[clang] [llvm] [X86] Fix x87 excess precision escaping into f32/f64 values (PR #215104)

[Trevor Gross via cfe-commits]

tgross35 wrote:

> I've tried to keep each commit as small as possible, with updated tests to show the impact of each change.

Note that this repo uses squash merging; as-is, this will all get squashed into one commit with the (rather lengthy) PR description as the commit message.

To get these committed in the style you're looking for, create a branch+PR at each commit and request the maintainers to merge them in that order. (GH stacked PRs should eventually make this less tedious, but it doesn't work cross-fork currently.)

https://github.com/llvm/llvm-project/pull/215104