https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50160
--- Comment #46 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Using this google/benchmark test:
#include <vector>
#include "benchmark/benchmark.h"
void BM_cmp(benchmark::State& state) {
std::vector<bool> v1(250);
std::vector<bool> v2(250);
v2.back() = true;
for (auto _ : state) {
benchmark::DoNotOptimize(v1 < v2);
}
}
BENCHMARK(BM_cmp);
BENCHMARK_MAIN();
The comparison is four times faster when compiled with -DFIX to enable the new
code.
