http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49263
--- Comment #6 from Kazumoto Kojima <kkojima at gcc dot gnu.org> 2011-06-27 05:14:36 UTC --- (In reply to comment #5) > Anyway, why not just add all the currently known-to-work cases? What are your > concerns regarding that? I can imagine that it is a maintenance burden to keep > all those definitions and special cases in the MD up-to-date (bit rot etc). Do > you have anything other than that in mind? Yep, maintenance burden but I don't mean ack/nak for anything. If it's enough fruitful, we should take that route. When it gives 5% improvement in the usual working set like as CSiBE, hundreds lines would be OK, but if it's ~0.5% or less, it doesn't look worth to add many patterns for that. > Isn't there a way to tell the combine pass not to do so, but instead first > look > deeper at what is in the MD? I don't know how to do it cleanly. > I guess this might generate wrong code for e.g. "if (x & -2)". When x has any > bits[31:1] set this must return true. The code after the peephole optimization > will look only at the lower 8 bits and would possibly return false for x = > 0xFF000000, which is wrong. So it should be satisfies_constraint_K08 only, > shouldn't it? You are right. That peephole was simply 'something like this'.
