------- Comment #6 from dominiq at lps dot ens dot fr 2010-05-15 18:01 ------- > If we can improve dependency analysis in the front end, where we (potentially) > have some more information, it should not hurt the middle-end. > > OTOH, if somebody is already doing work in this direction, it would be > a waste to duplicate. > > So... is any work going on with this?
I think Paul Thomas has something in his files, but I don't have any pointer at hand (IIRC along the line a(2:n)=a(1:n-1) replaced with a(n:2:-1)=a(n-1:1:-1)). I have given some thoughts to the problem a while ago, but did not go further than the easy cases, such as the one in the example, are easy;-) but writing something general quite difficult. In top of that what is already implemented is documented only in the code. I think if you start this problem, it would be worth to look at what is done in the middle-end (or in graphite) and what can be reused. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36928
