Update aom to commit id e87fb2378f01103d5d6e477a4ef6892dc714e614

1 files changed, 324 insertions, 2 deletions

diff --git a/third_party/aom/aom_dsp/loopfilter.c b/third_party/aom/aom_dsp/loopfilter.c
index 7ea1e6b89..69f131378 100644
--- a/third_party/aom/aom_dsp/loopfilter.c
+++ b/third_party/aom/aom_dsp/loopfilter.c
@@ -23,6 +23,14 @@ static INLINE int8_t signed_char_clamp(int t) {
 #define PARALLEL_DEBLOCKING_11_TAP 0
 #define PARALLEL_DEBLOCKING_9_TAP 0
 
+#if CONFIG_DEBLOCK_13TAP
+#define PARALLEL_DEBLOCKING_13_TAP 1
+#define PARALLEL_DEBLOCKING_5_TAP_CHROMA 1
+#else
+#define PARALLEL_DEBLOCKING_13_TAP 0
+#define PARALLEL_DEBLOCKING_5_TAP_CHROMA 0
+#endif
+
 #if CONFIG_HIGHBITDEPTH
 static INLINE int16_t signed_char_clamp_high(int t, int bd) {
   switch (bd) {
@@ -58,6 +66,19 @@ static INLINE int8_t filter_mask(uint8_t limit, uint8_t blimit, uint8_t p3,
   return ~mask;
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+static INLINE int8_t flat_mask3_chroma(uint8_t thresh, uint8_t p2, uint8_t p1,
+                                       uint8_t p0, uint8_t q0, uint8_t q1,
+                                       uint8_t q2) {
+  int8_t mask = 0;
+  mask |= (abs(p1 - p0) > thresh) * -1;
+  mask |= (abs(q1 - q0) > thresh) * -1;
+  mask |= (abs(p2 - p0) > thresh) * -1;
+  mask |= (abs(q2 - q0) > thresh) * -1;
+  return ~mask;
+}
+#endif
+
 static INLINE int8_t flat_mask4(uint8_t thresh, uint8_t p3, uint8_t p2,
                                 uint8_t p1, uint8_t p0, uint8_t q0, uint8_t q1,
                                 uint8_t q2, uint8_t q3) {
@@ -216,6 +237,25 @@ void aom_lpf_vertical_4_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0,
   aom_lpf_vertical_4_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1);
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+static INLINE void filter6(int8_t mask, uint8_t thresh, int8_t flat,
+                           uint8_t *op2, uint8_t *op1, uint8_t *op0,
+                           uint8_t *oq0, uint8_t *oq1, uint8_t *oq2) {
+  if (flat && mask) {
+    const uint8_t p2 = *op2, p1 = *op1, p0 = *op0;
+    const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2;
+
+    // 5-tap filter [1, 2, 2, 2, 1]
+    *op1 = ROUND_POWER_OF_TWO(p2 * 3 + p1 * 2 + p0 * 2 + q0, 3);
+    *op0 = ROUND_POWER_OF_TWO(p2 + p1 * 2 + p0 * 2 + q0 * 2 + q1, 3);
+    *oq0 = ROUND_POWER_OF_TWO(p1 + p0 * 2 + q0 * 2 + q1 * 2 + q2, 3);
+    *oq1 = ROUND_POWER_OF_TWO(p0 + q0 * 2 + q1 * 2 + q2 * 3, 3);
+  } else {
+    filter4(mask, thresh, op1, op0, oq0, oq1);
+  }
+}
+#endif
+
 static INLINE void filter8(int8_t mask, uint8_t thresh, int8_t flat,
                            uint8_t *op3, uint8_t *op2, uint8_t *op1,
                            uint8_t *op0, uint8_t *oq0, uint8_t *oq1,
@@ -236,6 +276,32 @@ static INLINE void filter8(int8_t mask, uint8_t thresh, int8_t flat,
   }
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+void aom_lpf_horizontal_6_c(uint8_t *s, int p, const uint8_t *blimit,
+                            const uint8_t *limit, const uint8_t *thresh) {
+  int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+  int count = 4;
+#else
+  int count = 8;
+#endif
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < count; ++i) {
+    const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
+    const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
+
+    const int8_t mask =
+        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t flat = flat_mask3_chroma(1, p2, p1, p0, q0, q1, q2);
+    filter6(mask, *thresh, flat, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p,
+            s + 2 * p);
+    ++s;
+  }
+}
+#endif
+
 void aom_lpf_horizontal_8_c(uint8_t *s, int p, const uint8_t *blimit,
                             const uint8_t *limit, const uint8_t *thresh) {
   int i;
@@ -268,6 +334,28 @@ void aom_lpf_horizontal_8_dual_c(uint8_t *s, int p, const uint8_t *blimit0,
   aom_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1);
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+void aom_lpf_vertical_6_c(uint8_t *s, int pitch, const uint8_t *blimit,
+                          const uint8_t *limit, const uint8_t *thresh) {
+  int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+  int count = 4;
+#else
+  int count = 8;
+#endif
+
+  for (i = 0; i < count; ++i) {
+    const uint8_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
+    const uint8_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
+    const int8_t mask =
+        filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
+    const int8_t flat = flat_mask3_chroma(1, p2, p1, p0, q0, q1, q2);
+    filter6(mask, *thresh, flat, s - 3, s - 2, s - 1, s, s + 1, s + 2);
+    s += pitch;
+  }
+}
+#endif
+
 void aom_lpf_vertical_8_c(uint8_t *s, int pitch, const uint8_t *blimit,
                           const uint8_t *limit, const uint8_t *thresh) {
   int i;
@@ -297,6 +385,56 @@ void aom_lpf_vertical_8_dual_c(uint8_t *s, int pitch, const uint8_t *blimit0,
   aom_lpf_vertical_8_c(s + 8 * pitch, pitch, blimit1, limit1, thresh1);
 }
 
