Austin Schuh | 36244a1 | 2019-09-21 17:52:38 -0700 | [diff] [blame^] | 1 | // Copyright 2017 The Abseil Authors. |
| 2 | // |
| 3 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | // you may not use this file except in compliance with the License. |
| 5 | // You may obtain a copy of the License at |
| 6 | // |
| 7 | // https://www.apache.org/licenses/LICENSE-2.0 |
| 8 | // |
| 9 | // Unless required by applicable law or agreed to in writing, software |
| 10 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | // See the License for the specific language governing permissions and |
| 13 | // limitations under the License. |
| 14 | |
| 15 | #include "absl/numeric/int128.h" |
| 16 | |
| 17 | #include <algorithm> |
| 18 | #include <cstdint> |
| 19 | #include <random> |
| 20 | #include <vector> |
| 21 | |
| 22 | #include "benchmark/benchmark.h" |
| 23 | #include "absl/base/config.h" |
| 24 | |
| 25 | namespace { |
| 26 | |
| 27 | constexpr size_t kSampleSize = 1000000; |
| 28 | |
| 29 | std::mt19937 MakeRandomEngine() { |
| 30 | std::random_device r; |
| 31 | std::seed_seq seed({r(), r(), r(), r(), r(), r(), r(), r()}); |
| 32 | return std::mt19937(seed); |
| 33 | } |
| 34 | |
| 35 | std::vector<std::pair<absl::uint128, absl::uint128>> |
| 36 | GetRandomClass128SampleUniformDivisor() { |
| 37 | std::vector<std::pair<absl::uint128, absl::uint128>> values; |
| 38 | std::mt19937 random = MakeRandomEngine(); |
| 39 | std::uniform_int_distribution<uint64_t> uniform_uint64; |
| 40 | values.reserve(kSampleSize); |
| 41 | for (size_t i = 0; i < kSampleSize; ++i) { |
| 42 | absl::uint128 a = |
| 43 | absl::MakeUint128(uniform_uint64(random), uniform_uint64(random)); |
| 44 | absl::uint128 b = |
| 45 | absl::MakeUint128(uniform_uint64(random), uniform_uint64(random)); |
| 46 | values.emplace_back(std::max(a, b), |
| 47 | std::max(absl::uint128(2), std::min(a, b))); |
| 48 | } |
| 49 | return values; |
| 50 | } |
| 51 | |
| 52 | void BM_DivideClass128UniformDivisor(benchmark::State& state) { |
| 53 | auto values = GetRandomClass128SampleUniformDivisor(); |
| 54 | while (state.KeepRunningBatch(values.size())) { |
| 55 | for (const auto& pair : values) { |
| 56 | benchmark::DoNotOptimize(pair.first / pair.second); |
| 57 | } |
| 58 | } |
| 59 | } |
| 60 | BENCHMARK(BM_DivideClass128UniformDivisor); |
| 61 | |
| 62 | std::vector<std::pair<absl::uint128, uint64_t>> |
| 63 | GetRandomClass128SampleSmallDivisor() { |
| 64 | std::vector<std::pair<absl::uint128, uint64_t>> values; |
| 65 | std::mt19937 random = MakeRandomEngine(); |
| 66 | std::uniform_int_distribution<uint64_t> uniform_uint64; |
| 67 | values.reserve(kSampleSize); |
| 68 | for (size_t i = 0; i < kSampleSize; ++i) { |
| 69 | absl::uint128 a = |
| 70 | absl::MakeUint128(uniform_uint64(random), uniform_uint64(random)); |
| 71 | uint64_t b = std::max(uint64_t{2}, uniform_uint64(random)); |
| 72 | values.emplace_back(std::max(a, absl::uint128(b)), b); |
| 73 | } |
| 74 | return values; |
| 75 | } |
| 76 | |
| 77 | void BM_DivideClass128SmallDivisor(benchmark::State& state) { |
| 78 | auto values = GetRandomClass128SampleSmallDivisor(); |
| 79 | while (state.KeepRunningBatch(values.size())) { |
| 80 | for (const auto& pair : values) { |
| 81 | benchmark::DoNotOptimize(pair.first / pair.second); |
| 82 | } |
| 83 | } |
| 84 | } |
| 85 | BENCHMARK(BM_DivideClass128SmallDivisor); |
| 86 | |
| 87 | std::vector<std::pair<absl::uint128, absl::uint128>> GetRandomClass128Sample() { |
| 88 | std::vector<std::pair<absl::uint128, absl::uint128>> values; |
| 89 | std::mt19937 random = MakeRandomEngine(); |
| 90 | std::uniform_int_distribution<uint64_t> uniform_uint64; |
| 91 | values.reserve(kSampleSize); |
| 92 | for (size_t i = 0; i < kSampleSize; ++i) { |
| 93 | values.emplace_back( |
| 94 | absl::MakeUint128(uniform_uint64(random), uniform_uint64(random)), |
| 95 | absl::MakeUint128(uniform_uint64(random), uniform_uint64(random))); |
| 96 | } |
| 97 | return values; |
| 98 | } |
| 99 | |
| 100 | void BM_MultiplyClass128(benchmark::State& state) { |
| 101 | auto values = GetRandomClass128Sample(); |
| 102 | while (state.KeepRunningBatch(values.size())) { |
| 103 | for (const auto& pair : values) { |
| 104 | benchmark::DoNotOptimize(pair.first * pair.second); |
| 105 | } |
| 106 | } |
| 107 | } |
| 108 | BENCHMARK(BM_MultiplyClass128); |
| 109 | |
| 110 | void BM_AddClass128(benchmark::State& state) { |
| 111 | auto values = GetRandomClass128Sample(); |
