xref: /aosp_15_r20/external/libaom/aom_dsp/x86/intrapred_sse4.c (revision 77c1e3ccc04c968bd2bc212e87364f250e820521)

/*
 * Copyright (c) 2021, 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 <emmintrin.h>  // SSE2
#include <smmintrin.h>  /* SSE4.1 */

#include "config/av1_rtcd.h"
#include "aom_dsp/x86/intrapred_x86.h"
#include "aom_dsp/x86/intrapred_utils.h"
#include "aom_dsp/x86/lpf_common_sse2.h"

// Low bit depth functions
static DECLARE_ALIGNED(16, uint8_t, Mask[2][33][16]) = {
  { { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0,
      0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0,
      0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0,
      0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0,
      0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0 },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff },
    { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
      0xff, 0xff, 0xff, 0xff } },
  {
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0, 0,
        0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0, 0, 0,
        0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0, 0, 0,
        0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0,
        0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
        0, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
        0xff, 0, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
        0xff, 0xff, 0, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
        0xff, 0xff, 0xff, 0 },
      { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
        0xff, 0xff, 0xff, 0xff },
  },
};

/* clang-format on */
static AOM_FORCE_INLINE void dr_prediction_z1_HxW_internal_sse4_1(
    int H, int W, __m128i *dst, const uint8_t *above, int upsample_above,
    int dx) {
  const int frac_bits = 6 - upsample_above;
  const int max_base_x = ((W + H) - 1) << upsample_above;

  assert(dx > 0);
  // pre-filter above pixels
  // store in temp buffers:
  //   above[x] * 32 + 16
  //   above[x+1] - above[x]
  // final pixels will be calculated as:
  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
  __m128i a0, a1, a32, a16;
  __m128i diff, c3f;
  __m128i a_mbase_x;

  a16 = _mm_set1_epi16(16);
  a_mbase_x = _mm_set1_epi8((char)above[max_base_x]);
  c3f = _mm_set1_epi16(0x3f);

  int x = dx;
  for (int r = 0; r < W; r++) {
    __m128i b, res, res1, shift;
    __m128i a0_above, a1_above;

    int base = x >> frac_bits;
    int base_max_diff = (max_base_x - base) >> upsample_above;
    if (base_max_diff <= 0) {
      for (int i = r; i < W; ++i) {
        dst[i] = a_mbase_x;  // save 4 values
      }
      return;
    }
    if (base_max_diff > H) base_max_diff = H;
    a0_above = _mm_loadu_si128((__m128i *)(above + base));
    a1_above = _mm_loadu_si128((__m128i *)(above + base + 1));

    if (upsample_above) {
      a0_above = _mm_shuffle_epi8(a0_above, *(__m128i *)EvenOddMaskx[0]);
      a1_above = _mm_srli_si128(a0_above, 8);

      shift = _mm_srli_epi16(
          _mm_and_si128(_mm_slli_epi16(_mm_set1_epi16(x), upsample_above), c3f),
          1);
    } else {
      shift = _mm_srli_epi16(_mm_and_si128(_mm_set1_epi16(x), c3f), 1);
    }
    // lower half
    a0 = _mm_cvtepu8_epi16(a0_above);
    a1 = _mm_cvtepu8_epi16(a1_above);

    diff = _mm_sub_epi16(a1, a0);   // a[x+1] - a[x]
    a32 = _mm_slli_epi16(a0, 5);    // a[x] * 32
    a32 = _mm_add_epi16(a32, a16);  // a[x] * 32 + 16

    b = _mm_mullo_epi16(diff, shift);
    res = _mm_add_epi16(a32, b);
    res = _mm_srli_epi16(res, 5);

    // uppar half
    a0 = _mm_cvtepu8_epi16(_mm_srli_si128(a0_above, 8));
    a1 = _mm_cvtepu8_epi16(_mm_srli_si128(a1_above, 8));

    diff = _mm_sub_epi16(a1, a0);   // a[x+1] - a[x]
    a32 = _mm_slli_epi16(a0, 5);    // a[x] * 32
    a32 = _mm_add_epi16(a32, a16);  // a[x] * 32 + 16

    b = _mm_mullo_epi16(diff, shift);
    res1 = _mm_add_epi16(a32, b);
    res1 = _mm_srli_epi16(res1, 5);

    res = _mm_packus_epi16(res, res1);

    dst[r] =
        _mm_blendv_epi8(a_mbase_x, res, *(__m128i *)Mask[0][base_max_diff]);
    x += dx;
  }
}

static void dr_prediction_z1_4xN_sse4_1(int N, uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *above,
                                        int upsample_above, int dx) {
  __m128i dstvec[16];

  dr_prediction_z1_HxW_internal_sse4_1(4, N, dstvec, above, upsample_above, dx);
  for (int i = 0; i < N; i++) {
    *(int *)(dst + stride * i) = _mm_cvtsi128_si32(dstvec[i]);
  }
}

static void dr_prediction_z1_8xN_sse4_1(int N, uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *above,
                                        int upsample_above, int dx) {
  __m128i dstvec[32];

  dr_prediction_z1_HxW_internal_sse4_1(8, N, dstvec, above, upsample_above, dx);
  for (int i = 0; i < N; i++) {
    _mm_storel_epi64((__m128i *)(dst + stride * i), dstvec[i]);
  }
}

static void dr_prediction_z1_16xN_sse4_1(int N, uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *above,
                                         int upsample_above, int dx) {
  __m128i dstvec[64];

  dr_prediction_z1_HxW_internal_sse4_1(16, N, dstvec, above, upsample_above,
                                       dx);
  for (int i = 0; i < N; i++) {
    _mm_storeu_si128((__m128i *)(dst + stride * i), dstvec[i]);
  }
}

static AOM_FORCE_INLINE void dr_prediction_z1_32xN_internal_sse4_1(
    int N, __m128i *dstvec, __m128i *dstvec_h, const uint8_t *above,
    int upsample_above, int dx) {
  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
  (void)upsample_above;
  const int frac_bits = 6;
  const int max_base_x = ((32 + N) - 1);

