[MPlayer-dev-eng] Re: Compile options

[Andrew Savchenko]

Hi

> > > > > It was tested with the following command:
> > > > > for -O4:
> > > > > for (( i=0; i-10; i++ )); do ( echo phase $i: >> ../O4.res;
> > > > > time ./mplayer file.avi -osdlevel 3 -stop-xscreensaver >>
> > > > > ../O4.res
> > > >
> > > > without -benchmark -nosound the time decoding a file obviously
> > > > will be >= the length of the file timewise which makes the
> > > > results pretty meaningless
> > >
> > > Why? I've tested mplayer as whole application, not only as video
> > > decoder. I've measured real and sys time required in both cases
> > > (-O2 vs -O4).
> >
> > Sorry, typo: I've measured user and sys time, imho they indicates
> > real cpu usage despite movie duration.
>
> indeed, that should be ok too

Well, I have done the same testing as earlier but with -benchmark 
-nosound -quiet options, other conditions remain the same.

New test only shows more significant difference between O2 and O4 in 
favor to O4. Full test results (output of mplayer and time) as well as 
computation results are in attachment, so anyone can verify them.

Now I want to discuss results:
*********************************************
Results for normal distribution, reliability 0.95, 10 measurements,
t(0.95,10)=2.26:
======================================
for -O4:

BENCHMARKs: VC:   (8.951 \pm 0.035)s VO: (178.3 \pm 0.7)s    A:  0.000s  
Sys:  (0.599 \pm 0.004)s   = (187.8 \pm 0.7)s
BENCHMARK%: VC:   (4.766 \pm 0.033)% VO: (94.914 \pm 0.034)% A:  0.0000% 
Sys:  (0.3192 \pm 0.0026)% = 100.0000%

real    (3m8.0s \pm 0m0.7s)
user    (1m48.75s \pm 0m0.16s)
sys     (0m0.799s \pm 0m0.034s)

user+sys     (1m49.55s \pm 0m0.16s)

======================================
for -O2:

BENCHMARKs: VC:   (9.012 \pm 0.012)s VO: (179.02 \pm 0.13)s  A:  0.000s  
Sys:  (0.594 \pm 0.004)s   = (188.62 \pm 0.14)s
BENCHMARK%: VC:   (4.778 \pm 0.006)% VO: (94.907 \pm 0.007)% A:  0.0000% 
Sys:  (0.3150 \pm 0.0022)% = 100.0000%

real    (3m8.77s \pm 0m0.14s)
user    (1m49.19s \pm 0m0.12s)
sys     (0m0.802s \pm 0m0.043s)

user+sys     (1m49.99s \pm 0m0.13s)

======================================
diffrences in performance between O2 and O4:
\Delta(O2-O4):

BENCHMARKs: VC:   (0.061 \pm 0.037)s VO: (0.7 \pm 0.7)s      A:  0.000s  
Sys:  (-0.005 \pm 0.006)s   = (0.8 \pm 0.7)s
BENCHMARK%: VC:   (0.012 \pm 0.034)% VO: (-0.007 \pm 0.035)% A:  0.0000% 
Sys:  (-0.0042 \pm 0.0034)% = 100.0000%

real    (0m0.77s \pm 0m0.7s)
user    (0m0.44s \pm 0m0.20s)
sys     (0m0.003s \pm 0m0.055s)

user+sys     (0m0.44s \pm 0m0.21s)
*********************************************

BENCHMARK% is not interesting, as it shows only relative distribution of 
CPU usage for mplayer.

BENCHMARKs shows _clearly speedup_ for VC when using -O4; possibly 
performance gain for VO (and possibly not, but it is not a performance 
loss); sys CPU performance change is unclean within errors: in may be 
higher or lower, or the same.

According to time, sys CPU usage is not changed within errors -- it is 
highly predictable result.

real, user and user+sys time clearly shows advantage of -O4 with respect 
to statistical errors. So -O4 must be enabled by default.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.tar.bz2
Type: application/x-tbz
Size: 3461 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20060917/be48b93a/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 191 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20060917/be48b93a/attachment.pgp>