[gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Eugenia Loli-Queru]

IMO, GCC 3 is better because it generates BETTER and FASTER code. I am all 
to see it included in the next Gentoo as default. The compilation time is 
indeed slower than GCC 2.95.X, because it does things that are more complex 
(hey, they have fixed their previously half-baked C++ support) and because 
it has not been optimized yet. However, the trade off is acceptable, because 
you got Konqueror or other (especially) C++ applications to run TIMES 
faster, *especially* when optimized for PPRO and above (i686). GCC 3.x does 
not generate as good and fast code as the Intel compiler ("Proton") does, 
but it certainly beats GCC 2.95.x as much Proton beats GCC 3.x. ;-D

So, regarding your benchmarks Spider. There is something wrong, definately. 
And I think our gentoo kernel heads around here should take a close look at 
it. Sure, GCC 3.X *is* slower on compilation time, however, your tests show 
a very disturbing fact: Under some circumstances, your CPU seems to spend 
unreasonable amount of time not doing anything. This could be an indication 
of a bigger issue, possibly a configuration or a hardware issue. There might 
be an  issue going on with the cache or the filesystem or even the loader. 
How much memory the PC you used has and what kind of drive and filesystem 
did you use? (I hope that all this is not a side effect of one of the Gentoo 
kernel patches...)

On a completely different issue, I just emerged Netscape 4.x. Netscape's 
fonts when loading any page are extremely small while Opera's fonts are ugly 
as nothing else on this planet. I think some Xft configuration is not done 
properly...
Also, when one emerges XFCE, XFCE does not load its window manager, because 
the xfce little script on /etc/X11/Sessions/ asks for the 'xfce' binary 
instead of the standard 'startxfce'.

As for the KDM problem that someone also mentioned in this list, I have the 
same problem. The user "eugenia" (part of the 'wheel' and 'audio' groups) 
cannot load KDM (it just doesn't load anything, with no errors displayed), 
while everything runs fine as a root. Any ideas? (standard X with twm works 
fine for user 'eugenia' over here)

Best Regards,
Eugenia


               -----------------
    Editor-in-Chief at http://www.OSNews.com
    Email: eugenia@osnews.com - ICQ: 6070904
     Home page at http://www.eugenia.co.uk







.

_________________________________________________________________
MSN Photos is the easiest way to share and print your photos: 
http://photos.msn.com/support/worldwide.aspx

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Spider]

[-- Attachment #1: Type: text/plain, Size: 2348 bytes --]

I know well that this is was completely unscientific and unreproductible
behaviour, with only one run and so on.

PDC20265: chipset revision 2
PDC20265: not 100% native mode: will probe irqs later
PDC20265: (U)DMA Burst Bit ENABLED Primary PCI Mode Secondary PCI Mode.
    ide2: BM-DMA at 0x7400-0x7407, BIOS settings: hde:DMA, hdf:pio
    ide3: BM-DMA at 0x7408-0x740f, BIOS settings: hdg:pio, hdh:pio

hde: Maxtor 5T030H3, ATA DISK drive
its an Athlon t-bird 1GHz
MemTotal:       288548 kB
(PC-100 SDRAM)

Filesystem on the drive used for compilations are ReiserFS.

Using r5 hash to sort names
ReiserFS version 3.6.25

also, the fact that I dont use the same compiler flags for both
compilers are a dead giveaway.

Better code, I can't speak for. More tests, I can, I've had to patch up
some c++ code in order to fit the stricter tests, something I consider
good.


cpu idle time doesn't matter much when diskaccess is ventured, should I
ever intend to do a good benchmark I'd use tmpfs for the whole process,
and make sure I dont run out of RAM while doing it. This is a user
comparsion, the feeling of how long things take to compile c++. 