  // pre-filter above pixels
  // store in temp buffers:
  //   above[x] * 32 + 16
  //   above[x+1] - above[x]
  // final pixels will be calculated as:
  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
  __m128i a0, a1, a32, a16;
  __m128i a_mbase_x, diff, c3f;

  a16 = _mm_set1_epi16(16);
  a_mbase_x = _mm_set1_epi8((char)above[max_base_x]);
  c3f = _mm_set1_epi16(0x3f);

  int x = dx;
  for (int r = 0; r < N; r++) {
    __m128i b, res, res1, res16[2];
    __m128i a0_above, a1_above;

    int base = x >> frac_bits;
    int base_max_diff = (max_base_x - base);
    if (base_max_diff <= 0) {
      for (int i = r; i < N; ++i) {
        dstvec[i] = a_mbase_x;  // save 32 values
        dstvec_h[i] = a_mbase_x;
      }
      return;
    }
    if (base_max_diff > 32) base_max_diff = 32;
    __m128i shift = _mm_srli_epi16(_mm_and_si128(_mm_set1_epi16(x), c3f), 1);

    for (int j = 0, jj = 0; j < 32; j += 16, jj++) {
      int mdiff = base_max_diff - j;
      if (mdiff <= 0) {
        res16[jj] = a_mbase_x;
      } else {
        a0_above = _mm_loadu_si128((__m128i *)(above + base + j));
        a1_above = _mm_loadu_si128((__m128i *)(above + base + j + 1));

        // lower half
        a0 = _mm_cvtepu8_epi16(a0_above);
        a1 = _mm_cvtepu8_epi16(a1_above);

        diff = _mm_sub_epi16(a1, a0);   // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0, 5);    // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);  // a[x] * 32 + 16
        b = _mm_mullo_epi16(diff, shift);

        res = _mm_add_epi16(a32, b);
        res = _mm_srli_epi16(res, 5);

        // uppar half
        a0 = _mm_cvtepu8_epi16(_mm_srli_si128(a0_above, 8));
        a1 = _mm_cvtepu8_epi16(_mm_srli_si128(a1_above, 8));

        diff = _mm_sub_epi16(a1, a0);   // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0, 5);    // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);  // a[x] * 32 + 16

        b = _mm_mullo_epi16(diff, shift);
        res1 = _mm_add_epi16(a32, b);
        res1 = _mm_srli_epi16(res1, 5);

        res16[jj] = _mm_packus_epi16(res, res1);  // 16 8bit values
      }
    }

    dstvec[r] =
        _mm_blendv_epi8(a_mbase_x, res16[0],
                        *(__m128i *)Mask[0][base_max_diff]);  // 16 8bit values

    dstvec_h[r] =
        _mm_blendv_epi8(a_mbase_x, res16[1],
                        *(__m128i *)Mask[1][base_max_diff]);  // 16 8bit values
    x += dx;
  }
}

static void dr_prediction_z1_32xN_sse4_1(int N, uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *above,
                                         int upsample_above, int dx) {
  __m128i dstvec[64], dstvec_h[64];
  dr_prediction_z1_32xN_internal_sse4_1(N, dstvec, dstvec_h, above,
                                        upsample_above, dx);
  for (int i = 0; i < N; i++) {
    _mm_storeu_si128((__m128i *)(dst + stride * i), dstvec[i]);
    _mm_storeu_si128((__m128i *)(dst + stride * i + 16), dstvec_h[i]);
  }
}

static void dr_prediction_z1_64xN_sse4_1(int N, uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *above,
                                         int upsample_above, int dx) {
  // here upsample_above is 0 by design of av1_use_intra_edge_upsample
  (void)upsample_above;
  const int frac_bits = 6;
  const int max_base_x = ((64 + N) - 1);

  // pre-filter above pixels
  // store in temp buffers:
  //   above[x] * 32 + 16
  //   above[x+1] - above[x]
  // final pixels will be calculated as:
  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
  __m128i a0, a1, a32, a16;
  __m128i a_mbase_x, diff, c3f;
  __m128i max_base, base_inc, mask;

  a16 = _mm_set1_epi16(16);
  a_mbase_x = _mm_set1_epi8((char)above[max_base_x]);
  max_base = _mm_set1_epi8(max_base_x);
  c3f = _mm_set1_epi16(0x3f);

  int x = dx;
  for (int r = 0; r < N; r++, dst += stride) {
    __m128i b, res, res1;
    int base = x >> frac_bits;
    if (base >= max_base_x) {
      for (int i = r; i < N; ++i) {
        _mm_storeu_si128((__m128i *)dst, a_mbase_x);  // save 32 values
        _mm_storeu_si128((__m128i *)(dst + 16), a_mbase_x);
        _mm_storeu_si128((__m128i *)(dst + 32), a_mbase_x);
        _mm_storeu_si128((__m128i *)(dst + 48), a_mbase_x);
        dst += stride;
      }
      return;
    }

    __m128i shift =
        _mm_srli_epi16(_mm_and_si128(_mm_set1_epi16(x), c3f), 1);  // 8 element

    __m128i a0_above, a1_above, res_val;
    for (int j = 0; j < 64; j += 16) {
      int mdif = max_base_x - (base + j);
      if (mdif <= 0) {
        _mm_storeu_si128((__m128i *)(dst + j), a_mbase_x);
      } else {
        a0_above =
            _mm_loadu_si128((__m128i *)(above + base + j));  // load 16 element
        a1_above = _mm_loadu_si128((__m128i *)(above + base + 1 + j));

        // lower half
        a0 = _mm_cvtepu8_epi16(a0_above);
        a1 = _mm_cvtepu8_epi16(a1_above);

        diff = _mm_sub_epi16(a1, a0);   // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0, 5);    // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);  // a[x] * 32 + 16
        b = _mm_mullo_epi16(diff, shift);

        res = _mm_add_epi16(a32, b);
        res = _mm_srli_epi16(res, 5);

        // uppar half
        a0 = _mm_cvtepu8_epi16(_mm_srli_si128(a0_above, 8));
        a1 = _mm_cvtepu8_epi16(_mm_srli_si128(a1_above, 8));

        diff = _mm_sub_epi16(a1, a0);   // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0, 5);    // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);  // a[x] * 32 + 16

        b = _mm_mullo_epi16(diff, shift);
        res1 = _mm_add_epi16(a32, b);
        res1 = _mm_srli_epi16(res1, 5);

        res = _mm_packus_epi16(res, res1);  // 16 8bit values

        base_inc =
            _mm_setr_epi8((int8_t)(base + j), (int8_t)(base + j + 1),
                          (int8_t)(base + j + 2), (int8_t)(base + j + 3),
                          (int8_t)(base + j + 4), (int8_t)(base + j + 5),
                          (int8_t)(base + j + 6), (int8_t)(base + j + 7),
                          (int8_t)(base + j + 8), (int8_t)(base + j + 9),
                          (int8_t)(base + j + 10), (int8_t)(base + j + 11),
                          (int8_t)(base + j + 12), (int8_t)(base + j + 13),
                          (int8_t)(base + j + 14), (int8_t)(base + j + 15));

        mask = _mm_cmpgt_epi8(_mm_subs_epu8(max_base, base_inc),
                              _mm_setzero_si128());
        res_val = _mm_blendv_epi8(a_mbase_x, res, mask);
        _mm_storeu_si128((__m128i *)(dst + j), res_val);
      }
    }
    x += dx;
  }
}

