Is the FPU control word setting per-thread or per-process?

I need to change the FPU control word from its default setting in a multithreaded application. Is this setting per-thread or per-process? Does it have different scopes under Mac OS X and Windows?

From stackoverflow

• It is per-thread on Windows. Not sure about OS-X, surely it does.

  Beware the nastiness you can run into if you are using libraries that expect the control word to be set at the default. Almost all of them do.