[comp.windows.x] x11perf version 1.2 available on expo.lcs.mit.edu

joel@decwrl.dec.com (Joel McCormack) (12/30/89)

The latest x11perf, version 1.2, is now available. x11perf measures the
performance of X11 servers in excruciating detail.  (There are so many tests,
if you make a mistake on one of the options, you'll have to pipe it through
``more''.)

Version 1.2 has far more tests than version 1.0 did.  Wide lines, dashed,
double-dashed, even wide double-dashed arcs, slow as they are.  Just about
every aspect of X11 drawing and window management gets at least some coverage.

You cannot make comparisons between numbers obtained from earlier versions of
x11perf and the 1.2 version.  I have changed too many tests in small and
subtle ways.  (For example, 1-pixel segments didn't really measure segments of
1 pixel...the algorithm got messed up.  Rectangles did not measure rectangles
at all possible alignments, as I had squished everything together to enable me
to find bugs in server stipple code.)

Mark Moraes from the University of Toronto donated some shells for comparing
x11perf outputs, which have been glued together as x11perfcomp (which calls
another script, x11perffill).  He also provided some useful options to x11perf
for dealing with servers that crash on some x11perf tests.  (Not DEC servers,
of course :)

Running time using the -all option varies, even with the self-calibration.  On
a pmax (DECStation 3100), it takes a little under two hours.  My stupid gpx
took over 4 hours.  Some of the tests just take forever (especially, say, wide
double-dashed arcs), even though they draw hardly anything.

If people will run x11perf on various hardware and servers, I will collect the
results and post whatever is not proprietary.  Also, if x11perf doesn't
measure something that makes a difference in your life, please send suggestions.

Where to get the new x11perf:

There is a compressed tar file available for anonymous ftp on
expo.lcs.mit.edu:/contrib/x11perf.tar.Z.

An slightly earlier version is available on the upcoming R4 tapes.  The tests
performed by that version are the same.  The newer version I've just put on
expo has a more correct and informative man page, a man page for x11perfcomp,
and a few other minor differences.

- Joel McCormack (decwrl!joel, joel@decwrl.dec.com)