www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - performance of structs vs classes

reply =?ISO-8859-1?Q?Christian_K=F6stlin?= <christian.koestlin gmail.com> writes:
I have this small program here, which compares the call-overhead of 
methods from classes to structs. I guess this always has to be, because 
of the "polymorphism" of methods of objects.

Is there something I can do to improve on the performance of methods 
calls with classes?

please see the source here:
https://gist.github.com/1242911


thanks in advance

christian
Sep 26 2011
next sibling parent Andrej Mitrovic <andrej.mitrovich gmail.com> writes:
You didn't call sw.reset() before calling sw.stop()!
Sep 26 2011
prev sibling next sibling parent Andrej Mitrovic <andrej.mitrovich gmail.com> writes:
On 9/26/11, Andrej Mitrovic <andrej.mitrovich gmail.com> wrote:

 You didn't call sw.reset() before calling sw.stop()!


Ehh, I mean before sw.start().
Sep 26 2011
prev sibling next sibling parent reply Andrej Mitrovic <andrej.mitrovich gmail.com> writes:
In the class test reset the timer:
    {
        auto h = new HClass();
        sw.reset(); // <-
        sw.start();
    }
Sep 26 2011
parent reply =?ISO-8859-1?Q?Christian_K=F6stlin?= <christian.koestlin gmail.com> writes:
On 09/26/2011 08:29 PM, Andrej Mitrovic wrote:

 In the class test reset the timer:
      {
          auto h = new HClass();
          sw.reset(); //<-
          sw.start();
      }


thanks a lot ... i totally misused the stopwatch api.

i compiled with:
   dmd -release -O -inline -m64 -oftarget/dmd/structs_vs_classes 
experimental/structs_vs_classes.d
and
   gdc -O3 -inline -o target/gdc/structs_vs_classes 
experimental/structs_vs_classes.d

wich resulted in:
target/dmd/structs_vs_classes
time with struct: 11326 for 1
time with class: 11323 for 1

and
target/gdc/structs_vs_classes
time with struct: 4011 for 1
time with class: 6880 for 1

which is much better than my wrong measurements...

thanks for the answer!!!

christian
Sep 26 2011
next sibling parent reply bearophile <bearophileHUGS lycos.com> writes:
Christian K.:


 i totally misused the stopwatch api.


I have misused it the same way, the first time. If other people will find the
same problem then it means its API has problems.

Bye,
bearophile
Sep 26 2011
parent reply "Steven Schveighoffer" <schveiguy yahoo.com> writes:
On Mon, 26 Sep 2011 15:58:48 -0400, bearophile <bearophileHUGS lycos.com>  
wrote:


 Christian K.:


 i totally misused the stopwatch api.


 I have misused it the same way, the first time. If other people will  
 find the same problem then it means its API has problems.


Have you ever used a real stopwatch?  It works the same way :)  Start/stop  
does not reset the time to 0, it just adds more time to the total.  Reset  
clears to 0.

I think probably the documentation needs to be better.  The example also  
suggests start resets the stopwatch.

-Steve
Sep 26 2011
parent =?UTF-8?B?Q2hyaXN0aWFuIEvDtnN0bGlu?= <christian.koestlin gmail.com> writes:
On 09/26/2011 10:30 PM, Steven Schveighoffer wrote:

 On Mon, 26 Sep 2011 15:58:48 -0400, bearophile
 <bearophileHUGS lycos.com> wrote:


 Christian K.:


 i totally misused the stopwatch api.


 I have misused it the same way, the first time. If other people will
 find the same problem then it means its API has problems.


 Have you ever used a real stopwatch? It works the same way :) Start/stop
 does not reset the time to 0, it just adds more time to the total. Reset
 clears to 0.

 I think probably the documentation needs to be better. The example also
 suggests start resets the stopwatch.

 -Steve


you are totally right ... it is the same as a real stopwatch ... so 
there should be no problem (i also used it correctly in another 
benchmark, but it was quite late and ...)

thanks

christian
Sep 26 2011
prev sibling parent reply "Steven Schveighoffer" <schveiguy yahoo.com> writes:
On Mon, 26 Sep 2011 15:36:04 -0400, Christian K=C3=B6stlin  =

<christian.koestlin gmail.com> wrote:


 On 09/26/2011 08:29 PM, Andrej Mitrovic wrote:

 In the class test reset the timer:
      {
          auto h =3D new HClass();
          sw.reset(); //<-
          sw.start();
      }


 thanks a lot ... i totally misused the stopwatch api.

 i compiled with:
    dmd -release -O -inline -m64 -oftarget/dmd/structs_vs_classes  =



 experimental/structs_vs_classes.d
 and
    gdc -O3 -inline -o target/gdc/structs_vs_classes  =



 experimental/structs_vs_classes.d

 wich resulted in:
 target/dmd/structs_vs_classes
 time with struct: 11326 for 1
 time with class: 11323 for 1

 and
 target/gdc/structs_vs_classes
 time with struct: 4011 for 1
 time with class: 6880 for 1


use -frelease to disable invariant calls on gdc.

Very important!

-Steve
Sep 26 2011
parent =?UTF-8?B?Q2hyaXN0aWFuIEvDtnN0bGlu?= <christian.koestlin gmail.com> writes:
On 09/26/2011 10:34 PM, Steven Schveighoffer wrote:

 On Mon, 26 Sep 2011 15:36:04 -0400, Christian Köstlin
 <christian.koestlin gmail.com> wrote:


 On 09/26/2011 08:29 PM, Andrej Mitrovic wrote:

 In the class test reset the timer:
 {
 auto h = new HClass();
 sw.reset(); //<-
 sw.start();
 }


 thanks a lot ... i totally misused the stopwatch api.

 i compiled with:
 dmd -release -O -inline -m64 -oftarget/dmd/structs_vs_classes
 experimental/structs_vs_classes.d
 and
 gdc -O3 -inline -o target/gdc/structs_vs_classes
 experimental/structs_vs_classes.d

 wich resulted in:
 target/dmd/structs_vs_classes
 time with struct: 11326 for 1
 time with class: 11323 for 1

 and
 target/gdc/structs_vs_classes
 time with struct: 4011 for 1
 time with class: 6880 for 1


 use -frelease to disable invariant calls on gdc.

 Very important!

 -Steve


you are right ... this leads to 6183 for gdc with class

thanks!

christian
Sep 26 2011
prev sibling parent "Steven Schveighoffer" <schveiguy yahoo.com> writes:
On Mon, 26 Sep 2011 14:14:09 -0400, Christian K=C3=B6stlin  =

<christian.koestlin gmail.com> wrote:


 I have this small program here, which compares the call-overhead of  =



 methods from classes to structs. I guess this always has to be, becaus=


e  =


 of the "polymorphism" of methods of objects.

 Is there something I can do to improve on the performance of methods  =



 calls with classes?

 please see the source here:
 https://gist.github.com/1242911


In addition to Andrej's information (call sw.reset()),  make *sure* you =
 =

compile in -release mode, because in non -release mode, all object metho=
ds  =

are followed by a call to Object's invariant, which is never inlined!   =

However, no invariants are called in release mode.

-Steve
Sep 26 2011