[OT] gcc 2.95.2 vs. 3.0 (fwd)

[OT] gcc 2.95.2 vs. 3.0 (fwd)

[Sasi Peter]

first of all, if you are not interested in compiler comparision, just
skip.

I though posting a little demonstration of how well performing code gcc
2.95.2 generates (vs. gcc 3.0) won't hurt. Also shown, that further
optimatizations are anything but impossible (using Intel's commercial c
compiler).

I know opendivx code is not like kernel code at all, but on the other hand
it is well suited for benchmark testing.

Maybe this will cool down thoose a bit, who were to move to gcc 3.0 as
fast as possible, for the _way_ better generated code...

-- 
SaPE - Peter, Sasi - mailto:sape@sch.hu - http://sape.iq.rulez.org/

---------- Forwarded message ----------
Date: Sun, 24 Jun 2001 04:31:56 +0200 (CEST)
From: Arpi <arpi@thot.banki.hu>
Reply-To: mplayer-users@lists.sourceforge.net
To: mplayer-users@lists.sourceforge.net
Subject: [Mplayer-users] gcc 2.95.2 vs. 3.0

Hi,

Just compiled and try gcc 3.0 stable.
And the results is very big surprise! It makes *SLOWER* code!

test file:  (opendivx with postprocessing, this stuff is written in C)
# mplayer -osdlevel 0 -nosound -benchmark 1800.avi -vo null -pp 15
VIDEO:  [divx]  640x352  24bpp  23.98 fps  1865.1 kbps (227.7 kbyte/s)

hardware: celeron-2 600 @ 900MHz (100MHz FSB), 256MB RAM (PC133, at 2/2/2)
(sse,mmx,mmxext enabled)

gcc 2.95.2 result:  (flags: -O4 -march=i686 -mcpu=i686 -pipe -ffast-math)
BENCHMARKs: V:   5.087s VO:   0.001s A:   0.000s Sys:   0.013s =    5.101s
BENCHMARK%: V: 99.7160% VO:  0.0258% A:  0.0000% Sys:  0.2582% = 100.0000%

gcc 3.0 result:  (flags: -O4 -march=i686 -mcpu=i686 -pipe -ffast-math)
BENCHMARKs: V:   5.194s VO:   0.001s A:   0.000s Sys:   0.013s =    5.208s
BENCHMARK%: V: 99.7309% VO:  0.0221% A:  0.0000% Sys:  0.2471% = 100.0000%

yes, speed diff isn't big, but i was prepared for big speed boost by 3.0,
and then i got speed loss! :(

if i recompile C-only parts of decore with intel's commercial C compiler
(free beta available for linux), it's about 15% faster:

icc result:  (optim. flags: -O2 -fp -rcd -tpp6 -xiMK)
BENCHMARKs: V:   4.415s VO:   0.001s A:   0.000s Sys:   0.013s =    4.429s
BENCHMARK%: V: 99.6676% VO:  0.0290% A:  0.0000% Sys:  0.3034% = 100.0000%


A'rpi / Astral & ESP-team

--
mailto:arpi@thot.banki.hu
http://esp-team.scene.hu

_______________________________________________
Mplayer-users mailing list
Mplayer-users@lists.sourceforge.net
http://lists.sourceforge.net/lists/listinfo/mplayer-users

Re: [OT] gcc 2.95.2 vs. 3.0 (fwd)

[J . A . Magallon]

On 20010624 Sasi Peter wrote:
>
>I know opendivx code is not like kernel code at all, but on the other hand
>it is well suited for benchmark testing.
>
>
>test file:  (opendivx with postprocessing, this stuff is written in C)
># mplayer -osdlevel 0 -nosound -benchmark 1800.avi -vo null -pp 15
>VIDEO:  [divx]  640x352  24bpp  23.98 fps  1865.1 kbps (227.7 kbyte/s)
>

Sure it is opendivx ? I think you are just using gcc compiled code for
the 'interface' and 'glue' to windows divx decoders (/usr/lib/win32/*.dll)
that do the real hard work.

Redo the tests with am MPEG2 movie.

-- 
J.A. Magallon                           #  Let the source be with you...        
mailto:jamagallon@able.es
Mandrake Linux release 8.1 (Cooker) for i586
Linux werewolf 2.4.5-ac17 #2 SMP Fri Jun 22 01:36:07 CEST 2001 i686

Re: [OT] gcc 2.95.2 vs. 3.0 (fwd)

[Sasi Peter]

On Mon, 25 Jun 2001, J . A . Magallon wrote:

> Sure it is opendivx ? I think you are just using gcc compiled code for
> the 'interface' and 'glue' to windows divx decoders (/usr/lib/win32/*.dll)
> that do the real hard work.

Have a look at mplayer.sourceforge.net. MPlayer besides DLL loading also
features native Opendivx en/decoding, and native MPEG1/2 decoding.
Actually the tests were performed by the leader of the development of
mplayer, and he did compile the whole opendivx encore/decore code used in
this player, taken from the original sources.

> Redo the tests with am MPEG2 movie.

Actually since the original posting, on the mplayer-devel list the
maintainer of mpeg2play (the portable parts of mplayer as a separate
player, w/o the dll stuff) also tested how well MPEG1/2 decoding works if
compiled w/ gcc 2.95.2 vs 3.0, and he was disappointed too, because there
was a slight decrease in the performanceof the generated code...

If you want proof of these, grab the C sources from the mentioned
sourceforge project site, and repeat the test yourself.

-- 
SaPE - Peter, Sasi - mailto:sape@sch.hu - http://sape.iq.rulez.org/