https://github.com/SamTebbs33 approved this pull request. Looks good to me. I do have to ask if there's a chance the CPU will see `fmov dX, dX` as a NOP and therefore not zero the high bits. https://github.com/llvm/llvm-project/pull/83107