------- Comment #10 from jakub at gcc dot gnu dot org 2007-10-31 22:45 ------- Thanks. Please see also PR31862. Looking at 456.hmmer it seems at least in the routines I have looked at there is always some unconditional write within the loop to the conditionally stored variable, so perhaps the best fix would be to not check for any kind of access, but check for store that dominates this conditional store or postdominates it and there can't be any kind of barrier point in between (volatile inline asm or non-const/pure function call), or, check if it is a local non-addressable variable.
-- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33961