Fix misaligned addresses while reading tensor attributes from raw data buffers

Author: n-v-k

onnxruntime/core/optimizer/common_subexpression_elimination.cc

@@ -186,11 +186,26 @@ bool AreScalarTensorAttributeEqual(const ONNX_NAMESPACE::TensorProto& lhs_t, con
   const void* rhs_value = rhs_t.raw_data().data();
   switch (lhs_t.data_type()) {
     case onnx::TensorProto_DataType_FLOAT:
-      return *reinterpret_cast<const float*>(lhs_value) == *reinterpret_cast<const float*>(rhs_value);
+    {
+      float lhs_float_value, rhs_float_value;
+      std::memcpy(&lhs_float_value, lhs_value, sizeof(lhs_float_value));
+      std::memcpy(&rhs_float_value, rhs_value, sizeof(rhs_float_value));
+      return lhs_float_value == rhs_float_value;
+    }
     case onnx::TensorProto_DataType_FLOAT16:
-      return *reinterpret_cast<const MLFloat16*>(lhs_value) == *reinterpret_cast<const MLFloat16*>(rhs_value);
+    {
+      MLFloat16 lhs_float16_value, rhs_float16_value;
+      std::memcpy(&lhs_float16_value, lhs_value, sizeof(lhs_float16_value));
+      std::memcpy(&rhs_float16_value, rhs_value, sizeof(rhs_float16_value));
+      return lhs_float16_value == rhs_float16_value;
+    }
     case onnx::TensorProto_DataType_INT64:
-      return *reinterpret_cast<const int64_t*>(lhs_value) == *reinterpret_cast<const int64_t*>(rhs_value);
+    {
+      int64_t lhs_int64_value, rhs_int64_value;
+      std::memcpy(&lhs_int64_value, lhs_value, sizeof(lhs_int64_value));
+      std::memcpy(&rhs_int64_value, rhs_value, sizeof(rhs_int64_value));
+      return lhs_int64_value == rhs_int64_value;
+    }
     default:
       break;
   }
@@ -240,19 +255,32 @@ std::size_t GetTensorAttributeHash(const ONNX_NAMESPACE::TensorProto& attr_t) {
   std::size_t hash = 0;
   if (utils::HasDataType(attr_t) && attr_t.dims_size() == 1 && attr_t.dims()[0] == 1 && utils::HasRawData(attr_t)) {
     int data_type = attr_t.data_type();
+    const char* value = attr_t.raw_data().data();
     switch (data_type) {
       case onnx::TensorProto_DataType_FLOAT:
+      {
+        float float_value;
+        std::memcpy(&float_value, value, sizeof(float_value));
         UpdateHash(data_type, hash);
-        UpdateHash(*reinterpret_cast<const float*>(attr_t.raw_data().data()), hash);
+        UpdateHash(float_value, hash);
         break;
+      }
       case onnx::TensorProto_DataType_FLOAT16:
+      {
+        MLFloat16 float16_value;
+        std::memcpy(&float16_value, value, sizeof(float16_value));
         UpdateHash(data_type, hash);
-        UpdateHash(static_cast<float>(*reinterpret_cast<const MLFloat16*>(attr_t.raw_data().data())), hash);
+        UpdateHash(static_cast<float>(float16_value), hash);
         break;
+      }
       case onnx::TensorProto_DataType_INT64:
+      {
+        int64_t int64_value;
+        std::memcpy(&int64_value, value, sizeof(int64_value));
         UpdateHash(data_type, hash);
-        UpdateHash(*reinterpret_cast<const int64_t*>(attr_t.raw_data().data()), hash);
+        UpdateHash(int64_value, hash);
         break;
+      }
       default:
         break;
     }