IEEE-754 floating point: first split or multiply for better accuracy?

Question

Which is better if I want to keep as much accuracy as possible when calculating using IEEE-754 floating point values:

a = b * c / d

or

a = b / d * c

Is there any difference? If so, does it depend on the values ​​of the input values? And, if a quantity matters, how is the best order determined when the general values ​​of the values ​​are known?

Answer 1

, , , . , , (, , 2 ^ 156 ), , , Sipe.

, / , , .

, a ( ) , , FMA ( , ).

Answer 2

. , , , , .

, , , , , . (Goldman IEEE-754.)