[clang] [llvm] [APFloat] Signal overflow when rounding delivers the largest finite value (PR #220020)

[Hung-Kuan Tseng via cfe-commits]

Tim096 wrote:

Thanks for the update, @Garfield1002. I checked it on top of current main: the APFloat unit tests pass, and with `PPCDoubleDouble`, `convertFromAPInt(2^1100)`, `max + max` and `max * max` under `rmTowardZero` now report `opOverflow` like `IEEEdouble` does. One nit in the clang test: the `// {{pragma STDC FENV_ROUND is not supported}}` after the pragma isn't a `-verify` directive (there's no `expected-warning`), so nothing checks it and it can go. I didn't build clang, so I haven't run that test myself.

https://github.com/llvm/llvm-project/pull/220020