Skip to content

Remove Arrow C++ dependency from C++ tests #23708

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

Draft
vyasr wants to merge 1 commit into
base: main
Choose a base branch
from
+175 −91
Draft

Remove Arrow C++ dependency from C++ tests #23708

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
7 changes: 1 addition & 6 deletions cpp/tests/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,8 +88,6 @@ endfunction()
set(CUDF_INSTALL_LIBRARY_DEPS OFF)
set(CUDF_EXCLUDE_DEPS_FROM_ALL ON)
set(CUDF_EXCLUDE_DEPS_FROM_ALL_FLAG EXCLUDE_FROM_ALL)
set(CUDF_ENABLE_ARROW_COMPUTE ON)
include(../cmake/thirdparty/get_arrow.cmake)

# ##################################################################################################
# test sources ##################################################################################
Expand Down Expand Up @@ -223,7 +221,7 @@ ConfigureTest(
QUANTILES_TEST quantiles/percentile_approx_test.cpp quantiles/quantile_test.cpp
quantiles/quantiles_test.cpp
GPUS 1
PERCENT 70 EXTRA_LIBS ${ARROW_LIBRARIES}
PERCENT 70
)

# ##################################################################################################
Expand Down Expand Up @@ -305,16 +303,13 @@ ConfigureTest(
INTEROP_TEST
interop/arrow_data_structures_test.cpp
interop/to_arrow_device_test.cpp
interop/to_arrow_test.cpp
interop/to_arrow_host_test.cpp
interop/from_arrow_test.cpp
interop/from_arrow_device_test.cpp
interop/from_arrow_host_test.cpp
interop/from_arrow_stream_test.cpp
interop/dlpack_test.cpp
EXTRA_LIBS
nanoarrow::nanoarrow
${ARROW_LIBRARIES}
)

# ##################################################################################################
Expand Down
101 changes: 101 additions & 0 deletions cpp/tests/interop/from_arrow_host_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,92 @@

#include <cuda/iterator>

#include <array>
#include <cstring>

namespace {

void release_schema(ArrowSchema* schema) { schema->release = nullptr; }

void release_array(ArrowArray* array) { array->release = nullptr; }

struct direct_arrow_c_producer {
static constexpr int64_t num_rows = 5;

std::array<int32_t, num_rows> int_values{1, 2, 5, 2, 7};
std::array<uint8_t, 1> int_validity{0b00011101};

std::array<int32_t, num_rows + 1> string_offsets{0, 3, 6, 6, 6, 9};
std::array<char, 9> string_chars{'f', 'f', 'f', 'a', 'a', 'a', 'c', 'c', 'c'};
std::array<uint8_t, 1> string_validity{0b00010111};

ArrowSchema schema{};
std::array<ArrowSchema, 2> child_schemas{};
std::array<ArrowSchema*, 2> child_schema_ptrs{};

ArrowArray array{};
std::array<ArrowArray, 2> child_arrays{};
std::array<ArrowArray*, 2> child_array_ptrs{};
std::array<void const*, 1> parent_buffers{nullptr};
std::array<void const*, 2> int_buffers{int_validity.data(), int_values.data()};
std::array<void const*, 3> string_buffers{
string_validity.data(), string_offsets.data(), string_chars.data()};

direct_arrow_c_producer()
{
child_schema_ptrs = {&child_schemas[0], &child_schemas[1]};
child_array_ptrs = {&child_arrays[0], &child_arrays[1]};

schema.format = "+s";
schema.name = "";
schema.flags = 0;
schema.n_children = child_schemas.size();
schema.children = child_schema_ptrs.data();
schema.release = release_schema;

child_schemas[0].format = "i";
child_schemas[0].name = "ints";
child_schemas[0].flags = ARROW_FLAG_NULLABLE;
child_schemas[0].release = release_schema;

child_schemas[1].format = "u";
child_schemas[1].name = "strings";
child_schemas[1].flags = ARROW_FLAG_NULLABLE;
child_schemas[1].release = release_schema;

array.length = num_rows;
array.null_count = 0;
array.n_buffers = parent_buffers.size();
array.n_children = child_arrays.size();
array.buffers = parent_buffers.data();
array.children = child_array_ptrs.data();
array.release = release_array;

child_arrays[0].length = num_rows;
child_arrays[0].null_count = 1;
child_arrays[0].n_buffers = int_buffers.size();
child_arrays[0].buffers = int_buffers.data();
child_arrays[0].release = release_array;

child_arrays[1].length = num_rows;
child_arrays[1].null_count = 1;
child_arrays[1].n_buffers = string_buffers.size();
child_arrays[1].buffers = string_buffers.data();
child_arrays[1].release = release_array;
}

ArrowDeviceArray device_array() const
{
ArrowDeviceArray out{};
std::memcpy(&out.array, &array, sizeof(ArrowArray));
out.device_type = ARROW_DEVICE_CPU;
out.device_id = -1;
return out;
}
};

