def test_iteration_index_order_matters_with_mat(self, backend, skip_opencl,
iterset, x, iter2ind1,
mat):
self.cache.clear()
assert len(self.cache) == 0
self.cache_hit.clear()
assert len(self.cache_hit) == 0
plan1 = plan.Plan(iterset.all_part,
mat(op2.INC,
(iter2ind1[op2.i[0]], iter2ind1[op2.i[1]])),
x(op2.READ, iter2ind1[0]),
partition_size=10,
matrix_coloring=True)
op2.base._trace.evaluate(set([mat]), set())
assert len(self.cache) == 1
assert self.cache_hit[plan1] == 1
plan2 = plan.Plan(iterset.all_part,
mat(op2.INC,
(iter2ind1[op2.i[1]], iter2ind1[op2.i[0]])),
x(op2.READ, iter2ind1[0]),
partition_size=10,
matrix_coloring=True)
op2.base._trace.evaluate(set([mat]), set())
assert len(self.cache) == 2
assert plan1 is not plan2
assert self.cache_hit[plan1] == 1
assert self.cache_hit[plan2] == 1
def test_thread_coloring(self, backend, skip_opencl, elements,
elem_node_map, elem_node, mat, x):
assert NUM_ELE % 2 == 0, "NUM_ELE must be even."
plan = _plan.Plan(elements.all_part,
mat(op2.INC,
(elem_node[op2.i[0]], elem_node[op2.i[1]])),
x(op2.WRITE, elem_node[0]),
partition_size=NUM_ELE / 2,
matrix_coloring=True)
assert plan.nblocks == 2
eidx = 0
for p in range(plan.nblocks):
for thrcol in range(plan.nthrcol[p]):
counter = numpy.zeros(NUM_NODES, dtype=numpy.uint32)
for e in range(eidx, eidx + plan.nelems[p]):
if plan.thrcol[e] == thrcol:
counter[elem_node.values[e][0]] += 1
assert (counter < 2).all()
eidx += plan.nelems[p]