| 112 | while (state.KeepRunningBatch(values.size())) { |
| 113 | for (const auto& pair : values) { |
| 114 | benchmark::DoNotOptimize(pair.first + pair.second); |
| 115 | } |
| 116 | } |
| 117 | } |
| 118 | BENCHMARK(BM_AddClass128); |
| 119 | |
| 120 | #ifdef ABSL_HAVE_INTRINSIC_INT128 |
| 121 | |
| 122 | // Some implementations of <random> do not support __int128 when it is |
| 123 | // available, so we make our own uniform_int_distribution-like type. |
| 124 | class UniformIntDistribution128 { |
| 125 | public: |
| 126 | // NOLINTNEXTLINE: mimicking std::uniform_int_distribution API |
| 127 | unsigned __int128 operator()(std::mt19937& generator) { |
| 128 | return (static_cast<unsigned __int128>(dist64_(generator)) << 64) | |
| 129 | dist64_(generator); |
| 130 | } |
| 131 | |
| 132 | private: |
| 133 | std::uniform_int_distribution<uint64_t> dist64_; |
| 134 | }; |
| 135 | |
| 136 | std::vector<std::pair<unsigned __int128, unsigned __int128>> |
| 137 | GetRandomIntrinsic128SampleUniformDivisor() { |
| 138 | std::vector<std::pair<unsigned __int128, unsigned __int128>> values; |
| 139 | std::mt19937 random = MakeRandomEngine(); |
| 140 | UniformIntDistribution128 uniform_uint128; |
| 141 | values.reserve(kSampleSize); |
| 142 | for (size_t i = 0; i < kSampleSize; ++i) { |
| 143 | unsigned __int128 a = uniform_uint128(random); |
| 144 | unsigned __int128 b = uniform_uint128(random); |
| 145 | values.emplace_back( |
| 146 | std::max(a, b), |
| 147 | std::max(static_cast<unsigned __int128>(2), std::min(a, b))); |
| 148 | } |
| 149 | return values; |
| 150 | } |
| 151 | |
| 152 | void BM_DivideIntrinsic128UniformDivisor(benchmark::State& state) { |
| 153 | auto values = GetRandomIntrinsic128SampleUniformDivisor(); |
| 154 | while (state.KeepRunningBatch(values.size())) { |
| 155 | for (const auto& pair : values) { |
| 156 | benchmark::DoNotOptimize(pair.first / pair.second); |
| 157 | } |
| 158 | } |
| 159 | } |
| 160 | BENCHMARK(BM_DivideIntrinsic128UniformDivisor); |
| 161 | |
| 162 | std::vector<std::pair<unsigned __int128, uint64_t>> |
| 163 | GetRandomIntrinsic128SampleSmallDivisor() { |
| 164 | std::vector<std::pair<unsigned __int128, uint64_t>> values; |
| 165 | std::mt19937 random = MakeRandomEngine(); |
| 166 | UniformIntDistribution128 uniform_uint128; |
| 167 | std::uniform_int_distribution<uint64_t> uniform_uint64; |
| 168 | values.reserve(kSampleSize); |
| 169 | for (size_t i = 0; i < kSampleSize; ++i) { |
| 170 | unsigned __int128 a = uniform_uint128(random); |
| 171 | uint64_t b = std::max(uint64_t{2}, uniform_uint64(random)); |
| 172 | values.emplace_back(std::max(a, static_cast<unsigned __int128>(b)), b); |
| 173 | } |
| 174 | return values; |
| 175 | } |
| 176 | |
| 177 | void BM_DivideIntrinsic128SmallDivisor(benchmark::State& state) { |
| 178 | auto values = GetRandomIntrinsic128SampleSmallDivisor(); |
| 179 | while (state.KeepRunningBatch(values.size())) { |
| 180 | for (const auto& pair : values) { |
| 181 | benchmark::DoNotOptimize(pair.first / pair.second); |
| 182 | } |
| 183 | } |
| 184 | } |
| 185 | BENCHMARK(BM_DivideIntrinsic128SmallDivisor); |
| 186 | |
| 187 | std::vector<std::pair<unsigned __int128, unsigned __int128>> |
| 188 | GetRandomIntrinsic128Sample() { |
| 189 | std::vector<std::pair<unsigned __int128, unsigned __int128>> values; |
| 190 | std::mt19937 random = MakeRandomEngine(); |
| 191 | UniformIntDistribution128 uniform_uint128; |
| 192 | values.reserve(kSampleSize); |
| 193 | for (size_t i = 0; i < kSampleSize; ++i) { |
| 194 | values.emplace_back(uniform_uint128(random), uniform_uint128(random)); |
| 195 | } |
| 196 | return values; |
| 197 | } |
| 198 | |
| 199 | void BM_MultiplyIntrinsic128(benchmark::State& state) { |
| 200 | auto values = GetRandomIntrinsic128Sample(); |
| 201 | while (state.KeepRunningBatch(values.size())) { |
| 202 | for (const auto& pair : values) { |
| 203 | benchmark::DoNotOptimize(pair.first * pair.second); |
| 204 | } |
| 205 | } |
| 206 | } |
| 207 | BENCHMARK(BM_MultiplyIntrinsic128); |
| 208 | |
| 209 | void BM_AddIntrinsic128(benchmark::State& state) { |
| 210 | auto values = GetRandomIntrinsic128Sample(); |
| 211 | while (state.KeepRunningBatch(values.size())) { |
| 212 | for (const auto& pair : values) { |
| 213 | benchmark::DoNotOptimize(pair.first + pair.second); |
| 214 | } |
| 215 | } |
| 216 | } |
| 217 | BENCHMARK(BM_AddIntrinsic128); |
| 218 | |
| 219 | #endif // ABSL_HAVE_INTRINSIC_INT128 |
| 220 | |
| 221 | } // namespace |