[Haskell-cafe] Re: newbie optimization question

[Peter Hercek]

OK, if somebody wants to speculate (and if it even makes sense for
such a microbenchmark) here are some more data.
Except different OS and C++ compiler also processor is different.
On my side it was AMD Athlon 64 X2 4800+ (2.4GHz, 2x1MiB L2 cache
non-shared; C&Q was not switched off during the test). The system has
2GiB RAM. The C++ version had working set about 1.7 MiB, ghc version
had it about 2 times bigger.

Peter.

Dusan Kolar wrote:
> Hello all,
> 
>  just to compare the stuff, I get quite other results being on other OS. 
> Thus, the result of C++ compiler may not be that interesting as I do not 
> have the one presented below.
> 
> My machine:
> Linux 2.6.23-ARCH #1 SMP PREEMPT Mon Oct 22 12:50:26 CEST 2007 x86_64 
> Intel(R) Core(TM)2 CPU  6600 @ 2.40GHz GenuineIntel GNU/Linux
> 
> Compilers:
> g++ --version
> g++ (GCC) 4.2.2
> Copyright (C) 2007 Free Software Foundation, Inc.
> This is free software; see the source for copying conditions.  There is NO
> warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
> 
> ghc --version
> The Glorious Glasgow Haskell Compilation System, version 6.6.1
> 
> Measurement:
> compiled with ghc -O2
> time ./mainInteger
> real    0m4.866s
> user    0m4.843s
> sys     0m0.020s
> 
> compiled with ghc -O2
> time ./mainInt64
> real    0m2.213s
> user    0m2.210s
> sys     0m0.003s
> 
> compiled with ghc -O2
> time ./mainInt
> real    0m1.149s
> user    0m1.143s
> sys     0m0.003s
> 
> compiled with g++ -O3
> time ./mainC
> real    0m0.271s
> user    0m0.270s
> sys     0m0.000s
> 
> I don't know what is the reason, but the difference between Int, Int64 
> and Integer is not that dramatic as in example below, nevertheless, the 
> difference between GHC and GNU C++ is very bad.... :-\
> 
> Dusan
> 
> 
> Peter Hercek wrote:
>> Derek Elkins wrote:
>>>
>>> Try with rem instead of mod.
>>>
>>> (What the heck is with bottom?)
>>
>> The bottom was there by error and I was lazy to redo
>>  the tests so I rather posted exactly what I was
>>  doing. I do not know the compiler that good to be
>>  absolutely sure it cannot have impact on result
>>  ... so I rather did not doctor what I did :-)
>>
>> Ok, rem did help quite a bit. Any comments why it is so?
>>
>> Here are summary of results for those interested. I run
>>  all the tests once again. Haskell was about 13% slower
>>  than C++.
>>
>> MS cl.exe options: /Ox /G7 /MD
>> ghc options: -O2
>>
>> C++ version:  1.000; 0.984; 0.984
>> Haskell version specialized to Int: 1.125; 1.125; 1.109
>> Haskell version specialized to Integer: 8.781; 8.813; 8.813
>> Haskell version specialized to Int64: 9.781; 9.766; 9.831
>>
>> The code:
>>
>> % cat b.hs
>> module Main (divisors, perfect, main) where
>> import Data.Int
>>
>> divisors :: Int -> [Int]
>> divisors i = [j | j<-[1..i-1], i `rem` j == 0]
>>
>> perfect :: [Int]
>> perfect = [i | i<-[1..10000], i == sum (divisors i)]
>>
>> main = print perfect
>>
>> % cat b.cpp
>> #include <iostream>
>> using namespace std;
>>
>> int main() {
>>   for (int i = 1; i <= 10000; i++) {
>>     int sum = 0;
>>     for (int j = 1; j < i; j++)
>>       if (i % j == 0)
>>         sum += j;
>>     if (sum == i)
>>       cout << i << " ";
>>   }
>>   return 0;
>> }
>>
>> %
>>
>> _______________________________________________
>> Haskell-Cafe mailing list
>> Haskell-Cafe at haskell.org
>> http://www.haskell.org/mailman/listinfo/haskell-cafe
>