Excellent! Now here comes my number 1 feature request for the next version: IO/ST arrays/computations in accelerate, so that I can implement stuff that can work on a 1GB device array with interleaved IO, without having to download and-re-upload that array every time. Keep up the good work!