2. Re: Inconsistent exw.exe (was RE: Looking for the best "JOIN")
Chris Bensler writes:
> ...Third, I discovered that I would get extremely different
> results using exw.exe, depending if a variable was
> declared before the functions or not.
> This shouldn't be!?
> In DOS, there is no difference.
> You might also note that Derek's routine was unaffected.
I wasn't able to duplicate your results exactly, but I did
see some large variability in timing when I tried
various other things that shouldn't affect speed.
About a year ago I thoroughly analyzed a benchmark
very similar to yours. I noticed that when you have
a tight loop that creates and destroys small sequences,
many many times (only a benchmark would do this), then
the performance can be strongly affected by trivial changes in
the heap, such as declaring an extra variable.
I put some changes into 2.3 as a result, but I don't think
"real" programs are likely to be affected by this.
If you have a wider variety of sequence lengths, and if you actually
use the sequences for some purpose, then the performance
should be pretty stable.
Regards,
Rob Craig
Rapid Deployment Software
http://www.RapidEuphoria.com
3. Re: Inconsistent exw.exe (was RE: Looking for the best "JOIN")
Ýòî ñîîáùåíèå â ôîðìàòå MIME ñîñòîèò èç íåñêîëüêèõ ÷àñòåé.
------=_NextPart_000_01C15C29.6737A8C0
4. Re: Inconsistent exw.exe (was RE: Looking for the best "JOIN")
- Posted by rforno at tutopia.com
Oct 23, 2001
This is a multi-part message in MIME format.
------=_NextPart_000_0074_01C15C13.9FF126C0
charset="iso-8859-1"
I slightly modified Derek's algorithm and made it marginally faster. I am
attaching the modified TEST.EX and TEST2.EX files. But what really surprised
me is the big
difference in speed between EX and EXW. EXW is many times faster than EX, at
least for these examples! Is there an explanation, Rob?
Just to complicate the issue, I am attaching also a TEST3.EX file that deals
with randomly varying lengths of sequences.
----- Original Message -----
From: <bensler at mail.com>
To: "EUforum" <EUforum at topica.com>
Sent: Monday, October 22, 2001 11:49 PM
Subject: Inconsistent exw.exe (was RE: Looking for the best "JOIN")
>
> First of all, I found an algorithm that is slightly faster than Jiri's,
just barely.
>
> Second, doing some bench testing, I found that Derek's is the fastest of
the three when run in a DOS box. Jiri's suffers
> most.
>
> Third, I discovered that I would get extremely different results using
exw.exe, depending if a variable was declared before
> the functions or not.
> This shouldn't be!?
> In DOS, there is no difference.
> You might also note that Derek's routine was unaffected.
>
> I've attached my bench tests(which includes my version of join). TEST.EX
and TEST2.EX are identical except TEST2.EX
> declares a variable at the top of the file.
>
> Here are my bench results.... (iterations per second)
>
> TEST.EX | TEST2.EX
> ------------|--------------
> WIN | WIN
> ------------|--------------
> 318229 | 219441 -- my version
> 210934 | 212506 -- Derek's
> 305709 | 186280 -- Jiri's
> ------------|--------------
> DOS | DOS
> ------------|--------------
> 53299 | 54770
> 56681 | 56467
> 52735 | 52475
> ------------|--------------
>
> Chris
>
>
>
------=_NextPart_000_0074_01C15C13.9FF126C0
Content-Type: application/octet-stream;
name="test2.ex"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
filename="test2.ex"
integer junk
-------------------------------------------------------------------------=
--------
-- FUNCTIONS --
function join(sequence s)
sequence r
if length(s)>1 then
r=3Ds[1]
for i =3D 2 to length(s) do
r &=3Ds[i]
end for
return r
elsif length(s) then
return s[1]
end if
return {}
end function
function joinStrings(sequence s)
integer l,a,b
sequence t
=20
if length(s)=3D0 then return s end if
=20
l=3D0
for i =3D 1 to length(s) do
l+=3Dlength(s[i])
end for
t =3Drepeat(0,l)
a=3D1
for i =3D 1 to length(s) do
b =3D a + length(s[i]) - 1
t[a..b] =3D s[i]
a =3D b+1
end for
return t
end function
function join2(sequence s)
sequence r
r=3D{}
for i =3D 1 to length(s) do
r &=3Ds[i]
end for
return r
end function
function join3(sequence s)
integer l,a,b
sequence t
l =3D 0
for i =3D 1 to length(s) do
l +=3Dlength(s[i])
end for
t =3D repeat(0,l)
a =3D 0
for i =3D 1 to length(s) do
b =3D a + length(s[i])
t[a + 1..b] =3D s[i]
a =3D b
end for
return t
end function
-------------------------------------------------------------------------=
--------
-- TEST CODE --
constant iteration_time=3D10
constant words1=3D{"abc","defghi","jklmnopqr","","st","u","vwxyz"}
printf(1,"bench time: =
%d:%02d\n",{floor(iteration_time*4/60),remainder(iteration_time*4,60)})
atom e
sequence k
integer x
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin(words1)
end while
printf(1,"join() <new> : %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3DjoinStrings(words1)
end while
printf(1,"joinStrings() <Derek's>: %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin2(words1)
end while
printf(1,"join2() <Jiri's> : %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin3(words1)
end while
