On Fri, Oct 18, 2013 at 6:34 AM, Bennie Kloosteman <[email protected]>wrote:
> But I can't help but wonder if there might be a way to explicitly identify > (at the source level) objects having the property that every in-bound > pointer to the object is either (a) a root pointer, or (b) a pointer field > of an object in the same allocation region. The goal is to deal with large > data structures that die as a group *without* requiring a full-heap GC. > So the idea is that: > > Shouldnt region analysis find this anyway ? If there is only 1 pointer > instead of 0 then see if its a root （ which LLVM needs to know anyway ). > Maybe. There are lots of kinds of region analysis, and its certainly one possibility. But at the moment I want to think harder about the story in which reference counts are only used for cross-block pointers. There are pros and cons, obviously. The thing I need to work through is what it does to the "delay to realloc" latency. But today is a day to spend mostly with my son (he's off school), so that may not happen today. shap
_______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
