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
|
/*
* Copyright (c) 2017 The WebM project authors. 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 "./vpx_dsp_rtcd.h"
#include "./vpx_config.h"
#include "vpx_dsp/arm/mem_neon.h"
#include "vpx_dsp/arm/sum_neon.h"
void vpx_fdct4x4_1_neon(const int16_t *input, tran_low_t *output, int stride) {
int16x4_t a0, a1, a2, a3;
int16x8_t b0, b1;
int16x8_t c;
a0 = vld1_s16(input);
input += stride;
a1 = vld1_s16(input);
input += stride;
a2 = vld1_s16(input);
input += stride;
a3 = vld1_s16(input);
b0 = vcombine_s16(a0, a1);
b1 = vcombine_s16(a2, a3);
c = vaddq_s16(b0, b1);
output[0] = (tran_low_t)(horizontal_add_int16x8(c) << 1);
output[1] = 0;
}
// Visual Studio 2022 (cl.exe) targeting AArch64 with optimizations enabled
// will fail with an internal compiler error.
// See:
// https://developercommunity.visualstudio.com/t/Compiler-crash-C1001-when-building-a-for/10346110
// TODO(jzern): check the compiler version after a fix for the issue is
// released.
#if defined(_MSC_VER) && defined(_M_ARM64) && !defined(__clang__)
#pragma optimize("", off)
#endif
void vpx_fdct8x8_1_neon(const int16_t *input, tran_low_t *output, int stride) {
int r;
int16x8_t sum = vld1q_s16(&input[0]);
for (r = 1; r < 8; ++r) {
const int16x8_t input_00 = vld1q_s16(&input[r * stride]);
sum = vaddq_s16(sum, input_00);
}
output[0] = (tran_low_t)horizontal_add_int16x8(sum);
output[1] = 0;
}
#if defined(_MSC_VER) && defined(_M_ARM64) && !defined(__clang__)
#pragma optimize("", on)
#endif
void vpx_fdct16x16_1_neon(const int16_t *input, tran_low_t *output,
int stride) {
int r;
int16x8_t left = vld1q_s16(input);
int16x8_t right = vld1q_s16(input + 8);
int32_t sum;
input += stride;
for (r = 1; r < 16; ++r) {
const int16x8_t a = vld1q_s16(input);
const int16x8_t b = vld1q_s16(input + 8);
input += stride;
left = vaddq_s16(left, a);
right = vaddq_s16(right, b);
}
sum = horizontal_add_int16x8(left) + horizontal_add_int16x8(right);
output[0] = (tran_low_t)(sum >> 1);
output[1] = 0;
}
void vpx_fdct32x32_1_neon(const int16_t *input, tran_low_t *output,
int stride) {
int r;
int16x8_t a0 = vld1q_s16(input);
int16x8_t a1 = vld1q_s16(input + 8);
int16x8_t a2 = vld1q_s16(input + 16);
int16x8_t a3 = vld1q_s16(input + 24);
int32_t sum;
input += stride;
for (r = 1; r < 32; ++r) {
const int16x8_t b0 = vld1q_s16(input);
const int16x8_t b1 = vld1q_s16(input + 8);
const int16x8_t b2 = vld1q_s16(input + 16);
const int16x8_t b3 = vld1q_s16(input + 24);
input += stride;
a0 = vaddq_s16(a0, b0);
a1 = vaddq_s16(a1, b1);
a2 = vaddq_s16(a2, b2);
a3 = vaddq_s16(a3, b3);
}
sum = horizontal_add_int16x8(a0);
sum += horizontal_add_int16x8(a1);
sum += horizontal_add_int16x8(a2);
sum += horizontal_add_int16x8(a3);
output[0] = (tran_low_t)(sum >> 3);
output[1] = 0;
}
#if CONFIG_VP9_HIGHBITDEPTH
void vpx_highbd_fdct16x16_1_neon(const int16_t *input, tran_low_t *output,
int stride) {
int32x4_t partial_sum[4] = { vdupq_n_s32(0), vdupq_n_s32(0), vdupq_n_s32(0),
vdupq_n_s32(0) };
int32_t sum;
int r = 0;
do {
const int16x8_t a = vld1q_s16(input);
const int16x8_t b = vld1q_s16(input + 8);
input += stride;
partial_sum[0] = vaddw_s16(partial_sum[0], vget_low_s16(a));
partial_sum[1] = vaddw_s16(partial_sum[1], vget_high_s16(a));
partial_sum[2] = vaddw_s16(partial_sum[2], vget_low_s16(b));
partial_sum[3] = vaddw_s16(partial_sum[3], vget_high_s16(b));
r++;
} while (r < 16);
partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[1]);
partial_sum[2] = vaddq_s32(partial_sum[2], partial_sum[3]);
partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[2]);
sum = horizontal_add_int32x4(partial_sum[0]);
output[0] = (tran_low_t)(sum >> 1);
output[1] = 0;
}
void vpx_highbd_fdct32x32_1_neon(const int16_t *input, tran_low_t *output,
int stride) {
int32x4_t partial_sum[4] = { vdupq_n_s32(0), vdupq_n_s32(0), vdupq_n_s32(0),
vdupq_n_s32(0) };
int32_t sum;
int r = 0;
do {
const int16x8_t a0 = vld1q_s16(input);
const int16x8_t a1 = vld1q_s16(input + 8);
const int16x8_t a2 = vld1q_s16(input + 16);
const int16x8_t a3 = vld1q_s16(input + 24);
input += stride;
partial_sum[0] = vaddw_s16(partial_sum[0], vget_low_s16(a0));
partial_sum[0] = vaddw_s16(partial_sum[0], vget_high_s16(a0));
partial_sum[1] = vaddw_s16(partial_sum[1], vget_low_s16(a1));
partial_sum[1] = vaddw_s16(partial_sum[1], vget_high_s16(a1));
partial_sum[2] = vaddw_s16(partial_sum[2], vget_low_s16(a2));
partial_sum[2] = vaddw_s16(partial_sum[2], vget_high_s16(a2));
partial_sum[3] = vaddw_s16(partial_sum[3], vget_low_s16(a3));
partial_sum[3] = vaddw_s16(partial_sum[3], vget_high_s16(a3));
r++;
} while (r < 32);
partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[1]);
partial_sum[2] = vaddq_s32(partial_sum[2], partial_sum[3]);
partial_sum[0] = vaddq_s32(partial_sum[0], partial_sum[2]);
sum = horizontal_add_int32x4(partial_sum[0]);
output[0] = (tran_low_t)(sum >> 3);
output[1] = 0;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
|
