my brain says 'you know what, let\'s go golfing, except instead of code golf, it's code pachinko' :3

aaaaa I don't understand my brain says 'no' lol

rounded to 4 decimals

That's generally a very bad idea. Tests for all translations should use equals with delta (say, 1e-3): that will make it compatible with existing solutions, but also will get rid of the usual problems when working with floating-point numbers.

what 0_0

a = {2680, 7880}, b = {9062, 4467}, c = {4982, 6348}
expected {5574.6666999999998 ± 1e-04, 6231.6666999999998 ± 1e-04}
got {5574.6665999999996, 6231.6665999999996}

I mean the difference is 1e-4...

wow I didn't even know this one exists 0_0 nooice

yes it is. the correct way to do it is via either union type punning, or memcpy()

this is technically UB

check out the comment below by haadam from 7 years ago

please show how it is derived

Please add info how the arguments are passed, in nasm... :)