Documentation

The uninterpreted implementation of our 2x2 multiplier. We simply receive two 2-bit values, and return the high and the low bit of the resulting multiplication via two uninterpreted functions that we called mul22_hi and mul22_lo. Note that there is absolutely no computation going on here, aside from simply passing the arguments to the uninterpreted functions and stitching it back together.

NB. While definining mul22_lo we used our domain knowledge that the low-bit of the multiplication only depends on the low bits of the inputs. However, this is merely a simplifying assumption; we could have passed all the arguments as well.

Synthesize a 2x2 multiplier. We use 8-bit inputs merely because that is the lowest bit-size SBV supports but that is more or less irrelevant. (Larger sizes would work too.) We simply assert this for all input values, extract the bottom two bits, and assert that our "uninterpreted" implementation in mul22 is precisely the same. We have:

>>> sat synthMul22
Satisfiable. Model:
  mul22_hi :: Bool -> Bool -> Bool -> Bool -> Bool
  mul22_hi False True  True  False = True
  mul22_hi False True  True  True  = True
  mul22_hi True  False False True  = True
  mul22_hi True  False True  True  = True
  mul22_hi True  True  False True  = True
  mul22_hi True  True  True  False = True
  mul22_hi _     _     _     _     = False

  mul22_lo :: Bool -> Bool -> Bool
  mul22_lo True True = True
  mul22_lo _    _    = False

It is easy to see that the low bit is simply the logical-and of the low bits. It takes a moment of staring, but you can see that the high bit is correct as well: The logical formula is a1b xor a0b1, and if you work out the truth-table presented, you'll see that it is exactly that. Of course, you can use SBV to prove this. First, define the model we were given to make it symbolic: