1 /* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15 #include "tensorflow/lite/c/builtin_op_data.h"
16 #include "tensorflow/lite/c/common.h"
17 #include "tensorflow/lite/kernels/internal/reference/reference_ops.h"
18 #include "tensorflow/lite/kernels/internal/tensor.h"
19 #include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
20 #include "tensorflow/lite/kernels/internal/types.h"
21 #include "tensorflow/lite/kernels/kernel_util.h"
22
23 namespace tflite {
24 namespace ops {
25 namespace builtin {
26 namespace fake_quant {
27
28 // This file has reference implementation of FakeQuant.
29 enum KernelType {
30 kReference,
31 };
32
33 struct OpContext {
OpContexttflite::ops::builtin::fake_quant::OpContext34 OpContext(TfLiteContext* context, TfLiteNode* node) {
35 input = GetInput(context, node, 0);
36 output = GetOutput(context, node, 0);
37 }
38 const TfLiteTensor* input;
39 TfLiteTensor* output;
40 };
41
Prepare(TfLiteContext * context,TfLiteNode * node)42 TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
43 TF_LITE_ENSURE_EQ(context, NumInputs(node), 1);
44 TF_LITE_ENSURE_EQ(context, NumOutputs(node), 1);
45
46 const auto* params =
47 reinterpret_cast<TfLiteFakeQuantParams*>(node->builtin_data);
48
49 if (params->narrow_range) {
50 TF_LITE_KERNEL_LOG(
51 context,
52 "narrow_range FakeQuant is not currently supported at runtime. "
53 "narrow_range is only meant to be applied to weights, not activations");
54 return kTfLiteError;
55 }
56
57 OpContext op_context(context, node);
58 TfLiteIntArray* output_dims = TfLiteIntArrayCopy(op_context.input->dims);
59 op_context.output->type = op_context.input->type;
60 return context->ResizeTensor(context, op_context.output, output_dims);
61 }
62
63 template <KernelType kernel_type>
Eval(TfLiteContext * context,TfLiteNode * node)64 TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
65 OpContext op_context(context, node);
66
67 const auto* params =
68 reinterpret_cast<TfLiteFakeQuantParams*>(node->builtin_data);
69
70 tflite::FakeQuantParams op_params;
71 op_params.num_bits = params->num_bits;
72 op_params.minmax.min = params->min;
73 op_params.minmax.max = params->max;
74 reference_ops::FakeQuant(op_params, GetTensorShape(op_context.input),
75 GetTensorData<float>(op_context.input),
76 GetTensorShape(op_context.output),
77 GetTensorData<float>(op_context.output));
78
79 return kTfLiteOk;
80 }
81
82 } // namespace fake_quant
83
Register_FAKE_QUANT_REF()84 TfLiteRegistration* Register_FAKE_QUANT_REF() {
85 static TfLiteRegistration r = {nullptr, nullptr, fake_quant::Prepare,
86 fake_quant::Eval<fake_quant::kReference>};
87 return &r;
88 }
89
Register_FAKE_QUANT()90 TfLiteRegistration* Register_FAKE_QUANT() { return Register_FAKE_QUANT_REF(); }
91
92 } // namespace builtin
93 } // namespace ops
94 } // namespace tflite
95