> > On Sun, Aug 4, 2013 at 1:38 PM, David Jeske <dav...@gmail.com> wrote: > >> On Sat, Aug 3, 2013 at 10:17 PM, Bennie Kloosteman <bkloo...@gmail.com>wrote: >> >>> ..since regions are just glorified stacks and i presume the stack scan >>> code can be used on regions . >>> >> >> I don't follow this.. stacks don't allow dynamic allocation per >> stack-frame, while dynamic regions do. >> > Actually, stacks *do* allow dynamic allocation in many cases. That's what alloca() is all about. There are two cases where you can't do this:
1. A "churned" region. 2. A region where a younger procedure allocates a dynamically sized object in a parent region. To me this makes them seem like entirely different beasts. With >> call-stacks, allocation and deallocation is mutually bound to the >> stack-frame-return. While with dynamic-mutable-regions, we quickly can >> witness situations where we would like to reclaim independently of any >> statically analyzable motion between stack-frames or blocks. >> > You are correct. That said, you note [correctly] that the important concern is [certain cases of] dynamic allocation. Where dynamic allocation is not a concern, or where use of a region meets the constraints required for alloca(), we can and should implement *that* region within the stack. shap
_______________________________________________ bitc-dev mailing list bitc-dev@coyotos.org http://www.coyotos.org/mailman/listinfo/bitc-dev
