Revision history for Math::BigInt::Pari

2011-02-26 v1.16 pjacklam (6153 tests)

• Change bigintpm.inc to reflect recent changes in the Math::BigInt distribution (Peter John Acklam).

2011-02-10 v1.15 pjacklam (6151 tests)

• Include latest version of test scripts from Math-BigInt (closes RT #65580) (Peter John Acklam).
• Add 00sig.t for testing SIGNATURE, and 01load.t for basic module loading and giving diagnostics useful for bug tracking (Peter John Acklam).
• Rename pod.t to 02pod.t and pod_cov.t to 03podcov.t and use more generic code (Peter John Acklam).
• Clean up whitespace (Peter John Acklam).

2010-09-10 v1.14 rafl 5716 tests

• Fix tests with Math::BigInt >= 1.90 and depend on it.

2007-07-01 v1.13 Tels 5704 tests

• require Math::BigInt v1.87
• testsuite from MBI v1.87
• logint() calculates a guess, and then use this to find the real result, this makes it almost independent from the results size and much faster:

  Using Math::BigInt v1.87 and Math::BigInt::Pari v1.12

00 len 2: 3s (3.11 usr + 0.00 sys = 3.11 CPU) @ 4714/s (n=14661) 01 len 2 big: 3s (3.00 usr + 0.01 sys = 3.01 CPU) @ 532/s (n=1604)

  02 len 3:     3s (3.19 usr + 0.01 sys = 3.20 CPU) @   728/s (n=2332)
  03 len 7:     3s (3.01 usr + 0.16 sys = 3.17 CPU) @  1960/s (n=6216)
  04 len 8:     3s (3.21 usr + 0.00 sys = 3.21 CPU) @  2247/s (n=7215)

05 len 11: 3s (3.18 usr + 0.00 sys = 3.18 CPU) @ 2862/s (n=9104) 06 len 14: 3s (3.23 usr + 0.00 sys = 3.23 CPU) @ 3620/s (n=11695) 06 len 20: 3s (3.25 usr + 0.00 sys = 3.25 CPU) @ 4882/s (n=15868) 0a len 3 big: 3s (3.16 usr + 0.00 sys = 3.16 CPU) @ 168/s (n=534)

Using Math::BigInt v1.87 and Math::BigInt::Pari v1.13

00 len 2: 2s (3.24 usr + 0.00 sys = 3.24 CPU) @ 13064/s (n=42328) 01 len 2 big: 3s (3.17 usr + 0.00 sys = 3.17 CPU) @ 4853/s (n=15387)

  02 len 3:     3s (3.23 usr + 0.00 sys = 3.23 CPU) @  4831/s (n=15607)
  03 len 7:     3s (3.24 usr + 0.00 sys = 3.24 CPU) @  4813/s (n=15597)
  04 len 8:     3s (3.22 usr + 0.00 sys = 3.22 CPU) @  4778/s (n=15387)

05 len 11: 3s (3.02 usr + 0.16 sys = 3.18 CPU) @ 4754/s (n=15119) 06 len 14: 3s (3.15 usr + 0.00 sys = 3.15 CPU) @ 4740/s (n=14934) 06 len 20: 3s (3.24 usr + 0.00 sys = 3.24 CPU) @ 4749/s (n=15387) 0a len 3 big: 3s (3.22 usr + 0.00 sys = 3.22 CPU) @ 877/s (n=2825)

For reference:

Using Math::BigInt v1.87 and Math::BigInt::GMP 1.22

00 len 2: 3s (3.14 usr + 0.00 sys = 3.14 CPU) @ 16853/s (n=52920) 01 len 2 big: 3s (3.13 usr + 0.00 sys = 3.13 CPU) @ 16391/s (n=51306)

  02 len 3:     3s (3.17 usr + 0.03 sys = 3.20 CPU) @ 16036/s (n=51316)
  03 len 7:     3s (2.97 usr + 0.13 sys = 3.10 CPU) @ 16063/s (n=49797)
  04 len 8:     3s (3.10 usr + 0.01 sys = 3.11 CPU) @ 16011/s (n=49797)

05 len 11: 3s (3.14 usr + 0.00 sys = 3.14 CPU) @ 15858/s (n=49797) 06 len 14: 3s (3.14 usr + 0.01 sys = 3.15 CPU) @ 15808/s (n=49797) 06 len 20: 3s (3.15 usr + 0.01 sys = 3.16 CPU) @ 15761/s (n=49807) 0a len 3 big: 4s (3.20 usr + 0.04 sys = 3.24 CPU) @ 15838/s (n=51316)

2007-04-17 v1.12 Tels 5258 tests

• require Math::BigInt v1.83, Math::Pari v2.010709
• use native sqrtint() for _sqrt(), making it faster
• use native sqrtn() for root(), making it faster
• support apiversion() 2: add: + fromoct(), asoct() + _10ex(), _alen()

2005-01-02 v1.11 Tels 5055 tests

• require Math::BigInt v1.74 and Math::Pari v2.010602
• testsuite from MBI v1.74
• support API from BigInt v1.70 and better
• add istwo, isten, _two, _ten, logint, _zeros, _sqrt, _root() etc

2004-01-07 v1.10 Tels

• testsuite from MBI v1.68 - 4583 tests
• ashex() and asbin() handle zero, tests for that
• bacmp() is about 11% faster due to _acmp() using gcmp() vs. "<=>"
• added _modinv() and tests

