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: Kill last use of ggc_push_context/ggc_pop_context


On Tue, 2004-10-05 at 12:47, Zack Weinberg wrote:
> Jeffrey A Law <law@redhat.com> writes:
> 
> > On Tue, 2004-10-05 at 12:33, Zack Weinberg wrote:
> >> Jeffrey A Law <law@redhat.com> writes:
> >> 
> >> > Not until someone confirms that cse no longer generates tons of
> >> > test RTL with the testcases from Plum Hall.  See:
> >> 
> >> He didn't take out the call to ggc_collect, only the push/pop context,
> >> so Plum Hall should still be fine even if it does generate tons of
> >> test RTL.
> > But wouldn't we have to do a full GC scan rather than the reduced
> > scan implied by pushing the GC context?  Or am I misunderstanding
> > how the GC push/pop context stuff works?
> 
> The GC always does a full mark phase.  The only effect of a pushed
> context is that the sweep phase does not consider objects to be
> garbage, even if unmarked, if they are not in the topmost context.
> 
> The compiler should be strictly more efficient in the absence of
> push/pop_context calls, assuming that there is no correctness issue,
> and I believe Jan has addressed the correctness issue adequately by
> doing a complete GCAC bootstrap.
OK.  Thanks for the clarification.  Consider my objection removed and
the patch approved.

jeff



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