+#if PARALLEL_DEBLOCKING_13_TAP
+static INLINE void filter14(int8_t mask, uint8_t thresh, int8_t flat,
+                            int8_t flat2, uint8_t *op6, uint8_t *op5,
+                            uint8_t *op4, uint8_t *op3, uint8_t *op2,
+                            uint8_t *op1, uint8_t *op0, uint8_t *oq0,
+                            uint8_t *oq1, uint8_t *oq2, uint8_t *oq3,
+                            uint8_t *oq4, uint8_t *oq5, uint8_t *oq6) {
+  if (flat2 && flat && mask) {
+    const uint8_t p6 = *op6, p5 = *op5, p4 = *op4, p3 = *op3, p2 = *op2,
+                  p1 = *op1, p0 = *op0;
+    const uint8_t q0 = *oq0, q1 = *oq1, q2 = *oq2, q3 = *oq3, q4 = *oq4,
+                  q5 = *oq5, q6 = *oq6;
+
+    // 13-tap filter [1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1]
+    *op5 = ROUND_POWER_OF_TWO(p6 * 7 + p5 * 2 + p4 * 2 + p3 + p2 + p1 + p0 + q0,
+                              4);
+    *op4 = ROUND_POWER_OF_TWO(
+        p6 * 5 + p5 * 2 + p4 * 2 + p3 * 2 + p2 + p1 + p0 + q0 + q1, 4);
+    *op3 = ROUND_POWER_OF_TWO(
+        p6 * 4 + p5 + p4 * 2 + p3 * 2 + p2 * 2 + p1 + p0 + q0 + q1 + q2, 4);
+    *op2 = ROUND_POWER_OF_TWO(
+        p6 * 3 + p5 + p4 + p3 * 2 + p2 * 2 + p1 * 2 + p0 + q0 + q1 + q2 + q3,
+        4);
+    *op1 = ROUND_POWER_OF_TWO(p6 * 2 + p5 + p4 + p3 + p2 * 2 + p1 * 2 + p0 * 2 +
+                                  q0 + q1 + q2 + q3 + q4,
+                              4);
+    *op0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 * 2 +
+                                  q0 * 2 + q1 + q2 + q3 + q4 + q5,
+                              4);
+    *oq0 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 * 2 + q0 * 2 +
+                                  q1 * 2 + q2 + q3 + q4 + q5 + q6,
+                              4);
+    *oq1 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 + q0 * 2 + q1 * 2 +
+                                  q2 * 2 + q3 + q4 + q5 + q6 * 2,
+                              4);
+    *oq2 = ROUND_POWER_OF_TWO(
+        p3 + p2 + p1 + p0 + q0 + q1 * 2 + q2 * 2 + q3 * 2 + q4 + q5 + q6 * 3,
+        4);
+    *oq3 = ROUND_POWER_OF_TWO(
+        p2 + p1 + p0 + q0 + q1 + q2 * 2 + q3 * 2 + q4 * 2 + q5 + q6 * 4, 4);
+    *oq4 = ROUND_POWER_OF_TWO(
+        p1 + p0 + q0 + q1 + q2 + q3 * 2 + q4 * 2 + q5 * 2 + q6 * 5, 4);
+    *oq5 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + q2 + q3 + q4 * 2 + q5 * 2 + q6 * 7,
+                              4);
+  } else {
+    filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3);
+  }
+}
+#endif
+
 #if PARALLEL_DEBLOCKING_11_TAP
 static INLINE void filter12(int8_t mask, uint8_t thresh, int8_t flat,
                             int8_t flat2, uint8_t *op5, uint8_t *op4,
@@ -428,7 +566,16 @@ static void mb_lpf_horizontal_edge_w(uint8_t *s, int p, const uint8_t *blimit,
         filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
     const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
 
