Comments on Wondering around: Microbenchmarking Scala vs Java

Can you compare it with machine code? I always wo...

2010-02-06T18:20:29.770-08:00

Can you compare it with machine code?

I always wonder if I programmed it in assembly, if it would be very much faster or not.

Sorry, here is the link: http://blog.objectmentor....

2009-07-11T23:23:04.896-07:00

Sorry, here is the link: http://blog.objectmentor.com/articles/2009/06/05/bay-area-scala-enthusiasts-base-meeting-whats-new-in-scala-2-8

Hi Stefan, check out Dean's ...

2009-07-11T23:21:05.726-07:00

Hi Stefan, check out Dean's post where he discuss @specialized.

Is there additional information available regardin...

2009-07-06T13:10:41.155-07:00

Is there additional information available regarding the @specialized annotation? From what I understand it would have to be used when defining the generic type. Currently I can't find any usage in the scala 2.8 library. Who can tell more about that interesting thing?

Yours,
Stefan

You got it Ismael! Only the last suggestion (the ...

2009-07-03T23:15:29.318-07:00

You got it Ismael!

Only the last suggestion (the bug you found) made any difference. After solving it Scala and Java matched up. Actually Scala was 2ms but considering the non-realtime OS, 0.1% performance difference is meaningless. I'll update the post.

Eishay, I understand that while loops are used in...

2009-07-03T12:44:10.809-07:00

Eishay,

I understand that while loops are used in this benchmark. That is why I said that specialized would not help.

What does help is using the same code for both versions. :)

Here are some suggestions (they make the Scala code closer to the Java one and made Scala's version faster when I tested):

- Replace the inner methods with normal ones.

- Change Scala's version to use a class instead of object and create an object that simply creates the class and calls main on it.

- The line that starts with "if (j < r)" is different from the Java code. The Java code is "if (R > i)".

Does that make any difference?

Ismael

But Ismael, the Scala code in the example does use...

2009-07-03T10:58:27.623-07:00

But Ismael, the Scala code in the example does use a while loop (compiled down to a for loop).
I'm trying to change the Scala code to close the performance gap between it and the Java code. Any ideas?

Of course not, Eishay. That would be ridiculous. ...

2009-07-03T10:25:34.973-07:00

Of course not, Eishay. That would be ridiculous.

I said that you have to use a while loop in Scala. A while loop (in Scala or Java) performs as well as a for loop in Java.

Ismael

@Nick I figured its a typo, just teasing ;-) @Is...

2009-07-03T09:59:28.888-07:00

@Nick I figured its a typo, just teasing ;-)

@Ismael did you mean you need to cripple Java so it won't use for loops in order to make it as slow as Scala? Is there a way to make similar Scala code as fast as Java in spite of the lack of for loops?

Hi Eishay, It's not surprising that @speciali...

2009-07-03T00:34:36.013-07:00

Hi Eishay,

It's not surprising that @specialized doesn't help much here. As far as I can see, while loops and Array[Int] are used almost everywhere (I only found one exception).

Scala is as fast as Java if you write it like Java (this means using while loops since Scala doesn't have an equivalent of Java's for loops).

The point of @specialized is to help cases where the code is higher level and you use anonymous functions extensively as well as higher-level collections. In the latter case, @specialized has the potential to allow Scala to do better than Java when using collections (not arrays) from the standard library (due to boxing on the Java side). Note that I said _potential_. :)

Best,
Ismael

Okay, that was just a typo in my published code se...

2009-07-02T22:14:42.524-07:00

Okay, that was just a typo in my published code segment. I really compared 100.000.

I also tried to figure out the influence of HotSpot, but I couldn't see any remarkable improvements. And also, in my benchmark Java is 30% faster.