compiler: restructure tests and pass

devito/core/cpu.py

@@ -3,8 +3,8 @@
 from devito.core.operator import CoreOperator, CustomOperator
 from devito.exceptions import InvalidOperator
 from devito.passes.equations import buffering, collect_derivatives
-from devito.passes.clusters import (Blocking, Lift, cire, cse, eliminate_arrays, fuse,
-                                    extract_increments, factorize, optimize_pows, Skewing)
+from devito.passes.clusters import (Blocking, Skewing, Lift, cire, cse, eliminate_arrays,
+                                    fuse, extract_increments, factorize, optimize_pows)
 from devito.passes.iet import (CTarget, OmpTarget, avoid_denormals, mpiize,
                                optimize_halospots, hoist_prodders, relax_incr_dimensions)
 from devito.tools import timed_pass
@@ -325,7 +325,7 @@ def _make_clusters_passes_mapper(cls, **kwargs):
             'cire-divs': lambda i: cire(i, 'divs', sregistry, options, platform),
             'cse': lambda i: cse(i, sregistry),
             'opt-pows': optimize_pows,
-            'topofuse': lambda i: fuse(i, toposort=True),
+            'topofuse': lambda i: fuse(i, toposort=True)
         }
 
     @classmethod
@@ -354,8 +354,8 @@ def _make_iet_passes_mapper(cls, **kwargs):
         # Expressions
         'buffering',
         # Clusters
-        'blocking', 'topofuse', 'fuse', 'factorize', 'cire-sops', 'cire-divs',
-        'cse', 'lift', 'opt-pows', 'skewing',
+        'blocking', 'skewing', 'topofuse', 'fuse', 'factorize', 'cire-sops', 'cire-divs',
+        'cse', 'lift', 'opt-pows'
         # IET
         'denormals', 'optcomms', 'openmp', 'mpi', 'simd', 'prodders',
     )

devito/core/gpu.py

@@ -293,8 +293,8 @@ def _make_iet_passes_mapper(cls, **kwargs):
         # Expressions
         'buffering',
         # Clusters
-        'blocking', 'tasking', 'streaming', 'factorize', 'fuse', 'lift',
-        'cire-sops', 'cire-divs', 'cse', 'opt-pows', 'topofuse', 'skewing',
+        'blocking', 'skewing', 'tasking', 'streaming', 'factorize', 'fuse',
+        'lift', 'cire-sops', 'cire-divs', 'cse', 'opt-pows', 'topofuse',
         # IET
         'optcomms', 'orchestrate', 'parallel', 'mpi', 'prodders', 'gpu-direct'
     )

devito/mpi/halo_scheme.py

@@ -429,31 +429,4 @@ def classify(exprs, ispace):
 
         mapper[f] = HaloSchemeEntry(frozendict(loc_indices), frozenset(halos))
 
-def compute_local_indices(f, dims, ispace, scope):
-    """
-    Map the Dimensions in ``dims`` to the local indices necessary
-    to perform a halo exchange, as described in HaloScheme.__doc__.
-
-    Examples
-    --------
-    1) u[t+1, x] = f(u[t, x])   => shift == 1
-    2) u[t-1, x] = f(u[t, x])   => shift == 1
-    3) u[t+1, x] = f(u[t+1, x]) => shift == 0
-    In the first and second cases, the x-halo should be inserted at `t`,
-    while in the last case it should be inserted at `t+1`.
-    """
-    loc_indices = {}
-    for d in dims:
-        try:
-            func = Max if ispace.is_forward(d.root) else Min
-        except KeyError:
-            raise HaloSchemeException("Don't know how to build a HaloScheme as `%s` "
-                                      "doesn't appear in `%s`" % (d, ispace))
-        if d.is_Stepping:
-            candidates = {i[d].origin - d: i[d] for i in scope.getreads(f)
-                          if not is_integer(i[d])}
-        else:
-            candidates = {i[d] - d: i[d] for i in scope.getreads(f)
-                          if not is_integer(i[d])}
-        loc_indices[d] = candidates[func(*candidates.keys())]
-    return loc_indices
+    return mapper

devito/passes/clusters/blocking.py

@@ -166,15 +166,11 @@ class Skewing(Queue):
 
     def __init__(self, options):
 
