refactor(cast): use mlu cast as default

lite/core/mir/mlu_postprocess_pass.cc

@@ -888,14 +888,14 @@ void MLUPostprocessPass::Apply(const std::unique_ptr<SSAGraph>& graph) {
 #endif
 
   g_stream_id = static_cast<int>(reinterpret_cast<int64_t>(graph.get()));
-  bool use_mlu_cast = GetBoolFromEnv("LITE_MLU_CAST");
-  ModifyValidPlaces(graph.get(), use_mlu_cast);
+  bool disable_mlu_cast = GetBoolFromEnv("LITE_DISABLE_MLU_CAST");
+  ModifyValidPlaces(graph.get(), !disable_mlu_cast);
   // insert io_copy, layout and precision cast of subgraph's inputs and outputs
   for (auto& node : graph->mutable_nodes()) {
     if (node.IsStmt() && node.AsStmt().op_type() == "subgraph") {
       const Type* subgraph_arg_type = nullptr;
       GetSubgraphOpArgType(&node, &subgraph_arg_type, graph.get());
-      if (use_mlu_cast) {
+      if (!disable_mlu_cast) {
         AdjustSubgraph(&node, subgraph_arg_type);
       }
 
@@ -903,14 +903,14 @@ void MLUPostprocessPass::Apply(const std::unique_ptr<SSAGraph>& graph) {
       for (auto p_in : links_tmp) {
         if (NeedInsert(p_in, subgraph_arg_type)) {
           InsertBefore(
-              graph.get(), p_in, &node, subgraph_arg_type, use_mlu_cast);
+              graph.get(), p_in, &node, subgraph_arg_type, !disable_mlu_cast);
         }
       }
       links_tmp.assign(node.outlinks.begin(), node.outlinks.end());
       for (auto p_out : links_tmp) {
         if (NeedInsert(p_out, subgraph_arg_type)) {
           InsertAfter(
-              graph.get(), p_out, &node, subgraph_arg_type, use_mlu_cast);
+              graph.get(), p_out, &node, subgraph_arg_type, !disable_mlu_cast);
         }
       }
     }

lite/kernels/mlu/subgraph_compute.h

@@ -343,7 +343,7 @@ class SubgraphEngine : public subgraph::Engine {
     CHECK_EQ(graph_input->size(), origin_itensors_.size());
     CHECK_EQ(graph_output->size(), origin_otensors_.size());
 
-    bool use_mlu_cast = GetBoolFromEnv("LITE_MLU_CAST");
+    bool disable_mlu_cast = GetBoolFromEnv("LITE_DISABLE_MLU_CAST");
 
     if (!disable_batch_size_changeable_) {
       std::vector<std::shared_ptr<paddle::lite::subgraph::mlu::MLUTensor>>
@@ -386,7 +386,7 @@ class SubgraphEngine : public subgraph::Engine {
         for (size_t i = 0; i < origin_otensors_.size(); ++i) {
           // origin_otensors_[i]->Resize(new_output_size.at(i));
           graph_out[i]->set_mlu_ptr(
-              GetOutputDataPtr(origin_otensors_[i], use_mlu_cast));
+              GetOutputDataPtr(origin_otensors_[i], !disable_mlu_cast));
         }
       } else {
         graph_out.reserve(origin_otensors_.size());
@@ -395,7 +395,8 @@ class SubgraphEngine : public subgraph::Engine {
           paddle::lite::subgraph::mlu::MLUTensor tmp(
               origin_otensors_[i]->dims().Vectorize());
           tmp.set_mlu_dtype(graph_output->at(i)->dtype());
-          tmp.set_mlu_ptr(GetOutputDataPtr(origin_otensors_[i], use_mlu_cast));
+          tmp.set_mlu_ptr(
+              GetOutputDataPtr(origin_otensors_[i], !disable_mlu_cast));
           graph_out.push_back(
               std::make_shared<paddle::lite::subgraph::mlu::MLUTensor>(tmp));
         }
@@ -410,7 +411,7 @@ class SubgraphEngine : public subgraph::Engine {
       for (size_t i = 0; i < origin_otensors_.size(); ++i) {
         origin_otensors_[i]->Resize(graph_output->at(i)->get_origin_shape());
         graph_output->at(i)->set_mlu_ptr(
-            GetOutputDataPtr(origin_otensors_[i], use_mlu_cast));
+            GetOutputDataPtr(origin_otensors_[i], !disable_mlu_cast));
       }
       graph->Compute(forward_param, exec_queue);
     }