RE: [Haskell-cafe] x86 code generation going wrong?
From: Chris Kuklewicz [EMAIL PROTECTED]
To: haskell-cafe@haskell.org
Subject: [Haskell-cafe] x86 code generation going wrong?
Date: Sat, 07 Jan 2006 16:18:59 +
Hello,
I need to ask for some help to test x86 code generation.
There is a factor of two runtime difference between the code I am
benchmarking on my OS X powerbook G4 (ghc 6.4.1) and shootout's speed on
a linux x86 machine (ghc 6.4.1).
Could someone else running on x86 test the three versions pasted below
before I think about submitting another one to the shootout?
Here are the tests on P4 2.4 ghz and athlon 64 3000 linux test1-3 in
respective order
of appearance (note:OPTIONS didn't do anything I have to
compile -O2 -fglasgow-exts explicitely, because I've got compile error for
test3.hs )
[EMAIL PROTECTED] ~/haskell/myhaskell] $ time ./test1 sum-file-test-input
400
real0m3.550s
user0m3.440s
sys 0m0.080s
[EMAIL PROTECTED] ~/haskell/myhaskell] $ time ./test2 sum-file-test-input
400
real0m3.708s
user0m3.660s
sys 0m0.060s
[EMAIL PROTECTED] ~/haskell/myhaskell] $ time ./test3 sum-file-test-input
400
real0m3.678s
user0m3.620s
sys 0m0.050s
This is on athlon64 3000 , linux :
[EMAIL PROTECTED] ~]$ time ./test1 sum-file-test-input
400
real0m5.782s
user0m5.724s
sys 0m0.056s
[EMAIL PROTECTED] ~]$ time ./test2 sum-file-test-input
400
real0m5.953s
user0m5.900s
sys 0m0.052s
[EMAIL PROTECTED] ~]$ time ./test3 sum-file-test-input
400
real0m5.403s
user0m5.332s
sys 0m0.072s
Greetings, Bane.
To compile ghc --make filename.hs -o program
To run cat input-file | time ./program
where to save space, the gzip'd input file is at
http://paradosso.mit.edu/~ckuklewicz/sum-file-test-input.gz
-
-- Original version
{-# OPTIONS -O2 #-}
import Char( ord )
main :: IO ()
main = getContents = print . accP 0 0
accP :: Int - Int - String - Int
accP before this [] = before+this
accP before this ('\n':xs) = accP (before+this) 0xs
accP before this ('-' :xs) = accN before this xs
accP before this (x :xs) = accP before (this*10+ord(x)-ord('0')) xs
accN :: Int - Int - String - Int
accN before this [] = before-this
accN before this ('\n':xs) = accP (before-this) 0xs
accN before this (x :xs) = accN before (this*10+ord(x)-ord('0')) xs
-
-- Faster on G4, 2x slower on x86
{-# OPTIONS -O2 -funbox-strict-fields #-}
import GHC.Base
data I = I !Int
main = print . new (I 0) = getContents
new (I i) [] = i
new (I i) ('-':xs) = neg (I 0) xs
where neg (I n) ('\n':xs) = new (I (i - n)) xs
neg (I n) (x :xs) = neg (I (parse x + (10 * n))) xs
new (I i) (x:xs) = pos (I (parse x)) xs
where pos (I n) ('\n':xs) = new (I (i + n)) xs
pos (I n) (x :xs) = pos (I (parse x + (10 * n))) xs
parse c = ord c - ord '0'
-
-- Explicitly unboxed proposal, faster on G4
{-# OPTIONS -fglasgow-exts -O2 #-}
import GHC.Base
main = print . sumFile = getContents
where sumFile = (\rest - newLine rest 0#)
newLine [] rt = (I# rt)
newLine ('-':rest) rt = negLine rest 0#
where negLine ('\n':rest) soFar = newLine rest (rt -# soFar)
negLine ( x :rest) soFar = negLine rest (d2i x +# (10# *#
soFar))
newLine (x:rest) rt = posLine rest (d2i x)
where posLine ('\n':rest) soFar = newLine rest (rt +# soFar)
posLine ( x :rest) soFar = posLine rest (d2i x +# (10# *#
soFar))
d2i (C# c) = (ord# c) -# z
where z = ord# '0'#
-
Thanks,
Chris
___
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
_
FREE pop-up blocking with the new MSN Toolbar - get it now!
http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/
___
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] x86 code generation going wrong?