-#if PARALLEL_DEBLOCKING_11_TAP
+#if PARALLEL_DEBLOCKING_13_TAP
+    (void)p7;
+    (void)q7;
+    const int8_t flat2 = flat_mask4(1, p6, p5, p4, p0, q0, q4, q5, q6);
+
+    filter14(mask, *thresh, flat, flat2, s - 7 * p, s - 6 * p, s - 5 * p,
+             s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p,
+             s + 2 * p, s + 3 * p, s + 4 * p, s + 5 * p, s + 6 * p);
+
+#elif PARALLEL_DEBLOCKING_11_TAP
     const int8_t flat2 = flat_mask3(1, p5, p4, p0, q0, q4, q5);
 
     filter12(mask, *thresh, flat, flat2, s - 6 * p, s - 5 * p, s - 4 * p,
@@ -482,7 +629,14 @@ static void mb_lpf_vertical_edge_w(uint8_t *s, int p, const uint8_t *blimit,
         filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3);
     const int8_t flat = flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3);
 
-#if PARALLEL_DEBLOCKING_11_TAP
+#if PARALLEL_DEBLOCKING_13_TAP
+    (void)p7;
+    (void)q7;
+    const int8_t flat2 = flat_mask4(1, p6, p5, p4, p0, q0, q4, q5, q6);
+
+    filter14(mask, *thresh, flat, flat2, s - 7, s - 6, s - 5, s - 4, s - 3,
+             s - 2, s - 1, s, s + 1, s + 2, s + 3, s + 4, s + 5, s + 6);
+#elif PARALLEL_DEBLOCKING_11_TAP
     const int8_t flat2 = flat_mask3(1, p5, p4, p0, q0, q4, q5);
 
