def test_div_test_2(self):
# Test with two same inputs from one placeholder
graph = build_graph(nodes_attributes,
[('placeholder_1', 'Div'),
('placeholder_1', 'Div'),
('Div', 'last')
],
{'placeholder_1': {'shape': np.array([1, 227, 227, 3])},
}, nodes_with_edges_only=True)
graph_ref = build_graph(nodes_attributes,
[('power_1', 'mul_1'),
('placeholder_1', 'mul_1'),
('placeholder_1', 'power_1'),
('mul_1', 'last'),
],
{'placeholder_1': {'shape': np.array([1, 227, 227, 3])},
'power_1': {'scale': np.array(1), 'power': np.array(-1), 'shift': np.array(0),
'type': 'Power'},
'mul_1': {'type': 'Eltwise', 'op': 'Mul'},
}, nodes_with_edges_only=True)
graph.stage = 'front'
tested_class = Div()
tested_class.find_and_replace_pattern(graph)
(flag, resp) = compare_graphs(graph, graph_ref, 'last', check_op_attrs=True)
self.assertTrue(flag, resp)
def test_div_with_integer(self):
# Test where transformation should not be applied because the divisor is integer
graph = build_graph(
{
**regular_op_with_shaped_data('parameter', [1, 227, 227, 3], {
'type': 'Parameter',
'data_type': np.int32
}),
**valued_const_with_data('const',
np.array([-1.], dtype=np.int32)),
**regular_op_with_shaped_data('div', None, {
'op': 'Div',
'type': 'Divide',
'name': 'my_div'
}),
**result()
}, [
*connect('parameter:0', '0:div'),
*connect_data('const:0', '1:div'),
*connect('div', 'output'),
])
graph_ref = graph.copy()
Div().find_and_replace_pattern(graph)
(flag, resp) = compare_graphs(graph,
graph_ref,
'output',
check_op_attrs=True)
self.assertTrue(flag, resp)
def test_div_test_2(self):
# Test with two same inputs from one placeholder
graph = build_graph(nodes, [
*connect('placeholder_1:0', '0:div'),
*connect_data('placeholder_1:0', '1:div'),
*connect('div', 'output'),
],
nodes_with_edges_only=True)
Div().find_and_replace_pattern(graph)
graph_ref = build_graph(nodes, [
*connect('placeholder_1:0', '0:mul'),
*connect_data('placeholder_1:0', '0:reciprocal'),
*connect('minus_one', '1:reciprocal'),
*connect('reciprocal', '1:mul'),
*connect('mul', 'output'),
],
nodes_with_edges_only=True)
(flag, resp) = compare_graphs(graph,
graph_ref,
'output',
check_op_attrs=True)
self.assertTrue(flag, resp)
self.assertTrue(graph.node[graph.get_nodes_with_attributes(
type='Multiply')[0]]['name'] == 'my_div')
def find_and_replace_pattern(self, graph: Graph):
fw = graph.graph['fw']
argv = graph.graph['cmd_params']
layout = graph.graph['layout']
for_graph_and_each_sub_graph_recursively(graph, fuse_pad)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
# Mark nodes with attr 'can_be_fused': False to disable fusing for specified nodes
for_graph_and_each_sub_graph_recursively(graph, lambda graph: mark_unfused_nodes(graph, argv.finegrain_fusing))
# Converting FusedBatchNorm layer to Mul->Add->Mul->Add sequence
# IE doesn't support batchNormInference with 4 inputs, so we have to split it to two ScaleShift
for_graph_and_each_sub_graph_recursively(graph, convert_batch_norm)
if fw == 'caffe':
# Converting ScaleShift layer to Mul->Add
for_graph_and_each_sub_graph_recursively(graph, convert_scale_shift_to_mul_add)
for_graph_and_each_sub_graph_recursively(graph, Div().find_and_replace_pattern)
for_graph_and_each_sub_graph_recursively(graph, Sub().find_and_replace_pattern)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
if not argv.disable_fusing:
if fw != 'caffe':
# Converting ScaleShift layer to Mul->Add
for_graph_and_each_sub_graph_recursively(graph, convert_scale_shift_to_mul_add)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
# Fusing the sequences of Mul/Add operations
for_graph_and_each_sub_graph_recursively(graph, fuse_mul_add_sequence)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
normalize_eltwise_inputs(graph)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
# Fusing linear operation to Convolution
for_graph_and_each_sub_graph_recursively(graph, fuse_linear_ops)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
if not argv.disable_gfusing:
for_graph_and_each_sub_graph_recursively(graph, grouped_convolutions_fusing)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
if not argv.disable_fusing:
for_graph_and_each_sub_graph_recursively(graph, fuse_linear_ops)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
for_graph_and_each_sub_graph_recursively(graph, normalize_eltwise_inputs)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
MarkNodesToFuseUpToFakeQuantize().find_and_replace_pattern(graph)
FakeQuantizeFuse().find_and_replace_pattern(graph)
AddFakeQuantizeFuse().find_and_replace_pattern(graph)
MulFakeQuantizeFuse().find_and_replace_pattern(graph)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
for_graph_and_each_sub_graph_recursively(graph, fuse_pad)
for_graph_and_each_sub_graph_recursively(graph, lambda G: G.clean_up())
if layout != 'NHWC' and not argv.disable_resnet_optimization:
stride_optimization(graph)
