1 files changed, 109 insertions, 0 deletions
diff --git a/compiler/oneco/src/Onnxutil.cpp b/compiler/oneco/src/Onnxutil.cpp
new file mode 100644
index 000000000..93f06677f
--- /dev/null
+++ b/compiler/oneco/src/Onnxutil.cpp
@@ -0,0 +1,109 @@
+/*
+ * Copyright (c) 2019 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 "Convert.h"
+
+#include <cassert>
+#include <stdexcept>
+
+namespace
+{
+
+/**
+ * @note If the platform is little endian, 0x00112233 would be saved as [0x33, 0x22, 0x11, 0x00]
+ *       If not, it would be saved as [0x00, 0x11, 0x22, 0x33]
+ * @return Whether platform is little endian or not
+ */
+bool is_platform_little_endian()
+{
+  int32_t num = 0x00112233;
+  return (*(char *)&num == 0x33);
+}
+
+} // namespace
+
+namespace moco
+{
+namespace onnx
+{
+
+bool is_default_domain(const std::string domain)
+{
+  return (domain.compare("") == 0 || domain.compare("onnx.ai") == 0);
+}
+
+std::vector<float> get_float_data(const ::onnx::TensorProto &tensor)
+{
+  std::vector<float> data;
+
+  // Exactly one of the fields is used to store the elements of the tensor
+  assert(!(tensor.has_raw_data() && (tensor.float_data_size() > 0)));
+  assert(tensor.has_raw_data() || (tensor.float_data_size() > 0));
+
+  if (tensor.has_raw_data())
+  {
+    const std::string raw_data = tensor.raw_data();
+
+    // If platform is big endian, we should convert data as big endian
+    if (!is_platform_little_endian())
+    {
+      // TODO Revise implementation of this logic. This is too complex.
+      const char *little_endian_bytes = raw_data.c_str();
+      char *big_endian_bytes = reinterpret_cast<char *>(std::malloc(raw_data.size()));
+
+      for (int i = 0; i < raw_data.size(); ++i)
+        big_endian_bytes[i] = little_endian_bytes[i];
+
+      const size_t element_size = sizeof(float);
+      const size_t num_elements = raw_data.size() / element_size;
+      for (size_t i = 0; i < num_elements; ++i)
+      {
+        char *start_byte = big_endian_bytes + i * element_size;
+        char *end_byte = start_byte + element_size - 1;
+
+        for (size_t count = 0; count < element_size / 2; ++count)
+        {
+          char temp = *start_byte;
+          *start_byte = *end_byte;
+          *end_byte = temp;
+          ++start_byte;
+          --end_byte;
+        }
+      }
+
+      data.insert(data.end(), reinterpret_cast<const float *>(big_endian_bytes),
+                  reinterpret_cast<const float *>(big_endian_bytes + raw_data.size()));
+
+      std::free(big_endian_bytes);
+    }
+    else
+    {
+      const char *bytes = raw_data.c_str();
+      data.insert(data.end(), reinterpret_cast<const float *>(bytes),
+                  reinterpret_cast<const float *>(bytes + raw_data.size()));
+    }
+  }
+  else
+  {
+    for (int i = 0; i < tensor.float_data_size(); ++i)
+      data.push_back(tensor.float_data(i));
+  }
+
+  return data;
+}
+
+} // namespace onnx
+} // namespace moco