[PATCH] Towards fixing PR47663
Jan Hubicka
hubicka@ucw.cz
Wed Feb 16 23:51:00 GMT 2011
>
> The function is called from the IPA inliner, I simply copied the above
> from the callsite which now recieves the original edge instead of a
> node mangled like above.
Ah then the code should not beneccesary in the IPA inliner too. The fibheap should
not have any edges that are already inlined in it - we remove them at the moment we
consider inlining and they should not be re-added afterwards.
It is probably remainder of the time inlier was function based not edge based.
>
> These checks seem to be to avoid issuing diagnostics for recursive calls
> (the call will reset inline_failed for non-always-inline functions).
> That's better done in tree-inline.c.
Hmm, yep, these are there to keep reason as recursive call as that seem sanier diagnostics than
the others. When we make tree-inline to test this, I guess we can drop all the logic.
>
> Well, we set the reason to CIF_UNSPECIFIED in the original call anyway ;)
Indeed, I guess back then I concluded that -Winline should not complain here most
probably for backward compatibility with previous implementation ;)
I guess we could just make tree-inline to report recursive calls.
Honza
More information about the Gcc-patches
mailing list