2013/3/8 GW <gw.2...@tnode.com> > ** > > Hello, > > I am wondering why is there a proposal to provide only a compatible > working NumPy CPU implementation for PyPy? Aren't NumPy operations more > suited for GPGPU (eg. OpenCL or CUDA) and supporting them would result in a > greater performance boost than just CPU? > > A great project thet seems to have implemented all aspects of > parallization, also uses JIT ideas, and has a NumPy-compatible interface > (with exchangable backend/context implementation) is Clyther ( > http://srossross.github.com/Clyther/). In my opinion it would be great to > join the efforts, combine the knowledge and code, and create something > really powerful. > Clythere itself is a pure Python library, but it is based on opencl-for-python, which is written in Cython. opencl-for-python could probably be rewritten on top of cffi, but it's a fair amount of work. Volunteers are welcome!
-- Amaury Forgeot d'Arc
_______________________________________________ pypy-dev mailing list pypy-dev@python.org http://mail.python.org/mailman/listinfo/pypy-dev