// Directional prediction, zone 1: 0 < angle < 90
void av1_dr_prediction_z1_sse4_1(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                                 const uint8_t *above, const uint8_t *left,
                                 int upsample_above, int dx, int dy) {
  (void)left;
  (void)dy;
  switch (bw) {
    case 4:
      dr_prediction_z1_4xN_sse4_1(bh, dst, stride, above, upsample_above, dx);
      break;
    case 8:
      dr_prediction_z1_8xN_sse4_1(bh, dst, stride, above, upsample_above, dx);
      break;
    case 16:
      dr_prediction_z1_16xN_sse4_1(bh, dst, stride, above, upsample_above, dx);
      break;
    case 32:
      dr_prediction_z1_32xN_sse4_1(bh, dst, stride, above, upsample_above, dx);
      break;
    case 64:
      dr_prediction_z1_64xN_sse4_1(bh, dst, stride, above, upsample_above, dx);
      break;
    default: assert(0 && "Invalid block size");
  }
  return;
}

static void dr_prediction_z2_Nx4_sse4_1(int N, uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *above,
                                        const uint8_t *left, int upsample_above,
                                        int upsample_left, int dx, int dy) {
  const int min_base_x = -(1 << upsample_above);
  const int min_base_y = -(1 << upsample_left);
  const int frac_bits_x = 6 - upsample_above;
  const int frac_bits_y = 6 - upsample_left;

  assert(dx > 0);
  // pre-filter above pixels
  // store in temp buffers:
  //   above[x] * 32 + 16
  //   above[x+1] - above[x]
  // final pixels will be calculated as:
  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
  __m128i a0_x, a1_x, a32, diff;

  const __m128i c3f = _mm_set1_epi16(0x3f);
  const __m128i min_y_base = _mm_set1_epi16(min_base_y);
  const __m128i c1234 = _mm_setr_epi16(0, 1, 2, 3, 4, 0, 0, 0);
  const __m128i dy_reg = _mm_set1_epi16(dy);
  const __m128i a16 = _mm_set1_epi16(16);

  for (int r = 0; r < N; r++) {
    __m128i b, res, shift, r6, ydx;
    __m128i resx, resy, resxy;
    __m128i a0_above, a1_above;
    int y = r + 1;
    int base_x = (-y * dx) >> frac_bits_x;
    int base_shift = 0;
    if (base_x < (min_base_x - 1)) {
      base_shift = (min_base_x - base_x - 1) >> upsample_above;
    }
    int base_min_diff =
        (min_base_x - base_x + upsample_above) >> upsample_above;
    if (base_min_diff > 4) {
      base_min_diff = 4;
    } else {
      if (base_min_diff < 0) base_min_diff = 0;
    }

    if (base_shift > 3) {
      a0_x = _mm_setzero_si128();
      a1_x = _mm_setzero_si128();
      shift = _mm_setzero_si128();
    } else {
      a0_above = _mm_loadu_si128((__m128i *)(above + base_x + base_shift));
      ydx = _mm_set1_epi16(y * dx);
      r6 = _mm_slli_epi16(c1234, 6);

      if (upsample_above) {
        a0_above =
            _mm_shuffle_epi8(a0_above, *(__m128i *)EvenOddMaskx[base_shift]);
        a1_above = _mm_srli_si128(a0_above, 8);

        shift = _mm_srli_epi16(
            _mm_and_si128(
                _mm_slli_epi16(_mm_sub_epi16(r6, ydx), upsample_above), c3f),
            1);
      } else {
        a0_above =
            _mm_shuffle_epi8(a0_above, *(__m128i *)LoadMaskx[base_shift]);
        a1_above = _mm_srli_si128(a0_above, 1);

        shift = _mm_srli_epi16(_mm_and_si128(_mm_sub_epi16(r6, ydx), c3f), 1);
      }
      a0_x = _mm_cvtepu8_epi16(a0_above);
      a1_x = _mm_cvtepu8_epi16(a1_above);
    }
    // y calc
    __m128i a0_y, a1_y, shifty;
    if (base_x < min_base_x) {
      DECLARE_ALIGNED(32, int16_t, base_y_c[8]);
      __m128i y_c, base_y_c_reg, mask, c1234_;
      c1234_ = _mm_srli_si128(c1234, 2);
      r6 = _mm_set1_epi16(r << 6);
      y_c = _mm_sub_epi16(r6, _mm_mullo_epi16(c1234_, dy_reg));
      base_y_c_reg = _mm_srai_epi16(y_c, frac_bits_y);
      mask = _mm_cmpgt_epi16(min_y_base, base_y_c_reg);
      base_y_c_reg = _mm_andnot_si128(mask, base_y_c_reg);
      _mm_store_si128((__m128i *)base_y_c, base_y_c_reg);

      a0_y = _mm_setr_epi16(left[base_y_c[0]], left[base_y_c[1]],
                            left[base_y_c[2]], left[base_y_c[3]], 0, 0, 0, 0);
      base_y_c_reg = _mm_add_epi16(base_y_c_reg, _mm_srli_epi16(a16, 4));
      _mm_store_si128((__m128i *)base_y_c, base_y_c_reg);
      a1_y = _mm_setr_epi16(left[base_y_c[0]], left[base_y_c[1]],
                            left[base_y_c[2]], left[base_y_c[3]], 0, 0, 0, 0);

      if (upsample_left) {
        shifty = _mm_srli_epi16(
            _mm_and_si128(_mm_slli_epi16(y_c, upsample_left), c3f), 1);
      } else {
        shifty = _mm_srli_epi16(_mm_and_si128(y_c, c3f), 1);
      }
      a0_x = _mm_unpacklo_epi64(a0_x, a0_y);
      a1_x = _mm_unpacklo_epi64(a1_x, a1_y);
      shift = _mm_unpacklo_epi64(shift, shifty);
    }

    diff = _mm_sub_epi16(a1_x, a0_x);  // a[x+1] - a[x]
    a32 = _mm_slli_epi16(a0_x, 5);     // a[x] * 32
    a32 = _mm_add_epi16(a32, a16);     // a[x] * 32 + 16

    b = _mm_mullo_epi16(diff, shift);
    res = _mm_add_epi16(a32, b);
    res = _mm_srli_epi16(res, 5);

    resx = _mm_packus_epi16(res, res);
    resy = _mm_srli_si128(resx, 4);

    resxy = _mm_blendv_epi8(resx, resy, *(__m128i *)Mask[0][base_min_diff]);
    *(int *)(dst) = _mm_cvtsi128_si32(resxy);
    dst += stride;
  }
}

static void dr_prediction_z2_Nx8_sse4_1(int N, uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *above,
                                        const uint8_t *left, int upsample_above,
                                        int upsample_left, int dx, int dy) {
  const int min_base_x = -(1 << upsample_above);
  const int min_base_y = -(1 << upsample_left);
  const int frac_bits_x = 6 - upsample_above;
  const int frac_bits_y = 6 - upsample_left;

