gistfile1.py

import numpy as np
import pyopencl as cl
import pyopencl.array as array

ctx = cl.create_some_context()
queue = cl.CommandQueue(ctx)

a = np.random.rand(128).astype(np.float32)
a_gpu = array.to_device(queue, a)
dest_gpu = array.empty_like(a_gpu)

prg = cl.Program(ctx, """
__kernel void test(__global float *dest, __global float *src)
{
    __local float lmem[128];
    int gid = get_global_id(0);

    lmem[gid] = src[gid];
    barrier(CLK_LOCAL_MEM_FENCE);
    dest[gid] = lmem[gid];
}
""").build()

prg.test(queue, a_gpu.shape, None, dest_gpu.data, a_gpu.data)
dest = dest_gpu.get()

assert (dest == a).all()