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) 2021-2023 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 <cstdint>
#include <cstddef>
#if defined(ARM_COMPUTE_ENABLE_SVE)
namespace arm_conv {
namespace pooling {
void sve_s8_nhwc_max_generic_depthfirst_impl(
const uint64_t,
const uint64_t n_valid_cells,
uint64_t n_channels,
const int8_t *const *const inptrs,
int8_t *outptr
)
{
__asm__ __volatile__(
"mov x9, #0x0\n"
"cntb x28\n"
"cntb x27, ALL, MUL #2\n"
"cntb x26, ALL, MUL #3\n"
"whilelt p4.b, x9, %x[n_channels]\n"
"whilelt p3.b, x28, %x[n_channels]\n"
"whilelt p2.b, x27, %x[n_channels]\n"
"whilelt p1.b, x26, %x[n_channels]\n"
"ptrue p0.b\n"
"b.none 7f\n"
"1:" // 4-vectors of channels
"lsr x25, %x[n_valid_cells], #0x2\n"
"mov z8.b, #0x80\n"
"mov z7.b, #0x80\n"
"mov x24, %x[inptrs]\n"
"mov z6.b, #0x80\n"
"mov z5.b, #0x80\n"
"cbz x25, 4f\n"
"ldp x23, x22, [x24, #0x0]\n"
"ldp x21, x20, [x24, #0x10]\n"
"subs x25, x25, #0x1\n"
"add x24, x24, #0x20\n"
"ld1b { z4.b }, p4/Z, [x23, x9]\n"
"ld1b { z3.b }, p4/Z, [x22, x9]\n"
"ld1b { z2.b }, p4/Z, [x21, x9]\n"
"ld1b { z1.b }, p4/Z, [x20, x9]\n"
"ld1b { z0.b }, p3/Z, [x23, x28]\n"
"ld1b { z31.b }, p3/Z, [x22, x28]\n"
"ld1b { z22.b }, p3/Z, [x21, x28]\n"
"ld1b { z30.b }, p3/Z, [x20, x28]\n"
"ld1b { z29.b }, p2/Z, [x23, x27]\n"
"ld1b { z28.b }, p2/Z, [x22, x27]\n"
"ld1b { z21.b }, p2/Z, [x21, x27]\n"
"ld1b { z27.b }, p2/Z, [x20, x27]\n"
"ld1b { z26.b }, p1/Z, [x23, x26]\n"
"ld1b { z25.b }, p1/Z, [x22, x26]\n"
"ld1b { z20.b }, p1/Z, [x21, x26]\n"
"ld1b { z24.b }, p1/Z, [x20, x26]\n"
"beq 3f\n"
"2:" // 4-vectors of channels: 4 inputs loop
"movprfx z19, z4\n smax z19.b, p0/M, z19.b, z3.b\n"
"movprfx z23, z2\n smax z23.b, p0/M, z23.b, z1.b\n"
"ldp x23, x22, [x24, #0x0]\n"
"ldp x21, x20, [x24, #0x10]\n"
"movprfx z18, z0\n smax z18.b, p0/M, z18.b, z31.b\n"
"smax z22.b, p0/M, z22.b, z30.b\n"
"ld1b { z4.b }, p4/Z, [x23, x9]\n"
"ld1b { z3.b }, p4/Z, [x22, x9]\n"
"movprfx z17, z29\n smax z17.b, p0/M, z17.b, z28.b\n"
"smax z21.b, p0/M, z21.b, z27.b\n"
"ld1b { z2.b }, p4/Z, [x21, x9]\n"
"ld1b { z1.b }, p4/Z, [x20, x9]\n"
"movprfx z16, z26\n smax z16.b, p0/M, z16.b, z25.b\n"
"smax z20.b, p0/M, z20.b, z24.b\n"
"ld1b { z0.b }, p3/Z, [x23, x28]\n"
"ld1b { z31.b }, p3/Z, [x22, x28]\n"
"smax z19.b, p0/M, z19.b, z23.b\n"
"smax z18.b, p0/M, z18.b, z22.b\n"
"ld1b { z22.b }, p3/Z, [x21, x28]\n"
"ld1b { z30.b }, p3/Z, [x20, x28]\n"
"smax z17.b, p0/M, z17.b, z21.b\n"
"smax z16.b, p0/M, z16.b, z20.b\n"
"ld1b { z29.b }, p2/Z, [x23, x27]\n"
"ld1b { z28.b }, p2/Z, [x22, x27]\n"
"subs x25, x25, #0x1\n"
"smax z8.b, p0/M, z8.b, z19.b\n"
"ld1b { z21.b }, p2/Z, [x21, x27]\n"
"ld1b { z27.b }, p2/Z, [x20, x27]\n"
"smax z7.b, p0/M, z7.b, z18.b\n"
"smax z6.b, p0/M, z6.b, z17.b\n"
"ld1b { z26.b }, p1/Z, [x23, x26]\n"
"ld1b { z25.b }, p1/Z, [x22, x26]\n"
"smax z5.b, p0/M, z5.b, z16.b\n"
"add x24, x24, #0x20\n"
"ld1b { z20.b }, p1/Z, [x21, x26]\n"
"ld1b { z24.b }, p1/Z, [x20, x26]\n"
"bgt 2b\n"
"3:" // 4-vectors of channels: 4 inputs tail
"movprfx z19, z4\n smax z19.b, p0/M, z19.b, z3.b\n"
