1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
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
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
|
/*
* Copyright (c) 2023, Alliance for Open Media. 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 <arm_neon.h>
#include "config/aom_dsp_rtcd.h"
#include "aom_dsp/arm/mem_neon.h"
#include "aom_dsp/arm/sum_neon.h"
static INLINE void sse_16x1_neon_dotprod(const uint8_t *src, const uint8_t *ref,
uint32x4_t *sse) {
uint8x16_t s = vld1q_u8(src);
uint8x16_t r = vld1q_u8(ref);
uint8x16_t abs_diff = vabdq_u8(s, r);
*sse = vdotq_u32(*sse, abs_diff, abs_diff);
}
static INLINE void sse_8x1_neon_dotprod(const uint8_t *src, const uint8_t *ref,
uint32x2_t *sse) {
uint8x8_t s = vld1_u8(src);
uint8x8_t r = vld1_u8(ref);
uint8x8_t abs_diff = vabd_u8(s, r);
*sse = vdot_u32(*sse, abs_diff, abs_diff);
}
static INLINE void sse_4x2_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
uint32x2_t *sse) {
uint8x8_t s = load_unaligned_u8(src, src_stride);
uint8x8_t r = load_unaligned_u8(ref, ref_stride);
uint8x8_t abs_diff = vabd_u8(s, r);
*sse = vdot_u32(*sse, abs_diff, abs_diff);
}
static INLINE uint32_t sse_wxh_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
int width, int height) {
uint32x2_t sse[2] = { vdup_n_u32(0), vdup_n_u32(0) };
if ((width & 0x07) && ((width & 0x07) < 5)) {
int i = height;
do {
int j = 0;
do {
sse_8x1_neon_dotprod(src + j, ref + j, &sse[0]);
sse_8x1_neon_dotprod(src + j + src_stride, ref + j + ref_stride,
&sse[1]);
j += 8;
} while (j + 4 < width);
sse_4x2_neon_dotprod(src + j, src_stride, ref + j, ref_stride, &sse[0]);
src += 2 * src_stride;
ref += 2 * ref_stride;
i -= 2;
} while (i != 0);
} else {
int i = height;
do {
int j = 0;
do {
sse_8x1_neon_dotprod(src + j, ref + j, &sse[0]);
sse_8x1_neon_dotprod(src + j + src_stride, ref + j + ref_stride,
&sse[1]);
j += 8;
} while (j < width);
src += 2 * src_stride;
ref += 2 * ref_stride;
i -= 2;
} while (i != 0);
}
return horizontal_add_u32x4(vcombine_u32(sse[0], sse[1]));
}
static INLINE uint32_t sse_128xh_neon_dotprod(const uint8_t *src,
int src_stride,
const uint8_t *ref,
int ref_stride, int height) {
uint32x4_t sse[2] = { vdupq_n_u32(0), vdupq_n_u32(0) };
int i = height;
do {
sse_16x1_neon_dotprod(src, ref, &sse[0]);
sse_16x1_neon_dotprod(src + 16, ref + 16, &sse[1]);
sse_16x1_neon_dotprod(src + 32, ref + 32, &sse[0]);
sse_16x1_neon_dotprod(src + 48, ref + 48, &sse[1]);
sse_16x1_neon_dotprod(src + 64, ref + 64, &sse[0]);
sse_16x1_neon_dotprod(src + 80, ref + 80, &sse[1]);
sse_16x1_neon_dotprod(src + 96, ref + 96, &sse[0]);
sse_16x1_neon_dotprod(src + 112, ref + 112, &sse[1]);
src += src_stride;
ref += ref_stride;
} while (--i != 0);
return horizontal_add_u32x4(vaddq_u32(sse[0], sse[1]));
}
static INLINE uint32_t sse_64xh_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
int height) {
uint32x4_t sse[2] = { vdupq_n_u32(0), vdupq_n_u32(0) };
int i = height;
do {
sse_16x1_neon_dotprod(src, ref, &sse[0]);
sse_16x1_neon_dotprod(src + 16, ref + 16, &sse[1]);
sse_16x1_neon_dotprod(src + 32, ref + 32, &sse[0]);
sse_16x1_neon_dotprod(src + 48, ref + 48, &sse[1]);
src += src_stride;
ref += ref_stride;
} while (--i != 0);
return horizontal_add_u32x4(vaddq_u32(sse[0], sse[1]));
}
static INLINE uint32_t sse_32xh_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
int height) {
uint32x4_t sse[2] = { vdupq_n_u32(0), vdupq_n_u32(0) };
int i = height;
do {
sse_16x1_neon_dotprod(src, ref, &sse[0]);
sse_16x1_neon_dotprod(src + 16, ref + 16, &sse[1]);
src += src_stride;
ref += ref_stride;
} while (--i != 0);
return horizontal_add_u32x4(vaddq_u32(sse[0], sse[1]));
}
static INLINE uint32_t sse_16xh_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
int height) {
uint32x4_t sse[2] = { vdupq_n_u32(0), vdupq_n_u32(0) };
int i = height;
do {
sse_16x1_neon_dotprod(src, ref, &sse[0]);
src += src_stride;
ref += ref_stride;
sse_16x1_neon_dotprod(src, ref, &sse[1]);
src += src_stride;
ref += ref_stride;
i -= 2;
} while (i != 0);
return horizontal_add_u32x4(vaddq_u32(sse[0], sse[1]));
}
static INLINE uint32_t sse_8xh_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
int height) {
uint32x2_t sse[2] = { vdup_n_u32(0), vdup_n_u32(0) };
int i = height;
do {
sse_8x1_neon_dotprod(src, ref, &sse[0]);
src += src_stride;
ref += ref_stride;
sse_8x1_neon_dotprod(src, ref, &sse[1]);
src += src_stride;
ref += ref_stride;
i -= 2;
} while (i != 0);
return horizontal_add_u32x4(vcombine_u32(sse[0], sse[1]));
}
static INLINE uint32_t sse_4xh_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
int height) {
uint32x2_t sse = vdup_n_u32(0);
int i = height;
do {
sse_4x2_neon_dotprod(src, src_stride, ref, ref_stride, &sse);
src += 2 * src_stride;
ref += 2 * ref_stride;
i -= 2;
} while (i != 0);
return horizontal_add_u32x2(sse);
}
int64_t aom_sse_neon_dotprod(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride, int width,
int height) {
switch (width) {
case 4:
return sse_4xh_neon_dotprod(src, src_stride, ref, ref_stride, height);
case 8:
return sse_8xh_neon_dotprod(src, src_stride, ref, ref_stride, height);
case 16:
return sse_16xh_neon_dotprod(src, src_stride, ref, ref_stride, height);
case 32:
return sse_32xh_neon_dotprod(src, src_stride, ref, ref_stride, height);
case 64:
return sse_64xh_neon_dotprod(src, src_stride, ref, ref_stride, height);
case 128:
return sse_128xh_neon_dotprod(src, src_stride, ref, ref_stride, height);
default:
return sse_wxh_neon_dotprod(src, src_stride, ref, ref_stride, width,
height);
}
}
|
