| 1 | //===----------------------------------------------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | // UNSUPPORTED: c++03, c++11, c++14, c++17 |
| 10 | |
| 11 | #include <algorithm> |
| 12 | #include <cstdlib> |
| 13 | #include <iterator> |
| 14 | #include <set> |
| 15 | #include <vector> |
| 16 | |
| 17 | #include "common.h" |
| 18 | #include "test_iterators.h" |
| 19 | |
| 20 | namespace { |
| 21 | |
| 22 | // types of containers we'll want to test, covering interesting iterator types |
| 23 | struct VectorContainer { |
| 24 | template <typename... Args> |
| 25 | using type = std::vector<Args...>; |
| 26 | |
| 27 | static constexpr const char* Name = "Vector" ; |
| 28 | }; |
| 29 | |
| 30 | struct SetContainer { |
| 31 | template <typename... Args> |
| 32 | using type = std::set<Args...>; |
| 33 | |
| 34 | static constexpr const char* Name = "Set" ; |
| 35 | }; |
| 36 | |
| 37 | using AllContainerTypes = std::tuple<VectorContainer, SetContainer>; |
| 38 | |
| 39 | // set_intersection performance may depend on where matching values lie |
| 40 | enum class OverlapPosition { |
| 41 | None, |
| 42 | Front, |
| 43 | // performance-wise, matches at the back are identical to ones at the front |
| 44 | Interlaced, |
| 45 | }; |
| 46 | |
| 47 | struct AllOverlapPositions : EnumValuesAsTuple<AllOverlapPositions, OverlapPosition, 3> { |
| 48 | static constexpr const char* Names[] = {"None" , "Front" , "Interlaced" }; |
| 49 | }; |
| 50 | |
| 51 | // forward_iterator wrapping which, for each increment, moves the underlying iterator forward Stride elements |
| 52 | template <typename Wrapped> |
| 53 | struct StridedFwdIt { |
| 54 | Wrapped base_; |
| 55 | unsigned stride_; |
| 56 | |
| 57 | using iterator_category = std::forward_iterator_tag; |
| 58 | using difference_type = typename Wrapped::difference_type; |
| 59 | using value_type = typename Wrapped::value_type; |
| 60 | using pointer = typename Wrapped::pointer; |
| 61 | using reference = typename Wrapped::reference; |
| 62 | |
| 63 | StridedFwdIt(Wrapped base, unsigned stride) : base_(base), stride_(stride) { assert(stride_ != 0); } |
| 64 | |
| 65 | StridedFwdIt operator++() { |
| 66 | for (unsigned i = 0; i < stride_; ++i) |
| 67 | ++base_; |
| 68 | return *this; |
| 69 | } |
| 70 | StridedFwdIt operator++(int) { |
| 71 | auto tmp = *this; |
| 72 | ++*this; |
| 73 | return tmp; |
| 74 | } |
| 75 | value_type& operator*() { return *base_; } |
| 76 | const value_type& operator*() const { return *base_; } |
| 77 | value_type& operator->() { return *base_; } |
| 78 | const value_type& operator->() const { return *base_; } |
| 79 | bool operator==(const StridedFwdIt& o) const { return base_ == o.base_; } |
| 80 | bool operator!=(const StridedFwdIt& o) const { return !operator==(o); } |
| 81 | }; |
| 82 | template <typename Wrapped> |
| 83 | StridedFwdIt(Wrapped, unsigned) -> StridedFwdIt<Wrapped>; |
| 84 | |
| 85 | template <typename T> |
| 86 | std::vector<T> getVectorOfRandom(size_t N) { |
| 87 | std::vector<T> v; |
| 88 | fillValues(v, N, Order::Random); |
| 89 | sortValues(v, Order::Random); |
| 90 | return std::vector<T>(v); |
| 91 | } |
| 92 | |
| 93 | // Realistically, data won't all be nicely contiguous in a container, |
| 94 | // we'll go through some effort to ensure that it's shuffled through memory |
| 95 | // this is especially important for containers with non-contiguous element |
| 96 | // storage, but it will affect even a std::vector, because when you copy a |
| 97 | // std::vector<std::string> the underlying data storage position for the char |
| 98 | // arrays of the copy are likely to have high locality |
| 99 | template <class Container> |
| 100 | std::pair<Container, Container> genCacheUnfriendlyData(size_t size1, size_t size2, OverlapPosition pos) { |
