def test_default_rules(self, order):
"""
Test that the default replacement rules return the same
as standard FD.
"""
grid = Grid(shape=(20, 20))
u0 = TimeFunction(name='u', grid=grid, time_order=order, space_order=order)
u1 = TimeFunction(name='u', grid=grid, time_order=order, space_order=order,
coefficients='symbolic')
eq0 = Eq(-u0.dx+u0.dt)
eq1 = Eq(u1.dt-u1.dx)
assert(eq0.evalf(_PRECISION).__repr__() == eq1.evalf(_PRECISION).__repr__())
def test_dynamic_nthreads(self):
grid = Grid(shape=(16, 16, 16))
f = TimeFunction(name='f', grid=grid)
op = Operator(Eq(f.forward, f + 1.), dle='openmp')
# Check num_threads appears in the generated code
# Not very elegant, but it does the trick
assert 'num_threads(nthreads)' in str(op)
# Check `op` accepts the `nthreads` kwarg
op.apply(time=0)
op.apply(time_m=1, time_M=1, nthreads=4)
assert np.all(f.data[0] == 2.)
# Check the actual value assumed by `nthreads`
assert op.arguments(time=0)['nthreads'] == NThreads.default_value()
assert op.arguments(time=0,
nthreads=123)['nthreads'] == 123 # user supplied
def test_precomputed2(self):
shape = (101, 101)
grid = Grid(shape=shape)
x, y = grid.dimensions
r = 2 # Constant for linear interpolation
# because we interpolate across 2 neighbouring points in each dimension
nt = 10
m = TimeFunction(name="m",
grid=grid,
space_order=0,
save=None,
time_order=1)
m.data[:] = 0.0
m.data[:, 40, 40] = 1.0
matrix = scipy.sparse.eye(1, dtype=np.float32)
sf = MatrixSparseTimeFunction(name="s",
grid=grid,
r=r,
matrix=matrix,
nt=nt)
# Lookup the exact point
sf.gridpoints.data[0, 0] = 40
sf.gridpoints.data[0, 1] = 40
sf.interpolation_coefficients[x].data[0, 0] = 1.0
sf.interpolation_coefficients[x].data[0, 1] = 2.0
sf.interpolation_coefficients[y].data[0, 0] = 1.0
sf.interpolation_coefficients[y].data[0, 1] = 2.0
sf.data[:] = 0.0
step = [Eq(m.forward, m)]
interp = sf.interpolate(m)
op = Operator(step + interp)
sf.manual_scatter()
op(time_m=0, time_M=0)
sf.manual_gather()
assert sf.data[0, 0] == 1.0
def setup(self):
grid = Grid(shape=(5, 5, 5))
funcs = [Function(name='f%d' % n, grid=grid) for n in range(30)]
tfuncs = [TimeFunction(name='u%d' % n, grid=grid) for n in range(30)]
stfuncs = [
SparseTimeFunction(name='su%d' % n, grid=grid, npoint=1, nt=100)
for n in range(30)
]
v = TimeFunction(name='v', grid=grid, space_order=2)
eq = Eq(v.forward,
v.laplace + sum(funcs) + sum(tfuncs) + sum(stfuncs),
subdomain=grid.interior)
self.op = Operator(eq, opt='noop')
# Allocate data, populate cached properties, etc.
self.op.arguments(time_M=98)
def test_discarding_runs():
grid = Grid(shape=(64, 64, 64))
f = TimeFunction(name='f', grid=grid)
op = Operator(Eq(f.forward, f + 1.), dle=('advanced', {'openmp': True}))
op.apply(time=100, nthreads=4, autotune='aggressive')
assert op._state['autotuning'][0]['runs'] == 18
assert op._state['autotuning'][0]['tpr'] == options['squeezer'] + 1
assert len(op._state['autotuning'][0]['tuned']) == 3
assert op._state['autotuning'][0]['tuned']['nthreads'] == 4
# With 1 < 4 threads, the AT eventually tries many more combinations
op.apply(time=100, nthreads=1, autotune='aggressive')
assert op._state['autotuning'][1]['runs'] == 25
assert op._state['autotuning'][1]['tpr'] == options['squeezer'] + 1
assert len(op._state['autotuning'][1]['tuned']) == 3
assert op._state['autotuning'][1]['tuned']['nthreads'] == 1
def test_explicit_run(self):
time_dim = 6
grid = Grid(shape=(11, 11))
a = TimeFunction(name='a',
grid=grid,
time_order=1,
time_dim=time_dim,
save=True)
eqn = Eq(a.forward, a + 1.)
