[Mingw-users] GCC compiled code very slow compared to MS Visual C
Jeremy Hall
I wonder if there is any obvious thing I should be doing here?
Simple benchmarks show that the same program compiled with MS VC 2003
or MS VC 2010
are about twice as fast as the program compiled with GCC.
I find this strange as GCC clearly produces better quality code. I
have compared the assembler
output from both compilers.
Even when compiled with -O3 the run is slower with GCC.
What am I missing ?
Regards
Jeremy
------------------------------------------------------------------------------
Protect Your Site and Customers from Malware Attacks
Learn about various malware tactics and how to avoid them. Understand
malware threats, the impact they can have on your business, and how you
can protect your company and customers by using code signing.
http://p.sf.net/sfu/oracle-sfdevnl
_______________________________________________
MinGW-users mailing list
MinGW...@lists.sourceforge.net
This list observes the Etiquette found at
http://www.mingw.org/Mailing_Lists.
We ask that you be polite and do the same. Disregard for the list etiquette may cause your account to be moderated.
_______________________________________________
You may change your MinGW Account Options or unsubscribe at:
https://lists.sourceforge.net/lists/listinfo/mingw-users
Also: mailto:mingw-use...@lists.sourceforge.net?subject=unsubscribe
Tor Lillqvist
Telling us more details what the code does?
--tml
Jeremy Hall
Its a console text editor that supports a scripting language.
The editor is about 11,250 lines of C and is self contained on Windows
(doesn't use any extra libraries). Its been heavily optimized over
many years.
The "benchmark" is a (complex) regression test script reports the
elapsed time at the end. It does a great deal
of work including a lot of arithmetic and memory access, but very
little disk access.
I realise its too big an example for specific suggestions. I'll try
and create a small example with simple compiler
options for both compilers. I was really wondering if anyone else had
seen this and if so what were their conclusions?
FYI, the timings I had were (Windows XP SP3, on 1.6GHz Athlon XP, 1GB) :
MinGW GCC 4.5.2 -Os 173 seconds
MinGW GCC 4.5.2 -O3 105 seconds
MS Visual C++ 2010 -O1s 89 seconds
MS Visual C++ 2010 -O2t 87 seconds
The same script on a 32 bit Linux box (GCC) runs a little faster (143
secs -Os), but still not as fast as VC++.
The Linux 64 bit version on a 4GHz Core i7 executes in 20 seconds :-)
The VS 2010 compile looks like this:
D:\ge>cl -DFILE_TABS=4 -DSCREEN_TABS=2 /O1isyb2 /J /fp:fast /GFrS-
/Qfast_transcendentals /arch:SSE /Zp4 /W2
g.c /Feg.exe /FAs
Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 16.00.30319.01 for 80x86
The MinGW GCC (4.5.2) options are:-
gcc -pipe -std=c99 -funsigned-char -DFILE_TABS=4 -DSCREEN_TABS=2 g.c
-Os -Wall -Wextra -Wno-char-subscripts
-ffast-math -mfpmath=387 -mpc80 -mhard-float -s -fno-ident
-fshort-enums -fwhole-program -minline-all-stringops
-fmerge-all-constants -fomit-frame-pointer -fno-non-call-exceptions
-fno-asynchronous-unwind-tables -fno-exceptions
-fno-unwind-tables -fira-loop-pressure -fstrict-aliasing
-fstrict-overflow -march=athlon-xp -mtune=athlon-xp -o g
Regards
Jeremy
Earnie
>
> The same script on a 32 bit Linux box (GCC) runs a little faster (143
> secs -Os), but still not as fast as VC++.
> The Linux 64 bit version on a 4GHz Core i7 executes in 20 seconds :-)
>
Everything runs faster on Linux so if your goal is fastest speed of
execution your choice would be to use the OS that runs faster. As for
Windows OS, I do not know the reasons for the differences.
--
Earnie
-- http://www.for-my-kids.com
Ross Ridge
> gcc [...] -mfpmath=387 -mpc80 [...]
I'm not sure if MinGW actually supports the -mpc80 option, but Visual C++
doesn't support 80-bit floating point arithmetic. It puts the processor
in 64-bit rounding mode by default and long doubles are only 64-bits.
