-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
97 lines (84 loc) · 3.28 KB
/
utils.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
from __future__ import annotations
import numpy as np
import onnx
from onnx.mapping import TENSOR_TYPE_TO_NP_TYPE, NP_TYPE_TO_TENSOR_TYPE
# names are variable names in generated code, so shouldn't result in invalid var names
def clean_name(name: str) -> str:
name = name.replace("::", "_") # some var names are onnx::gemm_input
name = name.replace(".", "_") # some var names conatins dots
name = name.replace("/", "_")
if (not name[0].isalpha()) and name[0]!="_": # some var names only contain numbers
name = "_" + name
return name
def map_onnx_dtype_to_numpy(onnx_dtype: int):
return TENSOR_TYPE_TO_NP_TYPE[onnx_dtype]
def map_np_type_to_onnx(np_type):
np_type = np.dtype(np_type)
return NP_TYPE_TO_TENSOR_TYPE[np_type]
def map_type(elem_type: int) -> str:
np_type = TENSOR_TYPE_TO_NP_TYPE[elem_type]
# TODO: add support for types as needed!
# https://github.com/onnx/onnx/blob/4e0b7197a015549f6773d22d174f854d7782295d/onnx/mapping.py#L13
if np_type == np.dtype("float16"):
return "float16"
elif np_type == np.dtype("float32"):
return "float"
elif np_type == np.dtype("uint8"):
return "uint8"
elif np_type == np.dtype("int64"):
return "int64"
elif np_type == np.dtype("int32"):
return "int32"
else:
raise NotImplementedError(f"type mapping for {elem_type} (np type = {np_type}) is not implemented yet")
def map_type_to_onnx_str(type) -> str:
if type == "float16":
return "ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16"
elif type == "float":
return "ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT"
elif type == "uint8":
return "ONNX_TENSOR_ELEMENT_DATA_TYPE_UINT8"
elif type == "int64":
return "ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64"
elif type == "int32":
return "ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32"
else:
raise NotImplementedError(f"type mapping for {type} is not implemented yet")
# #TODO: numpy class to str mapping a bit ugly
# type = str(type)
# if type.startswith("<class"):
# type = type.split("'")[1]
# if type.startswith("numpy"):
# type = type.split(".")[1]
# if type == "float":
# return "ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT"
# elif type == "int64":
# return "ONNX_TENSOR_ELEMENT_DATA_TYPE_INT64"
# elif type == "float16":
# return "ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT16"
# else:
# raise NotImplementedError(f"type mapping for {type} is not implemented yet")
def map_type_to_ait_str(type) -> str:
# TODO: need to check np types
if type == np.dtype("float16"):
return "kHalf"
elif type == np.dtype("float"):
return "kFloat"
elif type == np.dtype("int32"):
return "kInt"
elif type == np.dtype("int64"):
return "kLong"
else:
raise NotImplementedError(f"type mapping for {type} is not implemented yet")
def to_attribute_dict(attributes: list[onnx.AttributeProto]) -> dict:
d = {}
for attr in attributes:
d[attr.name] = attr
return d
def add_input_to_node(node, init_name, pos: int):
if len(node.input) > pos:
node.input[pos] = init_name
else:
for _ in range(len(node.input), pos):
node.input.append("")
node.input.append(init_name)