op = Operator(eqn)
assert isinstance(op, OperatorForeign)
args = OrderedDict(op.arguments())
assert args['a'] is None
# Emulate data feeding from outside
array = np.ndarray(shape=a.shape, dtype=np.float32)
array.fill(0.0)
args['a'] = array
op.cfunction(*list(args.values()))
assert all(np.allclose(args['a'][i], i) for i in range(time_dim))
def td_born_forward_op(model, geometry, time_order, space_order):
nt = geometry.nt
# Define the wavefields with the size of the model and the time dimension
u0 = TimeFunction(
name='u0',
grid=model.grid,
time_order=time_order,
space_order=space_order,
save=nt
)
u = TimeFunction(
name='u',
grid=model.grid,
time_order=time_order,
space_order=space_order,
save=nt
)
dm = TimeFunction(
name='dm',
grid=model.grid,
time_order=time_order,
space_order=space_order,
save=nt
)
# Define the wave equation
pde = model.m * u.dt2 - u.laplace + model.damp * u.dt - dm * u0
# Use `solve` to rearrange the equation into a stencil expression
stencil = Eq(u.forward, solve(pde, u.forward), subdomain=model.grid.subdomains['physdomain'])
# Sample at receivers
born_data_rec = PointSource(
name='born_data_rec',
grid=model.grid,
time_range=geometry.time_axis,
coordinates=geometry.rec_positions
)
rec_term = born_data_rec.interpolate(expr=u)
return Operator([stencil] + rec_term, subs=model.spacing_map)
def test_interior(self):
"""
Tests application of an Operator consisting of a single equation
over the ``INTERIOR`` region.
"""
grid = Grid(shape=(4, 4, 4))
x, y, z = grid.dimensions
u = TimeFunction(name='u', grid=grid)
eqn = [Eq(u.forward, u + 2, region=INTERIOR)]
op = Operator(eqn, dle='noop')
op.apply(time_M=2)
assert np.all(u.data[1, 1:-1, 1:-1, 1:-1] == 6.)
assert np.all(u.data[1, :, 0] == 0.)
assert np.all(u.data[1, :, -1] == 0.)
assert np.all(u.data[1, :, :, 0] == 0.)
assert np.all(u.data[1, :, :, -1] == 0.)
def test_mixed_space_order(self):
"""
Make sure that no matter whether data objects have different space order,
as long as they have same domain, the Operator will be executed correctly.
"""
grid = Grid(shape=(8, 8, 8))
u = TimeFunction(name='yu4D', grid=grid, space_order=0)
v = TimeFunction(name='yv4D', grid=grid, space_order=1)
u.data_with_halo[:] = 1.
v.data_with_halo[:] = 2.
op = Operator(Eq(v.forward, u + v))
op(yu4D=u, yv4D=v, t=0)
assert 'run_solution' in str(op)
# Chech that the domain size has actually been written to
assert np.all(v.data[1] == 3.)
# Check that the halo planes are untouched
assert np.all(v.data_with_halo[1, 0, :, :] == 2)
assert np.all(v.data_with_halo[1, :, 0, :] == 2)
assert np.all(v.data_with_halo[1, :, :, 0] == 2)
def test_increasing_halo_wo_ofs(self, space_order, nosimd):
"""
Apply the trivial equation ``u[t+1,x,y,z] = u[t,x,y,z] + 1`` and check
that increasing space orders lead to proportionately larger halo regions,
which are *not* written by the Operator.
For example, with ``space_order = 0``, produce (in 2D view):
1 1 1 ... 1 1
1 1 1 ... 1 1
1 1 1 ... 1 1
1 1 1 ... 1 1
1 1 1 ... 1 1
With ``space_order = 1``, produce:
0 0 0 0 0 0 0 0 0
0 1 1 1 ... 1 1 0
0 1 1 1 ... 1 1 0
0 1 1 1 ... 1 1 0
0 1 1 1 ... 1 1 0
0 1 1 1 ... 1 1 0
0 0 0 0 0 0 0 0 0
And so on and so forth.
"""
# SIMD on/off
configuration['develop-mode'] = nosimd
grid = Grid(shape=(16, 16, 16))
u = TimeFunction(name='yu4D', grid=grid, space_order=space_order)
u.data_with_halo[:] = 0.
op = Operator(Eq(u.forward, u + 1.))
op(yu4D=u, time=0)
assert 'run_solution' in str(op)
# Chech that the domain size has actually been written to
assert np.all(u.data[1] == 1.)
