Migrate loop filter functions from vp9/ to vpx_dsp/

The various tap loop filter operations are common functions across
codec. This commit moves them along with SIMD optimizations to
vpx_dsp folder.

Change-Id: Ia5fa0b2e5289cdb98467502a549c380b9c60e92c

4 files changed, 0 insertions, 4399 deletions

diff --git a/vp9/common/x86/vp9_high_loopfilter_intrin_sse2.c b/vp9/common/x86/vp9_high_loopfilter_intrin_sse2.c
deleted file mode 100644
index b40669c63..000000000
--- a/vp9/common/x86/vp9_high_loopfilter_intrin_sse2.c
+++ /dev/null
@@ -1,1215 +0,0 @@
-/*
- *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <emmintrin.h>  // SSE2
-
-#include "./vp9_rtcd.h"
-#include "vpx_ports/mem.h"
-#include "vp9/common/vp9_loopfilter.h"
-#include "vpx_ports/emmintrin_compat.h"
-
-static INLINE __m128i signed_char_clamp_bd_sse2(__m128i value, int bd) {
-  __m128i ubounded;
-  __m128i lbounded;
-  __m128i retval;
-
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i one = _mm_set1_epi16(1);
-  __m128i t80, max, min;
-
-  if (bd == 8) {
-    t80 = _mm_set1_epi16(0x80);
-    max = _mm_subs_epi16(
-              _mm_subs_epi16(_mm_slli_epi16(one, 8), one), t80);
-  } else if (bd == 10) {
-    t80 = _mm_set1_epi16(0x200);
-    max = _mm_subs_epi16(
-              _mm_subs_epi16(_mm_slli_epi16(one, 10), one), t80);
-  } else {  // bd == 12
-    t80 = _mm_set1_epi16(0x800);
-    max = _mm_subs_epi16(
-              _mm_subs_epi16(_mm_slli_epi16(one, 12), one), t80);
-  }
-
-  min = _mm_subs_epi16(zero, t80);
-
-  ubounded = _mm_cmpgt_epi16(value, max);
-  lbounded = _mm_cmplt_epi16(value, min);
-  retval = _mm_andnot_si128(_mm_or_si128(ubounded, lbounded), value);
-  ubounded = _mm_and_si128(ubounded, max);
-  lbounded = _mm_and_si128(lbounded, min);
-  retval = _mm_or_si128(retval, ubounded);
-  retval = _mm_or_si128(retval, lbounded);
-  return retval;
-}
-
-// TODO(debargha, peter): Break up large functions into smaller ones
-// in this file.
-static void highbd_mb_lpf_horizontal_edge_w_sse2_8(uint16_t *s,
-                                                   int p,
-                                                   const uint8_t *_blimit,
-                                                   const uint8_t *_limit,
-                                                   const uint8_t *_thresh,
-                                                   int bd) {
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i one = _mm_set1_epi16(1);
-  __m128i blimit, limit, thresh;
-  __m128i q7, p7, q6, p6, q5, p5, q4, p4, q3, p3, q2, p2, q1, p1, q0, p0;
-  __m128i mask, hev, flat, flat2, abs_p1p0, abs_q1q0;
-  __m128i ps1, qs1, ps0, qs0;
-  __m128i abs_p0q0, abs_p1q1, ffff, work;
-  __m128i filt, work_a, filter1, filter2;
-  __m128i flat2_q6, flat2_p6, flat2_q5, flat2_p5, flat2_q4, flat2_p4;
-  __m128i flat2_q3, flat2_p3, flat2_q2, flat2_p2, flat2_q1, flat2_p1;
-  __m128i flat2_q0, flat2_p0;
-  __m128i flat_q2, flat_p2, flat_q1, flat_p1, flat_q0, flat_p0;
-  __m128i pixelFilter_p, pixelFilter_q;
-  __m128i pixetFilter_p2p1p0, pixetFilter_q2q1q0;
-  __m128i sum_p7, sum_q7, sum_p3, sum_q3;
-  __m128i t4, t3, t80, t1;
-  __m128i eight, four;
-
-  if (bd == 8) {
-    blimit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero);
-    limit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero);
-    thresh = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero);
-  } else if (bd == 10) {
-    blimit = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero), 2);
-    limit = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero), 2);
-    thresh = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero), 2);
-  } else {  // bd == 12
-    blimit = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero), 4);
-    limit = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero), 4);
-    thresh = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero), 4);
-  }
-
-  q4 = _mm_load_si128((__m128i *)(s + 4 * p));
-  p4 = _mm_load_si128((__m128i *)(s - 5 * p));
-  q3 = _mm_load_si128((__m128i *)(s + 3 * p));
-  p3 = _mm_load_si128((__m128i *)(s - 4 * p));
-  q2 = _mm_load_si128((__m128i *)(s + 2 * p));
-  p2 = _mm_load_si128((__m128i *)(s - 3 * p));
-  q1 = _mm_load_si128((__m128i *)(s + 1 * p));
-  p1 = _mm_load_si128((__m128i *)(s - 2 * p));
-  q0 = _mm_load_si128((__m128i *)(s + 0 * p));
-  p0 = _mm_load_si128((__m128i *)(s - 1 * p));
-
-  //  highbd_filter_mask
-  abs_p1p0 = _mm_or_si128(_mm_subs_epu16(p1, p0), _mm_subs_epu16(p0, p1));
-  abs_q1q0 = _mm_or_si128(_mm_subs_epu16(q1, q0), _mm_subs_epu16(q0, q1));
-
-  ffff = _mm_cmpeq_epi16(abs_p1p0, abs_p1p0);
-
-  abs_p0q0 = _mm_or_si128(_mm_subs_epu16(p0, q0), _mm_subs_epu16(q0, p0));
-  abs_p1q1 = _mm_or_si128(_mm_subs_epu16(p1, q1), _mm_subs_epu16(q1, p1));
-
-  //  highbd_hev_mask (in C code this is actually called from highbd_filter4)
-  flat = _mm_max_epi16(abs_p1p0, abs_q1q0);
-  hev = _mm_subs_epu16(flat, thresh);
-  hev = _mm_xor_si128(_mm_cmpeq_epi16(hev, zero), ffff);
-
-  abs_p0q0 =_mm_adds_epu16(abs_p0q0, abs_p0q0);  // abs(p0 - q0) * 2
-  abs_p1q1 = _mm_srli_epi16(abs_p1q1, 1);  // abs(p1 - q1) / 2
-  mask = _mm_subs_epu16(_mm_adds_epu16(abs_p0q0, abs_p1q1), blimit);
-  mask = _mm_xor_si128(_mm_cmpeq_epi16(mask, zero), ffff);
-  mask = _mm_and_si128(mask, _mm_adds_epu16(limit, one));
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p1, p0),
-                                    _mm_subs_epu16(p0, p1)),
-                       _mm_or_si128(_mm_subs_epu16(q1, q0),
-                                    _mm_subs_epu16(q0, q1)));
-  mask = _mm_max_epi16(work, mask);
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p1),
-                                    _mm_subs_epu16(p1, p2)),
-                       _mm_or_si128(_mm_subs_epu16(q2, q1),
-                                    _mm_subs_epu16(q1, q2)));
-  mask = _mm_max_epi16(work, mask);
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p3, p2),
-                                    _mm_subs_epu16(p2, p3)),
-                       _mm_or_si128(_mm_subs_epu16(q3, q2),
-                                    _mm_subs_epu16(q2, q3)));
-  mask = _mm_max_epi16(work, mask);
-
-  mask = _mm_subs_epu16(mask, limit);
-  mask = _mm_cmpeq_epi16(mask, zero);  // return ~mask
-
-  // lp filter
-  // highbd_filter4
-  t4 = _mm_set1_epi16(4);
-  t3 = _mm_set1_epi16(3);
-  if (bd == 8)
-    t80 = _mm_set1_epi16(0x80);
-  else if (bd == 10)
-    t80 = _mm_set1_epi16(0x200);
-  else  // bd == 12
-    t80 = _mm_set1_epi16(0x800);
-
-  t1 = _mm_set1_epi16(0x1);
-
-  ps1 = _mm_subs_epi16(p1, t80);
-  qs1 = _mm_subs_epi16(q1, t80);
-  ps0 = _mm_subs_epi16(p0, t80);
-  qs0 = _mm_subs_epi16(q0, t80);
-
-  filt = _mm_and_si128(
-      signed_char_clamp_bd_sse2(_mm_subs_epi16(ps1, qs1), bd), hev);
-  work_a = _mm_subs_epi16(qs0, ps0);
-  filt = _mm_adds_epi16(filt, work_a);
-  filt = _mm_adds_epi16(filt, work_a);
-  filt = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, work_a), bd);
-  filt = _mm_and_si128(filt, mask);
-  filter1 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t4), bd);
-  filter2 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t3), bd);
-
-  // Filter1 >> 3
-  filter1 = _mm_srai_epi16(filter1, 0x3);
-  filter2 = _mm_srai_epi16(filter2, 0x3);
-
-  qs0 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs0, filter1), bd),
-      t80);
-  ps0 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps0, filter2), bd),
-      t80);
-  filt = _mm_adds_epi16(filter1, t1);
-  filt = _mm_srai_epi16(filt, 1);
-  filt = _mm_andnot_si128(hev, filt);
-  qs1 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs1, filt), bd),
-      t80);
-  ps1 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps1, filt), bd),
-      t80);
-
-  // end highbd_filter4
-  // loopfilter done
-
-  // highbd_flat_mask4
-  flat = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p0),
-                                    _mm_subs_epu16(p0, p2)),
-                       _mm_or_si128(_mm_subs_epu16(p3, p0),
-                                    _mm_subs_epu16(p0, p3)));
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(q2, q0),
-                                    _mm_subs_epu16(q0, q2)),
-                       _mm_or_si128(_mm_subs_epu16(q3, q0),
-                                    _mm_subs_epu16(q0, q3)));
-  flat = _mm_max_epi16(work, flat);
-  work = _mm_max_epi16(abs_p1p0, abs_q1q0);
-  flat = _mm_max_epi16(work, flat);
-
-  if (bd == 8)
-    flat = _mm_subs_epu16(flat, one);
-  else if (bd == 10)
-    flat = _mm_subs_epu16(flat, _mm_slli_epi16(one, 2));
-  else  // bd == 12
-    flat = _mm_subs_epu16(flat, _mm_slli_epi16(one, 4));
-
-  flat = _mm_cmpeq_epi16(flat, zero);
-  // end flat_mask4
-
-  // flat & mask = flat && mask (as used in filter8)
-  // (because, in both vars, each block of 16 either all 1s or all 0s)
-  flat = _mm_and_si128(flat, mask);
-
-  p5 = _mm_load_si128((__m128i *)(s - 6 * p));
-  q5 = _mm_load_si128((__m128i *)(s + 5 * p));
-  p6 = _mm_load_si128((__m128i *)(s - 7 * p));
-  q6 = _mm_load_si128((__m128i *)(s + 6 * p));
-  p7 = _mm_load_si128((__m128i *)(s - 8 * p));
-  q7 = _mm_load_si128((__m128i *)(s + 7 * p));
-
-  // highbd_flat_mask5 (arguments passed in are p0, q0, p4-p7, q4-q7
-  // but referred to as p0-p4 & q0-q4 in fn)
-  flat2 = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p4, p0),
-                                     _mm_subs_epu16(p0, p4)),
-                        _mm_or_si128(_mm_subs_epu16(q4, q0),
-                                     _mm_subs_epu16(q0, q4)));
-
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p5, p0),
-                                    _mm_subs_epu16(p0, p5)),
-                       _mm_or_si128(_mm_subs_epu16(q5, q0),
-                                    _mm_subs_epu16(q0, q5)));
-  flat2 = _mm_max_epi16(work, flat2);
-
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p6, p0),
-                                    _mm_subs_epu16(p0, p6)),
-                       _mm_or_si128(_mm_subs_epu16(q6, q0),
-                                    _mm_subs_epu16(q0, q6)));
-  flat2 = _mm_max_epi16(work, flat2);
-
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p7, p0),
-                                    _mm_subs_epu16(p0, p7)),
-                       _mm_or_si128(_mm_subs_epu16(q7, q0),
-                                    _mm_subs_epu16(q0, q7)));
-  flat2 = _mm_max_epi16(work, flat2);
-
-  if (bd == 8)
-    flat2 = _mm_subs_epu16(flat2, one);
-  else if (bd == 10)
-    flat2 = _mm_subs_epu16(flat2, _mm_slli_epi16(one, 2));
-  else  // bd == 12
-    flat2 = _mm_subs_epu16(flat2, _mm_slli_epi16(one, 4));
-
-  flat2 = _mm_cmpeq_epi16(flat2, zero);
-  flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
-  // end highbd_flat_mask5
-
-  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-  // flat and wide flat calculations
-  eight = _mm_set1_epi16(8);
-  four = _mm_set1_epi16(4);
-
-  pixelFilter_p = _mm_add_epi16(_mm_add_epi16(p6, p5),
-                                _mm_add_epi16(p4, p3));
-  pixelFilter_q = _mm_add_epi16(_mm_add_epi16(q6, q5),
-                                _mm_add_epi16(q4, q3));
-
-  pixetFilter_p2p1p0 = _mm_add_epi16(p0, _mm_add_epi16(p2, p1));
-  pixelFilter_p = _mm_add_epi16(pixelFilter_p, pixetFilter_p2p1p0);
-
-  pixetFilter_q2q1q0 = _mm_add_epi16(q0, _mm_add_epi16(q2, q1));
-  pixelFilter_q = _mm_add_epi16(pixelFilter_q, pixetFilter_q2q1q0);
-  pixelFilter_p = _mm_add_epi16(eight, _mm_add_epi16(pixelFilter_p,
-                                                      pixelFilter_q));
-  pixetFilter_p2p1p0 =   _mm_add_epi16(four,
-                                       _mm_add_epi16(pixetFilter_p2p1p0,
-                                                     pixetFilter_q2q1q0));
-  flat2_p0 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                          _mm_add_epi16(p7, p0)), 4);
-  flat2_q0 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                          _mm_add_epi16(q7, q0)), 4);
-  flat_p0 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                                         _mm_add_epi16(p3, p0)), 3);
-  flat_q0 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                                         _mm_add_epi16(q3, q0)), 3);
-
-  sum_p7 = _mm_add_epi16(p7, p7);
-  sum_q7 = _mm_add_epi16(q7, q7);
-  sum_p3 = _mm_add_epi16(p3, p3);
-  sum_q3 = _mm_add_epi16(q3, q3);
-
-  pixelFilter_q = _mm_sub_epi16(pixelFilter_p, p6);
-  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q6);
-  flat2_p1 = _mm_srli_epi16(
-      _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p1)), 4);
-  flat2_q1 = _mm_srli_epi16(
-      _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q1)), 4);
-
-  pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_p2p1p0, p2);
-  pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q2);
-  flat_p1 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                                         _mm_add_epi16(sum_p3, p1)), 3);
-  flat_q1 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0,
-                                         _mm_add_epi16(sum_q3, q1)), 3);
-
-  sum_p7 = _mm_add_epi16(sum_p7, p7);
-  sum_q7 = _mm_add_epi16(sum_q7, q7);
-  sum_p3 = _mm_add_epi16(sum_p3, p3);
-  sum_q3 = _mm_add_epi16(sum_q3, q3);
-
-  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q5);
-  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p5);
-  flat2_p2 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                          _mm_add_epi16(sum_p7, p2)), 4);
-  flat2_q2 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                                          _mm_add_epi16(sum_q7, q2)), 4);
-
-  pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q1);
-  pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_q2q1q0, p1);
-  flat_p2 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                                         _mm_add_epi16(sum_p3, p2)), 3);
-  flat_q2 = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0,
-                                         _mm_add_epi16(sum_q3, q2)), 3);
-
-  sum_p7 = _mm_add_epi16(sum_p7, p7);
-  sum_q7 = _mm_add_epi16(sum_q7, q7);
-  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q4);
-  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p4);
-  flat2_p3 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                          _mm_add_epi16(sum_p7, p3)), 4);
-  flat2_q3 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                                          _mm_add_epi16(sum_q7, q3)), 4);
-
-  sum_p7 = _mm_add_epi16(sum_p7, p7);
-  sum_q7 = _mm_add_epi16(sum_q7, q7);
-  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q3);
-  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p3);
-  flat2_p4 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                          _mm_add_epi16(sum_p7, p4)), 4);
-  flat2_q4 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                                          _mm_add_epi16(sum_q7, q4)), 4);
-
-  sum_p7 = _mm_add_epi16(sum_p7, p7);
-  sum_q7 = _mm_add_epi16(sum_q7, q7);
-  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q2);
-  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p2);
-  flat2_p5 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                          _mm_add_epi16(sum_p7, p5)), 4);
-  flat2_q5 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                                          _mm_add_epi16(sum_q7, q5)), 4);
-
-  sum_p7 = _mm_add_epi16(sum_p7, p7);
-  sum_q7 = _mm_add_epi16(sum_q7, q7);
-  pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q1);
-  pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p1);
-  flat2_p6 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                          _mm_add_epi16(sum_p7, p6)), 4);
-  flat2_q6 = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                                          _mm_add_epi16(sum_q7, q6)), 4);
-
-  //  wide flat
-  //  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-  //  highbd_filter8
-  p2 = _mm_andnot_si128(flat, p2);
-  //  p2 remains unchanged if !(flat && mask)
-  flat_p2 = _mm_and_si128(flat, flat_p2);
-  //  when (flat && mask)
-  p2 = _mm_or_si128(p2, flat_p2);  // full list of p2 values
-  q2 = _mm_andnot_si128(flat, q2);
-  flat_q2 = _mm_and_si128(flat, flat_q2);
-  q2 = _mm_or_si128(q2, flat_q2);  // full list of q2 values
-
-  ps1 = _mm_andnot_si128(flat, ps1);
-  //  p1 takes the value assigned to in in filter4 if !(flat && mask)
-  flat_p1 = _mm_and_si128(flat, flat_p1);
-  //  when (flat && mask)
-  p1 = _mm_or_si128(ps1, flat_p1);  // full list of p1 values
-  qs1 = _mm_andnot_si128(flat, qs1);
-  flat_q1 = _mm_and_si128(flat, flat_q1);
-  q1 = _mm_or_si128(qs1, flat_q1);  // full list of q1 values
-
-  ps0 = _mm_andnot_si128(flat, ps0);
-  //  p0 takes the value assigned to in in filter4 if !(flat && mask)
-  flat_p0 = _mm_and_si128(flat, flat_p0);
-  //  when (flat && mask)
-  p0 = _mm_or_si128(ps0, flat_p0);  // full list of p0 values
-  qs0 = _mm_andnot_si128(flat, qs0);
-  flat_q0 = _mm_and_si128(flat, flat_q0);
-  q0 = _mm_or_si128(qs0, flat_q0);  // full list of q0 values
-  // end highbd_filter8
-
-  // highbd_filter16
-  p6 = _mm_andnot_si128(flat2, p6);
-  //  p6 remains unchanged if !(flat2 && flat && mask)
-  flat2_p6 = _mm_and_si128(flat2, flat2_p6);
-  //  get values for when (flat2 && flat && mask)
-  p6 = _mm_or_si128(p6, flat2_p6);  // full list of p6 values
-  q6 = _mm_andnot_si128(flat2, q6);
-  //  q6 remains unchanged if !(flat2 && flat && mask)
-  flat2_q6 = _mm_and_si128(flat2, flat2_q6);
-  //  get values for when (flat2 && flat && mask)
-  q6 = _mm_or_si128(q6, flat2_q6);  // full list of q6 values
-  _mm_store_si128((__m128i *)(s - 7 * p), p6);
-  _mm_store_si128((__m128i *)(s + 6 * p), q6);
-
-  p5 = _mm_andnot_si128(flat2, p5);
-  //  p5 remains unchanged if !(flat2 && flat && mask)
-  flat2_p5 = _mm_and_si128(flat2, flat2_p5);
-  //  get values for when (flat2 && flat && mask)
-  p5 = _mm_or_si128(p5, flat2_p5);
-  //  full list of p5 values
-  q5 = _mm_andnot_si128(flat2, q5);
-  //  q5 remains unchanged if !(flat2 && flat && mask)
-  flat2_q5 = _mm_and_si128(flat2, flat2_q5);
-  //  get values for when (flat2 && flat && mask)
-  q5 = _mm_or_si128(q5, flat2_q5);
-  //  full list of q5 values
-  _mm_store_si128((__m128i *)(s - 6 * p), p5);
-  _mm_store_si128((__m128i *)(s + 5 * p), q5);
-
-  p4 = _mm_andnot_si128(flat2, p4);
-  //  p4 remains unchanged if !(flat2 && flat && mask)
-  flat2_p4 = _mm_and_si128(flat2, flat2_p4);
-  //  get values for when (flat2 && flat && mask)
-  p4 = _mm_or_si128(p4, flat2_p4);  // full list of p4 values
-  q4 = _mm_andnot_si128(flat2, q4);
-  //  q4 remains unchanged if !(flat2 && flat && mask)
-  flat2_q4 = _mm_and_si128(flat2, flat2_q4);
-  //  get values for when (flat2 && flat && mask)
-  q4 = _mm_or_si128(q4, flat2_q4);  // full list of q4 values
-  _mm_store_si128((__m128i *)(s - 5 * p), p4);
-  _mm_store_si128((__m128i *)(s + 4 * p), q4);
-
-  p3 = _mm_andnot_si128(flat2, p3);
-  //  p3 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_p3 = _mm_and_si128(flat2, flat2_p3);
-  //  get values for when (flat2 && flat && mask)
-  p3 = _mm_or_si128(p3, flat2_p3);  // full list of p3 values
-  q3 = _mm_andnot_si128(flat2, q3);
-  //  q3 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_q3 = _mm_and_si128(flat2, flat2_q3);
-  //  get values for when (flat2 && flat && mask)
-  q3 = _mm_or_si128(q3, flat2_q3);  // full list of q3 values
-  _mm_store_si128((__m128i *)(s - 4 * p), p3);
-  _mm_store_si128((__m128i *)(s + 3 * p), q3);
-
-  p2 = _mm_andnot_si128(flat2, p2);
-  //  p2 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_p2 = _mm_and_si128(flat2, flat2_p2);
-  //  get values for when (flat2 && flat && mask)
-  p2 = _mm_or_si128(p2, flat2_p2);
-  //  full list of p2 values
-  q2 = _mm_andnot_si128(flat2, q2);
-  //  q2 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_q2 = _mm_and_si128(flat2, flat2_q2);
-  //  get values for when (flat2 && flat && mask)
-  q2 = _mm_or_si128(q2, flat2_q2);  // full list of q2 values
-  _mm_store_si128((__m128i *)(s - 3 * p), p2);
-  _mm_store_si128((__m128i *)(s + 2 * p), q2);
-
-  p1 = _mm_andnot_si128(flat2, p1);
-  //  p1 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_p1 = _mm_and_si128(flat2, flat2_p1);
-  //  get values for when (flat2 && flat && mask)
-  p1 = _mm_or_si128(p1, flat2_p1);  // full list of p1 values
-  q1 = _mm_andnot_si128(flat2, q1);
-  //  q1 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_q1 = _mm_and_si128(flat2, flat2_q1);
-  //  get values for when (flat2 && flat && mask)
-  q1 = _mm_or_si128(q1, flat2_q1);  // full list of q1 values
-  _mm_store_si128((__m128i *)(s - 2 * p), p1);
-  _mm_store_si128((__m128i *)(s + 1 * p), q1);
-
-  p0 = _mm_andnot_si128(flat2, p0);
-  //  p0 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_p0 = _mm_and_si128(flat2, flat2_p0);
-  //  get values for when (flat2 && flat && mask)
-  p0 = _mm_or_si128(p0, flat2_p0);  // full list of p0 values
-  q0 = _mm_andnot_si128(flat2, q0);
-  //  q0 takes value from highbd_filter8 if !(flat2 && flat && mask)
-  flat2_q0 = _mm_and_si128(flat2, flat2_q0);
-  //  get values for when (flat2 && flat && mask)
-  q0 = _mm_or_si128(q0, flat2_q0);  // full list of q0 values
-  _mm_store_si128((__m128i *)(s - 1 * p), p0);
-  _mm_store_si128((__m128i *)(s - 0 * p), q0);
-}
-
-static void highbd_mb_lpf_horizontal_edge_w_sse2_16(uint16_t *s,
-                                                    int p,
-                                                    const uint8_t *_blimit,
-                                                    const uint8_t *_limit,
-                                                    const uint8_t *_thresh,
-                                                    int bd) {
-  highbd_mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh, bd);
-  highbd_mb_lpf_horizontal_edge_w_sse2_8(s + 8, p, _blimit, _limit, _thresh,
-                                         bd);
-}
-
-// TODO(yunqingwang): remove count and call these 2 functions(8 or 16) directly.
