diff --git a/src/layer/arm/gru_int8.h b/src/layer/arm/gru_int8.h
index 388d754677b..78696de401e 100644
--- a/src/layer/arm/gru_int8.h
+++ b/src/layer/arm/gru_int8.h
@@ -503,7 +503,7 @@ static void gru_int8_gate_output(const Mat& gates, Mat& hidden_state, Mat& top_b
                 "st1    {v0.4h}, [%0]       \n"
                 : "=r"(outptr) // %0
                 : "0"(outptr),
-                  "w"(_gru_H0)
+                "w"(_gru_H0)
                 : "memory", "v0");
 #else  // __aarch64__
             asm volatile(
@@ -511,7 +511,7 @@ static void gru_int8_gate_output(const Mat& gates, Mat& hidden_state, Mat& top_b
                 "vst1.u16   {d0}, [%0]      \n"
                 : "=r"(outptr) // %0
                 : "0"(outptr),
-                  "w"(_gru_H0)
+                "w"(_gru_H0)
                 : "memory", "q0");
 #endif // __aarch64__
 #else  // NCNN_GNU_INLINE_ASM
diff --git a/src/layer/arm/lstm_int8.h b/src/layer/arm/lstm_int8.h
index e4cebf0ab22..b58ebfde398 100644
--- a/src/layer/arm/lstm_int8.h
+++ b/src/layer/arm/lstm_int8.h
@@ -256,7 +256,7 @@ static void lstm_int8_gate_output(const Mat& gates, const Mat& weight_hr, Mat& h
                     "st1    {v0.4h}, [%0]       \n"
                     : "=r"(outptr) // %0
                     : "0"(outptr),
-                      "w"(_lstm_H)
+                    "w"(_lstm_H)
                     : "memory", "v0");
 #else  // __aarch64__
                 asm volatile(
@@ -264,7 +264,7 @@ static void lstm_int8_gate_output(const Mat& gates, const Mat& weight_hr, Mat& h
                     "vst1.u16   {d0}, [%0]      \n"
                     : "=r"(outptr) // %0
                     : "0"(outptr),
-                      "w"(_lstm_H)
+                    "w"(_lstm_H)
                     : "memory", "q0");
 #endif // __aarch64__
 #else  // NCNN_GNU_INLINE_ASM
diff --git a/src/layer/arm/rnn_int8.h b/src/layer/arm/rnn_int8.h
index dc96be3e64a..0dbf849fb85 100644
--- a/src/layer/arm/rnn_int8.h
+++ b/src/layer/arm/rnn_int8.h
@@ -259,7 +259,7 @@ static void rnn_int8_gate_output(const Mat& gates, Mat& hidden_state, Mat& top_b
                 "st1    {v0.4h}, [%0]       \n"
                 : "=r"(outptr) // %0
                 : "0"(outptr),
-                  "w"(_rnn_H)
+                "w"(_rnn_H)
                 : "memory", "v0");
 #else  // __aarch64__
             asm volatile(
@@ -267,7 +267,7 @@ static void rnn_int8_gate_output(const Mat& gates, Mat& hidden_state, Mat& top_b
                 "vst1.u16   {d0}, [%0]      \n"
                 : "=r"(outptr) // %0
                 : "0"(outptr),
-                  "w"(_rnn_H)
+                "w"(_rnn_H)
                 : "memory", "q0");
 #endif // __aarch64__
 #else  // NCNN_GNU_INLINE_ASM