See https://lists.llvm.org/pipermail/llvm-dev/2018-June/124164.html for the full rationale.
The tl;dr story is that this causes jumps in the emitted line tables, even at -O0.
We could at some point consider more fancy solutions to preserve locations for them, but it doesn't seem to be worth the effort for now.
@davide It looks like there are a number of sites in this file which require the same fix (e.g; the isVector case, the G_FCONSTANT case).
What's the plan for tackling that? @dsanders introduced a lost location observer for GISel; we could use that + -mir-debugify to find issues.
For this file in particular, I think we should consider introducing a buildInstrNoLoc method and using that where appropriate.