Fastest get2fold algorithm

[Riven]

Ouch. System.currentTimeMillis() is way to inaccurate for benchmark-results under 100ms (as it might snap to 15ms values).

Java 1.4 has sun.misc.Perf to do what System.nanoTime does in 1.5+

[DzzD]

it is true that System.currentTimeMillis() can have about 16ms error, but, if the complete test is lasting enough time,the final result will have a very good precision.

example: if the full loop lasts more than 5 seconds and your are testing 1000000 values you will only have the following error precision per op: EDIT not homogen 16ms / (5*1000000) = 3.2 ns max error

EDIT:
Total time error : 0,016/5 = 0,32%

EDIT2 false again ;-)
per op error : 0,32%/nbop=0,32%/1000000= 0,00000032%

per op error : 0,32%



conclusion making a bench with enought values will give the same result with System.currentMillis than any other precision timer ;-), (because the arror of System.currentMillis does not grow with time)

[Riven]

Yes, but the loop takes less than 70ms..

Games published by our own members! Check 'em out!

[DzzD]

ok, maybe it is too short even with good timer

@oravecz : thanks for developping this full bench with all differents methods

[DzzD]

1   2   3   4   5   6   7   8   9   10   11   12   13   14   15   16   17   18   19   20   21   22

public static final int lookup(int val)    {       if (val < 65536)       {          if (val < 256)          {             if(val<0)                return -1;             else                return lookup[val];          }          else             return lookup[val >> 8] << 8;       }       else       {          if (val < 16777216)             return lookup[val >> 16] << 16;          else             return lookup[val >> 24] << 24;       }    }

review of the lookup method with a  better implementation of if/else statements , seems to give better results

[Riven]

You can logicly remove a lot of 'else' statements from that code:

1   2   3   4   5   6   7   8   9   10   11   12   13   14   15   16   17

public static final int lookup(int val) {    if (val < 65536)    {       if (val < 256)       {          if(val<0)             return -1;          return lookup[val];       }       return lookup[val >> 8] << 8;    }    if (val < 16777216)       return lookup[val >> 16] << 16;    return lookup[val >> 24] << 24; }

Very tempting to inline one if/else statement:

1   2   3   4   5   6   7   8   9   10   11   12   13

public static final int lookup(int val) {    if (val < 65536)    {       if (val < 256)          return (val<0) -1 ? lookup[val];       return lookup[val >> 8] << 8;    }    if (val < 16777216)       return lookup[val >> 16] << 16;    return lookup[val >> 24] << 24; }

But probably just as fast

[DzzD]

maybe but your code looks much cleaner,I am  really supid to put a else statement after a return :-)

[Riven]

You can setup Eclipse to whine about it