Quelle  intrapred8_dspr2.c   Sprache: C

/*
 *  Copyright (c) 2015 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include "vpx_dsp/mips/common_dspr2.h"

#if HAVE_DSPR2
void vpx_h_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
                               const uint8_t *above, const uint8_t *left) {
  int32_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
  (void)above;

  __asm__ __volatile__(
      "lb %[tmp1], (%[left]) \n\t"
      "lb %[tmp2], 1(%[left]) \n\t"
      "lb %[tmp3], 2(%[left]) \n\t"
      "lb %[tmp4], 3(%[left]) \n\t"
      "lb %[tmp5], 4(%[left]) \n\t"
      "lb %[tmp6], 5(%[left]) \n\t"
      "lb %[tmp7], 6(%[left]) \n\t"
      "lb %[tmp8], 7(%[left]) \n\t"

      "replv.qb %[tmp1], %[tmp1] \n\t"
      "replv.qb %[tmp2], %[tmp2] \n\t"
      "replv.qb %[tmp3], %[tmp3] \n\t"
      "replv.qb %[tmp4], %[tmp4] \n\t"
      "replv.qb %[tmp5], %[tmp5] \n\t"
      "replv.qb %[tmp6], %[tmp6] \n\t"
      "replv.qb %[tmp7], %[tmp7] \n\t"
      "replv.qb %[tmp8], %[tmp8] \n\t"

      "sw %[tmp1], (%[dst]) \n\t"
      "sw %[tmp1], 4(%[dst]) \n\t"
      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[tmp2], (%[dst]) \n\t"
      "sw %[tmp2], 4(%[dst]) \n\t"
      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[tmp3], (%[dst]) \n\t"
      "sw %[tmp3], 4(%[dst]) \n\t"
      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[tmp4], (%[dst]) \n\t"
      "sw %[tmp4], 4(%[dst]) \n\t"
      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[tmp5], (%[dst]) \n\t"
      "sw %[tmp5], 4(%[dst]) \n\t"
      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[tmp6], (%[dst]) \n\t"
      "sw %[tmp6], 4(%[dst]) \n\t"
      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[tmp7], (%[dst]) \n\t"
      "sw %[tmp7], 4(%[dst]) \n\t"
      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[tmp8], (%[dst]) \n\t"
      "sw %[tmp8], 4(%[dst]) \n\t"

      : [tmp1] "=&r"(tmp1), [tmp2] "=&r"(tmp2), [tmp3] "=&r"(tmp3),
        [tmp4] "=&r"(tmp4), [tmp5] "=&r"(tmp5), [tmp7] "=&r"(tmp7),
        [tmp6] "=&r"(tmp6), [tmp8] "=&r"(tmp8)
      : [left] "r"(left), [dst] "r"(dst), [stride] "r"(stride));
}

void vpx_dc_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
                                const uint8_t *above, const uint8_t *left) {
  int32_t expected_dc;
  int32_t average;
  int32_t tmp, above1, above_l1, above_r1, left1, left_r1, left_l1;
  int32_t above2, above_l2, above_r2, left2, left_r2, left_l2;

  __asm__ __volatile__(
      "lw %[above1], (%[above]) \n\t"
      "lw %[above2], 4(%[above]) \n\t"
      "lw %[left1], (%[left]) \n\t"
      "lw %[left2], 4(%[left]) \n\t"

      "preceu.ph.qbl %[above_l1], %[above1] \n\t"
      "preceu.ph.qbr %[above_r1], %[above1] \n\t"
      "preceu.ph.qbl %[left_l1], %[left1] \n\t"
      "preceu.ph.qbr %[left_r1], %[left1] \n\t"

      "preceu.ph.qbl %[above_l2], %[above2] \n\t"
      "preceu.ph.qbr %[above_r2], %[above2] \n\t"
      "preceu.ph.qbl %[left_l2], %[left2] \n\t"
      "preceu.ph.qbr %[left_r2], %[left2] \n\t"

      "addu.ph %[average], %[above_r1], %[above_l1] \n\t"
      "addu.ph %[average], %[average], %[left_l1] \n\t"
      "addu.ph %[average], %[average], %[left_r1] \n\t"

      "addu.ph %[average], %[average], %[above_l2] \n\t"
      "addu.ph %[average], %[average], %[above_r2] \n\t"
      "addu.ph %[average], %[average], %[left_l2] \n\t"
      "addu.ph %[average], %[average], %[left_r2] \n\t"

      "addiu %[average], %[average], 8 \n\t"

      "srl %[tmp], %[average], 16 \n\t"
      "addu.ph %[average], %[tmp], %[average] \n\t"
      "srl %[expected_dc], %[average], 4 \n\t"
      "replv.qb %[expected_dc], %[expected_dc] \n\t"

