Thanks Andreas. Yes, I agree it is more of a numpy question since I need to control allocation of numpy array. But I really need for the host pointer in opencl to be aligned to a particular boundary. I will check out the link, hopefully it will get me an 'aligned_malloc' kind of functionality... dont mind throwing in few extra bytes either.
On Thu, Dec 20, 2012 at 2:00 PM, Andreas Kloeckner <[email protected]>wrote: > Dear Sushma, > > sushma r <[email protected]> writes: > > I would like to ensure that host memory pointers are aligned on a > > particular boundary to ensure good cache use (say by using the device > query > > MEM_BASE_ADDR_ALIGN to get best alignment for a device) > > Is there a way to specify that? Since numpy is used to define the array, > > for e.g. a = numpy.random.rand(50000).astype(numpy.float32) I don't > really > > have control over how the allocation is done. Is there pre-set alignment? > > If yes, what is it? > > This seems more like a numpy issue than a (Py)OpenCL issue. Here's a > message from the numpy list that discusses this: > > http://mail.scipy.org/pipermail/scipy-user/2009-March/020283.html > > As far as OpenCL is concerned, CL buffers are always aligned to the > largest granularity that the hardware cares about. > > Andreas >
_______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
