summaryrefslogtreecommitdiffstats
path: root/libcxx/benchmarks/algorithms.bench.cpp
diff options
context:
space:
mode:
authorEric Fiselier <eric@efcs.ca>2018-10-29 19:25:02 +0000
committerEric Fiselier <eric@efcs.ca>2018-10-29 19:25:02 +0000
commit8c40d81d4f019b8b1ae02c154be657b949c2cf4d (patch)
treeb472b1e17c4943123544714cf9fea379064e9b2f /libcxx/benchmarks/algorithms.bench.cpp
parentdd4be53b20a8e3ad43ed5b4e14f6c93d1a23ae34 (diff)
downloadbcm5719-llvm-8c40d81d4f019b8b1ae02c154be657b949c2cf4d.tar.gz
bcm5719-llvm-8c40d81d4f019b8b1ae02c154be657b949c2cf4d.zip
Bug 39129: Speeding up partition_point/lower_bound/upper_bound/ by using unsigned division by 2 when possible.
Patch by Denis Yaroshevskiy (denis.yaroshevskij@gmail.com) The rational and measurements can be found in the bug description: https://bugs.llvm.org/show_bug.cgi?id=39129 Reviewed as https://reviews.llvm.org/D52697 llvm-svn: 345525
Diffstat (limited to 'libcxx/benchmarks/algorithms.bench.cpp')
-rw-r--r--libcxx/benchmarks/algorithms.bench.cpp64
1 files changed, 64 insertions, 0 deletions
diff --git a/libcxx/benchmarks/algorithms.bench.cpp b/libcxx/benchmarks/algorithms.bench.cpp
index 86315390e0d..ab0e81b0cac 100644
--- a/libcxx/benchmarks/algorithms.bench.cpp
+++ b/libcxx/benchmarks/algorithms.bench.cpp
@@ -58,5 +58,69 @@ BENCHMARK_CAPTURE(BM_Sort, sorted_descending_strings,
BENCHMARK_CAPTURE(BM_Sort, single_element_strings,
getDuplicateStringInputs)->Arg(TestNumInputs);
+template <typename GenInputs, typename Alg>
+void do_binary_search_benchmark(benchmark::State& st, GenInputs gen, Alg alg)
+{
+ using ValueType = typename decltype(gen(0))::value_type;
+ auto in = gen(st.range(0));
+ std::sort(in.begin(), in.end());
+
+ const auto every_10_percentile = [&]() -> std::vector<ValueType*> {
+ size_t step = in.size() / 10;
+
+ if (step == 0) {
+ st.SkipWithError("Input doesn't contain enough elements");
+ return {};
+ }
+
+ std::vector<ValueType*> res;
+ for (size_t i = 0; i < in.size(); i += step)
+ res.push_back(&in[i]);
+
+ return res;
+ }();
+
+ for (auto _ : st)
+ {
+ for (auto* test : every_10_percentile)
+ benchmark::DoNotOptimize(alg(in.begin(), in.end(), *test));
+ }
+}
+
+template <typename GenInputs>
+void BM_LowerBound(benchmark::State& st, GenInputs gen)
+{
+ do_binary_search_benchmark(st, gen, [](auto f, auto l, const auto& v) {
+ return std::lower_bound(f, l, v);
+ });
+}
+
+BENCHMARK_CAPTURE(BM_LowerBound, random_int32, getRandomIntegerInputs<int32_t>)
+ ->Arg(TestNumInputs) // Small int32_t vector
+ ->Arg(TestNumInputs * TestNumInputs); // Big int32_t vector
+
+BENCHMARK_CAPTURE(BM_LowerBound, random_int64, getRandomIntegerInputs<int64_t>)
+ ->Arg(TestNumInputs); // Small int64_t vector. Should also represent pointers.
+
+BENCHMARK_CAPTURE(BM_LowerBound, random_strings, getRandomStringInputs)
+ ->Arg(TestNumInputs); // Small string vector. What happens if the comparison is not very cheap.
+
+template <typename GenInputs>
+void BM_EqualRange(benchmark::State& st, GenInputs gen)
+{
+ do_binary_search_benchmark(st, gen, [](auto f, auto l, const auto& v) {
+ return std::equal_range(f, l, v);
+ });
+}
+
+BENCHMARK_CAPTURE(BM_EqualRange, random_int32, getRandomIntegerInputs<int32_t>)
+ ->Arg(TestNumInputs) // Small int32_t vector
+ ->Arg(TestNumInputs * TestNumInputs); // Big int32_t vector
+
+BENCHMARK_CAPTURE(BM_EqualRange, random_int64, getRandomIntegerInputs<int64_t>)
+ ->Arg(TestNumInputs); // Small int64_t vector. Should also represent pointers.
+
+BENCHMARK_CAPTURE(BM_EqualRange, random_strings, getRandomStringInputs)
+ ->Arg(TestNumInputs); // Small string vector. What happens if the comparison is not very cheap.
BENCHMARK_MAIN();
OpenPOWER on IntegriCloud