And yes, the machine was in "normal use" at the time. Xchat, sylpheed
and some aterm's.  bad behaviour for a benchmarker. But standard for me
whenever I compile things, and thats how I wanted the comparsion done.

kernel is for once the default gentoo one, something I seldom use
normally. (I prefer -jam series)

//Spider



> 
> So, regarding your benchmarks Spider. There is something wrong,
> definately. And I think our gentoo kernel heads around here should
> take a close look at it. Sure, GCC 3.X *is* slower on compilation
> time, however, your tests show a very disturbing fact: Under some
> circumstances, your CPU seems to spend unreasonable amount of time not
> doing anything. This could be an indication of a bigger issue,
> possibly a configuration or a hardware issue. There might be an  issue
> going on with the cache or the filesystem or even the loader. How much
> memory the PC you used has and what kind of drive and filesystem did
> you use? (I hope that all this is not a side effect of one of the
> Gentoo kernel patches...)
> 


--
begin  happy99.exe
This is a .signature virus! Please copy me into your .signature!
See Microsoft KB Article Q265230 for more information.
end

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Stacey Keast]

On Sun, 2002-04-07 at 20:46, Spider wrote:
<snip>

Of course it is going to take gcc3 longer to compile things with the new
profiling code (see http://gcc.gnu.org/news/profiledriven.html)

All this basically is is the compiler running through code branches and
identifying blocks which should be optimized to produce faster
EXECUTABLES, this does not come at the price of faster COMPILE TIMES, as
the compiler has to build these profiles to do more advanced
optimizations.  So, you are really benchmarking the wrong thing here.

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Geert Bevin]

What were you benchmarking since you never even mentioned that, the
compilation of the software or the resulting executable's performance.
The first one seems pretty useless to me, while the latter does make
sense.

On Mon, 2002-04-08 at 04:46, Spider wrote:
> I know well that this is was completely unscientific and unreproductible
> behaviour, with only one run and so on.
> 
> PDC20265: chipset revision 2
> PDC20265: not 100% native mode: will probe irqs later
> PDC20265: (U)DMA Burst Bit ENABLED Primary PCI Mode Secondary PCI Mode.
>     ide2: BM-DMA at 0x7400-0x7407, BIOS settings: hde:DMA, hdf:pio
>     ide3: BM-DMA at 0x7408-0x740f, BIOS settings: hdg:pio, hdh:pio
> 
> hde: Maxtor 5T030H3, ATA DISK drive
> its an Athlon t-bird 1GHz
> MemTotal:       288548 kB
> (PC-100 SDRAM)
> 
> Filesystem on the drive used for compilations are ReiserFS.
> 
> Using r5 hash to sort names
> ReiserFS version 3.6.25
> 
> also, the fact that I dont use the same compiler flags for both
> compilers are a dead giveaway.
> 
> Better code, I can't speak for. More tests, I can, I've had to patch up
> some c++ code in order to fit the stricter tests, something I consider
> good.
> 
> 
> cpu idle time doesn't matter much when diskaccess is ventured, should I
> ever intend to do a good benchmark I'd use tmpfs for the whole process,
> and make sure I dont run out of RAM while doing it. This is a user
> comparsion, the feeling of how long things take to compile c++. 
> 
> And yes, the machine was in "normal use" at the time. Xchat, sylpheed
> and some aterm's.  bad behaviour for a benchmarker. But standard for me
> whenever I compile things, and thats how I wanted the comparsion done.
> 
> kernel is for once the default gentoo one, something I seldom use
> normally. (I prefer -jam series)
> 
> //Spider
> 
> 
> 
> > 
> > So, regarding your benchmarks Spider. There is something wrong,
> > definately. And I think our gentoo kernel heads around here should
> > take a close look at it. Sure, GCC 3.X *is* slower on compilation
> > time, however, your tests show a very disturbing fact: Under some
> > circumstances, your CPU seems to spend unreasonable amount of time not
> > doing anything. This could be an indication of a bigger issue,
> > possibly a configuration or a hardware issue. There might be an  issue
> > going on with the cache or the filesystem or even the loader. How much
> > memory the PC you used has and what kind of drive and filesystem did
> > you use? (I hope that all this is not a side effect of one of the
> > Gentoo kernel patches...)
> > 
> 
> 
> --
> begin  happy99.exe
> This is a .signature virus! Please copy me into your .signature!
> See Microsoft KB Article Q265230 for more information.
> end
-- 
Geert Bevin             Uwyn
"Use what you need"     Lambermontlaan 148
http://www.uwyn.com     1030 Brussels
gbevin@uwyn.com         Tel & Fax +32 2 245 41 06

