNER_JOIN:
case join_kind::LEFT_SEMI_JOIN:
return std::make_pair(std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr),
std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr));
// Full joins need to return the trivial complement.
case join_kind::FULL_JOIN: {
auto ret_flipped = get_trivial_left_join_indices(right, stream);
return std::make_pair(std::move(ret_flipped.second), std::move(ret_flipped.first));
}
vyasr marked this conversation as resolved.
Show resolved Hide resolved
default: CUDF_FAIL("Invalid join kind."); break;
}
}

Expand Down Expand Up @@ -118,8 +134,14 @@ conditional_join(table_view const& left,

// If the output size will be zero, we can return immediately.
if (join_size == 0) {
return std::make_pair(std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr),
std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr));
auto join_indices{
std::make_pair(std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr),
std::make_unique<rmm::device_uvector<size_type>>(0, stream, mr))};
if (join_type == join_kind::FULL_JOIN) {
auto complement_indices = detail::get_left_join_indices_complement(
join_indices.second, left.num_rows(), right.num_rows(), stream, mr);
return detail::concatenate_vector_pairs(join_indices, complement_indices, stream);
}
vyasr marked this conversation as resolved.
Show resolved Hide resolved
}

rmm::device_scalar<size_type> write_index(0, stream);
Expand Down Expand Up @@ -177,20 +199,31 @@ std::size_t compute_conditional_join_output_size(table_view const& left,
rmm::cuda_stream_view stream,
rmm::mr::device_memory_resource* mr)
{
// We can immediately filter out cases where the right table is empty. In
// some cases, we return all the rows of the left table with a corresponding
// null index for the right table; in others, we return an empty output.
// We can immediately filter out cases where one table is empty. In
// some cases, we return all the rows of the other table with a corresponding
// null index for the empty table; in others, we return an empty output.
if (right.num_rows() == 0) {
switch (join_type) {
// Left, left anti, and full (which are effectively left because we are
// guaranteed that left has more rows than right) all return a all the
// row indices from left with a corresponding NULL from the right.
// Left, left anti, and full all return all the row indices from left
// with a corresponding NULL from the right.
case join_kind::LEFT_JOIN:
case join_kind::LEFT_ANTI_JOIN:
case join_kind::FULL_JOIN: return left.num_rows();
// Inner and left semi joins return empty output because no matches can exist.
case join_kind::INNER_JOIN:
case join_kind::LEFT_SEMI_JOIN: return 0;
default: CUDF_FAIL("Invalid join kind."); break;
}
} else if (left.num_rows() == 0) {
switch (join_type) {
// Left, left anti, left semi, and inner joins all return empty sets.
case join_kind::LEFT_JOIN:
case join_kind::LEFT_ANTI_JOIN:
case join_kind::INNER_JOIN:
case join_kind::LEFT_SEMI_JOIN: return 0;
// Full joins need to return the trivial complement.
case join_kind::FULL_JOIN: return right.num_rows();
default: CUDF_FAIL("Invalid join kind."); break;
}
}

Expand Down
18 changes: 18 additions & 0 deletions cpp/tests/join/conditional_join_tests.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -304,6 +304,11 @@ TYPED_TEST(ConditionalInnerJoinTest, TestOneColumnOneRowAllEqual)
this->test({{0}}, {{0}}, left_zero_eq_right_zero, {{0, 0}});
};

TYPED_TEST(ConditionalInnerJoinTest, TestOneColumnLeftEmpty)
{
this->test({{}}, {{3, 4, 5}}, left_zero_eq_right_zero, {});
};

TYPED_TEST(ConditionalInnerJoinTest, TestOneColumnTwoRowAllEqual)
{
this->test({{0, 1}}, {{0, 0}}, left_zero_eq_right_zero, {{0, 0}, {0, 1}});
Expand Down Expand Up @@ -489,6 +494,11 @@ TYPED_TEST(ConditionalLeftJoinTest, TestTwoColumnThreeRowSomeEqual)
{{0, 0}, {1, 1}, {2, JoinNoneValue}});
};

TYPED_TEST(ConditionalLeftJoinTest, TestOneColumnLeftEmpty)
{
this->test({{}}, {{3, 4, 5}}, left_zero_eq_right_zero, {});
};

TYPED_TEST(ConditionalLeftJoinTest, TestCompareRandomToHash)
{
// Generate columns of 10 repeats of the integer range [0, 10), then merge
Expand Down Expand Up @@ -560,6 +570,14 @@ TYPED_TEST(ConditionalFullJoinTest, TestOneColumnNoneEqual)
{JoinNoneValue, 2}});
};

TYPED_TEST(ConditionalFullJoinTest, TestOneColumnLeftEmpty)
{
this->test({{}},
{{3, 4, 5}},
left_zero_eq_right_zero,
{{JoinNoneValue, 0}, {JoinNoneValue, 1}, {JoinNoneValue, 2}});
};

TYPED_TEST(ConditionalFullJoinTest, TestTwoColumnThreeRowSomeEqual)
{
this->test({{0, 1, 2}, {10, 20, 30}},
Expand Down