This is the mail archive of the gcc-patches@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: Fix IMA ICE in get_tmt_for really (PR 16954)


On Aug 11, 2004, at 12:58 PM, Diego Novillo wrote:
On Wed, 2004-08-11 at 15:49, Dale Johannesen wrote:

Did this also fix mesa?
I don't like this approach, the types are compatible and the renaming
ought to be
done.   Leaving extra copies around in these cases could be a
significant
performance hit.

This is not a problem in copy propagation.  It's in the copy renamer,
where we try and coalesce several variables to the same root.  Which is
a way of preserving user variables as much as possible.

This is something that I expect to trigger very rarely.

I don't think the case in the example, where variables become coalescable after
inlining, is rare at all. I'll go try a Spec run and get some data.



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