-        self.nskewed = Counter()
-
         super(Skewing, self).__init__()
 
     @timed_pass(name='skewing')
     def process(self, clusters):
-        processed = super(Skewing, self).process(clusters)
-
-        return processed
+        return super(Skewing, self).process(clusters)
 
     def _process_fdta(self, clusters, level, prefix=None):
 

tests/test_autotuner.py

@@ -142,9 +142,10 @@ def test_mixed_blocking_w_skewing(openmp, expected):
         assert 'nthreads' not in op._state['autotuning'][0]['tuned']
 
 
-def test_tti_aggressive():
+@pytest.mark.parametrize('opt', ['advanced', ('blocking', 'skewing')])
+def test_tti_aggressive(opt):
     from test_dse import TestTTI
-    wave_solver = TestTTI().tti_operator(opt='advanced')
+    wave_solver = TestTTI().tti_operator(opt=opt)
     op = wave_solver.op_fwd(kernel='centered')
     op.apply(time=0, autotune='aggressive', dt=0.1)
     assert op._state['autotuning'][0]['runs'] == 30
@@ -262,13 +263,14 @@ def test_multiple_blocking():
     assert len(op._state['autotuning'][0]['tuned']) == 5
 
 
-def test_hierarchical_blocking():
+@pytest.mark.parametrize('opt', ['blocking', ('blocking', 'skewing')])
+def test_hierarchical_blocking(opt):
     grid = Grid(shape=(64, 64, 64))
 
     u = TimeFunction(name='u', grid=grid, space_order=2)
 
-    op = Operator(Eq(u.forward, u + 1), opt=('blocking', {'openmp': False,
-                                                          'blocklevels': 2}))
+    op = Operator(Eq(u.forward, u + 1), opt=(opt, {'openmp': False,
+                                                   'blocklevels': 2}))
 
     # 'basic' mode
     op.apply(time_M=0, autotune='basic')
@@ -284,7 +286,8 @@ def test_hierarchical_blocking():
 
 
 @switchconfig(platform='cpu64-dummy')  # To fix the core count
-def test_multiple_threads():
+@pytest.mark.parametrize('opt', ['blocking', ('blocking', 'skewing')])
+def test_multiple_threads(opt):
     """
     Test autotuning when different ``num_threads`` for a given OpenMP parallel
     region are attempted.
@@ -293,7 +296,8 @@ def test_multiple_threads():
 
     v = TimeFunction(name='v', grid=grid)
 
-    op = Operator(Eq(v.forward, v + 1), opt=('blocking', {'openmp': True}))
+    op = Operator(Eq(v.forward, v + 1), opt=(opt, {'openmp': True}))
+
     op.apply(time_M=0, autotune='max')
     assert op._state['autotuning'][0]['runs'] == 60  # Would be 30 with `aggressive`
     assert op._state['autotuning'][0]['tpr'] == options['squeezer'] + 1

tests/test_ir.py

@@ -539,7 +539,6 @@ def fa(self, grid):
         ('u[x + z,x + y,z*z]', (IRREGULAR, IRREGULAR, REGULAR)),
         ('u[x+1,u[2,2,2],z-1]', (AFFINE, IRREGULAR, AFFINE)),
         ('u[y,x,z]', (IRREGULAR, IRREGULAR, AFFINE)),
-        ('u[x-time, y-time, z]', (IRREGULAR, IRREGULAR, AFFINE)),
     ])
     def test_index_mode_detection(self, indexed, expected):
         """
@@ -550,7 +549,6 @@ def test_index_mode_detection(self, indexed, expected):
         """
         grid = Grid(shape=(4, 4, 4))
         x, y, z = grid.dimensions  # noqa
-        time = grid.time_dim  # noqa
 
         sx = SubDimension.middle('sx', x, 1, 1)  # noqa