     filter12(mask, *thresh, flat, flat2, s - 6, s - 5, s - 4, s - 3, s - 2,
@@ -553,6 +707,21 @@ static INLINE int8_t highbd_filter_mask(uint8_t limit, uint8_t blimit,
   return ~mask;
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+static INLINE int8_t highbd_flat_mask3_chroma(uint8_t thresh, uint16_t p2,
+                                              uint16_t p1, uint16_t p0,
+                                              uint16_t q0, uint16_t q1,
+                                              uint16_t q2, int bd) {
+  int8_t mask = 0;
+  int16_t thresh16 = (uint16_t)thresh << (bd - 8);
+  mask |= (abs(p1 - p0) > thresh16) * -1;
+  mask |= (abs(q1 - q0) > thresh16) * -1;
+  mask |= (abs(p2 - p0) > thresh16) * -1;
+  mask |= (abs(q2 - q0) > thresh16) * -1;
+  return ~mask;
+}
+#endif
+
 static INLINE int8_t highbd_flat_mask4(uint8_t thresh, uint16_t p3, uint16_t p2,
                                        uint16_t p1, uint16_t p0, uint16_t q0,
                                        uint16_t q1, uint16_t q2, uint16_t q3,
@@ -708,6 +877,26 @@ void aom_highbd_lpf_vertical_4_dual_c(
                               bd);
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+static INLINE void highbd_filter6(int8_t mask, uint8_t thresh, int8_t flat,
+                                  uint16_t *op2, uint16_t *op1, uint16_t *op0,
+                                  uint16_t *oq0, uint16_t *oq1, uint16_t *oq2,
+                                  int bd) {
+  if (flat && mask) {
+    const uint16_t p2 = *op2, p1 = *op1, p0 = *op0;
+    const uint16_t q0 = *oq0, q1 = *oq1, q2 = *oq2;
+
+    // 5-tap filter [1, 2, 2, 2, 1]
+    *op1 = ROUND_POWER_OF_TWO(p2 * 3 + p1 * 2 + p0 * 2 + q0, 3);
+    *op0 = ROUND_POWER_OF_TWO(p2 + p1 * 2 + p0 * 2 + q0 * 2 + q1, 3);
+    *oq0 = ROUND_POWER_OF_TWO(p1 + p0 * 2 + q0 * 2 + q1 * 2 + q2, 3);
+    *oq1 = ROUND_POWER_OF_TWO(p0 + q0 * 2 + q1 * 2 + q2 * 3, 3);
+  } else {
+    highbd_filter4(mask, thresh, op1, op0, oq0, oq1, bd);
+  }
+}
+#endif
+
 static INLINE void highbd_filter8(int8_t mask, uint8_t thresh, int8_t flat,
                                   uint16_t *op3, uint16_t *op2, uint16_t *op1,
                                   uint16_t *op0, uint16_t *oq0, uint16_t *oq1,
@@ -754,6 +943,33 @@ void aom_highbd_lpf_horizontal_8_c(uint16_t *s, int p, const uint8_t *blimit,
   }
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+void aom_highbd_lpf_horizontal_6_c(uint16_t *s, int p, const uint8_t *blimit,
+                                   const uint8_t *limit, const uint8_t *thresh,
+                                   int bd) {
+  int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+  int count = 4;
+#else
+  int count = 8;
+#endif
+
+  // loop filter designed to work using chars so that we can make maximum use
+  // of 8 bit simd instructions.
+  for (i = 0; i < count; ++i) {
+    const uint16_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
+    const uint16_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
+
+    const int8_t mask =
+        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat = highbd_flat_mask3_chroma(1, p2, p1, p0, q0, q1, q2, bd);
+    highbd_filter6(mask, *thresh, flat, s - 3 * p, s - 2 * p, s - 1 * p, s,
+                   s + 1 * p, s + 2 * p, bd);
+    ++s;
+  }
+}
+#endif
+
 void aom_highbd_lpf_horizontal_8_dual_c(
     uint16_t *s, int p, const uint8_t *blimit0, const uint8_t *limit0,
     const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1,
@@ -762,6 +978,30 @@ void aom_highbd_lpf_horizontal_8_dual_c(
   aom_highbd_lpf_horizontal_8_c(s + 8, p, blimit1, limit1, thresh1, bd);
 }
 
