[onert/train] Add PadLayer op to train backend

runtime/onert/backend/train/ops/OperationUtils.h

@@ -35,6 +35,7 @@ using cpu::ops::getShape;
 using cpu::ops::getNumberOfDimensions;
 using cpu::ops::getNumberOfElements;
 using cpu::ops::getSizeOfDimension;
+using cpu::ops::ConstDataPtr;
 
 /**
  * @brief backpropagate acitvation

runtime/onert/backend/train/ops/PadLayer.cc

@@ -0,0 +1,118 @@
+/*
+ * Copyright (c) 2024 Samsung Electronics Co., Ltd. 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 "PadLayer.h"
+
+#include <cker/train/operation/Pad.h>
+
+namespace onert
+{
+namespace backend
+{
+namespace train
+{
+namespace ops
+{
+
+PadLayer::PadLayer()
+  : _input(nullptr), _output(nullptr), _padData(), _padRank(), _constantValueData(),
+    _back_prop_input{nullptr}, _back_prop_output{nullptr}
+{
+  // DO NOTHING
+}
+
+template <typename T> void PadLayer::padImpl(const T *constant_value_data)
+{
+  nnfw::cker::Pad<T>(_padData, _padRank, getShape(_input), getBuffer<T>(_input), getShape(_output),
+                     getBuffer<T>(_output), constant_value_data);
+}
+
+template <typename T> void PadLayer::depad()
+{
+  nnfw::cker::train::Depad<T>(_padData, _padRank, getShape(_back_prop_output),
+                              getBuffer<T>(_back_prop_output), getShape(_back_prop_input),
+                              getBuffer<T>(_back_prop_input));
+}
+
+void PadLayer::configure(const IPortableTensor *input, IPortableTensor *output,
+                         const int32_t *padData, int32_t padRank, const void *constantValueData,
+                         IPortableTensor *back_prop_input, const IPortableTensor *back_prop_output)
+{
+  _input = input;
+  _output = output;
+  memcpy(_padData, padData, sizeof(_padData));
+  _padRank = padRank;
+  _constantValueData.v = constantValueData;
+  _back_prop_input = back_prop_input;
+  _back_prop_output = back_prop_output;
+}
+
+void PadLayer::forward(bool)
+{
+  switch (_input->data_type())
+  {
+    case OperandType::FLOAT32:
+      padImpl<float>(_constantValueData.f);
+      break;
+    case OperandType::QUANT_UINT8_ASYMM:
+      if (_constantValueData.u8 == nullptr)
+      {
+        uint8_t pad_value = static_cast<uint8_t>(_output->data_zero_point());
+        padImpl<uint8_t>(&pad_value);
+      }
+      else
+      {
+        padImpl<uint8_t>(_constantValueData.u8);
+      }
+      break;
+    case OperandType::QUANT_INT8_ASYMM:
+      if (_constantValueData.i8 == nullptr)
+      {
+        int8_t pad_value = static_cast<int8_t>(_output->data_zero_point());
+        padImpl<int8_t>(&pad_value);
+      }
+      else
+      {
+        padImpl<int8_t>(_constantValueData.i8);
+      }
+      break;
+    default:
+      throw std::runtime_error{"Pad: unsupported data type"};
+  }
+}
+
+void PadLayer::backward()
+{
+  switch (_back_prop_output->data_type())
+  {
+    case OperandType::FLOAT32:
+      depad<float>();
+      break;
+    case OperandType::QUANT_UINT8_ASYMM:
+      depad<uint8_t>();
+      break;
+    case OperandType::QUANT_INT8_ASYMM:
+      depad<int8_t>();
+      break;
+    default:
+      throw std::runtime_error{"Pad: unsupported data type"};
+  }
+}
+
+} // namespace ops
+} // namespace train
+} // namespace backend
+} // namespace onert

runtime/onert/backend/train/ops/PadLayer.h

@@ -0,0 +1,68 @@
+/*
+ * Copyright (c) 2024 Samsung Electronics Co., Ltd. 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.
+ */
+
+#ifndef __ONERT_BACKEND_TRAIN_OPS_PADLAYER_H__
+#define __ONERT_BACKEND_TRAIN_OPS_PADLAYER_H__
+
+#include <backend/IPortableTensor.h>
+#include "OperationUtils.h"
+
+#include <exec/train/ITrainableFunction.h>
+
+namespace onert
+{
+namespace backend
+{
+namespace train
+{
+namespace ops
+{
+
+// Note, this is pad with mode=`CONSTANT`: it doesn't support `REFLECT` and
+// `SYMMETRIC`
+class PadLayer : public ::onert::exec::train::ITrainableFunction
+{
+public:
+  PadLayer();
+
+public:
+  template <typename T> void padImpl(const T *constant_value_data);
+  template <typename T> void depad();
+
+  void configure(const IPortableTensor *input, IPortableTensor *output, const int32_t *padData,
+                 int32_t padRank, const void *constantValueData, IPortableTensor *back_prop_input,
+                 const IPortableTensor *back_prop_output);
+  void forward(bool training) override;
+  void backward() override;
+
+private:
+  const IPortableTensor *_input;
+  IPortableTensor *_output;
+
+  int32_t _padData[8];
+  int32_t _padRank;
+  ConstDataPtr _constantValueData;
+
+  IPortableTensor *_back_prop_input;
+  const IPortableTensor *_back_prop_output;
+};
+
+} // namespace ops
+} // namespace train
+} // namespace backend
+} // namespace onert
+
+#endif // __ONERT_BACKEND_TRAIN_OPS_PADLAYER_H__