Evgeny Lazarev
970b1301b5
* Removed back phase transformations related to IRv7 * Fixed setting value for the input port using the 'set_value' method * Removed front and middle phase transformations related to IRv7 * Cleanup the rest of the Model Optimizer transformations from IRv7 specific transformations * Final cleanup of the deprecated IR v7 related code * Removed 'blobs_as_input' usage in the Model Optimizer. * Removed function '_fuse_add' from the Model Optimizer since it is not used anymore. * Removed 'keep_in_IR' node attribute for FakeQuantize ops in the MO * Disabled failing gpu_engine.user_context test
101 lines
4.3 KiB
Python
101 lines
4.3 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 logging as log
|
|
|
|
import numpy as np
|
|
|
|
from extensions.ops.fakequantize import FakeQuantize
|
|
from mo.front.common.replacement import FrontReplacementSubgraph
|
|
from mo.front.subgraph_matcher import SubgraphMatch
|
|
from mo.graph.graph import Graph, rename_nodes
|
|
from mo.ops.const import Const
|
|
from mo.utils.error import Error
|
|
|
|
|
|
class QuantizeDequantizeLinear(FrontReplacementSubgraph):
|
|
"""
|
|
Fuses QuantizeLinear and DequantizeLinear nodes into single FakeQuantize.
|
|
Covers cases when the values for zero point and scale are same in both QuantizeLinear and DequantizeLinear.
|
|
"""
|
|
enabled = True
|
|
|
|
def pattern(self):
|
|
return dict(
|
|
nodes=[
|
|
('quantize', dict(op='QuantizeLinear')),
|
|
('dequantize', dict(op='DequantizeLinear')),
|
|
],
|
|
edges=[
|
|
('quantize', 'dequantize', {'in': 0}),
|
|
]
|
|
)
|
|
|
|
def replace_sub_graph(self, graph: Graph, match: [dict, SubgraphMatch]):
|
|
|
|
q = match['quantize']
|
|
dq = match['dequantize']
|
|
|
|
q_scale = q.in_port(1).get_source().node
|
|
q_zerop = q.in_port(2).get_source().node
|
|
dq_scale = dq.in_port(1).get_source().node
|
|
dq_zerop = dq.in_port(2).get_source().node
|
|
|
|
inp_port = q.in_port(0).get_source()
|
|
name = inp_port.node.soft_get('name', inp_port.node.id)
|
|
|
|
# only constant as for zero_point/scale supported
|
|
if q_scale.soft_get('type') == 'Const' and dq_scale.soft_get('type') == 'Const' and \
|
|
q_zerop.soft_get('type') == 'Const' and dq_zerop.soft_get('type') == 'Const':
|
|
|
|
# only patterns with same scale/zero_point values for Q and DQ are supported
|
|
if q_scale.value == dq_scale.value and q_zerop.value == dq_zerop.value:
|
|
log.debug('Found Q-DQ pattern after {}'.format(name))
|
|
|
|
zero_point_type = q_zerop.value.dtype
|
|
# data type affects range of output values: [-128..127] or [0..255]
|
|
if zero_point_type == np.int8:
|
|
output_min_value = -128.0
|
|
output_max_value = 127.0
|
|
elif zero_point_type == np.uint8:
|
|
output_min_value = 0.0
|
|
output_max_value = 255.0
|
|
else:
|
|
raise Error('Not supported type {} for zero point value in node {}'.format(
|
|
zero_point_type, q_zerop.soft_get('name')))
|
|
min_value = q_scale.value * (output_min_value - q_zerop.value)
|
|
max_value = q_scale.value * (output_max_value - q_zerop.value)
|
|
input_min = Const(graph, {'value': np.array(min_value)}).create_node()
|
|
input_max = Const(graph, {'value': np.array(max_value)}).create_node()
|
|
|
|
FQ = FakeQuantize(graph, {
|
|
'levels': 256,
|
|
'name': match['quantize'].name + '_Dequantize/FakeQuantize'
|
|
}).create_node()
|
|
|
|
FQ.in_port(0).connect(match['quantize'].in_port(0).get_source())
|
|
FQ.in_port(1).connect(input_min.out_port(0))
|
|
FQ.in_port(2).connect(input_max.out_port(0))
|
|
FQ.in_port(3).connect(input_min.out_port(0))
|
|
FQ.in_port(4).connect(input_max.out_port(0))
|
|
|
|
match['dequantize'].out_port(0).get_connection().set_source(FQ.out_port(0))
|
|
dq_name = match['dequantize'].soft_get('name', match['dequantize'].id)
|
|
rename_nodes([(match['dequantize'], dq_name + '/to_be_removed'), (FQ, dq_name)])
|
|
else:
|
|
raise Error('QuantizeLinear and DequantizeLinear (after {}) have different scale or zero-point values, '
|
|
'cannot fuse into FakeQuantize!'.format(name))
|