to handle the 2d indexing, i replaced readArray m (i,j) by readMatrix m (i,j):
{-# INLINE readMatrix #-} readMatrix m ij = unsafeRead m (unsafeIndex matrixBounds ij)
matrixBounds :: ((Int,Int),(Int,Int)) matrixBounds = ((1,1),(n,n))
i'm still trying to understand why unsafeIndex is so much faster than index. to prepare for further experiments, i tried to include the default implementation for method index in my source (which includes all the other optimisations discussed here), as myindex: {-# INLINE readMatrix #-} readMatrix m ij = unsafeRead m (index matrixBounds ij) myindex b i | inRange b i = unsafeIndex b i | otherwise = error "Error in array index" now, the measurement that confuses me is this: time ./CG array 100000 readMatrix calls index: 16s readMatrix calls myindex: 9s so just calling an in-module copy of the default code for index, with bounds-check, is almost as fast as calling unsafeIndex, and almost twice as fast as calling the official index.. can anyone explain this effect? claus