Quelle  convolve2_dspr2.c   Sprache: C

/*
 *  Copyright (c) 2013 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 <assert.h>
#include <stdio.h>

#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/convolve_common_dspr2.h"
#include "vpx_dsp/vpx_dsp_common.h"
#include "vpx_dsp/vpx_filter.h"
#include "vpx_ports/mem.h"

#if HAVE_DSPR2
static void convolve_bi_horiz_4_transposed_dspr2(
    const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
    const int16_t *filter_x0, int32_t h) {
  int32_t y;
  uint8_t *cm = vpx_ff_cropTbl;
  uint8_t *dst_ptr;
  int32_t Temp1, Temp2;
  uint32_t vector4a = 64;
  uint32_t tp1, tp2;
  uint32_t p1, p2;
  const int16_t *filter = &filter_x0[3];
  uint32_t filter45;

  filter45 = ((const int32_t *)filter)[0];

  for (y = h; y--;) {
    dst_ptr = dst;
    /* prefetch data to cache memory */
    prefetch_load(src + src_stride);
    prefetch_load(src + src_stride + 32);

    __asm__ __volatile__(
        "ulw %[tp1], 0(%[src]) \n\t"
        "ulw %[tp2], 4(%[src]) \n\t"

        /* even 1. pixel */
        "mtlo %[vector4a], $ac3 \n\t"
        "mthi $zero, $ac3 \n\t"
        "preceu.ph.qbr %[p1], %[tp1] \n\t"
        "preceu.ph.qbl %[p2], %[tp1] \n\t"
        "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
        "extp %[Temp1], $ac3, 31 \n\t"

        /* even 2. pixel */
        "mtlo %[vector4a], $ac2 \n\t"
        "mthi $zero, $ac2 \n\t"
        "balign %[tp2], %[tp1], 3 \n\t"
        "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
        "extp %[Temp2], $ac2, 31 \n\t"

        /* odd 1. pixel */
        "lbux %[tp1], %[Temp1](%[cm]) \n\t"
        "mtlo %[vector4a], $ac3 \n\t"
        "mthi $zero, $ac3 \n\t"
        "preceu.ph.qbr %[p1], %[tp2] \n\t"
        "preceu.ph.qbl %[p2], %[tp2] \n\t"
        "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
        "extp %[Temp1], $ac3, 31 \n\t"

        /* odd 2. pixel */
        "lbux %[tp2], %[Temp2](%[cm]) \n\t"
        "mtlo %[vector4a], $ac2 \n\t"
        "mthi $zero, $ac2 \n\t"
        "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
        "extp %[Temp2], $ac2, 31 \n\t"

        /* clamp */
        "lbux %[p1], %[Temp1](%[cm]) \n\t"
        "lbux %[p2], %[Temp2](%[cm]) \n\t"

        /* store bytes */
        "sb %[tp1], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"

        "sb %[p1], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"

        "sb %[tp2], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"

        "sb %[p2], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_stride] \n\t"

        : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [p1] "=&r"(p1), [p2] "=&r"(p2),
          [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2), [dst_ptr] "+r"(dst_ptr)
        : [filter45] "r"(filter45), [vector4a] "r"(vector4a), [cm] "r"(cm),
          [src] "r"(src), [dst_stride] "r"(dst_stride));

    /* Next row... */
    src += src_stride;
    dst += 1;
  }
}

static void convolve_bi_horiz_8_transposed_dspr2(
    const uint8_t *src, int32_t src_stride, uint8_t *dst, int32_t dst_stride,
    const int16_t *filter_x0, int32_t h) {
  int32_t y;
  uint8_t *cm = vpx_ff_cropTbl;
  uint8_t *dst_ptr;
  uint32_t vector4a = 64;
  int32_t Temp1, Temp2, Temp3;
  uint32_t tp1, tp2, tp3;
  uint32_t p1, p2, p3, p4;
  uint8_t *odd_dst;
  uint32_t dst_pitch_2 = (dst_stride << 1);
  const int16_t *filter = &filter_x0[3];
  uint32_t filter45;

  filter45 = ((const int32_t *)filter)[0];

  for (y = h; y--;) {
    /* prefetch data to cache memory */
    prefetch_load(src + src_stride);
    prefetch_load(src + src_stride + 32);

    dst_ptr = dst;
    odd_dst = (dst_ptr + dst_stride);