-void vp9_highbd_lpf_horizontal_16_sse2(uint16_t *s, int p,
-                                       const uint8_t *_blimit,
-                                       const uint8_t *_limit,
-                                       const uint8_t *_thresh,
-                                       int count, int bd) {
-  if (count == 1)
-    highbd_mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh, bd);
-  else
-    highbd_mb_lpf_horizontal_edge_w_sse2_16(s, p, _blimit, _limit, _thresh, bd);
-}
-
-void vp9_highbd_lpf_horizontal_8_sse2(uint16_t *s, int p,
-                                      const uint8_t *_blimit,
-                                      const uint8_t *_limit,
-                                      const uint8_t *_thresh,
-                                      int count, int bd) {
-  DECLARE_ALIGNED(16, uint16_t, flat_op2[16]);
-  DECLARE_ALIGNED(16, uint16_t, flat_op1[16]);
-  DECLARE_ALIGNED(16, uint16_t, flat_op0[16]);
-  DECLARE_ALIGNED(16, uint16_t, flat_oq2[16]);
-  DECLARE_ALIGNED(16, uint16_t, flat_oq1[16]);
-  DECLARE_ALIGNED(16, uint16_t, flat_oq0[16]);
-  const __m128i zero = _mm_set1_epi16(0);
-  __m128i blimit, limit, thresh;
-  __m128i mask, hev, flat;
-  __m128i p3 = _mm_load_si128((__m128i *)(s - 4 * p));
-  __m128i q3 = _mm_load_si128((__m128i *)(s + 3 * p));
-  __m128i p2 = _mm_load_si128((__m128i *)(s - 3 * p));
-  __m128i q2 = _mm_load_si128((__m128i *)(s + 2 * p));
-  __m128i p1 = _mm_load_si128((__m128i *)(s - 2 * p));
-  __m128i q1 = _mm_load_si128((__m128i *)(s + 1 * p));
-  __m128i p0 = _mm_load_si128((__m128i *)(s - 1 * p));
-  __m128i q0 = _mm_load_si128((__m128i *)(s + 0 * p));
-  const __m128i one = _mm_set1_epi16(1);
-  const __m128i ffff = _mm_cmpeq_epi16(one, one);
-  __m128i abs_p1q1, abs_p0q0, abs_q1q0, abs_p1p0, work;
-  const __m128i four = _mm_set1_epi16(4);
-  __m128i workp_a, workp_b, workp_shft;
-
-  const __m128i t4 = _mm_set1_epi16(4);
-  const __m128i t3 = _mm_set1_epi16(3);
-  __m128i t80;
-  const __m128i t1 = _mm_set1_epi16(0x1);
-  __m128i ps1, ps0, qs0, qs1;
-  __m128i filt;
-  __m128i work_a;
-  __m128i filter1, filter2;
-
-  (void)count;
-
-  if (bd == 8) {
-    blimit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero);
-    limit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero);
-    thresh = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero);
-    t80 = _mm_set1_epi16(0x80);
-  } else if (bd == 10) {
-    blimit = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero), 2);
-    limit = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero), 2);
-    thresh = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero), 2);
-    t80 = _mm_set1_epi16(0x200);
-  } else {  // bd == 12
-    blimit = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero), 4);
-    limit = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero), 4);
-    thresh = _mm_slli_epi16(
-          _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero), 4);
-    t80 = _mm_set1_epi16(0x800);
-  }
-
-  ps1 = _mm_subs_epi16(p1, t80);
-  ps0 = _mm_subs_epi16(p0, t80);
-  qs0 = _mm_subs_epi16(q0, t80);
-  qs1 = _mm_subs_epi16(q1, t80);
-
-  // filter_mask and hev_mask
-  abs_p1p0 = _mm_or_si128(_mm_subs_epu16(p1, p0),
-                          _mm_subs_epu16(p0, p1));
-  abs_q1q0 = _mm_or_si128(_mm_subs_epu16(q1, q0),
-                          _mm_subs_epu16(q0, q1));
-
-  abs_p0q0 = _mm_or_si128(_mm_subs_epu16(p0, q0),
-                          _mm_subs_epu16(q0, p0));
-  abs_p1q1 = _mm_or_si128(_mm_subs_epu16(p1, q1),
-                          _mm_subs_epu16(q1, p1));
-  flat = _mm_max_epi16(abs_p1p0, abs_q1q0);
-  hev = _mm_subs_epu16(flat, thresh);
-  hev = _mm_xor_si128(_mm_cmpeq_epi16(hev, zero), ffff);
-
-  abs_p0q0 =_mm_adds_epu16(abs_p0q0, abs_p0q0);
-  abs_p1q1 = _mm_srli_epi16(abs_p1q1, 1);
-  mask = _mm_subs_epu16(_mm_adds_epu16(abs_p0q0, abs_p1q1), blimit);
-  mask = _mm_xor_si128(_mm_cmpeq_epi16(mask, zero), ffff);
-  // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-  // So taking maximums continues to work:
-  mask = _mm_and_si128(mask, _mm_adds_epu16(limit, one));
-  mask = _mm_max_epi16(abs_p1p0, mask);
-  // mask |= (abs(p1 - p0) > limit) * -1;
-  mask = _mm_max_epi16(abs_q1q0, mask);
-  // mask |= (abs(q1 - q0) > limit) * -1;
-
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p1),
-                                    _mm_subs_epu16(p1, p2)),
-                       _mm_or_si128(_mm_subs_epu16(q2, q1),
-                                    _mm_subs_epu16(q1, q2)));
-  mask = _mm_max_epi16(work, mask);
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p3, p2),
-                                    _mm_subs_epu16(p2, p3)),
-                       _mm_or_si128(_mm_subs_epu16(q3, q2),
-                                    _mm_subs_epu16(q2, q3)));
-  mask = _mm_max_epi16(work, mask);
-  mask = _mm_subs_epu16(mask, limit);
-  mask = _mm_cmpeq_epi16(mask, zero);
-
-  // flat_mask4
-  flat = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p0),
-                                    _mm_subs_epu16(p0, p2)),
-                       _mm_or_si128(_mm_subs_epu16(q2, q0),
-                                    _mm_subs_epu16(q0, q2)));
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p3, p0),
-                                    _mm_subs_epu16(p0, p3)),
-                       _mm_or_si128(_mm_subs_epu16(q3, q0),
-                                    _mm_subs_epu16(q0, q3)));
-  flat = _mm_max_epi16(work, flat);
-  flat = _mm_max_epi16(abs_p1p0, flat);
-  flat = _mm_max_epi16(abs_q1q0, flat);
-
-  if (bd == 8)
-    flat = _mm_subs_epu16(flat, one);
-  else if (bd == 10)
-    flat = _mm_subs_epu16(flat, _mm_slli_epi16(one, 2));
-  else  // bd == 12
-    flat = _mm_subs_epu16(flat, _mm_slli_epi16(one, 4));
-
-  flat = _mm_cmpeq_epi16(flat, zero);
-  flat = _mm_and_si128(flat, mask);  // flat & mask
-
-  // Added before shift for rounding part of ROUND_POWER_OF_TWO
-
-  workp_a = _mm_add_epi16(_mm_add_epi16(p3, p3), _mm_add_epi16(p2, p1));
-  workp_a = _mm_add_epi16(_mm_add_epi16(workp_a, four), p0);
-  workp_b = _mm_add_epi16(_mm_add_epi16(q0, p2), p3);
-  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-  _mm_store_si128((__m128i *)&flat_op2[0], workp_shft);
-
-  workp_b = _mm_add_epi16(_mm_add_epi16(q0, q1), p1);
-  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-  _mm_store_si128((__m128i *)&flat_op1[0], workp_shft);
-
-  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q2);
-  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p1), p0);
-  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-  _mm_store_si128((__m128i *)&flat_op0[0], workp_shft);
-
-  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q3);
-  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p0), q0);
-  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-  _mm_store_si128((__m128i *)&flat_oq0[0], workp_shft);
-
-  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p2), q3);
-  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q0), q1);
-  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-  _mm_store_si128((__m128i *)&flat_oq1[0], workp_shft);
-
-  workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p1), q3);
-  workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q1), q2);
-  workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-  _mm_store_si128((__m128i *)&flat_oq2[0], workp_shft);
-
-  // lp filter
-  filt = signed_char_clamp_bd_sse2(_mm_subs_epi16(ps1, qs1), bd);
-  filt = _mm_and_si128(filt, hev);
-  work_a = _mm_subs_epi16(qs0, ps0);
-  filt = _mm_adds_epi16(filt, work_a);
-  filt = _mm_adds_epi16(filt, work_a);
-  filt = _mm_adds_epi16(filt, work_a);
-  // (vp9_filter + 3 * (qs0 - ps0)) & mask
-  filt = signed_char_clamp_bd_sse2(filt, bd);
-  filt = _mm_and_si128(filt, mask);
-
-  filter1 = _mm_adds_epi16(filt, t4);
-  filter2 = _mm_adds_epi16(filt, t3);
-
-  // Filter1 >> 3
-  filter1 = signed_char_clamp_bd_sse2(filter1, bd);
-  filter1 = _mm_srai_epi16(filter1, 3);
-
-  // Filter2 >> 3
-  filter2 = signed_char_clamp_bd_sse2(filter2, bd);
-  filter2 = _mm_srai_epi16(filter2, 3);
-
-  // filt >> 1
-  filt = _mm_adds_epi16(filter1, t1);
-  filt = _mm_srai_epi16(filt, 1);
-  // filter = ROUND_POWER_OF_TWO(filter1, 1) & ~hev;
-  filt = _mm_andnot_si128(hev, filt);
-
-  work_a = signed_char_clamp_bd_sse2(_mm_subs_epi16(qs0, filter1), bd);
-  work_a = _mm_adds_epi16(work_a, t80);
-  q0 = _mm_load_si128((__m128i *)flat_oq0);
-  work_a = _mm_andnot_si128(flat, work_a);
-  q0 = _mm_and_si128(flat, q0);
-  q0 = _mm_or_si128(work_a, q0);
-
-  work_a = signed_char_clamp_bd_sse2(_mm_subs_epi16(qs1, filt), bd);
-  work_a = _mm_adds_epi16(work_a, t80);
-  q1 = _mm_load_si128((__m128i *)flat_oq1);
-  work_a = _mm_andnot_si128(flat, work_a);
-  q1 = _mm_and_si128(flat, q1);
-  q1 = _mm_or_si128(work_a, q1);
-
-  work_a = _mm_loadu_si128((__m128i *)(s + 2 * p));
-  q2 = _mm_load_si128((__m128i *)flat_oq2);
-  work_a = _mm_andnot_si128(flat, work_a);
-  q2 = _mm_and_si128(flat, q2);
-  q2 = _mm_or_si128(work_a, q2);
-
-  work_a = signed_char_clamp_bd_sse2(_mm_adds_epi16(ps0, filter2), bd);
-  work_a = _mm_adds_epi16(work_a, t80);
-  p0 = _mm_load_si128((__m128i *)flat_op0);
-  work_a = _mm_andnot_si128(flat, work_a);
-  p0 = _mm_and_si128(flat, p0);
-  p0 = _mm_or_si128(work_a, p0);
-
-  work_a = signed_char_clamp_bd_sse2(_mm_adds_epi16(ps1, filt), bd);
-  work_a = _mm_adds_epi16(work_a, t80);
-  p1 = _mm_load_si128((__m128i *)flat_op1);
-  work_a = _mm_andnot_si128(flat, work_a);
-  p1 = _mm_and_si128(flat, p1);
-  p1 = _mm_or_si128(work_a, p1);
-
-  work_a = _mm_loadu_si128((__m128i *)(s - 3 * p));
-  p2 = _mm_load_si128((__m128i *)flat_op2);
-  work_a = _mm_andnot_si128(flat, work_a);
-  p2 = _mm_and_si128(flat, p2);
-  p2 = _mm_or_si128(work_a, p2);
-
-  _mm_store_si128((__m128i *)(s - 3 * p), p2);
-  _mm_store_si128((__m128i *)(s - 2 * p), p1);
-  _mm_store_si128((__m128i *)(s - 1 * p), p0);
-  _mm_store_si128((__m128i *)(s + 0 * p), q0);
-  _mm_store_si128((__m128i *)(s + 1 * p), q1);
-  _mm_store_si128((__m128i *)(s + 2 * p), q2);
-}
-
-void vp9_highbd_lpf_horizontal_8_dual_sse2(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,
-                                           const uint8_t *_thresh1,
-                                           int bd) {
-  vp9_highbd_lpf_horizontal_8_sse2(s, p, _blimit0, _limit0, _thresh0, 1, bd);
-  vp9_highbd_lpf_horizontal_8_sse2(s + 8, p, _blimit1, _limit1, _thresh1,
-                                   1, bd);
-}
-
-void vp9_highbd_lpf_horizontal_4_sse2(uint16_t *s, int p,
-                                      const uint8_t *_blimit,
-                                      const uint8_t *_limit,
-                                      const uint8_t *_thresh,
-                                      int count, int bd) {
-  const __m128i zero = _mm_set1_epi16(0);
-  __m128i blimit, limit, thresh;
-  __m128i mask, hev, flat;
-  __m128i p3 = _mm_loadu_si128((__m128i *)(s - 4 * p));
-  __m128i p2 = _mm_loadu_si128((__m128i *)(s - 3 * p));
-  __m128i p1 = _mm_loadu_si128((__m128i *)(s - 2 * p));
-  __m128i p0 = _mm_loadu_si128((__m128i *)(s - 1 * p));
-  __m128i q0 = _mm_loadu_si128((__m128i *)(s - 0 * p));
-  __m128i q1 = _mm_loadu_si128((__m128i *)(s + 1 * p));
-  __m128i q2 = _mm_loadu_si128((__m128i *)(s + 2 * p));
-  __m128i q3 = _mm_loadu_si128((__m128i *)(s + 3 * p));
-  const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu16(p1, p0),
-                                        _mm_subs_epu16(p0, p1));
-  const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu16(q1, q0),
-                                        _mm_subs_epu16(q0, q1));
-  const __m128i ffff = _mm_cmpeq_epi16(abs_p1p0, abs_p1p0);
-  const __m128i one = _mm_set1_epi16(1);
-  __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu16(p0, q0),
-                                  _mm_subs_epu16(q0, p0));
-  __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu16(p1, q1),
-                                  _mm_subs_epu16(q1, p1));
-  __m128i work;
-  const __m128i t4 = _mm_set1_epi16(4);
-  const __m128i t3 = _mm_set1_epi16(3);
-  __m128i t80;
-  __m128i tff80;
-  __m128i tffe0;
-  __m128i t1f;
-  // equivalent to shifting 0x1f left by bitdepth - 8
-  // and setting new bits to 1
-  const __m128i t1 = _mm_set1_epi16(0x1);
-  __m128i t7f;
-  // equivalent to shifting 0x7f left by bitdepth - 8
-  // and setting new bits to 1
-  __m128i ps1, ps0, qs0, qs1;
-  __m128i filt;
-  __m128i work_a;
-  __m128i filter1, filter2;
-
-  (void)count;
-
-  if (bd == 8) {
-    blimit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero);
-    limit = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero);
-    thresh = _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero);
-    t80 = _mm_set1_epi16(0x80);
-    tff80 = _mm_set1_epi16(0xff80);
-    tffe0 = _mm_set1_epi16(0xffe0);
-    t1f = _mm_srli_epi16(_mm_set1_epi16(0x1fff), 8);
-    t7f = _mm_srli_epi16(_mm_set1_epi16(0x7fff), 8);
-  } else if (bd == 10) {
-    blimit = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero), 2);
-    limit = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero), 2);
-    thresh = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero), 2);
-    t80 = _mm_slli_epi16(_mm_set1_epi16(0x80), 2);
-    tff80 = _mm_slli_epi16(_mm_set1_epi16(0xff80), 2);
-    tffe0 = _mm_slli_epi16(_mm_set1_epi16(0xffe0), 2);
-    t1f = _mm_srli_epi16(_mm_set1_epi16(0x1fff), 6);
-    t7f = _mm_srli_epi16(_mm_set1_epi16(0x7fff), 6);
-  } else {  // bd == 12
-    blimit = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_blimit), zero), 4);
-    limit = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_limit), zero), 4);
-    thresh = _mm_slli_epi16(
-        _mm_unpacklo_epi8(_mm_load_si128((const __m128i *)_thresh), zero), 4);
-    t80 = _mm_slli_epi16(_mm_set1_epi16(0x80), 4);
-    tff80 = _mm_slli_epi16(_mm_set1_epi16(0xff80), 4);
-    tffe0 = _mm_slli_epi16(_mm_set1_epi16(0xffe0), 4);
-    t1f = _mm_srli_epi16(_mm_set1_epi16(0x1fff), 4);
-    t7f = _mm_srli_epi16(_mm_set1_epi16(0x7fff), 4);
-  }
-
-  ps1 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s - 2 * p)), t80);
-  ps0 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s - 1 * p)), t80);
-  qs0 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s + 0 * p)), t80);
-  qs1 = _mm_subs_epi16(_mm_loadu_si128((__m128i *)(s + 1 * p)), t80);
-
-  // filter_mask and hev_mask
-  flat = _mm_max_epi16(abs_p1p0, abs_q1q0);
-  hev = _mm_subs_epu16(flat, thresh);
-  hev = _mm_xor_si128(_mm_cmpeq_epi16(hev, zero), ffff);
-
-  abs_p0q0 =_mm_adds_epu16(abs_p0q0, abs_p0q0);
-  abs_p1q1 = _mm_srli_epi16(abs_p1q1, 1);
-  mask = _mm_subs_epu16(_mm_adds_epu16(abs_p0q0, abs_p1q1), blimit);
-  mask = _mm_xor_si128(_mm_cmpeq_epi16(mask, zero), ffff);
-  // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-  // So taking maximums continues to work:
-  mask = _mm_and_si128(mask, _mm_adds_epu16(limit, one));
-  mask = _mm_max_epi16(flat, mask);
-  // mask |= (abs(p1 - p0) > limit) * -1;
-  // mask |= (abs(q1 - q0) > limit) * -1;
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(p2, p1),
-                                    _mm_subs_epu16(p1, p2)),
-                       _mm_or_si128(_mm_subs_epu16(p3, p2),
-                                    _mm_subs_epu16(p2, p3)));
-  mask = _mm_max_epi16(work, mask);
-  work = _mm_max_epi16(_mm_or_si128(_mm_subs_epu16(q2, q1),
-                                    _mm_subs_epu16(q1, q2)),
-                       _mm_or_si128(_mm_subs_epu16(q3, q2),
-                                    _mm_subs_epu16(q2, q3)));
-  mask = _mm_max_epi16(work, mask);
-  mask = _mm_subs_epu16(mask, limit);
-  mask = _mm_cmpeq_epi16(mask, zero);
-
-  // filter4
-  filt = signed_char_clamp_bd_sse2(_mm_subs_epi16(ps1, qs1), bd);
-  filt = _mm_and_si128(filt, hev);
-  work_a = _mm_subs_epi16(qs0, ps0);
-  filt = _mm_adds_epi16(filt, work_a);
-  filt = _mm_adds_epi16(filt, work_a);
-  filt = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, work_a), bd);
-
-  // (vp9_filter + 3 * (qs0 - ps0)) & mask
-  filt = _mm_and_si128(filt, mask);
-
-  filter1 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t4), bd);
-  filter2 = signed_char_clamp_bd_sse2(_mm_adds_epi16(filt, t3), bd);
-
-  // Filter1 >> 3
-  work_a = _mm_cmpgt_epi16(zero, filter1);  // get the values that are <0
-  filter1 = _mm_srli_epi16(filter1, 3);
-  work_a = _mm_and_si128(work_a, tffe0);  // sign bits for the values < 0
-  filter1 = _mm_and_si128(filter1, t1f);  // clamp the range
-  filter1 = _mm_or_si128(filter1, work_a);  // reinsert the sign bits
-
-  // Filter2 >> 3
-  work_a = _mm_cmpgt_epi16(zero, filter2);
-  filter2 = _mm_srli_epi16(filter2, 3);
-  work_a = _mm_and_si128(work_a, tffe0);
-  filter2 = _mm_and_si128(filter2, t1f);
-  filter2 = _mm_or_si128(filter2, work_a);
-
-  // filt >> 1
-  filt = _mm_adds_epi16(filter1, t1);
-  work_a = _mm_cmpgt_epi16(zero, filt);
-  filt = _mm_srli_epi16(filt, 1);
-  work_a = _mm_and_si128(work_a, tff80);
-  filt = _mm_and_si128(filt, t7f);
-  filt = _mm_or_si128(filt, work_a);
-
-  filt = _mm_andnot_si128(hev, filt);
-
-  q0 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs0, filter1), bd), t80);
-  q1 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_subs_epi16(qs1, filt), bd), t80);
-  p0 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps0, filter2), bd), t80);
-  p1 = _mm_adds_epi16(
-      signed_char_clamp_bd_sse2(_mm_adds_epi16(ps1, filt), bd), t80);
-
-  _mm_storeu_si128((__m128i *)(s - 2 * p), p1);
-  _mm_storeu_si128((__m128i *)(s - 1 * p), p0);
-  _mm_storeu_si128((__m128i *)(s + 0 * p), q0);
-  _mm_storeu_si128((__m128i *)(s + 1 * p), q1);
-}
-
-void vp9_highbd_lpf_horizontal_4_dual_sse2(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,
-                                           const uint8_t *_thresh1,
-                                           int bd) {
-  vp9_highbd_lpf_horizontal_4_sse2(s, p, _blimit0, _limit0, _thresh0, 1, bd);
-  vp9_highbd_lpf_horizontal_4_sse2(s + 8, p, _blimit1, _limit1, _thresh1, 1,
-                                   bd);
-}
-
-static INLINE void highbd_transpose(uint16_t *src[], int in_p,
-                                    uint16_t *dst[], int out_p,
-                                    int num_8x8_to_transpose) {
-  int idx8x8 = 0;
-  __m128i p0, p1, p2, p3, p4, p5, p6, p7, x0, x1, x2, x3, x4, x5, x6, x7;
-  do {
-    uint16_t *in = src[idx8x8];
-    uint16_t *out = dst[idx8x8];
-
-    p0 = _mm_loadu_si128((__m128i *)(in + 0*in_p));  // 00 01 02 03 04 05 06 07
-    p1 = _mm_loadu_si128((__m128i *)(in + 1*in_p));  // 10 11 12 13 14 15 16 17
-    p2 = _mm_loadu_si128((__m128i *)(in + 2*in_p));  // 20 21 22 23 24 25 26 27
-    p3 = _mm_loadu_si128((__m128i *)(in + 3*in_p));  // 30 31 32 33 34 35 36 37
-    p4 = _mm_loadu_si128((__m128i *)(in + 4*in_p));  // 40 41 42 43 44 45 46 47
-    p5 = _mm_loadu_si128((__m128i *)(in + 5*in_p));  // 50 51 52 53 54 55 56 57
-    p6 = _mm_loadu_si128((__m128i *)(in + 6*in_p));  // 60 61 62 63 64 65 66 67
-    p7 = _mm_loadu_si128((__m128i *)(in + 7*in_p));  // 70 71 72 73 74 75 76 77
-    // 00 10 01 11 02 12 03 13
-    x0 = _mm_unpacklo_epi16(p0, p1);
-    // 20 30 21 31 22 32 23 33
-    x1 = _mm_unpacklo_epi16(p2, p3);
-    // 40 50 41 51 42 52 43 53
-    x2 = _mm_unpacklo_epi16(p4, p5);
-    // 60 70 61 71 62 72 63 73
-    x3 = _mm_unpacklo_epi16(p6, p7);
-    // 00 10 20 30 01 11 21 31
-    x4 = _mm_unpacklo_epi32(x0, x1);
-    // 40 50 60 70 41 51 61 71
-    x5 = _mm_unpacklo_epi32(x2, x3);
-    // 00 10 20 30 40 50 60 70
-    x6 = _mm_unpacklo_epi64(x4, x5);
-    // 01 11 21 31 41 51 61 71
-    x7 = _mm_unpackhi_epi64(x4, x5);
-
-    _mm_storeu_si128((__m128i *)(out + 0*out_p), x6);
-    // 00 10 20 30 40 50 60 70
-    _mm_storeu_si128((__m128i *)(out + 1*out_p), x7);
-    // 01 11 21 31 41 51 61 71
-
-    // 02 12 22 32 03 13 23 33
-    x4 = _mm_unpackhi_epi32(x0, x1);
-    // 42 52 62 72 43 53 63 73
-    x5 = _mm_unpackhi_epi32(x2, x3);
-    // 02 12 22 32 42 52 62 72
-    x6 = _mm_unpacklo_epi64(x4, x5);
-    // 03 13 23 33 43 53 63 73
-    x7 = _mm_unpackhi_epi64(x4, x5);
-
-    _mm_storeu_si128((__m128i *)(out + 2*out_p), x6);
-    // 02 12 22 32 42 52 62 72
-    _mm_storeu_si128((__m128i *)(out + 3*out_p), x7);
-    // 03 13 23 33 43 53 63 73
-
-    // 04 14 05 15 06 16 07 17
-    x0 = _mm_unpackhi_epi16(p0, p1);
-    // 24 34 25 35 26 36 27 37
-    x1 = _mm_unpackhi_epi16(p2, p3);
-    // 44 54 45 55 46 56 47 57
-    x2 = _mm_unpackhi_epi16(p4, p5);
-    // 64 74 65 75 66 76 67 77
-    x3 = _mm_unpackhi_epi16(p6, p7);
-    // 04 14 24 34 05 15 25 35
-    x4 = _mm_unpacklo_epi32(x0, x1);
-    // 44 54 64 74 45 55 65 75
-    x5 = _mm_unpacklo_epi32(x2, x3);
-    // 04 14 24 34 44 54 64 74
-    x6 = _mm_unpacklo_epi64(x4, x5);
-    // 05 15 25 35 45 55 65 75
-    x7 = _mm_unpackhi_epi64(x4, x5);
-
-    _mm_storeu_si128((__m128i *)(out + 4*out_p), x6);
-    // 04 14 24 34 44 54 64 74
-    _mm_storeu_si128((__m128i *)(out + 5*out_p), x7);
-    // 05 15 25 35 45 55 65 75
-
-    // 06 16 26 36 07 17 27 37
-    x4 = _mm_unpackhi_epi32(x0, x1);
-    // 46 56 66 76 47 57 67 77
-    x5 = _mm_unpackhi_epi32(x2, x3);
-    // 06 16 26 36 46 56 66 76
-    x6 = _mm_unpacklo_epi64(x4, x5);
-    // 07 17 27 37 47 57 67 77
-    x7 = _mm_unpackhi_epi64(x4, x5);
-
-    _mm_storeu_si128((__m128i *)(out + 6*out_p), x6);
-    // 06 16 26 36 46 56 66 76
-    _mm_storeu_si128((__m128i *)(out + 7*out_p), x7);
-    // 07 17 27 37 47 57 67 77
-  } while (++idx8x8 < num_8x8_to_transpose);
-}
-
-static INLINE void highbd_transpose8x16(uint16_t *in0, uint16_t *in1,
-                                        int in_p, uint16_t *out, int out_p) {
-  uint16_t *src0[1];
-  uint16_t *src1[1];
-  uint16_t *dest0[1];
-  uint16_t *dest1[1];
-  src0[0] = in0;
-  src1[0] = in1;
-  dest0[0] = out;
-  dest1[0] = out + 8;
-  highbd_transpose(src0, in_p, dest0, out_p, 1);
-  highbd_transpose(src1, in_p, dest1, out_p, 1);
-}
-
-void vp9_highbd_lpf_vertical_4_sse2(uint16_t *s, int p,
-                                    const uint8_t *blimit,
-                                    const uint8_t *limit,
-                                    const uint8_t *thresh,
-                                    int count, int bd) {
-  DECLARE_ALIGNED(16, uint16_t, t_dst[8 * 8]);
-  uint16_t *src[1];
-  uint16_t *dst[1];
-  (void)count;
-
-  // Transpose 8x8
-  src[0] = s - 4;
-  dst[0] = t_dst;
-
