For general purpose threading, I'd go with boost threads. They're being adopted in to the next C++ standard (with a handful of small changes) so you'll be future proofing yourself somewhat by learning them.

OpenMP is best for loop-based parallelism. It can be used for other stuff too but it gets a little trickier (version 3 of the spec has made inroads in to more general parallelism, I hear). For example, OpenMP would be a great tool to use for basic ray-tracing, fractal rendering or any other "embarrassingly parallel" algorithm. The other thing about OpenMP is that it is not available everywhere, though it is available in newer versions of MSVC (>= v8) and GCC (>= 4.2), which realistically will get you pretty far.

Intel's threading building blocks made a bit of a stir a few months back (probably because it was open sourced more than anything else), but you could also look at that. I haven't had time to, yet.

SDL provides threads too, but since you're using Direct3D there is probably no point in using SDL for your project.

OpenMP in visual studio appearantly forces to have to provide redistributables with your project that must work with manifests, not really nice for the installer. Does anyone know a way to do it without?

Quote:Original post by texel3d
I will use multi-threading for Virtual texture and maybe physics (later).
Does boost has some interesting things that Windows threads do not have ?

Condition variables is the big one. I think most would agree that programming using condition variables as popularised by pthreads, is much less error prone than event variables as found in the windows API.

scoped locks are another incredibly important addition that help with correctness in the face of exceptions. Since the Win32 api is not C++, it cannot provide such a facility that inherently depends on destructors). Boost also has a thread_group class, which is some what easier to use than WaitForMultipleObjects.

By employing boost bind and boost function, it is also easy to call an arbitrary function in another thread. This is generally much more tedious with C-level APIs such as pthreads and Win32. I took this another stage further in my async library, based on boost threads.

There's also a very flexible thread pool implementation built on top of boost threads at http://threadpool.sourceforge.net/.

And of course, by using thr boost stuff and a little care you get portability for free, which may or may not matter to you.