Jump to content

  • Log In with Google      Sign In   
  • Create Account


Is there an overhead for performing chains of method calls?


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
8 replies to this topic

#1 warnexus   Prime Members   -  Reputation: 1399

Like
0Likes
Like

Posted 29 April 2013 - 09:58 AM

Is there an overhead for performing chains of method calls?

 

Example of my code from past projects:

 

Main.getPolynomialHolder().get(polynomialIndex).getPolynomialVector().get(i).setExponent(Main.getPolynomialHolder().get(polynomialIndex).getPolynomialVector().get(i).getExponent()-1);



Sponsor:

#2 SiCrane   Moderators   -  Reputation: 9561

Like
0Likes
Like

Posted 29 April 2013 - 10:32 AM

Maybe. It depends on your language, the quality of your compiler, compiler settings used, exactly how each function is implemented and so on.

#3 warnexus   Prime Members   -  Reputation: 1399

Like
0Likes
Like

Posted 29 April 2013 - 10:54 AM

Maybe. It depends on your language, the quality of your compiler, compiler settings used, exactly how each function is implemented and so on.

The code is in Java. My IDE is Eclipse. The code uses the built-in Vector Class from the standard Java library.  



#4 frob   Moderators   -  Reputation: 20302

Like
5Likes
Like

Posted 29 April 2013 - 11:53 AM

There are actually two things.

 

First, the average cost of those function call.

 

It is something very specific to the hardware and the compiler combination.  It is something that lots of people have researched and measured over the years.  

 

However, since compilers have converged on the near-optimal case for function calls, and since processor speeds have basically top-lined around 2.5-3.0 GHz range a decade ago, the timing remains basically unchanged.

 

The amortized cost of an optimized build function call is approximately 7 nanoseconds overhead per function call, vs. the cost of direct access.  There is an additional approximately 7 nanoseconds overhead for virtual functions.

 

Again, for an optimized build, non-virtual functions generally average around 7ns each, virtual functions generally average around 14ns each.

 

Assuming you have all optimizations enabled in your code and it has been either directly compiled or JIT-optimized, you will probably see similar results.

 

 

 

Now for the bigger issue.

 

Your code has a nasty code smell.  It is often called "inappropriate intimacy".  The solution pattern is called the Law of Demeter, where basically you should only talk to the objects immediately available to you.  The most you should do is a.b().  

 

You have a.b().c().d().e().f(), which is less of a a 'code smell' and more like a reeking 2-day-old diaper.  Don't do that.


Edited by frob, 29 April 2013 - 11:55 AM.

Check out my personal indie blog at bryanwagstaff.com.

#5 warnexus   Prime Members   -  Reputation: 1399

Like
0Likes
Like

Posted 29 April 2013 - 01:17 PM

There are actually two things.

 

First, the average cost of those function call.

 

It is something very specific to the hardware and the compiler combination.  It is something that lots of people have researched and measured over the years.  

 

However, since compilers have converged on the near-optimal case for function calls, and since processor speeds have basically top-lined around 2.5-3.0 GHz range a decade ago, the timing remains basically unchanged.

 

The amortized cost of an optimized build function call is approximately 7 nanoseconds overhead per function call, vs. the cost of direct access.  There is an additional approximately 7 nanoseconds overhead for virtual functions.

 

Again, for an optimized build, non-virtual functions generally average around 7ns each, virtual functions generally average around 14ns each.

 

Assuming you have all optimizations enabled in your code and it has been either directly compiled or JIT-optimized, you will probably see similar results.

 

 

 

Now for the bigger issue.

 

Your code has a nasty code smell.  It is often called "inappropriate intimacy".  The solution pattern is called the Law of Demeter, where basically you should only talk to the objects immediately available to you.  The most you should do is a.b().  

 

You have a.b().c().d().e().f(), which is less of a a 'code smell' and more like a reeking 2-day-old diaper.  Don't do that.

Thanks for the feedback on the code. The smell came from the design I had in mind. 



#6 Orangeatang   Members   -  Reputation: 1456

Like
0Likes
Like

Posted 29 April 2013 - 02:14 PM

The most you should do is a.b().  

I've always liked a quote from Brian Kernighan regarding debugging:

 

Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it

 

What frob said is spot on :)



#7 Waterlimon   Crossbones+   -  Reputation: 2473

Like
0Likes
Like

Posted 29 April 2013 - 02:26 PM

Is there any inlining taking place?

I would assume so in c++ in highly used code, but dont know about java.

o3o


#8 rip-off   Moderators   -  Reputation: 8119

Like
0Likes
Like

Posted 29 April 2013 - 04:23 PM

Is there any inlining taking place?

Almost certainly. A modern JVM will inline functions dynamically based on how they are actually used. An interesting consequence of this is that in some circumstances a JVM can inline in C++ what would be "virtual" calls.



#9 TheChubu   Crossbones+   -  Reputation: 4101

Like
0Likes
Like

Posted 30 April 2013 - 01:05 AM

Aren't all Java method calls "virtual" ? I think I read that when I was researching the final keyword for methods and their parameters. Which many concluded it basically does nothing since the most relevant work to see if it can be inlined is done by the VM by analyzing the behavior of the method directly rather than relying on keywords.


"I AM ZE EMPRAH OPENGL 3.3 THE CORE, I DEMAND FROM THEE ZE SHADERZ AND MATRIXEZ"

 

My journals: dustArtemis ECS framework and Making a Terrain Generator





Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS