On Mon, Jun 30, 2014 at 03:53:55PM -0400, Carter Schonwald wrote:
boxity has nothing to do with parallelism. none :)
No, but it may affect /when/ haskell computes a value, and if the computation is delayed until the value is requested, then --hey-- we just missed the parallel opportunity. Quite simply, Repa's computeP does not work as expected out of the box when it is applied to a boxed return type. If that amounts to «nothing to do with» in your book, that's ok; it is still a problem which requires a solution. BTW, I thought a bit more and the solution using NFData suggested in http://stackoverflow.com/questions/16097418/parallel-repa-code-doesnt-create... was easier to implement than I feared, and more importantly, it works. I don't find it entirely satisfactory though, so any hints on how to use unboxed arrays with composite constituent types are still very welcome. -- :-- Hans Georg