Careful, some engines deal w/ NaN bits differently. E.g. the JVM doesn't guarantee non-NaN bits won't move (I had trouble with this at one point [0]). For specs

Careful, some engines deal w/ NaN bits differently. E.g. the JVM doesn't guarantee non-NaN bits won't move (I had trouble with this at one point [0]). For specs like WebAssembly (and their conformance tests), they expect the backends preserve the bits during some reinterpret ops[1] IIRC. 0 - https://stackoverflow.com/questions/43129365/javas-math-rint-not-behaving-as-expected-when-using-nan https://stackoverflow.com/questions/43129365/javas-math-rint... 1 - http://webassembly.org/docs/rationale/#nan-bit-pattern-nondeterminism http://webassembly.org/docs/rationale/#nan-bit-pattern-nonde...