/*
 * Copyright (c) 2023, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <arm_neon.h>
#include <assert.h>
#include <stdint.h>

#include "aom_dsp/arm/mem_neon.h"
#include "aom_dsp/arm/sum_neon.h"
#include "av1/encoder/arm/pickrst_neon.h"
#include "av1/encoder/pickrst.h"

static INLINE void highbd_calc_proj_params_r0_r1_neon(
    const uint8_t *src8, int width, int height, int src_stride,
    const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
    int32_t *flt1, int flt1_stride, int64_t H[2][2], int64_t C[2]) {
  assert(width % 8 == 0);
  const int size = width * height;
  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);

  int64x2_t h00_lo = vdupq_n_s64(0);
  int64x2_t h00_hi = vdupq_n_s64(0);
  int64x2_t h11_lo = vdupq_n_s64(0);
  int64x2_t h11_hi = vdupq_n_s64(0);
  int64x2_t h01_lo = vdupq_n_s64(0);
  int64x2_t h01_hi = vdupq_n_s64(0);
  int64x2_t c0_lo = vdupq_n_s64(0);
  int64x2_t c0_hi = vdupq_n_s64(0);
  int64x2_t c1_lo = vdupq_n_s64(0);
  int64x2_t c1_hi = vdupq_n_s64(0);

  do {
    const uint16_t *src_ptr = src;
    const uint16_t *dat_ptr = dat;
    int32_t *flt0_ptr = flt0;
    int32_t *flt1_ptr = flt1;
    int w = width;

    do {
      uint16x8_t s = vld1q_u16(src_ptr);
      uint16x8_t d = vld1q_u16(dat_ptr);
      int32x4_t f0_lo = vld1q_s32(flt0_ptr);
      int32x4_t f0_hi = vld1q_s32(flt0_ptr + 4);
      int32x4_t f1_lo = vld1q_s32(flt1_ptr);
      int32x4_t f1_hi = vld1q_s32(flt1_ptr + 4);

      int32x4_t u_lo =
          vreinterpretq_s32_u32(vshll_n_u16(vget_low_u16(d), SGRPROJ_RST_BITS));
      int32x4_t u_hi = vreinterpretq_s32_u32(
          vshll_n_u16(vget_high_u16(d), SGRPROJ_RST_BITS));
      int32x4_t s_lo =
          vreinterpretq_s32_u32(vshll_n_u16(vget_low_u16(s), SGRPROJ_RST_BITS));
      int32x4_t s_hi = vreinterpretq_s32_u32(
          vshll_n_u16(vget_high_u16(s), SGRPROJ_RST_BITS));
      s_lo = vsubq_s32(s_lo, u_lo);
      s_hi = vsubq_s32(s_hi, u_hi);

      f0_lo = vsubq_s32(f0_lo, u_lo);
      f0_hi = vsubq_s32(f0_hi, u_hi);
      f1_lo = vsubq_s32(f1_lo, u_lo);
      f1_hi = vsubq_s32(f1_hi, u_hi);

      h00_lo = vmlal_s32(h00_lo, vget_low_s32(f0_lo), vget_low_s32(f0_lo));
      h00_lo = vmlal_s32(h00_lo, vget_high_s32(f0_lo), vget_high_s32(f0_lo));
      h00_hi = vmlal_s32(h00_hi, vget_low_s32(f0_hi), vget_low_s32(f0_hi));
      h00_hi = vmlal_s32(h00_hi, vget_high_s32(f0_hi), vget_high_s32(f0_hi));

      h11_lo = vmlal_s32(h11_lo, vget_low_s32(f1_lo), vget_low_s32(f1_lo));
      h11_lo = vmlal_s32(h11_lo, vget_high_s32(f1_lo), vget_high_s32(f1_lo));
      h11_hi = vmlal_s32(h11_hi, vget_low_s32(f1_hi), vget_low_s32(f1_hi));
      h11_hi = vmlal_s32(h11_hi, vget_high_s32(f1_hi), vget_high_s32(f1_hi));

      h01_lo = vmlal_s32(h01_lo, vget_low_s32(f0_lo), vget_low_s32(f1_lo));
      h01_lo = vmlal_s32(h01_lo, vget_high_s32(f0_lo), vget_high_s32(f1_lo));
      h01_hi = vmlal_s32(h01_hi, vget_low_s32(f0_hi), vget_low_s32(f1_hi));
      h01_hi = vmlal_s32(h01_hi, vget_high_s32(f0_hi), vget_high_s32(f1_hi));

      c0_lo = vmlal_s32(c0_lo, vget_low_s32(f0_lo), vget_low_s32(s_lo));
      c0_lo = vmlal_s32(c0_lo, vget_high_s32(f0_lo), vget_high_s32(s_lo));
      c0_hi = vmlal_s32(c0_hi, vget_low_s32(f0_hi), vget_low_s32(s_hi));
      c0_hi = vmlal_s32(c0_hi, vget_high_s32(f0_hi), vget_high_s32(s_hi));

      c1_lo = vmlal_s32(c1_lo, vget_low_s32(f1_lo), vget_low_s32(s_lo));
      c1_lo = vmlal_s32(c1_lo, vget_high_s32(f1_lo), vget_high_s32(s_lo));
      c1_hi = vmlal_s32(c1_hi, vget_low_s32(f1_hi), vget_low_s32(s_hi));
      c1_hi = vmlal_s32(c1_hi, vget_high_s32(f1_hi), vget_high_s32(s_hi));

      src_ptr += 8;
      dat_ptr += 8;
      flt0_ptr += 8;
      flt1_ptr += 8;
      w -= 8;
    } while (w != 0);

    src += src_stride;
    dat += dat_stride;
    flt0 += flt0_stride;
    flt1 += flt1_stride;
  } while (--height != 0);

  H[0][0] = horizontal_add_s64x2(vaddq_s64(h00_lo, h00_hi)) / size;
  H[0][1] = horizontal_add_s64x2(vaddq_s64(h01_lo, h01_hi)) / size;
  H[1][1] = horizontal_add_s64x2(vaddq_s64(h11_lo, h11_hi)) / size;
  H[1][0] = H[0][1];
  C[0] = horizontal_add_s64x2(vaddq_s64(c0_lo, c0_hi)) / size;
  C[1] = horizontal_add_s64x2(vaddq_s64(c1_lo, c1_hi)) / size;
}

static INLINE void highbd_calc_proj_params_r0_neon(
    const uint8_t *src8, int width, int height, int src_stride,
    const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
    int64_t H[2][2], int64_t C[2]) {
  assert(width % 8 == 0);
  const int size = width * height;
  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);

  int64x2_t h00_lo = vdupq_n_s64(0);
  int64x2_t h00_hi = vdupq_n_s64(0);
  int64x2_t c0_lo = vdupq_n_s64(0);
  int64x2_t c0_hi = vdupq_n_s64(0);

  do {
    const uint16_t *src_ptr = src;
    const uint16_t *dat_ptr = dat;
    int32_t *flt0_ptr = flt0;
    int w = width;

    do {
      uint16x8_t s = vld1q_u16(src_ptr);
      uint16x8_t d = vld1q_u16(dat_ptr);
      int32x4_t f0_lo = vld1q_s32(flt0_ptr);
      int32x4_t f0_hi = vld1q_s32(flt0_ptr + 4);

      int32x4_t u_lo =
          vreinterpretq_s32_u32(vshll_n_u16(vget_low_u16(d), SGRPROJ_RST_BITS));
      int32x4_t u_hi = vreinterpretq_s32_u32(
          vshll_n_u16(vget_high_u16(d), SGRPROJ_RST_BITS));
      int32x4_t s_lo =
          vreinterpretq_s32_u32(vshll_n_u16(vget_low_u16(s), SGRPROJ_RST_BITS));
      int32x4_t s_hi = vreinterpretq_s32_u32(
          vshll_n_u16(vget_high_u16(s), SGRPROJ_RST_BITS));
      s_lo = vsubq_s32(s_lo, u_lo);
      s_hi = vsubq_s32(s_hi, u_hi);