# Check that the halo planes are still 0
assert all(np.all(u.data_with_halo[1, i, :, :] == 0)
for i in range(u._size_halo.left[1]))
assert all(np.all(u.data_with_halo[1, :, i, :] == 0)
for i in range(u._size_halo.left[2]))
assert all(np.all(u.data_with_halo[1, :, :, i] == 0)
for i in range(u._size_halo.left[3]))
def test_default_rules_vs_string(self, so, expected):
"""
Test that default_rules generates correct symbolic expressions when used
with staggered grids.
"""
grid = Grid(shape=(11, ), extent=(10., ))
x = grid.dimensions[0]
f = Function(name='f',
grid=grid,
space_order=so,
staggered=NODE,
coefficients='symbolic')
g = Function(name='g',
grid=grid,
space_order=so,
staggered=x,
coefficients='symbolic')
eq = Eq(g, f.dx)
assert str(eq.evaluate.rhs) == expected
def test_conddim_w_shifting():
nt = 50
grid = Grid(shape=(5, 5))
time = grid.time_dim
factor = Constant(name='factor', value=5, dtype=np.int32)
t_sub = ConditionalDimension('t_sub', parent=time, factor=factor)
save_shift = Constant(name='save_shift', dtype=np.int32)
u = TimeFunction(name='u', grid=grid, time_order=0)
u1 = TimeFunction(name='u', grid=grid, time_order=0)
usave = TimeFunction(name='usave',
grid=grid,
time_order=0,
save=(int(nt // factor.data)),
time_dim=t_sub)
for i in range(usave.save):
usave.data[i, :] = i
eqns = Eq(u.forward, u + usave.subs(t_sub, t_sub - save_shift))
op0 = Operator(eqns, opt='noop')
op1 = Operator(eqns, opt='buffering')
# Check generated code
assert len(retrieve_iteration_tree(op1)) == 3
buffers = [i for i in FindSymbols().visit(op1) if i.is_Array]
assert len(buffers) == 1
# From time_m=15 to time_M=35 with a factor=5 -- it means that, thanks
# to t_sub, we enter the Eq exactly (35-15)/5 + 1 = 5 times. We set
# save_shift=1 so instead of accessing the range usave[15/5:35/5+1],
# we rather access the range usave[15/5-1:35:5], which means accessing
# the usave values 2, 3, 4, 5, 6.
op0.apply(time_m=15, time_M=35, save_shift=1)
op1.apply(time_m=15, time_M=35, save_shift=1, u=u1)
assert np.allclose(u.data, 20)
assert np.all(u.data == u1.data)
# Again, but with a different shift
op1.apply(time_m=15, time_M=35, save_shift=-2, u=u1)
assert np.allclose(u1.data, 20 + 35)
def test_misc_dims(self):
"""
Test MPI in presence of Functions with mixed distributed/replicated
Dimensions, with only a strict subset of the Grid dimensions used.
"""
dx = Dimension(name='dx')
grid = Grid(shape=(4, 4))
x, y = grid.dimensions
glb_pos_map = grid.distributor.glb_pos_map
time = grid.time_dim
u = TimeFunction(name='u',
grid=grid,
time_order=1,
space_order=2,
save=4)
c = Function(name='c', grid=grid, dimensions=(x, dx), shape=(4, 5))
step = Eq(u.forward,
(u[time, x - 2, y] * c[x, 0] + u[time, x - 1, y] * c[x, 1] +
u[time, x, y] * c[x, 2] + u[time, x + 1, y] * c[x, 3] +
u[time, x + 2, y] * c[x, 4]))
for i in range(4):
c.data[i, 0] = 1.0 + i
c.data[i, 1] = 1.0 + i
c.data[i, 2] = 3.0 + i
c.data[i, 3] = 6.0 + i
c.data[i, 4] = 5.0 + i
u.data[:] = 0.0
u.data[0, 2, :] = 2.0
op = Operator(step)
op(time_m=0, time_M=0)
if LEFT in glb_pos_map[x]:
assert (np.all(u.data[1, 0, :] == 10.0))
assert (np.all(u.data[1, 1, :] == 14.0))
else:
assert (np.all(u.data[1, 2, :] == 10.0))
assert (np.all(u.data[1, 3, :] == 8.0))
def test_hierarchical_blocking():
grid = Grid(shape=(64, 64, 64))
u = TimeFunction(name='u', grid=grid, space_order=2)
op = Operator(Eq(u.forward, u + 1), dle=('blocking', {'openmp': False,
'blocklevels': 2}))
# 'basic' mode
op.apply(time_M=0, autotune='basic')
assert op._state['autotuning'][0]['runs'] == 10
assert op._state['autotuning'][0]['tpr'] == options['squeezer'] + 1
assert len(op._state['autotuning'][0]['tuned']) == 4
# 'aggressive' mode
op.apply(time_M=0, autotune='aggressive')
assert op._state['autotuning'][1]['runs'] == 38
assert op._state['autotuning'][1]['tpr'] == options['squeezer'] + 1
assert len(op._state['autotuning'][1]['tuned']) == 4
def test_at_is_actually_working(shape, expected):
"""
Check that autotuning is actually running when switched on,
in both 2D and 3D operators.
