1 /*
2 * Copyright (C) 2018 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #define LOG_TAG "Operations"
18
19 #include "ChannelShuffle.h"
20
21 #include "OperationResolver.h"
22 #include "OperationsExecutionUtils.h"
23 #include "Tracing.h"
24
25 namespace android {
26 namespace nn {
27 namespace channel_shuffle {
28
29 template <typename T>
eval(const T * inputData,const Shape & inputShape,int32_t numGroups,int32_t axis,T * outputData)30 inline bool eval(const T* inputData, const Shape& inputShape, int32_t numGroups, int32_t axis,
31 T* outputData) {
32 const uint32_t outerSize = getNumberOfElements(inputShape, 0, axis);
33 const uint32_t axisSize = getSizeOfDimension(inputShape, axis);
34 const uint32_t innerSize =
35 getNumberOfElements(inputShape, axis + 1, getNumberOfDimensions(inputShape));
36 const uint32_t groupSize = axisSize / numGroups;
37 for (uint32_t outer = 0; outer < outerSize; ++outer) {
38 for (uint32_t inner = 0; inner < innerSize; ++inner) {
39 const T* inputBase = inputData + outer * axisSize * innerSize + inner;
40 T* outputBase = outputData + outer * axisSize * innerSize + inner;
41 for (uint32_t i = 0; i < groupSize; i++) {
42 for (uint32_t j = 0; j < static_cast<uint32_t>(numGroups);
43 j++, outputBase += innerSize) {
44 *outputBase = inputBase[innerSize * (i + j * groupSize)];
45 }
46 }
47 }
48 }
49 return true;
50 }
51
prepare(IOperationExecutionContext * context)52 bool prepare(IOperationExecutionContext* context) {
53 Shape input = context->getInputShape(kInputTensor);
54 int32_t numGroups = context->getInputValue<int32_t>(kNumGroups);
55 int32_t axis = context->getInputValue<int32_t>(kInputAxis);
56 NN_RET_CHECK(handleNegativeAxis(input, &axis));
57 NN_RET_CHECK(numGroups > 0);
58 NN_RET_CHECK(getSizeOfDimension(input, axis) % numGroups == 0);
59 return context->setOutputShape(kOutputTensor, input);
60 }
61
execute(IOperationExecutionContext * context)62 bool execute(IOperationExecutionContext* context) {
63 int32_t numGroups = context->getInputValue<int32_t>(kNumGroups);
64 int32_t axis = context->getInputValue<int32_t>(kInputAxis);
65 NN_RET_CHECK(handleNegativeAxis(context->getInputShape(kInputTensor), &axis));
66 switch (context->getInputType(kInputTensor)) {
67 case OperandType::TENSOR_FLOAT16:
68 return eval(context->getInputBuffer<_Float16>(kInputTensor),
69 context->getInputShape(kInputTensor), numGroups, axis,
70 context->getOutputBuffer<_Float16>(kOutputTensor));
71 case OperandType::TENSOR_FLOAT32:
72 return eval(context->getInputBuffer<float>(kInputTensor),
73 context->getInputShape(kInputTensor), numGroups, axis,
74 context->getOutputBuffer<float>(kOutputTensor));
75 case OperandType::TENSOR_QUANT8_ASYMM:
76 return eval(context->getInputBuffer<uint8_t>(kInputTensor),
77 context->getInputShape(kInputTensor), numGroups, axis,
78 context->getOutputBuffer<uint8_t>(kOutputTensor));
79 case OperandType::TENSOR_QUANT8_ASYMM_SIGNED:
80 return eval(context->getInputBuffer<int8_t>(kInputTensor),
81 context->getInputShape(kInputTensor), numGroups, axis,
82 context->getOutputBuffer<int8_t>(kOutputTensor));
83 default:
84 NN_RET_CHECK_FAIL() << "Unsupported tensor type for operation " << kOperationName;
85 }
86 }
87
88 } // namespace channel_shuffle
89
90 NN_REGISTER_OPERATION_DEFAULT_VALIDATION(CHANNEL_SHUFFLE, channel_shuffle::prepare,
91 channel_shuffle::execute);
92
93 } // namespace nn
94 } // namespace android
95