Brian Sniffen wrote: The first couldn't even complete on my 2.26 GHz Celeron! It's only got 512 MB of RAM, which may be part of the problem. I should not leak memory but it may be an optimization problem. Try explicitly using ghc -O2 -funbox-strict-fields. Stack space overflow: current size 1048576 bytes. Use `+RTS -Ksize' to increase it. ./test1 sum-file-test-input 25.33s user 2.89s system 18% cpu 2:32.02 total ./test2 sum-file-test-input 4.79s user 0.20s system 94% cpu 5.276 total ./test3 sum-file-test-input 4.46s user 0.14s system 99% cpu 4.623 total -- Brian T. Sniffen [EMAIL PROTECTED]or[EMAIL PROTECTED] http://www.evenmere.org/~bts ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] x86 code generation going wrong?
From: Chris Kuklewicz [EMAIL PROTECTED]
To: [EMAIL PROTECTED], Haskell Cafe haskell-cafe@haskell.org
Subject: Re: [Haskell-cafe] x86 code generation going wrong?
Date: Sun, 08 Jan 2006 20:33:57 +
Brian Sniffen wrote:
The first couldn't even complete on my 2.26 GHz Celeron! It's only got
512 MB of RAM, which may be part of the problem.
I should not leak memory but it may be an optimization problem.
Try explicitly using ghc -O2 -funbox-strict-fields.
On p4. 2.4 ghz 512mb first example takes about 2.5 mb of ram.
I've compiled explicitelly with -O2, because without optimisations it takes
ridicilously large amount of RAM.
I've changed {# OPTIONS to OPTIONS_GHC as Bulat sugested
but still no effect. Options have to be specified on command line in
order to work.
Greetings, Bane.
_
Express yourself instantly with MSN Messenger! Download today it's FREE!
http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/
___
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] x86 code generation going wrong?
bmaxa:
From: Chris Kuklewicz [EMAIL PROTECTED]
To: [EMAIL PROTECTED], Haskell Cafe haskell-cafe@haskell.org
Subject: Re: [Haskell-cafe] x86 code generation going wrong?
Date: Sun, 08 Jan 2006 20:33:57 +
Brian Sniffen wrote:
The first couldn't even complete on my 2.26 GHz Celeron! It's only got
512 MB of RAM, which may be part of the problem.
I should not leak memory but it may be an optimization problem.
Try explicitly using ghc -O2 -funbox-strict-fields.
On p4. 2.4 ghz 512mb first example takes about 2.5 mb of ram.
I've compiled explicitelly with -O2, because without optimisations it takes
ridicilously large amount of RAM.
I've changed {# OPTIONS to OPTIONS_GHC as Bulat sugested
but still no effect. Options have to be specified on command line in
order to work.
Ensure that the {-# OPTIONS ... #-} lines is the *first* line of the
file, and that no comments precede it.
-- Don
___
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] x86 code generation going wrong?
From: [EMAIL PROTECTED] (Donald Bruce Stewart)
To: Branimir Maksimovic [EMAIL PROTECTED]
CC: [EMAIL PROTECTED],
[EMAIL PROTECTED],haskell-cafe@haskell.org
Subject: Re: [Haskell-cafe] x86 code generation going wrong?
Date: Mon, 9 Jan 2006 11:15:51 +1100
bmaxa:
From: Chris Kuklewicz [EMAIL PROTECTED]
To: [EMAIL PROTECTED], Haskell Cafe haskell-cafe@haskell.org
Subject: Re: [Haskell-cafe] x86 code generation going wrong?
Date: Sun, 08 Jan 2006 20:33:57 +
Brian Sniffen wrote:
The first couldn't even complete on my 2.26 GHz Celeron! It's only
got
512 MB of RAM, which may be part of the problem.
I should not leak memory but it may be an optimization problem.
Try explicitly using ghc -O2 -funbox-strict-fields.
On p4. 2.4 ghz 512mb first example takes about 2.5 mb of ram.
I've compiled explicitelly with -O2, because without optimisations it
takes
ridicilously large amount of RAM.
I've changed {# OPTIONS to OPTIONS_GHC as Bulat sugested
but still no effect. Options have to be specified on command line in
order to work.
Ensure that the {-# OPTIONS ... #-} lines is the *first* line of the
file, and that no comments precede it.
Aaah, I didn't knew that. Now this works, thanks!
Greetings, Bane.
_
FREE pop-up blocking with the new MSN Toolbar - get it now!
http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/
___
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
