----------------------------------------------------------------------------- -- | -- Module : Data.SBV.Examples.BitPrecise.MergeSort -- Copyright : (c) Levent Erkok -- License : BSD3 -- Maintainer : erkokl@gmail.com -- Stability : experimental -- Portability : portable -- -- Symbolic implementation of merge-sort and its correctness. ----------------------------------------------------------------------------- module Data.SBV.Examples.BitPrecise.MergeSort where import Data.SBV ----------------------------------------------------------------------------- -- * Implementing Merge-Sort ----------------------------------------------------------------------------- -- | Element type of lists we'd like to sort. For simplicity, we'll just -- use 'SWord8' here, but we can pick any symbolic type. type E = SWord8 -- | Merging two given sorted lists, preserving the order. merge :: [E] -> [E] -> [E] merge [] ys = ys merge xs [] = xs merge xs@(x:xr) ys@(y:yr) = ite (x .< y) (x : merge xr ys) (y : merge xs yr) -- | Simple merge-sort implementation. We simply divide the input list -- in two two halves so long as it has at least two elements, sort -- each half on its own, and then merge. mergeSort :: [E] -> [E] mergeSort [] = [] mergeSort [x] = [x] mergeSort xs = merge (mergeSort th) (mergeSort bh) where (th, bh) = splitAt (length xs `div` 2) xs ----------------------------------------------------------------------------- -- * Proving correctness -- ${props} ----------------------------------------------------------------------------- {- $props There are two main parts to proving that a sorting algorithm is correct: * Prove that the output is non-decreasing * Prove that the output is a permutation of the input -} -- | Check whether a given sequence is non-decreasing. nonDecreasing :: [E] -> SBool nonDecreasing [] = true nonDecreasing [_] = true nonDecreasing (a:b:xs) = a .<= b &&& nonDecreasing (b:xs) -- | Check whether two given sequences are permutations. We simply check that each sequence -- is a subset of the other, when considered as a set. The check is slightly complicated -- for the need to account for possibly duplicated elements. isPermutationOf :: [E] -> [E] -> SBool isPermutationOf as bs = go as (zip bs (repeat true)) &&& go bs (zip as (repeat true)) where go [] _ = true go (x:xs) ys = let (found, ys') = mark x ys in found &&& go xs ys' -- Go and mark off an instance of 'x' in the list, if possible. We keep track -- of unmarked elements by associating a boolean bit. Note that we have to -- keep the lists equal size for the recursive result to merge properly. mark _ [] = (false, []) mark x ((y,v):ys) = ite (v &&& x .== y) (true, (y, bnot v):ys) (let (r, ys') = mark x ys in (r, (y,v):ys')) -- | Asserting correctness of merge-sort for a list of the given size. Note that we can -- only check correctness for fixed-size lists. Also, the proof will get more and more -- complicated for the backend SMT solver as 'n' increases. A value around 5 or 6 should -- be fairly easy to prove. For instance, we have: -- -- >>> correctness 5 -- Q.E.D. correctness :: Int -> IO ThmResult correctness n = prove $ do xs <- mkFreeVars n let ys = mergeSort xs return $ nonDecreasing ys &&& isPermutationOf xs ys ----------------------------------------------------------------------------- -- * Generating C code ----------------------------------------------------------------------------- -- | Generate C code for merge-sorting an array of size 'n'. Again, we're restricted -- to fixed size inputs. While the output is not how one would code merge sort in C -- by hand, it's a faithful rendering of all the operations merge-sort would do as -- described by it's Haskell counterpart. codeGen :: Int -> IO () codeGen n = compileToC (Just ("mergeSort" ++ show n)) "mergeSort" $ do xs <- cgInputArr n "xs" cgOutputArr "ys" (mergeSort xs)