      f0_lo = vsubq_s32(f0_lo, u_lo);
      f0_hi = vsubq_s32(f0_hi, u_hi);

      h00_lo = vmlal_s32(h00_lo, vget_low_s32(f0_lo), vget_low_s32(f0_lo));
      h00_lo = vmlal_s32(h00_lo, vget_high_s32(f0_lo), vget_high_s32(f0_lo));
      h00_hi = vmlal_s32(h00_hi, vget_low_s32(f0_hi), vget_low_s32(f0_hi));
      h00_hi = vmlal_s32(h00_hi, vget_high_s32(f0_hi), vget_high_s32(f0_hi));

      c0_lo = vmlal_s32(c0_lo, vget_low_s32(f0_lo), vget_low_s32(s_lo));
      c0_lo = vmlal_s32(c0_lo, vget_high_s32(f0_lo), vget_high_s32(s_lo));
      c0_hi = vmlal_s32(c0_hi, vget_low_s32(f0_hi), vget_low_s32(s_hi));
      c0_hi = vmlal_s32(c0_hi, vget_high_s32(f0_hi), vget_high_s32(s_hi));

      src_ptr += 8;
      dat_ptr += 8;
      flt0_ptr += 8;
      w -= 8;
    } while (w != 0);

    src += src_stride;
    dat += dat_stride;
    flt0 += flt0_stride;
  } while (--height != 0);

  H[0][0] = horizontal_add_s64x2(vaddq_s64(h00_lo, h00_hi)) / size;
  C[0] = horizontal_add_s64x2(vaddq_s64(c0_lo, c0_hi)) / size;
}

static INLINE void highbd_calc_proj_params_r1_neon(
    const uint8_t *src8, int width, int height, int src_stride,
    const uint8_t *dat8, int dat_stride, int32_t *flt1, int flt1_stride,
    int64_t H[2][2], int64_t C[2]) {
  assert(width % 8 == 0);
  const int size = width * height;
  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);

  int64x2_t h11_lo = vdupq_n_s64(0);
  int64x2_t h11_hi = vdupq_n_s64(0);
  int64x2_t c1_lo = vdupq_n_s64(0);
  int64x2_t c1_hi = vdupq_n_s64(0);

  do {
    const uint16_t *src_ptr = src;
    const uint16_t *dat_ptr = dat;
    int32_t *flt1_ptr = flt1;
    int w = width;

    do {
      uint16x8_t s = vld1q_u16(src_ptr);
      uint16x8_t d = vld1q_u16(dat_ptr);
      int32x4_t f1_lo = vld1q_s32(flt1_ptr);
      int32x4_t f1_hi = vld1q_s32(flt1_ptr + 4);

      int32x4_t u_lo =
          vreinterpretq_s32_u32(vshll_n_u16(vget_low_u16(d), SGRPROJ_RST_BITS));
      int32x4_t u_hi = vreinterpretq_s32_u32(
          vshll_n_u16(vget_high_u16(d), SGRPROJ_RST_BITS));
      int32x4_t s_lo =
          vreinterpretq_s32_u32(vshll_n_u16(vget_low_u16(s), SGRPROJ_RST_BITS));
      int32x4_t s_hi = vreinterpretq_s32_u32(
          vshll_n_u16(vget_high_u16(s), SGRPROJ_RST_BITS));
      s_lo = vsubq_s32(s_lo, u_lo);
      s_hi = vsubq_s32(s_hi, u_hi);

      f1_lo = vsubq_s32(f1_lo, u_lo);
      f1_hi = vsubq_s32(f1_hi, u_hi);

      h11_lo = vmlal_s32(h11_lo, vget_low_s32(f1_lo), vget_low_s32(f1_lo));
      h11_lo = vmlal_s32(h11_lo, vget_high_s32(f1_lo), vget_high_s32(f1_lo));
      h11_hi = vmlal_s32(h11_hi, vget_low_s32(f1_hi), vget_low_s32(f1_hi));
      h11_hi = vmlal_s32(h11_hi, vget_high_s32(f1_hi), vget_high_s32(f1_hi));

      c1_lo = vmlal_s32(c1_lo, vget_low_s32(f1_lo), vget_low_s32(s_lo));
      c1_lo = vmlal_s32(c1_lo, vget_high_s32(f1_lo), vget_high_s32(s_lo));
      c1_hi = vmlal_s32(c1_hi, vget_low_s32(f1_hi), vget_low_s32(s_hi));
      c1_hi = vmlal_s32(c1_hi, vget_high_s32(f1_hi), vget_high_s32(s_hi));

      src_ptr += 8;
      dat_ptr += 8;
      flt1_ptr += 8;
      w -= 8;
    } while (w != 0);

    src += src_stride;
    dat += dat_stride;
    flt1 += flt1_stride;
  } while (--height != 0);

  H[1][1] = horizontal_add_s64x2(vaddq_s64(h11_lo, h11_hi)) / size;
  C[1] = horizontal_add_s64x2(vaddq_s64(c1_lo, c1_hi)) / size;
}

// The function calls 3 subfunctions for the following cases :
// 1) When params->r[0] > 0 and params->r[1] > 0. In this case all elements
//    of C and H need to be computed.
// 2) When only params->r[0] > 0. In this case only H[0][0] and C[0] are
//    non-zero and need to be computed.
// 3) When only params->r[1] > 0. In this case only H[1][1] and C[1] are
//    non-zero and need to be computed.
void av1_calc_proj_params_high_bd_neon(const uint8_t *src8, int width,
                                       int height, int src_stride,
                                       const uint8_t *dat8, int dat_stride,
                                       int32_t *flt0, int flt0_stride,
                                       int32_t *flt1, int flt1_stride,
                                       int64_t H[2][2], int64_t C[2],
                                       const sgr_params_type *params) {
  if ((params->r[0] > 0) && (params->r[1] > 0)) {
    highbd_calc_proj_params_r0_r1_neon(src8, width, height, src_stride, dat8,
                                       dat_stride, flt0, flt0_stride, flt1,
                                       flt1_stride, H, C);
  } else if (params->r[0] > 0) {
    highbd_calc_proj_params_r0_neon(src8, width, height, src_stride, dat8,
                                    dat_stride, flt0, flt0_stride, H, C);
  } else if (params->r[1] > 0) {
    highbd_calc_proj_params_r1_neon(src8, width, height, src_stride, dat8,
                                    dat_stride, flt1, flt1_stride, H, C);
  }
}

static INLINE int16x8_t tbl2q(int16x8_t a, int16x8_t b, uint8x16_t idx) {
#if AOM_ARCH_AARCH64
  uint8x16x2_t table = { { vreinterpretq_u8_s16(a), vreinterpretq_u8_s16(b) } };
  return vreinterpretq_s16_u8(vqtbl2q_u8(table, idx));
#else
  uint8x8x4_t table = { { vreinterpret_u8_s16(vget_low_s16(a)),
                          vreinterpret_u8_s16(vget_high_s16(a)),
                          vreinterpret_u8_s16(vget_low_s16(b)),
                          vreinterpret_u8_s16(vget_high_s16(b)) } };
  return vreinterpretq_s16_u8(vcombine_u8(vtbl4_u8(table, vget_low_u8(idx)),
                                          vtbl4_u8(table, vget_high_u8(idx))));
#endif
}

static INLINE int16x8_t tbl3q(int16x8_t a, int16x8_t b, int16x8_t c,
                              uint8x16_t idx) {
#if AOM_ARCH_AARCH64
  uint8x16x3_t table = { { vreinterpretq_u8_s16(a), vreinterpretq_u8_s16(b),
                           vreinterpretq_u8_s16(c) } };
  return vreinterpretq_s16_u8(vqtbl3q_u8(table, idx));
#else
  // This is a specific implementation working only for compute stats with
  // wiener_win == 5.
  uint8x8x3_t table_lo = { { vreinterpret_u8_s16(vget_low_s16(a)),
                             vreinterpret_u8_s16(vget_high_s16(a)),
                             vreinterpret_u8_s16(vget_low_s16(b)) } };
  uint8x8x3_t table_hi = { { vreinterpret_u8_s16(vget_low_s16(b)),
                             vreinterpret_u8_s16(vget_high_s16(b)),
                             vreinterpret_u8_s16(vget_low_s16(c)) } };
  return vreinterpretq_s16_u8(vcombine_u8(
      vtbl3_u8(table_lo, vget_low_u8(idx)),
      vtbl3_u8(table_hi, vsub_u8(vget_high_u8(idx), vdup_n_u8(16)))));
#endif
}

static INLINE int64_t div_shift_s64(int64_t x, int power) {
  return (x < 0 ? x + (1ll << power) - 1 : x) >> power;
}

// The M matrix is accumulated in a bitdepth-dependent number of steps to
// speed up the computation. This function computes the final M from the
// accumulated (src_s64) and the residual parts (src_s32). It also transposes
// the result as the output needs to be column-major.
static INLINE void acc_transpose_M(int64_t *dst, const int64_t *src_s64,
                                   const int32_t *src_s32, const int wiener_win,
                                   int shift) {
  for (int i = 0; i < wiener_win; ++i) {
    for (int j = 0; j < wiener_win; ++j) {
      int tr_idx = j * wiener_win + i;
      *dst++ = div_shift_s64(src_s64[tr_idx] + src_s32[tr_idx], shift);
    }
  }
}

// The resulting H is a column-major matrix accumulated from the transposed
// (column-major) samples of the filter kernel (5x5 or 7x7) viewed as a single
// vector. For the 7x7 filter case: H(49x49) = [49 x 1] x [1 x 49]. This
// function transforms back to the originally expected format (double
// transpose). The H matrix is accumulated in a bitdepth-dependent number of
// steps to speed up the computation. This function computes the final H from
// the accumulated (src_s64) and the residual parts (src_s32). The computed H is
// only an upper triangle matrix, this function also fills the lower triangle of
// the resulting matrix.
static INLINE void update_H(int64_t *dst, const int64_t *src_s64,
                            const int32_t *src_s32, const int wiener_win,
                            int stride, int shift) {
  // For a simplified theoretical 3x3 case where `wiener_win` is 3 and
  // `wiener_win2` is 9, the M matrix is 3x3:
  // 0, 3, 6
  // 1, 4, 7
  // 2, 5, 8
  //
  // This is viewed as a vector to compute H (9x9) by vector outer product:
  // 0, 3, 6, 1, 4, 7, 2, 5, 8
  //
  // Double transpose and upper triangle remapping for 3x3 -> 9x9 case:
  // 0,    3,    6,    1,    4,    7,    2,    5,    8,
  // 3,   30,   33,   12,   31,   34,   21,   32,   35,
  // 6,   33,   60,   15,   42,   61,   24,   51,   62,
  // 1,   12,   15,   10,   13,   16,   11,   14,   17,
  // 4,   31,   42,   13,   40,   43,   22,   41,   44,
  // 7,   34,   61,   16,   43,   70,   25,   52,   71,
  // 2,   21,   24,   11,   22,   25,   20,   23,   26,
  // 5,   32,   51,   14,   41,   52,   23,   50,   53,
  // 8,   35,   62,   17,   44,   71,   26,   53,   80,
  const int wiener_win2 = wiener_win * wiener_win;

  // Loop through the indices according to the remapping above, along the
  // columns:
  // 0, wiener_win, 2 * wiener_win, ..., 1, 1 + 2 * wiener_win, ...,
  // wiener_win - 1, wiener_win - 1 + wiener_win, ...
  // For the 3x3 case `j` will be: 0, 3, 6, 1, 4, 7, 2, 5, 8.
  for (int i = 0; i < wiener_win; ++i) {
    for (int j = i; j < wiener_win2; j += wiener_win) {
      // These two inner loops are the same as the two outer loops, but running
      // along rows instead of columns. For the 3x3 case `l` will be:
      // 0, 3, 6, 1, 4, 7, 2, 5, 8.
      for (int k = 0; k < wiener_win; ++k) {
        for (int l = k; l < wiener_win2; l += wiener_win) {
          // The nominal double transpose indexing would be:
          // int idx = stride * j + l;
          // However we need the upper-right triangle, it is easy with some
          // min/max operations.
          int tr_idx = stride * AOMMIN(j, l) + AOMMAX(j, l);

          // Resulting matrix is filled by combining the 64-bit and the residual
          // 32-bit matrices together with scaling.
          *dst++ = div_shift_s64(src_s64[tr_idx] + src_s32[tr_idx], shift);
        }
      }
    }
  }
}

// Load 7x7 matrix into 7 128-bit vectors from consecutive rows, the last load
// address is offset to prevent out-of-bounds access.
static INLINE void load_and_pack_s16_8x7(int16x8_t dst[7], const int16_t *src,
                                         ptrdiff_t stride) {
  dst[0] = vld1q_s16(src);
  src += stride;
  dst[1] = vld1q_s16(src);
  src += stride;
  dst[2] = vld1q_s16(src);
  src += stride;
  dst[3] = vld1q_s16(src);
  src += stride;
  dst[4] = vld1q_s16(src);
  src += stride;
  dst[5] = vld1q_s16(src);
  src += stride;
  dst[6] = vld1q_s16(src - 1);
}

static INLINE void highbd_compute_stats_win7_neon(
    const uint16_t *dgd, const uint16_t *src, int avg, int width, int height,
    int dgd_stride, int src_stride, int64_t *M, int64_t *H,
    aom_bit_depth_t bit_depth) {
  // Matrix names are capitalized to help readability.
  DECLARE_ALIGNED(64, int16_t, DGD_AVG0[WIENER_WIN2_ALIGN3]);
  DECLARE_ALIGNED(64, int16_t, DGD_AVG1[WIENER_WIN2_ALIGN3]);
  DECLARE_ALIGNED(64, int32_t, M_s32[WIENER_WIN2_ALIGN3]);
  DECLARE_ALIGNED(64, int64_t, M_s64[WIENER_WIN2_ALIGN3]);
  DECLARE_ALIGNED(64, int32_t, H_s32[WIENER_WIN2 * WIENER_WIN2_ALIGN2]);
  DECLARE_ALIGNED(64, int64_t, H_s64[WIENER_WIN2 * WIENER_WIN2_ALIGN2]);

  memset(M_s32, 0, sizeof(M_s32));
  memset(M_s64, 0, sizeof(M_s64));
  memset(H_s32, 0, sizeof(H_s32));
  memset(H_s64, 0, sizeof(H_s64));

  // Look-up tables to create 8x6 matrix with consecutive elements from two 7x7
  // matrices.
  // clang-format off
  DECLARE_ALIGNED(16, static const uint8_t, shuffle_stats7_highbd[192]) = {
    0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 16, 17,
    2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 16, 17, 18, 19,
    4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 16, 17, 18, 19, 20, 21,
    6,  7,  8,  9,  10, 11, 12, 13, 16, 17, 18, 19, 20, 21, 22, 23,
    8,  9,  10, 11, 12, 13, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25,
    10, 11, 12, 13, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,
    2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 18, 19,
    4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 18, 19, 20, 21,
    6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 18, 19, 20, 21, 22, 23,
    8,  9,  10, 11, 12, 13, 14, 15, 18, 19, 20, 21, 22, 23, 24, 25,
    10, 11, 12, 13, 14, 15, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27,
    12, 13, 14, 15, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
  };
  // clang-format on

  const uint8x16_t lut0 = vld1q_u8(shuffle_stats7_highbd + 0);
  const uint8x16_t lut1 = vld1q_u8(shuffle_stats7_highbd + 16);
  const uint8x16_t lut2 = vld1q_u8(shuffle_stats7_highbd + 32);
  const uint8x16_t lut3 = vld1q_u8(shuffle_stats7_highbd + 48);
  const uint8x16_t lut4 = vld1q_u8(shuffle_stats7_highbd + 64);
  const uint8x16_t lut5 = vld1q_u8(shuffle_stats7_highbd + 80);
  const uint8x16_t lut6 = vld1q_u8(shuffle_stats7_highbd + 96);
  const uint8x16_t lut7 = vld1q_u8(shuffle_stats7_highbd + 112);
  const uint8x16_t lut8 = vld1q_u8(shuffle_stats7_highbd + 128);
  const uint8x16_t lut9 = vld1q_u8(shuffle_stats7_highbd + 144);
  const uint8x16_t lut10 = vld1q_u8(shuffle_stats7_highbd + 160);
  const uint8x16_t lut11 = vld1q_u8(shuffle_stats7_highbd + 176);

