https://bugs.winehq.org/show_bug.cgi?id=43959
--- Comment #5 from Hans Leidekker hans@meelstraat.net --- The code depends on the extra precision offered by the long double type on x86. It's an 80-bit quantity on x86 and 64-bit on ARM, AFAICT.
Before we make changes to the code I'd like to confirm that the current tests reflect ARM Windows behavior.