Bonjour Andreas, geat news! I'm very happy to see pyopencl advancing. Several people in my field (control systems for tokamaks) have developed nice prototypes using hand produced static code and CUDA.
I'm not an AMD fanboy, but using CUDA means 100% product locking and zero portability. For the moment only NVIDIA has serious GPU hardware ready (and I like it), but this situation will change in a few months. Static, hand produced, code is another big cultural issue. When I talk about "metaprogramming" most of the people in the room (PhD++, physicists, engineers, etc.) does not like the concept because they have spent most of their professional life developing software using dinosaur-like methods. Does we need to wait their natural exintion? I hope not: we don't have so much time. A big tanks to the people involved with pyopencl (and pycuda too ;-) ) to keep this tool alive and, above all, realeased under an OPEN SOURCE license ! Thanks a lot Andreas! Simone Mannori www.scicos.org, www.scicoslab.org http://www.brasimone.enea.it/ On 26 June 2011 03:31, Andreas Kloeckner <[email protected]> wrote: > Hi all, > > Version 2011.1 has made it out, finally! The list of changes is copied > below. This is a big release, and a huge step forward for PyOpenCL. As > time progresses, I'll try to move to smaller, more frequent > releases. Have fun, and if you find bugs in the release code, please > speak up on the list. > > Andreas > > Detailed list of changes: > > * All is_blocking parameters now default to True to avoid > crashy-by-default behavior. (suggested by Jan Meinke) In particular, > this change affects pyopencl.enqueue_read_buffer(), > pyopencl.enqueue_write_buffer(), pyopencl.enqueue_read_buffer_rect(), > pyopencl.enqueue_write_buffer_rect(), pyopencl.enqueue_read_image(), > pyopencl.enqueue_write_image(), pyopencl.enqueue_map_buffer(), > pyopencl.enqueue_map_image(). > > * Add pyopencl.reduction. > * Add Reductions. > * Add pyopencl.scan. > * Add pyopencl.MemoryObject.get_host_array(). > * Deprecate context arguments of pyopencl.array.to_device(), > pyopencl.array.zeros(), pyopencl.array.arange(). > * Make construction of pyopencl.array.Array more flexible (cqa argument.) > * Add Memory Pools. > * Add vector types, see pyopencl.array.vec. > * Add pyopencl.array.Array.strides, pyopencl.array.Array.flags. Allow the > creation of arrys in C and Fortran order. > * Add pyopencl.enqueue_copy(). Deprecate all other transfer functions. > * Add support for numerous extensions, among them device fission. > * Add a compiler cache. > > > > _______________________________________________ > PyOpenCL mailing list > [email protected] > http://lists.tiker.net/listinfo/pyopencl > > _______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
