parse-integers-benchmark

Benchmarking reading and parsing integers from a file in C++ using one thread only.

The methods benchmarked are:

• getchar-inline - call std::getchar and parse each int char-by-char.
• scanf - call std::scanf to parse each int.
• scanf-multi - call std::scanf to parse 64 ints in one call.
• iostream - call std::istream::operator>>(int&) to parse each int. No multi version is possible.
• mmap-inline - mmap the entire file into memory and parse each int char-by-char.
• mmap-charconv - mmap the entire file into memory and parse each int using std::from_chars.

Building and running

git clone https://github.com/max0x7ba/parse-integers-benchmark.git
cd parse-integers-benchmark
make -r run_parse_integers_fast

The benchmark compiles in C++17 mode by default. You can search-replace gnu++17 to gnu++11 in Makefile to make it work with C++11.

Example outputs

Ubuntu 18.04.4 LTS, g++-8.3.0, Intel Core i9-9900KS CPU @ 5.1GHz, C++17 charconv.

---- Best times ----
seconds,    percent, method
0.123534698,  100.0, scanf
0.121618955,   98.4, iostream
0.104887812,   84.9, scanf-multi
0.047167165,   38.2, getchar-inline
0.030817239,   24.9, mmap-charconv
0.025470340,   20.6, mmap-inline

Ubuntu 18.04.2 LTS, g++-8.2.0, Intel Core i7-7700K CPU @ 5GHz

---- Best times ----
seconds,    percent, method
0.133155952,  100.0, iostream
0.102128208,   76.7, scanf
0.082469185,   61.9, scanf-multi
0.048661004,   36.5, getchar-inline
0.025320109,   19.0, mmap-inline

CentOS release 6.10, g++-6.3.0, Intel Core i7-4790 CPU @ 3.6GHz

---- Best times ----
seconds,    percent, method
0.167985515,  100.0, getchar-inline
0.147258495,   87.7, scanf
0.137161991,   81.7, iostream
0.118859546,   70.8, scanf-multi
0.034033769,   20.3, mmap-inline

Notes

• mmap is a relatively expensive operation so that it may only be faster for relatively large files. Benchmark on your file size.
• getchar-inline and mmap-inline use a similar inline method of parsing an int from a stream of char, so that the time difference can be attributed to the latency of accessing the next char. mmap method maps the entire file into memory and reads it sequentially, which is the best case scenario for the CPU memory prefetcher.