diff --git a/src/layer/batchnorm.cpp b/src/layer/batchnorm.cpp
index 387c11242..636293faf 100644
--- a/src/layer/batchnorm.cpp
+++ b/src/layer/batchnorm.cpp
@@ -60,7 +60,7 @@ int BatchNorm::load_model(const ModelBin& mb)
 
     for (int i=0; i<channels; i++)
     {
-        float sqrt_var = sqrt(var_data[i] + eps);
+        float sqrt_var = static_cast<float>(sqrt(var_data[i] + eps));
         a_data[i] = bias_data[i] - slope_data[i] * mean_data[i] / sqrt_var;
         b_data[i] = slope_data[i] / sqrt_var;
     }
diff --git a/src/layer/normalize.cpp b/src/layer/normalize.cpp
index c84041726..bf34cceaa 100644
--- a/src/layer/normalize.cpp
+++ b/src/layer/normalize.cpp
@@ -86,7 +86,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
         float a;
         if (eps_mode == 0) // caffe/mxnet
         {
-            a = 1.f / sqrt(ssum + eps);
+            a = static_cast<float>(1.f / sqrt(ssum + eps));
         }
         else if (eps_mode == 1) // pytorch
         {
@@ -94,7 +94,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
         }
         else //if (eps_mode == 2) // tensorflow
         {
-            a = 1.f / sqrt(std::max(ssum, eps));
+            a = static_cast<float>(1.f / sqrt(std::max(ssum, eps)));
         }
 
         if (channel_shared)
@@ -146,7 +146,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
             float a;
             if (eps_mode == 0) // caffe/mxnet
             {
-                a = 1.f / sqrt(ssum + eps);
+                a = static_cast<float>(1.f / sqrt(ssum + eps));
             }
             else if (eps_mode == 1) // pytorch
             {
@@ -154,7 +154,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
             }
             else //if (eps_mode == 2) // tensorflow
             {
-                a = 1.f / sqrt(std::max(ssum, eps));
+                a = static_cast<float>(1.f / sqrt(std::max(ssum, eps)));
             }
 
             float scale = a * (channel_shared ? scale_data[0] : scale_data[q]);
@@ -193,7 +193,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
                 float a;
                 if (eps_mode == 0) // caffe/mxnet
                 {
-                    a = 1.f / sqrt(ssum + eps);
+                    a = static_cast<float>(1.f / sqrt(ssum + eps));
                 }
                 else if (eps_mode == 1) // pytorch
                 {
@@ -201,7 +201,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
                 }
                 else //if (eps_mode == 2) // tensorflow
                 {
-                    a = 1.f / sqrt(std::max(ssum, eps));
+                    a = static_cast<float>(1.f / sqrt(std::max(ssum, eps)));
                 }
 
                 square_sum_blob[i] = a * scale;
@@ -233,7 +233,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
                 float a;
                 if (eps_mode == 0) // caffe/mxnet
                 {
-                    a = 1.f / sqrt(ssum + eps);
+                    a = static_cast<float>(1.f / sqrt(ssum + eps));
                 }
                 else if (eps_mode == 1) // pytorch
                 {
@@ -241,7 +241,7 @@ int Normalize::forward_inplace(Mat& bottom_top_blob, const Option& opt) const
                 }
                 else //if (eps_mode == 2) // tensorflow
                 {
-                    a = 1.f / sqrt(std::max(ssum, eps));
+                    a = static_cast<float>(1.f / sqrt(std::max(ssum, eps)));
                 }
 
                 square_sum_blob[i] = a;
diff --git a/src/layer/padding.cpp b/src/layer/padding.cpp
index fc751616e..66e715f14 100644
--- a/src/layer/padding.cpp
+++ b/src/layer/padding.cpp
@@ -298,7 +298,7 @@ int Padding::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) c
             return -100;
 
         if (elemsize == 1)
-            copy_make_border_image<signed char>(bottom_blob, top_blob, 0, left, type, value);
+            copy_make_border_image<signed char>(bottom_blob, top_blob, 0, left, type, static_cast<signed char>(value));
         else if (elemsize == 4)
             copy_make_border_image<float>(bottom_blob, top_blob, 0, left, type, value);
 
