diff --git a/dnn/src/fallback/general_intrinsic/gi_common.h b/dnn/src/fallback/general_intrinsic/gi_common.h
index df076993af21d2fd020bfae152c2198ea0ed5313..5efca76fa4ffd8bbd8d000109947d58f26f5855c 100644
--- a/dnn/src/fallback/general_intrinsic/gi_common.h
+++ b/dnn/src/fallback/general_intrinsic/gi_common.h
@@ -60,15 +60,15 @@
 #define GI_NEON32_INTRINSICS
 #endif
 #elif defined(GI_TARGET_X86)
-//#if defined(__FMA__)
-//#define GI_FMA_INTRINSICS
-//#define GI_AVX2_INTRINSICS
-//#define GI_AVX_INTRINSICS
-//#elif defined(__AVX2__)
-//#define GI_AVX2_INTRINSICS
-//#define GI_AVX_INTRINSICS
-//#elif defined(__AVX__)
-//#define GI_AVX_INTRINSICS
+// #if defined(__FMA__)
+// #define GI_FMA_INTRINSICS
+// #define GI_AVX2_INTRINSICS
+// #define GI_AVX_INTRINSICS
+// #elif defined(__AVX2__)
+// #define GI_AVX2_INTRINSICS
+// #define GI_AVX_INTRINSICS
+// #elif defined(__AVX__)
+// #define GI_AVX_INTRINSICS
 #if defined(__SSE4_2__)
 #define GI_SSE42_INTRINSICS
 #define GI_SSE2_INTRINSICS
@@ -94,7 +94,7 @@
 
 //! Gi fp16 only support arm64 neon and rvv
 #if (defined(GI_NEON_INTRINSICS) && __ARM_FEATURE_FP16_VECTOR_ARITHMETIC && \
-     MEGDNN_AARCH64) ||                                                     \
+     defined(__aarch64__)) ||                                               \
         defined(GI_RVV_INTRINSICS)
 #define GI_SUPPORT_F16
 #endif
diff --git a/dnn/src/fallback/general_intrinsic/gi_float16.h b/dnn/src/fallback/general_intrinsic/gi_float16.h
index f85997208fa754222e09c7f0136d909f126e7868..3359c6b147fe333070ae43e781ea27cd8703b785 100644
--- a/dnn/src/fallback/general_intrinsic/gi_float16.h
+++ b/dnn/src/fallback/general_intrinsic/gi_float16.h
@@ -34,7 +34,7 @@ GI_FLOAT16_t GiLoadBroadcastFloat16(const gi_float16_t* Value) {
 }
 
 GI_FORCEINLINE
-GI_FLOAT32_V2_t GiCastFloat16ToFloat32(const GI_FLOAT16_t& fp16) {
+GI_FLOAT32_V2_t GiCastFloat16ToFloat32(const GI_FLOAT16_t fp16) {
 #if defined(GI_NEON_INTRINSICS)
     GI_FLOAT32_V2_t ret;
     GiSetSubVectorFloat32V2(ret, 0, vcvt_f32_f16(vget_low_f16(fp16)));
@@ -51,7 +51,7 @@ GI_FLOAT32_V2_t GiCastFloat16ToFloat32(const GI_FLOAT16_t& fp16) {
 }
 
 GI_FORCEINLINE
-GI_FLOAT16_t GiCastFloat32ToFloat16(const GI_FLOAT32_t& low, const GI_FLOAT32_t& high) {
+GI_FLOAT16_t GiCastFloat32ToFloat16(const GI_FLOAT32_t low, const GI_FLOAT32_t high) {
 #if defined(GI_NEON_INTRINSICS)
     return vcombine_f16(vcvt_f16_f32(low), vcvt_f16_f32(high));
 #elif defined(GI_RVV_INTRINSICS)