The support of R_PPC_ADDR16_HI improves ld compatibility and makes things on par with RuntimeDyldELF that already implements this relocation.
Well, initially I did not, since it was simple. Yet I have just checked bfd, and ensured that it is the same.
I wanted the HA/HI tests be closer to each other for readability reasons.
For the record, R_PPC64_ADDR_HI is defined as
R_PPC64_ADDR16_HI 5 half16 #hi(S + A)
#hi(x) = ((x >> 16) & 0xffff).