This is the mail archive of the 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]

Re: Projects for beginners: break up expand_expr

Momchil Velikov wrote:

You can also use -time and -Q to get more finely-divided numbers,
down to the compiler phase level.

> Patched compiler:
> [...]
> user    0m30.110s

> Unpatched compiler:
> [...]
> user    0m30.030s

OK, I work this out to be about a 2.6% slowdown, which is enough to
concern me. If we put in just 40 changes with this kind of effect,
that doubles compilation time, and we easily do 40 changes each and
every week (fortunately, most can't affect compilation speed).

Yes, I know this isn't particularly scientific - we would need a much
longer test run just to be sure this isn't just an artifact, and to
test more compilations, etc.  Also, just for the record, I think this
change is worth making anyway, and so I don't want to sound like I'm
picking on it.

But we do have a speed problem with GCC, and because we do incremental
development, it's easy for changes like this to seem like they don't
cost anything.  People evaluating proposed patches ought to at least
consider whether the patch has any effect on overall performance, and
to look a little deeper if there's any question.  It would also help
to develop more accurate timing procedures, since slowdowns are more
likely to happen a little bit at a time.


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