  // We can accumulate up to 65536/4096/256 8/10/12-bit multiplication results
  // in 32-bit. We are processing 2 pixels at a time, so the accumulator max can
  // be as high as 32768/2048/128 for the compute stats.
  const int acc_cnt_max = (1 << (32 - 2 * bit_depth)) >> 1;
  int acc_cnt = acc_cnt_max;
  const int src_next = src_stride - width;
  const int dgd_next = dgd_stride - width;
  const int16x8_t avg_s16 = vdupq_n_s16(avg);

  do {
    int j = width;
    while (j >= 2) {
      // Load two adjacent, overlapping 7x7 matrices: a 8x7 matrix with the
      // middle 6x7 elements being shared.
      int16x8_t dgd_rows[7];
      load_and_pack_s16_8x7(dgd_rows, (const int16_t *)dgd, dgd_stride);

      const int16_t *dgd_ptr = (const int16_t *)dgd + dgd_stride * 6;
      dgd += 2;

      dgd_rows[0] = vsubq_s16(dgd_rows[0], avg_s16);
      dgd_rows[1] = vsubq_s16(dgd_rows[1], avg_s16);
      dgd_rows[2] = vsubq_s16(dgd_rows[2], avg_s16);
      dgd_rows[3] = vsubq_s16(dgd_rows[3], avg_s16);
      dgd_rows[4] = vsubq_s16(dgd_rows[4], avg_s16);
      dgd_rows[5] = vsubq_s16(dgd_rows[5], avg_s16);
      dgd_rows[6] = vsubq_s16(dgd_rows[6], avg_s16);

      // Re-arrange the combined 8x7 matrix to have the 2 whole 7x7 matrices (1
      // for each of the 2 pixels) separated into distinct int16x8_t[6] arrays.
      // These arrays contain 48 elements of the 49 (7x7). Compute `dgd - avg`
      // for both buffers. Each DGD_AVG buffer contains 49 consecutive elements.
      int16x8_t dgd_avg0[6];
      int16x8_t dgd_avg1[6];

      dgd_avg0[0] = tbl2q(dgd_rows[0], dgd_rows[1], lut0);
      dgd_avg1[0] = tbl2q(dgd_rows[0], dgd_rows[1], lut6);
      dgd_avg0[1] = tbl2q(dgd_rows[1], dgd_rows[2], lut1);
      dgd_avg1[1] = tbl2q(dgd_rows[1], dgd_rows[2], lut7);
      dgd_avg0[2] = tbl2q(dgd_rows[2], dgd_rows[3], lut2);
      dgd_avg1[2] = tbl2q(dgd_rows[2], dgd_rows[3], lut8);
      dgd_avg0[3] = tbl2q(dgd_rows[3], dgd_rows[4], lut3);
      dgd_avg1[3] = tbl2q(dgd_rows[3], dgd_rows[4], lut9);
      dgd_avg0[4] = tbl2q(dgd_rows[4], dgd_rows[5], lut4);
      dgd_avg1[4] = tbl2q(dgd_rows[4], dgd_rows[5], lut10);
      dgd_avg0[5] = tbl2q(dgd_rows[5], dgd_rows[6], lut5);
      dgd_avg1[5] = tbl2q(dgd_rows[5], dgd_rows[6], lut11);

      vst1q_s16(DGD_AVG0, dgd_avg0[0]);
      vst1q_s16(DGD_AVG1, dgd_avg1[0]);
      vst1q_s16(DGD_AVG0 + 8, dgd_avg0[1]);
      vst1q_s16(DGD_AVG1 + 8, dgd_avg1[1]);
      vst1q_s16(DGD_AVG0 + 16, dgd_avg0[2]);
      vst1q_s16(DGD_AVG1 + 16, dgd_avg1[2]);
      vst1q_s16(DGD_AVG0 + 24, dgd_avg0[3]);
      vst1q_s16(DGD_AVG1 + 24, dgd_avg1[3]);
      vst1q_s16(DGD_AVG0 + 32, dgd_avg0[4]);
      vst1q_s16(DGD_AVG1 + 32, dgd_avg1[4]);
      vst1q_s16(DGD_AVG0 + 40, dgd_avg0[5]);
      vst1q_s16(DGD_AVG1 + 40, dgd_avg1[5]);

      // The remaining last (49th) elements of `dgd - avg`.
      DGD_AVG0[48] = dgd_ptr[6] - avg;
      DGD_AVG1[48] = dgd_ptr[7] - avg;

      // Accumulate into row-major variant of matrix M (cross-correlation) for 2
      // output pixels at a time. M is of size 7 * 7. It needs to be filled such
      // that multiplying one element from src with each element of a row of the
      // wiener window will fill one column of M. However this is not very
      // convenient in terms of memory access, as it means we do contiguous
      // loads of dgd but strided stores to M. As a result, we use an
      // intermediate matrix M_s32 which is instead filled such that one row of
      // the wiener window gives one row of M_s32. Once fully computed, M_s32 is
      // then transposed to return M.
      int src_avg0 = *src++ - avg;
      int src_avg1 = *src++ - avg;
      int16x4_t src_avg0_s16 = vdup_n_s16(src_avg0);
      int16x4_t src_avg1_s16 = vdup_n_s16(src_avg1);
      update_M_2pixels(M_s32 + 0, src_avg0_s16, src_avg1_s16, dgd_avg0[0],
                       dgd_avg1[0]);
      update_M_2pixels(M_s32 + 8, src_avg0_s16, src_avg1_s16, dgd_avg0[1],
                       dgd_avg1[1]);
      update_M_2pixels(M_s32 + 16, src_avg0_s16, src_avg1_s16, dgd_avg0[2],
                       dgd_avg1[2]);
      update_M_2pixels(M_s32 + 24, src_avg0_s16, src_avg1_s16, dgd_avg0[3],
                       dgd_avg1[3]);
      update_M_2pixels(M_s32 + 32, src_avg0_s16, src_avg1_s16, dgd_avg0[4],
                       dgd_avg1[4]);
      update_M_2pixels(M_s32 + 40, src_avg0_s16, src_avg1_s16, dgd_avg0[5],
                       dgd_avg1[5]);

      // Last (49th) element of M_s32 can be computed as scalar more efficiently
      // for 2 output pixels.
      M_s32[48] += DGD_AVG0[48] * src_avg0 + DGD_AVG1[48] * src_avg1;

      // Start accumulating into row-major version of matrix H
      // (auto-covariance), it expects the DGD_AVG[01] matrices to also be
      // row-major. H is of size 49 * 49. It is filled by multiplying every pair
      // of elements of the wiener window together (vector outer product). Since
      // it is a symmetric matrix, we only compute the upper-right triangle, and
      // then copy it down to the lower-left later. The upper triangle is
      // covered by 4x4 tiles. The original algorithm assumes the M matrix is
      // column-major and the resulting H matrix is also expected to be
      // column-major. It is not efficient to work with column-major matrices,
      // so we accumulate into a row-major matrix H_s32. At the end of the
      // algorithm a double transpose transformation will convert H_s32 back to
      // the expected output layout.
      update_H_7x7_2pixels(H_s32, DGD_AVG0, DGD_AVG1);

