[snek_case]

> Your only remaining option is to use a totally IEEE-754 compliant algorithm that is guaranteed to work the same on every CPU that implements IEEE-754 floats

OK but, doesn't intel internally use 80 bits of precision for float64 computations? If that's the case, you can't even guarantee that float64 multiplication and addition would behave the same on say x86 vs arm.

[josefx]

That was the old x87 stack, that has been (almost) completely replaced by SSE and other vector instruction sets that operate on 32/64 sized elements. Unless you need 80 bit floats or have a mountain of ancient 32 bit intel CPUs without SSE support lying around you should get the expected rounding behavior by default.

[kevingadd]

The higher internal precision only applies to the classic x87 operations under most circumstances, AFAIK.