The stopwatch gives different results depending on where the code is located.

Question

I get some confusing Stopwatch results in my C # project. Consider the following code:

 static void Main(string[] args) { byte[] myEventArray = GetEventByteArrayFromDatabase(); byte[] myEventItemsArray = GetEventItemByteArrayFromDatabase(); uint numEvents = 1000; uint numEventItems = 1000; Stopwatch sw1 = Stopwatch.StartNew(); TestFunction(ref myEventArray, numEvents, ref myEventItemsArray, numEventItems); sw1.Stop(); float timeTakenInSeconds = (float)sw2.ElapsedTicks / Stopwatch.Frequency; Console.WriteLine("Total time: " + timeTakenInSeconds + " seconds. "); } static void TestFunction(ref byte[] EventArray, uint numEvents, ref byte[] EventItemArray, uint numEventItems) { Calculator calc = new Calculator(); calc.Test(EventArray, numEvents, EventItemArray, numEventItems); } 

I run this and get a time of about 0.2 seconds. Now consider the following:

 static void Main(string[] args) { byte[] myEventArray = GetEventByteArrayFromDatabase(); byte[] myEventItemsArray = GetEventItemByteArrayFromDatabase(); uint numEvents = 1000; uint numEventItems = 1000; Stopwatch sw1 = Stopwatch.StartNew(); Calculator calc = new Calculator(); calc.Test(myEventArray , numEvents, myEventItemsArray , numEventItems); sw1.Stop(); float timeTakenInSeconds = (float)sw1.ElapsedTicks / Stopwatch.Frequency; Console.WriteLine("Total time: " + timeTakenInSeconds + " seconds. "); } 

I run this and get a similar result, as expected. Finally check this out:

 static void Main(string[] args) { byte[] myEventArray = GetEventByteArrayFromDatabase(); byte[] myEventItemsArray = GetEventItemByteArrayFromDatabase(); uint numEvents = 1000; uint numEventItems = 1000; TestFunction(ref myEventArray, numEvents, ref myEventItemsArray, numEventItems); } static void TestFunction(ref byte[] EventArray, uint numEvents, ref byte[] EventItemArray, uint numEventItems) { Stopwatch sw1 = Stopwatch.StartNew(); Calculator calc = new Calculator(); calc.Test(EventArray, numEvents, EventItemArray, numEventItems); sw1.Stop(); float timeTakenInSeconds = (float)sw1.ElapsedTicks / Stopwatch.Frequency; Console.WriteLine("Total time: " + timeTakenInSeconds + " seconds. "); } 

When I ran this, the synchronization result will be ten times faster for some reason. Any ideas why this could be?

A bit more information: The Calculator class is defined in C ++ / CLI. I use it as a wrapper for my own C ++ code, which ultimately works with byte arrays. I also compile the "unsafe" flag. Not sure if this could have any effect. All code compiled in release mode.

Answer 1

I found the reason for this. This is because in the first case, the creation of my Calculator object was included in the synchronization result, and in the third case it is not.

If I understand this correctly, the stack variables are not actually created on the line you type "new ()", the compiler moves the memory allocation to the prolog method.

See this page: https://msdn.microsoft.com/en-us/library/tawsa7cb.aspx

"The prolog saves the values ​​of the argument registers in its home addresses, if necessary, pushes non-volatile registers onto the stack, allocates a fixed part of the stack for local and temporary files, and possibly sets a pointer to the frame."

So, my “case 1” included “new” (because it happened in the TestFunction prologue), and “case 3” excluded it because “new” had already happened.