subtract_neon.c revision 7ce0a1d1337c01056ba24006efab21f00e179e04 
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
13#include "./vpx_config.h"
14#include "vpx/vpx_integer.h"
15
16void vpx_subtract_block_neon(int rows, int cols,
17                             int16_t *diff, ptrdiff_t diff_stride,
18                             const uint8_t *src, ptrdiff_t src_stride,
19                             const uint8_t *pred, ptrdiff_t pred_stride) {
20  int r, c;
21
22  if (cols > 16) {
23    for (r = 0; r < rows; ++r) {
24      for (c = 0; c < cols; c += 32) {
25        const uint8x16_t v_src_00 = vld1q_u8(&src[c + 0]);
26        const uint8x16_t v_src_16 = vld1q_u8(&src[c + 16]);
27        const uint8x16_t v_pred_00 = vld1q_u8(&pred[c +  0]);
28        const uint8x16_t v_pred_16 = vld1q_u8(&pred[c + 16]);
29        const uint16x8_t v_diff_lo_00 = vsubl_u8(vget_low_u8(v_src_00),
30                                                 vget_low_u8(v_pred_00));
31        const uint16x8_t v_diff_hi_00 = vsubl_u8(vget_high_u8(v_src_00),
32                                                 vget_high_u8(v_pred_00));
33        const uint16x8_t v_diff_lo_16 = vsubl_u8(vget_low_u8(v_src_16),
34                                                 vget_low_u8(v_pred_16));
35        const uint16x8_t v_diff_hi_16 = vsubl_u8(vget_high_u8(v_src_16),
36                                                 vget_high_u8(v_pred_16));
37        vst1q_s16(&diff[c +  0], vreinterpretq_s16_u16(v_diff_lo_00));
38        vst1q_s16(&diff[c +  8], vreinterpretq_s16_u16(v_diff_hi_00));
39        vst1q_s16(&diff[c + 16], vreinterpretq_s16_u16(v_diff_lo_16));
40        vst1q_s16(&diff[c + 24], vreinterpretq_s16_u16(v_diff_hi_16));
41      }
42      diff += diff_stride;
43      pred += pred_stride;
44      src  += src_stride;
45    }
46  } else if (cols > 8) {
47    for (r = 0; r < rows; ++r) {
48      const uint8x16_t v_src = vld1q_u8(&src[0]);
49      const uint8x16_t v_pred = vld1q_u8(&pred[0]);
50      const uint16x8_t v_diff_lo = vsubl_u8(vget_low_u8(v_src),
51                                            vget_low_u8(v_pred));
52      const uint16x8_t v_diff_hi = vsubl_u8(vget_high_u8(v_src),
53                                            vget_high_u8(v_pred));
54      vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff_lo));
55      vst1q_s16(&diff[8], vreinterpretq_s16_u16(v_diff_hi));
56      diff += diff_stride;
57      pred += pred_stride;
58      src  += src_stride;
59    }
60  } else if (cols > 4) {
61    for (r = 0; r < rows; ++r) {
62      const uint8x8_t v_src = vld1_u8(&src[0]);
63      const uint8x8_t v_pred = vld1_u8(&pred[0]);
64      const uint16x8_t v_diff = vsubl_u8(v_src, v_pred);
65      vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff));
66      diff += diff_stride;
67      pred += pred_stride;
68      src  += src_stride;
69    }
70  } else {
71    for (r = 0; r < rows; ++r) {
72      for (c = 0; c < cols; ++c)
73        diff[c] = src[c] - pred[c];
74
75      diff += diff_stride;
76      pred += pred_stride;
77      src  += src_stride;
78    }
79  }
80}
81