2 * Copyright 2015 Facebook, Inc.
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
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
17 #include <glog/logging.h>
21 #include <folly/Benchmark.h>
22 #include <folly/experimental/Bits.h>
24 std::random_device rd;
26 const size_t kBufferSize = 1 << 10;
27 std::vector<uint8_t> buffer(kBufferSize + 16);
30 void benchmarkSet(size_t n, T) {
31 size_t size = sizeof(T) * 6.9; // use 6.9 bits/byte
35 std::mt19937 gen(rd());
37 if (std::is_signed<T>::value) {
38 max = (T(1) << (size - 1)) - 1;
39 min = -(T(1) << (size - 1));
41 max = (T(1) << size) - 1;
44 CHECK_LE(folly::findLastSet(max), size);
45 CHECK_LE(folly::findLastSet(-min), size);
46 std::uniform_int_distribution<T> dis(min, max);
47 for (size_t i = 0; i < k; ++i) {
52 for (size_t i = 0; i < n; ++i) {
53 size_t bit = (i * 2973) % (kBufferSize * 8);
54 size_t drop = i % size;
55 folly::Bits<T>::set(reinterpret_cast<T *>(buffer.data()),
56 bit, size - drop, values[i % k] >> drop);
59 folly::doNotOptimizeAway(
60 folly::Bits<T>::test(reinterpret_cast<T *>(buffer.data()), 512));
63 BENCHMARK_NAMED_PARAM(benchmarkSet, u16, uint16_t());
64 BENCHMARK_RELATIVE_NAMED_PARAM(benchmarkSet, i16, int16_t());
65 BENCHMARK_NAMED_PARAM(benchmarkSet, u32, uint32_t());
66 BENCHMARK_RELATIVE_NAMED_PARAM(benchmarkSet, i32, int32_t());
67 BENCHMARK_NAMED_PARAM(benchmarkSet, u64, uint64_t());
68 BENCHMARK_RELATIVE_NAMED_PARAM(benchmarkSet, i64, int64_t());
70 BENCHMARK_DRAW_LINE();
72 std::atomic<int64_t> sum(0);
75 void benchmarkGet(size_t n, T x) {
76 size_t size = sizeof(T) * 6.9; // use 6.9 bits/byte
77 for (size_t i = 0; i < n; ++i) {
78 size_t bit = (i * 2973) % (kBufferSize * 8);
79 size_t drop = i % size;
80 x += folly::Bits<T>::get(
81 reinterpret_cast<T *>(buffer.data()), bit, size - drop);
83 folly::doNotOptimizeAway(x);
86 BENCHMARK_NAMED_PARAM(benchmarkGet, u16, uint16_t(0));
87 BENCHMARK_RELATIVE_NAMED_PARAM(benchmarkGet, i16, int16_t(0));
88 BENCHMARK_NAMED_PARAM(benchmarkGet, u32, uint32_t(0));
89 BENCHMARK_RELATIVE_NAMED_PARAM(benchmarkGet, i32, int32_t(0));
90 BENCHMARK_NAMED_PARAM(benchmarkGet, u64, uint64_t(0));
91 BENCHMARK_RELATIVE_NAMED_PARAM(benchmarkGet, i64, int64_t(0));
94 ============================================================================
95 folly/experimental/test/BitsBenchmark.cpp relative time/iter iters/s
96 ============================================================================
97 benchmarkSet(u16) 8.58ns 116.59M
98 benchmarkSet(i16) 88.42% 9.70ns 103.08M
99 benchmarkSet(u32) 8.37ns 119.45M
100 benchmarkSet(i32) 88.23% 9.49ns 105.39M
101 benchmarkSet(u64) 9.23ns 108.34M
102 benchmarkSet(i64) 82.77% 11.15ns 89.68M
103 ----------------------------------------------------------------------------
104 benchmarkGet(u16) 6.32ns 158.13M
105 benchmarkGet(i16) 80.40% 7.87ns 127.14M
106 benchmarkGet(u32) 6.34ns 157.65M
107 benchmarkGet(i32) 84.61% 7.50ns 133.39M
108 benchmarkGet(u64) 7.32ns 136.58M
109 benchmarkGet(i64) 85.78% 8.53ns 117.16M
110 ============================================================================
113 int main(int argc, char *argv[]) {
114 gflags::ParseCommandLineFlags(&argc, &argv, true);
115 folly::runBenchmarks();