    __asm__ __volatile__(
        "ulw %[tp1], 0(%[src]) \n\t"
        "ulw %[tp2], 4(%[src]) \n\t"

        /* even 1. pixel */
        "mtlo %[vector4a], $ac3 \n\t"
        "mthi $zero, $ac3 \n\t"
        "mtlo %[vector4a], $ac2 \n\t"
        "mthi $zero, $ac2 \n\t"
        "preceu.ph.qbr %[p1], %[tp1] \n\t"
        "preceu.ph.qbl %[p2], %[tp1] \n\t"
        "preceu.ph.qbr %[p3], %[tp2] \n\t"
        "preceu.ph.qbl %[p4], %[tp2] \n\t"
        "ulw %[tp3], 8(%[src]) \n\t"
        "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
        "extp %[Temp1], $ac3, 31 \n\t"

        /* even 2. pixel */
        "dpa.w.ph $ac2, %[p2], %[filter45] \n\t"
        "extp %[Temp3], $ac2, 31 \n\t"

        /* even 3. pixel */
        "lbux %[Temp2], %[Temp1](%[cm]) \n\t"
        "mtlo %[vector4a], $ac1 \n\t"
        "mthi $zero, $ac1 \n\t"
        "balign %[tp3], %[tp2], 3 \n\t"
        "balign %[tp2], %[tp1], 3 \n\t"
        "dpa.w.ph $ac1, %[p3], %[filter45] \n\t"
        "lbux %[tp1], %[Temp3](%[cm]) \n\t"
        "extp %[p3], $ac1, 31 \n\t"

        /* even 4. pixel */
        "mtlo %[vector4a], $ac2 \n\t"
        "mthi $zero, $ac2 \n\t"
        "mtlo %[vector4a], $ac3 \n\t"
        "mthi $zero, $ac3 \n\t"
        "sb %[Temp2], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
        "sb %[tp1], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"

        "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
        "extp %[Temp3], $ac2, 31 \n\t"

        "lbux %[Temp1], %[p3](%[cm]) "
        "\n\t"

        /* odd 1. pixel */
        "mtlo %[vector4a], $ac1 \n\t"
        "mthi $zero, $ac1 \n\t"
        "preceu.ph.qbr %[p1], %[tp2] \n\t"
        "preceu.ph.qbl %[p2], %[tp2] \n\t"
        "preceu.ph.qbr %[p3], %[tp3] \n\t"
        "preceu.ph.qbl %[p4], %[tp3] \n\t"
        "sb %[Temp1], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"

        "dpa.w.ph $ac3, %[p1], %[filter45] \n\t"
        "extp %[Temp2], $ac3, 31 \n\t"

        /* odd 2. pixel */
        "lbux %[tp1], %[Temp3](%[cm]) \n\t"
        "mtlo %[vector4a], $ac3 \n\t"
        "mthi $zero, $ac3 \n\t"
        "mtlo %[vector4a], $ac2 \n\t"
        "mthi $zero, $ac2 \n\t"
        "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
        "sb %[tp1], 0(%[dst_ptr]) \n\t"
        "addu %[dst_ptr], %[dst_ptr], %[dst_pitch_2] \n\t"
        "extp %[Temp3], $ac1, 31 \n\t"

        /* odd 3. pixel */
        "lbux %[tp3], %[Temp2](%[cm]) \n\t"
        "dpa.w.ph $ac3, %[p3], %[filter45] \n\t"
        "extp %[Temp2], $ac3, 31 \n\t"

        /* odd 4. pixel */
        "sb %[tp3], 0(%[odd_dst]) \n\t"
        "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"
        "dpa.w.ph $ac2, %[p4], %[filter45] \n\t"
        "extp %[Temp1], $ac2, 31 \n\t"

        /* clamp */
        "lbux %[p4], %[Temp3](%[cm]) \n\t"
        "lbux %[p2], %[Temp2](%[cm]) \n\t"
        "lbux %[p1], %[Temp1](%[cm]) \n\t"

        /* store bytes */
        "sb %[p4], 0(%[odd_dst]) \n\t"
        "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"

        "sb %[p2], 0(%[odd_dst]) \n\t"
        "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] \n\t"

