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: [PR debug/47106] account used vars only once


> > Now onto ipa-inline-stack-growth-referenced-vars-pr47106.patch
> > 
> > it looks ok apart from the new pass_inlinable stuff, that at least sounds
> > odd and I'd want Honza to have a look here.  The testcase you say
> 
> It is because early inliner depends on toporder to properly place always_inline functions, right?
> I guess it is sane solution.  We might try to avoid recomputing inlinable in inline_parameters
> done just before early_inline pass as we have nothing that should change inlinablity.
> But in longer run we might want to add some cleanups before early inlining there, so I guess it is
> easier to not worry here.
Also because we only care abou disregard_inline_limits that is computed based on the DECL flag,
perhaps we can move 
node->local.disregard_inline_limits
      = DECL_DISREGARD_INLINE_LIMITS (node->decl);
into cgraph_analyze_function and do not introduce the new pass. 
We would end up setting the flag on uninlinable always inlines, but those are bogus anyway.

Honza


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