1. append() versus &=
- Posted by Lewis Townsend <keroltarr at HOTMAIL.COM>
Sep 12, 2000
-
Last edited Sep 13, 2000
Hello,
Who was it that recently said that append() is faster than &= ?
I have written a benchmark program that seems to prove otherwise.
Let me know if there is a flaw in my program or if I misunderstood
or what.
----- TESTED CODE -------
constant its = 100000
sequence s
atom t
object temp
constant frmt = "%0.03f"
printf (1, "Appending benchmarks for Euphoria using %d iterations\n", its)
puts (1, "\nAppending integers...\n")
puts (1, "append():\t")
s = {}
t=time()
for i = 1 to its do
s = append (s, rand (its))
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "&=:\t\t")
s = {}
t = time ()
for i = 1 to its do
s &= rand (its)
end for
t = time() - t
printf (1, frmt & "\n", t)
constant chars = 20
printf (1, "\nAppending strings (%d characters)...\n", chars)
puts (1, "append():\t")
s = {}
temp = repeat (255,chars)
t=time()
for i = 1 to its do
s = append (s, rand (temp))
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "&=:\t\t")
s = {}
t = time ()
for i = 1 to its do
s &= {rand (temp)}
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "\nAppending complex sequences eg: {a,{a,{a}}}...\n")
puts (1, "append():\t")
s = {}
temp = {its,{its,{its}}}
t=time()
for i = 1 to its do
s = append (s, rand (temp))
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "&=:\t\t")
s = {}
t = time ()
for i = 1 to its do
s &= {rand (temp)}
end for
t = time() - t
printf (1, frmt & "\n", t)
-------END CODE-------
later,
Lewis Townsend
_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.
Share information about yourself, create your own public profile at
http://profiles.msn.com.
2. Re: append() versus &=
- Posted by Robert Craig <rds at ATTCANADA.NET>
Sep 12, 2000
-
Last edited Sep 13, 2000
Lewis Townsend writes:
> Who was it that recently said that append() is faster
> than &= ? I have written a benchmark program that
> seems to prove otherwise. Let me know if there is a flaw
> in my program or if I misunderstood or what.
Try your benchmark again, but this time run each
append() loop *after* the corresponding &= loop.
I think you'll be surprised.
Regards,
Rob Craig
Rapid Deployment Software
http://www.RapidEuphoria.com
3. Re: append() versus &=
Lewis,
I am probably too dense, but Robert's reply was not of much help, as far as
I am concern...
I think your benchmark is completely dominated by the troublesome rand()
function. Take it out of the loop and you will see Rob indeed optimised the
common case:
s1 = append(s1, s2)
I modified your code slightly, improving the timer resolution and bumping
the iterations up too:
include machine.e -- tick_rate()
constant its = 1000000
constant frmt = "%0.03f"
constant chars = 20
sequence s,temp
atom t
integer rits
rits = rand(its)
tick_rate(200)
printf (1, "Appending benchmarks for Euphoria using %d iterations\n", its)
puts (1, "\nAppending integers again...\n")
puts (1, "append():\t")
s = {}
t=time()
for i = 1 to its do
s = append (s, rits)
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "&=:\t\t")
s = {}
t = time ()
for i = 1 to its do
s &= rits
end for
t = time() - t
printf (1, frmt & "\n", t)
printf (1, "\nAppending strings (%d characters)...\n", chars)
puts (1, "append():\t")
s = {}
temp = rand(repeat (255,chars))
t=time()
for i = 1 to its do
s = append (s, temp)
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "&=:\t\t")
s = {}
t = time ()
for i = 1 to its do
s &= {temp}
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "\nAppending complex sequences eg: {a,{a,{a}}}...\n")
puts (1, "append():\t")
s = {}
temp = rand({its,{its,{its}}})
t=time()
for i = 1 to its do
s = append (s, temp)
end for
t = time() - t
printf (1, frmt & "\n", t)
puts (1, "&=:\t\t")
s = {}
t = time ()
for i = 1 to its do
s &= {temp}
end for
t = time() - t
printf (1, frmt & "\n", t)
_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.
Share information about yourself, create your own public profile at
http://profiles.msn.com.
4. Re: append() versus &=
Jiri Babor writes:
> I am probably too dense, but Robert's reply was not of much help,
> as far as I am concern...
In Lewis Townsend's program I found a big difference
in the timing of the last two (more complicated) tests,
depending on whether I moved the append() loop before
or after the corresponding &= loop. It's complicated, but
I think it has to do with the pattern of dynamic storage allocation.
I didn't find this with your modified version.
Regards,
Rob Craig
Rapid Deployment Software
http://www.RapidEuphoria.com
5. Re: append() versus &=
Hello Rob, Jiri
I did try switching the parts arround and it did make a difference.
But either way, append() was still slower on the "complex" append.
>I think your benchmark is completely dominated by the troublesome rand()
>function. Take it out of the loop and you will see Rob indeed optimised the
>common case:
> s1 = append(s1, s2)
I used rand() to add some variation in the data because
otherwise cacheing occurs which speeds up things in a
seemingly unpredictable way. I can't think of any useful
program (off the top of my head) that would simply append
thousands of identical items to each other. Besides, I
don't see how rand() was messing up the numbers since I
always did the same thing in both loops. Perhaps an
assignment and THEN the append would work better.
I am currently
working on 2 projects: LOOP (an OOP lib) and a real-time
game. Both will need all the speed they can get and both
will be operating on complex sequences with plenty of
randomness. I felt that my benchmarks adequately tested
for speed in this situation but perhaps not. I have tried testing
LOOP lib itself with either all append()s or all &=s. Sometimes
one seems faster and sometimes the other. I want to use the one
that's faster MOST of the time but I think your (jiri's) program
allows data caching to invalidate the benchmarks. I'm not sure
WHAT was going on with mine that always gave the second loop an
advantage but I'd like to know how to accurately test for the
speed of random complex data appending. Jiri, I'm sure you can
come up with something. I wouldn't just wait for you to do
all the work but I wont' be able to do any coding until after
this weekend as I'll be away from my comp for 4 days.
later,
Lewis Townsend
_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.
Share information about yourself, create your own public profile at
http://profiles.msn.com.