      // The last element of the triangle of H_s32 matrix can be computed as a
      // scalar more efficiently.
      H_s32[48 * WIENER_WIN2_ALIGN2 + 48] +=
          DGD_AVG0[48] * DGD_AVG0[48] + DGD_AVG1[48] * DGD_AVG1[48];

      // Accumulate into 64-bit after a bit depth dependent number of iterations
      // to prevent overflow.
      if (--acc_cnt == 0) {
        acc_cnt = acc_cnt_max;

        accumulate_and_clear(M_s64, M_s32, WIENER_WIN2_ALIGN2);

        // The widening accumulation is only needed for the upper triangle part
        // of the matrix.
        int64_t *lh = H_s64;
        int32_t *lh32 = H_s32;
        for (int k = 0; k < WIENER_WIN2; ++k) {
          // The widening accumulation is only run for the relevant parts
          // (upper-right triangle) in a row 4-element aligned.
          int k4 = k / 4 * 4;
          accumulate_and_clear(lh + k4, lh32 + k4, 48 - k4);

          // Last element of the row is computed separately.
          lh[48] += lh32[48];
          lh32[48] = 0;

          lh += WIENER_WIN2_ALIGN2;
          lh32 += WIENER_WIN2_ALIGN2;
        }
      }

      j -= 2;
    }

    // Computations for odd pixel in the row.
    if (width & 1) {
      // Load two adjacent, overlapping 7x7 matrices: a 8x7 matrix with the
      // middle 6x7 elements being shared.
      int16x8_t dgd_rows[7];
      load_and_pack_s16_8x7(dgd_rows, (const int16_t *)dgd, dgd_stride);

      const int16_t *dgd_ptr = (const int16_t *)dgd + dgd_stride * 6;
      ++dgd;

      // Re-arrange the combined 8x7 matrix to have a whole 7x7 matrix tightly
      // packed into a int16x8_t[6] array. This array contains 48 elements of
      // the 49 (7x7). Compute `dgd - avg` for the whole buffer. The DGD_AVG
      // buffer contains 49 consecutive elements.
      int16x8_t dgd_avg0[6];

      dgd_avg0[0] = vsubq_s16(tbl2q(dgd_rows[0], dgd_rows[1], lut0), avg_s16);
      dgd_avg0[1] = vsubq_s16(tbl2q(dgd_rows[1], dgd_rows[2], lut1), avg_s16);
      dgd_avg0[2] = vsubq_s16(tbl2q(dgd_rows[2], dgd_rows[3], lut2), avg_s16);
      dgd_avg0[3] = vsubq_s16(tbl2q(dgd_rows[3], dgd_rows[4], lut3), avg_s16);
      dgd_avg0[4] = vsubq_s16(tbl2q(dgd_rows[4], dgd_rows[5], lut4), avg_s16);
      dgd_avg0[5] = vsubq_s16(tbl2q(dgd_rows[5], dgd_rows[6], lut5), avg_s16);

      vst1q_s16(DGD_AVG0, dgd_avg0[0]);
      vst1q_s16(DGD_AVG0 + 8, dgd_avg0[1]);
      vst1q_s16(DGD_AVG0 + 16, dgd_avg0[2]);
      vst1q_s16(DGD_AVG0 + 24, dgd_avg0[3]);
      vst1q_s16(DGD_AVG0 + 32, dgd_avg0[4]);
      vst1q_s16(DGD_AVG0 + 40, dgd_avg0[5]);

      // The remaining last (49th) element of `dgd - avg`.
      DGD_AVG0[48] = dgd_ptr[6] - avg;

      // Accumulate into row-major order variant of matrix M (cross-correlation)
      // for 1 output pixel at a time. M is of size 7 * 7. It needs to be filled
      // such that multiplying one element from src with each element of a row
      // of the wiener window will fill one column of M. However this is not
      // very convenient in terms of memory access, as it means we do
      // contiguous loads of dgd but strided stores to M. As a result, we use an
      // intermediate matrix M_s32 which is instead filled such that one row of
      // the wiener window gives one row of M_s32. Once fully computed, M_s32 is
      // then transposed to return M.
      int src_avg0 = *src++ - avg;
      int16x4_t src_avg0_s16 = vdup_n_s16(src_avg0);
      update_M_1pixel(M_s32 + 0, src_avg0_s16, dgd_avg0[0]);
      update_M_1pixel(M_s32 + 8, src_avg0_s16, dgd_avg0[1]);
      update_M_1pixel(M_s32 + 16, src_avg0_s16, dgd_avg0[2]);
      update_M_1pixel(M_s32 + 24, src_avg0_s16, dgd_avg0[3]);
      update_M_1pixel(M_s32 + 32, src_avg0_s16, dgd_avg0[4]);
      update_M_1pixel(M_s32 + 40, src_avg0_s16, dgd_avg0[5]);

      // Last (49th) element of M_s32 can be computed as scalar more efficiently
      // for 1 output pixel.
      M_s32[48] += DGD_AVG0[48] * src_avg0;

      // Start accumulating into row-major order version of matrix H
      // (auto-covariance), it expects the DGD_AVG0 matrix to also be row-major.
      // H is of size 49 * 49. It is filled by multiplying every pair of
      // elements of the wiener window together (vector outer product). Since it
      // is a symmetric matrix, we only compute the upper-right triangle, and
      // then copy it down to the lower-left later. The upper triangle is
      // covered by 4x4 tiles. The original algorithm assumes the M matrix is
      // column-major and the resulting H matrix is also expected to be
      // column-major. It is not efficient to work column-major matrices, so we
      // accumulate into a row-major matrix H_s32. At the end of the algorithm a
      // double transpose transformation will convert H_s32 back to the expected
      // output layout.
      update_H_1pixel(H_s32, DGD_AVG0, WIENER_WIN2_ALIGN2, 48);

      // The last element of the triangle of H_s32 matrix can be computed as
      // scalar more efficiently.
      H_s32[48 * WIENER_WIN2_ALIGN2 + 48] += DGD_AVG0[48] * DGD_AVG0[48];
    }

    src += src_next;
    dgd += dgd_next;
  } while (--height != 0);

  int bit_depth_shift = bit_depth - AOM_BITS_8;

  acc_transpose_M(M, M_s64, M_s32, WIENER_WIN, bit_depth_shift);

  update_H(H, H_s64, H_s32, WIENER_WIN, WIENER_WIN2_ALIGN2, bit_depth_shift);
}

// Load 5x5 matrix into 5 128-bit vectors from consecutive rows, the last load
// address is offset to prevent out-of-bounds access.
static INLINE void load_and_pack_s16_6x5(int16x8_t dst[5], const int16_t *src,
                                         ptrdiff_t stride) {
  dst[0] = vld1q_s16(src);
  src += stride;
  dst[1] = vld1q_s16(src);
  src += stride;
  dst[2] = vld1q_s16(src);
  src += stride;
  dst[3] = vld1q_s16(src);
  src += stride;
  dst[4] = vld1q_s16(src - 3);
}

static void highbd_compute_stats_win5_neon(const uint16_t *dgd,
                                           const uint16_t *src, int avg,
                                           int width, int height,
                                           int dgd_stride, int src_stride,
                                           int64_t *M, int64_t *H,
                                           aom_bit_depth_t bit_depth) {
  // Matrix names are capitalized to help readability.
  DECLARE_ALIGNED(64, int16_t, DGD_AVG0[WIENER_WIN2_REDUCED_ALIGN3]);
  DECLARE_ALIGNED(64, int16_t, DGD_AVG1[WIENER_WIN2_REDUCED_ALIGN3]);
  DECLARE_ALIGNED(64, int32_t, M_s32[WIENER_WIN2_REDUCED_ALIGN3]);
  DECLARE_ALIGNED(64, int64_t, M_s64[WIENER_WIN2_REDUCED_ALIGN3]);
  DECLARE_ALIGNED(64, int32_t,
                  H_s32[WIENER_WIN2_REDUCED * WIENER_WIN2_REDUCED_ALIGN2]);
  DECLARE_ALIGNED(64, int64_t,
                  H_s64[WIENER_WIN2_REDUCED * WIENER_WIN2_REDUCED_ALIGN2]);

  memset(M_s32, 0, sizeof(M_s32));
  memset(M_s64, 0, sizeof(M_s64));
  memset(H_s32, 0, sizeof(H_s32));
  memset(H_s64, 0, sizeof(H_s64));