  // pre-filter above pixels
  // store in temp buffers:
  //   above[x] * 32 + 16
  //   above[x+1] - above[x]
  // final pixels will be calculated as:
  //   (above[x] * 32 + 16 + (above[x+1] - above[x]) * shift) >> 5
  __m128i diff, a32;
  __m128i a0_x, a1_x, a0_y, a1_y;
  __m128i a0_above, a1_above;

  const __m128i a16 = _mm_set1_epi16(16);
  const __m128i c3f = _mm_set1_epi16(0x3f);
  const __m128i min_y_base = _mm_set1_epi16(min_base_y);
  const __m128i dy_reg = _mm_set1_epi16(dy);
  const __m128i c1234 = _mm_setr_epi16(1, 2, 3, 4, 5, 6, 7, 8);

  for (int r = 0; r < N; r++) {
    __m128i b, res, res1, shift;
    __m128i resx, resy, resxy, r6, ydx;

    int y = r + 1;
    int base_x = (-y * dx) >> frac_bits_x;
    int base_shift = 0;
    if (base_x < (min_base_x - 1)) {
      base_shift = (min_base_x - base_x - 1) >> upsample_above;
    }
    int base_min_diff =
        (min_base_x - base_x + upsample_above) >> upsample_above;
    if (base_min_diff > 8) {
      base_min_diff = 8;
    } else {
      if (base_min_diff < 0) base_min_diff = 0;
    }

    if (base_shift > 7) {
      resx = _mm_setzero_si128();
    } else {
      a0_above = _mm_loadu_si128((__m128i *)(above + base_x + base_shift));
      ydx = _mm_set1_epi16(y * dx);
      r6 = _mm_slli_epi16(_mm_srli_si128(c1234, 2), 6);
      if (upsample_above) {
        a0_above =
            _mm_shuffle_epi8(a0_above, *(__m128i *)EvenOddMaskx[base_shift]);
        a1_above = _mm_srli_si128(a0_above, 8);

        shift = _mm_srli_epi16(
            _mm_and_si128(
                _mm_slli_epi16(_mm_sub_epi16(r6, ydx), upsample_above), c3f),
            1);
      } else {
        a1_above = _mm_srli_si128(a0_above, 1);
        a0_above =
            _mm_shuffle_epi8(a0_above, *(__m128i *)LoadMaskx[base_shift]);
        a1_above =
            _mm_shuffle_epi8(a1_above, *(__m128i *)LoadMaskx[base_shift]);

        shift = _mm_srli_epi16(_mm_and_si128(_mm_sub_epi16(r6, ydx), c3f), 1);
      }
      a0_x = _mm_cvtepu8_epi16(a0_above);
      a1_x = _mm_cvtepu8_epi16(a1_above);

      diff = _mm_sub_epi16(a1_x, a0_x);  // a[x+1] - a[x]
      a32 = _mm_slli_epi16(a0_x, 5);     // a[x] * 32
      a32 = _mm_add_epi16(a32, a16);     // a[x] * 32 + 16

      b = _mm_mullo_epi16(diff, shift);
      res = _mm_add_epi16(a32, b);
      res = _mm_srli_epi16(res, 5);
      resx = _mm_packus_epi16(res, res);
    }

    // y calc
    if (base_x < min_base_x) {
      DECLARE_ALIGNED(32, int16_t, base_y_c[16]);
      __m128i y_c, base_y_c_reg, mask;
      r6 = _mm_set1_epi16(r << 6);
      y_c = _mm_sub_epi16(r6, _mm_mullo_epi16(c1234, dy_reg));
      base_y_c_reg = _mm_srai_epi16(y_c, frac_bits_y);
      mask = _mm_cmpgt_epi16(min_y_base, base_y_c_reg);
      base_y_c_reg = _mm_andnot_si128(mask, base_y_c_reg);
      _mm_store_si128((__m128i *)base_y_c, base_y_c_reg);

      a0_y = _mm_setr_epi16(left[base_y_c[0]], left[base_y_c[1]],
                            left[base_y_c[2]], left[base_y_c[3]],
                            left[base_y_c[4]], left[base_y_c[5]],
                            left[base_y_c[6]], left[base_y_c[7]]);
      base_y_c_reg = _mm_add_epi16(base_y_c_reg, _mm_srli_epi16(a16, 4));
      _mm_store_si128((__m128i *)base_y_c, base_y_c_reg);

      a1_y = _mm_setr_epi16(left[base_y_c[0]], left[base_y_c[1]],
                            left[base_y_c[2]], left[base_y_c[3]],
                            left[base_y_c[4]], left[base_y_c[5]],
                            left[base_y_c[6]], left[base_y_c[7]]);

      if (upsample_left) {
        shift = _mm_srli_epi16(
            _mm_and_si128(_mm_slli_epi16(y_c, upsample_left), c3f), 1);
      } else {
        shift = _mm_srli_epi16(_mm_and_si128(y_c, c3f), 1);
      }

      diff = _mm_sub_epi16(a1_y, a0_y);  // a[x+1] - a[x]
      a32 = _mm_slli_epi16(a0_y, 5);     // a[x] * 32
      a32 = _mm_add_epi16(a32, a16);     // a[x] * 32 + 16

      b = _mm_mullo_epi16(diff, shift);
      res1 = _mm_add_epi16(a32, b);
      res1 = _mm_srli_epi16(res1, 5);

      resy = _mm_packus_epi16(res1, res1);
      resxy = _mm_blendv_epi8(resx, resy, *(__m128i *)Mask[0][base_min_diff]);
      _mm_storel_epi64((__m128i *)dst, resxy);
    } else {
      _mm_storel_epi64((__m128i *)dst, resx);
    }

    dst += stride;
  }
}

static void dr_prediction_z2_HxW_sse4_1(int H, int W, uint8_t *dst,
                                        ptrdiff_t stride, const uint8_t *above,
                                        const uint8_t *left, int upsample_above,
                                        int upsample_left, int dx, int dy) {
  // here upsample_above and upsample_left are 0 by design of
  // av1_use_intra_edge_upsample
  const int min_base_x = -1;
  const int min_base_y = -1;
  (void)upsample_above;
  (void)upsample_left;
  const int frac_bits_x = 6;
  const int frac_bits_y = 6;

