diff --git a/src/dsp/dec_neon.c b/src/dsp/dec_neon.c
index afdb48f1..d84e0d1c 100644
--- a/src/dsp/dec_neon.c
+++ b/src/dsp/dec_neon.c
@@ -479,6 +479,21 @@ static int8x16_t GetBaseDelta0(const int8x16_t p0, const int8x16_t q0) {
 
 //------------------------------------------------------------------------------
 
+static void ApplyFilter2NoFlip(const int8x16_t p0s, const int8x16_t q0s,
+                               const int8x16_t delta,
+                               int8x16_t* const op0, int8x16_t* const oq0) {
+  const int8x16_t kCst3 = vdupq_n_s8(0x03);
+  const int8x16_t kCst4 = vdupq_n_s8(0x04);
+  const int8x16_t delta_p3 = vqaddq_s8(delta, kCst3);
+  const int8x16_t delta_p4 = vqaddq_s8(delta, kCst4);
+  const int8x16_t delta3 = vshrq_n_s8(delta_p3, 3);
+  const int8x16_t delta4 = vshrq_n_s8(delta_p4, 3);
+  *op0 = vqaddq_s8(p0s, delta3);
+  *oq0 = vqsubq_s8(q0s, delta4);
+}
+
+#if defined(WEBP_USE_INTRINSICS)
+
 static void ApplyFilter2(const int8x16_t p0s, const int8x16_t q0s,
                          const int8x16_t delta,
                          uint8x16_t* const op0, uint8x16_t* const oq0) {
@@ -494,8 +509,6 @@ static void ApplyFilter2(const int8x16_t p0s, const int8x16_t q0s,
   *oq0 = FlipSignBack(sq0);
 }
 
-#if defined(WEBP_USE_INTRINSICS)
-
 static void DoFilter2(const uint8x16_t p1, const uint8x16_t p0,
                       const uint8x16_t q0, const uint8x16_t q1,
                       const uint8x16_t mask,
@@ -721,11 +734,7 @@ static void DoFilter4(
     const int8x16_t delta = GetBaseDelta(p1s, p0s, q0s, q1s);
     const int8x16_t simple_lf_delta =
         vandq_s8(delta, vreinterpretq_s8_u8(simple_lf_mask));
-    uint8x16_t tmp_p0, tmp_q0;
-    ApplyFilter2(p0s, q0s, simple_lf_delta, &tmp_p0, &tmp_q0);
-    // TODO(skal): avoid the double FlipSign() in ApplyFilter2() and here
-    p0s = FlipSign(tmp_p0);
-    q0s = FlipSign(tmp_q0);
+    ApplyFilter2NoFlip(p0s, q0s, simple_lf_delta, &p0s, &q0s);
   }
 
   // do_filter4 part (complex loopfilter on pixels without hev)
@@ -797,11 +806,7 @@ static void DoFilter6(
   {
     const int8x16_t simple_lf_delta =
         vandq_s8(delta0, vreinterpretq_s8_u8(simple_lf_mask));
-    uint8x16_t tmp_p0, tmp_q0;
-    ApplyFilter2(p0s, q0s, simple_lf_delta, &tmp_p0, &tmp_q0);
-    // TODO(skal): avoid the double FlipSign() in ApplyFilter2() and here
-    p0s = FlipSign(tmp_p0);
-    q0s = FlipSign(tmp_q0);
+    ApplyFilter2NoFlip(p0s, q0s, simple_lf_delta, &p0s, &q0s);
   }
 
   // do_filter6 part (complex loopfilter on pixels without hev)