On Fri, 29 Jan 2016, Jakub Jelinek wrote: > On Fri, Jan 29, 2016 at 09:40:48AM +0100, Richard Biener wrote: > > I am testing the following patch to fix a regression that we no longer > > remove some empty loops. > > Doesn't this mean that DCE will remove the clobbers as unnecessary, even > when they aren't in empty loops?
No, as with other cases (we never mark clobbers as necessary during propagation) they will be retained if all required operands are still there (SSA names used in the LHS). So the idea is that they should not keeping other stuff live but remain in the IL if all uses are still live after DCE. Richard.