  // Look-up tables to create 8x3 matrix with consecutive elements from 5x5
  // matrix.
  DECLARE_ALIGNED(16, static const uint8_t, shuffle_stats5_highbd[96]) = {
    0, 1, 2,  3,  4,  5,  6,  7,  8,  9,  16, 17, 18, 19, 20, 21,
    6, 7, 8,  9,  16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 32, 33,
    2, 3, 4,  5,  6,  7,  8,  9,  22, 23, 24, 25, 26, 27, 28, 29,
    2, 3, 4,  5,  6,  7,  8,  9,  10, 11, 18, 19, 20, 21, 22, 23,
    8, 9, 10, 11, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 34, 35,
    4, 5, 6,  7,  8,  9,  10, 11, 24, 25, 26, 27, 28, 29, 30, 31,
  };

  const uint8x16_t lut0 = vld1q_u8(shuffle_stats5_highbd + 0);
  const uint8x16_t lut1 = vld1q_u8(shuffle_stats5_highbd + 16);
  const uint8x16_t lut2 = vld1q_u8(shuffle_stats5_highbd + 32);
  const uint8x16_t lut3 = vld1q_u8(shuffle_stats5_highbd + 48);
  const uint8x16_t lut4 = vld1q_u8(shuffle_stats5_highbd + 64);
  const uint8x16_t lut5 = vld1q_u8(shuffle_stats5_highbd + 80);

  // We can accumulate up to 65536/4096/256 8/10/12-bit multiplication results
  // in 32-bit. We are processing 2 pixels at a time, so the accumulator max can
  // be as high as 32768/2048/128 for the compute stats.
  const int acc_cnt_max = (1 << (32 - 2 * bit_depth)) >> 1;
  int acc_cnt = acc_cnt_max;
  const int src_next = src_stride - width;
  const int dgd_next = dgd_stride - width;
  const int16x8_t avg_s16 = vdupq_n_s16(avg);

  do {
    int j = width;
    while (j >= 2) {
      // Load two adjacent, overlapping 5x5 matrices: a 6x5 matrix with the
      // middle 4x5 elements being shared.
      int16x8_t dgd_rows[5];
      load_and_pack_s16_6x5(dgd_rows, (const int16_t *)dgd, dgd_stride);

      const int16_t *dgd_ptr = (const int16_t *)dgd + dgd_stride * 4;
      dgd += 2;

      dgd_rows[0] = vsubq_s16(dgd_rows[0], avg_s16);
      dgd_rows[1] = vsubq_s16(dgd_rows[1], avg_s16);
      dgd_rows[2] = vsubq_s16(dgd_rows[2], avg_s16);
      dgd_rows[3] = vsubq_s16(dgd_rows[3], avg_s16);
      dgd_rows[4] = vsubq_s16(dgd_rows[4], avg_s16);

      // Re-arrange the combined 6x5 matrix to have the 2 whole 5x5 matrices (1
      // for each of the 2 pixels) separated into distinct int16x8_t[3] arrays.
      // These arrays contain 24 elements of the 25 (5x5). Compute `dgd - avg`
      // for both buffers. Each DGD_AVG buffer contains 25 consecutive elements.
      int16x8_t dgd_avg0[3];
      int16x8_t dgd_avg1[3];

      dgd_avg0[0] = tbl2q(dgd_rows[0], dgd_rows[1], lut0);
      dgd_avg1[0] = tbl2q(dgd_rows[0], dgd_rows[1], lut3);
      dgd_avg0[1] = tbl3q(dgd_rows[1], dgd_rows[2], dgd_rows[3], lut1);
      dgd_avg1[1] = tbl3q(dgd_rows[1], dgd_rows[2], dgd_rows[3], lut4);
      dgd_avg0[2] = tbl2q(dgd_rows[3], dgd_rows[4], lut2);
      dgd_avg1[2] = tbl2q(dgd_rows[3], dgd_rows[4], lut5);

      vst1q_s16(DGD_AVG0, dgd_avg0[0]);
      vst1q_s16(DGD_AVG1, dgd_avg1[0]);
      vst1q_s16(DGD_AVG0 + 8, dgd_avg0[1]);
      vst1q_s16(DGD_AVG1 + 8, dgd_avg1[1]);
      vst1q_s16(DGD_AVG0 + 16, dgd_avg0[2]);
      vst1q_s16(DGD_AVG1 + 16, dgd_avg1[2]);

      // The remaining last (25th) elements of `dgd - avg`.
      DGD_AVG0[24] = dgd_ptr[4] - avg;
      DGD_AVG1[24] = dgd_ptr[5] - avg;

      // Accumulate into row-major variant of matrix M (cross-correlation) for 2
      // output pixels at a time. M is of size 5 * 5. It needs to be filled such
      // that multiplying one element from src with each element of a row of the
      // wiener window will fill one column of M. However this is not very
      // convenient in terms of memory access, as it means we do contiguous
      // loads of dgd but strided stores to M. As a result, we use an
      // intermediate matrix M_s32 which is instead filled such that one row of
      // the wiener window gives one row of M_s32. Once fully computed, M_s32 is
      // then transposed to return M.
      int src_avg0 = *src++ - avg;
      int src_avg1 = *src++ - avg;
      int16x4_t src_avg0_s16 = vdup_n_s16(src_avg0);
      int16x4_t src_avg1_s16 = vdup_n_s16(src_avg1);
      update_M_2pixels(M_s32 + 0, src_avg0_s16, src_avg1_s16, dgd_avg0[0],
                       dgd_avg1[0]);
      update_M_2pixels(M_s32 + 8, src_avg0_s16, src_avg1_s16, dgd_avg0[1],
                       dgd_avg1[1]);
      update_M_2pixels(M_s32 + 16, src_avg0_s16, src_avg1_s16, dgd_avg0[2],
                       dgd_avg1[2]);

      // Last (25th) element of M_s32 can be computed as scalar more efficiently
      // for 2 output pixels.
      M_s32[24] += DGD_AVG0[24] * src_avg0 + DGD_AVG1[24] * src_avg1;

      // Start accumulating into row-major version of matrix H
      // (auto-covariance), it expects the DGD_AVG[01] matrices to also be
      // row-major. H is of size 25 * 25. It is filled by multiplying every pair
      // of elements of the wiener window together (vector outer product). Since
      // it is a symmetric matrix, we only compute the upper-right triangle, and
      // then copy it down to the lower-left later. The upper triangle is
      // covered by 4x4 tiles. The original algorithm assumes the M matrix is
      // column-major and the resulting H matrix is also expected to be
      // column-major. It is not efficient to work with column-major matrices,
      // so we accumulate into a row-major matrix H_s32. At the end of the
      // algorithm a double transpose transformation will convert H_s32 back to
      // the expected output layout.
      update_H_5x5_2pixels(H_s32, DGD_AVG0, DGD_AVG1);

      // The last element of the triangle of H_s32 matrix can be computed as a
      // scalar more efficiently.
      H_s32[24 * WIENER_WIN2_REDUCED_ALIGN2 + 24] +=
          DGD_AVG0[24] * DGD_AVG0[24] + DGD_AVG1[24] * DGD_AVG1[24];

      // Accumulate into 64-bit after a bit depth dependent number of iterations
      // to prevent overflow.
      if (--acc_cnt == 0) {
        acc_cnt = acc_cnt_max;

        accumulate_and_clear(M_s64, M_s32, WIENER_WIN2_REDUCED_ALIGN2);

