B4J Tutorial B4J Primes Benchmark

I found this benchmark that tests the same algorithm in multiple programming languages and decided to test it in B4J.

I've implemented the two algorithms.

Sub Process_Globals

End Sub

Sub AppStart (Args() As String)
    #if debug
    Log("Never test performance in debug mode!!!")
    ExitApplication
    #End If
 
    Dim start As Long = DateTime.Now
 
    'Algorithm #1
    Dim counter As Int = GetAllPrimesLessThan1(20000000) '20,000,000
    'Algorithm #2
'    Dim counter As Int = GetAllPrimesLessThan2(200000000) '200,000,000
    Dim EndTime As Long = DateTime.Now
    Log($"number of primes: ${NumberFormat2(counter, 0, 0, 0, True)}"$)
    Log($"Time: $1.4{(EndTime - start) / 1000}"$)
End Sub

Private Sub GetAllPrimesLessThan1 (MaxPrime As Int) As Int
    Dim counter As Int
    For i = 2 To MaxPrime - 1
        If IsPrime(i) Then counter = counter + 1
    Next
    Return counter
End Sub

Private Sub IsPrime(Num As Int) As Boolean
    If Num <= 1 Then Return False
    If Num <= 3 Then Return True
    If ((Num Mod 2) = 0 Or (Num Mod 3) = 0) Then Return False
    For c = 5 To Sqrt(Num) Step 6
        If ((Num Mod c) = 0 Or Num Mod (c + 2) = 0) Then Return False
    Next
    Return True
End Sub

Private Sub GetAllPrimesLessThan2 (MaxPrime As Int) As Int
    Dim Count As Int = 0
    Dim MaxSquareRoot As Int = Sqrt(MaxPrime)
    Dim eliminated As B4XBitSet 'B4XCollections (collections implemented in B4A)
    eliminated.Initialize(MaxPrime + 1)
    For i = 2 To MaxPrime
        If eliminated.Get(i) = False Then
            Count = Count + 1
            If i <= MaxSquareRoot Then
                For j = i * i To MaxPrime Step i
                    eliminated.Set(j, True)
                Next
            End If
        End If
    Next
    Return Count
End Sub

And the results (measured in seconds):

Algorithm #1

B4J: 2.78
Java: 3.07
Java with the limit calculated before the loop (as in B4J): 2.91
C#: 2.69

Algorithm #2 - sieve

The C# code was fixed to only return the count of primes without collecting the primes, and the maximum was set to 200M.
B4J - same algorithm as the C# algorithm: 1.80
Java: 1.90
C#: 2.50


What does it mean?

The exact numbers aren't important. Small changes in the code can affect the results in unexpected ways due to the underlying compiler optimizations.
It does show that B4J, Java and C# have similar core performance.

 

[Didier9]

I have been impressed by the performance of B4J.
Knowing (at least at a conceptual level) what goes on in a virtual machine, it is quite astonishing.
It's good to have numbers to attach to it
Thank you and congrats Erel!

 

[chikega]

These benchmarks are being showcased on Dave's Garage I believe. Interesting to see the differences.

 

[CaptKronos]

I have seen the benchmarks being discussed on Dave's Garage. Is someone going to submit the B4J code? It should be good publicity.