ce19e1fb9a
* Add hsigmoid fusing for MO * Update Bom file * Remove comments * Refactoring hsigmoid fusion according to review * Add div and mul patterns for hsigmoid fusion * Refactoring code according to review * Fix HSigmoid fusion transformation
182 lines
7.2 KiB
Python
182 lines
7.2 KiB
Python
"""
|
|
Copyright (C) 2018-2020 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 numpy as np
|
|
|
|
from extensions.front.AttributedClampNormalizer import AttributedClampNormalizer
|
|
from extensions.ops.activation_ops import HSigmoid
|
|
from mo.front.common.replacement import FrontReplacementSubgraph
|
|
from mo.front.subgraph_matcher import SubgraphMatch
|
|
from mo.graph.graph import Graph, rename_nodes
|
|
from mo.utils.graph import Node
|
|
|
|
|
|
def replace_with_hsigmoid(graph: Graph, first_node: Node, last_node: Node):
|
|
# determine the input port of first and last nodes which gets the 'input' node output
|
|
add_input_port_idx = int(first_node.in_port(0).get_connection().get_source().node.soft_get('op') == 'Const')
|
|
last_node_name = last_node.soft_get('name', last_node.id)
|
|
|
|
hsigmoid = HSigmoid(graph, {}).create_node()
|
|
hsigmoid.in_port(0).connect(first_node.in_port(add_input_port_idx).get_source())
|
|
last_node.out_port(0).get_connection().set_source(hsigmoid.out_port(0))
|
|
|
|
rename_nodes([(last_node, last_node_name + '/TBR'), (hsigmoid, last_node_name)])
|
|
|
|
|
|
class HSigmoidWithClamp(FrontReplacementSubgraph):
|
|
"""
|
|
The transformation looks for the pattern with ReLU6 (Clamp) defining the HSigmoid function:
|
|
HSigmoid(x) = Relu6(x + 3.0) / 6.0.
|
|
"""
|
|
enabled = True
|
|
|
|
def run_after(self):
|
|
return [AttributedClampNormalizer]
|
|
|
|
def pattern(self):
|
|
return dict(
|
|
nodes=[
|
|
('input', dict()),
|
|
('add', dict(op='Add')),
|
|
('const_0', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 0.0, atol=1e-6))),
|
|
('const_3', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 3.0, atol=1e-6))),
|
|
('const_6', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 6.0, atol=1e-6))),
|
|
('const_1_6', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 1.0 / 6.0, atol=1e-6))),
|
|
('clamp', dict(op='Clamp')),
|
|
('mul_2', dict(op='Mul')),
|
|
],
|
|
edges=[
|
|
('input', 'add', {}),
|
|
('const_3', 'add', {}),
|
|
('add', 'clamp', {'in': 0}),
|
|
('const_0', 'clamp', {'in': 1}),
|
|
('const_6', 'clamp', {'in': 2}),
|
|
('clamp', 'mul_2', {}),
|
|
('const_1_6', 'mul_2', {}),
|
|
])
|
|
|
|
def replace_sub_graph(self, graph: Graph, match: [dict, SubgraphMatch]):
|
|
replace_with_hsigmoid(graph, match['add'], match['mul_2'])
|
|
|
|
|
|
class HSigmoidWithMinMax(FrontReplacementSubgraph):
|
|
"""
|
|
The transformation looks for the pattern with Min/Max defining the HSigmoid function:
|
|
HSigmoid(x) = Min(Max(x + 3.0, 0), 6.0) / 6.0.