-  highbd_transpose(src, p, dst, 8, 1);
-
-  // Loop filtering
-  vp9_highbd_lpf_horizontal_4_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1,
-                                   bd);
-
-  src[0] = t_dst;
-  dst[0] = s - 4;
-
-  // Transpose back
-  highbd_transpose(src, 8, dst, p, 1);
-}
-
-void vp9_highbd_lpf_vertical_4_dual_sse2(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,
-                                         const uint8_t *thresh1,
-                                         int bd) {
-  DECLARE_ALIGNED(16, uint16_t, t_dst[16 * 8]);
-  uint16_t *src[2];
-  uint16_t *dst[2];
-
-  // Transpose 8x16
-  highbd_transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16);
-
-  // Loop filtering
-  vp9_highbd_lpf_horizontal_4_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0,
-                                        thresh0, blimit1, limit1, thresh1, bd);
-  src[0] = t_dst;
-  src[1] = t_dst + 8;
-  dst[0] = s - 4;
-  dst[1] = s - 4 + p * 8;
-
-  // Transpose back
-  highbd_transpose(src, 16, dst, p, 2);
-}
-
-void vp9_highbd_lpf_vertical_8_sse2(uint16_t *s, int p,
-                                    const uint8_t *blimit,
-                                    const uint8_t *limit,
-                                    const uint8_t *thresh,
-                                    int count, int bd) {
-  DECLARE_ALIGNED(16, uint16_t, t_dst[8 * 8]);
-  uint16_t *src[1];
-  uint16_t *dst[1];
-  (void)count;
-
-  // Transpose 8x8
-  src[0] = s - 4;
-  dst[0] = t_dst;
-
-  highbd_transpose(src, p, dst, 8, 1);
-
-  // Loop filtering
-  vp9_highbd_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1,
-                                   bd);
-
-  src[0] = t_dst;
-  dst[0] = s - 4;
-
-  // Transpose back
-  highbd_transpose(src, 8, dst, p, 1);
-}
-
-void vp9_highbd_lpf_vertical_8_dual_sse2(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,
-                                         const uint8_t *thresh1,
-                                         int bd) {
-  DECLARE_ALIGNED(16, uint16_t, t_dst[16 * 8]);
-  uint16_t *src[2];
-  uint16_t *dst[2];
-
-  // Transpose 8x16
-  highbd_transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16);
-
-  // Loop filtering
-  vp9_highbd_lpf_horizontal_8_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0,
-                                        thresh0, blimit1, limit1, thresh1, bd);
-  src[0] = t_dst;
-  src[1] = t_dst + 8;
-
-  dst[0] = s - 4;
-  dst[1] = s - 4 + p * 8;
-
-  // Transpose back
-  highbd_transpose(src, 16, dst, p, 2);
-}
-
-void vp9_highbd_lpf_vertical_16_sse2(uint16_t *s, int p,
-                                     const uint8_t *blimit,
-                                     const uint8_t *limit,
-                                     const uint8_t *thresh,
-                                     int bd) {
-  DECLARE_ALIGNED(16, uint16_t, t_dst[8 * 16]);
-  uint16_t *src[2];
-  uint16_t *dst[2];
-
-  src[0] = s - 8;
-  src[1] = s;
-  dst[0] = t_dst;
-  dst[1] = t_dst + 8 * 8;
-
-  // Transpose 16x8
-  highbd_transpose(src, p, dst, 8, 2);
-
-  // Loop filtering
-  highbd_mb_lpf_horizontal_edge_w_sse2_8(t_dst + 8 * 8, 8, blimit, limit,
-                                         thresh, bd);
-  src[0] = t_dst;
-  src[1] = t_dst + 8 * 8;
-  dst[0] = s - 8;
-  dst[1] = s;
-
-  // Transpose back
-  highbd_transpose(src, 8, dst, p, 2);
-}
-
-void vp9_highbd_lpf_vertical_16_dual_sse2(uint16_t *s,
-                                          int p,
-                                          const uint8_t *blimit,
-                                          const uint8_t *limit,
-                                          const uint8_t *thresh,
-                                          int bd) {
-  DECLARE_ALIGNED(16, uint16_t, t_dst[256]);
-
-  //  Transpose 16x16
-  highbd_transpose8x16(s - 8, s - 8 + 8 * p, p, t_dst, 16);
-  highbd_transpose8x16(s, s + 8 * p, p, t_dst + 8 * 16, 16);
-
-  //  Loop filtering
-  highbd_mb_lpf_horizontal_edge_w_sse2_16(t_dst + 8 * 16, 16, blimit, limit,
-                                          thresh, bd);
-
-  //  Transpose back
-  highbd_transpose8x16(t_dst, t_dst + 8 * 16, 16, s - 8, p);
-  highbd_transpose8x16(t_dst + 8, t_dst + 8 + 8 * 16, 16, s - 8 + 8 * p, p);
-}
diff --git a/vp9/common/x86/vp9_loopfilter_intrin_avx2.c b/vp9/common/x86/vp9_loopfilter_intrin_avx2.c
deleted file mode 100644
index 770a65f4c..000000000
--- a/vp9/common/x86/vp9_loopfilter_intrin_avx2.c
+++ /dev/null
@@ -1,986 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <immintrin.h>  /* AVX2 */
-
-#include "./vp9_rtcd.h"
-#include "vpx_ports/mem.h"
-
-static void mb_lpf_horizontal_edge_w_avx2_8(unsigned char *s, int p,
-        const unsigned char *_blimit, const unsigned char *_limit,
-        const unsigned char *_thresh) {
-    __m128i mask, hev, flat, flat2;
-    const __m128i zero = _mm_set1_epi16(0);
-    const __m128i one = _mm_set1_epi8(1);
-    __m128i q7p7, q6p6, q5p5, q4p4, q3p3, q2p2, q1p1, q0p0, p0q0, p1q1;
-    __m128i abs_p1p0;
-
-    const __m128i thresh = _mm_broadcastb_epi8(
-            _mm_cvtsi32_si128((int) _thresh[0]));
-    const __m128i limit = _mm_broadcastb_epi8(
-            _mm_cvtsi32_si128((int) _limit[0]));
-    const __m128i blimit = _mm_broadcastb_epi8(
-            _mm_cvtsi32_si128((int) _blimit[0]));
-
-    q4p4 = _mm_loadl_epi64((__m128i *) (s - 5 * p));
-    q4p4 = _mm_castps_si128(
-            _mm_loadh_pi(_mm_castsi128_ps(q4p4), (__m64 *) (s + 4 * p)));
-    q3p3 = _mm_loadl_epi64((__m128i *) (s - 4 * p));
-    q3p3 = _mm_castps_si128(
-            _mm_loadh_pi(_mm_castsi128_ps(q3p3), (__m64 *) (s + 3 * p)));
-    q2p2 = _mm_loadl_epi64((__m128i *) (s - 3 * p));
-    q2p2 = _mm_castps_si128(
-            _mm_loadh_pi(_mm_castsi128_ps(q2p2), (__m64 *) (s + 2 * p)));
-    q1p1 = _mm_loadl_epi64((__m128i *) (s - 2 * p));
-    q1p1 = _mm_castps_si128(
-            _mm_loadh_pi(_mm_castsi128_ps(q1p1), (__m64 *) (s + 1 * p)));
-    p1q1 = _mm_shuffle_epi32(q1p1, 78);
-    q0p0 = _mm_loadl_epi64((__m128i *) (s - 1 * p));
-    q0p0 = _mm_castps_si128(
-            _mm_loadh_pi(_mm_castsi128_ps(q0p0), (__m64 *) (s - 0 * p)));
-    p0q0 = _mm_shuffle_epi32(q0p0, 78);
-
-    {
-        __m128i abs_p1q1, abs_p0q0, abs_q1q0, fe, ff, work;
-        abs_p1p0 = _mm_or_si128(_mm_subs_epu8(q1p1, q0p0),
-                _mm_subs_epu8(q0p0, q1p1));
-        abs_q1q0 = _mm_srli_si128(abs_p1p0, 8);
-        fe = _mm_set1_epi8(0xfe);
-        ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
-        abs_p0q0 = _mm_or_si128(_mm_subs_epu8(q0p0, p0q0),
-                _mm_subs_epu8(p0q0, q0p0));
-        abs_p1q1 = _mm_or_si128(_mm_subs_epu8(q1p1, p1q1),
-                _mm_subs_epu8(p1q1, q1p1));
-        flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
-        hev = _mm_subs_epu8(flat, thresh);
-        hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
-
-        abs_p0q0 = _mm_adds_epu8(abs_p0q0, abs_p0q0);
-        abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
-        mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
-        mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
-        // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-        mask = _mm_max_epu8(abs_p1p0, mask);
-        // mask |= (abs(p1 - p0) > limit) * -1;
-        // mask |= (abs(q1 - q0) > limit) * -1;
-
-        work = _mm_max_epu8(
-                _mm_or_si128(_mm_subs_epu8(q2p2, q1p1),
-                        _mm_subs_epu8(q1p1, q2p2)),
-                _mm_or_si128(_mm_subs_epu8(q3p3, q2p2),
-                        _mm_subs_epu8(q2p2, q3p3)));
-        mask = _mm_max_epu8(work, mask);
-        mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8));
-        mask = _mm_subs_epu8(mask, limit);
-        mask = _mm_cmpeq_epi8(mask, zero);
-    }
-
-    // lp filter
-    {
-        const __m128i t4 = _mm_set1_epi8(4);
-        const __m128i t3 = _mm_set1_epi8(3);
-        const __m128i t80 = _mm_set1_epi8(0x80);
-        const __m128i t1 = _mm_set1_epi16(0x1);
-        __m128i qs1ps1 = _mm_xor_si128(q1p1, t80);
-        __m128i qs0ps0 = _mm_xor_si128(q0p0, t80);
-        __m128i qs0 = _mm_xor_si128(p0q0, t80);
-        __m128i qs1 = _mm_xor_si128(p1q1, t80);
-        __m128i filt;
-        __m128i work_a;
-        __m128i filter1, filter2;
-        __m128i flat2_q6p6, flat2_q5p5, flat2_q4p4, flat2_q3p3, flat2_q2p2;
-        __m128i flat2_q1p1, flat2_q0p0, flat_q2p2, flat_q1p1, flat_q0p0;
-
-        filt = _mm_and_si128(_mm_subs_epi8(qs1ps1, qs1), hev);
-        work_a = _mm_subs_epi8(qs0, qs0ps0);
-        filt = _mm_adds_epi8(filt, work_a);
-        filt = _mm_adds_epi8(filt, work_a);
-        filt = _mm_adds_epi8(filt, work_a);
-        /* (vp9_filter + 3 * (qs0 - ps0)) & mask */
-        filt = _mm_and_si128(filt, mask);
-
-        filter1 = _mm_adds_epi8(filt, t4);
-        filter2 = _mm_adds_epi8(filt, t3);
-
-        filter1 = _mm_unpacklo_epi8(zero, filter1);
-        filter1 = _mm_srai_epi16(filter1, 0xB);
-        filter2 = _mm_unpacklo_epi8(zero, filter2);
-        filter2 = _mm_srai_epi16(filter2, 0xB);
-
-        /* Filter1 >> 3 */
-        filt = _mm_packs_epi16(filter2, _mm_subs_epi16(zero, filter1));
-        qs0ps0 = _mm_xor_si128(_mm_adds_epi8(qs0ps0, filt), t80);
-
-        /* filt >> 1 */
-        filt = _mm_adds_epi16(filter1, t1);
-        filt = _mm_srai_epi16(filt, 1);
-        filt = _mm_andnot_si128(
-                _mm_srai_epi16(_mm_unpacklo_epi8(zero, hev), 0x8), filt);
-        filt = _mm_packs_epi16(filt, _mm_subs_epi16(zero, filt));
-        qs1ps1 = _mm_xor_si128(_mm_adds_epi8(qs1ps1, filt), t80);
-        // loopfilter done
-
-        {
-            __m128i work;
-            flat = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(q2p2, q0p0),
-                            _mm_subs_epu8(q0p0, q2p2)),
-                    _mm_or_si128(_mm_subs_epu8(q3p3, q0p0),
-                            _mm_subs_epu8(q0p0, q3p3)));
-            flat = _mm_max_epu8(abs_p1p0, flat);
-            flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8));
-            flat = _mm_subs_epu8(flat, one);
-            flat = _mm_cmpeq_epi8(flat, zero);
-            flat = _mm_and_si128(flat, mask);
-
-            q5p5 = _mm_loadl_epi64((__m128i *) (s - 6 * p));
-            q5p5 = _mm_castps_si128(
-                    _mm_loadh_pi(_mm_castsi128_ps(q5p5),
-                            (__m64 *) (s + 5 * p)));
-
-            q6p6 = _mm_loadl_epi64((__m128i *) (s - 7 * p));
-            q6p6 = _mm_castps_si128(
-                    _mm_loadh_pi(_mm_castsi128_ps(q6p6),
-                            (__m64 *) (s + 6 * p)));
-
-            flat2 = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(q4p4, q0p0),
-                            _mm_subs_epu8(q0p0, q4p4)),
-                    _mm_or_si128(_mm_subs_epu8(q5p5, q0p0),
-                            _mm_subs_epu8(q0p0, q5p5)));
-
-            q7p7 = _mm_loadl_epi64((__m128i *) (s - 8 * p));
-            q7p7 = _mm_castps_si128(
-                    _mm_loadh_pi(_mm_castsi128_ps(q7p7),
-                            (__m64 *) (s + 7 * p)));
-
-            work = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(q6p6, q0p0),
-                            _mm_subs_epu8(q0p0, q6p6)),
-                    _mm_or_si128(_mm_subs_epu8(q7p7, q0p0),
-                            _mm_subs_epu8(q0p0, q7p7)));
-
-            flat2 = _mm_max_epu8(work, flat2);
-            flat2 = _mm_max_epu8(flat2, _mm_srli_si128(flat2, 8));
-            flat2 = _mm_subs_epu8(flat2, one);
-            flat2 = _mm_cmpeq_epi8(flat2, zero);
-            flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
-        }
-
-        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-        // flat and wide flat calculations
-        {
-            const __m128i eight = _mm_set1_epi16(8);
-            const __m128i four = _mm_set1_epi16(4);
-            __m128i p7_16, p6_16, p5_16, p4_16, p3_16, p2_16, p1_16, p0_16;
-            __m128i q7_16, q6_16, q5_16, q4_16, q3_16, q2_16, q1_16, q0_16;
-            __m128i pixelFilter_p, pixelFilter_q;
-            __m128i pixetFilter_p2p1p0, pixetFilter_q2q1q0;
-            __m128i sum_p7, sum_q7, sum_p3, sum_q3, res_p, res_q;
-
-            p7_16 = _mm_unpacklo_epi8(q7p7, zero);
-            p6_16 = _mm_unpacklo_epi8(q6p6, zero);
-            p5_16 = _mm_unpacklo_epi8(q5p5, zero);
-            p4_16 = _mm_unpacklo_epi8(q4p4, zero);
-            p3_16 = _mm_unpacklo_epi8(q3p3, zero);
-            p2_16 = _mm_unpacklo_epi8(q2p2, zero);
-            p1_16 = _mm_unpacklo_epi8(q1p1, zero);
-            p0_16 = _mm_unpacklo_epi8(q0p0, zero);
-            q0_16 = _mm_unpackhi_epi8(q0p0, zero);
-            q1_16 = _mm_unpackhi_epi8(q1p1, zero);
-            q2_16 = _mm_unpackhi_epi8(q2p2, zero);
-            q3_16 = _mm_unpackhi_epi8(q3p3, zero);
-            q4_16 = _mm_unpackhi_epi8(q4p4, zero);
-            q5_16 = _mm_unpackhi_epi8(q5p5, zero);
-            q6_16 = _mm_unpackhi_epi8(q6p6, zero);
-            q7_16 = _mm_unpackhi_epi8(q7p7, zero);
-
-            pixelFilter_p = _mm_add_epi16(_mm_add_epi16(p6_16, p5_16),
-                    _mm_add_epi16(p4_16, p3_16));
-            pixelFilter_q = _mm_add_epi16(_mm_add_epi16(q6_16, q5_16),
-                    _mm_add_epi16(q4_16, q3_16));
-
-            pixetFilter_p2p1p0 = _mm_add_epi16(p0_16,
-                    _mm_add_epi16(p2_16, p1_16));
-            pixelFilter_p = _mm_add_epi16(pixelFilter_p, pixetFilter_p2p1p0);
-
-            pixetFilter_q2q1q0 = _mm_add_epi16(q0_16,
-                    _mm_add_epi16(q2_16, q1_16));
-            pixelFilter_q = _mm_add_epi16(pixelFilter_q, pixetFilter_q2q1q0);
-            pixelFilter_p = _mm_add_epi16(eight,
-                    _mm_add_epi16(pixelFilter_p, pixelFilter_q));
-            pixetFilter_p2p1p0 = _mm_add_epi16(four,
-                    _mm_add_epi16(pixetFilter_p2p1p0, pixetFilter_q2q1q0));
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(p7_16, p0_16)),
-                    4);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(q7_16, q0_16)),
-                    4);
-            flat2_q0p0 = _mm_packus_epi16(res_p, res_q);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixetFilter_p2p1p0,
-                            _mm_add_epi16(p3_16, p0_16)), 3);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixetFilter_p2p1p0,
-                            _mm_add_epi16(q3_16, q0_16)), 3);
-
-            flat_q0p0 = _mm_packus_epi16(res_p, res_q);
-
-            sum_p7 = _mm_add_epi16(p7_16, p7_16);
-            sum_q7 = _mm_add_epi16(q7_16, q7_16);
-            sum_p3 = _mm_add_epi16(p3_16, p3_16);
-            sum_q3 = _mm_add_epi16(q3_16, q3_16);
-
-            pixelFilter_q = _mm_sub_epi16(pixelFilter_p, p6_16);
-            pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q6_16);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p1_16)),
-                    4);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q1_16)),
-                    4);
-            flat2_q1p1 = _mm_packus_epi16(res_p, res_q);
-
-            pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_p2p1p0, p2_16);
-            pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q2_16);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixetFilter_p2p1p0,
-                            _mm_add_epi16(sum_p3, p1_16)), 3);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixetFilter_q2q1q0,
-                            _mm_add_epi16(sum_q3, q1_16)), 3);
-            flat_q1p1 = _mm_packus_epi16(res_p, res_q);
-
-            sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-            sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-            sum_p3 = _mm_add_epi16(sum_p3, p3_16);
-            sum_q3 = _mm_add_epi16(sum_q3, q3_16);
-
-            pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q5_16);
-            pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p5_16);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p2_16)),
-                    4);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q2_16)),
-                    4);
-            flat2_q2p2 = _mm_packus_epi16(res_p, res_q);
-
-            pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q1_16);
-            pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_q2q1q0, p1_16);
-
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixetFilter_p2p1p0,
-                            _mm_add_epi16(sum_p3, p2_16)), 3);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixetFilter_q2q1q0,
-                            _mm_add_epi16(sum_q3, q2_16)), 3);
-            flat_q2p2 = _mm_packus_epi16(res_p, res_q);
-
-            sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-            sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-            pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q4_16);
-            pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p4_16);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p3_16)),
-                    4);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q3_16)),
-                    4);
-            flat2_q3p3 = _mm_packus_epi16(res_p, res_q);
-
-            sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-            sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-            pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q3_16);
-            pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p3_16);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p4_16)),
-                    4);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q4_16)),
-                    4);
-            flat2_q4p4 = _mm_packus_epi16(res_p, res_q);
-
-            sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-            sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-            pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q2_16);
-            pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p2_16);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p5_16)),
-                    4);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q5_16)),
-                    4);
-            flat2_q5p5 = _mm_packus_epi16(res_p, res_q);
-
-            sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-            sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-            pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q1_16);
-            pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p1_16);
-            res_p = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_p, _mm_add_epi16(sum_p7, p6_16)),
-                    4);
-            res_q = _mm_srli_epi16(
-                    _mm_add_epi16(pixelFilter_q, _mm_add_epi16(sum_q7, q6_16)),
-                    4);
-            flat2_q6p6 = _mm_packus_epi16(res_p, res_q);
-        }
-        // wide flat
-        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-        flat = _mm_shuffle_epi32(flat, 68);
-        flat2 = _mm_shuffle_epi32(flat2, 68);
-
-        q2p2 = _mm_andnot_si128(flat, q2p2);
-        flat_q2p2 = _mm_and_si128(flat, flat_q2p2);
-        q2p2 = _mm_or_si128(q2p2, flat_q2p2);
-
-        qs1ps1 = _mm_andnot_si128(flat, qs1ps1);
-        flat_q1p1 = _mm_and_si128(flat, flat_q1p1);
-        q1p1 = _mm_or_si128(qs1ps1, flat_q1p1);
-
-        qs0ps0 = _mm_andnot_si128(flat, qs0ps0);
-        flat_q0p0 = _mm_and_si128(flat, flat_q0p0);
-        q0p0 = _mm_or_si128(qs0ps0, flat_q0p0);
-
-        q6p6 = _mm_andnot_si128(flat2, q6p6);
-        flat2_q6p6 = _mm_and_si128(flat2, flat2_q6p6);
-        q6p6 = _mm_or_si128(q6p6, flat2_q6p6);
-        _mm_storel_epi64((__m128i *) (s - 7 * p), q6p6);
-        _mm_storeh_pi((__m64 *) (s + 6 * p), _mm_castsi128_ps(q6p6));
-
-        q5p5 = _mm_andnot_si128(flat2, q5p5);
-        flat2_q5p5 = _mm_and_si128(flat2, flat2_q5p5);
-        q5p5 = _mm_or_si128(q5p5, flat2_q5p5);
-        _mm_storel_epi64((__m128i *) (s - 6 * p), q5p5);
-        _mm_storeh_pi((__m64 *) (s + 5 * p), _mm_castsi128_ps(q5p5));
-
-        q4p4 = _mm_andnot_si128(flat2, q4p4);
-        flat2_q4p4 = _mm_and_si128(flat2, flat2_q4p4);
-        q4p4 = _mm_or_si128(q4p4, flat2_q4p4);
-        _mm_storel_epi64((__m128i *) (s - 5 * p), q4p4);
-        _mm_storeh_pi((__m64 *) (s + 4 * p), _mm_castsi128_ps(q4p4));
-
-        q3p3 = _mm_andnot_si128(flat2, q3p3);
-        flat2_q3p3 = _mm_and_si128(flat2, flat2_q3p3);
-        q3p3 = _mm_or_si128(q3p3, flat2_q3p3);
-        _mm_storel_epi64((__m128i *) (s - 4 * p), q3p3);
-        _mm_storeh_pi((__m64 *) (s + 3 * p), _mm_castsi128_ps(q3p3));
-
-        q2p2 = _mm_andnot_si128(flat2, q2p2);
-        flat2_q2p2 = _mm_and_si128(flat2, flat2_q2p2);
-        q2p2 = _mm_or_si128(q2p2, flat2_q2p2);
-        _mm_storel_epi64((__m128i *) (s - 3 * p), q2p2);
-        _mm_storeh_pi((__m64 *) (s + 2 * p), _mm_castsi128_ps(q2p2));
-
-        q1p1 = _mm_andnot_si128(flat2, q1p1);
-        flat2_q1p1 = _mm_and_si128(flat2, flat2_q1p1);
-        q1p1 = _mm_or_si128(q1p1, flat2_q1p1);
-        _mm_storel_epi64((__m128i *) (s - 2 * p), q1p1);
-        _mm_storeh_pi((__m64 *) (s + 1 * p), _mm_castsi128_ps(q1p1));
-
-        q0p0 = _mm_andnot_si128(flat2, q0p0);
-        flat2_q0p0 = _mm_and_si128(flat2, flat2_q0p0);
-        q0p0 = _mm_or_si128(q0p0, flat2_q0p0);
-        _mm_storel_epi64((__m128i *) (s - 1 * p), q0p0);
-        _mm_storeh_pi((__m64 *) (s - 0 * p), _mm_castsi128_ps(q0p0));
-    }
-}
-
-DECLARE_ALIGNED(32, static const uint8_t, filt_loopfilter_avx2[32]) = {
-  0, 128, 1, 128, 2, 128, 3, 128, 4, 128, 5, 128, 6, 128, 7, 128,
-  8, 128, 9, 128, 10, 128, 11, 128, 12, 128, 13, 128, 14, 128, 15, 128
-};
-
-static void mb_lpf_horizontal_edge_w_avx2_16(unsigned char *s, int p,
-        const unsigned char *_blimit, const unsigned char *_limit,
-        const unsigned char *_thresh) {
-    __m128i mask, hev, flat, flat2;
-    const __m128i zero = _mm_set1_epi16(0);
-    const __m128i one = _mm_set1_epi8(1);
-    __m128i p7, p6, p5;
-    __m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4;
-    __m128i q5, q6, q7;
-    __m256i p256_7, q256_7, p256_6, q256_6, p256_5, q256_5, p256_4,
-            q256_4, p256_3, q256_3, p256_2, q256_2, p256_1, q256_1,
-            p256_0, q256_0;
-
-    const __m128i thresh = _mm_broadcastb_epi8(
-            _mm_cvtsi32_si128((int) _thresh[0]));
-    const __m128i limit = _mm_broadcastb_epi8(
-            _mm_cvtsi32_si128((int) _limit[0]));
-    const __m128i blimit = _mm_broadcastb_epi8(
-            _mm_cvtsi32_si128((int) _blimit[0]));
-
-    p256_4 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s - 5 * p)));
-    p256_3 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s - 4 * p)));
-    p256_2 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s - 3 * p)));
-    p256_1 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s - 2 * p)));
-    p256_0 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s - 1 * p)));
-    q256_0 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s - 0 * p)));
-    q256_1 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s + 1 * p)));
-    q256_2 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s + 2 * p)));
-    q256_3 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s + 3 * p)));
-    q256_4 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                (__m128d const *)(s + 4 * p)));
-
-    p4 = _mm256_castsi256_si128(p256_4);
-    p3 = _mm256_castsi256_si128(p256_3);
-    p2 = _mm256_castsi256_si128(p256_2);
-    p1 = _mm256_castsi256_si128(p256_1);
-    p0 = _mm256_castsi256_si128(p256_0);
-    q0 = _mm256_castsi256_si128(q256_0);
-    q1 = _mm256_castsi256_si128(q256_1);
-    q2 = _mm256_castsi256_si128(q256_2);
-    q3 = _mm256_castsi256_si128(q256_3);
-    q4 = _mm256_castsi256_si128(q256_4);
-
-    {
-        const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0),
-                _mm_subs_epu8(p0, p1));
-        const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0),
-                _mm_subs_epu8(q0, q1));
-        const __m128i fe = _mm_set1_epi8(0xfe);
-        const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
-        __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0),
-                _mm_subs_epu8(q0, p0));
-        __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1),
-                _mm_subs_epu8(q1, p1));
-        __m128i work;
-        flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
-        hev = _mm_subs_epu8(flat, thresh);
-        hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
-
-        abs_p0q0 = _mm_adds_epu8(abs_p0q0, abs_p0q0);
-        abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
-        mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
-        mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
-        // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-        mask = _mm_max_epu8(flat, mask);
-        // mask |= (abs(p1 - p0) > limit) * -1;
-        // mask |= (abs(q1 - q0) > limit) * -1;
-        work = _mm_max_epu8(
-                _mm_or_si128(_mm_subs_epu8(p2, p1), _mm_subs_epu8(p1, p2)),
-                _mm_or_si128(_mm_subs_epu8(p3, p2), _mm_subs_epu8(p2, p3)));
-        mask = _mm_max_epu8(work, mask);
-        work = _mm_max_epu8(
-                _mm_or_si128(_mm_subs_epu8(q2, q1), _mm_subs_epu8(q1, q2)),
-                _mm_or_si128(_mm_subs_epu8(q3, q2), _mm_subs_epu8(q2, q3)));
-        mask = _mm_max_epu8(work, mask);
-        mask = _mm_subs_epu8(mask, limit);
-        mask = _mm_cmpeq_epi8(mask, zero);
-    }
-
-    // lp filter
-    {
-        const __m128i t4 = _mm_set1_epi8(4);
-        const __m128i t3 = _mm_set1_epi8(3);
-        const __m128i t80 = _mm_set1_epi8(0x80);
-        const __m128i te0 = _mm_set1_epi8(0xe0);
-        const __m128i t1f = _mm_set1_epi8(0x1f);
-        const __m128i t1 = _mm_set1_epi8(0x1);
-        const __m128i t7f = _mm_set1_epi8(0x7f);
-
-        __m128i ps1 = _mm_xor_si128(p1, t80);
-        __m128i ps0 = _mm_xor_si128(p0, t80);
-        __m128i qs0 = _mm_xor_si128(q0, t80);
-        __m128i qs1 = _mm_xor_si128(q1, t80);
-        __m128i filt;
-        __m128i work_a;
-        __m128i filter1, filter2;
-        __m128i flat2_p6, flat2_p5, flat2_p4, flat2_p3, flat2_p2, flat2_p1,
-                flat2_p0, flat2_q0, flat2_q1, flat2_q2, flat2_q3, flat2_q4,
-                flat2_q5, flat2_q6, flat_p2, flat_p1, flat_p0, flat_q0, flat_q1,
-                flat_q2;
-
-        filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev);
-        work_a = _mm_subs_epi8(qs0, ps0);
-        filt = _mm_adds_epi8(filt, work_a);
-        filt = _mm_adds_epi8(filt, work_a);
-        filt = _mm_adds_epi8(filt, work_a);
-        /* (vp9_filter + 3 * (qs0 - ps0)) & mask */
-        filt = _mm_and_si128(filt, mask);
-
-        filter1 = _mm_adds_epi8(filt, t4);
-        filter2 = _mm_adds_epi8(filt, t3);
-
-        /* Filter1 >> 3 */
-        work_a = _mm_cmpgt_epi8(zero, filter1);
-        filter1 = _mm_srli_epi16(filter1, 3);
-        work_a = _mm_and_si128(work_a, te0);
-        filter1 = _mm_and_si128(filter1, t1f);
-        filter1 = _mm_or_si128(filter1, work_a);
-        qs0 = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80);
-
-        /* Filter2 >> 3 */
-        work_a = _mm_cmpgt_epi8(zero, filter2);
-        filter2 = _mm_srli_epi16(filter2, 3);
-        work_a = _mm_and_si128(work_a, te0);
-        filter2 = _mm_and_si128(filter2, t1f);
-        filter2 = _mm_or_si128(filter2, work_a);
-        ps0 = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80);
-
-        /* filt >> 1 */
-        filt = _mm_adds_epi8(filter1, t1);
-        work_a = _mm_cmpgt_epi8(zero, filt);
-        filt = _mm_srli_epi16(filt, 1);
-        work_a = _mm_and_si128(work_a, t80);
-        filt = _mm_and_si128(filt, t7f);
-        filt = _mm_or_si128(filt, work_a);
-        filt = _mm_andnot_si128(hev, filt);
-        ps1 = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80);
-        qs1 = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80);
-        // loopfilter done
-
-        {
-            __m128i work;
-            work = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(p2, p0), _mm_subs_epu8(p0, p2)),
-                    _mm_or_si128(_mm_subs_epu8(q2, q0), _mm_subs_epu8(q0, q2)));
-            flat = _mm_max_epu8(work, flat);
-            work = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(p3, p0), _mm_subs_epu8(p0, p3)),
-                    _mm_or_si128(_mm_subs_epu8(q3, q0), _mm_subs_epu8(q0, q3)));
-            flat = _mm_max_epu8(work, flat);
-            work = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(p4, p0), _mm_subs_epu8(p0, p4)),
-                    _mm_or_si128(_mm_subs_epu8(q4, q0), _mm_subs_epu8(q0, q4)));
-            flat = _mm_subs_epu8(flat, one);
-            flat = _mm_cmpeq_epi8(flat, zero);
-            flat = _mm_and_si128(flat, mask);
-
-            p256_5 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                        (__m128d const *)(s - 6 * p)));
-            q256_5 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                        (__m128d const *)(s + 5 * p)));
-            p5 = _mm256_castsi256_si128(p256_5);
-            q5 = _mm256_castsi256_si128(q256_5);
-            flat2 = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(p5, p0), _mm_subs_epu8(p0, p5)),
-                    _mm_or_si128(_mm_subs_epu8(q5, q0), _mm_subs_epu8(q0, q5)));
-
-            flat2 = _mm_max_epu8(work, flat2);
-            p256_6 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                        (__m128d const *)(s - 7 * p)));
-            q256_6 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                        (__m128d const *)(s + 6 * p)));
-            p6 = _mm256_castsi256_si128(p256_6);
-            q6 = _mm256_castsi256_si128(q256_6);
-            work = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(p6, p0), _mm_subs_epu8(p0, p6)),
-                    _mm_or_si128(_mm_subs_epu8(q6, q0), _mm_subs_epu8(q0, q6)));
-
-            flat2 = _mm_max_epu8(work, flat2);
-
-            p256_7 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                        (__m128d const *)(s - 8 * p)));
-            q256_7 = _mm256_castpd_si256(_mm256_broadcast_pd(
-                                        (__m128d const *)(s + 7 * p)));
-            p7 = _mm256_castsi256_si128(p256_7);
-            q7 = _mm256_castsi256_si128(q256_7);
-            work = _mm_max_epu8(
-                    _mm_or_si128(_mm_subs_epu8(p7, p0), _mm_subs_epu8(p0, p7)),
-                    _mm_or_si128(_mm_subs_epu8(q7, q0), _mm_subs_epu8(q0, q7)));
-
-            flat2 = _mm_max_epu8(work, flat2);
-            flat2 = _mm_subs_epu8(flat2, one);
-            flat2 = _mm_cmpeq_epi8(flat2, zero);
-            flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
-        }
-
-        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-        // flat and wide flat calculations
-        {
-            const __m256i eight = _mm256_set1_epi16(8);
-            const __m256i four = _mm256_set1_epi16(4);
-            __m256i pixelFilter_p, pixelFilter_q, pixetFilter_p2p1p0,
-                    pixetFilter_q2q1q0, sum_p7, sum_q7, sum_p3, sum_q3, res_p,
-                    res_q;
-
-            const __m256i filter = _mm256_load_si256(
-                                  (__m256i const *)filt_loopfilter_avx2);
-            p256_7 = _mm256_shuffle_epi8(p256_7, filter);
-            p256_6 = _mm256_shuffle_epi8(p256_6, filter);
-            p256_5 = _mm256_shuffle_epi8(p256_5, filter);
-            p256_4 = _mm256_shuffle_epi8(p256_4, filter);
-            p256_3 = _mm256_shuffle_epi8(p256_3, filter);
-            p256_2 = _mm256_shuffle_epi8(p256_2, filter);
-            p256_1 = _mm256_shuffle_epi8(p256_1, filter);
-            p256_0 = _mm256_shuffle_epi8(p256_0, filter);
-            q256_0 = _mm256_shuffle_epi8(q256_0, filter);
-            q256_1 = _mm256_shuffle_epi8(q256_1, filter);
-            q256_2 = _mm256_shuffle_epi8(q256_2, filter);
-            q256_3 = _mm256_shuffle_epi8(q256_3, filter);
-            q256_4 = _mm256_shuffle_epi8(q256_4, filter);
-            q256_5 = _mm256_shuffle_epi8(q256_5, filter);
-            q256_6 = _mm256_shuffle_epi8(q256_6, filter);
-            q256_7 = _mm256_shuffle_epi8(q256_7, filter);
-
-            pixelFilter_p = _mm256_add_epi16(_mm256_add_epi16(p256_6, p256_5),
-                    _mm256_add_epi16(p256_4, p256_3));
-            pixelFilter_q = _mm256_add_epi16(_mm256_add_epi16(q256_6, q256_5),
-                    _mm256_add_epi16(q256_4, q256_3));
-
-            pixetFilter_p2p1p0 = _mm256_add_epi16(p256_0,
-                    _mm256_add_epi16(p256_2, p256_1));
-            pixelFilter_p = _mm256_add_epi16(pixelFilter_p, pixetFilter_p2p1p0);
-
-            pixetFilter_q2q1q0 = _mm256_add_epi16(q256_0,
-                    _mm256_add_epi16(q256_2, q256_1));
-            pixelFilter_q = _mm256_add_epi16(pixelFilter_q, pixetFilter_q2q1q0);
-
-            pixelFilter_p = _mm256_add_epi16(eight,
-                    _mm256_add_epi16(pixelFilter_p, pixelFilter_q));
-
-            pixetFilter_p2p1p0 = _mm256_add_epi16(four,
-                    _mm256_add_epi16(pixetFilter_p2p1p0, pixetFilter_q2q1q0));
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(p256_7, p256_0)), 4);
-
-            flat2_p0 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(q256_7, q256_0)), 4);
-
-            flat2_q0 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixetFilter_p2p1p0,
-                            _mm256_add_epi16(p256_3, p256_0)), 3);
-
-            flat_p0 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixetFilter_p2p1p0,
-                            _mm256_add_epi16(q256_3, q256_0)), 3);
-
-            flat_q0 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            sum_p7 = _mm256_add_epi16(p256_7, p256_7);
-
-            sum_q7 = _mm256_add_epi16(q256_7, q256_7);
-
-            sum_p3 = _mm256_add_epi16(p256_3, p256_3);
-
-            sum_q3 = _mm256_add_epi16(q256_3, q256_3);
-
-            pixelFilter_q = _mm256_sub_epi16(pixelFilter_p, p256_6);
-
-            pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_6);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(sum_p7, p256_1)), 4);
-
-            flat2_p1 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_q,
-                            _mm256_add_epi16(sum_q7, q256_1)), 4);
-
-            flat2_q1 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            pixetFilter_q2q1q0 = _mm256_sub_epi16(pixetFilter_p2p1p0, p256_2);
-
-            pixetFilter_p2p1p0 = _mm256_sub_epi16(pixetFilter_p2p1p0, q256_2);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixetFilter_p2p1p0,
-                            _mm256_add_epi16(sum_p3, p256_1)), 3);
-
-            flat_p1 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixetFilter_q2q1q0,
-                            _mm256_add_epi16(sum_q3, q256_1)), 3);
-
-            flat_q1 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            sum_p7 = _mm256_add_epi16(sum_p7, p256_7);
-
-            sum_q7 = _mm256_add_epi16(sum_q7, q256_7);
-
-            sum_p3 = _mm256_add_epi16(sum_p3, p256_3);
-
-            sum_q3 = _mm256_add_epi16(sum_q3, q256_3);
-
-            pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_5);
-
-            pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_5);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(sum_p7, p256_2)), 4);
-
-            flat2_p2 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_q,
-                            _mm256_add_epi16(sum_q7, q256_2)), 4);
-
-            flat2_q2 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            pixetFilter_p2p1p0 = _mm256_sub_epi16(pixetFilter_p2p1p0, q256_1);
-
-            pixetFilter_q2q1q0 = _mm256_sub_epi16(pixetFilter_q2q1q0, p256_1);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixetFilter_p2p1p0,
-                            _mm256_add_epi16(sum_p3, p256_2)), 3);
-
-            flat_p2 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixetFilter_q2q1q0,
-                            _mm256_add_epi16(sum_q3, q256_2)), 3);
-
-            flat_q2 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            sum_p7 = _mm256_add_epi16(sum_p7, p256_7);
-
-            sum_q7 = _mm256_add_epi16(sum_q7, q256_7);
-
-            pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_4);
-
-            pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_4);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(sum_p7, p256_3)), 4);
-
-            flat2_p3 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_q,
-                            _mm256_add_epi16(sum_q7, q256_3)), 4);
-
-            flat2_q3 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            sum_p7 = _mm256_add_epi16(sum_p7, p256_7);
-
-            sum_q7 = _mm256_add_epi16(sum_q7, q256_7);
-
-            pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_3);
-
-            pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_3);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(sum_p7, p256_4)), 4);
-
-            flat2_p4 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_q,
-                            _mm256_add_epi16(sum_q7, q256_4)), 4);
-
-            flat2_q4 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            sum_p7 = _mm256_add_epi16(sum_p7, p256_7);
-
-            sum_q7 = _mm256_add_epi16(sum_q7, q256_7);
-
-            pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_2);
-
-            pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_2);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(sum_p7, p256_5)), 4);
-
-            flat2_p5 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_q,
-                            _mm256_add_epi16(sum_q7, q256_5)), 4);
-
-            flat2_q5 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-
-            sum_p7 = _mm256_add_epi16(sum_p7, p256_7);
-
-            sum_q7 = _mm256_add_epi16(sum_q7, q256_7);
-
-            pixelFilter_p = _mm256_sub_epi16(pixelFilter_p, q256_1);
-
-            pixelFilter_q = _mm256_sub_epi16(pixelFilter_q, p256_1);
-
-            res_p = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_p,
-                            _mm256_add_epi16(sum_p7, p256_6)), 4);
-
-            flat2_p6 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_p, res_p),
-                            168));
-
-            res_q = _mm256_srli_epi16(
-                    _mm256_add_epi16(pixelFilter_q,
-                            _mm256_add_epi16(sum_q7, q256_6)), 4);
-
-            flat2_q6 = _mm256_castsi256_si128(
-                    _mm256_permute4x64_epi64(_mm256_packus_epi16(res_q, res_q),
-                            168));
-        }
-
-        // wide flat
-        // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-        p2 = _mm_andnot_si128(flat, p2);
-        flat_p2 = _mm_and_si128(flat, flat_p2);
-        p2 = _mm_or_si128(flat_p2, p2);
-
-        p1 = _mm_andnot_si128(flat, ps1);
-        flat_p1 = _mm_and_si128(flat, flat_p1);
-        p1 = _mm_or_si128(flat_p1, p1);
-
-        p0 = _mm_andnot_si128(flat, ps0);
-        flat_p0 = _mm_and_si128(flat, flat_p0);
-        p0 = _mm_or_si128(flat_p0, p0);
-
-        q0 = _mm_andnot_si128(flat, qs0);
-        flat_q0 = _mm_and_si128(flat, flat_q0);
-        q0 = _mm_or_si128(flat_q0, q0);
-
-        q1 = _mm_andnot_si128(flat, qs1);
-        flat_q1 = _mm_and_si128(flat, flat_q1);
-        q1 = _mm_or_si128(flat_q1, q1);
-
-        q2 = _mm_andnot_si128(flat, q2);
-        flat_q2 = _mm_and_si128(flat, flat_q2);
-        q2 = _mm_or_si128(flat_q2, q2);
-
-        p6 = _mm_andnot_si128(flat2, p6);
-        flat2_p6 = _mm_and_si128(flat2, flat2_p6);
-        p6 = _mm_or_si128(flat2_p6, p6);
-        _mm_storeu_si128((__m128i *) (s - 7 * p), p6);
-
-        p5 = _mm_andnot_si128(flat2, p5);
-        flat2_p5 = _mm_and_si128(flat2, flat2_p5);
-        p5 = _mm_or_si128(flat2_p5, p5);
-        _mm_storeu_si128((__m128i *) (s - 6 * p), p5);
-
-        p4 = _mm_andnot_si128(flat2, p4);
-        flat2_p4 = _mm_and_si128(flat2, flat2_p4);
-        p4 = _mm_or_si128(flat2_p4, p4);
-        _mm_storeu_si128((__m128i *) (s - 5 * p), p4);
-
-        p3 = _mm_andnot_si128(flat2, p3);
-        flat2_p3 = _mm_and_si128(flat2, flat2_p3);
-        p3 = _mm_or_si128(flat2_p3, p3);
-        _mm_storeu_si128((__m128i *) (s - 4 * p), p3);
-
-        p2 = _mm_andnot_si128(flat2, p2);
-        flat2_p2 = _mm_and_si128(flat2, flat2_p2);
-        p2 = _mm_or_si128(flat2_p2, p2);
-        _mm_storeu_si128((__m128i *) (s - 3 * p), p2);
-
-        p1 = _mm_andnot_si128(flat2, p1);
-        flat2_p1 = _mm_and_si128(flat2, flat2_p1);
-        p1 = _mm_or_si128(flat2_p1, p1);
-        _mm_storeu_si128((__m128i *) (s - 2 * p), p1);
-
-        p0 = _mm_andnot_si128(flat2, p0);
-        flat2_p0 = _mm_and_si128(flat2, flat2_p0);
-        p0 = _mm_or_si128(flat2_p0, p0);
-        _mm_storeu_si128((__m128i *) (s - 1 * p), p0);
-
-        q0 = _mm_andnot_si128(flat2, q0);
-        flat2_q0 = _mm_and_si128(flat2, flat2_q0);
-        q0 = _mm_or_si128(flat2_q0, q0);
-        _mm_storeu_si128((__m128i *) (s - 0 * p), q0);
-
-        q1 = _mm_andnot_si128(flat2, q1);
-        flat2_q1 = _mm_and_si128(flat2, flat2_q1);
-        q1 = _mm_or_si128(flat2_q1, q1);
-        _mm_storeu_si128((__m128i *) (s + 1 * p), q1);
-
-        q2 = _mm_andnot_si128(flat2, q2);
-        flat2_q2 = _mm_and_si128(flat2, flat2_q2);
-        q2 = _mm_or_si128(flat2_q2, q2);
-        _mm_storeu_si128((__m128i *) (s + 2 * p), q2);
-
-        q3 = _mm_andnot_si128(flat2, q3);
-        flat2_q3 = _mm_and_si128(flat2, flat2_q3);
-        q3 = _mm_or_si128(flat2_q3, q3);
-        _mm_storeu_si128((__m128i *) (s + 3 * p), q3);
-
-        q4 = _mm_andnot_si128(flat2, q4);
-        flat2_q4 = _mm_and_si128(flat2, flat2_q4);
-        q4 = _mm_or_si128(flat2_q4, q4);
-        _mm_storeu_si128((__m128i *) (s + 4 * p), q4);
-
-        q5 = _mm_andnot_si128(flat2, q5);
-        flat2_q5 = _mm_and_si128(flat2, flat2_q5);
-        q5 = _mm_or_si128(flat2_q5, q5);
-        _mm_storeu_si128((__m128i *) (s + 5 * p), q5);
-
-        q6 = _mm_andnot_si128(flat2, q6);
-        flat2_q6 = _mm_and_si128(flat2, flat2_q6);
-        q6 = _mm_or_si128(flat2_q6, q6);
-        _mm_storeu_si128((__m128i *) (s + 6 * p), q6);
-    }
-}
-
-void vp9_lpf_horizontal_16_avx2(unsigned char *s, int p,
-        const unsigned char *_blimit, const unsigned char *_limit,
-        const unsigned char *_thresh, int count) {
-    if (count == 1)
-        mb_lpf_horizontal_edge_w_avx2_8(s, p, _blimit, _limit, _thresh);
-    else
-        mb_lpf_horizontal_edge_w_avx2_16(s, p, _blimit, _limit, _thresh);
-}
diff --git a/vp9/common/x86/vp9_loopfilter_intrin_sse2.c b/vp9/common/x86/vp9_loopfilter_intrin_sse2.c
deleted file mode 100644
index fe8af5463..000000000
--- a/vp9/common/x86/vp9_loopfilter_intrin_sse2.c
+++ /dev/null
@@ -1,1587 +0,0 @@
-/*
- *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- *  Use of this source code is governed by a BSD-style license
- *  that can be found in the LICENSE file in the root of the source
- *  tree. An additional intellectual property rights grant can be found
- *  in the file PATENTS.  All contributing project authors may
- *  be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <emmintrin.h>  // SSE2
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/vp9_loopfilter.h"
-#include "vpx_ports/emmintrin_compat.h"
-
-static INLINE __m128i abs_diff(__m128i a, __m128i b) {
-  return _mm_or_si128(_mm_subs_epu8(a, b), _mm_subs_epu8(b, a));
-}
-
-static void mb_lpf_horizontal_edge_w_sse2_8(unsigned char *s,
-                                            int p,
-                                            const unsigned char *_blimit,
-                                            const unsigned char *_limit,
-                                            const unsigned char *_thresh) {
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i one = _mm_set1_epi8(1);
-  const __m128i blimit = _mm_load_si128((const __m128i *)_blimit);
-  const __m128i limit = _mm_load_si128((const __m128i *)_limit);
-  const __m128i thresh = _mm_load_si128((const __m128i *)_thresh);
-  __m128i mask, hev, flat, flat2;
-  __m128i q7p7, q6p6, q5p5, q4p4, q3p3, q2p2, q1p1, q0p0, p0q0, p1q1;
-  __m128i abs_p1p0;
-
-  q4p4 = _mm_loadl_epi64((__m128i *)(s - 5 * p));
-  q4p4 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q4p4),
-                                       (__m64 *)(s + 4 * p)));
-  q3p3 = _mm_loadl_epi64((__m128i *)(s - 4 * p));
-  q3p3 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q3p3),
-                                       (__m64 *)(s + 3 * p)));
-  q2p2 = _mm_loadl_epi64((__m128i *)(s - 3 * p));
-  q2p2 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q2p2),
