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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
|
#!/usr/bin/env python
# Copyright (c) 2022 Samsung Electronics Co., Ltd. All Rights Reserved
#
# 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.
class CONSTANT:
__slots__ = () # This prevents access via __dict__.
# Basic optimization passes
# These passes do not change the execution result of the model
O1 = (
# Constant folding
'fold_add_v2',
'fold_cast',
'fold_densify',
'fold_dequantize',
'fold_dwconv',
'fold_fully_connected',
'fold_gather',
'fold_sparse_to_dense',
# Operator fusion
'fuse_add_with_tconv',
'fuse_add_with_fully_connected',
'fuse_batchnorm_with_conv',
'fuse_batchnorm_with_dwconv',
'fuse_batchnorm_with_tconv',
'fuse_activation_function',
'fuse_instnorm',
'fuse_prelu',
'fuse_gelu',
'fuse_mean_with_mean',
'fuse_transpose_with_mean',
'transform_min_max_to_relu6',
'transform_min_relu_to_relu6',
# Remove redundant operators
'remove_redundant_reshape',
'remove_redundant_transpose',
'remove_unnecessary_reshape',
'remove_unnecessary_slice',
'remove_unnecessary_strided_slice',
'remove_unnecessary_split',
# Canonicalization
# (passes to help further optimization)
'resolve_customop_add',
'resolve_customop_batchmatmul',
'resolve_customop_matmul',
'resolve_customop_max_pool_with_argmax',
'resolve_customop_splitv',
'substitute_pack_to_reshape',
'substitute_padv2_to_pad',
'substitute_splitv_to_split',
'substitute_squeeze_to_reshape',
'substitute_strided_slice_to_reshape',
'substitute_transpose_to_reshape',
'forward_reshape_to_unaryop',
'forward_transpose_op',
'replace_non_const_fc_with_batch_matmul', # For quantization
)
OPTIMIZATION_OPTS = (
# (OPTION_NAME, HELP_MESSAGE)
('convert_nchw_to_nhwc',
'Experimental: This will convert NCHW operators to NHWC under the assumption that input model is NCHW.'
),
('expand_broadcast_const', 'expand broadcastable constant node inputs'),
('nchw_to_nhwc_input_shape',
'convert the input shape of the model (argument for convert_nchw_to_nhwc)'),
('nchw_to_nhwc_output_shape',
'convert the output shape of the model (argument for convert_nchw_to_nhwc)'),
('fold_add_v2', 'fold AddV2 op with constant inputs'),
('fold_cast', 'fold Cast op with constant input'),
('fold_densify', 'fold Densify op with sparse constant input'),
('fold_dequantize', 'fold Dequantize op'),
('fold_dwconv', 'fold Depthwise Convolution op with constant inputs'),
('fold_fully_connected', 'fold FullyConnected op with constant inputs'),
('fold_gather', 'fold Gather op'),
('fold_sparse_to_dense', 'fold SparseToDense op'),
('forward_reshape_to_unaryop', 'Forward Reshape op'),
('forward_transpose_op', 'Forward Transpose op'),
('fuse_add_with_tconv', 'fuse Add op to Transposed'),
('fuse_add_with_fully_connected', 'fuse Add op to FullyConnected op'),
('fuse_batchnorm_with_conv', 'fuse BatchNorm op to Convolution op'),
('fuse_batchnorm_with_dwconv', 'fuse BatchNorm op to Depthwise Convolution op'),
('fuse_batchnorm_with_tconv', 'fuse BatchNorm op to Transposed Convolution op'),
('fuse_bcq', 'apply Binary Coded Quantization'),
('fuse_preactivation_batchnorm',
'fuse BatchNorm operators of pre-activations to Convolution op'),
('fuse_mean_with_mean', 'fuse two consecutive Mean ops'),
('fuse_transpose_with_mean',
'fuse Mean with a preceding Transpose under certain conditions'),
('make_batchnorm_gamma_positive',
'make negative gamma of BatchNorm to a small positive value (1e-10).'
' Note that this pass can change the execution result of the model.'
' So, use it only when the impact is known to be acceptable.'),
('fuse_activation_function', 'fuse Activation function to a preceding operator'),
('fuse_instnorm', 'fuse ops to InstanceNorm operator'),
('fuse_prelu', 'fuse ops to PReLU operator'),
('fuse_gelu', 'fuse ops to GeLU operator'),
('replace_cw_mul_add_with_depthwise_conv',
'replace channel-wise Mul/Add with DepthwiseConv2D'),
('remove_fakequant', 'remove FakeQuant ops'),
('remove_quantdequant', 'remove Quantize-Dequantize sequence'),
('remove_redundant_quantize', 'remove redundant Quantize ops'),
('remove_redundant_reshape', 'fuse or remove subsequent Reshape ops'),
('remove_redundant_transpose', 'fuse or remove subsequent Transpose ops'),
('remove_unnecessary_reshape', 'remove unnecessary reshape ops'),
('remove_unnecessary_slice', 'remove unnecessary slice ops'),
('remove_unnecessary_strided_slice', 'remove unnecessary strided slice ops'),
('remove_unnecessary_split', 'remove unnecessary split ops'),
('replace_non_const_fc_with_batch_matmul',
'replace FullyConnected op with non-const weights to BatchMatMul op'),
('replace_sub_with_add', 'replace Sub op with Add op'),
('resolve_customop_add', 'convert Custom(Add) op to Add op'),
('resolve_customop_batchmatmul',
'convert Custom(BatchMatmul) op to BatchMatmul op'),
('resolve_customop_matmul', 'convert Custom(Matmul) op to Matmul op'),
('resolve_customop_max_pool_with_argmax',
'convert Custom(MaxPoolWithArgmax) to net of builtin operators'),
('resolve_customop_splitv', 'convert Custom(SplitV) op to SplitV op'),
('shuffle_weight_to_16x1float32',
'convert weight format of FullyConnected op to SHUFFLED16x1FLOAT32.'
' Note that it only converts weights whose row is a multiple of 16'),
('substitute_pack_to_reshape', 'convert single input Pack op to Reshape op'),
('substitute_padv2_to_pad', 'convert certain condition PadV2 to Pad'),
('substitute_splitv_to_split', 'convert certain condition SplitV to Split'),
('substitute_squeeze_to_reshape', 'convert certain condition Squeeze to Reshape'),
('substitute_strided_slice_to_reshape',
'convert certain condition StridedSlice to Reshape'),
('substitute_transpose_to_reshape',
'convert certain condition Transpose to Reshape'),
('transform_min_max_to_relu6', 'transform Minimum-Maximum pattern to Relu6 op'),
('transform_min_relu_to_relu6', 'transform Minimum(6)-Relu pattern to Relu6 op'),
('decompose_hardswish', 'decompose the HardSwish op to Add, Mul and Relu6 ops'),
('unroll_unidirseqlstm', 'unroll UnidirectionalSequenceLSTM op'),
('dynamic_batch_to_single_batch',
'convert dynamic batch size (first dimension) of inputs to 1'))
CONSTANT = CONSTANT()
|
