RalphAS

Yes, simpler forms would be preferable. However this is quite difficult in general (I'll refer you to the literature on the Dragon and Grisu algorithms for Float32 and Float64). We...

Have you tried using the generic `Array -> Ptr` method of `unsafe_convert` (or a replica for whatever you use for `f`, `g`, etc.)?

After calling `fma` on Float128, overflow seems generally broken for `+,-,*,/` with Float128 or Float64 arguments (tested with mixtures of huge and moderate values). I surmise that some floating point...

I'm really here for the trouble it gave me in #30 and the challenge of a level-1 puzzle to pin that down, so either of your suggestions LGTM.

It appears to be unchanged for current and nightly Julia. The example at the top is included as a broken test for Windows in the commit shown here: https://github.com/RalphAS/Quadmath.jl/actions/runs/2178209027

@vchuravy I agree that in the long run we should generate fp128 IR where feasible. I tried that approach when I was working on the Windows ABI here without success....