+#if PARALLEL_DEBLOCKING_5_TAP_CHROMA
+void aom_highbd_lpf_vertical_6_c(uint16_t *s, int pitch, const uint8_t *blimit,
+                                 const uint8_t *limit, const uint8_t *thresh,
+                                 int bd) {
+  int i;
+#if CONFIG_PARALLEL_DEBLOCKING && CONFIG_CB4X4
+  int count = 4;
+#else
+  int count = 8;
+#endif
+
+  for (i = 0; i < count; ++i) {
+    const uint16_t p3 = s[-4], p2 = s[-3], p1 = s[-2], p0 = s[-1];
+    const uint16_t q0 = s[0], q1 = s[1], q2 = s[2], q3 = s[3];
+    const int8_t mask =
+        highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+    const int8_t flat = highbd_flat_mask3_chroma(1, p2, p1, p0, q0, q1, q2, bd);
+    highbd_filter6(mask, *thresh, flat, s - 3, s - 2, s - 1, s, s + 1, s + 2,
+                   bd);
+    s += pitch;
+  }
+}
+#endif
+
 void aom_highbd_lpf_vertical_8_c(uint16_t *s, int pitch, const uint8_t *blimit,
                                  const uint8_t *limit, const uint8_t *thresh,
                                  int bd) {
@@ -794,6 +1034,68 @@ void aom_highbd_lpf_vertical_8_dual_c(
                               bd);
 }
 
