less unroll for unaryop arm, fix padding arm warning

src/layer/arm/padding_arm.cpp

@@ -419,6 +419,8 @@ int Padding_arm::forward_bf16s_fp16s(const Mat& bottom_blob, Mat& top_blob, cons
                 else
 #endif
                 {
+                    // shall never reach here
+                    pad_value = vdup_n_u16(0);
                 }
                 // *INDENT-ON*
                 // clang-format on
@@ -464,6 +466,8 @@ int Padding_arm::forward_bf16s_fp16s(const Mat& bottom_blob, Mat& top_blob, cons
                 else
 #endif
                 {
+                    // shall never reach here
+                    pad_value = vdup_n_u16(0);
                 }
                 // *INDENT-ON*
                 // clang-format on
@@ -516,6 +520,8 @@ int Padding_arm::forward_bf16s_fp16s(const Mat& bottom_blob, Mat& top_blob, cons
                     else
 #endif
                     {
+                        // shall never reach here
+                        pad_value = vdup_n_u16(0);
                     }
                     // *INDENT-ON*
                     // clang-format on
@@ -574,6 +580,8 @@ int Padding_arm::forward_bf16s_fp16s(const Mat& bottom_blob, Mat& top_blob, cons
                     else
 #endif
                     {
+                        // shall never reach here
+                        pad_value = vdup_n_u16(0);
                     }
                     // *INDENT-ON*
                     // clang-format on

src/layer/arm/unaryop_arm.cpp

@@ -58,24 +58,6 @@ static int unary_op_inplace(Mat& a, const Option& opt)
 
         int i = 0;
 #if __ARM_NEON
-#if __aarch64__
-        for (; i + 15 < size; i += 16)
-        {
-            float32x4_t _p0 = vld1q_f32(ptr);
-            float32x4_t _p1 = vld1q_f32(ptr + 4);
-            float32x4_t _p2 = vld1q_f32(ptr + 8);
-            float32x4_t _p3 = vld1q_f32(ptr + 12);
-            _p0 = op.func_pack4(_p0);
-            _p1 = op.func_pack4(_p1);
-            _p2 = op.func_pack4(_p2);
-            _p3 = op.func_pack4(_p3);
-            vst1q_f32(ptr, _p0);
-            vst1q_f32(ptr + 4, _p1);
-            vst1q_f32(ptr + 8, _p2);
-            vst1q_f32(ptr + 12, _p3);
-            ptr += 16;
-        }
-#endif // __aarch64__
         for (; i + 7 < size; i += 8)
         {
             float32x4_t _p0 = vld1q_f32(ptr);

src/layer/arm/unaryop_arm_asimdhp.cpp

@@ -44,22 +44,6 @@ static int unary_op_inplace_fp16s(Mat& a, const Option& opt)
         __fp16* ptr = a.channel(q);
 
         int i = 0;
-        for (; i + 31 < size; i += 32)
-        {
-            float16x8_t _p0 = vld1q_f16(ptr);
-            float16x8_t _p1 = vld1q_f16(ptr + 8);
-            float16x8_t _p2 = vld1q_f16(ptr + 16);
-            float16x8_t _p3 = vld1q_f16(ptr + 24);
-            _p0 = op.func_pack8(_p0);
-            _p1 = op.func_pack8(_p1);
-            _p2 = op.func_pack8(_p2);
-            _p3 = op.func_pack8(_p3);
-            vst1q_f16(ptr, _p0);
-            vst1q_f16(ptr + 8, _p1);
-            vst1q_f16(ptr + 16, _p2);
-            vst1q_f16(ptr + 24, _p3);
-            ptr += 32;
-        }
         for (; i + 15 < size; i += 16)
         {
             float16x8_t _p0 = vld1q_f16(ptr);