def test_2d_double():
a = np.finfo(np.float64).tiny * np.abs(np.random.randn(20, 20))
aa = np.copy(a)
ftz(aa)
bb = np.copy(a)
_ftz_numpy(bb)
np.testing.assert_array_equal(aa, bb)
def test_2d_single():
a = np.finfo(np.float32).tiny * np.abs(np.random.randn(20, 20))
a = np.asarray(a, dtype=np.float32)
aa = np.copy(a)
ftz(aa)
bb = np.copy(a)
_ftz_numpy(bb)
np.testing.assert_array_equal(aa, bb)
def test_norm_float():
number = (np.finfo(np.float32).tiny*2)
# pick some sizes that are both powers of two and not powers of two
# offset in both directions
for size in [2, 3, 4, 5, 7, 8, 9, 15, 16, 17, 33, 67]:
for offset in [0, 1]:
norm = np.array(np.random.randn(size), dtype=np.float32)[offset:]
result = norm.copy()
ftz(norm)
np.testing.assert_array_equal(norm, result)
def test_denorm_float():
number = (np.finfo(np.float32).tiny / 2)
# pick some sizes that are both powers of two and not powers of two
# offset in both directions
for size in [2, 3, 4, 5, 7, 8, 9, 15, 16, 17, 33, 67]:
for offset in [0, 1]:
denorm = number * np.ones(size, dtype=np.float32)[offset:]
denorm[0] = 12345.0 # make the first entry not a denorm
ftz(denorm)
np.testing.assert_array_equal(denorm[1:], np.zeros_like(denorm)[1:])
np.testing.assert_array_equal(denorm[0], np.array([12345], dtype=np.float32))
