On Wed, Nov 23, 2016 at 02:45:23PM +0100, Michael Matz wrote: > > Shouldn't new_rtx be set to NULL_RTX if that condition is false? Otherwise > > it will be whatever make_compound_operation returned. What about the break? > > Shouldn't that be done only if the condition is true too? > > Hmm? The conditionial statement only guards the adjustment of len if that > happens to be too large to be completely contained inside new_rtx.
new_rtx shouldn't be set at all here for non-constant shifts, because those have the same problem as this new conditional tries to solve. I'm committing a similar patch, see other thread. Segher
