In this benchmark we are trying to find the fastest implementation of log10(value | 1) for integer values.
$ mkdir build $ cd build $ cmake -DCMAKE_BUILD_TYPE=Release .. $ make
To run benchmarking script you need to install R language and ggplot2.
Make sure that you set cpu frequency to fixed value and disable any turbo mode, so the results of the benchmark should be more accurate.
And then you can run the benchmark. It will generate .png files in the current directory.
$ cd scripts $ ./benchmark.sh
Each algorithm is compiled in its own benchmarking application and testing application.
Benchmarking application support positional arguments:
- inner loop counter [default=1]
- integer size (u8, u16, u32, u64) [default=u64]
- number of digits [default=iterate through all possible values]
Testing application can be launched in two modes: simple testing and random testing. For simple testing, just launch the application, and for random testing launch application with an integer argument, it will be used as a seed for random number generator.
This script run benchmarks in two modes(cold, hot) and collect data from it.
In the cold mode, it runs benchmark application for 100 times with arguments (inner loop counter=1, integer size=<all>, digits=<all values for current integer size>).
In the hot mode, it runs benchmark application for 100 times with arguments (inner loop counter=100, integer size=<all>).
And then generates graphical representation of the collected data. As a time value it uses minimum of all time values for each (benchmark, integer size, digits).