  __m128i a0_x, a1_x, a0_y, a1_y, a0_y_h, a1_y_h, a32;
  __m128i diff, shifty, shifty_h;
  __m128i a0_above, a1_above;

  DECLARE_ALIGNED(32, int16_t, base_y_c[16]);
  const __m128i a16 = _mm_set1_epi16(16);
  const __m128i c1 = _mm_srli_epi16(a16, 4);
  const __m128i min_y_base = _mm_set1_epi16(min_base_y);
  const __m128i c3f = _mm_set1_epi16(0x3f);
  const __m128i dy256 = _mm_set1_epi16(dy);
  const __m128i c0123 = _mm_setr_epi16(0, 1, 2, 3, 4, 5, 6, 7);
  const __m128i c0123_h = _mm_setr_epi16(8, 9, 10, 11, 12, 13, 14, 15);
  const __m128i c1234 = _mm_add_epi16(c0123, c1);
  const __m128i c1234_h = _mm_add_epi16(c0123_h, c1);

  for (int r = 0; r < H; r++) {
    __m128i b, res, res1, shift, reg_j, r6, ydx;
    __m128i resx, resy;
    __m128i resxy;
    int y = r + 1;
    ydx = _mm_set1_epi16((int16_t)(y * dx));

    int base_x = (-y * dx) >> frac_bits_x;
    for (int j = 0; j < W; j += 16) {
      reg_j = _mm_set1_epi16(j);
      int base_shift = 0;
      if ((base_x + j) < (min_base_x - 1)) {
        base_shift = (min_base_x - (base_x + j) - 1);
      }
      int base_min_diff = (min_base_x - base_x - j);
      if (base_min_diff > 16) {
        base_min_diff = 16;
      } else {
        if (base_min_diff < 0) base_min_diff = 0;
      }

      if (base_shift < 16) {
        a0_above =
            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + j));
        a1_above =
            _mm_loadu_si128((__m128i *)(above + base_x + base_shift + 1 + j));
        a0_above =
            _mm_shuffle_epi8(a0_above, *(__m128i *)LoadMaskx[base_shift]);
        a1_above =
            _mm_shuffle_epi8(a1_above, *(__m128i *)LoadMaskx[base_shift]);

        a0_x = _mm_cvtepu8_epi16(a0_above);
        a1_x = _mm_cvtepu8_epi16(a1_above);

        r6 = _mm_slli_epi16(_mm_add_epi16(c0123, reg_j), 6);
        shift = _mm_srli_epi16(_mm_and_si128(_mm_sub_epi16(r6, ydx), c3f), 1);

        diff = _mm_sub_epi16(a1_x, a0_x);  // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0_x, 5);     // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);     // a[x] * 32 + 16

        b = _mm_mullo_epi16(diff, shift);
        res = _mm_add_epi16(a32, b);
        res = _mm_srli_epi16(res, 5);  // 16 16-bit values

        a0_x = _mm_cvtepu8_epi16(_mm_srli_si128(a0_above, 8));
        a1_x = _mm_cvtepu8_epi16(_mm_srli_si128(a1_above, 8));

        r6 = _mm_slli_epi16(_mm_add_epi16(c0123_h, reg_j), 6);
        shift = _mm_srli_epi16(_mm_and_si128(_mm_sub_epi16(r6, ydx), c3f), 1);

        diff = _mm_sub_epi16(a1_x, a0_x);  // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0_x, 5);     // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);     // a[x] * 32 + 16

        b = _mm_mullo_epi16(diff, shift);
        res1 = _mm_add_epi16(a32, b);
        res1 = _mm_srli_epi16(res1, 5);  // 16 16-bit values

        resx = _mm_packus_epi16(res, res1);
      } else {
        resx = _mm_setzero_si128();
      }

      // y calc
      if (base_x < min_base_x) {
        __m128i c_reg, c_reg_h, y_reg, y_reg_h, base_y, base_y_h;
        __m128i mask, mask_h, mul16, mul16_h;
        r6 = _mm_set1_epi16(r << 6);
        c_reg = _mm_add_epi16(reg_j, c1234);
        c_reg_h = _mm_add_epi16(reg_j, c1234_h);
        mul16 = _mm_min_epu16(_mm_mullo_epi16(c_reg, dy256),
                              _mm_srli_epi16(min_y_base, 1));
        mul16_h = _mm_min_epu16(_mm_mullo_epi16(c_reg_h, dy256),
                                _mm_srli_epi16(min_y_base, 1));
        y_reg = _mm_sub_epi16(r6, mul16);
        y_reg_h = _mm_sub_epi16(r6, mul16_h);

        base_y = _mm_srai_epi16(y_reg, frac_bits_y);
        base_y_h = _mm_srai_epi16(y_reg_h, frac_bits_y);
        mask = _mm_cmpgt_epi16(min_y_base, base_y);
        mask_h = _mm_cmpgt_epi16(min_y_base, base_y_h);

        base_y = _mm_blendv_epi8(base_y, min_y_base, mask);
        base_y_h = _mm_blendv_epi8(base_y_h, min_y_base, mask_h);
        int16_t min_y = (int16_t)_mm_extract_epi16(base_y_h, 7);
        int16_t max_y = (int16_t)_mm_extract_epi16(base_y, 0);
        int16_t offset_diff = max_y - min_y;

        if (offset_diff < 16) {
          __m128i min_y_reg = _mm_set1_epi16(min_y);

          __m128i base_y_offset = _mm_sub_epi16(base_y, min_y_reg);
          __m128i base_y_offset_h = _mm_sub_epi16(base_y_h, min_y_reg);
          __m128i y_offset = _mm_packs_epi16(base_y_offset, base_y_offset_h);