-                                       (__m64 *)(s + 2 * p)));
-  q1p1 = _mm_loadl_epi64((__m128i *)(s - 2 * p));
-  q1p1 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q1p1),
-                                       (__m64 *)(s + 1 * p)));
-  p1q1 = _mm_shuffle_epi32(q1p1, 78);
-  q0p0 = _mm_loadl_epi64((__m128i *)(s - 1 * p));
-  q0p0 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q0p0),
-                                       (__m64 *)(s - 0 * p)));
-  p0q0 = _mm_shuffle_epi32(q0p0, 78);
-
-  {
-    __m128i abs_p1q1, abs_p0q0, abs_q1q0, fe, ff, work;
-    abs_p1p0 = abs_diff(q1p1, q0p0);
-    abs_q1q0 =  _mm_srli_si128(abs_p1p0, 8);
-    fe = _mm_set1_epi8(0xfe);
-    ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
-    abs_p0q0 = abs_diff(q0p0, p0q0);
-    abs_p1q1 = abs_diff(q1p1, p1q1);
-    flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
-    hev = _mm_subs_epu8(flat, thresh);
-    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
-
-    abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
-    abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
-    mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
-    mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
-    // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-    mask = _mm_max_epu8(abs_p1p0, mask);
-    // mask |= (abs(p1 - p0) > limit) * -1;
-    // mask |= (abs(q1 - q0) > limit) * -1;
-
-    work = _mm_max_epu8(abs_diff(q2p2, q1p1),
-                        abs_diff(q3p3, q2p2));
-    mask = _mm_max_epu8(work, mask);
-    mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8));
-    mask = _mm_subs_epu8(mask, limit);
-    mask = _mm_cmpeq_epi8(mask, zero);
-  }
-
-  // lp filter
-  {
-    const __m128i t4 = _mm_set1_epi8(4);
-    const __m128i t3 = _mm_set1_epi8(3);
-    const __m128i t80 = _mm_set1_epi8(0x80);
-    const __m128i t1 = _mm_set1_epi16(0x1);
-    __m128i qs1ps1 = _mm_xor_si128(q1p1, t80);
-    __m128i qs0ps0 = _mm_xor_si128(q0p0, t80);
-    __m128i qs0 = _mm_xor_si128(p0q0, t80);
-    __m128i qs1 = _mm_xor_si128(p1q1, t80);
-    __m128i filt;
-    __m128i work_a;
-    __m128i filter1, filter2;
-    __m128i flat2_q6p6, flat2_q5p5, flat2_q4p4, flat2_q3p3, flat2_q2p2;
-    __m128i flat2_q1p1, flat2_q0p0, flat_q2p2, flat_q1p1, flat_q0p0;
-
-    filt = _mm_and_si128(_mm_subs_epi8(qs1ps1, qs1), hev);
-    work_a = _mm_subs_epi8(qs0, qs0ps0);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    // (vp9_filter + 3 * (qs0 - ps0)) & mask
-    filt = _mm_and_si128(filt, mask);
-
-    filter1 = _mm_adds_epi8(filt, t4);
-    filter2 = _mm_adds_epi8(filt, t3);
-
-    filter1 = _mm_unpacklo_epi8(zero, filter1);
-    filter1 = _mm_srai_epi16(filter1, 0xB);
-    filter2 = _mm_unpacklo_epi8(zero, filter2);
-    filter2 = _mm_srai_epi16(filter2, 0xB);
-
-    // Filter1 >> 3
-    filt = _mm_packs_epi16(filter2, _mm_subs_epi16(zero, filter1));
-    qs0ps0 = _mm_xor_si128(_mm_adds_epi8(qs0ps0, filt), t80);
-
-    // filt >> 1
-    filt = _mm_adds_epi16(filter1, t1);
-    filt = _mm_srai_epi16(filt, 1);
-    filt = _mm_andnot_si128(_mm_srai_epi16(_mm_unpacklo_epi8(zero, hev), 0x8),
-                            filt);
-    filt = _mm_packs_epi16(filt, _mm_subs_epi16(zero, filt));
-    qs1ps1 = _mm_xor_si128(_mm_adds_epi8(qs1ps1, filt), t80);
-    // loopfilter done
-
-    {
-      __m128i work;
-      flat = _mm_max_epu8(abs_diff(q2p2, q0p0), abs_diff(q3p3, q0p0));
-      flat = _mm_max_epu8(abs_p1p0, flat);
-      flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8));
-      flat = _mm_subs_epu8(flat, one);
-      flat = _mm_cmpeq_epi8(flat, zero);
-      flat = _mm_and_si128(flat, mask);
-
-      q5p5 = _mm_loadl_epi64((__m128i *)(s - 6 * p));
-      q5p5 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q5p5),
-                                           (__m64 *)(s + 5 * p)));
-
-      q6p6 = _mm_loadl_epi64((__m128i *)(s - 7 * p));
-      q6p6 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q6p6),
-                                           (__m64 *)(s + 6 * p)));
-      flat2 = _mm_max_epu8(abs_diff(q4p4, q0p0), abs_diff(q5p5, q0p0));
-
-      q7p7 = _mm_loadl_epi64((__m128i *)(s - 8 * p));
-      q7p7 = _mm_castps_si128(_mm_loadh_pi(_mm_castsi128_ps(q7p7),
-                                           (__m64 *)(s + 7 * p)));
-      work = _mm_max_epu8(abs_diff(q6p6, q0p0), abs_diff(q7p7, q0p0));
-      flat2 = _mm_max_epu8(work, flat2);
-      flat2 = _mm_max_epu8(flat2, _mm_srli_si128(flat2, 8));
-      flat2 = _mm_subs_epu8(flat2, one);
-      flat2 = _mm_cmpeq_epi8(flat2, zero);
-      flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
-    }
-
-    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    // flat and wide flat calculations
-    {
-      const __m128i eight = _mm_set1_epi16(8);
-      const __m128i four = _mm_set1_epi16(4);
-      __m128i p7_16, p6_16, p5_16, p4_16, p3_16, p2_16, p1_16, p0_16;
-      __m128i q7_16, q6_16, q5_16, q4_16, q3_16, q2_16, q1_16, q0_16;
-      __m128i pixelFilter_p, pixelFilter_q;
-      __m128i pixetFilter_p2p1p0, pixetFilter_q2q1q0;
-      __m128i sum_p7, sum_q7, sum_p3, sum_q3, res_p, res_q;
-
-      p7_16 = _mm_unpacklo_epi8(q7p7, zero);;
-      p6_16 = _mm_unpacklo_epi8(q6p6, zero);
-      p5_16 = _mm_unpacklo_epi8(q5p5, zero);
-      p4_16 = _mm_unpacklo_epi8(q4p4, zero);
-      p3_16 = _mm_unpacklo_epi8(q3p3, zero);
-      p2_16 = _mm_unpacklo_epi8(q2p2, zero);
-      p1_16 = _mm_unpacklo_epi8(q1p1, zero);
-      p0_16 = _mm_unpacklo_epi8(q0p0, zero);
-      q0_16 = _mm_unpackhi_epi8(q0p0, zero);
-      q1_16 = _mm_unpackhi_epi8(q1p1, zero);
-      q2_16 = _mm_unpackhi_epi8(q2p2, zero);
-      q3_16 = _mm_unpackhi_epi8(q3p3, zero);
-      q4_16 = _mm_unpackhi_epi8(q4p4, zero);
-      q5_16 = _mm_unpackhi_epi8(q5p5, zero);
-      q6_16 = _mm_unpackhi_epi8(q6p6, zero);
-      q7_16 = _mm_unpackhi_epi8(q7p7, zero);
-
-      pixelFilter_p = _mm_add_epi16(_mm_add_epi16(p6_16, p5_16),
-                                    _mm_add_epi16(p4_16, p3_16));
-      pixelFilter_q = _mm_add_epi16(_mm_add_epi16(q6_16, q5_16),
-                                    _mm_add_epi16(q4_16, q3_16));
-
-      pixetFilter_p2p1p0 = _mm_add_epi16(p0_16, _mm_add_epi16(p2_16, p1_16));
-      pixelFilter_p =  _mm_add_epi16(pixelFilter_p, pixetFilter_p2p1p0);
-
-      pixetFilter_q2q1q0 = _mm_add_epi16(q0_16, _mm_add_epi16(q2_16, q1_16));
-      pixelFilter_q =  _mm_add_epi16(pixelFilter_q, pixetFilter_q2q1q0);
-      pixelFilter_p =  _mm_add_epi16(eight, _mm_add_epi16(pixelFilter_p,
-                                                         pixelFilter_q));
-      pixetFilter_p2p1p0 =   _mm_add_epi16(four,
-                                           _mm_add_epi16(pixetFilter_p2p1p0,
-                                                         pixetFilter_q2q1q0));
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                           _mm_add_epi16(p7_16, p0_16)), 4);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                                           _mm_add_epi16(q7_16, q0_16)), 4);
-      flat2_q0p0 = _mm_packus_epi16(res_p, res_q);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                                           _mm_add_epi16(p3_16, p0_16)), 3);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                                           _mm_add_epi16(q3_16, q0_16)), 3);
-
-      flat_q0p0 = _mm_packus_epi16(res_p, res_q);
-
-      sum_p7 = _mm_add_epi16(p7_16, p7_16);
-      sum_q7 = _mm_add_epi16(q7_16, q7_16);
-      sum_p3 = _mm_add_epi16(p3_16, p3_16);
-      sum_q3 = _mm_add_epi16(q3_16, q3_16);
-
-      pixelFilter_q = _mm_sub_epi16(pixelFilter_p, p6_16);
-      pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q6_16);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                             _mm_add_epi16(sum_p7, p1_16)), 4);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                             _mm_add_epi16(sum_q7, q1_16)), 4);
-      flat2_q1p1 = _mm_packus_epi16(res_p, res_q);
-
-      pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_p2p1p0, p2_16);
-      pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q2_16);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                             _mm_add_epi16(sum_p3, p1_16)), 3);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0,
-                             _mm_add_epi16(sum_q3, q1_16)), 3);
-      flat_q1p1 = _mm_packus_epi16(res_p, res_q);
-
-      sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-      sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-      sum_p3 = _mm_add_epi16(sum_p3, p3_16);
-      sum_q3 = _mm_add_epi16(sum_q3, q3_16);
-
-      pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q5_16);
-      pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p5_16);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                             _mm_add_epi16(sum_p7, p2_16)), 4);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                             _mm_add_epi16(sum_q7, q2_16)), 4);
-      flat2_q2p2 = _mm_packus_epi16(res_p, res_q);
-
-      pixetFilter_p2p1p0 = _mm_sub_epi16(pixetFilter_p2p1p0, q1_16);
-      pixetFilter_q2q1q0 = _mm_sub_epi16(pixetFilter_q2q1q0, p1_16);
-
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixetFilter_p2p1p0,
-                                           _mm_add_epi16(sum_p3, p2_16)), 3);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixetFilter_q2q1q0,
-                                           _mm_add_epi16(sum_q3, q2_16)), 3);
-      flat_q2p2 = _mm_packus_epi16(res_p, res_q);
-
-      sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-      sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-      pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q4_16);
-      pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p4_16);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                             _mm_add_epi16(sum_p7, p3_16)), 4);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                             _mm_add_epi16(sum_q7, q3_16)), 4);
-      flat2_q3p3 = _mm_packus_epi16(res_p, res_q);
-
-      sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-      sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-      pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q3_16);
-      pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p3_16);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                             _mm_add_epi16(sum_p7, p4_16)), 4);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                             _mm_add_epi16(sum_q7, q4_16)), 4);
-      flat2_q4p4 = _mm_packus_epi16(res_p, res_q);
-
-      sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-      sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-      pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q2_16);
-      pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p2_16);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                             _mm_add_epi16(sum_p7, p5_16)), 4);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                             _mm_add_epi16(sum_q7, q5_16)), 4);
-      flat2_q5p5 = _mm_packus_epi16(res_p, res_q);
-
-      sum_p7 = _mm_add_epi16(sum_p7, p7_16);
-      sum_q7 = _mm_add_epi16(sum_q7, q7_16);
-      pixelFilter_p = _mm_sub_epi16(pixelFilter_p, q1_16);
-      pixelFilter_q = _mm_sub_epi16(pixelFilter_q, p1_16);
-      res_p = _mm_srli_epi16(_mm_add_epi16(pixelFilter_p,
-                             _mm_add_epi16(sum_p7, p6_16)), 4);
-      res_q = _mm_srli_epi16(_mm_add_epi16(pixelFilter_q,
-                             _mm_add_epi16(sum_q7, q6_16)), 4);
-      flat2_q6p6 = _mm_packus_epi16(res_p, res_q);
-    }
-    // wide flat
-    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-    flat = _mm_shuffle_epi32(flat, 68);
-    flat2 = _mm_shuffle_epi32(flat2, 68);
-
-    q2p2 = _mm_andnot_si128(flat, q2p2);
-    flat_q2p2 = _mm_and_si128(flat, flat_q2p2);
-    q2p2 = _mm_or_si128(q2p2, flat_q2p2);
-
-    qs1ps1 = _mm_andnot_si128(flat, qs1ps1);
-    flat_q1p1 = _mm_and_si128(flat, flat_q1p1);
-    q1p1 = _mm_or_si128(qs1ps1, flat_q1p1);
-
-    qs0ps0 = _mm_andnot_si128(flat, qs0ps0);
-    flat_q0p0 = _mm_and_si128(flat, flat_q0p0);
-    q0p0 = _mm_or_si128(qs0ps0, flat_q0p0);
-
-    q6p6 = _mm_andnot_si128(flat2, q6p6);
-    flat2_q6p6 = _mm_and_si128(flat2, flat2_q6p6);
-    q6p6 = _mm_or_si128(q6p6, flat2_q6p6);
-    _mm_storel_epi64((__m128i *)(s - 7 * p), q6p6);
-    _mm_storeh_pi((__m64 *)(s + 6 * p), _mm_castsi128_ps(q6p6));
-
-    q5p5 = _mm_andnot_si128(flat2, q5p5);
-    flat2_q5p5 = _mm_and_si128(flat2, flat2_q5p5);
-    q5p5 = _mm_or_si128(q5p5, flat2_q5p5);
-    _mm_storel_epi64((__m128i *)(s - 6 * p), q5p5);
-    _mm_storeh_pi((__m64 *)(s + 5 * p), _mm_castsi128_ps(q5p5));
-
-    q4p4 = _mm_andnot_si128(flat2, q4p4);
-    flat2_q4p4 = _mm_and_si128(flat2, flat2_q4p4);
-    q4p4 = _mm_or_si128(q4p4, flat2_q4p4);
-    _mm_storel_epi64((__m128i *)(s - 5 * p), q4p4);
-    _mm_storeh_pi((__m64 *)(s + 4 * p), _mm_castsi128_ps(q4p4));
-
-    q3p3 = _mm_andnot_si128(flat2, q3p3);
-    flat2_q3p3 = _mm_and_si128(flat2, flat2_q3p3);
-    q3p3 = _mm_or_si128(q3p3, flat2_q3p3);
-    _mm_storel_epi64((__m128i *)(s - 4 * p), q3p3);
-    _mm_storeh_pi((__m64 *)(s + 3 * p), _mm_castsi128_ps(q3p3));
-
-    q2p2 = _mm_andnot_si128(flat2, q2p2);
-    flat2_q2p2 = _mm_and_si128(flat2, flat2_q2p2);
-    q2p2 = _mm_or_si128(q2p2, flat2_q2p2);
-    _mm_storel_epi64((__m128i *)(s - 3 * p), q2p2);
-    _mm_storeh_pi((__m64 *)(s + 2 * p), _mm_castsi128_ps(q2p2));
-
-    q1p1 = _mm_andnot_si128(flat2, q1p1);
-    flat2_q1p1 = _mm_and_si128(flat2, flat2_q1p1);
-    q1p1 = _mm_or_si128(q1p1, flat2_q1p1);
-    _mm_storel_epi64((__m128i *)(s - 2 * p), q1p1);
-    _mm_storeh_pi((__m64 *)(s + 1 * p), _mm_castsi128_ps(q1p1));
-
-    q0p0 = _mm_andnot_si128(flat2, q0p0);
-    flat2_q0p0 = _mm_and_si128(flat2, flat2_q0p0);
-    q0p0 = _mm_or_si128(q0p0, flat2_q0p0);
-    _mm_storel_epi64((__m128i *)(s - 1 * p), q0p0);
-    _mm_storeh_pi((__m64 *)(s - 0 * p),  _mm_castsi128_ps(q0p0));
-  }
-}
-
-static INLINE __m128i filter_add2_sub2(const __m128i *const total,
-                                       const __m128i *const a1,
-                                       const __m128i *const a2,
-                                       const __m128i *const s1,
-                                       const __m128i *const s2) {
-  __m128i x = _mm_add_epi16(*a1, *total);
-  x = _mm_add_epi16(_mm_sub_epi16(x, _mm_add_epi16(*s1, *s2)), *a2);
-  return x;
-}
-
-static INLINE __m128i filter8_mask(const __m128i *const flat,
-                                   const __m128i *const other_filt,
-                                   const __m128i *const f8_lo,
-                                   const __m128i *const f8_hi) {
-  const __m128i f8 = _mm_packus_epi16(_mm_srli_epi16(*f8_lo, 3),
-                                      _mm_srli_epi16(*f8_hi, 3));
-  const __m128i result = _mm_and_si128(*flat, f8);
-  return _mm_or_si128(_mm_andnot_si128(*flat, *other_filt), result);
-}
-
-static INLINE __m128i filter16_mask(const __m128i *const flat,
-                                    const __m128i *const other_filt,
-                                    const __m128i *const f_lo,
-                                    const __m128i *const f_hi) {
-  const __m128i f = _mm_packus_epi16(_mm_srli_epi16(*f_lo, 4),
-                                     _mm_srli_epi16(*f_hi, 4));
-  const __m128i result = _mm_and_si128(*flat, f);
-  return _mm_or_si128(_mm_andnot_si128(*flat, *other_filt), result);
-}
-
-static void mb_lpf_horizontal_edge_w_sse2_16(unsigned char *s,
-                                             int p,
-                                             const unsigned char *_blimit,
-                                             const unsigned char *_limit,
-                                             const unsigned char *_thresh) {
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i one = _mm_set1_epi8(1);
-  const __m128i blimit = _mm_load_si128((const __m128i *)_blimit);
-  const __m128i limit = _mm_load_si128((const __m128i *)_limit);
-  const __m128i thresh = _mm_load_si128((const __m128i *)_thresh);
-  __m128i mask, hev, flat, flat2;
-  __m128i p7, p6, p5;
-  __m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4;
-  __m128i q5, q6, q7;
-
-  __m128i op2, op1, op0, oq0, oq1, oq2;
-
-  __m128i max_abs_p1p0q1q0;
-
-  p7 = _mm_loadu_si128((__m128i *)(s - 8 * p));
-  p6 = _mm_loadu_si128((__m128i *)(s - 7 * p));
-  p5 = _mm_loadu_si128((__m128i *)(s - 6 * p));
-  p4 = _mm_loadu_si128((__m128i *)(s - 5 * p));
-  p3 = _mm_loadu_si128((__m128i *)(s - 4 * p));
-  p2 = _mm_loadu_si128((__m128i *)(s - 3 * p));
-  p1 = _mm_loadu_si128((__m128i *)(s - 2 * p));
-  p0 = _mm_loadu_si128((__m128i *)(s - 1 * p));
-  q0 = _mm_loadu_si128((__m128i *)(s - 0 * p));
-  q1 = _mm_loadu_si128((__m128i *)(s + 1 * p));
-  q2 = _mm_loadu_si128((__m128i *)(s + 2 * p));
-  q3 = _mm_loadu_si128((__m128i *)(s + 3 * p));
-  q4 = _mm_loadu_si128((__m128i *)(s + 4 * p));
-  q5 = _mm_loadu_si128((__m128i *)(s + 5 * p));
-  q6 = _mm_loadu_si128((__m128i *)(s + 6 * p));
-  q7 = _mm_loadu_si128((__m128i *)(s + 7 * p));
-
-  {
-    const __m128i abs_p1p0 = abs_diff(p1, p0);
-    const __m128i abs_q1q0 = abs_diff(q1, q0);
-    const __m128i fe = _mm_set1_epi8(0xfe);
-    const __m128i ff = _mm_cmpeq_epi8(zero, zero);
-    __m128i abs_p0q0 = abs_diff(p0, q0);
-    __m128i abs_p1q1 = abs_diff(p1, q1);
-    __m128i work;
-    max_abs_p1p0q1q0 = _mm_max_epu8(abs_p1p0, abs_q1q0);
-
-    abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
-    abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
-    mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
-    mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
-    // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-    mask = _mm_max_epu8(max_abs_p1p0q1q0, mask);
-    // mask |= (abs(p1 - p0) > limit) * -1;
-    // mask |= (abs(q1 - q0) > limit) * -1;
-    work = _mm_max_epu8(abs_diff(p2, p1), abs_diff(p3, p2));
-    mask = _mm_max_epu8(work, mask);
-    work = _mm_max_epu8(abs_diff(q2, q1), abs_diff(q3, q2));
-    mask = _mm_max_epu8(work, mask);
-    mask = _mm_subs_epu8(mask, limit);
-    mask = _mm_cmpeq_epi8(mask, zero);
-  }
-
-  {
-    __m128i work;
-    work = _mm_max_epu8(abs_diff(p2, p0), abs_diff(q2, q0));
-    flat = _mm_max_epu8(work, max_abs_p1p0q1q0);
-    work = _mm_max_epu8(abs_diff(p3, p0), abs_diff(q3, q0));
-    flat = _mm_max_epu8(work, flat);
-    work = _mm_max_epu8(abs_diff(p4, p0), abs_diff(q4, q0));
-    flat = _mm_subs_epu8(flat, one);
-    flat = _mm_cmpeq_epi8(flat, zero);
-    flat = _mm_and_si128(flat, mask);
-    flat2 = _mm_max_epu8(abs_diff(p5, p0), abs_diff(q5, q0));
-    flat2 = _mm_max_epu8(work, flat2);
-    work = _mm_max_epu8(abs_diff(p6, p0), abs_diff(q6, q0));
-    flat2 = _mm_max_epu8(work, flat2);
-    work = _mm_max_epu8(abs_diff(p7, p0), abs_diff(q7, q0));
-    flat2 = _mm_max_epu8(work, flat2);
-    flat2 = _mm_subs_epu8(flat2, one);
-    flat2 = _mm_cmpeq_epi8(flat2, zero);
-    flat2 = _mm_and_si128(flat2, flat);  // flat2 & flat & mask
-  }
-
-  // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-  // filter4
-  {
-    const __m128i t4 = _mm_set1_epi8(4);
-    const __m128i t3 = _mm_set1_epi8(3);
-    const __m128i t80 = _mm_set1_epi8(0x80);
-    const __m128i te0 = _mm_set1_epi8(0xe0);
-    const __m128i t1f = _mm_set1_epi8(0x1f);
-    const __m128i t1 = _mm_set1_epi8(0x1);
-    const __m128i t7f = _mm_set1_epi8(0x7f);
-    const __m128i ff = _mm_cmpeq_epi8(t4, t4);
-
-    __m128i filt;
-    __m128i work_a;
-    __m128i filter1, filter2;
-
-    op1 = _mm_xor_si128(p1, t80);
-    op0 = _mm_xor_si128(p0, t80);
-    oq0 = _mm_xor_si128(q0, t80);
-    oq1 = _mm_xor_si128(q1, t80);
-
-    hev = _mm_subs_epu8(max_abs_p1p0q1q0, thresh);
-    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
-    filt = _mm_and_si128(_mm_subs_epi8(op1, oq1), hev);
-
-    work_a = _mm_subs_epi8(oq0, op0);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    // (vp9_filter + 3 * (qs0 - ps0)) & mask
-    filt = _mm_and_si128(filt, mask);
-    filter1 = _mm_adds_epi8(filt, t4);
-    filter2 = _mm_adds_epi8(filt, t3);
-
-    // Filter1 >> 3
-    work_a = _mm_cmpgt_epi8(zero, filter1);
-    filter1 = _mm_srli_epi16(filter1, 3);
-    work_a = _mm_and_si128(work_a, te0);
-    filter1 = _mm_and_si128(filter1, t1f);
-    filter1 = _mm_or_si128(filter1, work_a);
-    oq0 = _mm_xor_si128(_mm_subs_epi8(oq0, filter1), t80);
-
-    // Filter2 >> 3
-    work_a = _mm_cmpgt_epi8(zero, filter2);
-    filter2 = _mm_srli_epi16(filter2, 3);
-    work_a = _mm_and_si128(work_a, te0);
-    filter2 = _mm_and_si128(filter2, t1f);
-    filter2 = _mm_or_si128(filter2, work_a);
-    op0 = _mm_xor_si128(_mm_adds_epi8(op0, filter2), t80);
-
-    // filt >> 1
-    filt = _mm_adds_epi8(filter1, t1);
-    work_a = _mm_cmpgt_epi8(zero, filt);
-    filt = _mm_srli_epi16(filt, 1);
-    work_a = _mm_and_si128(work_a, t80);
-    filt = _mm_and_si128(filt, t7f);
-    filt = _mm_or_si128(filt, work_a);
-    filt = _mm_andnot_si128(hev, filt);
-    op1 = _mm_xor_si128(_mm_adds_epi8(op1, filt), t80);
-    oq1 = _mm_xor_si128(_mm_subs_epi8(oq1, filt), t80);
-    // loopfilter done
-
-    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    // filter8
-    {
-      const __m128i four = _mm_set1_epi16(4);
-      const __m128i p3_lo = _mm_unpacklo_epi8(p3, zero);
-      const __m128i p2_lo = _mm_unpacklo_epi8(p2, zero);
-      const __m128i p1_lo = _mm_unpacklo_epi8(p1, zero);
-      const __m128i p0_lo = _mm_unpacklo_epi8(p0, zero);
-      const __m128i q0_lo = _mm_unpacklo_epi8(q0, zero);
-      const __m128i q1_lo = _mm_unpacklo_epi8(q1, zero);
-      const __m128i q2_lo = _mm_unpacklo_epi8(q2, zero);
-      const __m128i q3_lo = _mm_unpacklo_epi8(q3, zero);
-
-      const __m128i p3_hi = _mm_unpackhi_epi8(p3, zero);
-      const __m128i p2_hi = _mm_unpackhi_epi8(p2, zero);
-      const __m128i p1_hi = _mm_unpackhi_epi8(p1, zero);
-      const __m128i p0_hi = _mm_unpackhi_epi8(p0, zero);
-      const __m128i q0_hi = _mm_unpackhi_epi8(q0, zero);
-      const __m128i q1_hi = _mm_unpackhi_epi8(q1, zero);
-      const __m128i q2_hi = _mm_unpackhi_epi8(q2, zero);
-      const __m128i q3_hi = _mm_unpackhi_epi8(q3, zero);
-      __m128i f8_lo, f8_hi;
-
