diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/arithmetic_self_cpu_kernel.cc b/mindspore/ccsrc/backend/kernel_compiler/cpu/arithmetic_self_cpu_kernel.cc
index a73682a888..a654effa4c 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/arithmetic_self_cpu_kernel.cc
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/arithmetic_self_cpu_kernel.cc
@@ -267,8 +267,10 @@ bool ArithmeticSelfCPUKernel::Launch(const std::vector<kernel::AddressPtr> &inpu
                                      const std::vector<kernel::AddressPtr> &outputs) {
   if (dtype_ == kNumberTypeFloat32 || dtype_ == kNumberTypeFloat16 || dtype_ == kNumberTypeFloat64) {
     LaunchKernel<float>(inputs, outputs);
-  } else if (dtype_ == kNumberTypeInt32 || dtype_ == kNumberTypeInt16 || dtype_ == kNumberTypeInt64) {
+  } else if (dtype_ == kNumberTypeInt32 || dtype_ == kNumberTypeInt16) {
     LaunchKernel<int>(inputs, outputs);
+  } else if (dtype_ == kNumberTypeInt64) {
+    LaunchKernel<int64_t>(inputs, outputs);
   } else if (dtype_ == kNumberTypeBool) {
     LaunchKernelLogic<bool>(inputs, outputs);
   } else {
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/cpu_kernel.cc b/mindspore/ccsrc/backend/kernel_compiler/cpu/cpu_kernel.cc
index 5d5d8b9de2..98af09a999 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/cpu_kernel.cc
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/cpu_kernel.cc
@@ -29,6 +29,7 @@ void CPUKernel::InitInputOutputSize(const CNodePtr &kernel_node) {
     std::vector<size_t> shape = AnfAlgo::GetInputDeviceShape(kernel_node, input_index);
     size_t tensor_size =
       shape.empty() ? type_size : std::accumulate(shape.begin(), shape.end(), type_size, std::multiplies<size_t>());
+    tensor_size = std::max(tensor_size, type_size);
     input_size_list_.emplace_back(tensor_size);
   }
   size_t output_num = AnfAlgo::GetOutputTensorNum(kernel_node);
@@ -38,6 +39,7 @@ void CPUKernel::InitInputOutputSize(const CNodePtr &kernel_node) {
     std::vector<size_t> shape = AnfAlgo::GetOutputDeviceShape(kernel_node, output_index);
     size_t tensor_size =
       shape.empty() ? type_size : std::accumulate(shape.begin(), shape.end(), type_size, std::multiplies<size_t>());
+    tensor_size = std::max(tensor_size, type_size);
     output_size_list_.emplace_back(tensor_size);
   }
 }
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/topk_cpu_kernel.cc b/mindspore/ccsrc/backend/kernel_compiler/cpu/topk_cpu_kernel.cc
index 01f29b01a9..cbab0fc837 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/topk_cpu_kernel.cc
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/topk_cpu_kernel.cc
@@ -50,11 +50,11 @@ void TopKCPUKernel::LaunchKernel(const std::vector<AddressPtr> &inputs, const st
     std::vector<size_t> idx(inner_size_);
     auto base_input = i * inner_size_;
     std::iota(idx.begin(), idx.end(), base_input);
-    std::sort(idx.begin(), idx.end(),
-              [&input](size_t index_1, size_t index_2) { return input[index_1] > input[index_2]; });
+    std::stable_sort(idx.begin(), idx.end(),
+                     [&input](size_t index_1, size_t index_2) { return input[index_1] > input[index_2]; });
     auto base_output = i * k_num;
     if (!sorted_) {
-      std::sort(idx.begin(), idx.begin() + k_num);
+      std::stable_sort(idx.begin(), idx.begin() + k_num);
     }
     for (int j = 0; j < k_num; ++j) {
       indices[base_output + j] = idx[j] - base_input;