diff --git a/tools/ncnnoptimize.cpp b/tools/ncnnoptimize.cpp
index 3c6eadbb3..8242607dc 100644
--- a/tools/ncnnoptimize.cpp
+++ b/tools/ncnnoptimize.cpp
@@ -98,7 +98,7 @@ public:
     int fprintf_param_int_array(int id, const ncnn::Mat& m, FILE* pp);
     int fprintf_param_float_array(int id, const ncnn::Mat& m, FILE* pp);
 
-    int fwrite_weight_tag(int tag, FILE* bp);
+    int fwrite_weight_tag_data(int tag, const ncnn::Mat& data, FILE* bp);
     int fwrite_weight_data(const ncnn::Mat& data, FILE* bp);
 
     int save(const char* parampath, const char* binpath);
@@ -1158,16 +1158,21 @@ int NetOptimize::fprintf_param_float_array(int id, const ncnn::Mat& m, FILE* pp)
     return 0;
 }
 
-int NetOptimize::fwrite_weight_tag(int tag, FILE* bp)
+int NetOptimize::fwrite_weight_tag_data(int tag, const ncnn::Mat& data, FILE* bp)
 {
+    ncnn::Mat data_flattened = data.reshape(data.w * data.h * data.c);
     if (storage_type == 1 && tag == 0)
     {
         tag = 0x01306B47; // fp16 magic
         fwrite(&tag, sizeof(int), 1, bp);
+        ncnn::Mat data_flattened_fp16;
+        ncnn::cast_float32_to_float16(data_flattened, data_flattened_fp16);
+        fwrite(data_flattened_fp16.data, data_flattened_fp16.elemsize, data_flattened_fp16.w, bp);
     }
     else
     {
         fwrite(&tag, sizeof(int), 1, bp);
+        fwrite(data_flattened.data, data_flattened.elemsize, data_flattened.w, bp);
     }
     return 0;
 }
@@ -1175,16 +1180,7 @@ int NetOptimize::fwrite_weight_tag(int tag, FILE* bp)
 int NetOptimize::fwrite_weight_data(const ncnn::Mat& data, FILE* bp)
 {
     ncnn::Mat data_flattened = data.reshape(data.w * data.h * data.c);
-    if (storage_type == 1)
-    {
-        ncnn::Mat data_flattened_fp16;
-        ncnn::cast_float32_to_float16(data_flattened, data_flattened_fp16);
-        fwrite(data_flattened_fp16.data, data_flattened_fp16.elemsize, data_flattened_fp16.w, bp);
-    }
-    else
-    {
-        fwrite(data_flattened.data, data_flattened.elemsize, data_flattened.w, bp);
-    }
+    fwrite(data_flattened.data, data_flattened.elemsize, data_flattened.w, bp);
     return 0;
 }
 
@@ -1322,8 +1318,7 @@ int NetOptimize::save(const char* parampath, const char* binpath)
             fprintf_param_value(" 9=%d", activation_type)
             { if (!op->activation_params.empty()) fprintf_param_int_array(10, op->activation_params, pp); }
 
-            fwrite_weight_tag(0, bp);
-            fwrite_weight_data(op->weight_data, bp);
+            fwrite_weight_tag_data(0, op->weight_data, bp);
             fwrite_weight_data(op->bias_data, bp);
         }
         else if (layer->type == "ConvolutionDepthWise")
@@ -1347,8 +1342,7 @@ int NetOptimize::save(const char* parampath, const char* binpath)
             fprintf_param_value(" 9=%d", activation_type)
             { if (!op->activation_params.empty()) fprintf_param_int_array(10, op->activation_params, pp); }
 
-            fwrite_weight_tag(0, bp);
-            fwrite_weight_data(op->weight_data, bp);
+            fwrite_weight_tag_data(0, op->weight_data, bp);
             fwrite_weight_data(op->bias_data, bp);
         }
         else if (layer->type == "Crop")
@@ -1382,8 +1376,7 @@ int NetOptimize::save(const char* parampath, const char* binpath)
             fprintf_param_value(" 9=%d", activation_type)
             { if (!op->activation_params.empty()) fprintf_param_int_array(10, op->activation_params, pp); }
 
-            fwrite_weight_tag(0, bp);
-            fwrite_weight_data(op->weight_data, bp);
+            fwrite_weight_tag_data(0, op->weight_data, bp);
             fwrite_weight_data(op->bias_data, bp);
         }
         else if (layer->type == "DeconvolutionDepthWise")
@@ -1406,8 +1399,7 @@ int NetOptimize::save(const char* parampath, const char* binpath)
             fprintf_param_value(" 9=%d", activation_type)
             { if (!op->activation_params.empty()) fprintf_param_int_array(10, op->activation_params, pp); }
 
-            fwrite_weight_tag(0, bp);
-            fwrite_weight_data(op->weight_data, bp);
+            fwrite_weight_tag_data(0, op->weight_data, bp);
             fwrite_weight_data(op->bias_data, bp);
         }
         else if (layer->type == "DetectionOutput")
@@ -1468,8 +1460,7 @@ int NetOptimize::save(const char* parampath, const char* binpath)
             fprintf_param_value(" 9=%d", activation_type)
             { if (!op->activation_params.empty()) fprintf_param_int_array(10, op->activation_params, pp); }
 
-            fwrite_weight_tag(0, bp);
-            fwrite_weight_data(op->weight_data, bp);
+            fwrite_weight_tag_data(0, op->weight_data, bp);
             fwrite_weight_data(op->bias_data, bp);
         }
         else if (layer->type == "Input")