1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include <arm_neon.h>
13
14 #include "config/aom_config.h"
15
16 #include "aom/aom_integer.h"
17 #include "aom_ports/mem.h"
18
aom_subtract_block_neon(int rows,int cols,int16_t * diff,ptrdiff_t diff_stride,const uint8_t * src,ptrdiff_t src_stride,const uint8_t * pred,ptrdiff_t pred_stride)19 void aom_subtract_block_neon(int rows, int cols, int16_t *diff,
20 ptrdiff_t diff_stride, const uint8_t *src,
21 ptrdiff_t src_stride, const uint8_t *pred,
22 ptrdiff_t pred_stride) {
23 if (cols > 16) {
24 int r = rows;
25 do {
26 int c = 0;
27 do {
28 const uint8x16_t v_src_00 = vld1q_u8(&src[c + 0]);
29 const uint8x16_t v_src_16 = vld1q_u8(&src[c + 16]);
30 const uint8x16_t v_pred_00 = vld1q_u8(&pred[c + 0]);
31 const uint8x16_t v_pred_16 = vld1q_u8(&pred[c + 16]);
32 const uint16x8_t v_diff_lo_00 =
33 vsubl_u8(vget_low_u8(v_src_00), vget_low_u8(v_pred_00));
34 const uint16x8_t v_diff_hi_00 =
35 vsubl_u8(vget_high_u8(v_src_00), vget_high_u8(v_pred_00));
36 const uint16x8_t v_diff_lo_16 =
37 vsubl_u8(vget_low_u8(v_src_16), vget_low_u8(v_pred_16));
38 const uint16x8_t v_diff_hi_16 =
39 vsubl_u8(vget_high_u8(v_src_16), vget_high_u8(v_pred_16));
40 vst1q_s16(&diff[c + 0], vreinterpretq_s16_u16(v_diff_lo_00));
41 vst1q_s16(&diff[c + 8], vreinterpretq_s16_u16(v_diff_hi_00));
42 vst1q_s16(&diff[c + 16], vreinterpretq_s16_u16(v_diff_lo_16));
43 vst1q_s16(&diff[c + 24], vreinterpretq_s16_u16(v_diff_hi_16));
44 c += 32;
45 } while (c < cols);
46 diff += diff_stride;
47 pred += pred_stride;
48 src += src_stride;
49 } while (--r != 0);
50 } else if (cols > 8) {
51 int r = rows;
52 do {
53 const uint8x16_t v_src = vld1q_u8(&src[0]);
54 const uint8x16_t v_pred = vld1q_u8(&pred[0]);
55 const uint16x8_t v_diff_lo =
56 vsubl_u8(vget_low_u8(v_src), vget_low_u8(v_pred));
57 const uint16x8_t v_diff_hi =
58 vsubl_u8(vget_high_u8(v_src), vget_high_u8(v_pred));
59 vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff_lo));
60 vst1q_s16(&diff[8], vreinterpretq_s16_u16(v_diff_hi));
61 diff += diff_stride;
62 pred += pred_stride;
63 src += src_stride;
64 } while (--r != 0);
65 } else if (cols > 4) {
66 int r = rows;
67 do {
68 const uint8x8_t v_src = vld1_u8(&src[0]);
69 const uint8x8_t v_pred = vld1_u8(&pred[0]);
70 const uint16x8_t v_diff = vsubl_u8(v_src, v_pred);
71 vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff));
72 diff += diff_stride;
73 pred += pred_stride;
74 src += src_stride;
75 } while (--r != 0);
76 } else {
77 int r = rows;
78 do {
79 int c = 0;
80 do {
81 diff[c] = src[c] - pred[c];
82 } while (++c < cols);
83 diff += diff_stride;
84 pred += pred_stride;
85 src += src_stride;
86 } while (--r != 0);
87 }
88 }
89
90 #if CONFIG_AV1_HIGHBITDEPTH
aom_highbd_subtract_block_neon(int rows,int cols,int16_t * diff,ptrdiff_t diff_stride,const uint8_t * src8,ptrdiff_t src_stride,const uint8_t * pred8,ptrdiff_t pred_stride)91 void aom_highbd_subtract_block_neon(int rows, int cols, int16_t *diff,
