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]
Other format: [Raw text]

Re: [PATCH] do not lower a/b to a*(1/b)

On Friday 13 May 2005 16:22, Roger Sayle wrote:
>  In several recent posting gcc 4.0 has been
> reported slower than gcc 3.x, and it's a trend I'm not happy with, even
> if attempts to "stop the rot" make reviewers unpopular.


People are more likely to report slowdowns than speedups,
so all those postings in the end mean very little.  It is
only to be expected that there are some regressions if you
implement an entirely new, still-to-be-tuned optimization

What matters more in the end is what the *overall* quality
of code produced by GCC.  GCC4 is simply better than any
GCC3 most of the time.  See echristo's remarks about GCC4
on MIPS the other day, on PPC GCC4 rocks compared to GCC3,
and on AMD64 we are also doing better, see e.g. these two

So there really is not much rot to stop.  There is only a
lot of already rotten code to be cleaned up, and an entire
framework to be tuned.

So please stop spreading this "GCC4 is worse" FUD.

Dan, could PRE strength reduction catch this "optimization"
that we'd miss with Paolo's patch?


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