1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
|
"""
Copyright (c) 2018 Intel Corporation
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""
import networkx as nx
import numpy as np
from copy import deepcopy
from extensions.middle.EltwiseInputNormalization import EltwiseInputNormalize
from extensions.middle.EltwiseInputReshape import EltwiseInputReshape, Eltwise1DInputReshape
from mo.front.common.layout import get_features_dim, shape_for_layout
from mo.graph.graph import Node, get_sorted_inputs
from mo.middle.passes.fusing.helpers import get_value_id
from mo.middle.replacement import MiddleReplacementPattern
from mo.ops.op import Op
from mo.ops.reshape import Reshape
class EltwiseChecker(MiddleReplacementPattern):
# This pass checks for each eltwise, can it be ScaleShift or not
enabled = True
def run_after(self):
return [Eltwise1DInputReshape]
def find_and_replace_pattern(self, graph: nx.MultiDiGraph):
eltwise_nodes = [Node(graph, node) for node in graph.node if Node(graph, node).soft_get('type') == 'Eltwise']
for node in eltwise_nodes:
raw_inputs = [(inp, attr) for inp, attr in get_sorted_inputs(node)
if 'control_flow_edge' not in attr or not attr['control_flow_edge']]
shapes = [node.graph.node[inp]['shape'] for inp, attr in raw_inputs]
max_dims = None
max_dims_id = None
input_shape = None
for id, s in enumerate(shapes):
if max_dims is None or len(s) > max_dims:
max_dims = len(s)
input_shape = s
max_dims_id = id
feature_dim = 1 if node.graph.graph['layout'] == 'NCHW' else (max_dims - 1)
def check_shape(shape):
# Check that value has shape like 1,N,1,1
return np.prod(shape) == np.max(shape) and (max_dims - feature_dim) <= len(shape) and \
(input_shape[feature_dim] == shape[-1 * (max_dims - feature_dim)] or
(shape[-1 * (max_dims - feature_dim)] == 1 and np.max(shape) == 1))
# Make all input shapes of the same size by adding 1's
axis = node.axis if node.has_valid('axis') else None
for id, shape in enumerate(shapes):
if id != max_dims_id and len(shape) > 0 and not check_shape(shapes[id]) and np.prod(shape) != 1:
node['can_be_fused'] = False
node['can_be_scaleshift'] = False
|
