gistfile1.txt

import numpy as np
import pyopencl as cl

N = 32
its = 3

np.random.seed(123)
a = np.random.rand(N).astype(np.float32)

gpu_idx = 0

platforms = cl.get_platforms()
i = 0
for platform in platforms:
    gpu_devices = platform.get_devices(device_type=cl.device_type.GPU)
    if gpu_idx < i + len(gpu_devices):
        ctx = cl.Context(devices=[gpu_devices[gpu_idx - i]])
        break
    i += len(gpu_devices)

print('context', ctx)
q = cl.CommandQueue(ctx)

mf = cl.mem_flags
a_gpu = cl.Buffer(ctx, mf.READ_WRITE | mf.COPY_HOST_PTR, hostbuf=a)

prg = cl.Program(ctx, """
kernel void _z8setValuePfif(global float* data, long data_offset, int idx, float value) {
   data = (global float*)((global char *)data + data_offset);

   label0:;
   int v1 = get_local_id(0);
    bool v2 = v1 == 0;
    if(v2) {
        goto v4;
    } else {
        goto v5;
    }
    v4:;
    long v6 = idx;
    global float* v7 = (&data[v6]);
    v7[0] = value;
        goto v5;
    v5:;
    return;
}
""").build()

print('run kernel...')
workgroupsize = 32
global_size = ((N + workgroupsize - 1) // workgroupsize) * workgroupsize
cl.enqueue_copy(q, a_gpu, a)
for it in range(its):
    prg._z8setValuePfif(q, (global_size,), (workgroupsize,), a_gpu, np.int64(0), np.int32(0), np.float32(123))


a_res = np.empty_like(a)
cl.enqueue_copy(q, a_res, a_gpu)

q.finish()
print('kernel done')

print('a_res[:5]', a_res[:5])