[MPlayer-dev-eng] [PATCH] Drop support for gcc-2.95

[Ivo]

On Monday 10 July 2006 14:38, Diego Biurrun wrote:
> Search the archives, I have done extensive compiler comparisons on x86 a
> long time ago.  The result was that gcc 3.x prior to 3.4 just plain
> sucks.  3.4 finally managed to produce binaries that are as fast as those
> created by 2.95, but it takes considerably longer (40-60% IIRC).
>
> We can discuss this once a gcc >= 3.4 is part of all mainstream
> distributions (meaning at least Debian stable).  Until then it's
> completely out of the question.

I have done a small test myself this afternoon. I wanted to have the latest 
of each gcc branch (2.x, 3.x and 4.x) on my system anyway, so I thought I'd 
compile MPlayer a few times and have them decode some video. I disabled 
alsa because gcc 4.1.1 bugs on my version of /usr/include/alsa/pcm.h. Don't 
need it anyway. System is an AMD Sempron 2400+ (ia32), clocked at 1667MHz 
(no overclocking), with 512MB RAM.

make distclean
./configure --disable-alsa --cc=compiler
svnversion
time make

2.95.3      282.16user 15.26system 5:30.80elapsed 89%CPU	[100%]
3.4.6       409.11user 21.05system 8:02.67elapsed 89%CPU	[145%]
4.1.1       546.13user 23.55system 10:37.38elapsed 89%CPU	[194%]

binary size:
7332030     7.0M        mplayer-r19035-2.95.3		[106%]
6904957     6.6M        mplayer-r19035-3.4.6		[100%]
7950935     7.6M        mplayer-r19035-4.1.1		[115%]

playing typical files (XviD with lavc and a DVD with lavc):

preload file in cache
mpayer -vo null -nosound -benchmark The.Sopranos.S06E10.HDTV.XviD-LOL.avi

2.95.3      VC: 140.477s VO:   0.261s A:   0.000s Sys:  29.620s =  170.357s
3.4.6       VC: 131.340s VO:   0.201s A:   0.000s Sys:  29.473s =  161.014s
4.1.1       VC: 129.461s VO:   0.276s A:   0.000s Sys:  29.500s =  159.237s

clear file cache
mplayer \
-vo null -nosound -benchmark -vfm ffmpeg -dvd-device wallander6.img dvd://1

2.95.3      VC: 627.204s VO:   0.438s A:   0.000s Sys: 118.686s =  746.328s
3.4.6       VC: 579.631s VO:   0.421s A:   0.000s Sys: 113.791s =  693.843s
4.1.1       VC: 601.902s VO:   0.505s A:   0.000s Sys:  99.001s =  701.408s

I added -vfm ffmpeg because it defaults to mpegpes (pass through) if -vo 
null is specified.

percentages of speed (VC only):

		ffodivx		ffmpeg2
2.95.3		109%		108%
3.4.6		101%		100%
4.1.1		100%		104%


I'd think 2.95.3 still performs pretty well. The resulting binary is only a 
tad slower (<10% for typical use), but it compiles way faster.

--Ivo