Variable scope and performance

Question

Below is the code i am using

private void TestFunction()
{
  foreach (MySampleClass c in dictSampleClass)
  {
    String sText = c.VAR1 + c.VAR2 + c.VAR3
    PerformSomeTask(sText,c.VAR4);
  }
}

My friend has suggesed to change to (to improve performance. dictSampleClass is a dictionary. It has 10K objects)

private void TestFunction()
{
  String sText="";
  foreach (MySampleClass c in dictSampleClass)
  {
    sText = c.VAR1 + c.VAR2 + c.VAR3
    PerformSomeTask(sText,c.VAR4);
  }
}

My Question is, "Does above change improve performance? if yes, how?"

WOW thats more than expected response. Most guys said "C# compiler would take care of that". So what about c compiler??

I'd be very surprised, if this improved performance. But it hurts readability. — Henrik, Jul 02 '10 at 08:56
There is a (negligible) performance HIT as you assign the empty string only to reassign immediately within the foreach. — Hans Keﬆing, Jul 02 '10 at 08:59
You can avoid to set the sText to empty, since it is initialized anyway at every iteration. As others said, the compiler will produce the same code, but I find that declaring the variable outside the loop is cleaner. That said, note that concatenating 3 strings that way, produces 2 more strings than expected (tmp1=c.Var1+c.Var2 and tmp2=tmp1+c.Var3), so it could be worth (for the sake of using as few memory as needed) using a StringBuilder whose size is pre-initialized to (c.VAR1.Length+c.VAR2.Length+c.VAR3.Length). — BladeWise, Jul 02 '10 at 09:50
Sorry, but why would the C compiler have anything to do with it? — Adam Houldsworth, Jul 02 '10 at 10:20
@Nathan Baulch: Thanks for pointing out. I am sorry, i could not find that before asking :( — Manjoor, Jul 02 '10 at 10:33
@Adam: Does it means any language compiler would react same ? — Manjoor, Jul 02 '10 at 10:35
@Manjoor no its specific to the implementation of the compiler; however it is very likely that good compilers for popular languages have optimisations such as these. — Adam Houldsworth, Jul 02 '10 at 10:52
You can answer this question yourself. **Try it both ways and see if the performance is different.** If you can't tell a difference, then *performance is not different*. The whole point of better performance is that it is observably better. — Eric Lippert, Jul 02 '10 at 14:20

Adam Houldsworth · Accepted Answer · 2010-07-02T12:15:58.837

The compiler has intelligence to move variable declarations into/out of loops where required. In your example however, you are using a string, which is immutable. By declaring it outside I believe you are trying to "create once, use many", however strings are created each time they are modified so you can't achieve that.

Not to sound harse, but this is a premature optimisation, and likely a failing one - due to string immutability.

If the collection is large, go down the route of appending many strings into a StringBuilder - separated by a delimiter. Then split the string on this delimiter and iterate the array to add them, instead of concatenating them and adding them in one loop.

StringBuilder sb = new StringBuilder();

foreach (MySampleClass c in dictSampleClass)
{
    sb.Append(c.VAR1);
    sb.Append(c.VAR2);
    sb.Append(c.VAR3);
    sb.Append("|");
}

string[] results = sb.ToString().Split('|');

for (int i = 0; i < dictSampleClass.Count; i++)
{
    string s = results[i];
    MySampleClass c = dictSampleClass[i];
    PerformSomeTask(s,c.VAR4); 
}

I infer no benefits to using this code - most likely doesn't even work!

UPDATE: in light of the fast performance of string creation from multiple strings, if PerformSomeTask is your bottleneck, try to break the iteration of the strings into multiple threads - it won't improve the efficiency of the code but you'll be able to utilise multiple cores.

Ditto now realising that initialising a string from multiple strings is very quick. — Adam Houldsworth, Jul 02 '10 at 10:47

score 2 · Answer 2 · answered Jul 02 '10 at 08:56

2

Run the two functions through reflector and have a look at the generated assembly. It might give some insights but at best, the performance improval would be minimal.

answered Jul 02 '10 at 08:56

cristobalito

4,192
1
29
41

score 1 · Answer 3 · answered Jul 02 '10 at 10:21

1

I'd go for this instead:

private void TestFunction() 
{ 
  foreach (MySampleClass c in dictSampleClass) 
  { 
    PerformSomeTask(c.VAR1 + c.VAR2 + c.VAR3, c.VAR4); 
  } 
}

There's still probably no real performance benefit, but it removes creating a variable that you don't really need.

answered Jul 02 '10 at 10:21

cjk

45,739
9
81
112

Hmmmm.. My objective was only to know the diffrence not code optimization. – Manjoor Jul 02 '10 at 10:30

score 0 · Answer 4 · answered Jul 02 '10 at 08:56

0

No, it does not. Things like these are handled by the C# compiler which is very inteligent nad you basicaly do not need to care about these details.

Aways use code that promotes readability.

You can check this by disassembling the program.

answered Jul 02 '10 at 08:56

user381879

1

score 0 · Answer 5 · answered Jul 02 '10 at 09:13

0

Ii will not improve performance. But on another note: Assumed improvements are error prone. You have to measure you application and only optimise the slow part if needed. I don't believe that loop is your appliacations's bottleneck.

answered Jul 02 '10 at 09:13

aggsol

2,343
1
32
49

Not safe to assume that either, 10k items with string concatenations in a tight loop - could be memory hungry. – Adam Houldsworth Jul 02 '10 at 09:17
1

@Adam: It's not really a string concatenation in a loop, its a string initialisation from multiple sources. Its not the same as `String str = c.VAR1; str += c.VAR2; str += c.VAR3` which would cause performance problems. – cjk Jul 02 '10 at 10:22
@ck true, I've never considered the performance differences of that. – Adam Houldsworth Jul 02 '10 at 10:25

Variable scope and performance

5 Answers5