Lua 5.1.4: 2.95s
JRuby 1.1.7 (trunk)—fast—server: 4.93s
Python 2.6.1: 7.92s
PHP 5.2.8: 8.20s
Ruby EE 1.8.6: 10.08s

platform: Mac OS-X 10.5.6, PowerMac 2×2.8GHz

from time import clock

start = clock()
from math import sqrt

results = [(sqrt(c*c – b*b),b,c) for c in xrange(2,5001) for b in xrange(1,c) if sqrt(c*c – b*b) %1 0.0]

print clock() - start

And if I wanted to cheat and beat everything, including Fortran, in the speed department:

from time import clock

start = clock()
from math import sqrt

results = ((sqrt(c*c - b*b),b,c) for c in xrange(2,5001) for b in xrange(1,c) if sqrt(c*c - b*b) %1 0.0)

print clock() – start

This is going to be rather longish and slightly offtopic…

Here’s some code from a Russian site, http://rsdn.ru/forum/message/2515740.1.aspx (you may llok there to see pretty printed code):

Original author is not me

[start quote]
Haskell’s GHC compiler is pretty bad when it comes to double-precision floating point arythmetic (or maybe I’m doing something wrong)

If we directly rewrite the test:

pythags :: Double -> Int
pythags mp = pyth 2 0
where
pyth c i | c > mp = i
| otherwise = pyth (c+1) (i + (pyth’ 1 0))
where
pyth’ b j | b c = j
| otherwise = let a = sqrt (c*c - b*b)
in if a fromIntegral ((truncate a)::Int)
then pyth’ (b+1) (j+1)
else pyth’ (b+1) j

or like this:

pythags :: Double -> Int
pythags mp = length [ () | c

then it will be approx. 40 times slower than VisualC or 25 times slower than GCC:

int pyphags(double mp)
{
double c, b;
int i = 0;
double a;

for (c = 2; c

If we, however, rewrite the test using integers only:

pythags :: Int -> Int
pythags mp = length [ () | c

then the difference between GHC and GCC will be less than that between VisualC and GCC:

int pyphags(int mp)
{
int a, b, c, s, i=0;

for (c = 2; c

When mp = 5000 my computer gives:

Double Int
VisualC 8.0 0.9 sec 0.7 sec
GCC mingw 3.4.5 1.4 sec 1.0 sec
GHC 6.6 34 sec 1.3 sec

Note. GCC 6.2 that comes with GHC 6.6 (Win32) has a slow implementation of sqrt in math.h, so I defined fastDoubleSqrt and fastIntSqrt for GCC as follows:

__inline double fastDoubleSqrt(double x)
{
register double res;
__asm volatile (“fsqrt” : ”=t” (res) : “0” (x));
return res;
}

__inline int fastIntSqrt(int x)
{
return (int)fastDoubleSqrt((double)x);
}

For VisualC (it’s ok in this department):

inline double fastDoubleSqrt(double x)
{
return sqrt(x);
}

inline int fastIntSqrt(double x)
{
return ((int)sqrt((double)x);
}

[end quote]

Here’s some code from a Russian site, http://rsdn.ru/forum/message/2515740.1.aspx (you may llok there to see pretty printed code):

Original author is not me

[start quote]
Haskell’s GHC compiler is pretty bad when it comes to double-precision floating point arythmetic (or maybe I’m doing something wrong)

If we directly rewrite the test:

pythags :: Double -> Int
pythags mp = pyth 2 0
where
pyth c i | c > mp = i
| otherwise = pyth (c+1) (i + (pyth’ 1 0))
where
pyth’ b j | b c = j
| otherwise = let a = sqrt (c*c - b*b)
in if a fromIntegral ((truncate a)::Int)
then pyth’ (b+1) (j+1)
else pyth’ (b+1) j

or like this:

pythags :: Double -> Int
pythags mp = length [ () | c

then it will be approx. 40 times slower than VisualC or 25 times slower than GCC:

int pyphags(double mp)
{
double c, b;
int i = 0;
double a;

for (c = 2; c

If we, however, rewrite the test using integers only:

pythags :: Int -> Int
pythags mp = length [ () | c

then the difference between GHC and GCC will be less than that between VisualC and GCC:

int pyphags(int mp)
{
int a, b, c, s, i=0;

for (c = 2; c

When mp = 5000 my computer gives:

Double Int
VisualC 8.0 0.9 sec 0.7 sec
GCC mingw 3.4.5 1.4 sec 1.0 sec
GHC 6.6 34 sec 1.3 sec

Note. GCC 6.2 that comes with GHC 6.6 (Win32) has a slow implementation of sqrt in math.h, so I defined fastDoubleSqrt and fastIntSqrt for GCC as follows:

__inline double fastDoubleSqrt(double x)
{
register double res;
__asm volatile (“fsqrt” : ”=t” (res) : “0” (x));
return res;
}

__inline int fastIntSqrt(int x)
{
return (int)fastDoubleSqrt((double)x);
}

For VisualC (it’s ok in this department):

inline double fastDoubleSqrt(double x)
{
return sqrt(x);
}

inline int fastIntSqrt(double x)
{
return ((int)sqrt((double)x);
}

[end quote]

 tags and it's still lost the indentation.

OK - the line starting 'where a =' should be indented (by a tab-stop, or the equivalent in spaces), and the next line I have indented so that the start of 'diff =' lines up with the start of 'a ='.

module Main where

isPythag :: Int -> Int <del>> Bool
isPythag c b =
   (a*a)==diff
   where a = ratSqrt (diff)
         diff = (c*c)</del>(b*b)

ratSqrt :: Int <del>> Int
ratSqrt = truncate.sqrt.fromIntegral

main = (putStrLn.show.length) [ 1 | c<</del>[2..5000],b<-[1..c-1], isPythag c b ]

Put the code into a file called a.hs and compile with ‘ghc -O2 a.hs -o a’

main = (putStrLn.show.length) [ 1 | c< -[2..5000],b<-[1..c-1], isPythag c b ]