def test_timesteps_per_at_run():
"""
Check that each autotuning run (ie with a given block shape) takes
``autotuning.options['at_squeezer'] - data.time_order`` timesteps.
in an operator performing an increment such as
``a[t + timeorder, ...] = f(a[t, ...], ...)``.
"""
buffer = StringIO()
temporary_handler = logging.StreamHandler(buffer)
logger.addHandler(temporary_handler)
set_log_level('DEBUG')
shape = (30, 30, 30)
grid = Grid(shape=shape)
x, y, z = grid.dimensions
t = grid.stepping_dim
# Function
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', {'blockalways': True}))
op(infield=infield, outfield=outfield, autotune=True)
out = [i for i in buffer.getvalue().split('\n') if 'AutoTuner:' in i]
assert len(out) == 4
assert all('in 1 time steps' in i for i in out)
buffer.truncate(0)
# TimeFunction with increasing time order
for to in [1, 2, 4]:
infield = TimeFunction(name='infield', grid=grid, time_order=to)
infield.data[:] = np.arange(reduce(mul, infield.shape),
dtype=np.int32).reshape(infield.shape)
outfield = TimeFunction(name='outfield', grid=grid, time_order=to)
stencil = Eq(outfield.indexed[t + to, x, y, z],
outfield.indexify() + infield.indexify() * 3.0)
op = Operator(stencil, dle=('blocking', {'blockalways': True}))
op(infield=infield, outfield=outfield, autotune=True)
out = [i for i in buffer.getvalue().split('\n') if 'AutoTuner:' in i]
expected = options['at_squeezer'] - to
assert len(out) == 4
assert all('in %d time steps' % expected in i for i in out)
buffer.truncate(0)
logger.removeHandler(temporary_handler)
temporary_handler.flush()
temporary_handler.close()
buffer.flush()
buffer.close()
set_log_level('INFO')
def test_timesteps_per_at_run():
"""
Check that each autotuning run (ie with a given block shape) takes
``autotuning.core.options['at_squeezer']`` timesteps, for an operator
performing the increment ``a[t + timeorder, ...] = f(a[t, ...], ...)``.
"""
from devito.core.autotuning import options
buffer = StringIO()
temporary_handler = logging.StreamHandler(buffer)
logger.addHandler(temporary_handler)
shape = (30, 30, 30)
grid = Grid(shape=shape)
x, y, z = grid.dimensions
t = grid.stepping_dim
# Function
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', {'blockalways': True}))
op(infield=infield, outfield=outfield, autotune=True)
out = [i for i in buffer.getvalue().split('\n') if 'AutoTuner:' in i]
assert len(out) == 4
assert all('in 1 time steps' in i for i in out)
buffer.truncate(0)
# TimeFunction with increasing time order; increasing the time order
# shouldn't affect how many iterations the autotuner is gonna run
for to in [1, 2, 4]:
infield = TimeFunction(name='infield', grid=grid, time_order=to)
infield.data[:] = np.arange(reduce(mul, infield.shape),
dtype=np.int32).reshape(infield.shape)
outfield = TimeFunction(name='outfield', grid=grid, time_order=to)
stencil = Eq(outfield.indexed[t + to, x, y, z],
outfield.indexify() + infield.indexify() * 3.0)
op = Operator(stencil, dle=('blocking', {'blockalways': True}))
op(infield=infield, outfield=outfield, t=2, autotune=True)
out = [i for i in buffer.getvalue().split('\n') if 'AutoTuner:' in i]
assert len(out) == 4
assert all('in %d time steps' % options['at_squeezer'] in i
for i in out)
buffer.truncate(0)
logger.removeHandler(temporary_handler)
temporary_handler.flush()
temporary_handler.close()
buffer.flush()
buffer.close()
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)
set_log_level('DEBUG')
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()
set_log_level('INFO')