use memchr for searching for '%' in printf format string

[rimathia]

I agree that my contributions are licensed under the {fmt} license, and agree to future changes to the licensing.

This is the same change as in the closed pull request #1853 but this time with a simple microbenchmark to argue for it. I hope this isn't considered impolite.

The change starts to be noticeable at a format string length of about 100 characters.

The microbenchmark is the same as in #1982:

#include <benchmark/benchmark.h>
#include <fmt/printf.h>

void long_argument(benchmark::State& state) {
  std::string long_string(state.range(0), 'A');
  for (auto _ : state) {
    benchmark::DoNotOptimize(fmt::sprintf("%s", long_string));
  };
}
BENCHMARK(long_argument)->Arg(10)->Arg(100)->Arg(1000)->Arg(10000);

void long_format_string(benchmark::State& state) {
  std::string format_string = std::string(state.range(0), 'A') + "%s";
  std::string message = "Hi";
  for (auto _ : state) {
    benchmark::DoNotOptimize(fmt::sprintf(format_string, message));
  };
}
BENCHMARK(long_format_string)->Arg(10)->Arg(100)->Arg(1000)->Arg(10000);

BENCHMARK_MAIN();

The microbenchmark numbers:

master (038057e)

-------------------------------------------------------------------
Benchmark                         Time             CPU   Iterations
-------------------------------------------------------------------
long_argument/10               36.6 ns         36.6 ns     17808714
long_argument/100               188 ns          188 ns      3716840
long_argument/1000             1284 ns         1284 ns       538288
long_argument/10000           11769 ns        11767 ns        62667
long_format_string/10          39.6 ns         39.6 ns     17552922
long_format_string/100          237 ns          237 ns      2934014
long_format_string/1000        1882 ns         1882 ns       383536
long_format_string/10000      18333 ns        18328 ns        39354

branch (e662999)

-------------------------------------------------------------------
Benchmark                         Time             CPU   Iterations
-------------------------------------------------------------------
long_argument/10               39.7 ns         39.7 ns     16266435
long_argument/100               191 ns          191 ns      3609438
long_argument/1000             1304 ns         1304 ns       535373
long_argument/10000           11039 ns        11039 ns        62367
long_format_string/10          39.1 ns         39.1 ns     17829580
long_format_string/100          191 ns          191 ns      3675833
long_format_string/1000        1383 ns         1383 ns       493149
long_format_string/10000      12418 ns        12418 ns        55520

[vitaut]

Thanks for the PR. Please fix formatting in the PR description and post benchmark results as text, not as image.

[rimathia]

I have clicked on 'resolve comment' in the hope that this would make the red 'Changes requested' go away, I wasn't completely sure whether this is something I should do or you as the reviewer should do.