2002-07-07 v1.09 Tels

• testsuite from MBI v1.60 - 4061 tests
• prerequisite fixed
• _copy() uses $zero instead of PARI(0) - benchmark for full BigInt copy(): copy: 6 secs ( 5.16 usr + 0.04 sys = 5.20 CPU) @ 23649.23/s (n=122976) copy: 5 secs ( 5.19 usr + 0.06 sys = 5.25 CPU) @ 27047.05/s (n=141997)
• digit() now uses a method that scales much better when extracting digits counting from the right side ($n > 0)
• likeweise, _inc and _dec use $one instead of PARI(0). Benchmark over Math::BigInt->bdec() and binc() (not just _inc() and _dec()!):

  1.08, 100 and 1000 digit numbers (speed is largely irrel. of number size): dec 100: 4 secs ( 5.11 usr + 0.06 sys = 5.17 CPU) @ 17412.57/s (n=90023) dec 1000: 6 secs ( 5.17 usr + 0.08 sys = 5.25 CPU) @ 17147.24/s (n=90023) inc 100: 5 secs ( 5.05 usr + 0.13 sys = 5.18 CPU) @ 17463.51/s (n=90461) inc 1000: 6 secs ( 5.15 usr + 0.05 sys = 5.20 CPU) @ 17312.12/s (n=90023)

  v1.09, 100 and 1000 digit long numbers: dec 100: 6 secs ( 5.28 usr + 0.02 sys = 5.30 CPU) @ 24095.47/s (n=127706) dec 1000: 4 secs ( 5.00 usr + 0.00 sys = 5.00 CPU) @ 23957.20/s (n=119786) inc 100: 6 secs ( 5.39 usr + 0.00 sys = 5.39 CPU) @ 24157.70/s (n=130210) inc 1000: 5 secs ( 4.93 usr + 0.07 sys = 5.00 CPU) @ 24108.80/s (n=120544)

2002-03-23 v1.08 Tels

• testsuite from MBI v1.55 - 3881 tests
• prerequisite fixed

2002-02-21 v1.07 Tels

• Math::Pari has hexcvt(), which makes from_hex() fast, but doesn't have something for from_bin(). Now frombin() repacks the input as hex number and then passes it to hexcvt().
• tests from Math::BigInt v1.51
• replaced corelib() by config()->{lib}

2002-02-10 v1.06 Tels

• _lsft() and _rsft() do now bases != 2, too
• tests are now from v1.50
• added _fac() for faster factorial() (used from MBI v1.50 onwards)

        #!/usr/bin/perl -w
        use lib 'lib';
        use Math::BigInt lib => 'Pari';
        use Benchmark;
        my $c = 20000;
        my $x = Math::BigInt->new(1); my $z = $x->copy()->blsft($c);
        my $x1 = Math::BigInt->new(1); my $z1 = $x->copy()->blsft($c*10);
        timethese ( -3, {
          right_1_10 => sub { $x->brsft(1,10); },
          left_1_10 => sub { $z->blsft(1,10); },
          right_10_10 => sub { $x1->blsft(10,10); },
          left_10_10 => sub { $z1->brsft(10,10); }, } );

        Results (operations per second) on a 1Ghz Athlon:

                        v1.05   v1.06   Factor
          -------------------------------------
         left_10_10:     1990    6000   3.0
          left_1_10:     2280    5150   2.2
        right_10_10:     1130    1560   1.4
         right_1_10:     2570    6433   2.5

2001-11-11 v1.05 Tels

• fixed _inc() and _dec() so that they can be used by MBI v1.46

2001-11-01 v1.04 Tels

• tests from BigInt v1.45
• added _mod() for faster % operator (about two times faster now)
• added _lsft() and _rsft() for faster shifting

        #!/usr/bin/perl -w
        use lib 'lib';
        use Math::BigInt lib => 'Pari';
        use Benchmark;
        my $c = 20000;
        my $x = Math::BigInt->new(1); my $z = $x->copy()->blsft($c);
        my $x1 = Math::BigInt->new(1); my $z1 = $x->copy()->blsft($c*10);
        timethese ( $c, {
          right_1 => sub { $x->brsft(1,2); },
          left_1 => sub { $z->blsft(1,2); },
          right_10 => sub { $x1->blsft(10,2); },
          left_10 => sub { $z1->brsft(10,2); }, } );

        Results (operations per second) on a 1Ghz Athlon:

                        v1.03   v1.04   Factor
          -------------------------------------
            left_1:      2577    6450   2.5
           left_10:       881    3086   3.5
           right_1:      2928    6990   2.4
          right_10:      2020    5465   2.7

1.03 2001.09.02 btrott

• Incorporated Tels's changes: no longer need to EXPORT_OK functions; added empty import method.
• Added ashex and asbin implementations for Math::Pari integers, to support as_hex and as_bin Math::BigInt methods.

1.02 2001.08.05 btrott

• Added _gcd method; uses PARI library function to return gcd.
• Incorporated Tels's changes: _copy() is now linear; iszero, isone, is_even, and is_odd now use PARI library functions rather than overloaded ops and should thus be faster. Also added new tests from Math::BigInt 1.40.

1.01 2001.07.14 btrott

• Incorporated Tels's changes per calling args to plugin methods (now class is always the first arg, so no more checking). Fixed some tests per Math::BigInt changes.

1.00 2001.07.11 btrott

• Initial release.