aom_convolve_avg_neon.c 4.18 KB
Newer Older
Zoe Liu's avatar
Zoe Liu committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Zoe Liu's avatar
Zoe Liu committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
4 5 6 7 8 9
 * 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.
Zoe Liu's avatar
Zoe Liu committed
10 11 12 13
 */

#include <arm_neon.h>

Yaowu Xu's avatar
Yaowu Xu committed
14 15
#include "./aom_dsp_rtcd.h"
#include "aom/aom_integer.h"
Zoe Liu's avatar
Zoe Liu committed
16

Yaowu Xu's avatar
Yaowu Xu committed
17
void aom_convolve_avg_neon(const uint8_t *src,    // r0
clang-format's avatar
clang-format committed
18 19 20 21 22 23
                           ptrdiff_t src_stride,  // r1
                           uint8_t *dst,          // r2
                           ptrdiff_t dst_stride,  // r3
                           const int16_t *filter_x, int filter_x_stride,
                           const int16_t *filter_y, int filter_y_stride, int w,
                           int h) {
Zoe Liu's avatar
Zoe Liu committed
24 25 26 27
  uint8_t *d;
  uint8x8_t d0u8, d1u8, d2u8, d3u8;
  uint32x2_t d0u32, d2u32;
  uint8x16_t q0u8, q1u8, q2u8, q3u8, q8u8, q9u8, q10u8, q11u8;
clang-format's avatar
clang-format committed
28 29 30 31
  (void)filter_x;
  (void)filter_x_stride;
  (void)filter_y;
  (void)filter_y_stride;
Zoe Liu's avatar
Zoe Liu committed
32 33 34 35

  d = dst;
  if (w > 32) {  // avg64
    for (; h > 0; h -= 1) {
clang-format's avatar
clang-format committed
36 37 38 39
      q0u8 = vld1q_u8(src);
      q1u8 = vld1q_u8(src + 16);
      q2u8 = vld1q_u8(src + 32);
      q3u8 = vld1q_u8(src + 48);
Zoe Liu's avatar
Zoe Liu committed
40
      src += src_stride;
clang-format's avatar
clang-format committed
41 42
      q8u8 = vld1q_u8(d);
      q9u8 = vld1q_u8(d + 16);
Zoe Liu's avatar
Zoe Liu committed
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134
      q10u8 = vld1q_u8(d + 32);
      q11u8 = vld1q_u8(d + 48);
      d += dst_stride;

      q0u8 = vrhaddq_u8(q0u8, q8u8);
      q1u8 = vrhaddq_u8(q1u8, q9u8);
      q2u8 = vrhaddq_u8(q2u8, q10u8);
      q3u8 = vrhaddq_u8(q3u8, q11u8);

      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q1u8);
      vst1q_u8(dst + 32, q2u8);
      vst1q_u8(dst + 48, q3u8);
      dst += dst_stride;
    }
  } else if (w == 32) {  // avg32
    for (; h > 0; h -= 2) {
      q0u8 = vld1q_u8(src);
      q1u8 = vld1q_u8(src + 16);
      src += src_stride;
      q2u8 = vld1q_u8(src);
      q3u8 = vld1q_u8(src + 16);
      src += src_stride;
      q8u8 = vld1q_u8(d);
      q9u8 = vld1q_u8(d + 16);
      d += dst_stride;
      q10u8 = vld1q_u8(d);
      q11u8 = vld1q_u8(d + 16);
      d += dst_stride;

      q0u8 = vrhaddq_u8(q0u8, q8u8);
      q1u8 = vrhaddq_u8(q1u8, q9u8);
      q2u8 = vrhaddq_u8(q2u8, q10u8);
      q3u8 = vrhaddq_u8(q3u8, q11u8);

      vst1q_u8(dst, q0u8);
      vst1q_u8(dst + 16, q1u8);
      dst += dst_stride;
      vst1q_u8(dst, q2u8);
      vst1q_u8(dst + 16, q3u8);
      dst += dst_stride;
    }
  } else if (w > 8) {  // avg16
    for (; h > 0; h -= 2) {
      q0u8 = vld1q_u8(src);
      src += src_stride;
      q1u8 = vld1q_u8(src);
      src += src_stride;
      q2u8 = vld1q_u8(d);
      d += dst_stride;
      q3u8 = vld1q_u8(d);
      d += dst_stride;

      q0u8 = vrhaddq_u8(q0u8, q2u8);
      q1u8 = vrhaddq_u8(q1u8, q3u8);

      vst1q_u8(dst, q0u8);
      dst += dst_stride;
      vst1q_u8(dst, q1u8);
      dst += dst_stride;
    }
  } else if (w == 8) {  // avg8
    for (; h > 0; h -= 2) {
      d0u8 = vld1_u8(src);
      src += src_stride;
      d1u8 = vld1_u8(src);
      src += src_stride;
      d2u8 = vld1_u8(d);
      d += dst_stride;
      d3u8 = vld1_u8(d);
      d += dst_stride;

      q0u8 = vcombine_u8(d0u8, d1u8);
      q1u8 = vcombine_u8(d2u8, d3u8);
      q0u8 = vrhaddq_u8(q0u8, q1u8);

      vst1_u8(dst, vget_low_u8(q0u8));
      dst += dst_stride;
      vst1_u8(dst, vget_high_u8(q0u8));
      dst += dst_stride;
    }
  } else {  // avg4
    for (; h > 0; h -= 2) {
      d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 0);
      src += src_stride;
      d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 1);
      src += src_stride;
      d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 0);
      d += dst_stride;
      d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 1);
      d += dst_stride;

clang-format's avatar
clang-format committed
135
      d0u8 = vrhadd_u8(vreinterpret_u8_u32(d0u32), vreinterpret_u8_u32(d2u32));
Zoe Liu's avatar
Zoe Liu committed
136 137 138 139 140 141 142 143 144 145

      d0u32 = vreinterpret_u32_u8(d0u8);
      vst1_lane_u32((uint32_t *)dst, d0u32, 0);
      dst += dst_stride;
      vst1_lane_u32((uint32_t *)dst, d0u32, 1);
      dst += dst_stride;
    }
  }
  return;
}