-      f8_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, four),
-                            _mm_add_epi16(p3_lo, p2_lo));
-      f8_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, f8_lo),
-                            _mm_add_epi16(p2_lo, p1_lo));
-      f8_lo = _mm_add_epi16(_mm_add_epi16(p0_lo, q0_lo), f8_lo);
-
-      f8_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, four),
-                            _mm_add_epi16(p3_hi, p2_hi));
-      f8_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, f8_hi),
-                            _mm_add_epi16(p2_hi, p1_hi));
-      f8_hi = _mm_add_epi16(_mm_add_epi16(p0_hi, q0_hi), f8_hi);
-
-      op2 = filter8_mask(&flat, &p2, &f8_lo, &f8_hi);
-
-      f8_lo = filter_add2_sub2(&f8_lo, &q1_lo, &p1_lo, &p2_lo, &p3_lo);
-      f8_hi = filter_add2_sub2(&f8_hi, &q1_hi, &p1_hi, &p2_hi, &p3_hi);
-      op1 = filter8_mask(&flat, &op1, &f8_lo, &f8_hi);
-
-      f8_lo = filter_add2_sub2(&f8_lo, &q2_lo, &p0_lo, &p1_lo, &p3_lo);
-      f8_hi = filter_add2_sub2(&f8_hi, &q2_hi, &p0_hi, &p1_hi, &p3_hi);
-      op0 = filter8_mask(&flat, &op0, &f8_lo, &f8_hi);
-
-      f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q0_lo, &p0_lo, &p3_lo);
-      f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q0_hi, &p0_hi, &p3_hi);
-      oq0 = filter8_mask(&flat, &oq0, &f8_lo, &f8_hi);
-
-      f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q1_lo, &q0_lo, &p2_lo);
-      f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q1_hi, &q0_hi, &p2_hi);
-      oq1 = filter8_mask(&flat, &oq1, &f8_lo, &f8_hi);
-
-      f8_lo = filter_add2_sub2(&f8_lo, &q3_lo, &q2_lo, &q1_lo, &p1_lo);
-      f8_hi = filter_add2_sub2(&f8_hi, &q3_hi, &q2_hi, &q1_hi, &p1_hi);
-      oq2 = filter8_mask(&flat, &q2, &f8_lo, &f8_hi);
-    }
-
-    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-    // wide flat calculations
-    {
-      const __m128i eight = _mm_set1_epi16(8);
-      const __m128i p7_lo = _mm_unpacklo_epi8(p7, zero);
-      const __m128i p6_lo = _mm_unpacklo_epi8(p6, zero);
-      const __m128i p5_lo = _mm_unpacklo_epi8(p5, zero);
-      const __m128i p4_lo = _mm_unpacklo_epi8(p4, zero);
-      const __m128i p3_lo = _mm_unpacklo_epi8(p3, zero);
-      const __m128i p2_lo = _mm_unpacklo_epi8(p2, zero);
-      const __m128i p1_lo = _mm_unpacklo_epi8(p1, zero);
-      const __m128i p0_lo = _mm_unpacklo_epi8(p0, zero);
-      const __m128i q0_lo = _mm_unpacklo_epi8(q0, zero);
-      const __m128i q1_lo = _mm_unpacklo_epi8(q1, zero);
-      const __m128i q2_lo = _mm_unpacklo_epi8(q2, zero);
-      const __m128i q3_lo = _mm_unpacklo_epi8(q3, zero);
-      const __m128i q4_lo = _mm_unpacklo_epi8(q4, zero);
-      const __m128i q5_lo = _mm_unpacklo_epi8(q5, zero);
-      const __m128i q6_lo = _mm_unpacklo_epi8(q6, zero);
-      const __m128i q7_lo = _mm_unpacklo_epi8(q7, zero);
-
-      const __m128i p7_hi = _mm_unpackhi_epi8(p7, zero);
-      const __m128i p6_hi = _mm_unpackhi_epi8(p6, zero);
-      const __m128i p5_hi = _mm_unpackhi_epi8(p5, zero);
-      const __m128i p4_hi = _mm_unpackhi_epi8(p4, zero);
-      const __m128i p3_hi = _mm_unpackhi_epi8(p3, zero);
-      const __m128i p2_hi = _mm_unpackhi_epi8(p2, zero);
-      const __m128i p1_hi = _mm_unpackhi_epi8(p1, zero);
-      const __m128i p0_hi = _mm_unpackhi_epi8(p0, zero);
-      const __m128i q0_hi = _mm_unpackhi_epi8(q0, zero);
-      const __m128i q1_hi = _mm_unpackhi_epi8(q1, zero);
-      const __m128i q2_hi = _mm_unpackhi_epi8(q2, zero);
-      const __m128i q3_hi = _mm_unpackhi_epi8(q3, zero);
-      const __m128i q4_hi = _mm_unpackhi_epi8(q4, zero);
-      const __m128i q5_hi = _mm_unpackhi_epi8(q5, zero);
-      const __m128i q6_hi = _mm_unpackhi_epi8(q6, zero);
-      const __m128i q7_hi = _mm_unpackhi_epi8(q7, zero);
-
-      __m128i f_lo;
-      __m128i f_hi;
-
-      f_lo = _mm_sub_epi16(_mm_slli_epi16(p7_lo, 3), p7_lo);  // p7 * 7
-      f_lo = _mm_add_epi16(_mm_slli_epi16(p6_lo, 1),
-                           _mm_add_epi16(p4_lo, f_lo));
-      f_lo = _mm_add_epi16(_mm_add_epi16(p3_lo, f_lo),
-                           _mm_add_epi16(p2_lo, p1_lo));
-      f_lo = _mm_add_epi16(_mm_add_epi16(p0_lo, q0_lo), f_lo);
-      f_lo = _mm_add_epi16(_mm_add_epi16(p5_lo, eight), f_lo);
-
-      f_hi = _mm_sub_epi16(_mm_slli_epi16(p7_hi, 3), p7_hi);  // p7 * 7
-      f_hi = _mm_add_epi16(_mm_slli_epi16(p6_hi, 1),
-                           _mm_add_epi16(p4_hi, f_hi));
-      f_hi = _mm_add_epi16(_mm_add_epi16(p3_hi, f_hi),
-                           _mm_add_epi16(p2_hi, p1_hi));
-      f_hi = _mm_add_epi16(_mm_add_epi16(p0_hi, q0_hi), f_hi);
-      f_hi = _mm_add_epi16(_mm_add_epi16(p5_hi, eight), f_hi);
-
-      p6 = filter16_mask(&flat2, &p6, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 7 * p), p6);
-
-      f_lo = filter_add2_sub2(&f_lo, &q1_lo, &p5_lo, &p6_lo, &p7_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q1_hi, &p5_hi, &p6_hi, &p7_hi);
-      p5 = filter16_mask(&flat2, &p5, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 6 * p), p5);
-
-      f_lo = filter_add2_sub2(&f_lo, &q2_lo, &p4_lo, &p5_lo, &p7_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q2_hi, &p4_hi, &p5_hi, &p7_hi);
-      p4 = filter16_mask(&flat2, &p4, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 5 * p), p4);
-
-      f_lo = filter_add2_sub2(&f_lo, &q3_lo, &p3_lo, &p4_lo, &p7_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q3_hi, &p3_hi, &p4_hi, &p7_hi);
-      p3 = filter16_mask(&flat2, &p3, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 4 * p), p3);
-
-      f_lo = filter_add2_sub2(&f_lo, &q4_lo, &p2_lo, &p3_lo, &p7_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q4_hi, &p2_hi, &p3_hi, &p7_hi);
-      op2 = filter16_mask(&flat2, &op2, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 3 * p), op2);
-
-      f_lo = filter_add2_sub2(&f_lo, &q5_lo, &p1_lo, &p2_lo, &p7_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q5_hi, &p1_hi, &p2_hi, &p7_hi);
-      op1 = filter16_mask(&flat2, &op1, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 2 * p), op1);
-
-      f_lo = filter_add2_sub2(&f_lo, &q6_lo, &p0_lo, &p1_lo, &p7_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q6_hi, &p0_hi, &p1_hi, &p7_hi);
-      op0 = filter16_mask(&flat2, &op0, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 1 * p), op0);
-
-      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q0_lo, &p0_lo, &p7_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q0_hi, &p0_hi, &p7_hi);
-      oq0 = filter16_mask(&flat2, &oq0, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s - 0 * p), oq0);
-
-      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q1_lo, &p6_lo, &q0_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q1_hi, &p6_hi, &q0_hi);
-      oq1 = filter16_mask(&flat2, &oq1, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s + 1 * p), oq1);
-
-      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q2_lo, &p5_lo, &q1_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q2_hi, &p5_hi, &q1_hi);
-      oq2 = filter16_mask(&flat2, &oq2, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s + 2 * p), oq2);
-
-      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q3_lo, &p4_lo, &q2_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q3_hi, &p4_hi, &q2_hi);
-      q3 = filter16_mask(&flat2, &q3, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s + 3 * p), q3);
-
-      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q4_lo, &p3_lo, &q3_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q4_hi, &p3_hi, &q3_hi);
-      q4 = filter16_mask(&flat2, &q4, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s + 4 * p), q4);
-
-      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q5_lo, &p2_lo, &q4_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q5_hi, &p2_hi, &q4_hi);
-      q5 = filter16_mask(&flat2, &q5, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s + 5 * p), q5);
-
-      f_lo = filter_add2_sub2(&f_lo, &q7_lo, &q6_lo, &p1_lo, &q5_lo);
-      f_hi = filter_add2_sub2(&f_hi, &q7_hi, &q6_hi, &p1_hi, &q5_hi);
-      q6 = filter16_mask(&flat2, &q6, &f_lo, &f_hi);
-      _mm_storeu_si128((__m128i *)(s + 6 * p), q6);
-    }
-    // wide flat
-    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-  }
-}
-
-// TODO(yunqingwang): remove count and call these 2 functions(8 or 16) directly.
-void vp9_lpf_horizontal_16_sse2(unsigned char *s, int p,
-                                const unsigned char *_blimit,
-                                const unsigned char *_limit,
-                                const unsigned char *_thresh, int count) {
-  if (count == 1)
-    mb_lpf_horizontal_edge_w_sse2_8(s, p, _blimit, _limit, _thresh);
-  else
-    mb_lpf_horizontal_edge_w_sse2_16(s, p, _blimit, _limit, _thresh);
-}
-
-void vp9_lpf_horizontal_8_sse2(unsigned char *s, int p,
-                               const unsigned char *_blimit,
-                               const unsigned char *_limit,
-                               const unsigned char *_thresh, int count) {
-  DECLARE_ALIGNED(16, unsigned char, flat_op2[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_op1[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_op0[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_oq2[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_oq1[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_oq0[16]);
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i blimit = _mm_load_si128((const __m128i *)_blimit);
-  const __m128i limit = _mm_load_si128((const __m128i *)_limit);
-  const __m128i thresh = _mm_load_si128((const __m128i *)_thresh);
-  __m128i mask, hev, flat;
-  __m128i p3, p2, p1, p0, q0, q1, q2, q3;
-  __m128i q3p3, q2p2, q1p1, q0p0, p1q1, p0q0;
-
-  (void)count;
-
-  q3p3 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 4 * p)),
-                            _mm_loadl_epi64((__m128i *)(s + 3 * p)));
-  q2p2 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 3 * p)),
-                            _mm_loadl_epi64((__m128i *)(s + 2 * p)));
-  q1p1 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 2 * p)),
-                            _mm_loadl_epi64((__m128i *)(s + 1 * p)));
-  q0p0 = _mm_unpacklo_epi64(_mm_loadl_epi64((__m128i *)(s - 1 * p)),
-                            _mm_loadl_epi64((__m128i *)(s - 0 * p)));
-  p1q1 = _mm_shuffle_epi32(q1p1, 78);
-  p0q0 = _mm_shuffle_epi32(q0p0, 78);
-
-  {
-    // filter_mask and hev_mask
-    const __m128i one = _mm_set1_epi8(1);
-    const __m128i fe = _mm_set1_epi8(0xfe);
-    const __m128i ff = _mm_cmpeq_epi8(fe, fe);
-    __m128i abs_p1q1, abs_p0q0, abs_q1q0, abs_p1p0, work;
-    abs_p1p0 = abs_diff(q1p1, q0p0);
-    abs_q1q0 =  _mm_srli_si128(abs_p1p0, 8);
-
-    abs_p0q0 = abs_diff(q0p0, p0q0);
-    abs_p1q1 = abs_diff(q1p1, p1q1);
-    flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
-    hev = _mm_subs_epu8(flat, thresh);
-    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
-
-    abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
-    abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
-    mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
-    mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
-    // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-    mask = _mm_max_epu8(abs_p1p0, mask);
-    // mask |= (abs(p1 - p0) > limit) * -1;
-    // mask |= (abs(q1 - q0) > limit) * -1;
-
-    work = _mm_max_epu8(abs_diff(q2p2, q1p1),
-                        abs_diff(q3p3, q2p2));
-    mask = _mm_max_epu8(work, mask);
-    mask = _mm_max_epu8(mask, _mm_srli_si128(mask, 8));
-    mask = _mm_subs_epu8(mask, limit);
-    mask = _mm_cmpeq_epi8(mask, zero);
-
-    // flat_mask4
-
-    flat = _mm_max_epu8(abs_diff(q2p2, q0p0),
-                        abs_diff(q3p3, q0p0));
-    flat = _mm_max_epu8(abs_p1p0, flat);
-    flat = _mm_max_epu8(flat, _mm_srli_si128(flat, 8));
-    flat = _mm_subs_epu8(flat, one);
-    flat = _mm_cmpeq_epi8(flat, zero);
-    flat = _mm_and_si128(flat, mask);
-  }
-
-  {
-    const __m128i four = _mm_set1_epi16(4);
-    unsigned char *src = s;
-    {
-      __m128i workp_a, workp_b, workp_shft;
-      p3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 4 * p)), zero);
-      p2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 3 * p)), zero);
-      p1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 2 * p)), zero);
-      p0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 1 * p)), zero);
-      q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 0 * p)), zero);
-      q1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 1 * p)), zero);
-      q2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 2 * p)), zero);
-      q3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 3 * p)), zero);
-
-      workp_a = _mm_add_epi16(_mm_add_epi16(p3, p3), _mm_add_epi16(p2, p1));
-      workp_a = _mm_add_epi16(_mm_add_epi16(workp_a, four), p0);
-      workp_b = _mm_add_epi16(_mm_add_epi16(q0, p2), p3);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_op2[0],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_b = _mm_add_epi16(_mm_add_epi16(q0, q1), p1);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_op1[0],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q2);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p1), p0);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_op0[0],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q3);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p0), q0);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_oq0[0],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p2), q3);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q0), q1);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_oq1[0],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p1), q3);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q1), q2);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_oq2[0],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-    }
-  }
-  // lp filter
-  {
-    const __m128i t4 = _mm_set1_epi8(4);
-    const __m128i t3 = _mm_set1_epi8(3);
-    const __m128i t80 = _mm_set1_epi8(0x80);
-    const __m128i t1 = _mm_set1_epi8(0x1);
-    const __m128i ps1 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s - 2 * p)),
-                                      t80);
-    const __m128i ps0 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s - 1 * p)),
-                                      t80);
-    const __m128i qs0 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s + 0 * p)),
-                                      t80);
-    const __m128i qs1 = _mm_xor_si128(_mm_loadl_epi64((__m128i *)(s + 1 * p)),
-                                      t80);
-    __m128i filt;
-    __m128i work_a;
-    __m128i filter1, filter2;
-
-    filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev);
-    work_a = _mm_subs_epi8(qs0, ps0);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    // (vp9_filter + 3 * (qs0 - ps0)) & mask
-    filt = _mm_and_si128(filt, mask);
-
-    filter1 = _mm_adds_epi8(filt, t4);
-    filter2 = _mm_adds_epi8(filt, t3);
-
-    // Filter1 >> 3
-    filter1 = _mm_unpacklo_epi8(zero, filter1);
-    filter1 = _mm_srai_epi16(filter1, 11);
-    filter1 = _mm_packs_epi16(filter1, filter1);
-
-    // Filter2 >> 3
-    filter2 = _mm_unpacklo_epi8(zero, filter2);
-    filter2 = _mm_srai_epi16(filter2, 11);
-    filter2 = _mm_packs_epi16(filter2, zero);
-
-    // filt >> 1
-    filt = _mm_adds_epi8(filter1, t1);
-    filt = _mm_unpacklo_epi8(zero, filt);
-    filt = _mm_srai_epi16(filt, 9);
-    filt = _mm_packs_epi16(filt, zero);
-
-    filt = _mm_andnot_si128(hev, filt);
-
-    work_a = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80);
-    q0 = _mm_loadl_epi64((__m128i *)flat_oq0);
-    work_a = _mm_andnot_si128(flat, work_a);
-    q0 = _mm_and_si128(flat, q0);
-    q0 = _mm_or_si128(work_a, q0);
-
-    work_a = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80);
-    q1 = _mm_loadl_epi64((__m128i *)flat_oq1);
-    work_a = _mm_andnot_si128(flat, work_a);
-    q1 = _mm_and_si128(flat, q1);
-    q1 = _mm_or_si128(work_a, q1);
-
-    work_a = _mm_loadu_si128((__m128i *)(s + 2 * p));
-    q2 = _mm_loadl_epi64((__m128i *)flat_oq2);
-    work_a = _mm_andnot_si128(flat, work_a);
-    q2 = _mm_and_si128(flat, q2);
-    q2 = _mm_or_si128(work_a, q2);
-
-    work_a = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80);
-    p0 = _mm_loadl_epi64((__m128i *)flat_op0);
-    work_a = _mm_andnot_si128(flat, work_a);
-    p0 = _mm_and_si128(flat, p0);
-    p0 = _mm_or_si128(work_a, p0);
-
-    work_a = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80);
-    p1 = _mm_loadl_epi64((__m128i *)flat_op1);
-    work_a = _mm_andnot_si128(flat, work_a);
-    p1 = _mm_and_si128(flat, p1);
-    p1 = _mm_or_si128(work_a, p1);
-
-    work_a = _mm_loadu_si128((__m128i *)(s - 3 * p));
-    p2 = _mm_loadl_epi64((__m128i *)flat_op2);
-    work_a = _mm_andnot_si128(flat, work_a);
-    p2 = _mm_and_si128(flat, p2);
-    p2 = _mm_or_si128(work_a, p2);
-
-    _mm_storel_epi64((__m128i *)(s - 3 * p), p2);
-    _mm_storel_epi64((__m128i *)(s - 2 * p), p1);
-    _mm_storel_epi64((__m128i *)(s - 1 * p), p0);
-    _mm_storel_epi64((__m128i *)(s + 0 * p), q0);
-    _mm_storel_epi64((__m128i *)(s + 1 * p), q1);
-    _mm_storel_epi64((__m128i *)(s + 2 * p), q2);
-  }
-}
-
-void vp9_lpf_horizontal_8_dual_sse2(uint8_t *s, int p,
-                                    const uint8_t *_blimit0,
-                                    const uint8_t *_limit0,
-                                    const uint8_t *_thresh0,
-                                    const uint8_t *_blimit1,
-                                    const uint8_t *_limit1,
-                                    const uint8_t *_thresh1) {
-  DECLARE_ALIGNED(16, unsigned char, flat_op2[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_op1[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_op0[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_oq2[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_oq1[16]);
-  DECLARE_ALIGNED(16, unsigned char, flat_oq0[16]);
-  const __m128i zero = _mm_set1_epi16(0);
-  const __m128i blimit =
-      _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_blimit0),
-                         _mm_load_si128((const __m128i *)_blimit1));
-  const __m128i limit =
-      _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_limit0),
-                         _mm_load_si128((const __m128i *)_limit1));
-  const __m128i thresh =
-      _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_thresh0),
-                         _mm_load_si128((const __m128i *)_thresh1));
-
-  __m128i mask, hev, flat;
-  __m128i p3, p2, p1, p0, q0, q1, q2, q3;
-
-  p3 = _mm_loadu_si128((__m128i *)(s - 4 * p));
-  p2 = _mm_loadu_si128((__m128i *)(s - 3 * p));
-  p1 = _mm_loadu_si128((__m128i *)(s - 2 * p));
-  p0 = _mm_loadu_si128((__m128i *)(s - 1 * p));
-  q0 = _mm_loadu_si128((__m128i *)(s - 0 * p));
-  q1 = _mm_loadu_si128((__m128i *)(s + 1 * p));
-  q2 = _mm_loadu_si128((__m128i *)(s + 2 * p));
-  q3 = _mm_loadu_si128((__m128i *)(s + 3 * p));
-  {
-    const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0),
-                                          _mm_subs_epu8(p0, p1));
-    const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0),
-                                          _mm_subs_epu8(q0, q1));
-    const __m128i one = _mm_set1_epi8(1);
-    const __m128i fe = _mm_set1_epi8(0xfe);
-    const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
-    __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0),
-                                    _mm_subs_epu8(q0, p0));
-    __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1),
-                                    _mm_subs_epu8(q1, p1));
-    __m128i work;
-
-    // filter_mask and hev_mask
-    flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
-    hev = _mm_subs_epu8(flat, thresh);
-    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
-
-    abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
-    abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
-    mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
-    mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
-    // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-    mask = _mm_max_epu8(flat, mask);
-    // mask |= (abs(p1 - p0) > limit) * -1;
-    // mask |= (abs(q1 - q0) > limit) * -1;
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1),
-                                     _mm_subs_epu8(p1, p2)),
-                         _mm_or_si128(_mm_subs_epu8(p3, p2),
-                                      _mm_subs_epu8(p2, p3)));
-    mask = _mm_max_epu8(work, mask);
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2, q1),
-                                     _mm_subs_epu8(q1, q2)),
-                         _mm_or_si128(_mm_subs_epu8(q3, q2),
-                                      _mm_subs_epu8(q2, q3)));
-    mask = _mm_max_epu8(work, mask);
-    mask = _mm_subs_epu8(mask, limit);
-    mask = _mm_cmpeq_epi8(mask, zero);
-
-    // flat_mask4
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p0),
-                                     _mm_subs_epu8(p0, p2)),
-                         _mm_or_si128(_mm_subs_epu8(q2, q0),
-                                      _mm_subs_epu8(q0, q2)));
-    flat = _mm_max_epu8(work, flat);
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p3, p0),
-                                     _mm_subs_epu8(p0, p3)),
-                         _mm_or_si128(_mm_subs_epu8(q3, q0),
-                                      _mm_subs_epu8(q0, q3)));
-    flat = _mm_max_epu8(work, flat);
-    flat = _mm_subs_epu8(flat, one);
-    flat = _mm_cmpeq_epi8(flat, zero);
-    flat = _mm_and_si128(flat, mask);
-  }
-  {
-    const __m128i four = _mm_set1_epi16(4);
-    unsigned char *src = s;
-    int i = 0;
-
-    do {
-      __m128i workp_a, workp_b, workp_shft;
-      p3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 4 * p)), zero);
-      p2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 3 * p)), zero);
-      p1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 2 * p)), zero);
-      p0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 1 * p)), zero);
-      q0 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src - 0 * p)), zero);
-      q1 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 1 * p)), zero);
