Il 01/07/2013 22:30, Richard Henderson ha scritto: > On 07/01/2013 12:25 PM, Peter Maydell wrote: >> Does any OS have a __thread which compiles but is broken, or can >> we just have a configure test for this? That would let MacOSX+clang >> use __thread. > > I suspect that this will work. Some targets may succeed in using gcc's > "emutls" path, which while slower than TLS is pretty much exactly the pthread > get/setcontext fallback that's been proposed elsewhere on this list.
We do not want to hit emutls on Windows, but that can be done simply by reordering the three implementation. Paolo