"""
grid = Grid(shape=shape)
infield = Function(name='infield', grid=grid)
infield.data[:] = np.arange(reduce(mul, shape),
dtype=np.int32).reshape(shape)
outfield = Function(name='outfield', grid=grid)
stencil = Eq(outfield.indexify(),
outfield.indexify() + infield.indexify() * 3.0)
op = Operator(stencil,
dle=('blocking', {
'openmp': False,
'blockinner': True,
'blockalways': True
}))
# Run with whatever `configuration` says (by default, basic+preemptive)
op(infield=infield, outfield=outfield, autotune=True)
assert op._state['autotuning'][-1]['runs'] == 4
assert op._state['autotuning'][-1]['tpr'] == 1
# Now try `aggressive` autotuning
configuration['autotuning'] = 'aggressive'
op(infield=infield, outfield=outfield, autotune=True)
assert op._state['autotuning'][-1]['runs'] == expected
assert op._state['autotuning'][-1]['tpr'] == 1
configuration['autotuning'] = configuration._defaults['autotuning']
# Try again, but using the Operator API directly
op(infield=infield, outfield=outfield, autotune='aggressive')
assert op._state['autotuning'][-1]['runs'] == expected
assert op._state['autotuning'][-1]['tpr'] == 1
# Similar to above
op(infield=infield,
outfield=outfield,
autotune=('aggressive', 'preemptive'))
assert op._state['autotuning'][-1]['runs'] == expected
assert op._state['autotuning'][-1]['tpr'] == 1
def test_collapsing(self):
grid = Grid(shape=(3, 3, 3))
u = TimeFunction(name='u', grid=grid)
op = Operator(Eq(u.forward, u + 1), opt=('blocking', 'openmp'))
# Does it compile? Honoring the OpenMP specification isn't trivial
assert op.cfunction
# Does it produce the right result
op.apply(t_M=9)
assert np.all(u.data[0] == 10)
iterations = FindNodes(Iteration).visit(op._func_table['bf0'])
assert iterations[0].pragmas[0].value == 'omp for collapse(2) schedule(dynamic,1)'
assert iterations[2].pragmas[0].value == ('omp parallel for collapse(2) '
'schedule(dynamic,1) '
'num_threads(nthreads_nested)')
def _new_operator3(shape, blockshape0=None, blockshape1=None, opt=None):
blockshape0 = as_tuple(blockshape0)
blockshape1 = as_tuple(blockshape1)
grid = Grid(shape=shape, extent=shape, dtype=np.float64)
# Allocate the grid and set initial condition
# Note: This should be made simpler through the use of defaults
u = TimeFunction(name='u', grid=grid, time_order=1, space_order=(2, 2, 2))
u.data[0, :] = np.linspace(-1, 1, reduce(mul, shape)).reshape(shape)
# Derive the stencil according to devito conventions
op = Operator(Eq(u.forward, 0.5 * u.laplace + u), opt=opt)
blocksizes0 = get_blocksizes(op, opt, grid, blockshape0, 0)
blocksizes1 = get_blocksizes(op, opt, grid, blockshape1, 1)
op.apply(u=u, t=10, **blocksizes0, **blocksizes1)
return u.data[1, :], op
def test_arguments_subrange(self):
"""
Test op.apply when a subrange is specified for a distributed dimension.