-      q2 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 2 * p)), zero);
-      q3 = _mm_unpacklo_epi8(_mm_loadl_epi64((__m128i *)(src + 3 * p)), zero);
-
-      workp_a = _mm_add_epi16(_mm_add_epi16(p3, p3), _mm_add_epi16(p2, p1));
-      workp_a = _mm_add_epi16(_mm_add_epi16(workp_a, four), p0);
-      workp_b = _mm_add_epi16(_mm_add_epi16(q0, p2), p3);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_op2[i * 8],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_b = _mm_add_epi16(_mm_add_epi16(q0, q1), p1);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_op1[i * 8],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q2);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p1), p0);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_op0[i * 8],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p3), q3);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, p0), q0);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_oq0[i * 8],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p2), q3);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q0), q1);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_oq1[i * 8],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      workp_a = _mm_add_epi16(_mm_sub_epi16(workp_a, p1), q3);
-      workp_b = _mm_add_epi16(_mm_sub_epi16(workp_b, q1), q2);
-      workp_shft = _mm_srli_epi16(_mm_add_epi16(workp_a, workp_b), 3);
-      _mm_storel_epi64((__m128i *)&flat_oq2[i * 8],
-                       _mm_packus_epi16(workp_shft, workp_shft));
-
-      src += 8;
-    } while (++i < 2);
-  }
-  // lp filter
-  {
-    const __m128i t4 = _mm_set1_epi8(4);
-    const __m128i t3 = _mm_set1_epi8(3);
-    const __m128i t80 = _mm_set1_epi8(0x80);
-    const __m128i te0 = _mm_set1_epi8(0xe0);
-    const __m128i t1f = _mm_set1_epi8(0x1f);
-    const __m128i t1 = _mm_set1_epi8(0x1);
-    const __m128i t7f = _mm_set1_epi8(0x7f);
-
-    const __m128i ps1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 2 * p)),
-                                      t80);
-    const __m128i ps0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 1 * p)),
-                                      t80);
-    const __m128i qs0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 0 * p)),
-                                      t80);
-    const __m128i qs1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 1 * p)),
-                                      t80);
-    __m128i filt;
-    __m128i work_a;
-    __m128i filter1, filter2;
-
-    filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev);
-    work_a = _mm_subs_epi8(qs0, ps0);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    // (vp9_filter + 3 * (qs0 - ps0)) & mask
-    filt = _mm_and_si128(filt, mask);
-
-    filter1 = _mm_adds_epi8(filt, t4);
-    filter2 = _mm_adds_epi8(filt, t3);
-
-    // Filter1 >> 3
-    work_a = _mm_cmpgt_epi8(zero, filter1);
-    filter1 = _mm_srli_epi16(filter1, 3);
-    work_a = _mm_and_si128(work_a, te0);
-    filter1 = _mm_and_si128(filter1, t1f);
-    filter1 = _mm_or_si128(filter1, work_a);
-
-    // Filter2 >> 3
-    work_a = _mm_cmpgt_epi8(zero, filter2);
-    filter2 = _mm_srli_epi16(filter2, 3);
-    work_a = _mm_and_si128(work_a, te0);
-    filter2 = _mm_and_si128(filter2, t1f);
-    filter2 = _mm_or_si128(filter2, work_a);
-
-    // filt >> 1
-    filt = _mm_adds_epi8(filter1, t1);
-    work_a = _mm_cmpgt_epi8(zero, filt);
-    filt = _mm_srli_epi16(filt, 1);
-    work_a = _mm_and_si128(work_a, t80);
-    filt = _mm_and_si128(filt, t7f);
-    filt = _mm_or_si128(filt, work_a);
-
-    filt = _mm_andnot_si128(hev, filt);
-
-    work_a = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80);
-    q0 = _mm_load_si128((__m128i *)flat_oq0);
-    work_a = _mm_andnot_si128(flat, work_a);
-    q0 = _mm_and_si128(flat, q0);
-    q0 = _mm_or_si128(work_a, q0);
-
-    work_a = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80);
-    q1 = _mm_load_si128((__m128i *)flat_oq1);
-    work_a = _mm_andnot_si128(flat, work_a);
-    q1 = _mm_and_si128(flat, q1);
-    q1 = _mm_or_si128(work_a, q1);
-
-    work_a = _mm_loadu_si128((__m128i *)(s + 2 * p));
-    q2 = _mm_load_si128((__m128i *)flat_oq2);
-    work_a = _mm_andnot_si128(flat, work_a);
-    q2 = _mm_and_si128(flat, q2);
-    q2 = _mm_or_si128(work_a, q2);
-
-    work_a = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80);
-    p0 = _mm_load_si128((__m128i *)flat_op0);
-    work_a = _mm_andnot_si128(flat, work_a);
-    p0 = _mm_and_si128(flat, p0);
-    p0 = _mm_or_si128(work_a, p0);
-
-    work_a = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80);
-    p1 = _mm_load_si128((__m128i *)flat_op1);
-    work_a = _mm_andnot_si128(flat, work_a);
-    p1 = _mm_and_si128(flat, p1);
-    p1 = _mm_or_si128(work_a, p1);
-
-    work_a = _mm_loadu_si128((__m128i *)(s - 3 * p));
-    p2 = _mm_load_si128((__m128i *)flat_op2);
-    work_a = _mm_andnot_si128(flat, work_a);
-    p2 = _mm_and_si128(flat, p2);
-    p2 = _mm_or_si128(work_a, p2);
-
-    _mm_storeu_si128((__m128i *)(s - 3 * p), p2);
-    _mm_storeu_si128((__m128i *)(s - 2 * p), p1);
-    _mm_storeu_si128((__m128i *)(s - 1 * p), p0);
-    _mm_storeu_si128((__m128i *)(s + 0 * p), q0);
-    _mm_storeu_si128((__m128i *)(s + 1 * p), q1);
-    _mm_storeu_si128((__m128i *)(s + 2 * p), q2);
-  }
-}
-
-void vp9_lpf_horizontal_4_dual_sse2(unsigned char *s, int p,
-                                    const unsigned char *_blimit0,
-                                    const unsigned char *_limit0,
-                                    const unsigned char *_thresh0,
-                                    const unsigned char *_blimit1,
-                                    const unsigned char *_limit1,
-                                    const unsigned char *_thresh1) {
-  const __m128i blimit =
-      _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_blimit0),
-                         _mm_load_si128((const __m128i *)_blimit1));
-  const __m128i limit =
-      _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_limit0),
-                         _mm_load_si128((const __m128i *)_limit1));
-  const __m128i thresh =
-      _mm_unpacklo_epi64(_mm_load_si128((const __m128i *)_thresh0),
-                         _mm_load_si128((const __m128i *)_thresh1));
-  const __m128i zero = _mm_set1_epi16(0);
-  __m128i p3, p2, p1, p0, q0, q1, q2, q3;
-  __m128i mask, hev, flat;
-
-  p3 = _mm_loadu_si128((__m128i *)(s - 4 * p));
-  p2 = _mm_loadu_si128((__m128i *)(s - 3 * p));
-  p1 = _mm_loadu_si128((__m128i *)(s - 2 * p));
-  p0 = _mm_loadu_si128((__m128i *)(s - 1 * p));
-  q0 = _mm_loadu_si128((__m128i *)(s - 0 * p));
-  q1 = _mm_loadu_si128((__m128i *)(s + 1 * p));
-  q2 = _mm_loadu_si128((__m128i *)(s + 2 * p));
-  q3 = _mm_loadu_si128((__m128i *)(s + 3 * p));
-
-  // filter_mask and hev_mask
-  {
-    const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0),
-                                          _mm_subs_epu8(p0, p1));
-    const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0),
-                                          _mm_subs_epu8(q0, q1));
-    const __m128i fe = _mm_set1_epi8(0xfe);
-    const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
-    __m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0),
-                                    _mm_subs_epu8(q0, p0));
-    __m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1),
-                                    _mm_subs_epu8(q1, p1));
-    __m128i work;
-
-    flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
-    hev = _mm_subs_epu8(flat, thresh);
-    hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
-
-    abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
-    abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
-    mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
-    mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
-    // mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
-    mask = _mm_max_epu8(flat, mask);
-    // mask |= (abs(p1 - p0) > limit) * -1;
-    // mask |= (abs(q1 - q0) > limit) * -1;
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1),
-                                     _mm_subs_epu8(p1, p2)),
-                         _mm_or_si128(_mm_subs_epu8(p3, p2),
-                                      _mm_subs_epu8(p2, p3)));
-    mask = _mm_max_epu8(work, mask);
-    work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(q2, q1),
-                                     _mm_subs_epu8(q1, q2)),
-                         _mm_or_si128(_mm_subs_epu8(q3, q2),
-                                      _mm_subs_epu8(q2, q3)));
-    mask = _mm_max_epu8(work, mask);
-    mask = _mm_subs_epu8(mask, limit);
-    mask = _mm_cmpeq_epi8(mask, zero);
-  }
-
-  // filter4
-  {
-    const __m128i t4 = _mm_set1_epi8(4);
-    const __m128i t3 = _mm_set1_epi8(3);
-    const __m128i t80 = _mm_set1_epi8(0x80);
-    const __m128i te0 = _mm_set1_epi8(0xe0);
-    const __m128i t1f = _mm_set1_epi8(0x1f);
-    const __m128i t1 = _mm_set1_epi8(0x1);
-    const __m128i t7f = _mm_set1_epi8(0x7f);
-
-    const __m128i ps1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 2 * p)),
-                                      t80);
-    const __m128i ps0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s - 1 * p)),
-                                      t80);
-    const __m128i qs0 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 0 * p)),
-                                      t80);
-    const __m128i qs1 = _mm_xor_si128(_mm_loadu_si128((__m128i *)(s + 1 * p)),
-                                      t80);
-    __m128i filt;
-    __m128i work_a;
-    __m128i filter1, filter2;
-
-    filt = _mm_and_si128(_mm_subs_epi8(ps1, qs1), hev);
-    work_a = _mm_subs_epi8(qs0, ps0);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    filt = _mm_adds_epi8(filt, work_a);
-    // (vp9_filter + 3 * (qs0 - ps0)) & mask
-    filt = _mm_and_si128(filt, mask);
-
-    filter1 = _mm_adds_epi8(filt, t4);
-    filter2 = _mm_adds_epi8(filt, t3);
-
-    // Filter1 >> 3
-    work_a = _mm_cmpgt_epi8(zero, filter1);
-    filter1 = _mm_srli_epi16(filter1, 3);
-    work_a = _mm_and_si128(work_a, te0);
-    filter1 = _mm_and_si128(filter1, t1f);
-    filter1 = _mm_or_si128(filter1, work_a);
-
-    // Filter2 >> 3
-    work_a = _mm_cmpgt_epi8(zero, filter2);
-    filter2 = _mm_srli_epi16(filter2, 3);
-    work_a = _mm_and_si128(work_a, te0);
-    filter2 = _mm_and_si128(filter2, t1f);
-    filter2 = _mm_or_si128(filter2, work_a);
-
-    // filt >> 1
-    filt = _mm_adds_epi8(filter1, t1);
-    work_a = _mm_cmpgt_epi8(zero, filt);
-    filt = _mm_srli_epi16(filt, 1);
-    work_a = _mm_and_si128(work_a, t80);
-    filt = _mm_and_si128(filt, t7f);
-    filt = _mm_or_si128(filt, work_a);
-
-    filt = _mm_andnot_si128(hev, filt);
-
-    q0 = _mm_xor_si128(_mm_subs_epi8(qs0, filter1), t80);
-    q1 = _mm_xor_si128(_mm_subs_epi8(qs1, filt), t80);
-    p0 = _mm_xor_si128(_mm_adds_epi8(ps0, filter2), t80);
-    p1 = _mm_xor_si128(_mm_adds_epi8(ps1, filt), t80);
-
-    _mm_storeu_si128((__m128i *)(s - 2 * p), p1);
-    _mm_storeu_si128((__m128i *)(s - 1 * p), p0);
-    _mm_storeu_si128((__m128i *)(s + 0 * p), q0);
-    _mm_storeu_si128((__m128i *)(s + 1 * p), q1);
-  }
-}
-
-static INLINE void transpose8x16(unsigned char *in0, unsigned char *in1,
-                                 int in_p, unsigned char *out, int out_p) {
-  __m128i x0, x1, x2, x3, x4, x5, x6, x7;
-  __m128i x8, x9, x10, x11, x12, x13, x14, x15;
-
-  // 2-way interleave w/hoisting of unpacks
-  x0 = _mm_loadl_epi64((__m128i *)in0);  // 1
-  x1 = _mm_loadl_epi64((__m128i *)(in0 + in_p));  // 3
-  x0 = _mm_unpacklo_epi8(x0, x1);  // 1
-
-  x2 = _mm_loadl_epi64((__m128i *)(in0 + 2 * in_p));  // 5
-  x3 = _mm_loadl_epi64((__m128i *)(in0 + 3*in_p));  // 7
-  x1 = _mm_unpacklo_epi8(x2, x3);  // 2
-
-  x4 = _mm_loadl_epi64((__m128i *)(in0 + 4*in_p));  // 9
-  x5 = _mm_loadl_epi64((__m128i *)(in0 + 5*in_p));  // 11
-  x2 = _mm_unpacklo_epi8(x4, x5);  // 3
-
-  x6 = _mm_loadl_epi64((__m128i *)(in0 + 6*in_p));  // 13
-  x7 = _mm_loadl_epi64((__m128i *)(in0 + 7*in_p));  // 15
-  x3 = _mm_unpacklo_epi8(x6, x7);  // 4
-  x4 = _mm_unpacklo_epi16(x0, x1);  // 9
-
-  x8 = _mm_loadl_epi64((__m128i *)in1);  // 2
-  x9 = _mm_loadl_epi64((__m128i *)(in1 + in_p));  // 4
-  x8 = _mm_unpacklo_epi8(x8, x9);  // 5
-  x5 = _mm_unpacklo_epi16(x2, x3);  // 10
-
-  x10 = _mm_loadl_epi64((__m128i *)(in1 + 2 * in_p));  // 6
-  x11 = _mm_loadl_epi64((__m128i *)(in1 + 3*in_p));  // 8
-  x9 = _mm_unpacklo_epi8(x10, x11);  // 6
-
-  x12 = _mm_loadl_epi64((__m128i *)(in1 + 4*in_p));  // 10
-  x13 = _mm_loadl_epi64((__m128i *)(in1 + 5*in_p));  // 12
-  x10 = _mm_unpacklo_epi8(x12, x13);  // 7
-  x12 = _mm_unpacklo_epi16(x8, x9);  // 11
-
-  x14 = _mm_loadl_epi64((__m128i *)(in1 + 6*in_p));  // 14
-  x15 = _mm_loadl_epi64((__m128i *)(in1 + 7*in_p));  // 16
-  x11 = _mm_unpacklo_epi8(x14, x15);  // 8
-  x13 = _mm_unpacklo_epi16(x10, x11);  // 12
-
-  x6 = _mm_unpacklo_epi32(x4, x5);  // 13
-  x7 = _mm_unpackhi_epi32(x4, x5);  // 14
-  x14 = _mm_unpacklo_epi32(x12, x13);  // 15
-  x15 = _mm_unpackhi_epi32(x12, x13);  // 16
-
-  // Store first 4-line result
-  _mm_storeu_si128((__m128i *)out, _mm_unpacklo_epi64(x6, x14));
-  _mm_storeu_si128((__m128i *)(out + out_p), _mm_unpackhi_epi64(x6, x14));
-  _mm_storeu_si128((__m128i *)(out + 2 * out_p), _mm_unpacklo_epi64(x7, x15));
-  _mm_storeu_si128((__m128i *)(out + 3 * out_p), _mm_unpackhi_epi64(x7, x15));
-
-  x4 = _mm_unpackhi_epi16(x0, x1);
-  x5 = _mm_unpackhi_epi16(x2, x3);
-  x12 = _mm_unpackhi_epi16(x8, x9);
-  x13 = _mm_unpackhi_epi16(x10, x11);
-
-  x6 = _mm_unpacklo_epi32(x4, x5);
-  x7 = _mm_unpackhi_epi32(x4, x5);
-  x14 = _mm_unpacklo_epi32(x12, x13);
-  x15 = _mm_unpackhi_epi32(x12, x13);
-
-  // Store second 4-line result
-  _mm_storeu_si128((__m128i *)(out + 4 * out_p), _mm_unpacklo_epi64(x6, x14));
-  _mm_storeu_si128((__m128i *)(out + 5 * out_p), _mm_unpackhi_epi64(x6, x14));
-  _mm_storeu_si128((__m128i *)(out + 6 * out_p), _mm_unpacklo_epi64(x7, x15));
-  _mm_storeu_si128((__m128i *)(out + 7 * out_p), _mm_unpackhi_epi64(x7, x15));
-}
-
-static INLINE void transpose(unsigned char *src[], int in_p,
-                             unsigned char *dst[], int out_p,
-                             int num_8x8_to_transpose) {
-  int idx8x8 = 0;
-  __m128i x0, x1, x2, x3, x4, x5, x6, x7;
-  do {
-    unsigned char *in = src[idx8x8];
-    unsigned char *out = dst[idx8x8];
-
-    x0 = _mm_loadl_epi64((__m128i *)(in + 0*in_p));  // 00 01 02 03 04 05 06 07
-    x1 = _mm_loadl_epi64((__m128i *)(in + 1*in_p));  // 10 11 12 13 14 15 16 17
-    // 00 10 01 11 02 12 03 13 04 14 05 15 06 16 07 17
-    x0 = _mm_unpacklo_epi8(x0, x1);
-
-    x2 = _mm_loadl_epi64((__m128i *)(in + 2*in_p));  // 20 21 22 23 24 25 26 27
-    x3 = _mm_loadl_epi64((__m128i *)(in + 3*in_p));  // 30 31 32 33 34 35 36 37
-    // 20 30 21 31 22 32 23 33 24 34 25 35 26 36 27 37
-    x1 = _mm_unpacklo_epi8(x2, x3);
-
-    x4 = _mm_loadl_epi64((__m128i *)(in + 4*in_p));  // 40 41 42 43 44 45 46 47
-    x5 = _mm_loadl_epi64((__m128i *)(in + 5*in_p));  // 50 51 52 53 54 55 56 57
-    // 40 50 41 51 42 52 43 53 44 54 45 55 46 56 47 57
-    x2 = _mm_unpacklo_epi8(x4, x5);
-
-    x6 = _mm_loadl_epi64((__m128i *)(in + 6*in_p));  // 60 61 62 63 64 65 66 67
-    x7 = _mm_loadl_epi64((__m128i *)(in + 7*in_p));  // 70 71 72 73 74 75 76 77
-    // 60 70 61 71 62 72 63 73 64 74 65 75 66 76 67 77
-    x3 = _mm_unpacklo_epi8(x6, x7);
-
-    // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-    x4 = _mm_unpacklo_epi16(x0, x1);
-    // 40 50 60 70 41 51 61 71 42 52 62 72 43 53 63 73
-    x5 = _mm_unpacklo_epi16(x2, x3);
-    // 00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71
-    x6 = _mm_unpacklo_epi32(x4, x5);
-    _mm_storel_pd((double *)(out + 0*out_p),
-                  _mm_castsi128_pd(x6));  // 00 10 20 30 40 50 60 70
-    _mm_storeh_pd((double *)(out + 1*out_p),
-                  _mm_castsi128_pd(x6));  // 01 11 21 31 41 51 61 71
-    // 02 12 22 32 42 52 62 72 03 13 23 33 43 53 63 73
-    x7 = _mm_unpackhi_epi32(x4, x5);
-    _mm_storel_pd((double *)(out + 2*out_p),
-                  _mm_castsi128_pd(x7));  // 02 12 22 32 42 52 62 72
-    _mm_storeh_pd((double *)(out + 3*out_p),
-                  _mm_castsi128_pd(x7));  // 03 13 23 33 43 53 63 73
-
-    // 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37
-    x4 = _mm_unpackhi_epi16(x0, x1);
-    // 44 54 64 74 45 55 65 75 46 56 66 76 47 57 67 77
-    x5 = _mm_unpackhi_epi16(x2, x3);
-    // 04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75
-    x6 = _mm_unpacklo_epi32(x4, x5);
-    _mm_storel_pd((double *)(out + 4*out_p),
-                  _mm_castsi128_pd(x6));  // 04 14 24 34 44 54 64 74
-    _mm_storeh_pd((double *)(out + 5*out_p),
-                  _mm_castsi128_pd(x6));  // 05 15 25 35 45 55 65 75
-    // 06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77
-    x7 = _mm_unpackhi_epi32(x4, x5);
-
-    _mm_storel_pd((double *)(out + 6*out_p),
-                  _mm_castsi128_pd(x7));  // 06 16 26 36 46 56 66 76
-    _mm_storeh_pd((double *)(out + 7*out_p),
-                  _mm_castsi128_pd(x7));  // 07 17 27 37 47 57 67 77
-  } while (++idx8x8 < num_8x8_to_transpose);
-}
-
-void vp9_lpf_vertical_4_dual_sse2(uint8_t *s, int p, const uint8_t *blimit0,
-                                  const uint8_t *limit0,
-                                  const uint8_t *thresh0,
-                                  const uint8_t *blimit1,
-                                  const uint8_t *limit1,
-                                  const uint8_t *thresh1) {
-  DECLARE_ALIGNED(16, unsigned char, t_dst[16 * 8]);
-  unsigned char *src[2];
-  unsigned char *dst[2];
-
-  // Transpose 8x16
-  transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16);
-
-  // Loop filtering
-  vp9_lpf_horizontal_4_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0, thresh0,
-                                 blimit1, limit1, thresh1);
-  src[0] = t_dst;
-  src[1] = t_dst + 8;
-  dst[0] = s - 4;
-  dst[1] = s - 4 + p * 8;
-
-  // Transpose back
-  transpose(src, 16, dst, p, 2);
-}
-
-void vp9_lpf_vertical_8_sse2(unsigned char *s, int p,
-                             const unsigned char *blimit,
-                             const unsigned char *limit,
-                             const unsigned char *thresh, int count) {
-  DECLARE_ALIGNED(8, unsigned char, t_dst[8 * 8]);
-  unsigned char *src[1];
-  unsigned char *dst[1];
-  (void)count;
-
-  // Transpose 8x8
-  src[0] = s - 4;
-  dst[0] = t_dst;
-
-  transpose(src, p, dst, 8, 1);
-
-  // Loop filtering
-  vp9_lpf_horizontal_8_sse2(t_dst + 4 * 8, 8, blimit, limit, thresh, 1);
-
-  src[0] = t_dst;
-  dst[0] = s - 4;
-
-  // Transpose back
-  transpose(src, 8, dst, p, 1);
-}
-
-void vp9_lpf_vertical_8_dual_sse2(uint8_t *s, int p, const uint8_t *blimit0,
-                                  const uint8_t *limit0,
-                                  const uint8_t *thresh0,
-                                  const uint8_t *blimit1,
-                                  const uint8_t *limit1,
-                                  const uint8_t *thresh1) {
-  DECLARE_ALIGNED(16, unsigned char, t_dst[16 * 8]);
-  unsigned char *src[2];
-  unsigned char *dst[2];
-
-  // Transpose 8x16
-  transpose8x16(s - 4, s - 4 + p * 8, p, t_dst, 16);
-
-  // Loop filtering
-  vp9_lpf_horizontal_8_dual_sse2(t_dst + 4 * 16, 16, blimit0, limit0, thresh0,
-                                 blimit1, limit1, thresh1);
-  src[0] = t_dst;
-  src[1] = t_dst + 8;
-
-  dst[0] = s - 4;
-  dst[1] = s - 4 + p * 8;
-
-  // Transpose back
-  transpose(src, 16, dst, p, 2);
-}
-
-void vp9_lpf_vertical_16_sse2(unsigned char *s, int p,
-                              const unsigned char *blimit,
-                              const unsigned char *limit,
-                              const unsigned char *thresh) {
-  DECLARE_ALIGNED(8, unsigned char, t_dst[8 * 16]);
-  unsigned char *src[2];
-  unsigned char *dst[2];
-
-  src[0] = s - 8;
-  src[1] = s;
-  dst[0] = t_dst;
-  dst[1] = t_dst + 8 * 8;
-
-  // Transpose 16x8
-  transpose(src, p, dst, 8, 2);
-
-  // Loop filtering
-  mb_lpf_horizontal_edge_w_sse2_8(t_dst + 8 * 8, 8, blimit, limit, thresh);
-
-  src[0] = t_dst;
-  src[1] = t_dst + 8 * 8;
-  dst[0] = s - 8;
-  dst[1] = s;
-
-  // Transpose back
-  transpose(src, 8, dst, p, 2);
-}
-
-void vp9_lpf_vertical_16_dual_sse2(unsigned char *s, int p,
-                                   const uint8_t *blimit, const uint8_t *limit,
-                                   const uint8_t *thresh) {
-  DECLARE_ALIGNED(16, unsigned char, t_dst[256]);
-
-  // Transpose 16x16
-  transpose8x16(s - 8, s - 8 + 8 * p, p, t_dst, 16);
-  transpose8x16(s, s + 8 * p, p, t_dst + 8 * 16, 16);
-
-  // Loop filtering
-  mb_lpf_horizontal_edge_w_sse2_16(t_dst + 8 * 16, 16, blimit, limit,
-                                   thresh);
-
-  // Transpose back
-  transpose8x16(t_dst, t_dst + 8 * 16, 16, s - 8, p);
-  transpose8x16(t_dst + 8, t_dst + 8 + 8 * 16, 16, s - 8 + 8 * p, p);
-}
diff --git a/vp9/common/x86/vp9_loopfilter_mmx.asm b/vp9/common/x86/vp9_loopfilter_mmx.asm
deleted file mode 100644
index f5f7d5af7..000000000
--- a/vp9/common/x86/vp9_loopfilter_mmx.asm
+++ /dev/null
@@ -1,611 +0,0 @@
-;
-;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-;  Use of this source code is governed by a BSD-style license
-;  that can be found in the LICENSE file in the root of the source
-;  tree. An additional intellectual property rights grant can be found
-;  in the file PATENTS.  All contributing project authors may
-;  be found in the AUTHORS file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-
-
-;void vp9_lpf_horizontal_4_mmx
-;(
-;    unsigned char *src_ptr,
-;    int src_pixel_step,
-;    const char *blimit,
-;    const char *limit,
-;    const char *thresh,
-;    int  count
-;)
-global sym(vp9_lpf_horizontal_4_mmx) PRIVATE
-sym(vp9_lpf_horizontal_4_mmx):
-    push        rbp
-    mov         rbp, rsp
-    SHADOW_ARGS_TO_STACK 6
-    GET_GOT     rbx
-    push        rsi
-    push        rdi
-    ; end prolog
-
-    ALIGN_STACK 16, rax
-    sub         rsp, 32                         ; reserve 32 bytes
-    %define t0 [rsp + 0]    ;__declspec(align(16)) char t0[8];
-    %define t1 [rsp + 16]   ;__declspec(align(16)) char t1[8];
-
-        mov         rsi, arg(0) ;src_ptr
-        movsxd      rax, dword ptr arg(1) ;src_pixel_step     ; destination pitch?
