def test_raises_uequal_shapes_circ():
with slippy.OverRideCuda():
im = e_im('zz', (128, 128), (0.01, 0.01), 200e9, 0.3)
load_shapes = [(128, 129), (128, 129), (129, 128), (129, 128)]
circulars = [True, (False, True), True, (True, False)]
for l_s, circ in zip(load_shapes, circulars):
with npt.assert_raises(AssertionError):
loads = np.zeros(l_s)
_ = c.plan_convolve(loads, im, circular=circ)
def test_non_circ_convolve_vs_scipy():
with slippy.OverRideCuda():
for im_s, l_s in zip(im_shapes, loads_shapes):
# generate an influence matrix, pick a component which is not symmetric!
im = e_im('zx', im_s, (0.01, 0.01), 200e9, 0.3)
loads = 1000 * np.random.rand(*l_s)
scipy_result = fftconvolve(loads, im, mode='same')
conv_func = c.plan_convolve(loads, im)
slippy_result = conv_func(loads)
err_msg = f'Non circular convolution did not match scipy output for loads shape: {l_s} and IM shape: {im_s}'
npt.assert_allclose(slippy_result, scipy_result, err_msg=err_msg)
def test_raises_uequal_shapes_circ():
try:
import cupy # noqa: F401
slippy.CUDA = True
except ImportError:
return
im = e_im('zz', (128, 128), (0.01, 0.01), 200e9, 0.3)
load_shapes = [(128, 129), (128, 129), (129, 128), (129, 128)]
circulars = [True, (False, True), True, (True, False)]
for l_s, circ in zip(load_shapes, circulars):
with npt.assert_raises(AssertionError):
loads = np.zeros(l_s)
_ = c.plan_convolve(loads, im, circular=circ)
def test_dont_raise_equal_shapes_circ():
with slippy.OverRideCuda():
im = e_im('zz', (128, 128), (0.01, 0.01), 200e9, 0.3)
load_shapes = [(128, 128), (128, 129), (129, 128), (129, 128)]
circulars = [True, (True, False), False, (False, True)]
for l_s, circ in zip(load_shapes, circulars):
loads = np.zeros(l_s)
try:
_ = c.plan_convolve(loads, im, circular=circ)
except: # noqa: E722
raise AssertionError(
f"Plan convolve raised wrong error for mixed "
f"convolution load shape: {l_s}, circ: {circ}")
def test_non_circ_convolve_vs_scipy():
try:
import cupy as cp
slippy.CUDA = True
except ImportError:
return
for im_s, l_s in zip(im_shapes, loads_shapes):
# generate an influence matrix, pick a component which is not symmetric!
im = e_im('zx', im_s, (0.01, 0.01), 200e9, 0.3)
loads = 1000*np.random.rand(*l_s)
scipy_result = fftconvolve(loads, im, mode='same')
conv_func = c.plan_convolve(loads, im)
slippy_result = cp.asnumpy(conv_func(loads))
err_msg = f'Non circular convolution did not match scipy output for loads shape: {l_s} and IM shape: {im_s}'
npt.assert_allclose(slippy_result, scipy_result, err_msg=err_msg)
def test_mixed_convolve():
with slippy.OverRideCuda():
for circ in [[True, False], [False, True]]:
im = e_im('zz', (128, 128), (0.01, 0.01), 200e9, 0.3)
loads = np.zeros([128, 128])
loads[64, 64] = 1000
conv_func = c.plan_convolve(loads, im, circular=circ)
slippy_result = conv_func(loads)
loc_load = np.argmax(loads)
loc_result = np.argmax(slippy_result)
err_msg = f'Mixed circular convolution, location of load dosn\'t match displacement' \
f'for circular: {circ} \n ' \
f'expected: {np.unravel_index(loc_load, loads.shape)}, ' \
f'found: {np.unravel_index(loc_result, loads.shape)}'
assert loc_load == loc_result, err_msg
def test_non_circ_convolve_location():
with slippy.OverRideCuda():
for im_s, l_s in zip(im_shapes, loads_shapes):
# generate an influence matrix, pick a component which is not symmetric!
im = e_im('zz', im_s, (0.01, 0.01), 200e9, 0.3)
loads = np.zeros(l_s)
loads[64, 64] = 1000
conv_func = c.plan_convolve(loads, im)
slippy_result = conv_func(loads)
loc_load = np.argmax(loads)
loc_result = np.argmax(slippy_result)
err_msg = f'Non circular convolution, location of load dosn\'t match displacement' \
f'for loads shape: {l_s} and IM shape: {im_s} \n ' \
f'expected: {np.unravel_index(loc_load,l_s)}, found: {np.unravel_index(loc_result,l_s)}'
assert loc_load == loc_result, err_msg
def test_mixed_convolve():
try:
import cupy as cp
slippy.CUDA = True
except ImportError:
return
for circ in [[True, False], [False, True]]:
im = e_im('zz', (128, 128), (0.01, 0.01), 200e9, 0.3)
loads = np.zeros([128, 128])
loads[64, 64] = 1000
conv_func = c.plan_convolve(loads, im, circular=circ)
slippy_result = cp.asnumpy(conv_func(loads))
loc_load = np.argmax(loads)
loc_result = np.argmax(slippy_result)
err_msg = f'Mixed circular convolution, location of load dosn\'t match displacement' \
f'for circular: {circ} \n ' \
f'expected: {np.unravel_index(loc_load, loads.shape)}, ' \
f'found: {np.unravel_index(loc_result, loads.shape)}'
assert loc_load == loc_result, err_msg
def test_circ_convolve_location():
try:
import cupy as cp
slippy.CUDA = True
except ImportError:
return
for im_s, l_s in zip(shapes_circ, shapes_circ):
# generate an influence matrix, pick a component which is not symmetric!
im = e_im('zz', im_s, (0.01, 0.01), 200e9, 0.3)
loads = np.zeros(l_s)
loads[64, 64] = 1000
conv_func = c.plan_convolve(loads, im, circular=True)
slippy_result = cp.asnumpy(conv_func(loads))
loc_load = np.argmax(loads)
loc_result = np.argmax(slippy_result)
err_msg = f'Circular convolution, location of load dosn\'t match displacement' \
f'for loads shape: {l_s} and IM shape: {im_s} \n ' \
f'expected: {np.unravel_index(loc_load,l_s)}, found: {np.unravel_index(loc_result,l_s)}'
assert loc_load == loc_result, err_msg