|
|
"""
|
|
enabled = True
|
|
|
|
def run_after(self):
|
|
return [AttributedClampNormalizer]
|
|
|
|
def pattern(self):
|
|
return dict(
|
|
nodes=[
|
|
('input', dict()),
|
|
('add', dict(op='Add')),
|
|
('const_0', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 0.0, atol=1e-6))),
|
|
('const_3', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 3.0, atol=1e-6))),
|
|
('const_6', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 6.0, atol=1e-6))),
|
|
('const_1_6', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 1.0 / 6.0, atol=1e-6))),
|
|
('max', dict(op='Maximum')),
|
|
('min', dict(op='Minimum')),
|
|
('mul_2', dict(op='Mul')),
|
|
],
|
|
edges=[
|
|
('input', 'add', {'out': 0}),
|
|
('const_3', 'add', {}),
|
|
('add', 'max', {}),
|
|
('const_0', 'max', {}),
|
|
('max', 'min', {}),
|
|
('const_6', 'min', {}),
|
|
('min', 'mul_2', {}),
|
|
('const_1_6', 'mul_2', {}),
|
|
])
|
|
|
|
def replace_sub_graph(self, graph: Graph, match: [dict, SubgraphMatch]):
|
|
replace_with_hsigmoid(graph, match['add'], match['mul_2'])
|
|
|
|
|
|
class HSigmoidWithReluDiv(FrontReplacementSubgraph):
|
|
"""
|
|
The transformation looks for the pattern with Relu/Div defining the HSigmoid function:
|
|
HSigmoid(x) = Min(Relu(x + 3.0), 6.0) / 6.0
|
|
"""
|
|
enabled = True
|
|
|
|
def run_after(self):
|
|
return [AttributedClampNormalizer]
|
|
|
|
def pattern(self):
|
|
return dict(
|
|
nodes=[
|
|
('input', dict()),
|
|
('add_const', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 3.0, atol=1e-6))),
|
|
('add', dict(op='Add')),
|
|
('relu', dict(op='ReLU')),
|
|
('min_const', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 6.0, atol=1e-6))),
|
|
('min', dict(op='Minimum')),
|
|
('div_const', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 6.0, atol=1e-6))),
|
|
('div', dict(op='Div')),
|
|
],
|
|
edges=[
|
|
('input', 'add', {'out': 0}),
|
|
('add_const', 'add', {}),
|
|
('add', 'relu', {}),
|
|
('relu', 'min', {}),
|
|
('min_const', 'min', {}),
|
|
('min', 'div', {}),
|
|
('div_const', 'div', {}),
|
|
])
|
|
|
|
def replace_sub_graph(self, graph: Graph, match: [dict, SubgraphMatch]):
|
|
replace_with_hsigmoid(graph, match['add'], match['div'])
|
|
|
|
|
|
class HSigmoidWithReluMul(FrontReplacementSubgraph):
|
|
"""
|
|
The transformation looks for the pattern with Relu/Mul defining the HSigmoid function:
|
|
HSigmoid(x) = Min(Relu(x + 3.0), 6.0) * 1.0/6.0
|
|
"""
|
|
enabled = True
|
|
|
|
def run_after(self):
|
|
return [AttributedClampNormalizer]
|
|
|
|
def pattern(self):
|
|
return dict(
|
|
nodes=[
|
|
('input', dict()),
|
|
('add_const', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 3.0, atol=1e-6))),
|
|
('add', dict(op='Add')),
|
|
('relu', dict(op='ReLU')),
|
|
('min_const', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 6.0, atol=1e-6))),
|
|
('min', dict(op='Minimum')),
|
|
('mul_const', dict(op='Const', value=lambda v: v is not None and np.allclose(v, 1.0/6.0, atol=1e-6))),
|
|
('mul', dict(op='Mul')),
|
|
],
|
|
edges=[
|
|
('input', 'add', {'out': 0}),
|
|
('add_const', 'add', {}),
|
|
('add', 'relu', {}),
|
|
('relu', 'min', {}),
|
|
('min_const', 'min', {}),
|
|
('min', 'mul', {}),
|
|
('mul_const', 'mul', {}),
|
|
])
|
|
|
|
def replace_sub_graph(self, graph: Graph, match: [dict, SubgraphMatch]):
|
|
replace_with_hsigmoid(graph, match['add'], match['mul'])
|