          __m128i a0_mask = _mm_loadu_si128((__m128i *)(left + min_y));
          __m128i a1_mask = _mm_loadu_si128((__m128i *)(left + min_y + 1));
          __m128i LoadMask =
              _mm_loadu_si128((__m128i *)(LoadMaskz2[offset_diff / 4]));

          a0_mask = _mm_and_si128(a0_mask, LoadMask);
          a1_mask = _mm_and_si128(a1_mask, LoadMask);

          a0_mask = _mm_shuffle_epi8(a0_mask, y_offset);
          a1_mask = _mm_shuffle_epi8(a1_mask, y_offset);
          a0_y = _mm_cvtepu8_epi16(a0_mask);
          a1_y = _mm_cvtepu8_epi16(a1_mask);
          a0_y_h = _mm_cvtepu8_epi16(_mm_srli_si128(a0_mask, 8));
          a1_y_h = _mm_cvtepu8_epi16(_mm_srli_si128(a1_mask, 8));
        } else {
          base_y = _mm_andnot_si128(mask, base_y);
          base_y_h = _mm_andnot_si128(mask_h, base_y_h);
          _mm_store_si128((__m128i *)base_y_c, base_y);
          _mm_store_si128((__m128i *)&base_y_c[8], base_y_h);

          a0_y = _mm_setr_epi16(left[base_y_c[0]], left[base_y_c[1]],
                                left[base_y_c[2]], left[base_y_c[3]],
                                left[base_y_c[4]], left[base_y_c[5]],
                                left[base_y_c[6]], left[base_y_c[7]]);
          a0_y_h = _mm_setr_epi16(left[base_y_c[8]], left[base_y_c[9]],
                                  left[base_y_c[10]], left[base_y_c[11]],
                                  left[base_y_c[12]], left[base_y_c[13]],
                                  left[base_y_c[14]], left[base_y_c[15]]);
          base_y = _mm_add_epi16(base_y, c1);
          base_y_h = _mm_add_epi16(base_y_h, c1);
          _mm_store_si128((__m128i *)base_y_c, base_y);
          _mm_store_si128((__m128i *)&base_y_c[8], base_y_h);

          a1_y = _mm_setr_epi16(left[base_y_c[0]], left[base_y_c[1]],
                                left[base_y_c[2]], left[base_y_c[3]],
                                left[base_y_c[4]], left[base_y_c[5]],
                                left[base_y_c[6]], left[base_y_c[7]]);
          a1_y_h = _mm_setr_epi16(left[base_y_c[8]], left[base_y_c[9]],
                                  left[base_y_c[10]], left[base_y_c[11]],
                                  left[base_y_c[12]], left[base_y_c[13]],
                                  left[base_y_c[14]], left[base_y_c[15]]);
        }
        shifty = _mm_srli_epi16(_mm_and_si128(y_reg, c3f), 1);
        shifty_h = _mm_srli_epi16(_mm_and_si128(y_reg_h, c3f), 1);

        diff = _mm_sub_epi16(a1_y, a0_y);  // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0_y, 5);     // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);     // a[x] * 32 + 16

        b = _mm_mullo_epi16(diff, shifty);
        res = _mm_add_epi16(a32, b);
        res = _mm_srli_epi16(res, 5);  // 16 16-bit values

        diff = _mm_sub_epi16(a1_y_h, a0_y_h);  // a[x+1] - a[x]
        a32 = _mm_slli_epi16(a0_y_h, 5);       // a[x] * 32
        a32 = _mm_add_epi16(a32, a16);         // a[x] * 32 + 16

        b = _mm_mullo_epi16(diff, shifty_h);
        res1 = _mm_add_epi16(a32, b);
        res1 = _mm_srli_epi16(res1, 5);  // 16 16-bit values
        resy = _mm_packus_epi16(res, res1);
      } else {
        resy = _mm_setzero_si128();
      }
      resxy = _mm_blendv_epi8(resx, resy, *(__m128i *)Mask[0][base_min_diff]);
      _mm_storeu_si128((__m128i *)(dst + j), resxy);
    }  // for j
    dst += stride;
  }
}

// Directional prediction, zone 2: 90 < angle < 180
void av1_dr_prediction_z2_sse4_1(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                                 const uint8_t *above, const uint8_t *left,
                                 int upsample_above, int upsample_left, int dx,
                                 int dy) {
  assert(dx > 0);
  assert(dy > 0);
  switch (bw) {
    case 4:
      dr_prediction_z2_Nx4_sse4_1(bh, dst, stride, above, left, upsample_above,
                                  upsample_left, dx, dy);
      break;
    case 8:
      dr_prediction_z2_Nx8_sse4_1(bh, dst, stride, above, left, upsample_above,
                                  upsample_left, dx, dy);
      break;
    default:
      dr_prediction_z2_HxW_sse4_1(bh, bw, dst, stride, above, left,
                                  upsample_above, upsample_left, dx, dy);
  }
  return;
}

// z3 functions
static void dr_prediction_z3_4x4_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *left, int upsample_left,
                                        int dy) {
  __m128i dstvec[4], d[4];

  dr_prediction_z1_HxW_internal_sse4_1(4, 4, dstvec, left, upsample_left, dy);
  transpose4x8_8x4_low_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3],
                            &d[0], &d[1], &d[2], &d[3]);

  *(int *)(dst + stride * 0) = _mm_cvtsi128_si32(d[0]);
  *(int *)(dst + stride * 1) = _mm_cvtsi128_si32(d[1]);
  *(int *)(dst + stride * 2) = _mm_cvtsi128_si32(d[2]);
  *(int *)(dst + stride * 3) = _mm_cvtsi128_si32(d[3]);
  return;
}

static void dr_prediction_z3_8x8_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *left, int upsample_left,
                                        int dy) {
  __m128i dstvec[8], d[8];

  dr_prediction_z1_HxW_internal_sse4_1(8, 8, dstvec, left, upsample_left, dy);
  transpose8x8_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4],
                    &dstvec[5], &dstvec[6], &dstvec[7], &d[0], &d[1], &d[2],
                    &d[3]);

