import cupy as cp

a = cp.random.rand(4096, 4096, dtype=cp.float32)
b = cp.random.rand(4096, 4096, dtype=cp.float32)
c = a @ b        # dispatches to cuBLAS under the hood

add_kernel = cp.RawKernel(r'''
extern "C" __global__
void addOne(float *data, int n) {
    int i = blockIdx.x * blockDim.x + threadIdx.x;
    if (i < n) data[i] += 1.0f;
}
''', 'addOne')

data = cp.zeros(1_000_000, dtype=cp.float32)
threads = 256
blocks = (data.size + threads - 1) // threads
add_kernel((blocks,), (threads,), (data, data.size))
