Loading ...
Sorry, an error occurred while loading the content.
 

Re: [PrimeNumbers] big numbers library

Expand Messages
  • Alan Eliasen
    ... That s because Sun s Java implementation still uses horrible O(n^2) algorithms for multiplication. Their exponentiation routine already does
    Message 1 of 9 , Aug 7 5:34 AM
      fyatim wrote:
      > I wrote a small method using a well known algorithm “Exponentiation by
      > Squaring”, but it still take horrible execution time.

      That's because Sun's Java implementation still uses horrible O(n^2)
      algorithms for multiplication. Their exponentiation routine already
      does exponentiation by squaring, so you probably can't improve on it
      without fixing multiplication.

      > Where can I find the algorithm you are mentioning i.e. “bit shifting” ??

      "Bit shifting" works when your base contains powers of 2. This is
      because you can do powers of 2 by simply left-shifting the binary
      representation by the appropriate number of bits. If you're doing
      something like calculating large Mersenne numbers, this makes it about a
      thousand times faster or more than Sun's implementation. It's an easy
      and obvious optimization that Sun missed.

      In short, here's a code snippet that does it. The base is expected
      to be in a BigInteger called "big", and the exponent in an int called
      "exponent". It factors out powers of two quickly by the call to
      getLowestSetBit(), and does the exponentiation for powers of two rapidly
      with shiftLeft() and then multiplies it by the remaining part that isn't
      a power of 2.

      This only helps if your base contains powers of 2.

      if (big.signum() > 0)
      {
      // Get factor of two
      int bit = big.getLowestSetBit();

      if (bit > 0)
      {
      big = big.shiftRight(bit);
      BigInteger twoPower = FrinkBigInteger.ONE.shiftLeft(bit*exponent);
      if (big.equals(FrinkBigInteger.ONE))
      return FrinkInteger.construct(twoPower);
      else
      {
      big = big.pow(exponent);
      return FrinkInteger.construct(big.multiply(twoPower));
      }
      }
      }

      --
      Alan Eliasen | "It is not enough to do your best;
      eliasen@... | you must know what to do and THEN
      http://futureboy.homeip.net/ | do your best." -- W. Edwards Deming
    • Alan Eliasen
      If it wasn t clear (and it wasn t,) the algorithm I just posted was one to speed up exponentiation. I use it as a wrapper around BigInteger.pow(BigInteger
      Message 2 of 9 , Aug 7 5:39 AM
        If it wasn't clear (and it wasn't,) the algorithm I just posted was
        one to speed up exponentiation. I use it as a wrapper around
        BigInteger.pow(BigInteger big, int exponent) in Java.

        --
        Alan Eliasen | "It is not enough to do your best;
        eliasen@... | you must know what to do and THEN
        http://futureboy.homeip.net/ | do your best." -- W. Edwards Deming
      • fyatim
        Jan, Alan, Thank you for your support. I wrote a small method using a well known algorithm Exponentiation by Squaring , but it still take horrible execution
        Message 3 of 9 , Aug 20 4:19 AM
          Jan, Alan,

          Thank you for your support.

          I wrote a small method using a well known algorithm "Exponentiation by
          Squaring", but it still take horrible execution time.

          Where can I find the algorithm you are mentioning i.e. "bit shifting" ??

          Faysal





          _____

          From: primenumbers@yahoogroups.com [mailto:primenumbers@yahoogroups.com] On
          Behalf Of Jan van Oort
          Sent: Friday, August 05, 2005 8:57 PM
          To: Alan Eliasen
          Cc: Prime Number
          Subject: Re: [PrimeNumbers] big numbers library



          Alan,
          you are quite right. Anyway, if I were in Faysal's case, it probably would
          boil down - and that rather quickly - to option 1, with an eye on option 2.
          The next time I am in between two project, I am going to write a
          BigInteger.pow( BigInteger _exponent ) method. Just for fun. On a rainy day.

          :-D
          Jan
          PS I was not complaining about any limits. Faysal was.

          On 8/5/05, Alan Eliasen <eliasen@...> wrote:
          >
          > Jan van Oort wrote:
          > > 2) take Colin Plum's source code, study it closely, and use to implement
          > > your own method in a shared library ( .so or .dll compiled from C/C++ ).
          > > Probably closely related to 1)
          > > 3) download the sources of java.math.BigInteger and see if you can learn
          > > anything from it, i.e. from the non-native ( Java ) part.
          > > 4) look on the internet --- developer's forums etc --- if somebody has
          > > already had the same idea / need as you, and developed something
          > > 5) contact Colin Plum and ask him directly.
          >
          > I need to clarify that Sun's JVM hasn't used Colin Plumb's library
          > since version 1.2. Since then, BigInteger has been implented in pure Java.
          >
          > If you use the free Kaffe JVM, it can be compiled to use the GMP
          > library for BigInteger. You need to both compile with this and enable
          > it at runtime with a command-line switch. Still, there's no API to have
          > BigInteger exponents.
          >
          > In any case, BigInteger represents its bits as an int[] array. This
          > limits the size of the numbers that can be represented to about
          > 68 billion digits, or possibly less.
          >
          > But Sun's algorithms are horrible--they only use the most naive
          > O(n^2) algorithm for multiplication, and equally horrible algorithms for
          > radix conversion and exponentiation, so your program would be horribly
          > time-bound. For example, it takes about 16 hours just to convert one of
          > the larger Mersenne numbers, about 2^13000000, (which is *way, way*
          > smaller than the 2^2147483647 limit that you're complaining about) to
          > decimal on my computer. The exponentiation takes horribly long unless
          > you write your own bit-shifting algorithms to fix Sun's inadequacies too.
          >
          > In short, if you're going to use numbers that big, you're going to
          > have to write your own library. Or try Kaffe. NB: make sure you
          > compile Kaffe with the option that intermediate numbers are allocated on
          > the heap, not the stack, 'cause your stack isn't gigabytes in size.
          >
          > --
          > Alan Eliasen | "It is not enough to do your best;
          > eliasen@... | you must know what to do and THEN
          > http://futureboy.homeip.net/ | do your best." -- W. Edwards Deming
          >



