sk:
currently i'm working on stuff that looks something like this:
1 read soundfile from disk in blocks of N samples (IOCArray, hsndfile package) 2 convert to CArray with unsafeFreeze (simple O(1) cast, carray package) 3 perform FFT (CArray, fftw package) 4 convert to UArr (uvector package) 5 do some stuff with vectors
i'd like to minimize copying, and since the FFT returns a new array anyway, the only copying conversion is the one between CArray and UArr in step 4. the approach i've been following is defining a `stream' conversion for CArray, hoping that subsequent carray accesses will be fused with uvector operations without the need for allocating a vector in between. does that make sense? could this be a general strategy for avoiding copying at the boundary between the various array libraries?
If you can write a function from CArray a -> Stream a, then yep, it'll fuse, and that's a good way to avoid copying. Another way is to dive into the internals and coerce the CArray type to a MUArr. It would be helpful to see the programs people are writing with uvector, so I can polish up the API some more :) -- Don