} // namespace

// create a cudf::table and equivalent arrow table with host memory
std::tuple<std::unique_ptr<cudf::table>, nanoarrow::UniqueSchema, nanoarrow::UniqueArray>
get_nanoarrow_host_tables(cudf::size_type length)
Expand Down Expand Up @@ -106,6 +192,21 @@ TEST_F(FromArrowHostDeviceTest, EmptyTable)
CUDF_TEST_EXPECT_TABLES_EQUAL(expected_cudf_table, got_cudf_table->view());
}

TEST_F(FromArrowHostDeviceTest, DirectArrowCProducerTable)
{
direct_arrow_c_producer producer;
auto input = producer.device_array();

auto const expected_ints =
cudf::test::fixed_width_column_wrapper<int32_t>{{1, 2, 5, 2, 7}, {1, 0, 1, 1, 1}};
auto const expected_strings =
cudf::test::strings_column_wrapper{{"fff", "aaa", "", "xxx", "ccc"}, {1, 1, 1, 0, 1}};
auto const expected = cudf::table_view{{expected_ints, expected_strings}};

auto got_cudf_table = cudf::from_arrow_host(&producer.schema, &input);
CUDF_TEST_EXPECT_TABLES_EQUIVALENT(expected, got_cudf_table->view());
}

TEST_F(FromArrowHostDeviceTest, ZeroColumnsWithRows)
{
constexpr cudf::size_type num_rows = 5;
Expand Down
158 changes: 73 additions & 85 deletions cpp/tests/quantiles/percentile_approx_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION.
* SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: Apache-2.0
*/

Expand All @@ -14,113 +14,31 @@
#include <cudf/groupby.hpp>
#include <cudf/quantiles.hpp>
#include <cudf/reduction.hpp>
#include <cudf/sorting.hpp>
#include <cudf/tdigest/tdigest_column_view.hpp>
#include <cudf/transform.hpp>
#include <cudf/utilities/default_stream.hpp>
#include <cudf/utilities/error.hpp>
#include <cudf/utilities/memory_resource.hpp>

#include <cuda/iterator>

#include <arrow/api.h>
#include <arrow/compute/api.h>
#include <arrow/compute/initialize.h>

namespace {
std::unique_ptr<cudf::column> arrow_percentile_approx(cudf::column_view const& _values,
int delta,
std::vector<double> const& percentages)
{
static auto const _arrow_init_status = arrow::compute::Initialize();
EXPECT_TRUE(_arrow_init_status.ok());

auto stream = cudf::get_default_stream();

// sort the incoming values using the same settings that groupby does.
// this is a little weak because null_order::AFTER is hardcoded internally to groupby.
cudf::table_view t({_values});
auto sorted_t = cudf::sort(t, {}, {cudf::null_order::AFTER}, stream);
auto sorted_values = sorted_t->get_column(0).view();

std::vector<double> h_values(sorted_values.size());
CUDF_CUDA_TRY(cudaMemcpyAsync(h_values.data(),
sorted_values.data<double>(),
sizeof(double) * sorted_values.size(),
cudaMemcpyDefault,
stream.value()));
std::vector<char> h_validity(sorted_values.size());
if (sorted_values.null_mask() != nullptr) {
auto validity = cudf::mask_to_bools(sorted_values.null_mask(), 0, sorted_values.size(), stream);
CUDF_CUDA_TRY(cudaMemcpyAsync(h_validity.data(),
(validity->view().data<char>()),
sizeof(char) * sorted_values.size(),
cudaMemcpyDefault,
stream.value()));
}

// generate the tdigest
arrow::DoubleBuilder builder;
for (size_t idx = 0; idx < h_values.size(); idx++) {
if (sorted_values.null_mask() == nullptr || h_validity[idx]) {
EXPECT_TRUE(builder.Append(h_values[idx]).ok());
}
}
std::shared_ptr<arrow::Array> array;
EXPECT_TRUE(builder.Finish(&array).ok());

auto const udelta = static_cast<uint32_t>(delta);
auto const usize = static_cast<uint32_t>(h_values.size()) * 2;
arrow::compute::TDigestOptions options{percentages, udelta, usize};

auto arrow_result = arrow::compute::CallFunction("tdigest", {array}, &options);
auto result_array = arrow_result.ValueOrDie().array_as<arrow::DoubleArray>();

// copy the percentiles and stuff them into a list column
std::vector<double> h_result;
h_result.reserve(percentages.size());
std::transform(
result_array->begin(), result_array->end(), std::back_inserter(h_result), [](auto p) {
return p.value();
});
cudf::test::fixed_width_column_wrapper<double> result(h_result.begin(), h_result.end());
cudf::test::fixed_width_column_wrapper<cudf::size_type> offsets{
0, static_cast<cudf::size_type>(percentages.size())};
stream.synchronize();
return cudf::make_lists_column(1, offsets.release(), result.release(), 0, {});
}

struct percentile_approx_dispatch {
template <typename T, typename Func>
std::unique_ptr<cudf::column> operator()(Func op,
cudf::column_view const& values,
int delta,
std::vector<double> const& percentages,
cudf::size_type ulps)
[[maybe_unused]] cudf::size_type ulps)
requires(cudf::is_numeric<T>() || cudf::is_fixed_point<T>())
{
// gpu implementation
// gpu implementation.
auto agg_result = op(values, delta);

cudf::test::fixed_width_column_wrapper<double> g_percentages(percentages.begin(),
percentages.end());
cudf::tdigest::tdigest_column_view tdv(*agg_result);
auto result = cudf::percentile_approx(tdv, g_percentages);

