I just compiled mplayer2 from source (git://git.mplayer2.org/mplayer2-build.git) because the repository (ubuntu 12.04) version didn't work on my system. Since I have old hardware I was just wondering if there are some compiler flags I could use to optimize it for my hardware.
The CPU is an Athlon XP 2200+ (mobile), 1GB RAM, graphics: Nvidia GeForce4 420 Go.
I also want to do the same on another old system with similar specs:
CPU: Athlon 1,2 GHZ, 1GB RAM, graphics: [SiS] 65x/M650/740
If the version taken from your distribution repository of mplayer2 doesn't work for you, it is a good thing to report what didn't work to the bug tracking system, so that it has benefits for both you and others:
Recompiling the program specifically for your machines is likely to have better results than the "generic" flavor that a distribution releases (this is, BTW, one of the motivations that the Gentoo folks have when they recompile things to their own systems).
Of course, you may gain some improvements in speed, but you loose the portability of the binaries.
That being said, the general way recompiling a program in a current Debian/Ubuntu system is to take the source package and its build dependencies with something as:
sudo apt-get build-dep mplayer2
sudo apt-get install fakeroot
apt-get source mplayer2
then edit the file debian/rules inside the directory created by the last
command to change the values that go in the CFLAGS, CPPFLAGS,
CXXFLAGS, and LDFLAGS.
What can you do to tailor the application to your machines? You will have to
experiment (read: "measure/benchmark", see below) with which level of
optimization (like -O2, -Os, or -O3) the program runs faster.
To actually compile the program, you will need to run, inside the directory
created by the apt-get source mplayer2 command:
fakeroot debian/rules binary
sudo dpkg -i ../*.deb
With GCC versions 4.7 or newer, you can even experiment with the -Ofast
compilation level, which for playing videos is not going to cause too much
harm, but which can give you some improvements (enough to not accumulate
frames and get audio and video out of sync).
To compile the program specifically for the machine where you will be
executing it, it is a good thing to use GCC's -mach=native flag. This will
probably make the binaries produced unuseable in other systems, but as long
as you care only for your system, that's the way to go.
Just to give you an idea of what options are enabled on my Core i5-2410M
when I use -march=native, see (output reformatted to not destroy the
layout of the site):
gcc -march=native -E -v - < /dev/null 2>&1 | grep cc1
/usr/lib/gcc/i486-linux-gnu/4.7/cc1 -E -quiet -v -imultiarch i386-linux-gnu - \
-march=corei7-avx -mcx16 -msahf -mno-movbe -maes -mpclmul -mpopcnt \
-mno-abm -mno-lwp -mno-fma -mno-fma4 -mno-xop -mno-bmi -mno-bmi2 \
-mno-tbm -mavx -mno-avx2 -msse4.2 -msse4.1 -mno-lzcnt -mno-rdrnd \
-mno-f16c -mno-fsgsbase --param l1-cache-size=32 \
--param l1-cache-line-size=64 \
--param l2-cache-size=3072 -mtune=corei7-avx
From there you can see that GCC detected some "advanced" instructions that my computer has (AVX) and others that it doesn't have (AVX2).
As a hint, to benchmark, just play a short video, say, foo.mkv with:
mplayer -benchmark -vo null -nosound foo.mkv
This will "play" the video as fast as your system can and tell you how many seconds it took to "play" the video in its entirety. Note that I said "play" in quotes, because we are disabling:
-nosound. Usually minor time is spent here, in
comparison to the other parts of playing a video.-vo null).To see if the video card is getting in the way or not, you can omit the -vo
null part from the command above and see if the video you want plays faster
than real-time (or whatever your goal is).
That being said, much of mplayer2 (and regular mplayer, when this later one is taken from distributions) is that most of their processing is "offloaded" to libraries. In particular, a lot of the decoding is made by libav or ffmpeg, and those are the packages that should be compiled/optimized in the first place.
In the case of "vanilla" mplayer (not mplayer2) taken from upstream, it uses
embedded copies of many libraries, which means that, if you compile it from
the upstream sources (instead of the method that I gave you above with
apt-get source mplayer2 etc), it will also compile its own libav/ffmpeg
and have the potential of being much faster than the alternatives.
It is not always necessary to recompile the mplayer/mplayer2 binaries provided by your distribution if you change a few configuration parameters. To avoid all the work above, I would begin by playing the videos with something like:
mplayer -framedrop -lavdopts fast:skipframe=nonref:skiploopfilter=nonref foo.mkv
Of course, you can play with the options that I just gave you and the
manpage documents the possible values for skipframe and skiploopfilter,
among other things.
And happy watching videos!
