Francisco,
On Tue Dec 23 2014 at 3:31:32 AM Francisco José Tapia
We can do too a benchmark with variable lenght elements with strings.
If you prepare the operations for to do in that test, I can prepare an additional benchmark with all the algorithms and your operations.
After all, we see and take a decision
Yours
Francisco
The approach I've used with the boost::sort library is to set up the benchmark tests so that they read in a file of random data generated by a single application (randomgen), or any other file that is passed in for testing, and they write out their results to file. This enables better testing and debugging in case there are any problems (I directly diff the sort results when I can expect them to be identical; tune.pl does this automatically). I would prefer if you used stringsample.cpp and int64.cpp as examples of how to do this, and downloaded the boost sort library and added any modifications you need into that code. The library has a tune.pl script that runs the benchmarks. It may be helpful to write some utility templated funciton to switch between different algorithms, instead of the current approach of using a "-std" switch to use std::sort. https://github.com/spreadsort/sort There is a README there on how to install it using modular boost.