"""
grid = Grid(shape=(16, ))
x = grid.dimensions[0]
f = TimeFunction(name='f', grid=grid)
op = Operator(Eq(f.forward, f + 1.))
op.apply(time=0, x_m=4, x_M=11)
glb_pos_map = f.grid.distributor.glb_pos_map
if LEFT in glb_pos_map[x]:
assert np.all(f.data_ro_domain[1, :4] == 0.)
assert np.all(f.data_ro_domain[1, 4:] == 1.)
else:
assert np.all(f.data_ro_domain[1, :-4] == 1.)
assert np.all(f.data_ro_domain[1, -4:] == 0.)
def test_trivial_eq_1d_save(self):
grid = Grid(shape=(32, ))
x = grid.dimensions[0]
time = grid.time_dim
f = TimeFunction(name='f', grid=grid, save=5)
f.data_with_halo[:] = 1.
op = Operator(Eq(f.forward, f[time, x - 1] + f[time, x + 1] + 1))
op.apply()
time_M = op.prepare_arguments()['time_M']
assert np.all(f.data_ro_domain[1] == 3.)
glb_pos_map = f.grid.distributor.glb_pos_map
if LEFT in glb_pos_map[x]:
assert np.all(f.data_ro_domain[-1, time_M:] == 31.)
else:
assert np.all(f.data_ro_domain[-1, :-time_M] == 31.)
def test_mpi():
grid = Grid(shape=(4, 4))
u = TimeFunction(name='u', grid=grid, space_order=2)
u1 = TimeFunction(name='u', grid=grid, space_order=2)
eqn = Eq(u.forward, u.dx2 + 1.)
op0 = Operator(eqn)
op1 = Operator(eqn, opt=('advanced', {'linearize': True}))
# Check generated code
assert 'uL0' not in str(op0)
assert 'uL0' in str(op1)
op0.apply(time_M=10)
op1.apply(time_M=10, u=u1)
assert np.all(u.data == u1.data)
def test_at_is_actually_working(shape, expected):
"""
Check that autotuning is actually running when switched on,
in both 2D and 3D operators.
"""
grid = Grid(shape=shape)
buffer = StringIO()
temporary_handler = logging.StreamHandler(buffer)
logger.addHandler(temporary_handler)
infield = Function(name='infield', grid=grid)
infield.data[:] = np.arange(reduce(mul, shape),
dtype=np.int32).reshape(shape)
outfield = Function(name='outfield', grid=grid)
stencil = Eq(outfield.indexify(),
outfield.indexify() + infield.indexify() * 3.0)
op = Operator(stencil,
dle=('blocking', {
'blockinner': True,
'blockalways': True
}))
# Expected 3 AT attempts for the given shape
op(infield=infield, outfield=outfield, autotune=True)
out = [i for i in buffer.getvalue().split('\n') if 'AutoTuner:' in i]
assert len(out) == 4
# Now try the same with aggressive autotuning, which tries 9 more cases
configuration.core['autotuning'] = 'aggressive'
op(infield=infield, outfield=outfield, autotune=True)
out = [i for i in buffer.getvalue().split('\n') if 'AutoTuner:' in i]
assert len(out) == expected
configuration.core['autotuning'] = configuration.core._defaults[
'autotuning']
logger.removeHandler(temporary_handler)
temporary_handler.flush()
temporary_handler.close()
buffer.flush()
buffer.close()
def test_multiple_subnests_v1(self):
"""
Unlike ``test_multiple_subnestes_v0``, now we use the ``cire-rotate=True``
option, which trades some of the inner parallelism for a smaller working set.
"""
grid = Grid(shape=(3, 3, 3))
x, y, z = grid.dimensions
t = grid.stepping_dim
f = Function(name='f', grid=grid)
u = TimeFunction(name='u', grid=grid, space_order=3)
eqn = Eq(
u.forward,
((u[t, x, y, z] + u[t, x + 1, y + 1, z + 1]) * 3 * f +
(u[t, x + 2, y + 2, z + 2] + u[t, x + 3, y + 3, z + 3]) * 3 * f +
1))
op = Operator(eqn,
opt=('advanced', {
'openmp': True,
'cire-mincost-sops': 1,
'cire-rotate': True,
'par-nested': 0,
'par-collapse-ncores': 1,
'par-dynamic-work': 0
}))
trees = retrieve_iteration_tree(op._func_table['bf0'].root)
assert len(trees) == 2
assert trees[0][0] is trees[1][0]
assert trees[0][0].pragmas[0].value ==\
'omp for collapse(2) schedule(dynamic,1)'
assert not trees[0][2].pragmas
assert not trees[0][3].pragmas
assert trees[0][4].pragmas[0].value == ('omp parallel for collapse(1) '
'schedule(dynamic,1) '
'num_threads(nthreads_nested)')
assert not trees[1][2].pragmas
assert trees[1][3].pragmas[0].value == ('omp parallel for collapse(1) '
'schedule(dynamic,1) '
'num_threads(nthreads_nested)')
def test_collapsing_v2(self):
"""
MFE from issue #1478.
