OpenEuphoria

1. Comparison of Euphoria vs Perl, Python, PHP

• Posted by c.k.lester <euphoric at cklester.c??> Jun 11, 2008
• 855 views
• Last edited Jun 12, 2008

Can anybody create a Euphoria program based on the Perl/Python/PHP version
found on this page:
http://www.skitoy.com/p/performance-of-python-php-and-perl/160

It would be interesting to see how Euphoria fares against them these days.

3. Re: Comparison of Euphoria vs Perl, Python, PHP

• Posted by c.k.lester <euphoric at ckl?ster.?om> Jun 11, 2008
• 785 views
• Last edited Jun 12, 2008

Bernie Ryan wrote:
> c.k.lester wrote:
> > Can anybody create a Euphoria program based on the Perl/Python/PHP version
> > found on this page: <a
> > href="http://www.skitoy.com/p/performance-of-python-php-and-perl/160">http://www.skitoy.com/p/performance-of-python-php-and-perl/160</a>
> > It would be interesting to see how Euphoria fares against them these days.
> CK: 
> I wrote a program in Euphoria that was so fast that it ran before I
> could finish it.

HAHA! :P Video or it didn't happen. :D

5. Re: Comparison of Euphoria vs Perl, Python, PHP

• Posted by Jeremy Cowgar <jeremy at cowga?.?om> Jun 11, 2008
• 805 views
• Last edited Jun 12, 2008

CK, I benchmarked PHP as well. Installed it just for you against my better
judgment 

PHP: 11.940s

--
Jeremy Cowgar
http://jeremy.cowgar.com

7. Re: Comparison of Euphoria vs Perl, Python, PHP

• Posted by Matt Lewis <matthewwalkerlewis at gma?l.c?m> Jun 12, 2008
• 814 views

c.k.lester wrote:
> 
> Can anybody create a Euphoria program based on the Perl/Python/PHP version
> found on this page: <a
> href="http://www.skitoy.com/p/performance-of-python-php-and-perl/160">http://www.skitoy.com/p/performance-of-python-php-and-perl/160</a>
> 
> It would be interesting to see how Euphoria fares against them these days.

Not so good.  Now, I didn't do anything special to try to optimize.  I 
just used map.e and regex.e.  I suspect that some of the slow down may
be the conversion that has to happen between a euphoria sequence, and
the C-string before it gets passed to PCRE.  Also, the native perl hash
and python dictionaries probably give them an edge.
 
Here were my results:

perl:
real    0m26.361s
user    0m25.726s
sys     0m0.580s

python:
real    0m38.613s
user    0m37.818s
sys     0m0.760s

euphoria:
real    1m29.448s
user    1m28.722s
sys     0m0.620s

translated euphoria:
real    0m55.930s
user    0m55.231s
sys     0m0.636s

Here's the code:

#!/usr/bin/exu

include regex.e as re
include map.e as map

integer in
sequence cmd
cmd = command_line()
in = open( cmd[3], "r" )

map:map first
first = map:new(1000)

integer FULL
FULL = open( "full.txt", "w" )

regex MULTI_TOKEN, SINGLE_TOKEN

MULTI_TOKEN  = re:new( "^__MULTI_TOKEN__\\s+(\\S+)\\s+(.*)\\t?\\s*(\\d+)\\s*$" )
SINGLE_TOKEN = re:new( "^__SINGLE_TOKEN__\\s+(\\S+)\\s*\\t?\\s*(\\d+)\\s*$" )

object line
object matches
object one, two, three, three_plus

while sequence( line ) entry do
	line = trim_tail( line )
	matches = re:search( MULTI_TOKEN, line )
	if sequence( matches ) then
		one   = line[matches[2][1]..matches[2][2]]
		two   = line[matches[3][1]..matches[3][2]]
		three = value(line[matches[4][1]..matches[4][2]])
		three = three[2]
		three_plus = three + map:get( first, one, 0 )
		
		first = map:put( first, one, three_plus )
		printf( FULL, "%s %s\t%d\n", {one, two, three})
		
	else
		matches = re:search( SINGLE_TOKEN, line )
		if sequence( matches ) then
			one   = line[matches[2][1]..matches[2][2]]
			two   = value(line[matches[3][1]..matches[3][2]])
			first = map:put( first, one, map:get( first, one, 0 ) + two[2] )
			
		else
			printf( 1, "Unknown: {%s}\n", {line} )
		end if
	end if
entry
	line = gets(in)
end while

close( FULL )
close( in )

integer FIRST

FIRST = open( "first.txt", "w" )
sequence keys
keys = map:keys( first )

for i = 1 to length( keys ) do
	printf( FIRST, "%s\t%d\n", {keys[i], map:get( first, keys[i], 0)})
end for
close( FIRST )