| 101 | using ValueType = typename Container::value_type; |
| 102 | auto move_into = [](auto first, auto last) { |
| 103 | Container out; |
| 104 | std::move(first, last, std::inserter(out, out.begin())); |
| 105 | return out; |
| 106 | }; |
| 107 | const auto src_size = pos == OverlapPosition::None ? size1 + size2 : std::max(size1, size2); |
| 108 | std::vector<ValueType> src = getVectorOfRandom<ValueType>(src_size); |
| 109 | |
| 110 | if (pos == OverlapPosition::None) { |
| 111 | std::sort(src.begin(), src.end()); |
| 112 | return std::make_pair(move_into(src.begin(), src.begin() + size1), move_into(src.begin() + size1, src.end())); |
| 113 | } |
| 114 | |
| 115 | // All other overlap types will have to copy some part of the data, but if |
| 116 | // we copy after sorting it will likely have high locality, so we sort |
| 117 | // each copy separately |
| 118 | auto copy = src; |
| 119 | std::sort(src.begin(), src.end()); |
| 120 | std::sort(copy.begin(), copy.end()); |
| 121 | |
| 122 | switch (pos) { |
| 123 | case OverlapPosition::None: |
| 124 | // we like -Wswitch :) |
| 125 | break; |
| 126 | |
| 127 | case OverlapPosition::Front: |
| 128 | return std::make_pair(move_into(src.begin(), src.begin() + size1), move_into(copy.begin(), copy.begin() + size2)); |
| 129 | |
| 130 | case OverlapPosition::Interlaced: |
| 131 | const auto stride1 = size1 < size2 ? size2 / size1 : 1; |
| 132 | const auto stride2 = size2 < size1 ? size1 / size2 : 1; |
| 133 | return std::make_pair(move_into(StridedFwdIt(src.begin(), stride1), StridedFwdIt(src.end(), stride1)), |
| 134 | move_into(StridedFwdIt(copy.begin(), stride2), StridedFwdIt(copy.end(), stride2))); |
| 135 | } |
| 136 | std::abort(); // would be std::unreachable() if it could |
| 137 | return std::pair<Container, Container>(); |
| 138 | } |
| 139 | |
| 140 | template <class ValueType, class Container, class Overlap> |
| 141 | struct SetIntersection { |
| 142 | using ContainerType = typename Container::template type<Value<ValueType>>; |
| 143 | size_t size1_; |
| 144 | size_t size2_; |
| 145 | |
| 146 | SetIntersection(size_t size1, size_t size2) : size1_(size1), size2_(size2) {} |
| 147 | |
| 148 | bool skip() const noexcept { |
| 149 | // let's save some time and skip simmetrical runs |
| 150 | return size1_ < size2_; |
| 151 | } |
| 152 | |
| 153 | void run(benchmark::State& state) const { |
| 154 | auto input = genCacheUnfriendlyData<ContainerType>(size1_, size2_, Overlap()); |
| 155 | std::vector<Value<ValueType>> out(std::min(size1_, size2_)); |
| 156 | |
| 157 | const auto BATCH_SIZE = std::max(size_t{512}, (2 * TestSetElements) / (size1_ + size2_)); |
| 158 | for (const auto& _ : state) { |
| 159 | while (state.KeepRunningBatch(n: BATCH_SIZE)) { |
| 160 | for (unsigned i = 0; i < BATCH_SIZE; ++i) { |
| 161 | const auto& [c1, c2] = input; |
| 162 | auto res = std::set_intersection(c1.begin(), c1.end(), c2.begin(), c2.end(), out.begin()); |
| 163 | benchmark::DoNotOptimize(res); |
| 164 | } |
| 165 | } |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | std::string name() const { |
| 170 | return std::string("SetIntersection" ) + Overlap::name() + '_' + Container::Name + ValueType::name() + '_' + |
| 171 | std::to_string(val: size1_) + '_' + std::to_string(val: size2_); |
| 172 | } |
| 173 | }; |
| 174 | |
| 175 | } // namespace |
| 176 | |
| 177 | int main(int argc, char** argv) { /**/ |
| 178 | benchmark::Initialize(argc: &argc, argv); |
| 179 | if (benchmark::ReportUnrecognizedArguments(argc, argv)) |
| 180 | return 1; |
| 181 | |
| 182 | makeCartesianProductBenchmark<SetIntersection, AllValueTypes, AllContainerTypes, AllOverlapPositions>( |
| 183 | A: Quantities, A: Quantities); |
| 184 | benchmark::RunSpecifiedBenchmarks(); |
| 185 | return 0; |
| 186 | } |
| 187 | |