1 files changed, 171 insertions, 0 deletions
diff --git a/compiler/nnc/backends/interpreter/InterpreterBackend.cpp b/compiler/nnc/backends/interpreter/InterpreterBackend.cpp
new file mode 100644
index 000000000..923a7cfc7
--- /dev/null
+++ b/compiler/nnc/backends/interpreter/InterpreterBackend.cpp
@@ -0,0 +1,171 @@
+/*
+ * Copyright (c) 2018 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 <cstring>
+#include <utility>
+#include <vector>
+#include <fstream>
+#include <sstream>
+#include <algorithm>
+
+#include "Definitions.h"
+
+#ifdef NNC_HDF5_SUPPORTED
+#include <H5Cpp.h>
+#else
+#include <iostream>
+#endif // NNC_HDF5_SUPPORTED
+
+#include "mir/Shape.h"
+
+#include "MirInterpreter.h"
+#include "backends/interpreter/InterpreterBackend.h"
+
+#include "mir/Graph.h"
+
+#include "mir/ShapeRange.h"
+#include "mir/Tensor.h"
+
+#include <stdexcept>
+
+namespace nnc
+{
+
+using namespace mir;
+
+#ifdef NNC_HDF5_SUPPORTED
+
+/**
+ * @brief save tensor in file in '.hdf5' format
+ * @param tensor - tensor to save
+ * @param tensor_name - name, by wich tensor will be saved
+ * @param destination - path to file, in which tensor will be saved
+ */
+static void writeTensorToHDF5File(const TensorVariant &tensor, std::string tensor_name,
+                                  const std::string &destination)
+{
+
+  // Prepare shape, rank, dims, numElems
+  auto &shape = tensor.getShape();
+  const int32_t rank = shape.rank();
+  hsize_t dims[rank];
+  for (int32_t axis = 0; axis < rank; ++axis)
+  {
+    dims[axis] = static_cast<hsize_t>(shape.dim(axis));
+  }
+
+  // Create array from tensor
+  std::vector<char> values;
+  const auto elem_size = tensor.getElementSize();
+  values.resize(elem_size * shape.numElements());
+  char *values_ptr = values.data();
+  ShapeRange out_range(shape);
+  for (auto &out_idx : out_range)
+  {
+    std::memcpy(values_ptr, tensor.at(out_idx), elem_size);
+    values_ptr += elem_size;
+  }
+
+  // Backslashes are not allowed in tensor names
+  std::replace(tensor_name.begin(), tensor_name.end(), '/', '_');
+  std::string filename = destination + "/" + tensor_name + ".hdf5";
+
+  // Write to .hdf5 file
+  H5::H5File h5File(filename, H5F_ACC_TRUNC);
+  H5::DataSpace dataspace(rank, dims);
+  H5::DataType h5_data_type;
+
+  if (tensor.getDataType() == DataType::FLOAT32)
+    h5_data_type = H5::PredType::NATIVE_FLOAT;
+  else if (tensor.getDataType() == DataType::UINT8)
+    h5_data_type = H5::PredType::NATIVE_UINT8;
+  else
+    throw std::runtime_error("NYI writing that DataType!");
+
+  auto dataset = h5File.createDataSet(tensor_name, h5_data_type, dataspace);
+  dataset.write(values.data(), h5_data_type);
+}
+
+#endif // NNC_HDF5_SUPPORTED
+
+static TensorVariant readTensorFromFile(const std::string &filename, const TensorType &type)
+{
+  const std::size_t input_data_size =
+      type.getShape().numElements() * getDataTypeSize(type.getElementType());
+
+  std::ifstream stream(filename, std::ios::in | std::ios::binary);
+  if (stream.fail())
+    throw std::runtime_error("Couldn't open file \"" + filename + "\".");
+
+  stream.seekg(0, std::ios::end);
+  std::streampos end = stream.tellg();
+  stream.seekg(0, std::ios::beg);
+  std::streampos begin = stream.tellg();
+  int64_t file_size = end - begin;
+
+  if (static_cast<std::size_t>(file_size) != input_data_size)
+    throw std::runtime_error("File \"" + filename + "\" has incorrect size: " +
+                             std::to_string(file_size) + "(expected: " +
+                             std::to_string(input_data_size) + ").");
+
+  std::unique_ptr<char[]> data(new char[input_data_size]);
+  stream.read(data.get(), input_data_size);
+  if (stream.fail())
+    throw std::runtime_error("Couldn't read file \"" + filename + "\".");
+
+  return TensorVariant(type, data.get());
+}
+
+InterpreterBackend::InterpreterBackend(std::string input_dir, std::string output_dir)
+    : _input_dir(std::move(input_dir)), _output_dir(std::move(output_dir))
+{
+}
+
+void InterpreterBackend::run(mir::Graph *graph)
+{
+  assert(graph);
+
+  mir_interpreter::MIRInterpreter interpreter;
+
+  for (const auto *input_op : graph->getInputs())
+  {
+    const Operation::Output *input = input_op->getOutput(0);
+
+    std::string tensor_name = input->getName();
+    assert(!tensor_name.empty());
+    std::replace(tensor_name.begin(), tensor_name.end(), '/', '_');
+    std::string filename = _input_dir + "/" + tensor_name + ".dat";
+
+    TensorVariant tensor = readTensorFromFile(filename, input->getType());
+    interpreter.setTensor(input, std::move(tensor));
+  }
+
+  graph->accept(&interpreter);
+
+  for (const auto *output_op : graph->getOutputs())
+  {
+    const auto &output_name = output_op->getInput(0)->getName();
+
+#ifdef NNC_HDF5_SUPPORTED
+    const auto &tensor = interpreter.getTensor(output_op->getInput(0));
+    writeTensorToHDF5File(tensor, output_name, _output_dir);
+#else
+    std::cout << "Result <" << output_name << "> wasn't saved, due to lack of HDF5" << std::endl;
+#endif // NNC_HDF5_SUPPORTED
+  }
+}
+
+} // namespace nnc