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
|
/* Copyright 2017 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.
==============================================================================*/
template <typename T>
void Transpose(const TransposeParams& params,
const RuntimeShape& unextended_input_shape, const T* input_data,
const RuntimeShape& unextended_output_shape, T* output_data) {
const int unextended_output_size = unextended_output_shape.DimensionsCount();
TFLITE_DCHECK_LE(unextended_input_shape.DimensionsCount(), 4);
TFLITE_DCHECK_LE(unextended_output_size, 4);
TFLITE_DCHECK_EQ(unextended_output_size, params.perm_count);
const RuntimeShape input_shape =
RuntimeShape::ExtendedShape(4, unextended_input_shape);
const RuntimeShape output_shape =
RuntimeShape::ExtendedShape(4, unextended_output_shape);
const int input_ext_size = 4 - unextended_input_shape.DimensionsCount();
const int output_ext_size = 4 - unextended_output_size;
// The perm data is extended to match the output, each index incremented by
// the amount of front padding of the input shape.
int extended_perm[4];
for (int i = 0; i < output_ext_size; ++i) {
extended_perm[i] = i;
}
for (int i = 0; i < unextended_output_size; ++i) {
extended_perm[i + output_ext_size] = params.perm[i] + input_ext_size;
}
int out_sizes[4];
// Compute the inverse permutation array so we can do an output centered
// transpose. Also, check to make sure output_dims is matching input_dims.
for (int k = 0; k < 4; k++) {
out_sizes[k] = MatchingDim(input_shape, extended_perm[k], output_shape, k);
}
// Naive transpose loop (iterate on output index and compute input index).
int o[4]; // loop index (on output).
int i[4];
for (o[3] = 0; o[3] < out_sizes[3]; o[3]++) {
i[extended_perm[3]] = o[3];
for (o[2] = 0; o[2] < out_sizes[2]; o[2]++) {
i[extended_perm[2]] = o[2];
for (o[1] = 0; o[1] < out_sizes[1]; o[1]++) {
i[extended_perm[1]] = o[1];
for (o[0] = 0; o[0] < out_sizes[0]; o[0]++) {
i[extended_perm[0]] = o[0];
output_data[Offset(output_shape, o)] =
input_data[Offset(input_shape, i)];
}
}
}
}
}
|