@@ -314,7 +314,7 @@ int Padding::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) c
             return -100;
 
         if (elemsize == 1)
-            copy_make_border_image<signed char>(bottom_blob, top_blob, top, left, type, value);
+            copy_make_border_image<signed char>(bottom_blob, top_blob, top, left, type, static_cast<signed char>(value));
         else if (elemsize == 4)
             copy_make_border_image<float>(bottom_blob, top_blob, top, left, type, value);
 
@@ -334,7 +334,7 @@ int Padding::forward(const Mat& bottom_blob, Mat& top_blob, const Option& opt) c
             Mat borderm = top_blob.channel(q);
 
             if (elemsize == 1)
-                copy_make_border_image<signed char>(m, borderm, top, left, type, value);
+                copy_make_border_image<signed char>(m, borderm, top, left, type, static_cast<signed char>(value));
             else if (elemsize == 4)
                 copy_make_border_image<float>(m, borderm, top, left, type, value);
         }
@@ -386,7 +386,7 @@ int Padding::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top
             return -100;
 
         if (elemsize == 1)
-            copy_make_border_image<signed char>(bottom_blob, top_blob, 0, _left, type, value);
+            copy_make_border_image<signed char>(bottom_blob, top_blob, 0, _left, type, static_cast<signed char>(value));
         else if (elemsize == 4)
             copy_make_border_image<float>(bottom_blob, top_blob, 0, _left, type, value);
 
@@ -402,7 +402,7 @@ int Padding::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top
             return -100;
 
         if (elemsize == 1)
-            copy_make_border_image<signed char>(bottom_blob, top_blob, _top, _left, type, value);
+            copy_make_border_image<signed char>(bottom_blob, top_blob, _top, _left, type, static_cast<signed char>(value));
         else if (elemsize == 4)
             copy_make_border_image<float>(bottom_blob, top_blob, _top, _left, type, value);
 
@@ -422,7 +422,7 @@ int Padding::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top
             Mat borderm = top_blob.channel(q);
 
             if (elemsize == 1)
-                copy_make_border_image<signed char>(m, borderm, _top, _left, type, value);
+                copy_make_border_image<signed char>(m, borderm, _top, _left, type, static_cast<signed char>(value));
             else if (elemsize == 4)
                 copy_make_border_image<float>(m, borderm, _top, _left, type, value);
         }
diff --git a/src/layer/priorbox.cpp b/src/layer/priorbox.cpp
index 39f8aa45f..edbc888de 100644
--- a/src/layer/priorbox.cpp
+++ b/src/layer/priorbox.cpp
@@ -101,7 +101,7 @@ int PriorBox::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& to
                 float size = min_sizes[0];
                 for (int p = 1; p < num_ratios; p++)
                 {
-                    float ratio = sqrt(aspect_ratios[p]);
+                    float ratio = static_cast<float>(sqrt(aspect_ratios[p]));
                     float cw = size * h / w * ratio / 2;
                     float ch = size / ratio / 2;
 
@@ -141,13 +141,13 @@ int PriorBox::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& to
     {
         step_w = (float)image_w / w;
         if (step_mmdetection)
-            step_w = ceil((float)image_w / w);
+            step_w = static_cast<float>(ceil((float)image_w / w));
     }
     if (step_h == -233)
     {
         step_h = (float)image_h / h;
         if (step_mmdetection)
-            step_h = ceil((float)image_h / h);
+            step_h = static_cast<float>(ceil((float)image_h / h));
     }
 
     int num_min_size = min_sizes.w;
@@ -200,7 +200,7 @@ int PriorBox::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& to
                     float max_size = max_sizes[k];
 
