diff --git a/dnn/src/fallback/conv_bias/conv1x1/conv1x1_utils.cpp b/dnn/src/fallback/conv_bias/conv1x1/conv1x1_utils.cpp
index 2d899a6f325acbd61ced3680d4494177e779c371..e909443063f2d33d9c5ba1183140f59617b92857 100644
--- a/dnn/src/fallback/conv_bias/conv1x1/conv1x1_utils.cpp
+++ b/dnn/src/fallback/conv_bias/conv1x1/conv1x1_utils.cpp
@@ -17,7 +17,7 @@ WorkspaceBundle get_thread_bundle(
                        (param.src_type.enumv() == DTypeEnum::Quantized8Asymm &&
                         param.dst_type.enumv() == DTypeEnum::Quantized8Asymm);
     size_t matmul_dst_bytes_per_thread =
-            is_dst_8bit ? oc_tile_size * OH * OW * sizeof(param.bias_type) : 0;
+            is_dst_8bit ? oc_tile_size * OH * OW * param.bias_type.size() : 0;
     return WorkspaceBundle{nullptr, {matmul_c_size, matmul_dst_bytes_per_thread}};
 }
 
diff --git a/dnn/src/fallback/conv_bias/im2col/algos.cpp b/dnn/src/fallback/conv_bias/im2col/algos.cpp
index cdad66974c69b84cc105fc2638bcb7f05f565e8c..5b28e90434001796dbda68c2f378c389b98e3509 100644
--- a/dnn/src/fallback/conv_bias/im2col/algos.cpp
+++ b/dnn/src/fallback/conv_bias/im2col/algos.cpp
@@ -199,8 +199,7 @@ static WorkspaceBundle get_bundle(
     if (no_need_pading) {
         padding = 0;  //! not need  padding
     } else {
-        padding =
-                (GROUP * N * IC * IH2 * IW2) * sizeof(param.src_type);  //! for padding
+        padding = (GROUP * N * IC * IH2 * IW2) * param.src_type.size();  //! for padding
     }
 
     packa_size = GROUP * packa_group_size;  //! for packA  size = GROUP * a_size
diff --git a/dnn/src/fallback/conv_bias/im2col/im2col_kerns.h b/dnn/src/fallback/conv_bias/im2col/im2col_kerns.h
index fb86ff56e7e1b8b04fd07f10f7201c572a00fa7b..143e14fb1305ae3afe024596ef1058cf79d56877 100644
--- a/dnn/src/fallback/conv_bias/im2col/im2col_kerns.h
+++ b/dnn/src/fallback/conv_bias/im2col/im2col_kerns.h
@@ -146,15 +146,15 @@ public:
         size_t im2col = 0, packb = 0, bias_temp = 0;
         bool default_pack = matmul_algo->packmode() == Pack_Mode::DEFAULT;
         megdnn_assert(default_pack, "only support default packa");
-        size_t im2col_dst_size = IC * FH * FW * ohw_tile_size * sizeof(param.src_type);
+        size_t im2col_dst_size = IC * FH * FW * ohw_tile_size * param.src_type.size();
         size_t matmul_dst_size =
-                pack_oc_size * oc_tile_size * ohw_tile_size * sizeof(param.bias_type);
+                pack_oc_size * oc_tile_size * ohw_tile_size * param.bias_type.size();
         //! matmul_dst and im2col_dst use the same memory
         WorkspaceBundle wb = matmul_algo->get_bundle(im2col_kern_param);
         packb = wb.get_size(1);
         im2col = std::max(im2col_dst_size, matmul_dst_size);
         if (param.bias_mode == megdnn::BiasMode::BIAS) {
-            bias_temp = oc_tile_size * ohw_tile_size * sizeof(param.bias_type);
+            bias_temp = oc_tile_size * ohw_tile_size * param.bias_type.size();
         }
         return {nullptr, {packb, im2col, bias_temp}};
     }
@@ -231,15 +231,15 @@ public:
         size_t im2col = 0, packb = 0, matmul_dst = 0, bias_temp = 0;
         bool only_packA = matmul_algo->packmode() == Pack_Mode::ONLY_PACKA;
         megdnn_assert(only_packA, "onlysupport onlypackA mode");
-        size_t im2col_dst_size = IC * FH * FW * ohw_tile_size * sizeof(param.src_type);
-        size_t matmul_dst_size = oc_tile_size * ohw_tile_size * sizeof(param.bias_type);
+        size_t im2col_dst_size = IC * FH * FW * ohw_tile_size * param.src_type.size();
+        size_t matmul_dst_size = oc_tile_size * ohw_tile_size * param.bias_type.size();
         //! matmul_dst and im2col_dst use the same memory
         WorkspaceBundle wb = matmul_algo->get_bundle(im2col_kern_param);
         packb = wb.get_size(1);
         im2col = im2col_dst_size;
         matmul_dst = matmul_dst_size;
         if (param.bias_mode == megdnn::BiasMode::BIAS) {
-            bias_temp = oc_tile_size * ohw_tile_size * sizeof(param.bias_type);
+            bias_temp = oc_tile_size * ohw_tile_size * param.bias_type.size();
         }
 
         return {nullptr, {packb, im2col, matmul_dst, bias_temp}};
@@ -309,8 +309,8 @@ public:
                             param.dst_type.enumv() == DTypeEnum::QuantizedS8) ||
                            (param.src_type.enumv() == DTypeEnum::Quantized8Asymm &&
                             param.dst_type.enumv() == DTypeEnum::Quantized8Asymm);
-        size_t im2col_dst_size = IC * FH * FW * ohw_tile_size * sizeof(param.src_type);
-        size_t matmul_dst_size = oc_tile_size * ohw_tile_size * sizeof(param.bias_type);
+        size_t im2col_dst_size = IC * FH * FW * ohw_tile_size * param.src_type.size();
+        size_t matmul_dst_size = oc_tile_size * ohw_tile_size * param.bias_type.size();
         im2col = im2col_dst_size;
         if (is_dst_8bit) {
             matmul_dst = matmul_dst_size;
@@ -319,9 +319,8 @@ public:
         }
         matmul_compute = matmul_algo->get_workspace(im2col_kern_param);
         if (param.bias_mode == megdnn::BiasMode::BIAS) {
-            bias_temp = oc_tile_size * ohw_tile_size * sizeof(param.bias_type);
+            bias_temp = oc_tile_size * ohw_tile_size * param.bias_type.size();
         }
-
         return {nullptr, {im2col, matmul_dst, bias_temp, matmul_compute}};
     }
 };