"""
n = 8
m = 8
nx, ny, nchi, ncho = 12, 12, 1, 1
x, y = SpaceDimension("x"), SpaceDimension("y")
ci, co = Dimension("ci"), Dimension("co")
i, j = Dimension("i"), Dimension("j")
grid = Grid((nx, ny), dtype=np.float32, dimensions=(x, y))
X = Function(name="xin",
dimensions=(ci, x, y),
shape=(nchi, nx, ny),
grid=grid,
space_order=n // 2)
dy = Function(name="dy",
dimensions=(co, x, y),
shape=(ncho, nx, ny),
grid=grid,
space_order=n // 2)
dW = Function(name="dW",
dimensions=(co, ci, i, j),
shape=(ncho, nchi, n, m),
grid=grid)
eq = [
Eq(
dW[co, ci, i, j], dW[co, ci, i, j] +
dy[co, x, y] * X[ci, x + i - n // 2, y + j - m // 2])
for i in range(n) for j in range(m)
]
op = Operator(eq, opt=('advanced', {'openmp': True}))
iterations = FindNodes(Iteration).visit(op)
assert len(iterations) == 4
assert iterations[0].ncollapse == 1
assert iterations[1].is_Vectorized
assert iterations[2].is_Sequential
assert iterations[3].is_Sequential
def test_at_w_mpi():
"""Make sure autotuning works in presence of MPI. MPI ranks work
in isolation to determine the best block size, locally."""
grid = Grid(shape=(8, 8))
t = grid.stepping_dim
x, y = grid.dimensions
f = TimeFunction(name='f', grid=grid, time_order=1)
f.data_with_halo[:] = 1.
eq = Eq(f.forward, f[t, x-1, y] + f[t, x+1, y])
op = Operator(eq, dle=('advanced', {'openmp': False, 'blockinner': True}))
op.apply(time=-1, autotune=('basic', 'runtime'))
# Nothing really happened, as not enough timesteps
assert np.all(f.data_ro_domain[0] == 1.)
assert np.all(f.data_ro_domain[1] == 1.)
# The 'destructive' mode writes directly to `f` for whatever timesteps required
# to perform the autotuning. Eventually, the result is complete garbage; note
# also that this autotuning mode disables the halo exchanges
op.apply(time=-1, autotune=('basic', 'destructive'))
assert np.all(f._data_ro_with_inhalo.sum() == 904)
# Check the halo hasn't been touched during AT
glb_pos_map = grid.distributor.glb_pos_map
if LEFT in glb_pos_map[x]:
assert np.all(f._data_ro_with_inhalo[:, -1] == 1)
else:
assert np.all(f._data_ro_with_inhalo[:, 0] == 1)
# Finally, try running w/o AT, just to be sure nothing was broken
f.data_with_halo[:] = 1.
op.apply(time=2)
if LEFT in glb_pos_map[x]:
assert np.all(f.data_ro_domain[1, 0] == 5.)
assert np.all(f.data_ro_domain[1, 1] == 7.)
assert np.all(f.data_ro_domain[1, 2:4] == 8.)
else:
assert np.all(f.data_ro_domain[1, 4:6] == 8)
assert np.all(f.data_ro_domain[1, 6] == 7)
assert np.all(f.data_ro_domain[1, 7] == 5)
def test_misc_dims(self):
"""
Tests grid-independent :class:`Function`s, which require YASK's "misc"
dimensions.
