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
|
/*
*
* Copyright (c) 2018, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <arm_neon.h>
#include <assert.h>
#include "aom/aom_integer.h"
#include "aom_dsp/blend.h"
#include "aom_ports/mem.h"
#include "av1/common/arm/mem_neon.h"
#include "aom_dsp/aom_dsp_common.h"
#include "config/aom_dsp_rtcd.h"
void aom_blend_a64_vmask_neon(uint8_t *dst, uint32_t dst_stride,
const uint8_t *src0, uint32_t src0_stride,
const uint8_t *src1, uint32_t src1_stride,
const uint8_t *mask, int w, int h) {
uint8x8_t tmp0, tmp1;
uint8x16_t tmp0_q, tmp1_q, res_q;
uint16x8_t res, res_low, res_high;
uint32x2_t tmp0_32 = vdup_n_u32(0), tmp1_32 = vdup_n_u32(0);
uint16x4_t tmp0_16 = vdup_n_u16(0), tmp1_16 = vdup_n_u16(0);
assert(IMPLIES(src0 == dst, src0_stride == dst_stride));
assert(IMPLIES(src1 == dst, src1_stride == dst_stride));
assert(h >= 2);
assert(w >= 2);
assert(IS_POWER_OF_TWO(h));
assert(IS_POWER_OF_TWO(w));
if (w >= 16) {
for (int i = 0; i < h; ++i) {
const uint8x8_t m = vdup_n_u8((uint8_t)mask[i]);
const uint8x8_t max_minus_m = vdup_n_u8(64 - (uint8_t)mask[i]);
for (int j = 0; j < w; j += 16) {
__builtin_prefetch(src0);
__builtin_prefetch(src1);
tmp0_q = vld1q_u8(src0);
tmp1_q = vld1q_u8(src1);
res_low = vmull_u8(m, vget_low_u8(tmp0_q));
res_low = vmlal_u8(res_low, max_minus_m, vget_low_u8(tmp1_q));
res_high = vmull_u8(m, vget_high_u8(tmp0_q));
res_high = vmlal_u8(res_high, max_minus_m, vget_high_u8(tmp1_q));
res_q = vcombine_u8(vrshrn_n_u16(res_low, AOM_BLEND_A64_ROUND_BITS),
vrshrn_n_u16(res_high, AOM_BLEND_A64_ROUND_BITS));
vst1q_u8(dst, res_q);
src0 += 16;
src1 += 16;
dst += 16;
}
src0 += src0_stride - w;
src1 += src1_stride - w;
dst += dst_stride - w;
}
} else if (w == 8) {
for (int i = 0; i < h; ++i) {
__builtin_prefetch(src0);
__builtin_prefetch(src1);
const uint8x8_t m = vdup_n_u8((uint8_t)mask[i]);
const uint8x8_t max_minus_m = vdup_n_u8(64 - (uint8_t)mask[i]);
tmp0 = vld1_u8(src0);
tmp1 = vld1_u8(src1);
res = vmull_u8(m, tmp0);
res = vmlal_u8(res, max_minus_m, tmp1);
vst1_u8(dst, vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS));
src0 += src0_stride;
src1 += src1_stride;
dst += dst_stride;
}
} else if (w == 4) {
for (int i = 0; i < h; i += 2) {
__builtin_prefetch(src0 + 0 * src0_stride);
__builtin_prefetch(src0 + 1 * src0_stride);
__builtin_prefetch(src1 + 0 * src1_stride);
__builtin_prefetch(src1 + 1 * src1_stride);
const uint16x4_t m1 = vdup_n_u16((uint16_t)mask[i]);
const uint16x4_t m2 = vdup_n_u16((uint16_t)mask[i + 1]);
const uint8x8_t m = vmovn_u16(vcombine_u16(m1, m2));
const uint16x4_t max_minus_m1 = vdup_n_u16(64 - (uint16_t)mask[i]);
const uint16x4_t max_minus_m2 = vdup_n_u16(64 - (uint16_t)mask[i + 1]);
const uint8x8_t max_minus_m =
vmovn_u16(vcombine_u16(max_minus_m1, max_minus_m2));
load_unaligned_u8_4x2(src0, src0_stride, &tmp0_32);
tmp0 = vreinterpret_u8_u32(tmp0_32);
load_unaligned_u8_4x2(src1, src1_stride, &tmp1_32);
tmp1 = vreinterpret_u8_u32(tmp1_32);
res = vmull_u8(m, tmp0);
res = vmlal_u8(res, max_minus_m, tmp1);
vst1_lane_u32(
(uint32_t *)(dst + (0 * dst_stride)),
vreinterpret_u32_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 0);
vst1_lane_u32(
(uint32_t *)(dst + (1 * dst_stride)),
vreinterpret_u32_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 1);
src0 += (2 * src0_stride);
src1 += (2 * src1_stride);
dst += (2 * dst_stride);
}
} else if (w == 2) {
for (int i = 0; i < h; i += 2) {
__builtin_prefetch(src0 + 0 * src0_stride);
__builtin_prefetch(src0 + 1 * src0_stride);
__builtin_prefetch(src1 + 0 * src1_stride);
__builtin_prefetch(src1 + 1 * src1_stride);
const uint8x8_t m1 = vdup_n_u8(mask[i]);
const uint8x8_t m2 = vdup_n_u8(mask[i + 1]);
const uint16x4x2_t m_trn =
vtrn_u16(vreinterpret_u16_u8(m1), vreinterpret_u16_u8(m2));
const uint8x8_t m = vreinterpret_u8_u16(m_trn.val[0]);
const uint8x8_t max_minus_m1 = vdup_n_u8(64 - mask[i]);
const uint8x8_t max_minus_m2 = vdup_n_u8(64 - mask[i + 1]);
const uint16x4x2_t max_minus_m_trn = vtrn_u16(
vreinterpret_u16_u8(max_minus_m1), vreinterpret_u16_u8(max_minus_m2));
const uint8x8_t max_minus_m = vreinterpret_u8_u16(max_minus_m_trn.val[0]);
load_unaligned_u8_2x2(src0, src0_stride, &tmp0_16);
tmp0 = vreinterpret_u8_u16(tmp0_16);
load_unaligned_u8_2x2(src1, src1_stride, &tmp1_16);
tmp1 = vreinterpret_u8_u16(tmp1_16);
res = vmull_u8(m, tmp0);
res = vmlal_u8(res, max_minus_m, tmp1);
vst1_lane_u16(
(uint16_t *)(dst + (0 * dst_stride)),
vreinterpret_u16_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 0);
vst1_lane_u16(
(uint16_t *)(dst + (1 * dst_stride)),
vreinterpret_u16_u8(vrshrn_n_u16(res, AOM_BLEND_A64_ROUND_BITS)), 1);
src0 += (2 * src0_stride);
src1 += (2 * src1_stride);
dst += (2 * dst_stride);
}
}
}
|
