-
Notifications
You must be signed in to change notification settings - Fork 178
/
quant_tensor.py
160 lines (118 loc) · 5.75 KB
/
quant_tensor.py
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
155
156
157
158
159
160
# Copyright (c) 2018- Xilinx, Inc (Alessandro Pappalardo)
# Copyright (c) 2016- Facebook, Inc (Adam Paszke)
# Copyright (c) 2014- Facebook, Inc (Soumith Chintala)
# Copyright (c) 2011-2014 Idiap Research Institute (Ronan Collobert)
# Copyright (c) 2012-2014 Deepmind Technologies (Koray Kavukcuoglu)
# Copyright (c) 2011-2012 NEC Laboratories America (Koray Kavukcuoglu)
# Copyright (c) 2011-2013 NYU (Clement Farabet)
# Copyright (c) 2006-2010 NEC Laboratories America (Ronan Collobert, Leon Bottou, Iain Melvin, Jason Weston)
# Copyright (c) 2006 Idiap Research Institute (Samy Bengio)
# Copyright (c) 2001-2004 Idiap Research Institute (Ronan Collobert, Samy Bengio, Johnny Mariethoz)
# All rights reserved.
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
# 1. Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# 2. Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# 3. Neither the names of Xilinx, Facebook, Deepmind Technologies, NYU,
# NEC Laboratories America and IDIAP Research Institute nor the names
# of its contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
# POSSIBILITY OF SUCH DAMAGE.
from collections import namedtuple
import torch
from brevitas.function.ops_ste import round_ste, ceil_ste
from brevitas.function.ops import max_uint
def pack_quant_tensor(tensor, scale, bit_width):
return QuantTensor._make([tensor, scale, bit_width])
class QuantTensor(namedtuple("QuantTensor", ["tensor", "scale", "bit_width"])):
@staticmethod
def check_input_type(other):
if not isinstance(other, QuantTensor):
raise Exception("Other tensor is not a QuantTensor")
def check_scaling_factors_same(self, other):
if not torch.allclose(self.scale, other.scale):
raise Exception("Scalign factors are different")
# Reference: https://docs.python.org/3/reference/datamodel.html#emulating-numeric-types
def __neg__(self):
return QuantTensor._make([- self.tensor, self.scale, self.bit_width])
def __add__(self, other):
QuantTensor.check_input_type(other)
self.check_scaling_factors_same(other)
output_tensor = self.tensor + other.tensor
output_scale = (self.scale + other.scale) / 2
max_uint_val = max_uint(narrow_range=False, bit_width=self.bit_width)
max_uint_val += max_uint(narrow_range=False, bit_width=other.bit_width)
output_bit_width = ceil_ste(torch.log2(max_uint_val))
output = pack_quant_tensor(output_tensor, output_scale, output_bit_width)
return output
def __mul__(self, other):
QuantTensor.check_input_type(other)
output_tensor = self.tensor * other.tensor
output_scale = self.scale * other.scale
output_bit_width = self.bit_width + other.bit_width
output = pack_quant_tensor(output_tensor, output_scale, output_bit_width)
return output
def __sub__(self, other):
return self.__add__(- other)
def __truediv__(self, other):
QuantTensor.check_input_type(other)
output_tensor = self.tensor / other.tensor
output_scale = self.scale / other.scale
output_bit_width = self.bit_width - other.bit_width
output = pack_quant_tensor(output_tensor, output_scale, output_bit_width)
return output
def __abs__(self):
return QuantTensor._make([abs(self.tensor), self.scale, self.bit_width])
def __pos__(self):
return self
def __int__(self):
return round_ste(self.tensor / self.scale)
def __float__(self):
return self.tensor
def __index__(self):
raise NotImplementedError
def __round__(self):
raise NotImplementedError
def __trunc__(self):
raise NotImplementedError
def __floor__(self):
raise NotImplementedError
def __ceil__(self):
raise NotImplementedError
def __complex__(self):
raise NotImplementedError
def __invert__(self):
raise NotImplementedError
def __matmul__(self, other):
raise NotImplementedError
def __floordiv__(self, other):
raise NotImplementedError
def __mod__(self, other):
raise NotImplementedError
def __divmod__(self, other):
raise NotImplementedError
def __pow__(self, other):
raise NotImplementedError
def __lshift__(self, other):
raise NotImplementedError
def __rshift__(self, other):
raise NotImplementedError
def __and__(self, other):
raise NotImplementedError
def __xor__(self, other):
raise NotImplementedError
def __or__(self, other):
raise NotImplementedError