[llvm] Optimize fptrunc(x)>=C1 --> x>=C2 (PR #99475)

[via llvm-commits]

kissholic wrote:

https://alive2.llvm.org/ce/z/bTf3ZW
Converting from `max float` to` ppc_fp128 ` will be `0xM47EFFFFFF0000000C150000000000000` which should pass the last finite check in the commit but fails. This error seems to be coming from the precision loss of addition of two IEEE 64-bit doubles inside of `ppc_fp128`.  This case will be bypassed temporarily before the issue fixed.

https://github.com/llvm/llvm-project/pull/99475