diff --git a/src/pass/reduction_factor.cc b/src/pass/reduction_factor.cc
index 54476a76b15bbe320b70d8c2688da6533bc4c334..b8356bb81fc9bdef85e2431b02517dcbdc53eba0 100644
--- a/src/pass/reduction_factor.cc
+++ b/src/pass/reduction_factor.cc
@@ -138,7 +138,13 @@ class IdentifyReduceChance : public IRVisitor {
           reduce_index_++;
         }
       }
-      IRVisitor::Visit_(op);
+      const Block *block_ptr = op->body.as<Block>();
+      if (block_ptr && is_reduce_y_) {
+        IRVisitor::Visit(block_ptr->first);
+      } else {
+        IRVisitor::Visit_(op);
+      }
+
       if (cur_reduce_data_->outter_reduction_data) {
         auto inner_reduction_data = cur_reduce_data_;
         cur_reduce_data_ = cur_reduce_data_->outter_reduction_data;
@@ -187,7 +193,11 @@ class IdentifyReduceChance : public IRVisitor {
           break;
         case air::ir::ForType::Parallel:
           is_parallel_area_ = true;
-          cur_reduce_data_->vector_parallel_for = op;
+          if (is_reduce_y_) {
+            cur_reduce_data_->parallel_for = op;
+          } else {
+            cur_reduce_data_->vector_parallel_for = op;
+          }
           IRVisitor::Visit_(op);
           is_parallel_area_ = false;
           return;
@@ -260,6 +270,26 @@ class IdentifyReduceChance : public IRVisitor {
             Array<Expr> output_args = op->args;
             Expr input_expr = GetTheInputExpr(op);
             auto call_ptr = input_expr.as<Call>();
+            while(!call_ptr) {
+              if (auto min = input_expr.as<Min>()) {
+                input_expr = min->b;
+              } else if (auto max = input_expr.as<Max>()) {
+                input_expr = max->b;
+              } else if (auto and_op = input_expr.as<And>()) {
+                input_expr = and_op->b;
+              } else if (auto or_op = input_expr.as<Or>()) {
+                input_expr = or_op->b;
+              } else if (auto add_op = input_expr.as<Add>()) {
+                input_expr = add_op->b;
+              } else if (auto mul_op = input_expr.as<Mul>()) {
+                input_expr = mul_op->b;
+              } else if (auto div_op = input_expr.as<Div>()) {
+                input_expr = div_op->b;
+              } else {
+                CHECK(false) << "reduce type is invalid";
+              }
+              call_ptr = input_expr.as<Call>();
+            }
             CHECK(call_ptr) << "call_ptr is nullptr";
             Array<Expr> input_args = call_ptr->args;
             while(call_ptr) {
@@ -445,7 +475,8 @@ class MutateReduceBody : public IRMutator {
                            cur_reduce_data_->reduce_temp.as<Call>()->args);
     } else if (cur_reduce_data_->isolate_reduce_provide.count(op) &&
                (cur_reduce_data_->outter_reduction_data != nullptr ||
-                cur_reduce_data_->pre_reduction_data != nullptr)) {
+                cur_reduce_data_->pre_reduction_data != nullptr ||
+                cur_reduce_data_->is_reduce_y)) {
       auto value = op->value;
       auto temp_call = MakeUniqueTempCallForIsolate(cur_reduce_data_->reduce_temp, op);
       auto input_call = GetTheInputExpr(op);
@@ -540,10 +571,6 @@ class ReduceVectorizeEnable : public IRMutator {
         return body;
       }
 
-      if (cur_reduce_data_->is_reduce_y && reduce_index_ == INNER_REDUCTION_DATA_INDEX) {
-        return MutateReduceBody(cur_reduce_data_).Mutate(body);
-      }
-
       CHECK(cur_reduce_data_->body.defined()) << "cur_reduce_data_ body is not defined";
       Stmt stmt = MakeReduceStmt();
       stmt = VectorizedForAmend(cur_reduce_data_).Mutate(stmt);
@@ -565,13 +592,6 @@ class ReduceVectorizeEnable : public IRMutator {
       } else if (current_level_entry->pre_reduction_data) {
         cur_reduce_data_ = current_level_entry->pre_reduction_data;
       }
-      if (cur_reduce_data_->is_reduce_y) {
-        auto inner_reduce_data = reduce_datas_[reduce_index_];
-        inner_reduce_data->reduce_temp = cur_reduce_data_->reduce_temp;
-        inner_reduce_data->temp_tensor = cur_reduce_data_->temp_tensor;
-        inner_reduce_data->temp_buffer = cur_reduce_data_->temp_buffer;
-        return IRMutator::Mutate(stmt);
-      }
       return stmt;
     } else if (op->attr_key == REDUCE_Y_FLAG) {
       return IRMutator::Mutate(op->body);
@@ -592,6 +612,7 @@ class ReduceVectorizeEnable : public IRMutator {
       } else {
         CHECK(false) << "can not get output_shape";
       }
+      CHECK(reduce_data->parallel_for) << "parallel_for is nullptr";
       shapes.push_back(reduce_data->parallel_for->extent);
       for (size_t i = 0; i < output_shape.size(); i++) {
         if (reduce_data->reduce_axis_hidden || i != reduce_data->reduce_axis) {
@@ -780,9 +801,8 @@ class ReduceVectorizeEnable : public IRMutator {
     }
     // step 3: vectorize or parallel area
     Stmt reduce_body = cur_reduce_data_->body;
-    if (!cur_reduce_data_->is_reduce_y || reduce_index_ == 0) {
-      reduce_body = MutateReduceBody(cur_reduce_data_).Mutate(cur_reduce_data_->body);
-    }
+    reduce_body = MutateReduceBody(cur_reduce_data_).Mutate(reduce_body);
+    
     // step 4: init area
     const For *for_ptr = cur_reduce_data_->is_reduce_y ? cur_reduce_data_->parallel_for : vector_parallel_for;
     loop_var = Variable::make(for_ptr->loop_var->type,
diff --git a/src/poly/schedule_pass/tile_outer_band.cc b/src/poly/schedule_pass/tile_outer_band.cc
index 34b7d1d74c1b7594712112c4f1f76468c16f769a..27b3e57798b004552c38a84337da05292cfcdfee 100644
--- a/src/poly/schedule_pass/tile_outer_band.cc
+++ b/src/poly/schedule_pass/tile_outer_band.cc
@@ -1355,8 +1355,6 @@ isl::schedule_node TileOuterBand::InsertMarkerForReduceY(const isl::schedule_nod
     node = InsertMarkerForLoop(node, FOR_PARALLEL);
     bool is_parallel = !GetMarkerName(node, FOR_PARALLEL).empty();
     if (is_parallel) {
-      node = node.child(0).child(0).child(0).child(0);
-      node = node.insert_mark(REDUCE_AREA_FLAG);
       node = node.ancestor(node.get_tree_depth() - start_depth);
       node = node.insert_mark(REDUCE_AREA_FLAG);
       node = node.insert_mark(REDUCE_Y_FLAG);