http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45352
--- Comment #22 from Andrey Belevantsev <abel at gcc dot gnu.org> 2010-12-21 12:10:32 UTC --- Created attachment 22834 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22834 Another patch This patch should fix the last attachment from this bug and also the tests from 46521/46522. The patch solves two problems: - propagates the rescheduling bits also through empty blocks; - fixes the (wrong) assumption that if we need to stall for N cycles because of the DFA, and for M > N cycles because of the data dependency, then after M cycles the DFA will be ready to issue. Actually, we need to recheck the DFA after M cycles again. I will submit the patch tomorrow after further testing and I can commit it either until Friday or next January, so Zdenek, please let me know if there are further problems with this patch.
