This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: gcc compile-time performance take 2


Dara Hazeghi <dhazeghi@pacbell.net> writes:

> Hello,
>
> well I had no idea the thread would spark a firestorm of debate (is flamewar 
> the correct term?). In any case there are a number of opinions about the 
> matter. My point (although it seems to now be submerged in a discussion of 
> what is acceptable development hardware) is that things have gotten slower. I 
> am not particularly optimistic that gcc 3.2 is going to be faster or equal in 
> this respect to gcc 2.95, and I think it is rather unrealistic to think so. 
> What I would like to see though, is some method of ensuring that gcc 3.2 is 
> not particularly slower, given the same optimizations, than gcc 3.1. To this 
> end, I think Andi Kleen's comments are right on the mark.
>
> So my suggestion (again, as a non-developer, though a frequent user) would be 
> to time the SPEC builds. Considering that Andreas Jaeger and Diego Novillo 
> are already building and running the benchmarks, it seems that a few tweaks 
> to the scripts in question would easily allow one to follow compile-time 
> performance on top of run-time performance.

Jan Hubicka asked me last week to do this and yesterday evening I
started enhancing my scripts to visualize:
- the bootstrap time of GCC
- the build time for each SPEC program

Since a complete iteration takes half a day I haven't seen any results
yet - and therefore hadn't planned to publish this yet. I'll look
during the next days in changing all my graphs and fixing the problems
yesterday's work encountered.

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]