forked from tensorflow/tflite-micro
-
Notifications
You must be signed in to change notification settings - Fork 0
/
conv.cc
140 lines (121 loc) · 4.98 KB
/
conv.cc
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
/* Copyright 2021 The TensorFlow Authors. 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.
==============================================================================*/
#include "tensorflow/lite/micro/kernels/conv.h"
#include "tensorflow/lite/c/builtin_op_data.h"
#include "tensorflow/lite/c/common.h"
#include "tensorflow/lite/kernels/internal/common.h"
#include "tensorflow/lite/kernels/internal/quantization_util.h"
#include "tensorflow/lite/kernels/internal/reference/conv.h"
#include "tensorflow/lite/kernels/internal/reference/integer_ops/conv.h"
#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
#include "tensorflow/lite/kernels/kernel_util.h"
#include "tensorflow/lite/kernels/padding.h"
#include "tensorflow/lite/micro/kernels/kernel_util.h"
#include "tensorflow/lite/micro/kernels/xtensa/xtensa.h"
#include "tensorflow/lite/micro/kernels/xtensa/xtensa_conv.h"
#include "tensorflow/lite/micro/micro_log.h"
namespace tflite {
namespace {
void* Init(TfLiteContext* context, const char* buffer, size_t length) {
TFLITE_DCHECK(context->AllocatePersistentBuffer != nullptr);
void* data =
context->AllocatePersistentBuffer(context, sizeof(XtensaConvOpData));
#if defined(VISION_P6)
if (InitXtensaContext()) {
return nullptr;
}
#endif // defined(VISION_P6)
return data;
}
TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
TF_LITE_ENSURE_OK(context, ConvPrepare(context, node));
#if defined(HIFI4) || defined(HIFI5)
TF_LITE_ENSURE_OK(context, ConvPrepareHifi(context, node));
#endif
#if defined(VISION_P6)
TF_LITE_ENSURE_OK(context, ConvPrepareVision(context, node));
#endif // VISION_P6
return kTfLiteOk;
}
TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
TFLITE_DCHECK(node->user_data != nullptr);
TFLITE_DCHECK(node->builtin_data != nullptr);
const TfLiteEvalTensor* input =
tflite::micro::GetEvalInput(context, node, kConvInputTensor);
const auto& params =
*(reinterpret_cast<TfLiteConvParams*>(node->builtin_data));
const auto& op_data = *(reinterpret_cast<XtensaConvOpData*>(node->user_data));
TfLiteEvalTensor* output =
tflite::micro::GetEvalOutput(context, node, kConvOutputTensor);
const TfLiteEvalTensor* filter =
tflite::micro::GetEvalInput(context, node, kConvWeightsTensor);
const TfLiteEvalTensor* bias =
(NumInputs(node) == 3)
? tflite::micro::GetEvalInput(context, node, kConvBiasTensor)
: nullptr;
TfLiteEvalTensor filter_int8 = tflite::micro::MakeUnpackedInt4Tensor(
context, op_data.reference_op_data.filter_buffer_index, filter);
switch (input->type) {
case kTfLiteInt8: {
switch (filter_int8.type) {
case kTfLiteInt8: {
#if defined(HIFI4) || defined(HIFI5)
ConvEvalHifi(context, node, params, op_data, input, &filter_int8,
bias, output);
#elif defined(VISION_P6)
return ConvEvalVision(context, node, params, op_data, input,
&filter_int8, bias, output);
#else
reference_integer_ops::ConvPerChannel(
ConvParamsQuantized(params, op_data.reference_op_data),
op_data.reference_op_data.per_channel_output_multiplier,
op_data.reference_op_data.per_channel_output_shift,
tflite::micro::GetTensorShape(input),
tflite::micro::GetTensorData<int8_t>(input),
tflite::micro::GetTensorShape(filter),
tflite::micro::GetTensorData<int8_t>(&filter_int8),
tflite::micro::GetTensorShape(bias),
tflite::micro::GetOptionalTensorData<int32_t>(bias),
tflite::micro::GetTensorShape(output),
tflite::micro::GetTensorData<int8_t>(output));
return kTfLiteOk;
#endif
break;
}
default:
MicroPrintf("Filter type %s (%d) not supported.",
TfLiteTypeGetName(filter->type), filter->type);
return kTfLiteError;
}
return kTfLiteOk;
}
case kTfLiteInt16: {
#if defined(HIFI4)
ConvEvalHifi16(context, node, params, op_data, input, filter, bias,
output);
#else
return ConvReferenceEvalInt16(context, node);
#endif // defined(HIFI4)
break;
}
default:
MicroPrintf("Type %s (%d) not supported.", TfLiteTypeGetName(input->type),
input->type);
return kTfLiteError;
}
return kTfLiteOk;
}
} // namespace
TFLMRegistration Register_CONV_2D() {
return tflite::micro::RegisterOp(Init, Prepare, Eval);
}
} // namespace tflite