|
| 1 | +/* |
| 2 | + * Copyright (c) 2025, NVIDIA CORPORATION. |
| 3 | + * |
| 4 | + * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | + * you may not use this file except in compliance with the License. |
| 6 | + * You may obtain a copy of the License at |
| 7 | + * |
| 8 | + * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | + * |
| 10 | + * Unless required by applicable law or agreed to in writing, software |
| 11 | + * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | + * See the License for the specific language governing permissions and |
| 14 | + * limitations under the License. |
| 15 | + */ |
| 16 | + |
| 17 | +#include <test_utils.hpp> |
| 18 | + |
| 19 | +#include <cuco/bloom_filter.cuh> |
| 20 | +#include <cuco/utility/error.hpp> |
| 21 | + |
| 22 | +#include <cuda/functional> |
| 23 | +#include <thrust/device_vector.h> |
| 24 | +#include <thrust/iterator/counting_iterator.h> |
| 25 | + |
| 26 | +#include <catch2/catch_template_test_macros.hpp> |
| 27 | +#include <catch2/generators/catch_generators.hpp> |
| 28 | + |
| 29 | +#include <cstdint> |
| 30 | +#include <exception> |
| 31 | + |
| 32 | +using size_type = int32_t; |
| 33 | + |
| 34 | +template <typename Filter> |
| 35 | +void test_merge_intersect(Filter& filter_a, |
| 36 | + Filter& filter_b, |
| 37 | + Filter const& filter_c, |
| 38 | + size_type capacity) |
| 39 | +{ |
| 40 | + using Key = typename Filter::key_type; |
| 41 | + |
| 42 | + size_type num_keys = capacity; |
| 43 | + size_type half_keys = capacity / 2; |
| 44 | + |
| 45 | + // Set A: [0, capacity) |
| 46 | + auto keys_a_begin = thrust::counting_iterator<Key>{static_cast<Key>(0)}; |
| 47 | + auto keys_a_end = keys_a_begin + num_keys; |
| 48 | + |
| 49 | + // Set B: [capacity/2, capacity + capacity/2) (50% overlap with A) |
| 50 | + auto keys_b_begin = thrust::counting_iterator<Key>{static_cast<Key>(half_keys)}; |
| 51 | + auto keys_b_end = keys_b_begin + num_keys; |
| 52 | + |
| 53 | + // Intersection: [capacity/2, capacity) |
| 54 | + auto keys_intersection_begin = thrust::counting_iterator<Key>{static_cast<Key>(half_keys)}; |
| 55 | + auto keys_intersection_end = keys_intersection_begin + half_keys; |
| 56 | + |
| 57 | + // Union: [0, capacity + capacity/2) |
| 58 | + auto keys_union_begin = thrust::counting_iterator<Key>{static_cast<Key>(0)}; |
| 59 | + auto keys_union_end = keys_union_begin + num_keys + half_keys; |
| 60 | + |
| 61 | + // Unique A: [0, capacity/2) |
| 62 | + auto keys_unique_a_begin = thrust::counting_iterator<Key>{static_cast<Key>(0)}; |
| 63 | + auto keys_unique_a_end = keys_unique_a_begin + half_keys; |
| 64 | + |
| 65 | + // Unique B: [capacity, capacity + capacity/2) |
| 66 | + auto keys_unique_b_begin = thrust::counting_iterator<Key>{static_cast<Key>(num_keys)}; |
| 67 | + auto keys_unique_b_end = keys_unique_b_begin + half_keys; |
| 68 | + |
| 69 | + // Helper to fill filters |
| 70 | + auto refill_filters = [&]() { |
| 71 | + filter_a.clear(); |
| 72 | + filter_a.add(keys_a_begin, keys_a_end); |
| 73 | + |
| 74 | + filter_b.clear(); |
| 75 | + filter_b.add(keys_b_begin, keys_b_end); |
| 76 | + }; |
| 77 | + |
| 78 | + // Reusable output vector (sized for largest query: union) |
| 79 | + thrust::device_vector<bool> contained(num_keys + half_keys); |
| 80 | + |
| 81 | + SECTION("Merge B into A") |
| 82 | + { |
| 83 | + refill_filters(); |
| 84 | + filter_a.merge(filter_b); |
| 85 | + |
| 86 | + // Check A contains all of Union |
| 87 | + filter_a.contains(keys_union_begin, keys_union_end, contained.begin()); |
| 88 | + REQUIRE(cuco::test::all_of( |
| 89 | + contained.begin(), contained.begin() + num_keys + half_keys, cuda::std::identity{})); |
| 90 | + |
| 91 | + // Check B is unchanged |
| 92 | + filter_b.contains(keys_b_begin, keys_b_end, contained.begin()); |
| 93 | + REQUIRE( |
| 94 | + cuco::test::all_of(contained.begin(), contained.begin() + num_keys, cuda::std::identity{})); |
| 95 | + } |
| 96 | + |
| 97 | + SECTION("Intersect B into A") |
| 98 | + { |
| 99 | + refill_filters(); |
| 100 | + filter_a.intersect(filter_b); |
| 101 | + |
| 102 | + // Check A contains Intersection |
| 103 | + filter_a.contains(keys_intersection_begin, keys_intersection_end, contained.begin()); |
| 104 | + REQUIRE( |
| 105 | + cuco::test::all_of(contained.begin(), contained.begin() + half_keys, cuda::std::identity{})); |
| 106 | + |
| 107 | + // Check A does NOT contain Unique A (approximate) |
| 108 | + // We expect none_of, but due to false positives, we might get some. |