      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      "add %[dst], %[dst], %[stride] \n\t"
      "sw %[expected_dc], (%[dst]) \n\t"
      "sw %[expected_dc], 4(%[dst]) \n\t"

      : [above1] "=&r"(above1), [above_l1] "=&r"(above_l1),
        [above_r1] "=&r"(above_r1), [left1] "=&r"(left1),
        [left_l1] "=&r"(left_l1), [left_r1] "=&r"(left_r1),
        [above2] "=&r"(above2), [above_l2] "=&r"(above_l2),
        [above_r2] "=&r"(above_r2), [left2] "=&r"(left2),
        [left_l2] "=&r"(left_l2), [left_r2] "=&r"(left_r2),
        [average] "=&r"(average), [tmp] "=&r"(tmp),
        [expected_dc] "=&r"(expected_dc)
      : [above] "r"(above), [left] "r"(left), [dst] "r"(dst),
        [stride] "r"(stride));
}

void vpx_tm_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
                                const uint8_t *above, const uint8_t *left) {
  int32_t abovel, abover;
  int32_t abovel_1, abover_1;
  int32_t left0;
  int32_t res0, res1, res2, res3;
  int32_t reshw;
  int32_t top_left;
  uint8_t *cm = vpx_ff_cropTbl;

  __asm__ __volatile__(
      "ulw %[reshw], (%[above]) \n\t"
      "ulw %[top_left], 4(%[above]) \n\t"

      "lbu %[left0], (%[left]) \n\t"

      "preceu.ph.qbl %[abovel], %[reshw] \n\t"
      "preceu.ph.qbr %[abover], %[reshw] \n\t"
      "preceu.ph.qbl %[abovel_1], %[top_left] \n\t"
      "preceu.ph.qbr %[abover_1], %[top_left] \n\t"

      "lbu %[top_left], -1(%[above]) \n\t"
      "replv.ph %[left0], %[left0] \n\t"

      "replv.ph %[top_left], %[top_left] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbu %[left0], 1(%[left]) \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      "replv.ph %[left0], %[left0] \n\t"
      "add %[dst], %[dst], %[stride] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbu %[left0], 2(%[left]) \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      "replv.ph %[left0], %[left0] \n\t"
      "add %[dst], %[dst], %[stride] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbu %[left0], 3(%[left]) \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      "replv.ph %[left0], %[left0] \n\t"
      "add %[dst], %[dst], %[stride] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbu %[left0], 4(%[left]) \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      "replv.ph %[left0], %[left0] \n\t"
      "add %[dst], %[dst], %[stride] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbu %[left0], 5(%[left]) \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      "replv.ph %[left0], %[left0] \n\t"
      "add %[dst], %[dst], %[stride] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbu %[left0], 6(%[left]) \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      "replv.ph %[left0], %[left0] \n\t"
      "add %[dst], %[dst], %[stride] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbu %[left0], 7(%[left]) \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      "replv.ph %[left0], %[left0] \n\t"
      "add %[dst], %[dst], %[stride] \n\t"

      "addu.ph %[reshw], %[abovel], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], (%[dst]) \n\t"
      "sb %[res1], 1(%[dst]) \n\t"
      "sb %[res2], 2(%[dst]) \n\t"
      "sb %[res3], 3(%[dst]) \n\t"

      "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res2], %[reshw], 16 \n\t"
      "sra %[res2], %[res2], 16 \n\t"
      "sra %[res3], %[reshw], 16 \n\t"

      "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
      "subu.ph %[reshw], %[reshw], %[top_left] \n\t"

      "sll %[res0], %[reshw], 16 \n\t"
      "sra %[res0], %[res0], 16 \n\t"
      "sra %[res1], %[reshw], 16 \n\t"

      "lbux %[res0], %[res0](%[cm]) \n\t"
      "lbux %[res1], %[res1](%[cm]) \n\t"
      "lbux %[res2], %[res2](%[cm]) \n\t"
      "lbux %[res3], %[res3](%[cm]) \n\t"

      "sb %[res0], 4(%[dst]) \n\t"
      "sb %[res1], 5(%[dst]) \n\t"
      "sb %[res2], 6(%[dst]) \n\t"
      "sb %[res3], 7(%[dst]) \n\t"

      : [abovel] "=&r"(abovel), [abover] "=&r"(abover),
        [abovel_1] "=&r"(abovel_1), [abover_1] "=&r"(abover_1),
        [left0] "=&r"(left0), [res2] "=&r"(res2), [res3] "=&r"(res3),
        [res0] "=&r"(res0), [res1] "=&r"(res1), [reshw] "=&r"(reshw),
        [top_left] "=&r"(top_left)
      : [above] "r"(above), [left] "r"(left), [dst] "r"(dst),
        [stride] "r"(stride), [cm] "r"(cm));
}
#endif  // #if HAVE_DSPR2