-
-        movsxd      rcx, dword ptr arg(5) ;count
-.next8_h:
-        mov         rdx, arg(3) ;limit
-        movq        mm7, [rdx]
-        mov         rdi, rsi              ; rdi points to row +1 for indirect addressing
-        add         rdi, rax
-
-        ; calculate breakout conditions
-        movq        mm2, [rdi+2*rax]      ; q3
-        movq        mm1, [rsi+2*rax]      ; q2
-        movq        mm6, mm1              ; q2
-        psubusb     mm1, mm2              ; q2-=q3
-        psubusb     mm2, mm6              ; q3-=q2
-        por         mm1, mm2              ; abs(q3-q2)
-        psubusb     mm1, mm7              ;
-
-
-        movq        mm4, [rsi+rax]        ; q1
-        movq        mm3, mm4              ; q1
-        psubusb     mm4, mm6              ; q1-=q2
-        psubusb     mm6, mm3              ; q2-=q1
-        por         mm4, mm6              ; abs(q2-q1)
-
-        psubusb     mm4, mm7
-        por        mm1, mm4
-
-        movq        mm4, [rsi]            ; q0
-        movq        mm0, mm4              ; q0
-        psubusb     mm4, mm3              ; q0-=q1
-        psubusb     mm3, mm0              ; q1-=q0
-        por         mm4, mm3              ; abs(q0-q1)
-        movq        t0, mm4               ; save to t0
-        psubusb     mm4, mm7
-        por        mm1, mm4
-
-
-        neg         rax                   ; negate pitch to deal with above border
-
-        movq        mm2, [rsi+4*rax]      ; p3
-        movq        mm4, [rdi+4*rax]      ; p2
-        movq        mm5, mm4              ; p2
-        psubusb     mm4, mm2              ; p2-=p3
-        psubusb     mm2, mm5              ; p3-=p2
-        por         mm4, mm2              ; abs(p3 - p2)
-        psubusb     mm4, mm7
-        por        mm1, mm4
-
-
-        movq        mm4, [rsi+2*rax]      ; p1
-        movq        mm3, mm4              ; p1
-        psubusb     mm4, mm5              ; p1-=p2
-        psubusb     mm5, mm3              ; p2-=p1
-        por         mm4, mm5              ; abs(p2 - p1)
-        psubusb     mm4, mm7
-        por        mm1, mm4
-
-        movq        mm2, mm3              ; p1
-
-        movq        mm4, [rsi+rax]        ; p0
-        movq        mm5, mm4              ; p0
-        psubusb     mm4, mm3              ; p0-=p1
-        psubusb     mm3, mm5              ; p1-=p0
-        por         mm4, mm3              ; abs(p1 - p0)
-        movq        t1, mm4               ; save to t1
-        psubusb     mm4, mm7
-        por        mm1, mm4
-
-        movq        mm3, [rdi]            ; q1
-        movq        mm4, mm3              ; q1
-        psubusb     mm3, mm2              ; q1-=p1
-        psubusb     mm2, mm4              ; p1-=q1
-        por         mm2, mm3              ; abs(p1-q1)
-        pand        mm2, [GLOBAL(tfe)]    ; set lsb of each byte to zero
-        psrlw       mm2, 1                ; abs(p1-q1)/2
-
-        movq        mm6, mm5              ; p0
-        movq        mm3, [rsi]            ; q0
-        psubusb     mm5, mm3              ; p0-=q0
-        psubusb     mm3, mm6              ; q0-=p0
-        por         mm5, mm3              ; abs(p0 - q0)
-        paddusb     mm5, mm5              ; abs(p0-q0)*2
-        paddusb     mm5, mm2              ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
-        mov         rdx, arg(2) ;blimit           ; get blimit
-        movq        mm7, [rdx]            ; blimit
-
-        psubusb     mm5,    mm7           ; abs (p0 - q0) *2 + abs(p1-q1)/2  > blimit
-        por         mm1,    mm5
-        pxor        mm5,    mm5
-        pcmpeqb     mm1,    mm5           ; mask mm1
-
-        ; calculate high edge variance
-        mov         rdx, arg(4) ;thresh           ; get thresh
-        movq        mm7, [rdx]            ;
-        movq        mm4, t0               ; get abs (q1 - q0)
-        psubusb     mm4, mm7
-        movq        mm3, t1               ; get abs (p1 - p0)
-        psubusb     mm3, mm7
-        paddb       mm4, mm3              ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
-
-        pcmpeqb     mm4,        mm5
-
-        pcmpeqb     mm5,        mm5
-        pxor        mm4,        mm5
-
-
-        ; start work on filters
-        movq        mm2, [rsi+2*rax]      ; p1
-        movq        mm7, [rdi]            ; q1
-        pxor        mm2, [GLOBAL(t80)]    ; p1 offset to convert to signed values
-        pxor        mm7, [GLOBAL(t80)]    ; q1 offset to convert to signed values
-        psubsb      mm2, mm7              ; p1 - q1
-        pand        mm2, mm4              ; high var mask (hvm)(p1 - q1)
-        pxor        mm6, [GLOBAL(t80)]    ; offset to convert to signed values
-        pxor        mm0, [GLOBAL(t80)]    ; offset to convert to signed values
-        movq        mm3, mm0              ; q0
-        psubsb      mm0, mm6              ; q0 - p0
-        paddsb      mm2, mm0              ; 1 * (q0 - p0) + hvm(p1 - q1)
-        paddsb      mm2, mm0              ; 2 * (q0 - p0) + hvm(p1 - q1)
-        paddsb      mm2, mm0              ; 3 * (q0 - p0) + hvm(p1 - q1)
-        pand        mm1, mm2                  ; mask filter values we don't care about
-        movq        mm2, mm1
-        paddsb      mm1, [GLOBAL(t4)]     ; 3* (q0 - p0) + hvm(p1 - q1) + 4
-        paddsb      mm2, [GLOBAL(t3)]     ; 3* (q0 - p0) + hvm(p1 - q1) + 3
-
-        pxor        mm0, mm0             ;
-        pxor        mm5, mm5
-        punpcklbw   mm0, mm2            ;
-        punpckhbw   mm5, mm2            ;
-        psraw       mm0, 11             ;
-        psraw       mm5, 11
-        packsswb    mm0, mm5
-        movq        mm2, mm0            ;  (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
-
-        pxor        mm0, mm0              ; 0
-        movq        mm5, mm1              ; abcdefgh
-        punpcklbw   mm0, mm1              ; e0f0g0h0
-        psraw       mm0, 11               ; sign extended shift right by 3
-        pxor        mm1, mm1              ; 0
-        punpckhbw   mm1, mm5              ; a0b0c0d0
-        psraw       mm1, 11               ; sign extended shift right by 3
-        movq        mm5, mm0              ; save results
-
-        packsswb    mm0, mm1              ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
-        paddsw      mm5, [GLOBAL(ones)]
-        paddsw      mm1, [GLOBAL(ones)]
-        psraw       mm5, 1                ; partial shifted one more time for 2nd tap
-        psraw       mm1, 1                ; partial shifted one more time for 2nd tap
-        packsswb    mm5, mm1              ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
-        pandn       mm4, mm5              ; high edge variance additive
-
-        paddsb      mm6, mm2              ; p0+= p0 add
-        pxor        mm6, [GLOBAL(t80)]    ; unoffset
-        movq        [rsi+rax], mm6        ; write back
-
-        movq        mm6, [rsi+2*rax]      ; p1
-        pxor        mm6, [GLOBAL(t80)]    ; reoffset
-        paddsb      mm6, mm4              ; p1+= p1 add
-        pxor        mm6, [GLOBAL(t80)]    ; unoffset
-        movq        [rsi+2*rax], mm6      ; write back
-
-        psubsb      mm3, mm0              ; q0-= q0 add
-        pxor        mm3, [GLOBAL(t80)]    ; unoffset
-        movq        [rsi], mm3            ; write back
-
-        psubsb      mm7, mm4              ; q1-= q1 add
-        pxor        mm7, [GLOBAL(t80)]    ; unoffset
-        movq        [rdi], mm7            ; write back
-
-        add         rsi,8
-        neg         rax
-        dec         rcx
-        jnz         .next8_h
-
-    add rsp, 32
-    pop rsp
-    ; begin epilog
-    pop rdi
-    pop rsi
-    RESTORE_GOT
-    UNSHADOW_ARGS
-    pop         rbp
-    ret
-
-
-;void vp9_lpf_vertical_4_mmx
-;(
-;    unsigned char *src_ptr,
-;    int  src_pixel_step,
-;    const char *blimit,
-;    const char *limit,
-;    const char *thresh,
-;    int count
-;)
-global sym(vp9_lpf_vertical_4_mmx) PRIVATE
-sym(vp9_lpf_vertical_4_mmx):
-    push        rbp
-    mov         rbp, rsp
-    SHADOW_ARGS_TO_STACK 6
-    GET_GOT     rbx
-    push        rsi
-    push        rdi
-    ; end prolog
-
-    ALIGN_STACK 16, rax
-    sub          rsp, 64      ; reserve 64 bytes
-    %define t0   [rsp + 0]    ;__declspec(align(16)) char t0[8];
-    %define t1   [rsp + 16]   ;__declspec(align(16)) char t1[8];
-    %define srct [rsp + 32]   ;__declspec(align(16)) char srct[32];
-
-        mov         rsi,        arg(0) ;src_ptr
-        movsxd      rax,        dword ptr arg(1) ;src_pixel_step     ; destination pitch?
-
-        lea         rsi,        [rsi + rax*4 - 4]
-
-        movsxd      rcx,        dword ptr arg(5) ;count
-.next8_v:
-        mov         rdi,        rsi           ; rdi points to row +1 for indirect addressing
-        add         rdi,        rax
-
-
-        ;transpose
-        movq        mm6,        [rsi+2*rax]                 ; 67 66 65 64 63 62 61 60
-        movq        mm7,        mm6                         ; 77 76 75 74 73 72 71 70
-
-        punpckhbw   mm7,        [rdi+2*rax]                 ; 77 67 76 66 75 65 74 64
-        punpcklbw   mm6,        [rdi+2*rax]                 ; 73 63 72 62 71 61 70 60
-
-        movq        mm4,        [rsi]                       ; 47 46 45 44 43 42 41 40
-        movq        mm5,        mm4                         ; 47 46 45 44 43 42 41 40
-
-        punpckhbw   mm5,        [rsi+rax]                   ; 57 47 56 46 55 45 54 44
-        punpcklbw   mm4,        [rsi+rax]                   ; 53 43 52 42 51 41 50 40
-
-        movq        mm3,        mm5                         ; 57 47 56 46 55 45 54 44
-        punpckhwd   mm5,        mm7                         ; 77 67 57 47 76 66 56 46
-
-        punpcklwd   mm3,        mm7                         ; 75 65 55 45 74 64 54 44
-        movq        mm2,        mm4                         ; 53 43 52 42 51 41 50 40
-
-        punpckhwd   mm4,        mm6                         ; 73 63 53 43 72 62 52 42
-        punpcklwd   mm2,        mm6                         ; 71 61 51 41 70 60 50 40
-
-        neg         rax
-        movq        mm6,        [rsi+rax*2]                 ; 27 26 25 24 23 22 21 20
-
-        movq        mm1,        mm6                         ; 27 26 25 24 23 22 21 20
-        punpckhbw   mm6,        [rsi+rax]                   ; 37 27 36 36 35 25 34 24
-
-        punpcklbw   mm1,        [rsi+rax]                   ; 33 23 32 22 31 21 30 20
-        movq        mm7,        [rsi+rax*4];                ; 07 06 05 04 03 02 01 00
-
-        punpckhbw   mm7,        [rdi+rax*4]                 ; 17 07 16 06 15 05 14 04
-        movq        mm0,        mm7                         ; 17 07 16 06 15 05 14 04
-
-        punpckhwd   mm7,        mm6                         ; 37 27 17 07 36 26 16 06
-        punpcklwd   mm0,        mm6                         ; 35 25 15 05 34 24 14 04
-
-        movq        mm6,        mm7                         ; 37 27 17 07 36 26 16 06
-        punpckhdq   mm7,        mm5                         ; 77 67 57 47 37 27 17 07  = q3
-
-        punpckldq   mm6,        mm5                         ; 76 66 56 46 36 26 16 06  = q2
-
-        movq        mm5,        mm6                         ; 76 66 56 46 36 26 16 06
-        psubusb     mm5,        mm7                         ; q2-q3
-
-        psubusb     mm7,        mm6                         ; q3-q2
-        por         mm7,        mm5;                        ; mm7=abs (q3-q2)
-
-        movq        mm5,        mm0                         ; 35 25 15 05 34 24 14 04
-        punpckhdq   mm5,        mm3                         ; 75 65 55 45 35 25 15 05 = q1
-
-        punpckldq   mm0,        mm3                         ; 74 64 54 44 34 24 15 04 = q0
-        movq        mm3,        mm5                         ; 75 65 55 45 35 25 15 05 = q1
-
-        psubusb     mm3,        mm6                         ; q1-q2
-        psubusb     mm6,        mm5                         ; q2-q1
-
-        por         mm6,        mm3                         ; mm6=abs(q2-q1)
-        lea         rdx,        srct
-
-        movq        [rdx+24],   mm5                         ; save q1
-        movq        [rdx+16],   mm0                         ; save q0
-
-        movq        mm3,        [rsi+rax*4]                 ; 07 06 05 04 03 02 01 00
-        punpcklbw   mm3,        [rdi+rax*4]                 ; 13 03 12 02 11 01 10 00
-
-        movq        mm0,        mm3                         ; 13 03 12 02 11 01 10 00
-        punpcklwd   mm0,        mm1                         ; 31 21 11 01 30 20 10 00
-
-        punpckhwd   mm3,        mm1                         ; 33 23 13 03 32 22 12 02
-        movq        mm1,        mm0                         ; 31 21 11 01 30 20 10 00
-
-        punpckldq   mm0,        mm2                         ; 70 60 50 40 30 20 10 00  =p3
-        punpckhdq   mm1,        mm2                         ; 71 61 51 41 31 21 11 01  =p2
-
-        movq        mm2,        mm1                         ; 71 61 51 41 31 21 11 01  =p2
-        psubusb     mm2,        mm0                         ; p2-p3
-
-        psubusb     mm0,        mm1                         ; p3-p2
-        por         mm0,        mm2                         ; mm0=abs(p3-p2)
-
-        movq        mm2,        mm3                         ; 33 23 13 03 32 22 12 02
-        punpckldq   mm2,        mm4                         ; 72 62 52 42 32 22 12 02 = p1
-
-        punpckhdq   mm3,        mm4                         ; 73 63 53 43 33 23 13 03 = p0
-        movq        [rdx+8],    mm3                         ; save p0
-
-        movq        [rdx],      mm2                         ; save p1
-        movq        mm5,        mm2                         ; mm5 = p1
-
-        psubusb     mm2,        mm1                         ; p1-p2
-        psubusb     mm1,        mm5                         ; p2-p1
-
-        por         mm1,        mm2                         ; mm1=abs(p2-p1)
-        mov         rdx,        arg(3) ;limit
-
-        movq        mm4,        [rdx]                       ; mm4 = limit
-        psubusb     mm7,        mm4
-
-        psubusb     mm0,        mm4
-        psubusb     mm1,        mm4
-
-        psubusb     mm6,        mm4
-        por         mm7,        mm6
-
-        por         mm0,        mm1
-        por         mm0,        mm7                         ;   abs(q3-q2) > limit || abs(p3-p2) > limit ||abs(p2-p1) > limit || abs(q2-q1) > limit
-
-        movq        mm1,        mm5                         ; p1
-
-        movq        mm7,        mm3                         ; mm3=mm7=p0
-        psubusb     mm7,        mm5                         ; p0 - p1
-
-        psubusb     mm5,        mm3                         ; p1 - p0
-        por         mm5,        mm7                         ; abs(p1-p0)
-
-        movq        t0,         mm5                         ; save abs(p1-p0)
-        lea         rdx,        srct
-
-        psubusb     mm5,        mm4
-        por         mm0,        mm5                         ; mm0=mask
-
-        movq        mm5,        [rdx+16]                    ; mm5=q0
-        movq        mm7,        [rdx+24]                    ; mm7=q1
-
-        movq        mm6,        mm5                         ; mm6=q0
-        movq        mm2,        mm7                         ; q1
-        psubusb     mm5,        mm7                         ; q0-q1
-
-        psubusb     mm7,        mm6                         ; q1-q0
-        por         mm7,        mm5                         ; abs(q1-q0)
-
-        movq        t1,         mm7                         ; save abs(q1-q0)
-        psubusb     mm7,        mm4
-
-        por         mm0,        mm7                         ; mask
-
-        movq        mm5,        mm2                         ; q1
-        psubusb     mm5,        mm1                         ; q1-=p1
-        psubusb     mm1,        mm2                         ; p1-=q1
-        por         mm5,        mm1                         ; abs(p1-q1)
-        pand        mm5,        [GLOBAL(tfe)]               ; set lsb of each byte to zero
-        psrlw       mm5,        1                           ; abs(p1-q1)/2
-
-        mov         rdx,        arg(2) ;blimit                      ;
-
-        movq        mm4,        [rdx]                       ;blimit
-        movq        mm1,        mm3                         ; mm1=mm3=p0
-
-        movq        mm7,        mm6                         ; mm7=mm6=q0
-        psubusb     mm1,        mm7                         ; p0-q0
-
-        psubusb     mm7,        mm3                         ; q0-p0
-        por         mm1,        mm7                         ; abs(q0-p0)
-        paddusb     mm1,        mm1                         ; abs(q0-p0)*2
-        paddusb     mm1,        mm5                         ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
-        psubusb     mm1,        mm4                         ; abs (p0 - q0) *2 + abs(p1-q1)/2  > blimit
-        por         mm1,        mm0;                        ; mask
-
-        pxor        mm0,        mm0
-        pcmpeqb     mm1,        mm0
-
-        ; calculate high edge variance
-        mov         rdx,        arg(4) ;thresh            ; get thresh
-        movq        mm7,        [rdx]
-        ;
-        movq        mm4,        t0              ; get abs (q1 - q0)
-        psubusb     mm4,        mm7
-
-        movq        mm3,        t1              ; get abs (p1 - p0)
-        psubusb     mm3,        mm7
-
-        por         mm4,        mm3             ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
-        pcmpeqb     mm4,        mm0
-
-        pcmpeqb     mm0,        mm0
-        pxor        mm4,        mm0
-
-
-
-        ; start work on filters
-        lea         rdx,        srct
-
-        movq        mm2,        [rdx]           ; p1
-        movq        mm7,        [rdx+24]        ; q1
-
-        movq        mm6,        [rdx+8]         ; p0
-        movq        mm0,        [rdx+16]        ; q0
-
-        pxor        mm2,        [GLOBAL(t80)]   ; p1 offset to convert to signed values
-        pxor        mm7,        [GLOBAL(t80)]   ; q1 offset to convert to signed values
-
-        psubsb      mm2,        mm7             ; p1 - q1
-        pand        mm2,        mm4             ; high var mask (hvm)(p1 - q1)
-
-        pxor        mm6,        [GLOBAL(t80)]   ; offset to convert to signed values
-        pxor        mm0,        [GLOBAL(t80)]   ; offset to convert to signed values
-
-        movq        mm3,        mm0             ; q0
-        psubsb      mm0,        mm6             ; q0 - p0
-
-        paddsb      mm2,        mm0             ; 1 * (q0 - p0) + hvm(p1 - q1)
-        paddsb      mm2,        mm0             ; 2 * (q0 - p0) + hvm(p1 - q1)
-
-        paddsb      mm2,        mm0             ; 3 * (q0 - p0) + hvm(p1 - q1)
-        pand       mm1,        mm2              ; mask filter values we don't care about
-
-        movq        mm2,        mm1
-        paddsb      mm1,        [GLOBAL(t4)]      ; 3* (q0 - p0) + hvm(p1 - q1) + 4
-
-        paddsb      mm2,        [GLOBAL(t3)]      ; 3* (q0 - p0) + hvm(p1 - q1) + 3
-        pxor        mm0,        mm0          ;
-
-        pxor        mm5,        mm5
-        punpcklbw   mm0,        mm2         ;
-
-        punpckhbw   mm5,        mm2         ;
-        psraw       mm0,        11              ;
-
-        psraw       mm5,        11
-        packsswb    mm0,        mm5
-
-        movq        mm2,        mm0         ;  (3* (q0 - p0) + hvm(p1 - q1) + 3) >> 3;
-
-        pxor        mm0,        mm0           ; 0
-        movq        mm5,        mm1           ; abcdefgh
-
-        punpcklbw   mm0,        mm1           ; e0f0g0h0
-        psraw       mm0,        11                ; sign extended shift right by 3
-
-        pxor        mm1,        mm1           ; 0
-        punpckhbw   mm1,        mm5           ; a0b0c0d0
-
-        psraw       mm1,        11                ; sign extended shift right by 3
-        movq        mm5,        mm0              ; save results
-
-        packsswb    mm0,        mm1           ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>3
-        paddsw      mm5,        [GLOBAL(ones)]
-
-        paddsw      mm1,        [GLOBAL(ones)]
-        psraw       mm5,        1                 ; partial shifted one more time for 2nd tap
-
-        psraw       mm1,        1                 ; partial shifted one more time for 2nd tap
-        packsswb    mm5,        mm1           ; (3* (q0 - p0) + hvm(p1 - q1) + 4) >>4
-
-        pandn       mm4,        mm5             ; high edge variance additive
-
-        paddsb      mm6,        mm2             ; p0+= p0 add
-        pxor        mm6,        [GLOBAL(t80)]   ; unoffset
-
-        ; mm6=p0                               ;
-        movq        mm1,        [rdx]           ; p1
-        pxor        mm1,        [GLOBAL(t80)]   ; reoffset
-
-        paddsb      mm1,        mm4                 ; p1+= p1 add
-        pxor        mm1,        [GLOBAL(t80)]       ; unoffset
-        ; mm6 = p0 mm1 = p1
-
-        psubsb      mm3,        mm0                 ; q0-= q0 add
-        pxor        mm3,        [GLOBAL(t80)]       ; unoffset
-
-        ; mm3 = q0
-        psubsb      mm7,        mm4                 ; q1-= q1 add
-        pxor        mm7,        [GLOBAL(t80)]       ; unoffset
-        ; mm7 = q1
-
-        ; transpose and write back
-        ; mm1 =    72 62 52 42 32 22 12 02
-        ; mm6 =    73 63 53 43 33 23 13 03
-        ; mm3 =    74 64 54 44 34 24 14 04
-        ; mm7 =    75 65 55 45 35 25 15 05
-
-        movq        mm2,        mm1             ; 72 62 52 42 32 22 12 02
-        punpcklbw   mm2,        mm6             ; 33 32 23 22 13 12 03 02
-
-        movq        mm4,        mm3             ; 74 64 54 44 34 24 14 04
-        punpckhbw   mm1,        mm6             ; 73 72 63 62 53 52 43 42
-
-        punpcklbw   mm4,        mm7             ; 35 34 25 24 15 14 05 04
-        punpckhbw   mm3,        mm7             ; 75 74 65 64 55 54 45 44
-
-        movq        mm6,        mm2             ; 33 32 23 22 13 12 03 02
-        punpcklwd   mm2,        mm4             ; 15 14 13 12 05 04 03 02
-
-        punpckhwd   mm6,        mm4             ; 35 34 33 32 25 24 23 22
-        movq        mm5,        mm1             ; 73 72 63 62 53 52 43 42
-
-        punpcklwd   mm1,        mm3             ; 55 54 53 52 45 44 43 42
-        punpckhwd   mm5,        mm3             ; 75 74 73 72 65 64 63 62
-
-
-        ; mm2 = 15 14 13 12 05 04 03 02
-        ; mm6 = 35 34 33 32 25 24 23 22
-        ; mm5 = 55 54 53 52 45 44 43 42
-        ; mm1 = 75 74 73 72 65 64 63 62
-
-
-
-        movd        [rsi+rax*4+2], mm2
-        psrlq       mm2,        32
-
-        movd        [rdi+rax*4+2], mm2
-        movd        [rsi+rax*2+2], mm6
-
-        psrlq       mm6,        32
-        movd        [rsi+rax+2],mm6
-
-        movd        [rsi+2],    mm1
-        psrlq       mm1,        32
-
-        movd        [rdi+2],    mm1
-        neg         rax
-
-        movd        [rdi+rax+2],mm5
-        psrlq       mm5,        32
-
-        movd        [rdi+rax*2+2], mm5
-
-        lea         rsi,        [rsi+rax*8]
-        dec         rcx
-        jnz         .next8_v
-
-    add rsp, 64
-    pop rsp
-    ; begin epilog
-    pop rdi
-    pop rsi
-    RESTORE_GOT
-    UNSHADOW_ARGS
-    pop         rbp
-    ret
-
-SECTION_RODATA
-align 16
-tfe:
-    times 8 db 0xfe
-align 16
-t80:
-    times 8 db 0x80
-align 16
-t3:
-    times 8 db 0x03
-align 16
-t4:
-    times 8 db 0x04
-align 16
-ones:
-    times 4 dw 0x0001