Fastutils wrapping benchmark

[szymonm]

No description provided.

[szymonm]

When I run the code wrapped version of fastutils collection performs 50% worse than not wrapped:

time of pure map ops: 1862
time of wrapped map ops: 2678

This is possibly due to boxing.

[pankajgupta]

I verified on my computer as well, and it is indeed 50% slower on mine too...

[pankajgupta]

Interestingly, I get the same results when I do
val easyMap = (new Int2IntOpenHashMap()).asInstanceOf[java.util.Map[Int, Int]]

So it is not the scala wrapping, but just the boxing/unboxing at java level.

Any suggestions, @szymonm ? esp. in the light of #139 where also we want to rethink InfoKeeper.

[szymonm]

Yes, looks like using Int2IntOpenHashMap as Map[Int, Int] makes the methods boxed.

Right, we have go back with InfoKeepers.

And hope for efficient specialization in Scala one day.

[szymonm]

I'm afraid that this means whenever we use:

val nodes: Seq[Int] = new Array[Int](100)

then we have boxing...

[szymonm]

scala> :paste
// Entering paste mode (ctrl-D to finish)

def time(b : => Unit) {
  val start = java.lang.System.currentTimeMillis()
  b
  print(s"Time: ${java.lang.System.currentTimeMillis() - start}")
}

// Exiting paste mode, now interpreting.

time: (b: => Unit)Unit

scala> import scala.util.Random
import scala.util.Random

scala> val arr = new Array[Int](1000000)
scala> (0 until 1000000).foreach{ i => arr(i) = i }

scala> var b = 0
b: Int = 0

scala> time {(0 until 1000000).foreach{ i => b = arr(i) * 7 }}
Time: 7
scala> val seq: Seq[Int] = arr
scala> time {(0 until 1000000).foreach{ i => b = seq(i) * 7 }}
Time: 19

[szymonm]

@pankajgupta please have a look and check it too

[pankajgupta]

Interesting indeed.

In general, in cassovary the desire has been to be first and foremost
efficient on storage (so native arrays) and then be as performant as
possible without getting too low level. This clearly shows that there are
opportunities to improve speed.

On Mon, Feb 9, 2015 at 2:38 PM, Szymon [email protected] wrote:

@pankajgupta https://github.com/pankajgupta please have a look

—
Reply to this email directly or view it on GitHub
#142 (comment).