dont release

core/predictor/framework/infer.h

@@ -501,7 +501,6 @@ class FluidInferEngine : public CloneDBReloadableInferEngine<EngineCore> {
       paddle::PaddleBuf paddleBuf(databuf_char, databuf_size);
       tensor_out.data = paddleBuf;
       tensorVector_out_pointer->push_back(tensor_out);
-      MempoolWrapper::instance().free(databuf_char, databuf_size);
     }
     return 0;
   }

core/predictor/framework/memory.cpp

@@ -126,7 +126,7 @@ void MempoolWrapper::free(void* p, size_t size) {
                  << ", since mempool is not thread initialized";
     return;
   }
-  return mempool->free(p,size);
+  return mempool->free(p, size);
 }
 
 }  // namespace predictor

core/predictor/mempool/mempool.cpp

@@ -24,7 +24,7 @@ namespace fugue {
 namespace memory {
 
 void Region::init() {
-  _big_mem_capacity = 64 * 1024 * 1024;//64MB
+  _big_mem_capacity = 64 * 1024 * 1024;  // 64MB
   _big_mem_start = new char[_big_mem_capacity];
 }
 

core/predictor/mempool/mempool.h

@@ -341,8 +341,9 @@ class Region {
       2 * 1024 *
       1024;  // 2MB,means when you need less than 2M, get memory from Block.
 
-  // 64MB,means when you need less than 64MB, get memory from BigMemory instead of BigNode
-  static const int BIGNODE_MEM_THRESHOLD = (64 * 1024 * 1024 + 1);  
+  // 64MB,means when you need less than 64MB, get memory from BigMemory instead
+  // of BigNode
+  static const int BIGNODE_MEM_THRESHOLD = (64 * 1024 * 1024 + 1);
   static const int COUNTER_SIZE =
       BIGNODE_MEM_THRESHOLD / BIG_MEM_THRESHOLD + 1;  // this is not used