        "sb %[p1], 0(%[odd_dst]) \n\t"

        : [tp1] "=&r"(tp1), [tp2] "=&r"(tp2), [tp3] "=&r"(tp3), [p1] "=&r"(p1),
          [p2] "=&r"(p2), [p3] "=&r"(p3), [p4] "=&r"(p4), [Temp1] "=&r"(Temp1),
          [Temp2] "=&r"(Temp2), [Temp3] "=&r"(Temp3), [dst_ptr] "+r"(dst_ptr),
          [odd_dst] "+r"(odd_dst)
        : [filter45] "r"(filter45), [vector4a] "r"(vector4a), [cm] "r"(cm),
          [src] "r"(src), [dst_pitch_2] "r"(dst_pitch_2));

    /* Next row... */
    src += src_stride;
    dst += 1;
  }
}

static void convolve_bi_horiz_16_transposed_dspr2(
    const uint8_t *src_ptr, int32_t src_stride, uint8_t *dst_ptr,
    int32_t dst_stride, const int16_t *filter_x0, int32_t h, int32_t count) {
  int32_t c, y;
  const uint8_t *src;
  uint8_t *dst;
  uint8_t *cm = vpx_ff_cropTbl;
  uint32_t vector_64 = 64;
  int32_t Temp1, Temp2, Temp3;
  uint32_t qload1, qload2;
  uint32_t p1, p2, p3, p4, p5;
  uint32_t st1, st2, st3;
  uint32_t dst_pitch_2 = (dst_stride << 1);
  uint8_t *odd_dst;
  const int16_t *filter = &filter_x0[3];
  uint32_t filter45;

  filter45 = ((const int32_t *)filter)[0];

  for (y = h; y--;) {
    /* prefetch data to cache memory */
    prefetch_load(src_ptr + src_stride);
    prefetch_load(src_ptr + src_stride + 32);

    src = src_ptr;
    dst = dst_ptr;

    odd_dst = (dst + dst_stride);

    for (c = 0; c < count; c++) {
      __asm__ __volatile__(
          "ulw %[qload1], 0(%[src]) "
          "\n\t"
          "ulw %[qload2], 4(%[src]) "
          "\n\t"

          /* even 1. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* even 1 */
          "mthi $zero, $ac1 "
          "\n\t"
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* even 2 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload1] "
          "\n\t"
          "preceu.ph.qbl %[p2], %[qload1] "
          "\n\t"
          "preceu.ph.qbr %[p3], %[qload2] "
          "\n\t"
          "preceu.ph.qbl %[p4], %[qload2] "
          "\n\t"
          "ulw %[qload1], 8(%[src]) "
          "\n\t"
          "dpa.w.ph $ac1, %[p1], %[filter45] "
          "\n\t" /* even 1 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* even 1 */

          /* even 2. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* even 3 */
          "mthi $zero, $ac3 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload1] "
          "\n\t"
          "preceu.ph.qbl %[p5], %[qload1] "
          "\n\t"
          "ulw %[qload2], 12(%[src]) "
          "\n\t"
          "dpa.w.ph $ac2, %[p2], %[filter45] "
          "\n\t" /* even 1 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* even 1 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* even 1 */

          /* even 3. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* even 4 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbr %[p2], %[qload2] "
          "\n\t"
          "sb %[st1], 0(%[dst]) "
          "\n\t" /* even 1 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          " \n\t"
          "dpa.w.ph $ac3, %[p3], %[filter45] "
          "\n\t" /* even 3 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* even 3 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* even 1 */

          /* even 4. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* even 5 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbl %[p3], %[qload2] "
          "\n\t"
          "sb %[st2], 0(%[dst]) "
          "\n\t" /* even 2 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac1, %[p4], %[filter45] "
          "\n\t" /* even 4 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* even 4 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* even 3 */

          /* even 5. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* even 6 */
          "mthi $zero, $ac3 "
          "\n\t"
          "sb %[st3], 0(%[dst]) "
          "\n\t" /* even 3 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac2, %[p1], %[filter45] "
          "\n\t" /* even 5 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* even 5 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* even 4 */