"""
dx = Dimension(name='dx')
grid = Grid(shape=(10, 10))
x, y = grid.dimensions
time = grid.time_dim
u = TimeFunction(name='u',
grid=grid,
time_order=1,
space_order=4,
save=4)
c = Function(name='c', dimensions=(x, dx), shape=(10, 5))
step = Eq(u.forward,
(u[time, x - 2, y] * c[x, 0] + u[time, x - 1, y] * c[x, 1] +
u[time, x, y] * c[x, 2] + u[time, x + 1, y] * c[x, 3] +
u[time, x + 2, y] * c[x, 4]))
for i in range(10):
c.data[i, 0] = 1.0 + i
c.data[i, 1] = 1.0 + i
c.data[i, 2] = 3.0 + i
c.data[i, 3] = 6.0 + i
c.data[i, 4] = 5.0 + i
u.data[:] = 0.0
u.data[0, 2, :] = 2.0
op = Operator(step)
assert 'run_solution' in str(op)
op(time_m=0, time_M=0)
assert (np.all(u.data[1, 0, :] == 10.0))
assert (np.all(u.data[1, 1, :] == 14.0))
assert (np.all(u.data[1, 2, :] == 10.0))
assert (np.all(u.data[1, 3, :] == 8.0))
assert (np.all(u.data[1, 4, :] == 10.0))
assert (np.all(u.data[1, 5:10, :] == 0.0))
def test_full_shape_with_subdims(self):
"""
Like `test_full_alias_shape_after_blocking`, but SubDomains (and therefore
SubDimensions) are used. Nevertheless, the temporary shape should still be
dictated by the root Dimensions.
"""
grid = Grid(shape=(3, 3, 3))
x, y, z = grid.dimensions # noqa
t = grid.stepping_dim
f = Function(name='f', grid=grid)
f.data_with_halo[:] = 1.
u = TimeFunction(name='u', grid=grid, space_order=3)
u.data_with_halo[:] = 0.
# Leads to 3D aliases
eqn = Eq(u.forward, ((u[t, x, y, z] + u[t, x+1, y+1, z+1])*3*f +
(u[t, x+2, y+2, z+2] + u[t, x+3, y+3, z+3])*3*f + 1),
subdomain=grid.interior)
op0 = Operator(eqn, dse='noop', dle=('advanced', {'openmp': True}))
op1 = Operator(eqn, dse='aggressive', dle=('advanced', {'openmp': True}))
xi0_blk_size = op1.parameters[-3]
yi0_blk_size = op1.parameters[-2]
z_size = op1.parameters[4]
# Check Array shape
arrays = [i for i in FindSymbols().visit(op1._func_table['bf0'].root)
if i.is_Array]
assert len(arrays) == 1
a = arrays[0]
assert len(a.dimensions) == 3
assert a.halo == ((1, 1), (1, 1), (1, 1))
assert Add(*a.symbolic_shape[0].args) == xi0_blk_size + 2
assert Add(*a.symbolic_shape[1].args) == yi0_blk_size + 2
assert Add(*a.symbolic_shape[2].args) == z_size + 2
# Check numerical output
op0(time_M=1)
exp = np.copy(u.data[:])
u.data_with_halo[:] = 0.
op1(time_M=1)
assert np.all(u.data == exp)
def conv(nx, ny, nchi, ncho, n, m):
# Image size
dt = np.float32
x, y, ci, co = (SpaceDimension("x"), SpaceDimension("y"), Dimension("ci"),
Dimension("co"))
grid = Grid((nchi, ncho, nx, ny), dtype=dt, dimensions=(ci, co, x, y))
# Image
im_in = Function(name="imi", dimensions=(ci, x, y),
shape=(nchi, nx, ny), grid=grid, space_order=n//2)
# Output
im_out = Function(name="imo", dimensions=(co, x, y),
shape=(ncho, nx, ny), grid=grid, space_order=n//2)
# Weights
i, j = Dimension("i"), Dimension("j")
W = Function(name="W", dimensions=(co, ci, i, j), shape=(ncho, nchi, n, m),
grid=grid)
# Popuate weights with deterministic values
for i in range(ncho):
for j in range(nchi):
W.data[i, j, :, :] = np.linspace(i+j, i+j+(n*m), n*m).reshape(n, m)
# Convlution
conv = [Eq(im_out, im_out
+ sum([W[co, ci, i2, i1] * im_in[ci, x+i1-n//2, y+i2-m//2]
for i1 in range(n) for i2 in range(m)]))]
op = Operator(conv)
op.cfunction
# Initialize the input/output
input_data = np.linspace(-1, 1, nx*ny*nchi).reshape(nchi, nx, ny)
im_in.data[:] = input_data.astype(np.float32)
im_out.data
op()
return im_out.data, im_in.data
def test_subdimleft_parallel(self):
"""
Tests application of an Operator consisting of a subdimension
defined over different sub-regions, explicitly created through the
use of :class:`SubDimension`s.
This tests that flow direction is not being automatically inferred
from whether the subdimension is on the left or right boundary.
