moving the tick register to another register, but is fine with rd.
* properly mask out the upper 32 bits of the register containing the
lower 32 bits so that the or with the upper 32bits actually works
properly. Hopefully, gcc will optimize this into code that doesn't
suck so much (it's much easier to deal with when you get to control
the argument registers...)
This commit was SVN r7410.
b3cc58b681