List view
Write documentation and examples, compare performance to VexCL's MPI capability and top level splitting. Explore possibility of heterogeneous backends.
Overdue by 12 year(s)•Due by September 15, 2013•0/2 issues closedImplement resize and reduction operations, simple reduction first (reduce on each slave, send result to master), explore necessity/viability of more complex reductions away from master (i.e. reduce on each slave, send result to next slave and recurse, only send end result to master. Useful with large value type or very large number of slaves). Test with all possible backends, and nested with OpenMP backend. Look into getting access to a real cluster for performance measurements and debugging.
Overdue by 12 year(s)•Due by August 25, 2013•0/3 issues closedImplement state, algebra and operations for the MPI parallelized backend by refactoring the OpenMP backend, pulling out parts not specific to the parallelization method (i.e. splitting the state into sub-states).
Overdue by 12 year(s)•Due by July 28, 2013•0/2 issues closedImplement an OpenMP parallelized backend wrapping an arbitrary computational backend with state S, algebra A and operations O as `openmp_state<S>`, `openmp_algebra<A>` and `openmp_operations<O>`. Implement resize and reduction operations. Test correctness and performance with all viable backends and find user-friendly solution for parallelized system functions (i.e. allow easy use of system functions written for the computational backend with the parallelized backend, but non-linear operations like moving data between the threads' sub-states should be possible here, too, if required by user)
Overdue by 12 year(s)•Due by June 30, 2013•2/5 issues closed