Error with number output in C#?

[AlTech]

So yeah,

 

I'm having an issue with a numerical output coming out really weirdly.

 

Does anybody know why this is happening?

 

 

Here's the code from my mathematical calculations for working out the millions of calculations per second:

 

Dividing multiTime by 1000 is to turn miliseconds into seconds.

 

the 750 Million + 750 Million is because 2 types of calculations are done and each does 750 Million calculations.

 

The multi threaded does 750 Million calcs in one thread and the other 750 Million in a second thread.

       public string multiThreadedScorePerSecond()
        {
            //Convert to calcs per second
            multiTimePerSec = (750000000 + 750000000) / (multiTime / 1000);

            //Convert to millions of calcs per second
            multiTimePerSec = multiTimePerSec / 1000;
            multiTimePerSec = multiTimePerSec / 1000;

            multiTimePerSec = multiTimePerSec / multiTime;

            return multiTimePerSec.ToString() + " Millions of calculations per second.";
        }

 

The single threaded does the first type of calc and then the second type of calc sequentially.

        public string singleThreadedScorePerSecond()
        {
            //Convert to calcs per second
            singleTimePerSec = (750000000 + 750000000) / (singleTimePerSec / 1000);

            //Convert to millions of calcs per second
            singleTimePerSec = singleTimePerSec / 1000;
            singleTimePerSec = singleTimePerSec / 1000;

            singleTimePerSec = singleTimePerSec / singleTime;

            return singleTimePerSec.ToString() + " Millions of calculations per second.";
        }

Thanks.

[alca]

Can you post the full code? Also, why are you diving multiTimePerSec & singleTimePerSec by 1000, twice?

 

multiTimePerSec = multiTimePerSec / 1000;
multiTimePerSec = multiTimePerSec / 1000;

singleTimePerSec = singleTimePerSec / 1000;
singleTimePerSec = singleTimePerSec / 1000;

[AlTech]

15 minutes ago, alca said:

Can you post the full code? Also, why are you diving multiTimePerSec & singleTimePerSec by 1000, twice?

 

multiTimePerSec = multiTimePerSec / 1000;
multiTimePerSec = multiTimePerSec / 1000;

singleTimePerSec = singleTimePerSec / 1000;
singleTimePerSec = singleTimePerSec / 1000;

I want to turn the 1.5 Billion calcs / second to turn it into Millions per second and so I figured I could do it this way......

 

The full code is too much for here so i'll save you the trouble and give you a link to it -> https://github.com/AluminiumTech/CSMark/tree/master/CSMark

[alca]

1.5 bil / 1000 / 1000 ~ 15k calcs/s

1.5 bil / 1000 ~ 1.5 mil calcs/s

 

Even then I don't think this would cause the output thingy. 

Try looking at the singleTime variable and if it's maybe causing the issue. Other than that, I'm not sure and you should try posting the question @ http://www.stackoverflow.com

[AlTech]

30 minutes ago, alca said:

1.5 bil / 1000 / 1000 ~ 15k calcs/s

1.5 bil / 1000 ~ 1.5 mil calcs/s

 

Even then I don't think this would cause the output thingy. 

Try looking at the singleTime variable and if it's maybe causing the issue. Other than that, I'm not sure and you should try posting the question @ http://www.stackoverflow.com

So I did some testing and it turned out that my Single Threaded number was being divided too many times.

 

Do you know how I could take my existing single threaded benchmark and make it multi threaded? Cos my existing multi threaded benchmark sucks.

 

[AlTech]

[Mira Yurizaki]

It's possible you're not handling the data properly in a multithreaded environment, and so "multiTimePerSec" could mean anything by the time you're poking it.

 

I would have a better idea if more of the source code was available.

[AlTech]

17 hours ago, M.Yurizaki said:

It's possible you're not handling the data properly in a multithreaded environment, and so "multiTimePerSec" could mean anything by the time you're poking it.

 

I would have a better idea if more of the source code was available.

All the source code is available at the link. I'm not posting it here because it would be obnoxious to read the entirety of it.

 

https://github.com/AluminiumTech/CSMark/tree/master/CSMark

[madknight3]

6 hours ago, AluminiumTech said:

All the source code is available at the link. I'm not posting it here because it would be obnoxious to read the entirety of it.

 

https://github.com/AluminiumTech/CSMark/tree/master/CSMark

I don't think this code will do what you want it to do

multiTimeStopwatch.Start();

//Start the threads and make the benchmarking happen!
pyThread.Start();
trigThread.Start();

multiTimeStopwatch.Stop();

That's just timing how long it takes to start up the threads. It doesn't take into account how long the threads run for. You probably want something more like these answers describe - StackOverflow: How to wait for thread to finish with .NET?

 

edit (to answer your original question): So regarding the printing issue, I think it's because multiTimeStopwatch.ElapsedMilliseconds is zero because there's not enough precision to measure something that fast. Which means you're dividing by zero later. Here's some code to explain.

class Program
{
    static void Main(string[] args)
    {
        double multiTime = 0;
        double x = 1 / multiTime;
        Console.WriteLine(x); // prints weird character
      
        // If you were using integers it would throw an exception.
        // However double has the notion of Infinity and NaN so it doesn't throw an exception
        bool b = double.IsInfinity(x); // is true
        Console.WriteLine(b);
    }
}

Side note, some programs may properly print out an infinity character. Like LINQPad did when I tested with that.

 

Another side note, if ElapsedMilliseconds isn't enough in the future you may be able to get a value from ElapsedTicks - StackOverflow: How do you convert Stopwatch ticks to nanoseconds, milliseconds and seconds? (I've never looked into the accuracy of this though)

[Mira Yurizaki]

I can't seem to find anything obviously wrong.

 

My only thought would be to use String.Format instead of ToString. Though you might want to also verify in the debugger before exiting multiThreadedScorePerSecond that the value looks more or less correct.

[AlTech]

18 hours ago, madknight3 said:

I don't think this code will do what you want it to do

multiTimeStopwatch.Start();

//Start the threads and make the benchmarking happen!
pyThread.Start();
trigThread.Start();

multiTimeStopwatch.Stop();

That's just timing how long it takes to start up the threads. It doesn't take into account how long the threads run for. You probably want something more like these answers describe - StackOverflow: How to wait for thread to finish with .NET?

 

edit (to answer your original question): So regarding the printing issue, I think it's because multiTimeStopwatch.ElapsedMilliseconds is zero because there's not enough precision to measure something that fast. Which means you're dividing by zero later. Here's some code to explain.

class Program
{
    static void Main(string[] args)
    {
        double multiTime = 0;
        double x = 1 / multiTime;
        Console.WriteLine(x); // prints weird character
      
        // If you were using integers it would throw an exception.
        // However double has the notion of Infinity and NaN so it doesn't throw an exception
        bool b = double.IsInfinity(x); // is true
        Console.WriteLine(b);
    }
}

 

Thanks. I eventually managed to fix the single threaded test. The multi threaded test has yet to be fixed.

 

I'll take a look at the Stack Overflow article but I have a feeling I'll need to implement a while loop.