On Tue, Aug 4, 2015 at 10:16 AM, Mike Stump <mikest...@comcast.net> wrote: > On Aug 4, 2015, at 8:44 AM, H.J. Lu <hjl.to...@gmail.com> wrote: >> On Tue, Aug 4, 2015 at 8:40 AM, Mike Stump <mikest...@comcast.net> wrote: >>> On Aug 4, 2015, at 5:30 AM, H.J. Lu <hjl.to...@gmail.com> wrote: >>>> Where does this feature belong? >>> >>> I prefer the middle end. >> >> Any comments on my middle-end patch? > > So, if the answer is the same as frame_address (0), why not have the fallback > just expand to that? Then, one can use this builtin everywhere that frame > address is used today. People that want a faster, tighter port can then > implement the hook and achieve higher performance.
The motivation of __builtin_stack_top is that frame_address requires a frame pointer register, which isn't desirable for x86. __builtin_stack_top doesn't require a frame pointer register. -- H.J.
