🚀 go-pugleaf

"Jonathan Benedicto" <incorrect@no.server> wrote in message
news:4291ed05$1@newsgroups.borland.com...
> "Oliver Rutsch" <orutsch@sympatec.com> wrote in message
> news:4291c5b1$1@newsgroups.borland.com...
> >
> > I would never use boolean locks "by hand". Use instead a critical
> > section for your structure. When we come to hyper threading and
> > dual/multi core systems even a bool-write is not an atomic operation
> > anymore. And what do you do if one thread finds the structure
> > locked? It polls the bool and consumes time. A critical section
> > manages this all for you.
>
> Thank you for this input. I could use critical sections, but I thought
> that the bool would be enough. I really need as little delay as
> possible when trying to find the unlocked buffer.

A critical section is the most efficient (in terms of performance) kernel
mode synchronization primitive.  If you are looking for performance, a
critical section is the way to go.  Though it might seem slower than, say, a
boolean lock, it is safe, while a boolean lock is not.

On the other hand, depending on what the other threads are doing, it might
be faster to lock the whole vector, do your search, and then unlock (as
opposed to locking each element individually).

- Dennis