          --

          Non sunt multiplicanda entia praeter necessitatem


          [Non-text portions of this message have been removed]



          Unsubscribe by an email to: primenumbers-unsubscribe@yahoogroups.com
          The Prime Pages : http://www.primepages.org/






          SPONSORED LINKS


          Mathematics
          <http://groups.yahoo.com/gads?t=ms&k=Mathematics+and+computer+science&w1=Mat
          hematics+and+computer+science&w2=Music+theory&w3=Theory+of&c=3&s=71&.sig=SOh
          q77_rF4rHyNjUSRWLFg> and computer science

          Music
          <http://groups.yahoo.com/gads?t=ms&k=Music+theory&w1=Mathematics+and+compute
          r+science&w2=Music+theory&w3=Theory+of&c=3&s=71&.sig=6RLtLyQQ5iFfnkmGdOkbBA>
          theory

          Theory
          <http://groups.yahoo.com/gads?t=ms&k=Theory+of&w1=Mathematics+and+computer+s
          cience&w2=Music+theory&w3=Theory+of&c=3&s=71&.sig=O8mFVRYrr9LhUeQJhzbgFg>
          of



          _____

          YAHOO! GROUPS LINKS



          * Visit your group "primenumbers
          <http://groups.yahoo.com/group/primenumbers> " on the web.

          * To unsubscribe from this group, send an email to:
          primenumbers-unsubscribe@yahoogroups.com
          <mailto:primenumbers-unsubscribe@yahoogroups.com?subject=Unsubscribe>

          * Your use of Yahoo! Groups is subject to the Yahoo!
          <http://docs.yahoo.com/info/terms/> Terms of Service.



          _____



          [Non-text portions of this message have been removed]
        • fyatim
          Alan, Yes is clear .. Thanks. But, my problem still the same when the exponent is a prime 2... I m trying to find out the best (fastest) way to calculate it
          Message 4 of 9 , Aug 21 1:15 AM
            Alan,
            Yes is clear .. Thanks.
            But, my problem still the same when the exponent is a prime > 2...
            I'm trying to find out the best (fastest) way to calculate it ...
            I welcome any new ideas...
            Regards
            Faysal

            -----Original Message-----
            From: primenumbers@yahoogroups.com [mailto:primenumbers@yahoogroups.com] On
            Behalf Of Alan Eliasen
            Sent: Sunday, August 07, 2005 3:34 PM
            To: fyatim
            Cc: 'Jan van Oort'; 'Prime Number'
            Subject: Re: [PrimeNumbers] big numbers library

            fyatim wrote:
            > I wrote a small method using a well known algorithm "Exponentiation by
            > Squaring", but it still take horrible execution time.

            That's because Sun's Java implementation still uses horrible O(n^2)
            algorithms for multiplication. Their exponentiation routine already
            does exponentiation by squaring, so you probably can't improve on it
            without fixing multiplication.

            > Where can I find the algorithm you are mentioning i.e. "bit shifting" ??

            "Bit shifting" works when your base contains powers of 2. This is
            because you can do powers of 2 by simply left-shifting the binary
            representation by the appropriate number of bits. If you're doing
            something like calculating large Mersenne numbers, this makes it about a
            thousand times faster or more than Sun's implementation. It's an easy
            and obvious optimization that Sun missed.

            In short, here's a code snippet that does it. The base is expected
            to be in a BigInteger called "big", and the exponent in an int called
            "exponent". It factors out powers of two quickly by the call to
            getLowestSetBit(), and does the exponentiation for powers of two rapidly
            with shiftLeft() and then multiplies it by the remaining part that isn't
            a power of 2.

            This only helps if your base contains powers of 2.

            if (big.signum() > 0)
            {
            // Get factor of two
            int bit = big.getLowestSetBit();

            if (bit > 0)
            {
            big = big.shiftRight(bit);
            BigInteger twoPower = FrinkBigInteger.ONE.shiftLeft(bit*exponent);
            if (big.equals(FrinkBigInteger.ONE))
            return FrinkInteger.construct(twoPower);
            else
            {
            big = big.pow(exponent);
            return FrinkInteger.construct(big.multiply(twoPower));
            }
            }
            }

            --
            Alan Eliasen | "It is not enough to do your best;
            eliasen@... | you must know what to do and THEN
            http://futureboy.homeip.net/ | do your best." -- W. Edwards Deming


            Unsubscribe by an email to: primenumbers-unsubscribe@yahoogroups.com
            The Prime Pages : http://www.primepages.org/


            Yahoo! Groups Links
          Your message has been successfully submitted and would be delivered to recipients shortly.