92 ptrdiff_t diff_stride, const uint8_t *src8,
93 ptrdiff_t src_stride, const uint8_t *pred8,
94 ptrdiff_t pred_stride) {
95 uint16_t *src = CONVERT_TO_SHORTPTR(src8);
96 uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
97
98 if (cols > 16) {
99 int r = rows;
100 do {
101 int c = 0;
102 do {
103 const uint16x8_t v_src_00 = vld1q_u16(&src[c + 0]);
104 const uint16x8_t v_pred_00 = vld1q_u16(&pred[c + 0]);
105 const uint16x8_t v_diff_00 = vsubq_u16(v_src_00, v_pred_00);
106 const uint16x8_t v_src_08 = vld1q_u16(&src[c + 8]);
107 const uint16x8_t v_pred_08 = vld1q_u16(&pred[c + 8]);
108 const uint16x8_t v_diff_08 = vsubq_u16(v_src_08, v_pred_08);
109 vst1q_s16(&diff[c + 0], vreinterpretq_s16_u16(v_diff_00));
110 vst1q_s16(&diff[c + 8], vreinterpretq_s16_u16(v_diff_08));
111 c += 16;
112 } while (c < cols);
113 diff += diff_stride;
114 pred += pred_stride;
115 src += src_stride;
116 } while (--r != 0);
117 } else if (cols > 8) {
118 int r = rows;
119 do {
120 const uint16x8_t v_src_00 = vld1q_u16(&src[0]);
121 const uint16x8_t v_pred_00 = vld1q_u16(&pred[0]);
122 const uint16x8_t v_diff_00 = vsubq_u16(v_src_00, v_pred_00);
123 const uint16x8_t v_src_08 = vld1q_u16(&src[8]);
124 const uint16x8_t v_pred_08 = vld1q_u16(&pred[8]);
125 const uint16x8_t v_diff_08 = vsubq_u16(v_src_08, v_pred_08);
126 vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff_00));
127 vst1q_s16(&diff[8], vreinterpretq_s16_u16(v_diff_08));
128 diff += diff_stride;
129 pred += pred_stride;
130 src += src_stride;
131 } while (--r != 0);
132 } else if (cols > 4) {
133 int r = rows;
134 do {
135 const uint16x8_t v_src_r0 = vld1q_u16(&src[0]);
136 const uint16x8_t v_src_r1 = vld1q_u16(&src[src_stride]);
137 const uint16x8_t v_pred_r0 = vld1q_u16(&pred[0]);
138 const uint16x8_t v_pred_r1 = vld1q_u16(&pred[pred_stride]);
139 const uint16x8_t v_diff_r0 = vsubq_u16(v_src_r0, v_pred_r0);
140 const uint16x8_t v_diff_r1 = vsubq_u16(v_src_r1, v_pred_r1);
141 vst1q_s16(&diff[0], vreinterpretq_s16_u16(v_diff_r0));
142 vst1q_s16(&diff[diff_stride], vreinterpretq_s16_u16(v_diff_r1));
143 diff += diff_stride << 1;
144 pred += pred_stride << 1;
145 src += src_stride << 1;
146 r -= 2;
147 } while (r != 0);
148 } else {
149 int r = rows;
150 do {
151 const uint16x4_t v_src_r0 = vld1_u16(&src[0]);
152 const uint16x4_t v_src_r1 = vld1_u16(&src[src_stride]);
153 const uint16x4_t v_pred_r0 = vld1_u16(&pred[0]);
154 const uint16x4_t v_pred_r1 = vld1_u16(&pred[pred_stride]);
155 const uint16x4_t v_diff_r0 = vsub_u16(v_src_r0, v_pred_r0);
156 const uint16x4_t v_diff_r1 = vsub_u16(v_src_r1, v_pred_r1);
157 vst1_s16(&diff[0], vreinterpret_s16_u16(v_diff_r0));
158 vst1_s16(&diff[diff_stride], vreinterpret_s16_u16(v_diff_r1));
159 diff += diff_stride << 1;
160 pred += pred_stride << 1;
161 src += src_stride << 1;
162 r -= 2;
163 } while (r != 0);
164 }
165 }
166 #endif // CONFIG_AV1_HIGHBITDEPTH
167