CHERRY_PICK: 2109, 2008, 2098, 2102 (#2111)

cherry pick 2109, 2008, 2098, 2102

lite/api/model_test.cc

@@ -40,6 +40,7 @@ void OutputOptModel(const std::string& load_model_dir,
   config.set_valid_places({
       Place{TARGET(kX86), PRECISION(kFloat)},
       Place{TARGET(kARM), PRECISION(kFloat)},
+      Place{TARGET(kHost), PRECISION(kFloat)},
   });
   auto predictor = lite_api::CreatePaddlePredictor(config);
 

lite/core/mir/fusion/conv_elementwise_fuser.cc

@@ -27,8 +27,10 @@ void ConvElementwiseFuser::BuildPattern() {
       VarNode("input")->assert_is_op_input(conv_type_, "Input")->AsInput();
   auto* filter =
       VarNode("filter")->assert_is_op_input(conv_type_, "Filter")->AsInput();
-  auto* bias =
-      VarNode("bias")->assert_is_op_input("elementwise_add", "Y")->AsInput();
+  auto* bias = VarNode("bias")
+                   ->assert_is_op_input("elementwise_add", "Y")
+                   ->AsInput()
+                   ->assert_is_persistable_var();
 
   // create op nodes
   auto* conv2d =

lite/core/mir/memory_optimize_pass.cc

@@ -49,6 +49,7 @@ void MemoryOptimizePass::CollectLifeCycleByDevice(
                                         "equal",
                                         "lod_reset",
                                         "concat",
+                                        "yolo_box",
                                         "graph_op",
                                         "feed",
                                         "fetch"};

lite/kernels/arm/norm_compute.cc

@@ -47,4 +47,5 @@ REGISTER_LITE_KERNEL(
     norm, kARM, kFloat, kNCHW, paddle::lite::kernels::arm::NormCompute, def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM))})
+    .BindOutput("Norm", {LiteType::GetTensorTy(TARGET(kARM))})
     .Finalize();

lite/operators/op_params.h

@@ -600,6 +600,7 @@ struct SequenceSoftmaxParam {
 struct NormParam {
   const lite::Tensor* X{};
   lite::Tensor* Out{};
+  lite::Tensor* Norm{};
   int axis{1};
   float epsilon{1e-10};
 };

lite/tests/kernels/norm_compute_test.cc

@@ -46,7 +46,7 @@ class NormComputeTester : public arena::TestCase {
 
     auto* x = scope->FindTensor(input_);
     const auto* x_data = x->data<float>();
-    int axis = axis_ < 0 ? axis + dims_.size() : axis_;
+    int axis = axis_ < 0 ? axis_ + dims_.size() : axis_;
     int pre_n = dims_.count(0, axis);
     int n = dims_[axis];
     int post_n = dims_.count(axis + 1, dims_.size());

lite/tools/build.sh

@@ -221,7 +221,7 @@ function print_usage {
     echo -e "argument choices:"
     echo -e "--arm_os:\t android|ios|ios64"
     echo -e "--arm_abi:\t armv8|armv7"
-    echo -e "--arm_lang:\t gcc|clang (for android)"
+    echo -e "--arm_lang:\t only support gcc now, clang will be supported in future.(for android)"
     echo -e "--android_stl:\t c++_static|c++_shared (for android)"
     echo
     echo -e "tasks:"
@@ -252,6 +252,13 @@ function main {
                 ;;
             --arm_lang=*)
                 ARM_LANG="${i#*=}"
+                if [ ${ARM_LANG} == "clang" ]; then
+                     set +x
+                     echo
+                     echo -e "error: only support gcc now, clang will be supported in future."
+                     echo
+                     exit 1
+                fi
                 shift
                 ;;
             --android_stl=*)