5f3fb64ef0
We were doing so for x3 and x7, as well, which could reduce performance for compilers that happen to perform indirect jumps via t2 (x7).