[MPlayer-users] [BUG] Compile with ICC 10.1

[Reimar Döffinger]

Hello,
On Sun, Jan 27, 2008 at 09:56:31PM +0200, Dmitry Leonenko wrote:
> Reimar Döffinger wrote:
> > On Sun, Jan 27, 2008 at 06:56:03PM +0200, Dmitry Leonenko wrote:
> >   
> >>> I tested with 64 bit and I have no such problem with ICC 10.1.008 (was
> >>> too lazy to download a newer version).
> >>> But my previous patch was incomplete, here is the complete one.
> >>> Do not forget to make distclean and reconfigure after changing
> >>> something, just to be sure (it should not really be necessary most of
> >>> the time).
> >>>   
> >>>       
> >> My log after compiling with last patch (attached).
> >>     
> >
> > Remove the "static" in lines 51 and 52 of
> > libavcodec/i386/simple_idct_mmx.c as well.
> > _______________________________________________
> > MPlayer-users mailing list
> > MPlayer-users at mplayerhq.hu
> > http://lists.mplayerhq.hu/mailman/listinfo/mplayer-users
> >   
> I've made some test like:
>     time ./mencoder /tmp/samyjluchshijfilm_clip3_640.divx -o test.avi 
> -ovc lavc -oac lavc
> and fount that in any case with any flags ICC 10.1.008 performs worse 
> than GCC.
> Some results:
> 
> icc -O2 -xT -parallel -ip:
> real    1m49.636s
> user    1m48.803s
> sys     0m3.604s
> 
> Ubuntu's binary mplayer:
> real    1m23.876s
> user    1m20.149s
> sys     0m1.800s
> 
> icc -O3 -msse3 -ip -parallel -march=core2 -mtune=core2 -msse3:
> real    1m28.511s
> user    2m37.506s
> sys     0m3.640s

Using -parallel is really a very bad idea, multiple threads have a big
overhead and the cases that _really_ make a difference have already been
parallelize manually in much more efficient ways (see lavcopts/lavdopts
threads options).
Compile without -parallel and it should have at least the same speed
(though as said on a 64 bit system I would not expect much of a speedup).