printf(1,"join3() <Forno's>: %d iterations per =
second\n",x/iteration_time)
position(6,28)
puts(1,"PRESS A KEY")
while get_key()=3D-1 do end while
------=_NextPart_000_0074_01C15C13.9FF126C0
Content-Type: application/octet-stream;
name="test.ex"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
filename="test.ex"
-------------------------------------------------------------------------=
--------
-- FUNCTIONS --
function join(sequence s)
sequence r
if length(s)>1 then
r=3Ds[1]
for i =3D 2 to length(s) do
r &=3Ds[i]
end for
return r
elsif length(s) then
return s[1]
end if
return {}
end function
function joinStrings(sequence s)
integer l,a,b
sequence t
=20
if length(s)=3D0 then return s end if
=20
l=3D0
for i =3D 1 to length(s) do
l+=3Dlength(s[i])
end for
t =3Drepeat(0,l)
a=3D1
for i =3D 1 to length(s) do
b =3D a + length(s[i]) - 1
t[a..b] =3D s[i]
a =3D b+1
end for
return t
end function
function join2(sequence s)
sequence r
r=3D{}
for i =3D 1 to length(s) do
r &=3Ds[i]
end for
return r
end function
function join3(sequence s)
integer l,a,b
sequence t
l =3D 0
for i =3D 1 to length(s) do
l +=3Dlength(s[i])
end for
t =3D repeat(0,l)
a =3D 0
for i =3D 1 to length(s) do
b =3D a + length(s[i])
t[a + 1..b] =3D s[i]
a =3D b
end for
return t
end function
-------------------------------------------------------------------------=
--------
-- TEST CODE --
constant iteration_time=3D10
constant words1=3D{"abc","defghi","jklmnopqr","","st","u","vwxyz"}
printf(1,"bench time: =
%d:%02d\n",{floor(iteration_time*4/60),remainder(iteration_time*4,60)})
atom e
sequence k
integer x
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin(words1)
end while
printf(1,"join() <new> : %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3DjoinStrings(words1)
end while
printf(1,"joinStrings() <Derek's>: %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin2(words1)
end while
printf(1,"join2() <Jiri's> : %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin3(words1)
end while
printf(1,"join3() <Forno's>: %d iterations per =
second\n",x/iteration_time)
position(6,28)
puts(1,"PRESS A KEY")
while get_key()=3D-1 do end while
------=_NextPart_000_0074_01C15C13.9FF126C0
Content-Type: application/octet-stream;
name="test3.ex"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
filename="test3.ex"
integer junk
-------------------------------------------------------------------------=
--------
-- FUNCTIONS --
function join(sequence s)
sequence r
if length(s)>1 then
r=3Ds[1]
for i =3D 2 to length(s) do
r &=3Ds[i]
end for
return r
elsif length(s) then
return s[1]
end if
return {}
end function
function joinStrings(sequence s)
integer l,a,b
sequence t
=20
if length(s)=3D0 then return s end if
=20
l=3D0
for i =3D 1 to length(s) do
l+=3Dlength(s[i])
end for
t =3Drepeat(0,l)
a=3D1
for i =3D 1 to length(s) do
b =3D a + length(s[i]) - 1
t[a..b] =3D s[i]
a =3D b+1
end for
return t
end function
function join2(sequence s)
sequence r
r=3D{}
for i =3D 1 to length(s) do
r &=3Ds[i]
end for
return r
end function
function join3(sequence s)
integer l,a,b
sequence t
l =3D 0
for i =3D 1 to length(s) do
l +=3Dlength(s[i])
end for
t =3D repeat(0,l)
a =3D 0
for i =3D 1 to length(s) do
b =3D a + length(s[i])
t[a + 1..b] =3D s[i]
a =3D b
end for
return t
end function
-------------------------------------------------------------------------=
--------
-- TEST CODE --
constant iteration_time=3D10
sequence words1, w
words1 =3D {}
for i =3D 1 to rand(100) do
w =3D {}
for j =3D 1 to rand(100) do
w &=3D rand(256) - 1
end for
words1 =3D append(words1, w)
end for
printf(1,"bench time: =
%d:%02d\n",{floor(iteration_time*4/60),remainder(iteration_time*4,60)})
atom e
sequence k
integer x
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin(words1)
end while
printf(1,"join() <new> : %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3DjoinStrings(words1)
end while
printf(1,"joinStrings() <Derek's>: %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin2(words1)
end while
printf(1,"join2() <Jiri's> : %d iterations per =
second\n",x/iteration_time)
x=3D0
e=3Dtime()+iteration_time
while e >=3D time() do
x +=3D1
k =3Djoin3(words1)
end while
printf(1,"join3() <Forno's>: %d iterations per =
second\n",x/iteration_time)
position(6,28)
puts(1,"PRESS A KEY")
while get_key()=3D-1 do end while
------=_NextPart_000_0074_01C15C13.9FF126C0--
5. Re: Inconsistent exw.exe (was RE: Looking for the best "JOIN")
rforno writes:
> EXW is many times faster than EX, at
> least for these examples! Is there an explanation, Rob?
Yes, I already explained it.
exw is not faster. It's just luckier.
Under the right conditions ex will be just as fast.
For instance if I turn on time profiling, ex suddenly
becomes just as fast as exw. Your results may vary.
It just depends on a subtle property of the heap.
When you buy the source, you'll be able to see
what I'm talking about 
Remember - this phenomenon happens because
this is an artificial program that does nothing useful.
It just creates and destroys small sequences
over and over again in a tight loop.
Regards,
Rob Craig
Rapid Deployment Software
http://www.RapidEuphoria.com
