HDL Coder only supports division for fixed-point numbers, not for doubles. Fixed point numbers are denoted by specifying a word length (in bits), a fraction length (in bits), and whether the value is signed. The choices for these parameters determine what values the fixed point type can hold, and with what precision.
In x = a/b, a is the dividend, b is the divisor, and x is the output. The error message is stating that fraction(x) == fraction(a) - fraction(b). So, if the dividend (a) has 8 bits of fraction, and the divisor (b) has 3, then the result (x) must have 5 fractional bits.
