-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathSpeedTest.py
190 lines (169 loc) · 8.79 KB
/
SpeedTest.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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
import onnx
from onnx import numpy_helper
import numpy as np
from ctypes import cdll, c_int, c_float, POINTER, cast
import ctypes
def loadOnnxModel(path):
model = onnx.load(path)
return model
def Add(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape_list = []
output_shape = []
for i in range(len(model.graph.input)):
input_shape = [dim.dim_value for dim in model.graph.input[i].type.tensor_type.shape.dim]
input_shape_list.append(input_shape)
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./op/SpeedTest.so")
if len(input_shape_list) == 2:
cuda_add = lib.cuda_add_2
cuda_add.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int]
NumSize = 1
for i in input_shape_list[0]:
NumSize = NumSize * i
result_array = np.zeros(1, dtype=np.float32)
cuda_add(result_array, NumSize)
elif len(input_shape_list) == 1:
cuda_add = lib.cuda_add_1
cuda_add.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int, c_float]
NumSize = 1
for i in input_shape_list[0]:
NumSize = NumSize * i
result_array = np.zeros(1, dtype=np.float32)
Addinit = onnx.numpy_helper.to_array(model.graph.initializer[0])
cuda_add(result_array, NumSize, Addinit)
else :
print('error')
return result_array[0]
def LeakyRelu(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape = [dim.dim_value for dim in model.graph.input[0].type.tensor_type.shape.dim]
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./op/SpeedTest.so")
cuda_leakyrelu = lib.cuda_leaky_1
cuda_leakyrelu.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int, c_float]
NumSize = 1
for i in input_shape:
NumSize = NumSize * i
result_array = np.zeros(100, dtype=np.float32)
Leakyinit = model.graph.node[0].attribute[0].f
cuda_leakyrelu(result_array, NumSize, Leakyinit)
return result_array[0]
def Abs(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape = [dim.dim_value for dim in model.graph.input[0].type.tensor_type.shape.dim]
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./op/SpeedTest.so")
cuda_abs = lib.cuda_abs_1
cuda_abs.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int]
NumSize = 1
for i in input_shape:
NumSize = NumSize * i
result_array = np.zeros(1, dtype=np.float32)
cuda_abs(result_array, NumSize)
return result_array[0]
def Div(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape = [dim.dim_value for dim in model.graph.input[0].type.tensor_type.shape.dim]
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./op/SpeedTest.so")
cuda_div = lib.cuda_div_1
cuda_div.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int, c_float]
NumSize = 1
for i in input_shape:
NumSize = NumSize * i
result_array = np.zeros(1, dtype=np.float32)
Divinit = onnx.numpy_helper.to_array(model.graph.initializer[0])
cuda_div(result_array, NumSize, Divinit)
return result_array[0]
def Tanh(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape = [dim.dim_value for dim in model.graph.input[0].type.tensor_type.shape.dim]
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./op/SpeedTest.so")
cuda_tanh = lib.cuda_tanh_1
cuda_tanh.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int]
NumSize = 1
for i in input_shape:
NumSize = NumSize * i
result_array = np.zeros(1, dtype=np.float32)
cuda_tanh(result_array, NumSize)
return result_array[0]
#适配性待修改
def Slice(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape = [dim.dim_value for dim in model.graph.input[0].type.tensor_type.shape.dim]
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./op/SpeedTest.so")
cuda_slice = lib.cuda_slice_1
cuda_slice.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), POINTER(c_int), POINTER(c_int)]