        // The widening accumulation is only needed for the upper triangle part
        // of the matrix.
        int64_t *lh = H_s64;
        int32_t *lh32 = H_s32;
        for (int k = 0; k < WIENER_WIN2_REDUCED; ++k) {
          // The widening accumulation is only run for the relevant parts
          // (upper-right triangle) in a row 4-element aligned.
          int k4 = k / 4 * 4;
          accumulate_and_clear(lh + k4, lh32 + k4, 24 - k4);

          // Last element of the row is computed separately.
          lh[24] += lh32[24];
          lh32[24] = 0;

          lh += WIENER_WIN2_REDUCED_ALIGN2;
          lh32 += WIENER_WIN2_REDUCED_ALIGN2;
        }
      }

      j -= 2;
    }

    // Computations for odd pixel in the row.
    if (width & 1) {
      // Load two adjacent, overlapping 5x5 matrices: a 6x5 matrix with the
      // middle 4x5 elements being shared.
      int16x8_t dgd_rows[5];
      load_and_pack_s16_6x5(dgd_rows, (const int16_t *)dgd, dgd_stride);

      const int16_t *dgd_ptr = (const int16_t *)dgd + dgd_stride * 4;
      ++dgd;

      // Re-arrange (and widen) the combined 6x5 matrix to have a whole 5x5
      // matrix tightly packed into a int16x8_t[3] array. This array contains
      // 24 elements of the 25 (5x5). Compute `dgd - avg` for the whole buffer.
      // The DGD_AVG buffer contains 25 consecutive elements.
      int16x8_t dgd_avg0[3];

      dgd_avg0[0] = vsubq_s16(tbl2q(dgd_rows[0], dgd_rows[1], lut0), avg_s16);
      dgd_avg0[1] = vsubq_s16(
          tbl3q(dgd_rows[1], dgd_rows[2], dgd_rows[3], lut1), avg_s16);
      dgd_avg0[2] = vsubq_s16(tbl2q(dgd_rows[3], dgd_rows[4], lut2), avg_s16);

      vst1q_s16(DGD_AVG0, dgd_avg0[0]);
      vst1q_s16(DGD_AVG0 + 8, dgd_avg0[1]);
      vst1q_s16(DGD_AVG0 + 16, dgd_avg0[2]);

      // The remaining last (25th) element of `dgd - avg`.
      DGD_AVG0[24] = dgd_ptr[4] - avg;
      DGD_AVG1[24] = dgd_ptr[5] - avg;

      // Accumulate into row-major order variant of matrix M (cross-correlation)
      // for 1 output pixel at a time. M is of size 5 * 5. It needs to be filled
      // such that multiplying one element from src with each element of a row
      // of the wiener window will fill one column of M. However this is not
      // very convenient in terms of memory access, as it means we do
      // contiguous loads of dgd but strided stores to M. As a result, we use an
      // intermediate matrix M_s32 which is instead filled such that one row of
      // the wiener window gives one row of M_s32. Once fully computed, M_s32 is
      // then transposed to return M.
      int src_avg0 = *src++ - avg;
      int16x4_t src_avg0_s16 = vdup_n_s16(src_avg0);
      update_M_1pixel(M_s32 + 0, src_avg0_s16, dgd_avg0[0]);
      update_M_1pixel(M_s32 + 8, src_avg0_s16, dgd_avg0[1]);
      update_M_1pixel(M_s32 + 16, src_avg0_s16, dgd_avg0[2]);

      // Last (25th) element of M_s32 can be computed as scalar more efficiently
      // for 1 output pixel.
      M_s32[24] += DGD_AVG0[24] * src_avg0;

      // Start accumulating into row-major order version of matrix H
      // (auto-covariance), it expects the DGD_AVG0 matrix to also be row-major.
      // H is of size 25 * 25. It is filled by multiplying every pair of
      // elements of the wiener window together (vector outer product). Since it
      // is a symmetric matrix, we only compute the upper-right triangle, and
      // then copy it down to the lower-left later. The upper triangle is
      // covered by 4x4 tiles. The original algorithm assumes the M matrix is
      // column-major and the resulting H matrix is also expected to be
      // column-major. It is not efficient to work with column-major matrices,
      // so we accumulate into a row-major matrix H_s32. At the end of the
      // algorithm a double transpose transformation will convert H_s32 back to
      // the expected output layout.
      update_H_1pixel(H_s32, DGD_AVG0, WIENER_WIN2_REDUCED_ALIGN2, 24);

      // The last element of the triangle of H_s32 matrix can be computed as a
      // scalar more efficiently.
      H_s32[24 * WIENER_WIN2_REDUCED_ALIGN2 + 24] +=
          DGD_AVG0[24] * DGD_AVG0[24];
    }

    src += src_next;
    dgd += dgd_next;
  } while (--height != 0);

  int bit_depth_shift = bit_depth - AOM_BITS_8;

  acc_transpose_M(M, M_s64, M_s32, WIENER_WIN_REDUCED, bit_depth_shift);

  update_H(H, H_s64, H_s32, WIENER_WIN_REDUCED, WIENER_WIN2_REDUCED_ALIGN2,
           bit_depth_shift);
}

static uint16_t highbd_find_average_neon(const uint16_t *src, int src_stride,
                                         int width, int height) {
  assert(width > 0);
  assert(height > 0);

  uint64x2_t sum_u64 = vdupq_n_u64(0);
  uint64_t sum = 0;

  int h = height;
  do {
    uint32x4_t sum_u32[2] = { vdupq_n_u32(0), vdupq_n_u32(0) };

    int w = width;
    const uint16_t *row = src;
    while (w >= 32) {
      uint16x8_t s0 = vld1q_u16(row + 0);
      uint16x8_t s1 = vld1q_u16(row + 8);
      uint16x8_t s2 = vld1q_u16(row + 16);
      uint16x8_t s3 = vld1q_u16(row + 24);

      s0 = vaddq_u16(s0, s1);
      s2 = vaddq_u16(s2, s3);
      sum_u32[0] = vpadalq_u16(sum_u32[0], s0);
      sum_u32[1] = vpadalq_u16(sum_u32[1], s2);

      row += 32;
      w -= 32;
    }

    if (w >= 16) {
      uint16x8_t s0 = vld1q_u16(row + 0);
      uint16x8_t s1 = vld1q_u16(row + 8);

      s0 = vaddq_u16(s0, s1);
      sum_u32[0] = vpadalq_u16(sum_u32[0], s0);

      row += 16;
      w -= 16;
    }

    if (w >= 8) {
      uint16x8_t s0 = vld1q_u16(row);
      sum_u32[1] = vpadalq_u16(sum_u32[1], s0);

      row += 8;
      w -= 8;
    }

    if (w >= 4) {
      uint16x8_t s0 = vcombine_u16(vld1_u16(row), vdup_n_u16(0));
      sum_u32[0] = vpadalq_u16(sum_u32[0], s0);

      row += 4;
      w -= 4;
    }

    while (w-- > 0) {
      sum += *row++;
    }

    sum_u64 = vpadalq_u32(sum_u64, vaddq_u32(sum_u32[0], sum_u32[1]));

    src += src_stride;
  } while (--h != 0);

  return (uint16_t)((horizontal_add_u64x2(sum_u64) + sum) / (height * width));
}

void av1_compute_stats_highbd_neon(int wiener_win, const uint8_t *dgd8,
                                   const uint8_t *src8, int16_t *dgd_avg,
                                   int16_t *src_avg, int h_start, int h_end,
                                   int v_start, int v_end, int dgd_stride,
                                   int src_stride, int64_t *M, int64_t *H,
                                   aom_bit_depth_t bit_depth) {
  (void)dgd_avg;
  (void)src_avg;
  assert(wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_REDUCED);

  const int wiener_halfwin = wiener_win >> 1;
  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8);
  const int height = v_end - v_start;
  const int width = h_end - h_start;

  const uint16_t *dgd_start = dgd + h_start + v_start * dgd_stride;
  const uint16_t *src_start = src + h_start + v_start * src_stride;