If you're using long doubles then that could explain some of the
performace differnce. You're also using 387 math with GCC and it looks
like VC++ is using faster SSE math.
Anyways, 105 seconds versus 87 seconds seems like a reasonable difference
in performance between the two compilers.
Ross Ridge
Jeremy Hall
I only use doubles, not long doubles. I thought -mpc80 caused 80bit
intermediate arithmetic to be used with the
387; which is the default for Linux GCC anyway,
I tried removing SSE from the VC build and it did slow down the VC run
a little, but only a few percent.
Interesting!
Really, I was surprised GCC didn't produce faster code given the
effort that lots of very clever people
put into the optimizer, and you can see, in general, the assembler
produced is much more sophisticated.
MS hardly seem to develop VC, it doesn't even support C99 after 11 years,
Still, now I have started using MinGW, its so nice to be using the
same compiler on all platforms, have decent warnings, have C99 support
etc.
------------------------------------------------------------------------------
Special Offer-- Download ArcSight Logger for FREE (a $49 USD value)!
Finally, a world-class log management solution at an even better price-free!
Download using promo code Free_Logger_4_Dev2Dev. Offer expires
February 28th, so secure your free ArcSight Logger TODAY!
http://p.sf.net/sfu/arcsight-sfd2d
Chris Wilson
On Sun, 23 Jan 2011, Jeremy Hall wrote:
>> MinGW GCC 4.5.2 -Os 173 seconds
>> MinGW GCC 4.5.2 -O3 105 seconds
>> MS Visual C++ 2010 -O1s 89 seconds
>> MS Visual C++ 2010 -O2t 87 seconds
It's well known that GCC's -Os produces *much* slower code in return for
reducing executable size, so if you want to compare the highest
performances of GCC and VC then I would ignore -Os completely.
> Really, I was surprised GCC didn't produce faster code given the effort
> that lots of very clever people put into the optimizer, and you can see,
> in general, the assembler produced is much more sophisticated.
It may be because it's an intermediate language compiler that supports
several source languages other than C++, and several target architectures
other than i386. This may reduce the number of optimisations that it can
apply. It is also mainly developed by volunteers, whereas:
> MS hardly seem to develop VC, it doesn't even support C99 after 11 years,
Windows and most of its software (MS Office, IE, etc) have always been
compiled with VC or its ancestors, and being a commercial product, Windows
always had to be fast. They could make their programmers write bizarre
contortions to get through the compiler, but if that compiler produced
awful code then Windows would suffer in the marketplace.
So they had a business incentive to spend a lot of money on making the
compiler produce really fast code (maybe a long time ago), but they never
really had an incentive to add language features.
It would still be very interesting if you could reduce the problem to a
test case and compare the assembler code generated by the two compilers.
If GCC can be improved then all x86 Linuxes and FreeBSD would benefit as
well, and many people would thank you :)
Cheers, Chris.
--
_____ __ _
\ __/ / ,__(_)_ | Chris Wilson <chri...@qwirx.com> Cambs UK |
/ (_/ ,\/ _/ /_ \ | Security/C/C++/Java/Ruby/Perl/SQL Developer |
\__/_/_/_//_/___/ | We are GNU : free your mind & your software |
Bruce Cran
Chris Wilson <ch...@qwirx.com> wrote:
> So they had a business incentive to spend a lot of money on making
> the compiler produce really fast code (maybe a long time ago), but
> they never really had an incentive to add language features.
I think Microsoft haven't implemented C99 because most customers build
C code using the C++ compiler since it's seen as a "better C", so
there's very little demand now for anything other than C++0x.
http://connect.microsoft.com/VisualStudio/feedback/details/531634/support-iso-c99-standard
has a response from Microsoft.
--
Bruce Cran
Reini Urban
> I only use doubles, not long doubles. I thought -mpc80 caused 80bit
> intermediate arithmetic to be used with the
> 387; which is the default for Linux GCC anyway,
Try
-march=pentium4 -mfpmath=sse -mieee-fp -mmmx -msse -msse2
-fexcess-precision=standard
to get compareble sse math within gcc (-fexcess-precision=standard
only sinced 4.5)
See http://blog.andreas.org/display?id=9
--
Reini Urban
http://phpwiki.org/ http://murbreak.at/