Performance of fibonacci function

You can save another around 8-10% (on my machine) using:

def fib(i) {
    if (i == 0 || i == 1) return 1
    return fib(i-1)+fib(i-2)
}

long millis = new Date().getTime()
println fib(args[0])
println new Date().getTime() - millis

but there are plans in the works to save much more.

Update on the numbers: running on Java SE 6, the milliseconds for fib 30 and fib 40 are ~771 and ~71100 respectively. I had read about the Java 6 performance upgrades, but did not expect them to be this significant.

For the java version the change on fibj 30 is insignificant, but fibj 40 drops to ~1041ms.

Performance has definitely increased in 1.6 beta 2. After recompiling the examples:

Groovy Version: 1.6-beta-2 JVM: 1.5.0_16

Here:fib ricardo$ java fib 30
1346269
818
Here:fib ricardo$ java fib 40
165580141
95574

1.8 allows for significant improvements in the performance of the fibonacci example

I saw reporter claimed that to be 90 times slower. Jochen, what are the last numbers from your machine?

for fib.groovy using 1.8-beta4
n = 30 : 80-105
n = 40 : 3028-3084

for fib.java
n = 30 : 17 - 27
n = 40 : 1346-2018

I would not be surprised if Groovy 1.6 would be even slower than 90 times using the setup of this issue. The way of testing is in my opinion less than ideal. There is no warmup, there is no measuring through multiple runs and I think java.util.Date is less than ideal for this too. Considering the setup I think my numbers are still good enough to be shown and far from the 90 times. For n=30 in this setup Groovy looks like being java sped multiplied with 3-7. And for n=40 the factor is 2.5-1.5. In my measurements I usually get something between factor 2 and 3 for this kind of test, but with a different setup. Of course there are two significant differences to my testing using this setup and what your reporter is saying. And that is that I am using Groovy-1.8-beta4 and I am using a openjdk7 (1.7.0-b125) for this. It would really surprise me if you would get this using JDK1.5. JDK1.6 surely has improvement here, but 1.7 is simply a whole different story. Additionally it is important to have the JVM in server mode for Groovy. Client mode just doesn't bring it for such things in combination with Groovy. Since I am on a 64bit system I tend to forget such things as here the server mode is the only one... and I might be wrong but I think java7 has no real client mode as well.