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] PR15242

On Jan 27, 2005 02:54 PM, Paolo Bonzini <> wrote:

> 1) Are you sure you want !optimize_size?


> 300% slowdowns are not 
> something small, and such computed goto code can be found in embedded 
> software using -Os.

A couple of kilobytes less is also not easily found.  For my Prolog
test case, the object file grows >12% (back to 3.2-days sizes) with
the patch.

> 2) Also, it does not seem particularly expensive given that it only 
> affects a wide minority of functions (for others, it's just a single 
> FOR_EACH_BB iteration).
> Maybe
> if (flag_expensive_optimizations || (optimize && !optimize_size))
> (i.e. only do it with -Os -fexpensive-optimization, or with -O)
> would be better?

Also fine by me, but if you don't mind, I'm not going to do the
testing for that.  I really don't think you'd do -Os and then
worry about speed.  Especially because those large slowdowns are
only for cases with very many computed gotos (and I suspect, 
mostly for targets with branch target buffers, which you would
not typically find in embedded chips, /meguesses).


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