| 109 | + // However, for this test configuration, we expect 0 false positives if the filter is |
| 110 | + // reasonably sized. |
| 111 | + filter_a.contains(keys_unique_a_begin, keys_unique_a_end, contained.begin()); |
| 112 | + REQUIRE( |
| 113 | + cuco::test::none_of(contained.begin(), contained.begin() + half_keys, cuda::std::identity{})); |
| 114 | + |
| 115 | + // Check A does NOT contain Unique B |
| 116 | + filter_a.contains(keys_unique_b_begin, keys_unique_b_end, contained.begin()); |
| 117 | + REQUIRE( |
| 118 | + cuco::test::none_of(contained.begin(), contained.begin() + half_keys, cuda::std::identity{})); |
| 119 | + } |
| 120 | + |
| 121 | + SECTION("Merge empty filter into A") |
| 122 | + { |
| 123 | + filter_a.clear(); |
| 124 | + filter_a.add(keys_a_begin, keys_a_end); |
| 125 | + filter_b.clear(); // B is empty |
| 126 | + |
| 127 | + filter_a.merge(filter_b); |
| 128 | + |
| 129 | + // A should still contain all of Set A |
| 130 | + filter_a.contains(keys_a_begin, keys_a_end, contained.begin()); |
| 131 | + REQUIRE( |
| 132 | + cuco::test::all_of(contained.begin(), contained.begin() + num_keys, cuda::std::identity{})); |
| 133 | + } |
| 134 | + |
| 135 | + SECTION("Intersect empty filter into A") |
| 136 | + { |
| 137 | + filter_a.clear(); |
| 138 | + filter_a.add(keys_a_begin, keys_a_end); |
| 139 | + filter_b.clear(); // B is empty |
| 140 | + |
| 141 | + filter_a.intersect(filter_b); |
| 142 | + |
| 143 | + // A should now be empty (intersection with empty set) |
| 144 | + filter_a.contains(keys_a_begin, keys_a_end, contained.begin()); |
| 145 | + REQUIRE( |
| 146 | + cuco::test::none_of(contained.begin(), contained.begin() + num_keys, cuda::std::identity{})); |
| 147 | + } |
| 148 | + |
| 149 | + SECTION("Mismatched block counts") |
| 150 | + { |
| 151 | + // also test with custom stream |
| 152 | + cudaStream_t stream; |
| 153 | + cudaStreamCreate(&stream); |
| 154 | + REQUIRE_THROWS_AS(filter_a.merge(filter_c, stream), cuco::logic_error); |
| 155 | + REQUIRE_THROWS_AS(filter_a.intersect(filter_c, stream), cuco::logic_error); |
| 156 | + cudaStreamDestroy(stream); |
| 157 | + } |
| 158 | +} |
| 159 | + |
| 160 | +TEMPLATE_TEST_CASE_SIG( |
| 161 | + "bloom_filter merge and intersect tests", |
| 162 | + "", |
| 163 | + ((class Key, class Policy), Key, Policy), |
| 164 | + (int32_t, cuco::default_filter_policy<cuco::xxhash_64<int32_t>, uint32_t, 1>), |
| 165 | + (int32_t, cuco::default_filter_policy<cuco::xxhash_64<int32_t>, uint32_t, 8>), |
| 166 | + (int64_t, cuco::default_filter_policy<cuco::xxhash_64<int64_t>, uint64_t, 1>), |
| 167 | + (int64_t, cuco::default_filter_policy<cuco::xxhash_64<int64_t>, uint64_t, 8>)) |
| 168 | +{ |
| 169 | + using filter_type = |
| 170 | + cuco::bloom_filter<Key, cuco::extent<size_t>, cuda::thread_scope_device, Policy>; |
| 171 | + constexpr size_type capacity{1000}; |
| 172 | + |
| 173 | + uint32_t pattern_bits = Policy::words_per_block + GENERATE(0, 1); |
| 174 | + |
| 175 | + // some parameter combinations might be invalid so we skip them |
| 176 | + try { |
| 177 | + [[maybe_unused]] auto policy = Policy{pattern_bits}; |
| 178 | + } catch (std::exception const& e) { |
| 179 | + SKIP(e.what()); |
| 180 | + } |
| 181 | + |
| 182 | + auto filter_a = filter_type{capacity, {}, {pattern_bits}}; |
| 183 | + auto filter_b = filter_type{capacity, {}, {pattern_bits}}; |
| 184 | + auto filter_c = filter_type{static_cast<size_t>(capacity) * 2, {}, {pattern_bits}}; |
| 185 | + |
| 186 | + test_merge_intersect(filter_a, filter_b, filter_c, capacity); |
| 187 | +} |
| 188 | + |
| 189 | +TEMPLATE_TEST_CASE_SIG("bloom_filter merge and intersect arrow tests", |
| 190 | + "", |
| 191 | + ((class Key, class Policy), Key, Policy), |
| 192 | + (int32_t, cuco::arrow_filter_policy<int32_t>), |
| 193 | + (int64_t, cuco::arrow_filter_policy<int64_t>), |
| 194 | + (float, cuco::arrow_filter_policy<float>)) |
| 195 | +{ |
| 196 | + using filter_type = |
| 197 | + cuco::bloom_filter<Key, cuco::extent<size_t>, cuda::thread_scope_device, Policy>; |
| 198 | + constexpr size_type capacity{1000}; // Must match capacity used in helper logic |
| 199 | + |
| 200 | + auto filter_a = filter_type{capacity}; |
| 201 | + auto filter_b = filter_type{capacity}; |
| 202 | + auto filter_c = filter_type{static_cast<size_t>(capacity) * 2}; |
| 203 | + |
| 204 | + test_merge_intersect(filter_a, filter_b, filter_c, capacity); |
| 205 | +} |
0 commit comments