And here's how I generated the test data:

include machine.e

set_rand( 271828183 )
function make_word()
	integer len
	sequence word
	len = rand( 20 ) + 1
	word = ""
	for i = 1 to len do
		word &= 'a' + rand(26) - 1
	end for
	return word
end function
sequence words
words = repeat( {}, 1000 )
for k = 1 to 1000 do
	words[k] = make_word()
end for

function get_word()
	return words[rand(1000)]
end function

integer fn
fn = open( "line_test.txt", "w" )
for i = 1 to 5000000 do
	if rand(2) = 1 then
		-- multi token
printf( fn, "__MULTI_TOKEN__ %s %s %s\t%d\n", {get_word(), get_word(),
get_word(), rand(20)} )
		
	else
		-- single token
		printf( fn, "__SINGLE_TOKEN__ %s \t%d\n", {get_word(), rand(20)} )
		
	end if
end for
close( fn )

9. Re: Comparison of Euphoria vs Perl, Python, PHP

• Posted by Jeremy Cowgar <jeremy at cowg??.com> Jun 12, 2008
• 787 views

A very dumb check of the hash implementations between Python and Euphoria:

count = 100000
dict = {}
for i in range(count):
	dict[str(i)] = i
for i in range(count):
	c = dict[str(i)]
print(c)

--include sequence.e
include map.e as map

integer c
map:map m
constant count = 100000
m = map:new()

for i = 1 to count do
	m = map:put(m, sprint(i), i)
end for

for i = 1 to count do
	c = map:get(m, sprint(i), 0)
end for

? c

[jeremy@jdesk ~]$ time python hash.py 
99999

real    0m0.233s
user    0m0.220s
sys     0m0.010s

[jeremy@jdesk ~]$ time exu hash.ex 
100000

real    0m1.852s
user    0m1.850s
sys     0m0.003s

[jeremy at jdesk ~]$ time ./hash
100000

real    0m1.573s
user    0m1.563s
sys     0m0.010s

--
Jeremy Cowgar
http://jeremy.cowgar.com

11. Re: Comparison of Euphoria vs Perl, Python, PHP

• Posted by Jeremy Cowgar <jeremy at c?w?ar.com> Jun 12, 2008
• 778 views

Jeremy Cowgar wrote:
> 
> }}}
<eucode>
> --include sequence.e
> include map.e as map
> 
> integer c
> map:map m
> constant count = 100000
> m = map:new()
> 
> for i = 1 to count do
> 	m = map:put(m, sprint(i), i)
> end for
> 
> for i = 1 to count do
> 	c = map:get(m, sprint(i), 0)
> end for
> 
> ? c
> </eucode>
{{{

> 

Oh, I forgot to show the difference if not using sprint(i) but i directly.

[jeremy@jdesk ~]$ time exu hash.ex 
100000

real    0m1.010s
user    0m1.003s
sys     0m0.003s

vs.

[jeremy@jdesk ~]$ time exu hash.ex 
100000

real    0m1.852s
user    0m1.850s
sys     0m0.003s

That's quite a difference!

--
Jeremy Cowgar
http://jeremy.cowgar.com

13. Re: Comparison of Euphoria vs Perl, Python, PHP

• Posted by Jeremy Cowgar <jeremy at cowgar?co?> Jun 12, 2008
• 782 views

c.k.lester wrote:
> 
> Matt Lewis wrote:
> > c.k.lester wrote:
> > > Can anybody create a Euphoria program based on the Perl/Python/PHP version
> > > found on this page: <a
> > > href="http://www.skitoy.com/p/performance-of-python-php-and-perl/160">http://www.skitoy.com/p/performance-of-python-php-and-perl/160</a>
> > > It would be interesting to see how Euphoria fares against them these days.
> > Not so good.
> 
> Well, hopefully the devs will take this as a mandate to improve Euphoria core
> for speed in this domain.
> 
> Or no?

It gives us something to look at, but bear in mind this not new to 4.0, we have
lived with this probably since Euphoria's conception.

--
Jeremy Cowgar
http://jeremy.cowgar.com

15. Re: Comparison of Euphoria vs Perl, Python, PHP

• Posted by Jeremy Cowgar <jeremy at ?owgar?com> Jun 12, 2008
• 788 views

c.k.lester wrote:
>  
> We lived without GOTO also. And optional parameters. But those things 
> are said to improve Euphoria. I think a speed boost in core where we are
> competitive in the text processing domain is even better.
> 
> Of course, Rob maybe have tweaked the heck outta core and we're about 
> as fast as we'll ever be. But I hope not. :)

Yes, I would say the core is as fast as it is ever going to be. Now the speed
improvement that would be required to make this one test (again it's just 1 test)
shine would be to move hash tables into C code, but, many will think that is
bloating the interpreter. It's a balancing act.

For this one test, I can show you 100 others that we are faster. There is a
common belief that benchmarking is evil. Also, Matt did say that he did not try
to optimize anything. Maybe before we get too excited about making mass changes
we should concentrate on making what we have work.

goto has been in the feature request bin for years. optional parameters makes
coding much easier and less functions to learn and use. We *are* a volunteer
staff. If you would like to concentrate on making this faster, you can do so. You
are a dev on the Euphoria project and the major problem right now seems that map
is not as fast as Python's, so it's Euphoria code that needs updated. Jump in!

--
Jeremy Cowgar
http://jeremy.cowgar.com