def test_overlaps(self, refinement_boxes, expected):
print("GeometryTest.test_overlaps")
test_id = self._testMethodName.split("_")[-1]
dim, interp_order, nbr_cells = (2, 1, [20] * 2)
hierarchy = self.setup_hierarchy(dim,
interp_order,
nbr_cells,
refinement_boxes,
quantities="Bx")
level_overlaps = hierarchy_overlaps(hierarchy)
for ilvl, lvl in enumerate(hierarchy.patch_levels):
if ilvl not in expected:
continue
self.assertEqual(len(expected[ilvl]), len(level_overlaps[ilvl]))
for exp, actual in zip(expected[ilvl], level_overlaps[ilvl]):
self.assertEqual(actual["box"], exp["box"])
self.assertTrue((np.asarray(actual["offset"]) == np.asarray(
exp["offset"])).all())
if 1 in level_overlaps:
fig = hierarchy.plot_2d_patches(
1, [p.box for p in hierarchy.level(1).patches])
fig.savefig("hierarchy_2d_lvl_" + str(1) + "_simple_test_" +
str(test_id) + ".png")
def test_large_patchoverlaps(self, expected):
print(f"GeometryTest.{self._testMethodName}")
test_id = self._testMethodName.split("_")[-1]
dim, interp_order, nbr_cells = (2, 1, [20] * 2)
hierarchy = self.setup_hierarchy(
dim, interp_order, nbr_cells, {}, quantities="Bx", largest_patch_size=20
)
level_overlaps = hierarchy_overlaps(hierarchy)
ilvl = 0
lvl = hierarchy.level(ilvl)
overlap_boxes = []
self.assertEqual(len(expected), len(level_overlaps[ilvl]))
for exp, actual in zip(expected, level_overlaps[ilvl]):
self.assertEqual(actual["box"], exp["box"])
self.assertTrue(
(np.asarray(actual["offset"]) == np.asarray(exp["offset"])).all()
)
overlap_boxes += [actual["box"]]
fig = hierarchy.plot_2d_patches(
ilvl,
collections=[
{
"boxes": overlap_boxes,
"facecolor": "yellow",
},
{
"boxes": [p.box for p in hierarchy.level(ilvl).patches],
"facecolor": "grey",
},
],
)
fig.savefig(f"hierarchy_2d_lvl_0_large_patch_test_{test_id}.png")
def test_overlaps(self):
hierarchy = self.basic_hierarchy()
expected = {0 :
# Middle overlap, for all quantities
[ {"box":Box(28, 38),'offset':(0,0)},
{"box": Box(28, 37),"offset":(0,0)},
{"box": Box(28, 37),"offset":(0,0)},
{"box": Box(28, 37),'offset': (0,0)},
{"box": Box(28, 38),"offset":(0,0)},
{"box": Box(28, 38),"offset":(0,0)},
{"box": Box(32,33),"offset":(0,0)},
# left side overlap with periodicity, for all quantities
{"box": Box(-5, 5),"offset":(0,-65)},
# right side overlap with periodicity, for all quantities
{"box": Box(60, 70),"offset": (65,0)},
{"box": Box(-5, 4),"offset":(0,-65)},
{"box": Box(60, 69),"offset": (65,0 )},
{"box": Box(-5, 4),"offset":(0,-65)},
{"box": Box(60, 69),"offset": (65,0)},
{"box": Box(-5, 4),"offset":(0,-65)},
{"box": Box(60, 69),"offset":(65, 0)},
{"box": Box(-5, 5),"offset":(0,-65)},
{"box": Box(60, 70),"offset":(65, 0)},
{"box": Box(-5, 5),"offset":(0,-65)},
{"box": Box(60, 70),"offset":(65, 0)},
{"box": Box(-1,0),"offset":(0,-65)},
{"box": Box(64, 65),"offset":(65 ,0)}
],
1: # level 1
[
{"box": Box(59, 65),'offset': (0, 0)},
{"box": Box(59, 64),"offset": (0, 0)},
{"box": Box(59, 64),"offset": (0, 0)},
{"box": Box(59, 64),'offset': (0, 0)},
{"box": Box(59, 65),"offset": (0, 0)},
{"box": Box(59, 65),"offset": (0, 0)}
]
}
overlaps = hierarchy_overlaps(hierarchy)
for ilvl, lvl in enumerate(hierarchy.patch_levels):
self.assertEqual(len(expected[ilvl]), len(overlaps[ilvl]))
for exp, actual in zip(expected[ilvl], overlaps[ilvl]):
act_box = actual["box"]
act_offset = actual["offset"]
exp_box = exp["box"]
exp_offset = exp["offset"]
self.assertEqual(act_box, exp_box)
self.assertEqual(act_offset, exp_offset)
def _test_overlapped_particledatas_have_identical_particles(self, dim, interp_order, refinement_boxes):
print("test_overlapped_particledatas_have_identical_particles")
print("interporder : {}".format(interp_order))
from copy import copy
time_step_nbr=3
time_step=0.001
diag_outputs=f"phare_overlapped_particledatas_have_identical_particles_{self.ddt_test_id()}"
datahier = self.getHierarchy(interp_order, refinement_boxes, "particles", diag_outputs=diag_outputs,
time_step=time_step, time_step_nbr=time_step_nbr)
for time_step_idx in range(time_step_nbr + 1):
coarsest_time = time_step_idx * time_step
overlaps = hierarchy_overlaps(datahier, coarsest_time)
for ilvl, lvl in datahier.patch_levels.items():
print("testing level {}".format(ilvl))
for overlap in overlaps[ilvl]:
pd1, pd2 = overlap["pdatas"]
box = overlap["box"]
offsets = overlap["offset"]
self.assertEqual(pd1.quantity, pd2.quantity)
if "particles" in pd1.quantity:
# the following uses 'offset', we need to remember that offset
# is the quantity by which a patch has been moved to detect
# overlap with the other one.
# so shift by +offset when evaluating patch data in overlap box
# index space, and by -offset when we want to shift box indexes
# to the associated patch index space.
# overlap box must be shifted by -offset to select data in the patches
part1 = copy(pd1.dataset.select(boxm.shift(box, -offsets[0])))
part2 = copy(pd2.dataset.select(boxm.shift(box, -offsets[1])))
idx1 = np.argsort(part1.iCells + part1.deltas)
idx2 = np.argsort(part2.iCells + part2.deltas)
# if there is an overlap, there should be particles
# in these cells
assert(len(idx1) >0)
assert(len(idx2) >0)
print("respectively {} and {} in overlaped patchdatas".format(len(idx1), len(idx2)))
# particle iCells are in their patch AMR space
# so we need to shift them by +offset to move them to the box space
np.testing.assert_array_equal(part1.iCells[idx1]+offsets[0], part2.iCells[idx2]+offsets[1])
self.assertTrue(np.allclose(part1.deltas[idx1], part2.deltas[idx2], atol=1e-12))
self.assertTrue(np.allclose(part1.v[idx1,0], part2.v[idx2,0], atol=1e-12))
self.assertTrue(np.allclose(part1.v[idx1,1], part2.v[idx2,1], atol=1e-12))
self.assertTrue(np.allclose(part1.v[idx1,2], part2.v[idx2,2], atol=1e-12))
def _test_overlapped_particledatas_have_identical_particles(
self, ndim, interp_order, refinement_boxes, ppc=100, **kwargs):
print(
"test_overlapped_particledatas_have_identical_particles, interporder : {}"
.format(interp_order))
from copy import copy
time_step_nbr = 3
time_step = 0.001
diag_outputs = f"phare_overlapped_particledatas_have_identical_particles/{ndim}/{interp_order}/{self.ddt_test_id()}"
datahier = self.getHierarchy(interp_order,
refinement_boxes,
"particles",
diag_outputs=diag_outputs,
ndim=ndim,
time_step=time_step,
time_step_nbr=time_step_nbr,
nbr_part_per_cell=ppc,
**kwargs)
for time_step_idx in range(time_step_nbr + 1):
coarsest_time = time_step_idx * time_step
overlaps = hierarchy_overlaps(datahier, coarsest_time)
for ilvl, lvl in datahier.patch_levels.items():
print("testing level {}".format(ilvl))
for overlap in overlaps[ilvl]:
pd1, pd2 = overlap["pdatas"]
box = overlap["box"]
offsets = overlap["offset"]
self.assertEqual(pd1.quantity, pd2.quantity)
if "particles" in pd1.quantity:
# the following uses 'offset', we need to remember that offset
# is the quantity by which a patch has been moved to detect
# overlap with the other one.
# so shift by +offset when evaluating patch data in overlap box
# index space, and by -offset when we want to shift box indexes
# to the associated patch index space.
# overlap box must be shifted by -offset to select data in the patches
part1 = copy(
pd1.dataset.select(
boxm.shift(box, -np.asarray(offsets[0]))))
part2 = copy(
pd2.dataset.select(
boxm.shift(box, -np.asarray(offsets[1]))))
# periodic icell overlaps need shifting to be the same
part1.iCells = part1.iCells + offsets[0]
part2.iCells = part2.iCells + offsets[1]
self.assertEqual(part1, part2)
def base_test_overlaped_fields_are_equal(self, datahier, coarsest_time):
checks = 0
for ilvl, overlaps in hierarchy_overlaps(datahier,
coarsest_time).items():
for overlap in overlaps:
pd1, pd2 = overlap["pdatas"]
box = overlap["box"]
offsets = overlap["offset"]
self.assertEqual(pd1.quantity, pd2.quantity)
if pd1.quantity == 'field':
# we need to transform the AMR overlap box, which is thus
# (because AMR) common to both pd1 and pd2 into local index
# boxes that will allow to slice the data
# the patchData ghost box that serves as a reference box
# to transfrom AMR to local indexes first needs to be
# shifted by the overlap offset associated to it
# this is because the overlap box has been calculated from
# the intersection of possibly shifted patch data ghost boxes
loc_b1 = boxm.amr_to_local(
box, boxm.shift(pd1.ghost_box, offsets[0]))
loc_b2 = boxm.amr_to_local(
box, boxm.shift(pd2.ghost_box, offsets[1]))
data1 = pd1.dataset
data2 = pd2.dataset
if box.ndim == 1:
slice1 = data1[loc_b1.lower[0]:loc_b1.upper[0] + 1]
slice2 = data2[loc_b2.lower[0]:loc_b2.upper[0] + 1]
if box.ndim == 2:
slice1 = data1[loc_b1.lower[0]:loc_b1.upper[0] + 1,
loc_b1.lower[1]:loc_b1.upper[1] + 1]
slice2 = data2[loc_b2.lower[0]:loc_b2.upper[0] + 1,
loc_b2.lower[1]:loc_b2.upper[1] + 1]
try:
assert slice1.dtype == np.float64
np.testing.assert_equal(slice1, slice2)
checks += 1
except AssertionError as e:
print("AssertionError", pd1.name, e)
if self.rethrow_:
raise e
return diff_boxes(slice1, slice2, box)
return checks
def test_periodic_overlaps(self, refinement_boxes, expected):
dim, interp_order, nbr_cells = (1, 1, 20)
hierarchy = self.setup_hierarchy(
dim, interp_order, nbr_cells, refinement_boxes, quantities="Bx"
)
overlaps = hierarchy_overlaps(hierarchy)
for ilvl, lvl in enumerate(hierarchy.patch_levels):
if ilvl not in expected:
continue
self.assertEqual(len(expected[ilvl]), len(overlaps[ilvl]))
for exp, actual in zip(expected[ilvl], overlaps[ilvl]):
self.assertEqual(actual["box"], exp["box"])
self.assertEqual(actual["offset"], exp["offset"])
def _test_overlaped_fields_are_equal(self, time_step, time_step_nbr, datahier):
check=0
for time_step_idx in range(time_step_nbr + 1):
coarsest_time = time_step_idx * time_step
for ilvl, overlaps in hierarchy_overlaps(datahier, coarsest_time).items():
for overlap in overlaps:
pd1, pd2 = overlap["pdatas"]
box = overlap["box"]
offsets = overlap["offset"]
self.assertEqual(pd1.quantity, pd2.quantity)
if pd1.quantity == 'field':
check+=1
# we need to transform the AMR overlap box, which is thus
# (because AMR) common to both pd1 and pd2 into local index
# boxes that will allow to slice the data
# the patchData ghost box that serves as a reference box
# to transfrom AMR to local indexes first needs to be
# shifted by the overlap offset associated to it
# this is because the overlap box has been calculated from
# the intersection of possibly shifted patch data ghost boxes
loc_b1 = boxm.amr_to_local(box, boxm.shift(pd1.ghost_box, offsets[0]))
loc_b2 = boxm.amr_to_local(box, boxm.shift(pd2.ghost_box, offsets[1]))
data1 = pd1.dataset
data2 = pd2.dataset
slice1 = data1[loc_b1.lower[0]:loc_b1.upper[0] + 1]
slice2 = data2[loc_b2.lower[0]:loc_b2.upper[0] + 1]
try:
np.testing.assert_allclose(slice1, slice2, atol=1e-6)
except AssertionError as e:
print("error", coarsest_time, overlap)
raise e
self.assertGreater(check, time_step_nbr)
self.assertEqual(check % time_step_nbr, 0)
def test_overlaped_fields_are_equal(self, interp_order, refinement_boxes):
print("test_overlaped_fields_are_equal")
