Re: [Haskell-cafe] fast Array operations: foldl, drop

lemming:

On Fri, 30 Nov 2007, Ketil Malde wrote:

...

Bryan O'Sullivan writes:

...

For higher dimensions, there are enough options in terms of traversal direction and what exactly e.g. a fold should fold over (single elements? lower-dimensional slices?) that a sensible API doesn't exactly leap out.

How about a 'reduce' instead of 'foldl1'? I think that if you require a commutative operator, the order doesn't matter (except for efficiency and possible rounding issues, I guess).

For what I have in mind the order of execution matters.

I also think now that slices for higher dimensional arrays are useful, anyway. If you choose a subrange of indices in the most significant dimension this would be possible without copying. It would be also possible to 'reshape' (in MatLab terms) an array without copying, as long as the number elements remain the same. So you could first transform an array of arbitrary dimension to a two-dimensional one, say

I forgot to mention this early, but possibly you could use the ndp array library. There are some people using its UArr type for (non parallel) strict arrays, that support map/fold/zip et al. http://darcs.haskell.org/packages/ndp/ This blog post recently, http://sequence.complete.org/node/371 shows at least one non-developer is using it :) Roman, what do you think -- are the unlifted, non-parallel arrays usably `beta'? -- Don