[MPlayer-dev-eng] Re: Compile options

[Trent Piepho]

On Sun, 17 Sep 2006, Carl Eugen Hoyos wrote:
> Andrew Savchenko <Bircoph <at> list.ru> writes:
>
> [...]
>
> > Well, I have done the same testing as earlier but with -benchmark
> > -nosound -quiet options, other conditions remain the same.
>
> Could you provide a link to the movie clip you used? I repeated my tests a few
> times yesterday and I believe that -O2 is significantly faster than -O4 for
> Apple HD movie trailers (H264). Could you also test one of those?.
>
> Can somebody else confirm one or the other result?

mplayer svn compiled with:
-fno-PIC -march=athlon-xp -mtune=athlon-xp -ffast-math -fomit-frame-pointer

and -O2 or -O4.  I used the apple Serenity 1080p trailer, with codec
ffh264.  mplayer options were -quiet -benchmark -vo null -nosound

Results:
vc.O4 <- c( 137.928,139.243,138.744,138.226,139.541,138.695,138.198,137.495,137.840,136.083 );
vo.O4 <- c( 0.026,0.028,0.027,0.028,0.027,0.027,0.027,0.028,0.029,0.027 );
a.O4 <- c( 0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000 );
sys.O4 <- c( 0.772,0.788,0.779,0.781,0.779,0.778,0.782,0.776,0.776,0.770 );
user.O4 <- c( 138.57,139.92,139.41,138.89,140.23,139.38,138.94,138.15,138.5,136.8 );
elapsed.O4 <- c( 139.2,140.53,139.62,139.5,140.42,139.57,139.48,138.77,138.72,137.35 );

vc.O2 <- c( 136.208,133.413,134.538,133.513,133.833,130.967,134.642,131.339,133.438,134.247 );
vo.O2 <- c( 0.029,0.029,0.028,0.028,0.028,0.028,0.028,0.028,0.028,0.030 );
a.O2 <- c( 0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000 );
sys.O2 <- c( 0.790,0.784,0.783,0.782,0.782,0.770,0.781,0.771,0.781,0.785 );
user.O2 <- c( 136.85,134.06,135.25,134.21,134.53,131.67,135.34,132.05,134.13,134.93 );
elapsed.O2 <- c( 137.5,134.3,135.42,134.4,134.71,131.84,135.52,132.21,134.32,135.13 );

A Welch two sample t-test finds that O2 is faster for the vc, user, and
elapsed time.  Since the vc time makes up nearly all of the user/elpased
time, it's expected that they are all the same.  For sys time O2 and O4 are
the same.  And O4 is faster for vo, but since -vo null was used, that isn't
very interesting.  The data appears to have a roughly normal distribution,
and using a robust median based test, like Wilcoxon rank sum, doesn't find
anything different.

The output from comparing vc time with R looks like this:
> t.test(vc.O2, vc.O4, alt="less")

        Welch Two Sample t-test

data:  vc.O2 and vc.O4
t = -7.9474, df = 15.243, p-value = 4.191e-07
alternative hypothesis: true difference in means is less than 0
95 percent confidence interval:
      -Inf -3.575077
sample estimates:
mean of x mean of y
 133.6138  138.1993

In this case we have a "Null Hypothesis" that the mean value of vc.O2 minus
the mean value of vc.O4 is equal to greater than zero.  That is to say, our
null hypothesis is that O2 is slower than or the same as O4.  By comparing
the observed mean values, 133.6138 and 138.1993, and the observed variance
of those values, we find that the probility of the null hypothesis being
true and seeing the values we did is 0.00004191%.  That is extremely
unlikely, and so we reject the null hypothesis as being false.  In that
case, we must accept the alternative hypothesis, that the mean value of
vc.O2 is less than the mean value of vc.O2, ie. O2 is faster than O4.

The 95 percent confidence interval tells us that we can be 95% certain that
O2 is 3.575077 seconds or more faster than O4.