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
|
/*
* 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 "aom_dsp/arm/dot_sve.h"
#include "aom_dsp/arm/mem_neon.h"
#include "config/aom_dsp_rtcd.h"
static INLINE void highbd_sse_8x1_neon(const uint16_t *src, const uint16_t *ref,
uint64x2_t *sse) {
uint16x8_t s = vld1q_u16(src);
uint16x8_t r = vld1q_u16(ref);
uint16x8_t abs_diff = vabdq_u16(s, r);
*sse = aom_udotq_u16(*sse, abs_diff, abs_diff);
}
static INLINE int64_t highbd_sse_128xh_sve(const uint16_t *src, int src_stride,
const uint16_t *ref, int ref_stride,
int height) {
uint64x2_t sse[4] = { vdupq_n_u64(0), vdupq_n_u64(0), vdupq_n_u64(0),
vdupq_n_u64(0) };
do {
highbd_sse_8x1_neon(src + 0 * 8, ref + 0 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 1 * 8, ref + 1 * 8, &sse[1]);
highbd_sse_8x1_neon(src + 2 * 8, ref + 2 * 8, &sse[2]);
highbd_sse_8x1_neon(src + 3 * 8, ref + 3 * 8, &sse[3]);
highbd_sse_8x1_neon(src + 4 * 8, ref + 4 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 5 * 8, ref + 5 * 8, &sse[1]);
highbd_sse_8x1_neon(src + 6 * 8, ref + 6 * 8, &sse[2]);
highbd_sse_8x1_neon(src + 7 * 8, ref + 7 * 8, &sse[3]);
highbd_sse_8x1_neon(src + 8 * 8, ref + 8 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 9 * 8, ref + 9 * 8, &sse[1]);
highbd_sse_8x1_neon(src + 10 * 8, ref + 10 * 8, &sse[2]);
highbd_sse_8x1_neon(src + 11 * 8, ref + 11 * 8, &sse[3]);
highbd_sse_8x1_neon(src + 12 * 8, ref + 12 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 13 * 8, ref + 13 * 8, &sse[1]);
highbd_sse_8x1_neon(src + 14 * 8, ref + 14 * 8, &sse[2]);
highbd_sse_8x1_neon(src + 15 * 8, ref + 15 * 8, &sse[3]);
src += src_stride;
ref += ref_stride;
} while (--height != 0);
sse[0] = vaddq_u64(sse[0], sse[1]);
sse[2] = vaddq_u64(sse[2], sse[3]);
sse[0] = vaddq_u64(sse[0], sse[2]);
return vaddvq_u64(sse[0]);
}
static INLINE int64_t highbd_sse_64xh_sve(const uint16_t *src, int src_stride,
const uint16_t *ref, int ref_stride,
int height) {
uint64x2_t sse[4] = { vdupq_n_u64(0), vdupq_n_u64(0), vdupq_n_u64(0),
vdupq_n_u64(0) };
do {
highbd_sse_8x1_neon(src + 0 * 8, ref + 0 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 1 * 8, ref + 1 * 8, &sse[1]);
highbd_sse_8x1_neon(src + 2 * 8, ref + 2 * 8, &sse[2]);
highbd_sse_8x1_neon(src + 3 * 8, ref + 3 * 8, &sse[3]);
highbd_sse_8x1_neon(src + 4 * 8, ref + 4 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 5 * 8, ref + 5 * 8, &sse[1]);
highbd_sse_8x1_neon(src + 6 * 8, ref + 6 * 8, &sse[2]);
highbd_sse_8x1_neon(src + 7 * 8, ref + 7 * 8, &sse[3]);
src += src_stride;
ref += ref_stride;
} while (--height != 0);
sse[0] = vaddq_u64(sse[0], sse[1]);
sse[2] = vaddq_u64(sse[2], sse[3]);
sse[0] = vaddq_u64(sse[0], sse[2]);
return vaddvq_u64(sse[0]);
}
static INLINE int64_t highbd_sse_32xh_sve(const uint16_t *src, int src_stride,
const uint16_t *ref, int ref_stride,
int height) {
uint64x2_t sse[4] = { vdupq_n_u64(0), vdupq_n_u64(0), vdupq_n_u64(0),
vdupq_n_u64(0) };
do {
highbd_sse_8x1_neon(src + 0 * 8, ref + 0 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 1 * 8, ref + 1 * 8, &sse[1]);
