def func_nplp():
np = nplp.begin_computation_stack()
a = np.arange(8)
b = a.reshape((2, 4), order='F')
knl = np.end_computation_stack([a, b])
evt, (out_a, out_b) = knl(queue)
return out_a.get(), out_b.get()
def func_nplp():
np = nplp.begin_computation_stack()
a = np.ones(10)
b = np.arange(10)
c = 2 * a + 3 * b
knl = np.end_computation_stack([a, c])
evt, (out_a, out_c) = knl(queue)
return out_a.get(), out_c.get()
def func_nplp():
np = nplp.begin_computation_stack()
A = np.arange(9).reshape((3, 3)) # noqa: N806
x = np.arange(3)
y = np.sum(A * x, axis=1)
knl = np.end_computation_stack([A, x, y])
evt, (out_A, out_x, out_y) = knl(queue)
return out_A.get(), out_x.get(), out_y.get()
def func_nplp(A_in, B_in):
np = nplp.begin_computation_stack()
A = np.argument((n, n))
B = np.argument((n, n))
A1 = A.reshape((n, n, 1))
B1 = B.reshape((1, n, n))
C1 = A1 * B1
C = np.sum(C1, axis=1)
knl = np.end_computation_stack((C, ))
evt, (out_c, ) = knl(queue, **{A.name: A_in, B.name: B_in})
return out_c.get()
def func_nplp(x_in):
np = nplp.begin_computation_stack()
x = np.argument((1000, 2))
x_new = np.argument((1000, 2))
midpoint = np.sum(x, axis=0) / x.shape[0]
is_left = x[:, 0] < midpoint[0]
is_lower = x[:, 1] < midpoint[1]
is_ll = is_lower * is_left
is_ul = (1 - is_lower) * is_left
is_lr = is_lower * (1 - is_left)
is_ur = (1 - is_lower) * (1 - is_left)
num_ll = np.sum(is_ll)
num_ul = np.sum(is_ul)
num_lr = np.sum(is_lr)
idx_ll = np.cumsum(is_ll) * (is_ll)
idx_ul = (np.cumsum(is_ul) + num_ll) * (is_ul)
idx_lr = (np.cumsum(is_lr) + num_ll + num_ul) * (is_lr)
idx_ur = (np.cumsum(is_ur) + num_ll + num_ul + num_lr) * (is_ur)
indices = idx_ll + idx_ul + idx_lr + idx_ur - 1
x_new[indices, :] = x
knl = np.end_computation_stack([
midpoint, is_lower, is_left, num_ll, num_ul, num_lr, indices, x_new
])
knl = lp.set_options(knl, return_dict=True)
knl, out_dict = knl(queue, **{x.name: x_in})
return out_dict[x_new.name].get()
import loopy as lp import numloopy as nplp np = nplp.begin_computation_stack() a = np.arange(8) b = a.reshape((2, 4), order='F') knl = np.end_computation_stack([a, b]) print(lp.generate_code_v2(knl).device_code())