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
|
/*
* Copyright (c) 2021-2024 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <cstddef>
#include <cstdint>
#if defined(__aarch64__) && defined(__ARM_FP16_ARGS) && defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
namespace arm_conv {
namespace pooling {
void a64_fp16_nhwc_max_2x2_s1_output2x2_depthfirst_impl(
const unsigned int n_channels,
const __fp16 *const *const inptrs,
__fp16 *const *const outptrs,
const bool exclude_padding,
const unsigned int pad_left,
const unsigned int pad_top,
const unsigned int pad_right,
const unsigned int pad_bottom
)
{
struct KernelArgs
{
const uint64_t n_channels;
const __fp16 *const *const inptrs;
__fp16 *const *const outptrs;
KernelArgs(
unsigned int channels,
const __fp16 *const *input_ptrs,
__fp16 *const * output_ptrs,
bool, unsigned int, unsigned int, unsigned int, unsigned int
) : n_channels(channels),
inptrs(input_ptrs),
outptrs(output_ptrs)
{
}
};
const KernelArgs args(n_channels, inptrs, outptrs, exclude_padding,
pad_left, pad_top, pad_right, pad_bottom);
__asm__ __volatile__(
"ldr x16, [%x[args], %[offsetof_n_channels]]\n"
"ldr x21, [%x[args], %[offsetof_outptrs]]\n"
"mov x15, #0x0\n"
"mov x14, #0x0\n"
"ldr x20, [%x[args], %[offsetof_inptrs]]\n"
"cmp x16, #0x8\n"
"ldp x13, x12, [x21, #0x0]\n"
"ldp x11, x10, [x21, #0x10]\n"
"ldp x9, x28, [x20, #0x0]\n"
"ldp x27, x26, [x20, #0x10]\n"
"ldp x25, x24, [x20, #0x20]\n"
"ldp x23, x22, [x20, #0x30]\n"
"ldr x21, [x20, #0x40]\n"
"blt 3f\n"
"ldr q30, [x28, x15]\n"
"ldr q29, [x25, x15]\n"
"lsr x20, x16, #0x3\n"
"ldr q28, [x22, x15]\n"
"ldr q27, [x26, x15]\n"
"ldr q26, [x9, x15]\n"
"ldr q25, [x27, x15]\n"
"ldr q24, [x24, x15]\n"
"ldr q23, [x23, x15]\n"
"sub x16, x16, x20, LSL #3\n"
"subs x20, x20, #0x1\n"
"ldr q22, [x21, x15]\n"
"add x15, x15, #0x10\n"
"beq 2f\n"
"1:" // Vector: Loop
"fmax v21.8h, v30.8h, v29.8h\n"
"ldr q30, [x28, x15]\n"
"fmax v20.8h, v29.8h, v28.8h\n"
"ldr q29, [x25, x15]\n"
"ldr q28, [x22, x15]\n"
"fmax v19.8h, v27.8h, v26.8h\n"
"ldr q26, [x9, x15]\n"
"fmax v18.8h, v25.8h, v24.8h\n"
"ldr q25, [x27, x15]\n"
"fmax v17.8h, v27.8h, v23.8h\n"
"ldr q27, [x26, x15]\n"
"fmax v16.8h, v24.8h, v22.8h\n"
"ldr q24, [x24, x15]\n"
"ldr q23, [x23, x15]\n"
"subs x20, x20, #0x1\n"
"ldr q22, [x21, x15]\n"
"fmax v19.8h, v21.8h, v19.8h\n"
"fmax v18.8h, v18.8h, v21.8h\n"
"add x15, x15, #0x10\n"
"fmax v17.8h, v17.8h, v20.8h\n"
"fmax v16.8h, v20.8h, v16.8h\n"
"str q19, [x13, x14]\n"
"str q18, [x12, x14]\n"
"str q17, [x11, x14]\n"
"str q16, [x10, x14]\n"
"add x14, x14, #0x10\n"
"bgt 1b\n"
"2:" // Vector: Tail
"fmax v21.8h, v30.8h, v29.8h\n"
"fmax v20.8h, v29.8h, v28.8h\n"
"fmax v19.8h, v27.8h, v26.8h\n"
"fmax v18.8h, v25.8h, v24.8h\n"
"fmax v17.8h, v27.8h, v23.8h\n"
"fmax v16.8h, v24.8h, v22.8h\n"
"fmax v19.8h, v21.8h, v19.8h\n"
"fmax v18.8h, v18.8h, v21.8h\n"
"fmax v17.8h, v17.8h, v20.8h\n"
"fmax v16.8h, v20.8h, v16.8h\n"
"str q19, [x13, x14]\n"
"str q18, [x12, x14]\n"
"str q17, [x11, x14]\n"
"str q16, [x10, x14]\n"
"add x14, x14, #0x10\n"
"cbz x16, 4f\n"
"3:" // Oddments
"ldr h16, [x28, x15]\n"
"ldr h24, [x25, x15]\n"
"subs x16, x16, #0x1\n"
"ldr h20, [x22, x15]\n"
"ldr h23, [x26, x15]\n"
"ldr h19, [x9, x15]\n"
"ldr h18, [x27, x15]\n"
"ldr h22, [x24, x15]\n"
"ldr h17, [x23, x15]\n"
"fmax v21.8h, v16.8h, v24.8h\n"
"ldr h16, [x21, x15]\n"
"fmax v20.8h, v24.8h, v20.8h\n"
"add x15, x15, #0x2\n"
"fmax v19.8h, v23.8h, v19.8h\n"
"fmax v18.8h, v18.8h, v22.8h\n"
"fmax v17.8h, v23.8h, v17.8h\n"
"fmax v16.8h, v22.8h, v16.8h\n"
"fmax v19.8h, v21.8h, v19.8h\n"
"fmax v18.8h, v18.8h, v21.8h\n"
"fmax v17.8h, v17.8h, v20.8h\n"
"fmax v16.8h, v20.8h, v16.8h\n"
"str h19, [x13, x14]\n"
"str h18, [x12, x14]\n"
"str h17, [x11, x14]\n"
"str h16, [x10, x14]\n"
"add x14, x14, #0x2\n"
"bgt 3b\n"
"4:" // End
:
: [args] "r" (&args), [offsetof_inptrs] "I" (offsetof(KernelArgs, inptrs)), [offsetof_n_channels] "I" (offsetof(KernelArgs, n_channels)), [offsetof_outptrs] "I" (offsetof(KernelArgs, outptrs))
: "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28"
);
}
} // namespace pooling
} // namespace arm_conv
#endif // defined(__aarch64__) && defined(__ARM_FP16_ARGS) && defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
|