          /* even 6. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* even 7 */
          "mthi $zero, $ac1 "
          "\n\t"
          "sb %[st1], 0(%[dst]) "
          "\n\t" /* even 4 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "ulw %[qload1], 20(%[src]) "
          "\n\t"
          "dpa.w.ph $ac3, %[p5], %[filter45] "
          "\n\t" /* even 6 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* even 6 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* even 5 */

          /* even 7. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* even 8 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbr %[p5], %[qload1] "
          "\n\t"
          "sb %[st2], 0(%[dst]) "
          "\n\t" /* even 5 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac1, %[p2], %[filter45] "
          "\n\t" /* even 7 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* even 7 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* even 6 */

          /* even 8. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* odd 1 */
          "mthi $zero, $ac3 "
          "\n\t"
          "dpa.w.ph $ac2, %[p3], %[filter45] "
          "\n\t" /* even 8 */
          "sb %[st3], 0(%[dst]) "
          "\n\t" /* even 6 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* even 8 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* even 7 */

          /* ODD pixels */
          "ulw %[qload1], 1(%[src]) "
          "\n\t"
          "ulw %[qload2], 5(%[src]) "
          "\n\t"

          /* odd 1. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* odd 2 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload1] "
          "\n\t"
          "preceu.ph.qbl %[p2], %[qload1] "
          "\n\t"
          "preceu.ph.qbr %[p3], %[qload2] "
          "\n\t"
          "preceu.ph.qbl %[p4], %[qload2] "
          "\n\t"
          "sb %[st1], 0(%[dst]) "
          "\n\t" /* even 7 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "ulw %[qload2], 9(%[src]) "
          "\n\t"
          "dpa.w.ph $ac3, %[p1], %[filter45] "
          "\n\t" /* odd 1 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* odd 1 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* even 8 */

          /* odd 2. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* odd 3 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload2] "
          "\n\t"
          "preceu.ph.qbl %[p5], %[qload2] "
          "\n\t"
          "sb %[st2], 0(%[dst]) "
          "\n\t" /* even 8 */
          "ulw %[qload1], 13(%[src]) "
          "\n\t"
          "dpa.w.ph $ac1, %[p2], %[filter45] "
          "\n\t" /* odd 2 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* odd 2 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* odd 1 */

          /* odd 3. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* odd 4 */
          "mthi $zero, $ac3 "
          "\n\t"
          "preceu.ph.qbr %[p2], %[qload1] "
          "\n\t"
          "sb %[st3], 0(%[odd_dst]) "
          "\n\t" /* odd 1 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac2, %[p3], %[filter45] "
          "\n\t" /* odd 3 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* odd 3 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* odd 2 */

          /* odd 4. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* odd 5 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbl %[p3], %[qload1] "
          "\n\t"
          "sb %[st1], 0(%[odd_dst]) "
          "\n\t" /* odd 2 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac3, %[p4], %[filter45] "
          "\n\t" /* odd 4 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* odd 4 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* odd 3 */

          /* odd 5. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* odd 6 */
          "mthi $zero, $ac2 "
          "\n\t"
          "sb %[st2], 0(%[odd_dst]) "
          "\n\t" /* odd 3 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac1, %[p1], %[filter45] "
          "\n\t" /* odd 5 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* odd 5 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* odd 4 */

          /* odd 6. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* odd 7 */
          "mthi $zero, $ac3 "
          "\n\t"
          "sb %[st3], 0(%[odd_dst]) "
          "\n\t" /* odd 4 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "ulw %[qload1], 21(%[src]) "
          "\n\t"
          "dpa.w.ph $ac2, %[p5], %[filter45] "
          "\n\t" /* odd 6 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* odd 6 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* odd 5 */

          /* odd 7. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* odd 8 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbr %[p5], %[qload1] "
          "\n\t"
          "sb %[st1], 0(%[odd_dst]) "
          "\n\t" /* odd 5 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac3, %[p2], %[filter45] "
          "\n\t" /* odd 7 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* odd 7 */

          /* odd 8. pixel */
          "dpa.w.ph $ac1, %[p3], %[filter45] "
          "\n\t" /* odd 8 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* odd 8 */

          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* odd 6 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* odd 7 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* odd 8 */

          "sb %[st2], 0(%[odd_dst]) "
          "\n\t" /* odd 6 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"

          "sb %[st3], 0(%[odd_dst]) "
          "\n\t" /* odd 7 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"