  _mm_storel_epi64((__m128i *)(dst + 0 * stride), d[0]);
  _mm_storel_epi64((__m128i *)(dst + 1 * stride), _mm_srli_si128(d[0], 8));
  _mm_storel_epi64((__m128i *)(dst + 2 * stride), d[1]);
  _mm_storel_epi64((__m128i *)(dst + 3 * stride), _mm_srli_si128(d[1], 8));
  _mm_storel_epi64((__m128i *)(dst + 4 * stride), d[2]);
  _mm_storel_epi64((__m128i *)(dst + 5 * stride), _mm_srli_si128(d[2], 8));
  _mm_storel_epi64((__m128i *)(dst + 6 * stride), d[3]);
  _mm_storel_epi64((__m128i *)(dst + 7 * stride), _mm_srli_si128(d[3], 8));
}

static void dr_prediction_z3_4x8_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *left, int upsample_left,
                                        int dy) {
  __m128i dstvec[4], d[8];

  dr_prediction_z1_HxW_internal_sse4_1(8, 4, dstvec, left, upsample_left, dy);
  transpose4x8_8x4_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &d[0],
                        &d[1], &d[2], &d[3], &d[4], &d[5], &d[6], &d[7]);
  for (int i = 0; i < 8; i++) {
    *(int *)(dst + stride * i) = _mm_cvtsi128_si32(d[i]);
  }
}

static void dr_prediction_z3_8x4_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                        const uint8_t *left, int upsample_left,
                                        int dy) {
  __m128i dstvec[8], d[4];

  dr_prediction_z1_HxW_internal_sse4_1(4, 8, dstvec, left, upsample_left, dy);
  transpose8x8_low_sse2(&dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3],
                        &dstvec[4], &dstvec[5], &dstvec[6], &dstvec[7], &d[0],
                        &d[1], &d[2], &d[3]);
  _mm_storel_epi64((__m128i *)(dst + 0 * stride), d[0]);
  _mm_storel_epi64((__m128i *)(dst + 1 * stride), d[1]);
  _mm_storel_epi64((__m128i *)(dst + 2 * stride), d[2]);
  _mm_storel_epi64((__m128i *)(dst + 3 * stride), d[3]);
}

static void dr_prediction_z3_8x16_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *left, int upsample_left,
                                         int dy) {
  __m128i dstvec[8], d[8];

  dr_prediction_z1_HxW_internal_sse4_1(16, 8, dstvec, left, upsample_left, dy);
  transpose8x16_16x8_sse2(dstvec, dstvec + 1, dstvec + 2, dstvec + 3,
                          dstvec + 4, dstvec + 5, dstvec + 6, dstvec + 7, d,
                          d + 1, d + 2, d + 3, d + 4, d + 5, d + 6, d + 7);
  for (int i = 0; i < 8; i++) {
    _mm_storel_epi64((__m128i *)(dst + i * stride), d[i]);
    _mm_storel_epi64((__m128i *)(dst + (i + 8) * stride),
                     _mm_srli_si128(d[i], 8));
  }
}

static void dr_prediction_z3_16x8_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *left, int upsample_left,
                                         int dy) {
  __m128i dstvec[16], d[16];

  dr_prediction_z1_HxW_internal_sse4_1(8, 16, dstvec, left, upsample_left, dy);
  transpose16x8_8x16_sse2(
      &dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4], &dstvec[5],
      &dstvec[6], &dstvec[7], &dstvec[8], &dstvec[9], &dstvec[10], &dstvec[11],
      &dstvec[12], &dstvec[13], &dstvec[14], &dstvec[15], &d[0], &d[1], &d[2],
      &d[3], &d[4], &d[5], &d[6], &d[7]);

  for (int i = 0; i < 8; i++) {
    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
  }
}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
static void dr_prediction_z3_4x16_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *left, int upsample_left,
                                         int dy) {
  __m128i dstvec[4], d[16];

  dr_prediction_z1_HxW_internal_sse4_1(16, 4, dstvec, left, upsample_left, dy);
  transpose4x16_sse2(dstvec, d);
  for (int i = 0; i < 16; i++) {
    *(int *)(dst + stride * i) = _mm_cvtsi128_si32(d[i]);
  }
}

static void dr_prediction_z3_16x4_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *left, int upsample_left,
                                         int dy) {
  __m128i dstvec[16], d[8];

  dr_prediction_z1_HxW_internal_sse4_1(4, 16, dstvec, left, upsample_left, dy);
  for (int i = 4; i < 8; i++) {
    d[i] = _mm_setzero_si128();
  }
  transpose16x8_8x16_sse2(
      &dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4], &dstvec[5],
      &dstvec[6], &dstvec[7], &dstvec[8], &dstvec[9], &dstvec[10], &dstvec[11],
      &dstvec[12], &dstvec[13], &dstvec[14], &dstvec[15], &d[0], &d[1], &d[2],
      &d[3], &d[4], &d[5], &d[6], &d[7]);

  for (int i = 0; i < 4; i++) {
    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
  }
}

static void dr_prediction_z3_8x32_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *left, int upsample_left,
                                         int dy) {
  __m128i dstvec[16], d[16], dstvec_h[16], d_h[16];

  dr_prediction_z1_32xN_internal_sse4_1(8, dstvec, dstvec_h, left,
                                        upsample_left, dy);
  for (int i = 8; i < 16; i++) {
    dstvec[i] = _mm_setzero_si128();
    dstvec_h[i] = _mm_setzero_si128();
  }
  transpose16x16_sse2(dstvec, d);
  transpose16x16_sse2(dstvec_h, d_h);

  for (int i = 0; i < 16; i++) {
    _mm_storel_epi64((__m128i *)(dst + i * stride), d[i]);
  }
  for (int i = 0; i < 16; i++) {
    _mm_storel_epi64((__m128i *)(dst + (i + 16) * stride), d_h[i]);
  }
}

static void dr_prediction_z3_32x8_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                         const uint8_t *left, int upsample_left,
                                         int dy) {
  __m128i dstvec[32], d[16];

  dr_prediction_z1_HxW_internal_sse4_1(8, 32, dstvec, left, upsample_left, dy);

  transpose16x8_8x16_sse2(
      &dstvec[0], &dstvec[1], &dstvec[2], &dstvec[3], &dstvec[4], &dstvec[5],
      &dstvec[6], &dstvec[7], &dstvec[8], &dstvec[9], &dstvec[10], &dstvec[11],
      &dstvec[12], &dstvec[13], &dstvec[14], &dstvec[15], &d[0], &d[1], &d[2],
      &d[3], &d[4], &d[5], &d[6], &d[7]);
  transpose16x8_8x16_sse2(
      &dstvec[0 + 16], &dstvec[1 + 16], &dstvec[2 + 16], &dstvec[3 + 16],
      &dstvec[4 + 16], &dstvec[5 + 16], &dstvec[6 + 16], &dstvec[7 + 16],
      &dstvec[8 + 16], &dstvec[9 + 16], &dstvec[10 + 16], &dstvec[11 + 16],
      &dstvec[12 + 16], &dstvec[13 + 16], &dstvec[14 + 16], &dstvec[15 + 16],
      &d[0 + 8], &d[1 + 8], &d[2 + 8], &d[3 + 8], &d[4 + 8], &d[5 + 8],
      &d[6 + 8], &d[7 + 8]);