+#if PARALLEL_DEBLOCKING_13_TAP
+static INLINE void highbd_filter14(int8_t mask, uint8_t thresh, int8_t flat,
+                                   int8_t flat2, uint16_t *op6, uint16_t *op5,
+                                   uint16_t *op4, uint16_t *op3, uint16_t *op2,
+                                   uint16_t *op1, uint16_t *op0, uint16_t *oq0,
+                                   uint16_t *oq1, uint16_t *oq2, uint16_t *oq3,
+                                   uint16_t *oq4, uint16_t *oq5, uint16_t *oq6,
+                                   int bd) {
+  if (flat2 && flat && mask) {
+    const uint16_t p6 = *op6;
+    const uint16_t p5 = *op5;
+    const uint16_t p4 = *op4;
+    const uint16_t p3 = *op3;
+    const uint16_t p2 = *op2;
+    const uint16_t p1 = *op1;
+    const uint16_t p0 = *op0;
+    const uint16_t q0 = *oq0;
+    const uint16_t q1 = *oq1;
+    const uint16_t q2 = *oq2;
+    const uint16_t q3 = *oq3;
+    const uint16_t q4 = *oq4;
+    const uint16_t q5 = *oq5;
+    const uint16_t q6 = *oq6;
+
+    // 13-tap filter [1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1]
+    *op5 = ROUND_POWER_OF_TWO(p6 * 7 + p5 * 2 + p4 * 2 + p3 + p2 + p1 + p0 + q0,
+                              4);
+    *op4 = ROUND_POWER_OF_TWO(
+        p6 * 5 + p5 * 2 + p4 * 2 + p3 * 2 + p2 + p1 + p0 + q0 + q1, 4);
+    *op3 = ROUND_POWER_OF_TWO(
+        p6 * 4 + p5 + p4 * 2 + p3 * 2 + p2 * 2 + p1 + p0 + q0 + q1 + q2, 4);
+    *op2 = ROUND_POWER_OF_TWO(
+        p6 * 3 + p5 + p4 + p3 * 2 + p2 * 2 + p1 * 2 + p0 + q0 + q1 + q2 + q3,
+        4);
+    *op1 = ROUND_POWER_OF_TWO(p6 * 2 + p5 + p4 + p3 + p2 * 2 + p1 * 2 + p0 * 2 +
+                                  q0 + q1 + q2 + q3 + q4,
+                              4);
+    *op0 = ROUND_POWER_OF_TWO(p6 + p5 + p4 + p3 + p2 + p1 * 2 + p0 * 2 +
+                                  q0 * 2 + q1 + q2 + q3 + q4 + q5,
+                              4);
+    *oq0 = ROUND_POWER_OF_TWO(p5 + p4 + p3 + p2 + p1 + p0 * 2 + q0 * 2 +
+                                  q1 * 2 + q2 + q3 + q4 + q5 + q6,
+                              4);
+    *oq1 = ROUND_POWER_OF_TWO(p4 + p3 + p2 + p1 + p0 + q0 * 2 + q1 * 2 +
+                                  q2 * 2 + q3 + q4 + q5 + q6 * 2,
+                              4);
+    *oq2 = ROUND_POWER_OF_TWO(
+        p3 + p2 + p1 + p0 + q0 + q1 * 2 + q2 * 2 + q3 * 2 + q4 + q5 + q6 * 3,
+        4);
+    *oq3 = ROUND_POWER_OF_TWO(
+        p2 + p1 + p0 + q0 + q1 + q2 * 2 + q3 * 2 + q4 * 2 + q5 + q6 * 4, 4);
+    *oq4 = ROUND_POWER_OF_TWO(
+        p1 + p0 + q0 + q1 + q2 + q3 * 2 + q4 * 2 + q5 * 2 + q6 * 5, 4);
+    *oq5 = ROUND_POWER_OF_TWO(p0 + q0 + q1 + q2 + q3 + q4 * 2 + q5 * 2 + q6 * 7,
+                              4);
+  } else {
+    highbd_filter8(mask, thresh, flat, op3, op2, op1, op0, oq0, oq1, oq2, oq3,
+                   bd);
+  }
+}
+#endif
+
 static INLINE void highbd_filter16(int8_t mask, uint8_t thresh, int8_t flat,
                                    int8_t flat2, uint16_t *op7, uint16_t *op6,
                                    uint16_t *op5, uint16_t *op4, uint16_t *op3,
@@ -887,6 +1189,16 @@ static void highbd_mb_lpf_horizontal_edge_w(uint16_t *s, int p,
         highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
     const int8_t flat =
         highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+
+#if PARALLEL_DEBLOCKING_13_TAP
+    const int8_t flat2 =
+        highbd_flat_mask4(1, s[-7 * p], s[-6 * p], s[-5 * p], p0, q0, s[4 * p],
+                          s[5 * p], s[6 * p], bd);
+
+    highbd_filter14(mask, *thresh, flat, flat2, s - 7 * p, s - 6 * p, s - 5 * p,
+                    s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p,
+                    s + 2 * p, s + 3 * p, s + 4 * p, s + 5 * p, s + 6 * p, bd);
+#else
     const int8_t flat2 =
         highbd_flat_mask5(1, s[-8 * p], s[-7 * p], s[-6 * p], s[-5 * p], p0, q0,
                           s[4 * p], s[5 * p], s[6 * p], s[7 * p], bd);
@@ -895,6 +1207,7 @@ static void highbd_mb_lpf_horizontal_edge_w(uint16_t *s, int p,
                     s - 5 * p, s - 4 * p, s - 3 * p, s - 2 * p, s - 1 * p, s,
                     s + 1 * p, s + 2 * p, s + 3 * p, s + 4 * p, s + 5 * p,
                     s + 6 * p, s + 7 * p, bd);
+#endif
     ++s;
   }
 }
@@ -937,12 +1250,21 @@ static void highbd_mb_lpf_vertical_edge_w(uint16_t *s, int p,
         highbd_filter_mask(*limit, *blimit, p3, p2, p1, p0, q0, q1, q2, q3, bd);
     const int8_t flat =
         highbd_flat_mask4(1, p3, p2, p1, p0, q0, q1, q2, q3, bd);
+#if PARALLEL_DEBLOCKING_13_TAP
+    const int8_t flat2 =
+        highbd_flat_mask4(1, s[-7], s[-6], s[-5], p0, q0, s[4], s[5], s[6], bd);
+
+    highbd_filter14(mask, *thresh, flat, flat2, s - 7, s - 6, s - 5, s - 4,
+                    s - 3, s - 2, s - 1, s, s + 1, s + 2, s + 3, s + 4, s + 5,
+                    s + 6, bd);
+#else
     const int8_t flat2 = highbd_flat_mask5(1, s[-8], s[-7], s[-6], s[-5], p0,
                                            q0, s[4], s[5], s[6], s[7], bd);
 
     highbd_filter16(mask, *thresh, flat, flat2, s - 8, s - 7, s - 6, s - 5,
                     s - 4, s - 3, s - 2, s - 1, s, s + 1, s + 2, s + 3, s + 4,
                     s + 5, s + 6, s + 7, bd);
+#endif
     s += p;
   }
 }