          "sb %[st1], 0(%[odd_dst]) "
          "\n\t" /* odd 8 */

          : [qload1] "=&r"(qload1), [qload2] "=&r"(qload2), [p5] "=&r"(p5),
            [st1] "=&r"(st1), [st2] "=&r"(st2), [st3] "=&r"(st3),
            [p1] "=&r"(p1), [p2] "=&r"(p2), [p3] "=&r"(p3), [p4] "=&r"(p4),
            [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2), [Temp3] "=&r"(Temp3),
            [dst] "+r"(dst), [odd_dst] "+r"(odd_dst)
          : [filter45] "r"(filter45), [vector_64] "r"(vector_64), [cm] "r"(cm),
            [src] "r"(src), [dst_pitch_2] "r"(dst_pitch_2));

      src += 16;
      dst = (dst_ptr + ((c + 1) * 16 * dst_stride));
      odd_dst = (dst + dst_stride);
    }

    /* Next row... */
    src_ptr += src_stride;
    dst_ptr += 1;
  }
}

static void convolve_bi_horiz_64_transposed_dspr2(
    const uint8_t *src_ptr, int32_t src_stride, uint8_t *dst_ptr,
    int32_t dst_stride, const int16_t *filter_x0, int32_t h) {
  int32_t c, y;
  const uint8_t *src;
  uint8_t *dst;
  uint8_t *cm = vpx_ff_cropTbl;
  uint32_t vector_64 = 64;
  int32_t Temp1, Temp2, Temp3;
  uint32_t qload1, qload2;
  uint32_t p1, p2, p3, p4, p5;
  uint32_t st1, st2, st3;
  uint32_t dst_pitch_2 = (dst_stride << 1);
  uint8_t *odd_dst;
  const int16_t *filter = &filter_x0[3];
  uint32_t filter45;

  filter45 = ((const int32_t *)filter)[0];

  for (y = h; y--;) {
    /* prefetch data to cache memory */
    prefetch_load(src_ptr + src_stride);
    prefetch_load(src_ptr + src_stride + 32);
    prefetch_load(src_ptr + src_stride + 64);

    src = src_ptr;
    dst = dst_ptr;

    odd_dst = (dst + dst_stride);

    for (c = 0; c < 4; c++) {
      __asm__ __volatile__(
          "ulw %[qload1], 0(%[src]) "
          "\n\t"
          "ulw %[qload2], 4(%[src]) "
          "\n\t"

          /* even 1. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* even 1 */
          "mthi $zero, $ac1 "
          "\n\t"
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* even 2 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload1] "
          "\n\t"
          "preceu.ph.qbl %[p2], %[qload1] "
          "\n\t"
          "preceu.ph.qbr %[p3], %[qload2] "
          "\n\t"
          "preceu.ph.qbl %[p4], %[qload2] "
          "\n\t"
          "ulw %[qload1], 8(%[src]) "
          "\n\t"
          "dpa.w.ph $ac1, %[p1], %[filter45] "
          "\n\t" /* even 1 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* even 1 */

          /* even 2. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* even 3 */
          "mthi $zero, $ac3 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload1] "
          "\n\t"
          "preceu.ph.qbl %[p5], %[qload1] "
          "\n\t"
          "ulw %[qload2], 12(%[src]) "
          "\n\t"
          "dpa.w.ph $ac2, %[p2], %[filter45] "
          "\n\t" /* even 1 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* even 1 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* even 1 */

          /* even 3. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* even 4 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbr %[p2], %[qload2] "
          "\n\t"
          "sb %[st1], 0(%[dst]) "
          "\n\t" /* even 1 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          " \n\t"
          "dpa.w.ph $ac3, %[p3], %[filter45] "
          "\n\t" /* even 3 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* even 3 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* even 1 */

          /* even 4. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* even 5 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbl %[p3], %[qload2] "
          "\n\t"
          "sb %[st2], 0(%[dst]) "
          "\n\t" /* even 2 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac1, %[p4], %[filter45] "
          "\n\t" /* even 4 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* even 4 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* even 3 */

          /* even 5. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* even 6 */
          "mthi $zero, $ac3 "
          "\n\t"
          "sb %[st3], 0(%[dst]) "
          "\n\t" /* even 3 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac2, %[p1], %[filter45] "
          "\n\t" /* even 5 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* even 5 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* even 4 */

