Andy Ross wrote
> Vivian Meazza wrote:
> > Only a minute eh? Under Cygwin cvs takes nearly 5 minutes - time for
> > a brew a coffee - and that's on a pretty powerful machine.
>
> Time from execution to fade in of the cockpit display is about 10
> seconds on my laptop (1.8GHz Athlon64). I started trying to hunt this
> down.
>
> One issue that had been discovered earlier is that an AI-related patch
> to the the airport loading code in February causes the startup routine
> to try to load a rwyuse.xml and parking.xml file in every one of 24597
> potential directories under $FG_ROOT/Aiports/AI.
>
> Attached is a patch that pre-reads the directory contents ahead of
> time (currently that is a list of length zero) to avoid having to hit
> the kernel (twice!) for every airport.
>
> Under Linux, this doesn't provide much speedup. But Windows (and
> especially the cygwin libraries) has a somewhat less robust I/O system
> in the face of many tiny operations. Hopefully it will help there.
> Can someone on each of cygwin, mingw and/or MSVC try this out and see
> if it helps?
>
> Unfortunately, because there is no actual parking/runway AI data in
> the base package, much of this is currently "dead code" that cannot be
> tested. I can't promise I didn't break anything, because I have no
> way of knowing whether it worked in the first place. :)
>
The results of this patch are very encouraging:
Airports and Navigation Data Total
Before 2 min 40 sec 3 min 50 sec
After 1 min 40 sec 2 min 50 sec
It's still the most significant part of the load process - can we be even
more clever yet?
Nothing seems to be broken ... but I haven't tested exhaustively
I'll try the plib patch later.
Regards,
Vivian
_______________________________________________
Flightgear-devel mailing list
Flightgear-devel@flightgear.org
http://mail.flightgear.org/mailman/listinfo/flightgear-devel
2f585eeea02e2c79d7b1d8c4963bae2d