highbd_sse_8x1_neon(src + 2 * 8, ref + 2 * 8, &sse[2]);
highbd_sse_8x1_neon(src + 3 * 8, ref + 3 * 8, &sse[3]);
src += src_stride;
ref += ref_stride;
} while (--height != 0);
sse[0] = vaddq_u64(sse[0], sse[1]);
sse[2] = vaddq_u64(sse[2], sse[3]);
sse[0] = vaddq_u64(sse[0], sse[2]);
return vaddvq_u64(sse[0]);
}
static INLINE int64_t highbd_sse_16xh_sve(const uint16_t *src, int src_stride,
const uint16_t *ref, int ref_stride,
int height) {
uint64x2_t sse[2] = { vdupq_n_u64(0), vdupq_n_u64(0) };
do {
highbd_sse_8x1_neon(src + 0 * 8, ref + 0 * 8, &sse[0]);
highbd_sse_8x1_neon(src + 1 * 8, ref + 1 * 8, &sse[1]);
src += src_stride;
ref += ref_stride;
} while (--height != 0);
return vaddvq_u64(vaddq_u64(sse[0], sse[1]));
}
static INLINE int64_t highbd_sse_8xh_sve(const uint16_t *src, int src_stride,
const uint16_t *ref, int ref_stride,
int height) {
uint64x2_t sse[2] = { vdupq_n_u64(0), vdupq_n_u64(0) };
do {
highbd_sse_8x1_neon(src + 0 * src_stride, ref + 0 * ref_stride, &sse[0]);
highbd_sse_8x1_neon(src + 1 * src_stride, ref + 1 * ref_stride, &sse[1]);
src += 2 * src_stride;
ref += 2 * ref_stride;
height -= 2;
} while (height != 0);
return vaddvq_u64(vaddq_u64(sse[0], sse[1]));
}
static INLINE int64_t highbd_sse_4xh_sve(const uint16_t *src, int src_stride,
const uint16_t *ref, int ref_stride,
int height) {
uint64x2_t sse = vdupq_n_u64(0);
do {
uint16x8_t s = load_unaligned_u16_4x2(src, src_stride);
uint16x8_t r = load_unaligned_u16_4x2(ref, ref_stride);
uint16x8_t abs_diff = vabdq_u16(s, r);
sse = aom_udotq_u16(sse, abs_diff, abs_diff);
src += 2 * src_stride;
ref += 2 * ref_stride;
height -= 2;
} while (height != 0);
return vaddvq_u64(sse);
}
static INLINE int64_t highbd_sse_wxh_sve(const uint16_t *src, int src_stride,
const uint16_t *ref, int ref_stride,
int width, int height) {
svuint64_t sse = svdup_n_u64(0);
uint64_t step = svcnth();
do {
int w = 0;
const uint16_t *src_ptr = src;
const uint16_t *ref_ptr = ref;
do {
svbool_t pred = svwhilelt_b16_u32(w, width);
svuint16_t s = svld1_u16(pred, src_ptr);
svuint16_t r = svld1_u16(pred, ref_ptr);
svuint16_t abs_diff = svabd_u16_z(pred, s, r);
sse = svdot_u64(sse, abs_diff, abs_diff);
src_ptr += step;
ref_ptr += step;
w += step;
} while (w < width);
src += src_stride;
ref += ref_stride;
} while (--height != 0);
return svaddv_u64(svptrue_b64(), sse);
}
int64_t aom_highbd_sse_sve(const uint8_t *src8, int src_stride,
const uint8_t *ref8, int ref_stride, int width,
int height) {
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
switch (width) {
case 4: return highbd_sse_4xh_sve(src, src_stride, ref, ref_stride, height);
case 8: return highbd_sse_8xh_sve(src, src_stride, ref, ref_stride, height);
case 16:
return highbd_sse_16xh_sve(src, src_stride, ref, ref_stride, height);
case 32:
return highbd_sse_32xh_sve(src, src_stride, ref, ref_stride, height);
case 64:
return highbd_sse_64xh_sve(src, src_stride, ref, ref_stride, height);
case 128:
return highbd_sse_128xh_sve(src, src_stride, ref, ref_stride, height);
default:
return highbd_sse_wxh_sve(src, src_stride, ref, ref_stride, width,
height);
}
}
|