                     // max size box
-                    box_w = box_h = sqrt(min_size * max_size);
+                    box_w = box_h = static_cast<float>(sqrt(min_size * max_size));
 
                     box[0] = (center_x - box_w * 0.5f) / image_w;
                     box[1] = (center_y - box_h * 0.5f) / image_h;
@@ -215,8 +215,8 @@ int PriorBox::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& to
                 {
                     float ar = aspect_ratios[p];
 
-                    box_w = min_size * sqrt(ar);
-                    box_h = min_size / sqrt(ar);
+                    box_w = static_cast<float>(min_size * sqrt(ar));
+                    box_h = static_cast<float>(min_size / sqrt(ar));
 
                     box[0] = (center_x - box_w * 0.5f) / image_w;
                     box[1] = (center_y - box_h * 0.5f) / image_h;
diff --git a/src/layer/requantize.cpp b/src/layer/requantize.cpp
index 9f51d5237..8114d99f8 100644
--- a/src/layer/requantize.cpp
+++ b/src/layer/requantize.cpp
@@ -29,7 +29,7 @@ Requantize::Requantize()
 
 static inline signed char float2int8(float v)
 {
-    int int32 = round(v);
+    int int32 = static_cast<int>(round(v));
     if (int32 > 127) return 127;
     if (int32 < -127) return -127;
     return (signed char)int32;
diff --git a/src/mat_pixel.cpp b/src/mat_pixel.cpp
index fbe01f361..0752d1049 100644
--- a/src/mat_pixel.cpp
+++ b/src/mat_pixel.cpp
@@ -717,7 +717,7 @@ static int from_rgb2gray(const unsigned char* rgb, int w, int h, int stride, Mat
 #endif // __ARM_NEON
         for (; remain>0; remain--)
         {
-            *ptr = (rgb[0] * R2Y + rgb[1] * G2Y + rgb[2] * B2Y) >> Y_shift;
+            *ptr = static_cast<float>((rgb[0] * R2Y + rgb[1] * G2Y + rgb[2] * B2Y) >> Y_shift);
 
             rgb += 3;
             ptr++;
@@ -874,7 +874,7 @@ static int from_bgr2gray(const unsigned char* bgr, int w, int h, int stride, Mat
 #endif // __ARM_NEON
         for (; remain>0; remain--)
         {
-            *ptr = (bgr[2] * R2Y + bgr[1] * G2Y + bgr[0] * B2Y) >> Y_shift;
+            *ptr = static_cast<float>((bgr[2] * R2Y + bgr[1] * G2Y + bgr[0] * B2Y) >> Y_shift);
 
             bgr += 3;
             ptr++;
@@ -1428,7 +1428,7 @@ static int from_rgba2gray(const unsigned char* rgba, int w, int h, int stride, M
 #endif // __ARM_NEON
         for (; remain>0; remain--)
         {
-            *ptr = (rgba[0] * R2Y + rgba[1] * G2Y + rgba[2] * B2Y) >> Y_shift;
+            *ptr = static_cast<float>((rgba[0] * R2Y + rgba[1] * G2Y + rgba[2] * B2Y) >> Y_shift);
 
             rgba += 4;
             ptr++;
diff --git a/src/mat_pixel_resize.cpp b/src/mat_pixel_resize.cpp
index 79afc334e..1290043bf 100644
--- a/src/mat_pixel_resize.cpp
+++ b/src/mat_pixel_resize.cpp
@@ -71,7 +71,7 @@ void resize_bilinear_c1(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dx = 0; dx < w; dx++)
     {
         fx = (float)((dx + 0.5) * scale_x - 0.5);
-        sx = floor(fx);
+        sx = static_cast<int>(floor(fx));
         fx -= sx;
 
