In the 32-bit section of C2MacroAssembelr::fast_lock() we first CAS a stack-pointer and after this has been successful, fetch the current thread and store that into the stack-lock. With JDK-8291555 we can simply use the thread-register that's passed into fast_lock() and can avoid the intermediate store.
Also, fix the clearing of tmp register to use xorl, as the comment says.