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Spider]

[-- Attachment #1: Type: text/plain, Size: 1467 bytes --]

No, I'm not. :)

the "benchmark" as such was in response to peoples questions on wether
gcc3.0 was faster or slower during compilation than gcc2.9  

For my own use, I think the results are well worth it now with gcc
3.0.4, especially since gcc now can be used to compile the kernel ;)
(bugfix in gcc304) 

please take this the right way, I'm well aware of the improved results
of gcc3 compared to gcc 2.9 , and why the slowdowns occur. (Check the
memory usage of -Wall and other things and you'll be surprised)


//Spider


begin  quote
On 07 Apr 2002 23:00:55 -0600
Stacey Keast <slik@telusplanet.net> wrote:

> On Sun, 2002-04-07 at 20:46, Spider wrote:
> <snip>
> 
> Of course it is going to take gcc3 longer to compile things with the
> new profiling code (see http://gcc.gnu.org/news/profiledriven.html)
> 
> All this basically is is the compiler running through code branches
> and identifying blocks which should be optimized to produce faster
> EXECUTABLES, this does not come at the price of faster COMPILE TIMES,
> as the compiler has to build these profiles to do more advanced
> optimizations.  So, you are really benchmarking the wrong thing here.
> 
> _______________________________________________
> gentoo-dev mailing list
> gentoo-dev@gentoo.org
> http://lists.gentoo.org/mailman/listinfo/gentoo-dev


--
begin  happy99.exe
This is a .signature virus! Please copy me into your .signature!
See Microsoft KB Article Q265230 for more information.
end

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Bart Verwilst]

Hi

Euhm...
These benchmarks of spider were compile times? :o)
Damn.. I still don't know wether i will have a faster system with gcc 3.x
than i had with gcc 2.95.x... please tell me:
Is gcc 3.x faster than 2.95 during RUNTIME and execution of the program?

Thanks! :o)

-- 
Bart Verwilst
Gentoo Linux Developer, Desktop Team
Gent, Belgium

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Einar Karttunen]

On 08.04 11:54, Bart Verwilst wrote:
> These benchmarks of spider were compile times? :o)
> Damn.. I still don't know wether i will have a faster system with gcc 3.x
> than i had with gcc 2.95.x... please tell me:
> Is gcc 3.x faster than 2.95 during RUNTIME and execution of the program?
> 

it depends on the program :-)

In most cases the code generated by gcc 3.x is faster than that 
generated by 2.95.x. 

- Einar Karttunen

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Thilo Bangert]

On Monday, 8. April 2002 11:54, you wrote:
> Hi
>
> Euhm...
> These benchmarks of spider were compile times? :o)
> Damn.. I still don't know wether i will have a faster system with gcc
> 3.x than i had with gcc 2.95.x... please tell me:
> Is gcc 3.x faster than 2.95 during RUNTIME and execution of the
> program?

the german magazin ix ( http://www.heise.de/ix/ ) did some benchmarks 
(spec) on gcc 3.0 and gcc 2.95.3 which showed that gcc3.0 was faster - 
but not by much
in some cases (bzip) it was slower!

but then, i don't remember if they bootstrapped the compiler....

>
> Thanks! :o)

-- 
regards
Thilo

Re: [gentoo-dev] GCC 3 Vs GCC 2 and some other stuff

[Spider]

[-- Attachment #1: Type: text/plain, Size: 840 bytes --]

sorry for missing that vital point, it was a comparsion of the
compilation process, not the end results.

Thats the issue when talking about the -speed- of a compiler, when
comparing the -quality- on the other hand, then we are talking of how
well the compiled code is performing. There are several benchmarks of
this avaiable, just search the net and you shall find :)



//Spider


begin  quote
On 08 Apr 2002 07:33:13 +0200
Geert Bevin <gbevin@uwyn.com> wrote:

> What were you benchmarking since you never even mentioned that, the
> compilation of the software or the resulting executable's performance.
> The first one seems pretty useless to me, while the latter does make
> sense.
> 



--
begin  happy99.exe
This is a .signature virus! Please copy me into your .signature!
See Microsoft KB Article Q265230 for more information.
end

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]