         if (sx < 0)
@@ -97,7 +97,7 @@ void resize_bilinear_c1(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dy = 0; dy < h; dy++)
     {
         fy = (float)((dy + 0.5) * scale_y - 0.5);
-        sy = floor(fy);
+        sy = static_cast<int>(floor(fy));
         fy -= sy;
 
         if (sy < 0)
@@ -326,7 +326,7 @@ void resize_bilinear_c2(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dx = 0; dx < w; dx++)
     {
         fx = (float)((dx + 0.5) * scale_x - 0.5);
-        sx = floor(fx);
+        sx = static_cast<int>(floor(fx));
         fx -= sx;
 
         if (sx < 0)
@@ -352,7 +352,7 @@ void resize_bilinear_c2(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dy = 0; dy < h; dy++)
     {
         fy = (float)((dy + 0.5) * scale_y - 0.5);
-        sy = floor(fy);
+        sy = static_cast<int>(floor(fy));
         fy -= sy;
 
         if (sy < 0)
@@ -636,7 +636,7 @@ void resize_bilinear_c3(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dx = 0; dx < w; dx++)
     {
         fx = (float)((dx + 0.5) * scale_x - 0.5);
-        sx = floor(fx);
+        sx = static_cast<int>(floor(fx));
         fx -= sx;
 
         if (sx < 0)
@@ -662,7 +662,7 @@ void resize_bilinear_c3(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dy = 0; dy < h; dy++)
     {
         fy = (float)((dy + 0.5) * scale_y - 0.5);
-        sy = floor(fy);
+        sy = static_cast<int>(floor(fy));
         fy -= sy;
 
         if (sy < 0)
@@ -957,7 +957,7 @@ void resize_bilinear_c4(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dx = 0; dx < w; dx++)
     {
         fx = (float)((dx + 0.5) * scale_x - 0.5);
-        sx = floor(fx);
+        sx = static_cast<int>(floor(fx));
         fx -= sx;
 
         if (sx < 0)
@@ -983,7 +983,7 @@ void resize_bilinear_c4(const unsigned char* src, int srcw, int srch, int srcstr
     for (int dy = 0; dy < h; dy++)
     {
         fy = (float)((dy + 0.5) * scale_y - 0.5);
-        sy = floor(fy);
+        sy = static_cast<int>(floor(fy));
         fy -= sy;
 
         if (sy < 0)
diff --git a/src/modelbin.cpp b/src/modelbin.cpp
index 95d67bd70..51c35e2eb 100644
--- a/src/modelbin.cpp
+++ b/src/modelbin.cpp
@@ -77,7 +77,7 @@ Mat ModelBinFromDataReader::load(int w, int type) const
         if (flag_struct.tag == 0x01306B47)
         {
             // half-precision data
-            int align_data_size = alignSize(w * sizeof(unsigned short), 4);
+            size_t align_data_size = alignSize(w * sizeof(unsigned short), 4);
             std::vector<unsigned short> float16_weights;
             float16_weights.resize(align_data_size);
             nread = dr.read(float16_weights.data(), align_data_size);
@@ -92,7 +92,7 @@ Mat ModelBinFromDataReader::load(int w, int type) const
         else if (flag_struct.tag == 0x000D4B38)
         {
             // int8 data
-            int align_data_size = alignSize(w, 4);
+            size_t align_data_size = alignSize(w, 4);
             std::vector<signed char> int8_weights;
             int8_weights.resize(align_data_size);
             nread = dr.read(int8_weights.data(), align_data_size);
@@ -142,7 +142,7 @@ Mat ModelBinFromDataReader::load(int w, int type) const
                 return Mat();
             }
 
-            int align_weight_data_size = alignSize(w * sizeof(unsigned char), 4);
+            size_t align_weight_data_size = alignSize(w * sizeof(unsigned char), 4);
             std::vector<unsigned char> index_array;
             index_array.resize(align_weight_data_size);
             nread = dr.read(index_array.data(), align_weight_data_size);