Re: gcc compile-time performance

[Jan Hubicka <jh at suse dot cz>]

> 
> In this case a 6 cpu P3 box.
Hmm, my one is old good K6. Anyway I guess I will implement the simple fp
library for it.
I was just hoping that it will not be required (fewer code = easier maintenance)
and my profiles, both -Q and oprof looked good at Athlon.
Note that the propagation is, by nature, relatively expensive, as it is global
analysis of the CFG.  The algorithm used is fast and I consider 2% to be not
too bad, but 0.5% is better :)
> 
> Note, I see very bizarre numbers from the -Q dumps also -- I don't think 
> the timers are returning small enough increments as the machines get 
> faster -- note that often we are getting zero complete clock ticks within 
> a single optimization.

Perhaps we can try to use rdts on i386 machines, as -Q is definitly usefull
tool and should work.

Honza