// disable checking logic during a racecheck run
if (getenv("LIBCUDF_RACECHECK_ENABLED")) { return result; }

// arrow implementation.
auto expected = [&]() {
// we're explicitly casting back to doubles here but this is ok because that is
// exactly what happens inside of the cudf implementation as values are processed as well.
// so this should not affect results.
auto as_doubles = cudf::cast(values, cudf::data_type{cudf::type_id::FLOAT64});
return arrow_percentile_approx(*as_doubles, delta, percentages);
}();
cudf::test::detail::expect_columns_equivalent(
*expected, *result, cudf::test::debug_output_level::FIRST_ERROR, ulps);

return result;
}

Expand Down Expand Up @@ -494,6 +412,76 @@ TEST_F(PercentileApproxTest, NullPercentiles)
CUDF_TEST_EXPECT_COLUMNS_EQUAL(*result, expected);
}

TEST_F(PercentileApproxTest, ReductionGold)
{
auto const delta = 1000;

auto const values = cudf::test::fixed_width_column_wrapper<double>{9, 1, 7, 5, 2};

auto const tdigest =
cudf::reduce(values,
*cudf::make_tdigest_aggregation<cudf::reduce_aggregation>(delta),
cudf::data_type{cudf::type_id::STRUCT});
auto const tdigest_col = cudf::make_column_from_scalar(*tdigest, 1);

auto const percentiles =
cudf::test::fixed_width_column_wrapper<double>{0.0, 0.25, 0.5, 0.75, 1.0};
auto const result = cudf::percentile_approx(tdigest_col->view(), percentiles);

auto const expected = cudf::test::lists_column_wrapper<double>{{1, 2, 5, 7, 9}};
CUDF_TEST_EXPECT_COLUMNS_EQUAL(*result, expected);
}

TEST_F(PercentileApproxTest, GroupByGold)
{
auto const delta = 1000;

auto const values =
cudf::test::fixed_width_column_wrapper<double>{9, 1, 7, 5, 2, 50, 10, 40, 20, 30};
auto const keys = cudf::test::fixed_width_column_wrapper<int32_t>{0, 0, 0, 0, 0, 1, 1, 1, 1, 1};
auto const percentiles =
cudf::test::fixed_width_column_wrapper<double>{0.0, 0.25, 0.5, 0.75, 1.0};

cudf::groupby::groupby gb(cudf::table_view{{keys}});
std::vector<cudf::groupby::aggregation_request> requests;
std::vector<std::unique_ptr<cudf::groupby_aggregation>> aggregations;
aggregations.push_back(cudf::make_tdigest_aggregation<cudf::groupby_aggregation>(delta));
requests.push_back({values, std::move(aggregations)});
auto const tdigest_column = gb.aggregate(requests);

cudf::tdigest::tdigest_column_view tdv(*tdigest_column.second[0].results[0]);
auto const result = cudf::percentile_approx(tdv, percentiles);

auto const expected =
cudf::test::lists_column_wrapper<double>{{1, 2, 5, 7, 9}, {10, 20, 30, 40, 50}};
CUDF_TEST_EXPECT_COLUMNS_EQUAL(*result, expected);
}

TEST_F(PercentileApproxTest, GroupByWithNullsGold)
{
auto const delta = 1000;

auto const values = cudf::test::fixed_width_column_wrapper<double>{
{9, 99, 7, 5, 1, 50, 10, 40, 20, 30}, {1, 0, 1, 1, 1, 1, 1, 1, 1, 1}};
auto const keys = cudf::test::fixed_width_column_wrapper<int32_t>{0, 0, 0, 0, 0, 1, 1, 1, 1, 1};
auto const percentiles =
cudf::test::fixed_width_column_wrapper<double>{0.0, 0.25, 0.5, 0.75, 1.0};

cudf::groupby::groupby gb(cudf::table_view{{keys}});
std::vector<cudf::groupby::aggregation_request> requests;
std::vector<std::unique_ptr<cudf::groupby_aggregation>> aggregations;
aggregations.push_back(cudf::make_tdigest_aggregation<cudf::groupby_aggregation>(delta));
requests.push_back({values, std::move(aggregations)});
auto const tdigest_column = gb.aggregate(requests);

cudf::tdigest::tdigest_column_view tdv(*tdigest_column.second[0].results[0]);
auto const result = cudf::percentile_approx(tdv, percentiles);

auto const expected =
cudf::test::lists_column_wrapper<double>{{1, 1, 5, 7, 9}, {10, 20, 30, 40, 50}};
CUDF_TEST_EXPECT_COLUMNS_EQUAL(*result, expected);
}

TEST_F(PercentileApproxTest, ReductionWithLowRowCount)
{
// Test that the tdigest reduction with a low row count still produces the correct results.
Expand Down
Loading