  // The wiener window will slide along the dgd frame, centered on each pixel.
  // For the top left pixel and all the pixels on the side of the frame this
  // means half of the window will be outside of the frame. As such the actual
  // buffer that we need to subtract the avg from will be 2 * wiener_halfwin
  // wider and 2 * wiener_halfwin higher than the original dgd buffer.
  const int vert_offset = v_start - wiener_halfwin;
  const int horiz_offset = h_start - wiener_halfwin;
  const uint16_t *dgd_win = dgd + horiz_offset + vert_offset * dgd_stride;

  uint16_t avg = highbd_find_average_neon(dgd_start, dgd_stride, width, height);

  if (wiener_win == WIENER_WIN) {
    highbd_compute_stats_win7_neon(dgd_win, src_start, avg, width, height,
                                   dgd_stride, src_stride, M, H, bit_depth);
  } else {
    highbd_compute_stats_win5_neon(dgd_win, src_start, avg, width, height,
                                   dgd_stride, src_stride, M, H, bit_depth);
  }
}

int64_t av1_highbd_pixel_proj_error_neon(
    const uint8_t *src8, int width, int height, int src_stride,
    const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
    int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params) {
  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *dat = CONVERT_TO_SHORTPTR(dat8);
  int64_t sse = 0;
  int64x2_t sse_s64 = vdupq_n_s64(0);

  if (params->r[0] > 0 && params->r[1] > 0) {
    int32x2_t xq_v = vld1_s32(xq);
    int32x2_t xq_sum_v = vshl_n_s32(vpadd_s32(xq_v, xq_v), 4);

    do {
      int j = 0;
      int32x4_t sse_s32 = vdupq_n_s32(0);

      do {
        const uint16x8_t d = vld1q_u16(&dat[j]);
        const uint16x8_t s = vld1q_u16(&src[j]);
        int32x4_t flt0_0 = vld1q_s32(&flt0[j]);
        int32x4_t flt0_1 = vld1q_s32(&flt0[j + 4]);
        int32x4_t flt1_0 = vld1q_s32(&flt1[j]);
        int32x4_t flt1_1 = vld1q_s32(&flt1[j + 4]);

        int32x4_t d_s32_lo = vreinterpretq_s32_u32(
            vmull_lane_u16(vget_low_u16(d), vreinterpret_u16_s32(xq_sum_v), 0));
        int32x4_t d_s32_hi = vreinterpretq_s32_u32(vmull_lane_u16(
            vget_high_u16(d), vreinterpret_u16_s32(xq_sum_v), 0));

        int32x4_t v0 = vsubq_s32(
            vdupq_n_s32(1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1)),
            d_s32_lo);
        int32x4_t v1 = vsubq_s32(
            vdupq_n_s32(1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1)),
            d_s32_hi);

        v0 = vmlaq_lane_s32(v0, flt0_0, xq_v, 0);
        v1 = vmlaq_lane_s32(v1, flt0_1, xq_v, 0);
        v0 = vmlaq_lane_s32(v0, flt1_0, xq_v, 1);
        v1 = vmlaq_lane_s32(v1, flt1_1, xq_v, 1);

        int16x4_t vr0 = vshrn_n_s32(v0, SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS);
        int16x4_t vr1 = vshrn_n_s32(v1, SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS);

        int16x8_t e = vaddq_s16(vcombine_s16(vr0, vr1),
                                vreinterpretq_s16_u16(vsubq_u16(d, s)));
        int16x4_t e_lo = vget_low_s16(e);
        int16x4_t e_hi = vget_high_s16(e);

        sse_s32 = vmlal_s16(sse_s32, e_lo, e_lo);
        sse_s32 = vmlal_s16(sse_s32, e_hi, e_hi);

        j += 8;
      } while (j <= width - 8);

      for (int k = j; k < width; ++k) {
        int32_t v = 1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1);
        v += xq[0] * (flt0[k]) + xq[1] * (flt1[k]);
        v -= (xq[1] + xq[0]) * (int32_t)(dat[k] << 4);
        int32_t e =
            (v >> (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS)) + dat[k] - src[k];
        sse += ((int64_t)e * e);
      }

      sse_s64 = vpadalq_s32(sse_s64, sse_s32);

      dat += dat_stride;
      src += src_stride;
      flt0 += flt0_stride;
      flt1 += flt1_stride;
    } while (--height != 0);
  } else if (params->r[0] > 0 || params->r[1] > 0) {
    int xq_active = (params->r[0] > 0) ? xq[0] : xq[1];
    int32_t *flt = (params->r[0] > 0) ? flt0 : flt1;
    int flt_stride = (params->r[0] > 0) ? flt0_stride : flt1_stride;
    int32x4_t xq_v = vdupq_n_s32(xq_active);

    do {
      int j = 0;
      int32x4_t sse_s32 = vdupq_n_s32(0);
      do {
        const uint16x8_t d0 = vld1q_u16(&dat[j]);
        const uint16x8_t s0 = vld1q_u16(&src[j]);
        int32x4_t flt0_0 = vld1q_s32(&flt[j]);
        int32x4_t flt0_1 = vld1q_s32(&flt[j + 4]);

        uint16x8_t d_u16 = vshlq_n_u16(d0, 4);
        int32x4_t sub0 = vreinterpretq_s32_u32(
            vsubw_u16(vreinterpretq_u32_s32(flt0_0), vget_low_u16(d_u16)));
        int32x4_t sub1 = vreinterpretq_s32_u32(
            vsubw_u16(vreinterpretq_u32_s32(flt0_1), vget_high_u16(d_u16)));

        int32x4_t v0 = vmlaq_s32(
            vdupq_n_s32(1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1)), sub0,
            xq_v);
        int32x4_t v1 = vmlaq_s32(
            vdupq_n_s32(1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1)), sub1,
            xq_v);

        int16x4_t vr0 = vshrn_n_s32(v0, SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS);
        int16x4_t vr1 = vshrn_n_s32(v1, SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS);

        int16x8_t e = vaddq_s16(vcombine_s16(vr0, vr1),
                                vreinterpretq_s16_u16(vsubq_u16(d0, s0)));
        int16x4_t e_lo = vget_low_s16(e);
        int16x4_t e_hi = vget_high_s16(e);

        sse_s32 = vmlal_s16(sse_s32, e_lo, e_lo);
        sse_s32 = vmlal_s16(sse_s32, e_hi, e_hi);

        j += 8;
      } while (j <= width - 8);

      for (int k = j; k < width; ++k) {
        int32_t v = 1 << (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS - 1);
        v += xq_active * (int32_t)((uint32_t)flt[j] - (uint16_t)(dat[k] << 4));
        const int32_t e =
            (v >> (SGRPROJ_RST_BITS + SGRPROJ_PRJ_BITS)) + dat[k] - src[k];
        sse += ((int64_t)e * e);
      }

      sse_s64 = vpadalq_s32(sse_s64, sse_s32);

      dat += dat_stride;
      flt += flt_stride;
      src += src_stride;
    } while (--height != 0);
  } else {
    do {
      int j = 0;

      do {
        const uint16x8_t d = vld1q_u16(&dat[j]);
        const uint16x8_t s = vld1q_u16(&src[j]);

        uint16x8_t diff = vabdq_u16(d, s);
        uint16x4_t diff_lo = vget_low_u16(diff);
        uint16x4_t diff_hi = vget_high_u16(diff);

        uint32x4_t sqr_lo = vmull_u16(diff_lo, diff_lo);
        uint32x4_t sqr_hi = vmull_u16(diff_hi, diff_hi);

        sse_s64 = vpadalq_s32(sse_s64, vreinterpretq_s32_u32(sqr_lo));
        sse_s64 = vpadalq_s32(sse_s64, vreinterpretq_s32_u32(sqr_hi));

        j += 8;
      } while (j <= width - 8);

      for (int k = j; k < width; ++k) {
        int32_t e = dat[k] - src[k];
        sse += e * e;
      }

      dat += dat_stride;
      src += src_stride;
    } while (--height != 0);
  }

  sse += horizontal_add_s64x2(sse_s64);
  return sse;
}