Timing statistics are incorrect about celero

[DigitalInBlue]

Thank you for the issue. I have a fix for this along with a few other notes on timing that will be pushed up soon.

[bkloppenborg]

While making improvements to timing, it would be useful if we could include additional information in the report. For example inclusion of the total number of operations computed by the function being profiled. For example I have several convolution benchmarks with 2x2, 4x4, and 8x8 kernels that operate on 2D arrays of size NxN (with N coming from getExperimentValues).

[DigitalInBlue]

One thing you can do is to write your own main() function and register your own function to be called when results are collected. For instance, in order to add the j-unit output option, it was simply a matter of creating a singleton class which I wanted to collect all of the results, then register it with Celero to be called.

    // Has a JUnit output file been specified?
    argument = args.get<std::string>("junit");
    if(argument.empty() == false)
    {
        celero::JUnit::Instance().setFileName(argument);

        celero::AddExperimentResultCompleteFunction(
            [](std::shared_ptr<celero::Result> p)
            {
                celero::JUnit::Instance().add(p);
            });
    }