hier = self.getHierarchy(interp_order, refinement_boxes, "b")
overlaps = hierarchy_overlaps(hier)
check = 0
for ilvl, lvl in hier.levels().items():
for overlap in overlaps[ilvl]:
pd1, pd2 = overlap["pdatas"]
box = overlap["box"]
offsets = overlap["offset"]
self.assertEqual(pd1.quantity, pd2.quantity)
if pd1.quantity == 'field':
check += 1
# we need to transform the AMR overlap box, which is thus
# (because AMR) common to both pd1 and pd2 into local index
# boxes that will allow to slice the data
# the patchData ghost box that serves as a reference box
# to transfrom AMR to local indexes first needs to be
# shifted by the overlap offset associated to it
# this is because the overlap box has been calculated from
# the intersection of possibly shifted patch data ghost boxes
loc_b1 = boxm.amr_to_local(
box, boxm.shift(pd1.ghost_box, offsets[0]))
loc_b2 = boxm.amr_to_local(
box, boxm.shift(pd2.ghost_box, offsets[1]))
data1 = pd1.dataset
data2 = pd2.dataset
slice1 = data1[loc_b1.lower[0]:loc_b1.upper[0] + 1]
slice2 = data2[loc_b2.lower[0]:loc_b2.upper[0] + 1]
self.assertTrue(np.allclose(slice1, slice2, atol=1e-12))
self.assertTrue(check > 0)
def test_patch_ghost_particle_are_clone_of_overlaped_patch_domain_particles(
self, interp_order, refinement_boxes):
datahier = self.getHierarchy(interp_order, refinement_boxes,
"particles")
print(
"test_patch_ghost_particle_are_clone_of_overlaped_patch_domain_particles"
)
print("interporder : {}".format(interp_order))
overlaps = hierarchy_overlaps(datahier)
for ilvl, lvl_overlaps in overlaps.items():
print("level {}".format(ilvl))
for overlap in lvl_overlaps:
if ilvl != 0: #only root level tested here
continue
if "particles" not in overlap["pdatas"][0].quantity:
continue
ref_pd, cmp_pd = overlap["pdatas"]
box = overlap["box"]
print(
"overlap box : {}, reference patchdata box : {}, ghostbox {},"
" comp. patchdata box : {} ghostbox {}".format(
box, ref_pd.box, ref_pd.ghost_box, cmp_pd.box,
cmp_pd.ghost_box))
offsets = overlap["offset"]
# first let's shift the overlap box over the AMR
# indices of the patchdata. The box has been created
# by shifting the patchdata ghost box by 'offset' so here
# the box is shifted by -offset to get over patchdata
shift_refbox, shift_cmpbox = [
boxm.shift(box, -off) for off in offsets
]
# the overlap box overlaps both ghost and domain cells
# we need to extract the domain ones to later select domain
# particles
ovlped_refdom = ref_pd.box * shift_refbox
ovlped_cmpdom = cmp_pd.box * shift_cmpbox
# on lvl 0 patches are adjacent
# therefore the overlap box must overlap the
# patchData box. 1 cell in interporder1, 2 cells for higher
assert (ovlped_cmpdom is not None)
assert (ovlped_refdom is not None)
refdomain = ref_pd.dataset.select(ovlped_refdom)
cmpdomain = cmp_pd.dataset.select(ovlped_cmpdom)
# now get the ghost cells of each patch data overlaped by
# the overlap box. To do this we need to intersect the shifted
# overlap box with the patchdata ghost box, and remove interior cells
# note that in 1D we don't expect remove to return more than 1 box, hence [0]
ovlped_refghost = boxm.remove(ref_pd.ghost_box * shift_refbox,
ref_pd.box)[0]
ovlped_cmpghost = boxm.remove(cmp_pd.ghost_box * shift_cmpbox,
cmp_pd.box)[0]
refghost = ref_pd.dataset.select(ovlped_refghost)
cmpghost = cmp_pd.dataset.select(ovlped_cmpghost)
print("ghost box {} has {} particles".format(
ovlped_refghost, len(refghost.iCells)))
print("ghost box {} has {} particles".format(
ovlped_cmpghost, len(cmpghost.iCells)))
# before comparing the particles we need to be sure particles of both patchdatas
# are sorted in the same order. We do that by sorting by x position
sort_refdomain_idx = np.argsort(refdomain.iCells +
refdomain.deltas)
sort_cmpdomain_idx = np.argsort(cmpdomain.iCells +
cmpdomain.deltas)
sort_refghost_idx = np.argsort(refghost.iCells +
refghost.deltas)
sort_cmpghost_idx = np.argsort(cmpghost.iCells +
cmpghost.deltas)
assert (sort_refdomain_idx.size != 0)
assert (sort_cmpdomain_idx.size != 0)
assert (sort_refdomain_idx.size != 0)
assert (sort_cmpghost_idx.size != 0)
np.testing.assert_allclose(
refdomain.deltas[sort_refdomain_idx],
cmpghost.deltas[sort_cmpghost_idx],
atol=1e-12)
np.testing.assert_allclose(
cmpdomain.deltas[sort_cmpdomain_idx],
refghost.deltas[sort_refghost_idx],
atol=1e-12)
