Hmm, so did I get that totally wrong: (from: examples/matrix-multiply.py) [..] h_a = numpy.random.rand(a_height, a_width).astype(numpy.float32) h_b = numpy.random.rand(a_width, a_height).astype(numpy.float32) h_c = numpy.empty((a_height, a_height)).astype(numpy.float32) [..] d_a_buf = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=h_a) d_b_buf = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=h_b) d_c_buf = cl.Buffer(ctx, mf.WRITE_ONLY, size=h_c.nbytes) [..] event = kernel(queue, (a_height,a_height), d_c_buf, d_a_buf, d_b_buf, cl.LocalMemory(4* block_size**2), cl.LocalMemory(4* block_size**2), local_size=(block_size, block_size)) event.wait() [..] cl.enqueue_read_buffer(queue, d_c_buf, h_c).wait() [..] So there are 2D arrays (h_a,h_b,h_c) and the corresponding cl.Buffers d_... They are send to the kernel here: 'event = kernel(queue, (a_height,a_height), d_c_buf, d_a_buf, d_b_buf,...' But how is that connected to: __kernel __attribute__((reqd_work_group_size(16,16,1))) void matrixMul( __global float* C, __global float* A, __global float* B, __local float* As, __local float* Bs) My C experience is a 'bit' rusty so please: what does 'float* A' do and how does that differ from 'float *A'? Thanks Markus Am 12.02.2010 um 19:01 schrieb Rohit Garg:

Thanks for the answers. My confusion resulted from a gross misunderstanding about how opencl works. I hope I have it sorted out now. If there is ever an 'opencl for dummies' book I'll buy it. :) thanks Markus