1/*
2 *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
3 *
4 *  Use of this source code is governed by a BSD-style license
5 *  that can be found in the LICENSE file in the root of the source
6 *  tree. An additional intellectual property rights grant can be found
7 *  in the file PATENTS.  All contributing project authors may
8 *  be found in the AUTHORS file in the root of the source tree.
9 */
10
11#include <arm_neon.h>
12#include "./vp9_rtcd.h"
13#include "./vpx_config.h"
14
15#include "vpx/vpx_integer.h"
16
17void vp9_subtract_block_neon(int rows, int cols,
18                             int16_t *diff, ptrdiff_t diff_stride,
19                             const uint8_t *src, ptrdiff_t src_stride,
20                             const uint8_t *pred, ptrdiff_t pred_stride) {
21  int r, c;
22
23  if (cols > 16) {
24    for (r = 0; r < rows; ++r) {
25      for (c = 0; c < cols; c += 32) {
26        const uint8x16_t v_src_00 = vld1q_u8(&src[c + 0]);
27        const uint8x16_t v_src_16 = vld1q_u8(&src[c + 16]);
28        const uint8x16_t v_pred_00 = vld1q_u8(&pred[c +  0]);
29        const uint8x16_t v_pred_16 = vld1q_u8(&pred[c + 16]);
30        const uint16x8_t v_diff_lo_00 = vsubl_u8(vget_low_u8(v_src_00),
31                                                 vget_low_u8(v_pred_00));
32        const uint16x8_t v_diff_hi_00 = vsubl_u8(vget_high_u8(v_src_00),
33                                                 vget_high_u8(v_pred_00));
34        const uint16x8_t v_diff_lo_16 = vsubl_u8(vget_low_u8(v_src_16),
35                                                 vget_low_u8(v_pred_16));
36        const uint16x8_t v_diff_hi_16 = vsubl_u8(vget_high_u8(v_src_16),
37                                                 vget_high_u8(v_pred_16));
38        vst1q_s16(&diff[c +  0], vreinterpretq_s16_u16(v_diff_lo_00));
39        vst1q_s16(&diff[c +  8], vreinterpretq_s16_u16(v_diff_hi_00));
40        vst1q_s16(&diff[c + 16], vreinterpretq_s16_u16(v_diff_lo_16));
41        vst1q_s16(&diff[c + 24], vreinterpretq_s16_u16(v_diff_hi_16));
42      }
43      diff += diff_stride;
44      pred += pred_stride;
45      src  += src_stride;
46    }
47  } else if (cols > 8) {
48    for (r = 0; r < rows; ++r) {
49      const uint8x16_t v_src = vld1q_u8(&src[0]);
50      const uint8x16_t v_pred = vld1q_u8(&pred[0]);
51      const uint16x8_t v_diff_lo = vsubl_u8(vget_low_u8(v_src),
52                                            vget_low_u8(v_pred));
53      const uint16x8_t v_diff_hi = vsubl_u8(vget_high_u8(v_src),
54                                            vget_high_u8(v_pred));
55      vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff_lo));
56      vst1q_s16(&diff[8], vreinterpretq_s16_u16(v_diff_hi));
57      diff += diff_stride;
58      pred += pred_stride;
59      src  += src_stride;
60    }
61  } else if (cols > 4) {
62    for (r = 0; r < rows; ++r) {
63      const uint8x8_t v_src = vld1_u8(&src[0]);
64      const uint8x8_t v_pred = vld1_u8(&pred[0]);
65      const uint16x8_t v_diff = vsubl_u8(v_src, v_pred);
66      vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff));
67      diff += diff_stride;
68      pred += pred_stride;
69      src  += src_stride;
70    }
71  } else {
72    for (r = 0; r < rows; ++r) {
73      for (c = 0; c < cols; ++c)
74        diff[c] = src[c] - pred[c];
75
76      diff += diff_stride;
77      pred += pred_stride;
78      src  += src_stride;
79    }
80  }
81}
82