          /* even 6. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* even 7 */
          "mthi $zero, $ac1 "
          "\n\t"
          "sb %[st1], 0(%[dst]) "
          "\n\t" /* even 4 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "ulw %[qload1], 20(%[src]) "
          "\n\t"
          "dpa.w.ph $ac3, %[p5], %[filter45] "
          "\n\t" /* even 6 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* even 6 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* even 5 */

          /* even 7. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* even 8 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbr %[p5], %[qload1] "
          "\n\t"
          "sb %[st2], 0(%[dst]) "
          "\n\t" /* even 5 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac1, %[p2], %[filter45] "
          "\n\t" /* even 7 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* even 7 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* even 6 */

          /* even 8. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* odd 1 */
          "mthi $zero, $ac3 "
          "\n\t"
          "dpa.w.ph $ac2, %[p3], %[filter45] "
          "\n\t" /* even 8 */
          "sb %[st3], 0(%[dst]) "
          "\n\t" /* even 6 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* even 8 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* even 7 */

          /* ODD pixels */
          "ulw %[qload1], 1(%[src]) "
          "\n\t"
          "ulw %[qload2], 5(%[src]) "
          "\n\t"

          /* odd 1. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* odd 2 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload1] "
          "\n\t"
          "preceu.ph.qbl %[p2], %[qload1] "
          "\n\t"
          "preceu.ph.qbr %[p3], %[qload2] "
          "\n\t"
          "preceu.ph.qbl %[p4], %[qload2] "
          "\n\t"
          "sb %[st1], 0(%[dst]) "
          "\n\t" /* even 7 */
          "addu %[dst], %[dst], %[dst_pitch_2] "
          "\n\t"
          "ulw %[qload2], 9(%[src]) "
          "\n\t"
          "dpa.w.ph $ac3, %[p1], %[filter45] "
          "\n\t" /* odd 1 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* odd 1 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* even 8 */

          /* odd 2. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* odd 3 */
          "mthi $zero, $ac2 "
          "\n\t"
          "preceu.ph.qbr %[p1], %[qload2] "
          "\n\t"
          "preceu.ph.qbl %[p5], %[qload2] "
          "\n\t"
          "sb %[st2], 0(%[dst]) "
          "\n\t" /* even 8 */
          "ulw %[qload1], 13(%[src]) "
          "\n\t"
          "dpa.w.ph $ac1, %[p2], %[filter45] "
          "\n\t" /* odd 2 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* odd 2 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* odd 1 */

          /* odd 3. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* odd 4 */
          "mthi $zero, $ac3 "
          "\n\t"
          "preceu.ph.qbr %[p2], %[qload1] "
          "\n\t"
          "sb %[st3], 0(%[odd_dst]) "
          "\n\t" /* odd 1 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac2, %[p3], %[filter45] "
          "\n\t" /* odd 3 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* odd 3 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* odd 2 */

          /* odd 4. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* odd 5 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbl %[p3], %[qload1] "
          "\n\t"
          "sb %[st1], 0(%[odd_dst]) "
          "\n\t" /* odd 2 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac3, %[p4], %[filter45] "
          "\n\t" /* odd 4 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* odd 4 */
          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* odd 3 */

          /* odd 5. pixel */
          "mtlo %[vector_64], $ac2 "
          "\n\t" /* odd 6 */
          "mthi $zero, $ac2 "
          "\n\t"
          "sb %[st2], 0(%[odd_dst]) "
          "\n\t" /* odd 3 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac1, %[p1], %[filter45] "
          "\n\t" /* odd 5 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* odd 5 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* odd 4 */

          /* odd 6. pixel */
          "mtlo %[vector_64], $ac3 "
          "\n\t" /* odd 7 */
          "mthi $zero, $ac3 "
          "\n\t"
          "sb %[st3], 0(%[odd_dst]) "
          "\n\t" /* odd 4 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "ulw %[qload1], 21(%[src]) "
          "\n\t"
          "dpa.w.ph $ac2, %[p5], %[filter45] "
          "\n\t" /* odd 6 */
          "extp %[Temp2], $ac2, 31 "
          "\n\t" /* odd 6 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* odd 5 */

          /* odd 7. pixel */
          "mtlo %[vector_64], $ac1 "
          "\n\t" /* odd 8 */
          "mthi $zero, $ac1 "
          "\n\t"
          "preceu.ph.qbr %[p5], %[qload1] "
          "\n\t"
          "sb %[st1], 0(%[odd_dst]) "
          "\n\t" /* odd 5 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"
          "dpa.w.ph $ac3, %[p2], %[filter45] "
          "\n\t" /* odd 7 */
          "extp %[Temp3], $ac3, 31 "
          "\n\t" /* odd 7 */

          /* odd 8. pixel */
          "dpa.w.ph $ac1, %[p3], %[filter45] "
          "\n\t" /* odd 8 */
          "extp %[Temp1], $ac1, 31 "
          "\n\t" /* odd 8 */

          "lbux %[st2], %[Temp2](%[cm]) "
          "\n\t" /* odd 6 */
          "lbux %[st3], %[Temp3](%[cm]) "
          "\n\t" /* odd 7 */
          "lbux %[st1], %[Temp1](%[cm]) "
          "\n\t" /* odd 8 */

          "sb %[st2], 0(%[odd_dst]) "
          "\n\t" /* odd 6 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"

          "sb %[st3], 0(%[odd_dst]) "
          "\n\t" /* odd 7 */
          "addu %[odd_dst], %[odd_dst], %[dst_pitch_2] "
          "\n\t"

          "sb %[st1], 0(%[odd_dst]) "
          "\n\t" /* odd 8 */

          : [qload1] "=&r"(qload1), [qload2] "=&r"(qload2), [p5] "=&r"(p5),
            [st1] "=&r"(st1), [st2] "=&r"(st2), [st3] "=&r"(st3),
            [p1] "=&r"(p1), [p2] "=&r"(p2), [p3] "=&r"(p3), [p4] "=&r"(p4),
            [Temp1] "=&r"(Temp1), [Temp2] "=&r"(Temp2), [Temp3] "=&r"(Temp3),
            [dst] "+r"(dst), [odd_dst] "+r"(odd_dst)
          : [filter45] "r"(filter45), [vector_64] "r"(vector_64), [cm] "r"(cm),
            [src] "r"(src), [dst_pitch_2] "r"(dst_pitch_2));

      src += 16;
      dst = (dst_ptr + ((c + 1) * 16 * dst_stride));
      odd_dst = (dst + dst_stride);
    }

    /* Next row... */
    src_ptr += src_stride;
    dst_ptr += 1;
  }
}

void convolve_bi_horiz_transposed(const uint8_t *src, ptrdiff_t src_stride,
                                  uint8_t *dst, ptrdiff_t dst_stride,
                                  const int16_t *filter, int w, int h) {
  int x, y;

  for (y = 0; y < h; ++y) {
    for (x = 0; x < w; ++x) {
      int sum = 0;

      sum += src[x] * filter[3];
      sum += src[x + 1] * filter[4];

      dst[x * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
    }

    src += src_stride;
    dst += 1;
  }
}

void vpx_convolve2_dspr2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                         ptrdiff_t dst_stride, const int16_t *filter, int w,
                         int h) {
  uint32_t pos = 38;

  /* bit positon for extract from acc */
  __asm__ __volatile__("wrdsp %[pos], 1 \n\t"
                       :
                       : [pos] "r"(pos));

  /* prefetch data to cache memory */
  prefetch_load(src);
  prefetch_load(src + 32);

  switch (w) {
    case 4:
      convolve_bi_horiz_4_transposed_dspr2(src, src_stride, dst, dst_stride,
                                           filter, h);
      break;
    case 8:
      convolve_bi_horiz_8_transposed_dspr2(src, src_stride, dst, dst_stride,
                                           filter, h);
      break;
    case 16:
    case 32:
      convolve_bi_horiz_16_transposed_dspr2(src, src_stride, dst, dst_stride,
                                            filter, h, (w / 16));
      break;
    case 64:
      prefetch_load(src + 32);
      convolve_bi_horiz_64_transposed_dspr2(src, src_stride, dst, dst_stride,
                                            filter, h);
      break;
    default:
      convolve_bi_horiz_transposed(src, src_stride, dst, dst_stride, filter, w,
                                   h);
      break;
  }
}
#endif