result_array = np.zeros(1, dtype=np.float32)
starts = onnx.numpy_helper.to_array(model.graph.initializer[0])[0]
ends = onnx.numpy_helper.to_array(model.graph.initializer[1])[0]
axes = onnx.numpy_helper.to_array(model.graph.initializer[2])[0]
steps = onnx.numpy_helper.to_array(model.graph.initializer[3])[0]
argc = [starts, ends, axes, steps]
cuda_slice(result_array, (c_int * len(input_shape))(*input_shape), (c_int * len(argc))(*argc))
return result_array[0]
def Concat(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape_list = []
for i in range(len(model.graph.input)):
input_shape = [dim.dim_value for dim in model.graph.input[i].type.tensor_type.shape.dim]
input_shape_list.append(input_shape)
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./op/SpeedTest.so")
cuda_concat = lib.cuda_concat_1
cuda_concat.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), POINTER(POINTER(c_int)), c_int, POINTER(c_int), c_int]
result_array = np.zeros(1, dtype=np.float32)
Concatinit = model.graph.node[0].attribute[0].i
rows = len(input_shape_list)
cols = [len(row) for row in input_shape_list]
input_shape_list_c = (POINTER(c_int) * rows)()
input_shape_list_c[:] = [cast((c_int * len(row))(*row), POINTER(c_int)) for row in input_shape_list]
cols_c = (c_int * rows)(*cols)
cuda_concat(result_array, input_shape_list_c, rows, cols_c, Concatinit, (c_int * len(output_shape))(*output_shape))
return result_array[0]
def Conv(ModelPath):
model = loadOnnxModel(str(ModelPath))
input_shape = [dim.dim_value for dim in model.graph.input[0].type.tensor_type.shape.dim]
output_shape = [dim.dim_value for dim in model.graph.output[0].type.tensor_type.shape.dim]
lib = cdll.LoadLibrary("./SpeedTest.so")
if (len(model.graph.initializer) == 1):
cuda_conv2d = lib.cuda_conv2d_1
group = model.graph.node[0].attribute[1].i
kernel = model.graph.node[0].attribute[2].ints
pads = model.graph.node[0].attribute[3].ints
stride = model.graph.node[0].attribute[4].ints
weight = onnx.numpy_helper.to_array(model.graph.initializer[0])
data_ptr = weight.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
cuda_conv2d.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int, POINTER(c_int), POINTER(c_int), POINTER(c_int), POINTER(c_int),
POINTER(c_int), POINTER(c_float)]
result_array = np.zeros(1, dtype=np.float32)
cuda_conv2d(result_array, group, (c_int * len(input_shape))(*input_shape), (c_int * len(output_shape))(*output_shape), (c_int * len(weight.shape))(*weight.shape),
(c_int * len(pads))(*pads), (c_int * len(stride))(*stride), data_ptr)
return result_array[0]
elif (len(model.graph.initializer) == 2):
cuda_conv2d = lib.cuda_conv2d_2
group = model.graph.node[0].attribute[1].i
kernel= model.graph.node[0].attribute[2].ints
pads = model.graph.node[0].attribute[3].ints
stride= model.graph.node[0].attribute[4].ints
weight= onnx.numpy_helper.to_array(model.graph.initializer[0])
bias = onnx.numpy_helper.to_array(model.graph.initializer[1])
data_ptr = weight.ctypes.data_as(ctypes.POINTER(ctypes.c_float))
cuda_conv2d.argtypes = [np.ctypeslib.ndpointer(dtype=np.float32), c_int, POINTER(c_int), POINTER(c_int),
POINTER(c_int), POINTER(c_int), POINTER(c_int), POINTER(c_float), POINTER(c_float)]
result_array = np.zeros(1, dtype=np.float32)
cuda_conv2d(result_array, group, (c_int * len(input_shape))(*input_shape),(c_int * len(output_shape))(*output_shape), (c_int * len(weight.shape))(*weight.shape),
(c_int * len(pads))(*pads), (c_int * len(stride))(*stride), data_ptr, (c_float * len(bias))(*bias))
return result_array[0]
def CallBackTime(ModelPath, OpType):
function_dict = globals()
function_name = OpType
if function_name in function_dict:
function_to_call = function_dict[function_name]
return function_to_call(ModelPath)
if __name__=='__main__':
time = CallBackTime('vis_conv_conv.onnx', 'Conv')
print(time)