[LLVMdev] One way to support unwind on x86

[Luke Dalessandro]

Nicolas Geoffray wrote:
> It's *hardly* noticeable. If your language/runtime throws a lot of 
> exceptions, that's the way to go currently (there are other 
> sophisticated techniques, but much more complex). Dwarf is great if your 
> language/runtime basically never throws exceptions. I can't give you an 
> accurate measure of the real cost of Dwarf on Java applications because, 
> unfortunately, libgcc did not optimize its dwarf info lookup on 
> dynamically registered frames. It's doing a linear search of the info, 
> and that takes *a lot* of time.

Plus, for those of us who care, the linear search is protected by a 
single mutex because there appears to be the potential that dynamically 
registered frames can be removed from the list (in addition to being 
inserted) and a mutex is the easiest way to protect against inconsistency.

Unfortunately, this means that threads that throw a lot will end up 
serializing here, plus there's a decent amount of cache-line pinging for 
read-only use of the list.

There are other ways to deal with the potential concurrent removal from 
the list that force the remover to do much more overhead (rcu, node 
pools rather than deallocation, etc), but I've never gotten around to 
actually doing it.

Luke

> 
> Nicolas
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev