There was an issue reported against D111237 because for Arm we were promoting a PHI + ZExt chain to i64, causing the existing idiv to require a library routine to perform a 64-bit division. This patch prevents promoting to a width that is wider than the RegisterBitWidth as that is likely to not be beneficial.
Thanks a lot for the quick fix. This resolves all of the issues I noticed in my 32-bit ARM Linux kernel build tests. However, I don't think I am qualified enough to approve this.