  for (int i = 0; i < 8; i++) {
    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
    _mm_storeu_si128((__m128i *)(dst + i * stride + 16), d[i + 8]);
  }
}
#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

static void dr_prediction_z3_16x16_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  __m128i dstvec[16], d[16];

  dr_prediction_z1_HxW_internal_sse4_1(16, 16, dstvec, left, upsample_left, dy);
  transpose16x16_sse2(dstvec, d);

  for (int i = 0; i < 16; i++) {
    _mm_storeu_si128((__m128i *)(dst + i * stride), d[i]);
  }
}

static void dr_prediction_z3_32x32_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  __m128i dstvec[32], d[32], dstvec_h[32], d_h[32];

  dr_prediction_z1_32xN_internal_sse4_1(32, dstvec, dstvec_h, left,
                                        upsample_left, dy);
  transpose16x16_sse2(dstvec, d);
  transpose16x16_sse2(dstvec_h, d_h);
  transpose16x16_sse2(dstvec + 16, d + 16);
  transpose16x16_sse2(dstvec_h + 16, d_h + 16);
  for (int j = 0; j < 16; j++) {
    _mm_storeu_si128((__m128i *)(dst + j * stride), d[j]);
    _mm_storeu_si128((__m128i *)(dst + j * stride + 16), d[j + 16]);
  }
  for (int j = 0; j < 16; j++) {
    _mm_storeu_si128((__m128i *)(dst + (j + 16) * stride), d_h[j]);
    _mm_storeu_si128((__m128i *)(dst + (j + 16) * stride + 16), d_h[j + 16]);
  }
}

static void dr_prediction_z3_64x64_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  uint8_t dstT[64 * 64];
  dr_prediction_z1_64xN_sse4_1(64, dstT, 64, left, upsample_left, dy);
  transpose(dstT, 64, dst, stride, 64, 64);
}

static void dr_prediction_z3_16x32_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  __m128i dstvec[16], d[16], dstvec_h[16], d_h[16];

  dr_prediction_z1_32xN_internal_sse4_1(16, dstvec, dstvec_h, left,
                                        upsample_left, dy);
  transpose16x16_sse2(dstvec, d);
  transpose16x16_sse2(dstvec_h, d_h);
  // store
  for (int j = 0; j < 16; j++) {
    _mm_storeu_si128((__m128i *)(dst + j * stride), d[j]);
    _mm_storeu_si128((__m128i *)(dst + (j + 16) * stride), d_h[j]);
  }
}

static void dr_prediction_z3_32x16_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  __m128i dstvec[32], d[16];

  dr_prediction_z1_HxW_internal_sse4_1(16, 32, dstvec, left, upsample_left, dy);
  for (int i = 0; i < 32; i += 16) {
    transpose16x16_sse2((dstvec + i), d);
    for (int j = 0; j < 16; j++) {
      _mm_storeu_si128((__m128i *)(dst + j * stride + i), d[j]);
    }
  }
}

static void dr_prediction_z3_32x64_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  uint8_t dstT[64 * 32];
  dr_prediction_z1_64xN_sse4_1(32, dstT, 64, left, upsample_left, dy);
  transpose(dstT, 64, dst, stride, 32, 64);
}

static void dr_prediction_z3_64x32_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  uint8_t dstT[32 * 64];
  dr_prediction_z1_32xN_sse4_1(64, dstT, 32, left, upsample_left, dy);
  transpose(dstT, 32, dst, stride, 64, 32);
  return;
}

#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
static void dr_prediction_z3_16x64_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  uint8_t dstT[64 * 16];
  dr_prediction_z1_64xN_sse4_1(16, dstT, 64, left, upsample_left, dy);
  transpose(dstT, 64, dst, stride, 16, 64);
}

static void dr_prediction_z3_64x16_sse4_1(uint8_t *dst, ptrdiff_t stride,
                                          const uint8_t *left,
                                          int upsample_left, int dy) {
  __m128i dstvec[64], d[16];

  dr_prediction_z1_HxW_internal_sse4_1(16, 64, dstvec, left, upsample_left, dy);
  for (int i = 0; i < 64; i += 16) {
    transpose16x16_sse2(dstvec + i, d);
    for (int j = 0; j < 16; j++) {
      _mm_storeu_si128((__m128i *)(dst + j * stride + i), d[j]);
    }
  }
}
#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER

void av1_dr_prediction_z3_sse4_1(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
                                 const uint8_t *above, const uint8_t *left,
                                 int upsample_left, int dx, int dy) {
  (void)above;
  (void)dx;
  assert(dx == 1);
  assert(dy > 0);

  if (bw == bh) {
    switch (bw) {
      case 4:
        dr_prediction_z3_4x4_sse4_1(dst, stride, left, upsample_left, dy);
        break;
      case 8:
        dr_prediction_z3_8x8_sse4_1(dst, stride, left, upsample_left, dy);
        break;
      case 16:
        dr_prediction_z3_16x16_sse4_1(dst, stride, left, upsample_left, dy);
        break;
      case 32:
        dr_prediction_z3_32x32_sse4_1(dst, stride, left, upsample_left, dy);
        break;
      case 64:
        dr_prediction_z3_64x64_sse4_1(dst, stride, left, upsample_left, dy);
        break;
      default: assert(0 && "Invalid block size");
    }
  } else {
    if (bw < bh) {
      if (bw + bw == bh) {
        switch (bw) {
          case 4:
            dr_prediction_z3_4x8_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 8:
            dr_prediction_z3_8x16_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 16:
            dr_prediction_z3_16x32_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 32:
            dr_prediction_z3_32x64_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          default: assert(0 && "Invalid block size");
        }
      } else {
        switch (bw) {
#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
          case 4:
            dr_prediction_z3_4x16_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 8:
            dr_prediction_z3_8x32_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 16:
            dr_prediction_z3_16x64_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          default: assert(0 && "Invalid block size");
#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
        }
      }
    } else {
      if (bh + bh == bw) {
        switch (bh) {
          case 4:
            dr_prediction_z3_8x4_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 8:
            dr_prediction_z3_16x8_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 16:
            dr_prediction_z3_32x16_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 32:
            dr_prediction_z3_64x32_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          default: assert(0 && "Invalid block size");
        }
      } else {
        switch (bh) {
#if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
          case 4:
            dr_prediction_z3_16x4_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 8:
            dr_prediction_z3_32x8_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          case 16:
            dr_prediction_z3_64x16_sse4_1(dst, stride, left, upsample_left, dy);
            break;
          default: assert(0 && "Invalid block size");
#endif  // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER
        }
      }
    }
  }
}