"""
grid = Grid(shape=(20, 20))
x, y = grid.dimensions
t = grid.stepping_dim
thickness = 4
u = TimeFunction(name='u',
save=None,
grid=grid,
space_order=0,
time_order=1)
xl = SubDimension.left(name='xl', parent=x, thickness=thickness)
yi = SubDimension.middle(name='yi',
parent=y,
thickness_left=thickness,
thickness_right=thickness)
# Can be done in parallel
eq = Eq(u[t + 1, xl, yi], u[t, xl, yi] + 1)
op = Operator([eq])
iterations = FindNodes(Iteration).visit(op)
assert all(i.is_Affine and i.is_Parallel for i in iterations
if i.dim in [xl, yi])
op.apply(time_m=0, time_M=0)
assert np.all(u.data[1, 0:thickness, 0:thickness] == 0)
assert np.all(u.data[1, 0:thickness, -thickness:] == 0)
assert np.all(u.data[1, 0:thickness, thickness:-thickness] == 1)
assert np.all(u.data[1, thickness + 1:, :] == 0)
def test_cache_blocking_imperfect_nest_v2(blockinner):
"""
Test that a non-perfect Iteration nest is blocked correctly. This
is slightly different than ``test_cache_blocking_imperfect_nest``
as here only one Iteration gets blocked.
"""
shape = (16, 16, 16)
grid = Grid(shape=shape, dtype=np.float64)
u = TimeFunction(name='u', grid=grid, space_order=2)
u.data[:] = np.linspace(0, 1, reduce(mul, shape), dtype=np.float64).reshape(shape)
eq = Eq(u.forward, 0.01*u.dy.dy)
op0 = Operator(eq, opt='noop')
op1 = Operator(eq, opt=('cire-sops', {'blockinner': blockinner}))
op2 = Operator(eq, opt=('advanced-fsg', {'blockinner': blockinner}))
# First, check the generated code
trees = retrieve_iteration_tree(op2._func_table['bf0'].root)
assert len(trees) == 2
assert len(trees[0]) == len(trees[1])
assert all(i is j for i, j in zip(trees[0][:2], trees[1][:2]))
assert trees[0][2] is not trees[1][2]
assert trees[0].root.dim.is_Incr
assert trees[1].root.dim.is_Incr
assert op2.parameters[6] is trees[0].root.step
op0(time_M=0)
u1 = TimeFunction(name='u1', grid=grid, space_order=2)
u1.data[:] = np.linspace(0, 1, reduce(mul, shape), dtype=np.float64).reshape(shape)
op1(time_M=0, u=u1)
u2 = TimeFunction(name='u2', grid=grid, space_order=2)
u2.data[:] = np.linspace(0, 1, reduce(mul, shape), dtype=np.float64).reshape(shape)
op2(time_M=0, u=u2)
assert np.allclose(u.data, u1.data, rtol=1e-07)
assert np.allclose(u.data, u2.data, rtol=1e-07)
def test_argument_from_index_constant(self):
nx, ny = 30, 30
grid = Grid(shape=(nx, ny))
x, y = grid.dimensions
arbdim = Dimension('arb')
u = TimeFunction(name='u', grid=grid, save=None, time_order=2, space_order=0)
snap = Function(name='snap', dimensions=(arbdim, x, y), shape=(5, nx, ny),
space_order=0)
save_t = Constant(name='save_t', dtype=np.int32)
save_slot = Constant(name='save_slot', dtype=np.int32)
expr = Eq(snap.subs(arbdim, save_slot), u.subs(grid.stepping_dim, save_t))
op = Operator(expr)
u.data[:] = 0.0
snap.data[:] = 0.0
u.data[0, 10, 10] = 1.0
op.apply(save_t=0, save_slot=1)
assert snap.data[1, 10, 10] == 1.0
def test_default_functions(self):
"""
Test the default argument derivation for functions.
"""
grid = Grid(shape=(5, 6, 7))
f = TimeFunction(name='f', grid=grid)
g = Function(name='g', grid=grid)
op = Operator(Eq(g, g + f))
expected = {
'x_size': 5, 'x_m': 0, 'x_M': 4,
'y_size': 6, 'y_m': 0, 'y_M': 5,
'z_size': 7, 'z_m': 0, 'z_M': 6,
'f': f.data_allocated, 'g': g.data_allocated,
}
self.verify_arguments(op.arguments(time=4), expected)
exp_parameters = ['f', 'g', 'x_m', 'x_M', 'x_size', 'y_m',
'y_M', 'y_size', 'z_m', 'z_M', 'z_size',
'time_m', 'time_M']
self.verify_parameters(op.parameters, exp_parameters)