File size: 6,885 Bytes
b291f6a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
191
192
193
194
195
196
197
198
199
200
201
202
203
import warnings
from collections import namedtuple
from functools import partial
from pathlib import Path
from typing import List, Optional, Union

import numpy as np
import onnxruntime

try:
    import tensorrt as trt
except Exception:
    trt = None
import torch

warnings.filterwarnings(action='ignore', category=DeprecationWarning)


class TRTWrapper(torch.nn.Module):
    dtype_mapping = {}

    def __init__(self, weight: Union[str, Path],
                 device: Optional[torch.device]):
        super().__init__()
        weight = Path(weight) if isinstance(weight, str) else weight
        assert weight.exists() and weight.suffix in ('.engine', '.plan')
        if isinstance(device, str):
            device = torch.device(device)
        elif isinstance(device, int):
            device = torch.device(f'cuda:{device}')
        self.weight = weight
        self.device = device
        self.stream = torch.cuda.Stream(device=device)
        self.__update_mapping()
        self.__init_engine()
        self.__init_bindings()

    def __update_mapping(self):
        self.dtype_mapping.update({
            trt.bool: torch.bool,
            trt.int8: torch.int8,
            trt.int32: torch.int32,
            trt.float16: torch.float16,
            trt.float32: torch.float32
        })

    def __init_engine(self):
        logger = trt.Logger(trt.Logger.ERROR)
        self.log = partial(logger.log, trt.Logger.ERROR)
        trt.init_libnvinfer_plugins(logger, namespace='')
        self.logger = logger
        with trt.Runtime(logger) as runtime:
            model = runtime.deserialize_cuda_engine(self.weight.read_bytes())

        context = model.create_execution_context()

        names = [model.get_binding_name(i) for i in range(model.num_bindings)]

        num_inputs, num_outputs = 0, 0

        for i in range(model.num_bindings):
            if model.binding_is_input(i):
                num_inputs += 1
            else:
                num_outputs += 1

        self.is_dynamic = -1 in model.get_binding_shape(0)

        self.model = model
        self.context = context
        self.input_names = names[:num_inputs]
        self.output_names = names[num_inputs:]
        self.num_inputs = num_inputs
        self.num_outputs = num_outputs
        self.num_bindings = num_inputs + num_outputs
        self.bindings: List[int] = [0] * self.num_bindings

    def __init_bindings(self):
        Binding = namedtuple('Binding', ('name', 'dtype', 'shape'))
        inputs_info = []
        outputs_info = []

        for i, name in enumerate(self.input_names):
            assert self.model.get_binding_name(i) == name
            dtype = self.dtype_mapping[self.model.get_binding_dtype(i)]
            shape = tuple(self.model.get_binding_shape(i))
            inputs_info.append(Binding(name, dtype, shape))

        for i, name in enumerate(self.output_names):
            i += self.num_inputs
            assert self.model.get_binding_name(i) == name
            dtype = self.dtype_mapping[self.model.get_binding_dtype(i)]
            shape = tuple(self.model.get_binding_shape(i))
            outputs_info.append(Binding(name, dtype, shape))
        self.inputs_info = inputs_info
        self.outputs_info = outputs_info
        if not self.is_dynamic:
            self.output_tensor = [
                torch.empty(o.shape, dtype=o.dtype, device=self.device)
                for o in outputs_info
            ]

    def forward(self, *inputs):

        assert len(inputs) == self.num_inputs

        contiguous_inputs: List[torch.Tensor] = [
            i.contiguous() for i in inputs
        ]

        for i in range(self.num_inputs):
            self.bindings[i] = contiguous_inputs[i].data_ptr()
            if self.is_dynamic:
                self.context.set_binding_shape(
                    i, tuple(contiguous_inputs[i].shape))

        # create output tensors
        outputs: List[torch.Tensor] = []

        for i in range(self.num_outputs):
            j = i + self.num_inputs
            if self.is_dynamic:
                shape = tuple(self.context.get_binding_shape(j))
                output = torch.empty(
                    size=shape,
                    dtype=self.output_dtypes[i],
                    device=self.device)

            else:
                output = self.output_tensor[i]
            outputs.append(output)
            self.bindings[j] = output.data_ptr()

        self.context.execute_async_v2(self.bindings, self.stream.cuda_stream)
        self.stream.synchronize()

        return tuple(outputs)


class ORTWrapper(torch.nn.Module):

    def __init__(self, weight: Union[str, Path],
                 device: Optional[torch.device]):
        super().__init__()
        weight = Path(weight) if isinstance(weight, str) else weight
        assert weight.exists() and weight.suffix == '.onnx'

        if isinstance(device, str):
            device = torch.device(device)
        elif isinstance(device, int):
            device = torch.device(f'cuda:{device}')
        self.weight = weight
        self.device = device
        self.__init_session()
        self.__init_bindings()

    def __init_session(self):
        providers = ['CPUExecutionProvider']
        if 'cuda' in self.device.type:
            providers.insert(0, 'CUDAExecutionProvider')

        session = onnxruntime.InferenceSession(
            str(self.weight), providers=providers)
        self.session = session

    def __init_bindings(self):
        Binding = namedtuple('Binding', ('name', 'dtype', 'shape'))
        inputs_info = []
        outputs_info = []
        self.is_dynamic = False
        for i, tensor in enumerate(self.session.get_inputs()):
            if any(not isinstance(i, int) for i in tensor.shape):
                self.is_dynamic = True
            inputs_info.append(
                Binding(tensor.name, tensor.type, tuple(tensor.shape)))

        for i, tensor in enumerate(self.session.get_outputs()):
            outputs_info.append(
                Binding(tensor.name, tensor.type, tuple(tensor.shape)))
        self.inputs_info = inputs_info
        self.outputs_info = outputs_info
        self.num_inputs = len(inputs_info)

    def forward(self, *inputs):

        assert len(inputs) == self.num_inputs

        contiguous_inputs: List[np.ndarray] = [
            i.contiguous().cpu().numpy() for i in inputs
        ]

        if not self.is_dynamic:
            # make sure input shape is right for static input shape
            for i in range(self.num_inputs):
                assert contiguous_inputs[i].shape == self.inputs_info[i].shape

        outputs = self.session.run([o.name for o in self.outputs_info], {
            j.name: contiguous_inputs[i]
            for i, j in enumerate(self.inputs_info)
        })

        return tuple(torch.from_numpy(o).to(self.device) for o in outputs)