"movprfx z23, z2\n smax z23.b, p0/M, z23.b, z1.b\n"
"movprfx z18, z0\n smax z18.b, p0/M, z18.b, z31.b\n"
"smax z22.b, p0/M, z22.b, z30.b\n"
"movprfx z17, z29\n smax z17.b, p0/M, z17.b, z28.b\n"
"smax z21.b, p0/M, z21.b, z27.b\n"
"movprfx z16, z26\n smax z16.b, p0/M, z16.b, z25.b\n"
"smax z20.b, p0/M, z20.b, z24.b\n"
"smax z19.b, p0/M, z19.b, z23.b\n"
"smax z18.b, p0/M, z18.b, z22.b\n"
"smax z17.b, p0/M, z17.b, z21.b\n"
"smax z16.b, p0/M, z16.b, z20.b\n"
"smax z8.b, p0/M, z8.b, z19.b\n"
"smax z7.b, p0/M, z7.b, z18.b\n"
"smax z6.b, p0/M, z6.b, z17.b\n"
"smax z5.b, p0/M, z5.b, z16.b\n"
"4:" // 4-vectors of channels: After loop
"ands x21, %x[n_valid_cells], #0x3\n"
"beq 6f\n"
"5:" // 4-vectors of channels: Single input loop
"ldr x20, [x24], #0x8\n"
"ld1b { z16.b }, p4/Z, [x20, x9]\n"
"subs x21, x21, #0x1\n"
"smax z8.b, p0/M, z8.b, z16.b\n"
"ld1b { z17.b }, p3/Z, [x20, x28]\n"
"ld1b { z16.b }, p2/Z, [x20, x27]\n"
"smax z7.b, p0/M, z7.b, z17.b\n"
"smax z6.b, p0/M, z6.b, z16.b\n"
"ld1b { z16.b }, p1/Z, [x20, x26]\n"
"smax z5.b, p0/M, z5.b, z16.b\n"
"bgt 5b\n"
"6:" // 4-vectors of channels: Single input loop: End
"st1b { z8.b }, p4, [%x[outptr], x9]\n"
"incb x9, ALL, MUL #4\n"
"st1b { z7.b }, p3, [%x[outptr], x28]\n"
"incb x28, ALL, MUL #4\n"
"st1b { z6.b }, p2, [%x[outptr], x27]\n"
"incb x27, ALL, MUL #4\n"
"st1b { z5.b }, p1, [%x[outptr], x26]\n"
"incb x26, ALL, MUL #4\n"
"whilelt p1.b, x26, %x[n_channels]\n"
"b.any 1b\n"
"7:" // Single vector of channels
"whilelt p4.b, x9, %x[n_channels]\n"
"b.none 14f\n"
"8:" // Single vector of channels: Loop
"lsr x25, %x[n_valid_cells], #0x2\n"
"mov z8.b, #0x80\n"
"mov x24, %x[inptrs]\n"
"cbz x25, 11f\n"
"ldp x23, x22, [x24, #0x0]\n"
"ldp x21, x20, [x24, #0x10]\n"
"subs x25, x25, #0x1\n"
"add x24, x24, #0x20\n"
"ld1b { z4.b }, p4/Z, [x23, x9]\n"
"ld1b { z3.b }, p4/Z, [x22, x9]\n"
"ld1b { z2.b }, p4/Z, [x21, x9]\n"
"ld1b { z1.b }, p4/Z, [x20, x9]\n"
"beq 10f\n"
"9:" // Single vector of channels: Loop: 4 inputs loop
"movprfx z16, z4\n smax z16.b, p0/M, z16.b, z3.b\n"
"movprfx z17, z2\n smax z17.b, p0/M, z17.b, z1.b\n"
"ldp x23, x22, [x24, #0x0]\n"
"ldp x21, x20, [x24, #0x10]\n"
"smax z16.b, p0/M, z16.b, z17.b\n"
"subs x25, x25, #0x1\n"
"ld1b { z4.b }, p4/Z, [x23, x9]\n"
"ld1b { z3.b }, p4/Z, [x22, x9]\n"
"smax z8.b, p0/M, z8.b, z16.b\n"
"add x24, x24, #0x20\n"
"ld1b { z2.b }, p4/Z, [x21, x9]\n"
"ld1b { z1.b }, p4/Z, [x20, x9]\n"
"bgt 9b\n"
"10:" // Single vector of channels: Loop: 4 inputs tail
"movprfx z16, z4\n smax z16.b, p0/M, z16.b, z3.b\n"
"movprfx z17, z2\n smax z17.b, p0/M, z17.b, z1.b\n"
"smax z16.b, p0/M, z16.b, z17.b\n"
"smax z8.b, p0/M, z8.b, z16.b\n"
"11:" // Single vector of channels: Loop: After loop
"ands x21, %x[n_valid_cells], #0x3\n"
"beq 13f\n"
"12:" // Single vector of channels: Loop: Single input loop
"ldr x20, [x24], #0x8\n"
"ld1b { z16.b }, p4/Z, [x20, x9]\n"
"subs x21, x21, #0x1\n"
"smax z8.b, p0/M, z8.b, z16.b\n"
"bgt 12b\n"
"13:" // Single vector of channels: Loop: Single input loop: End
"st1b { z8.b }, p4, [%x[outptr], x9]\n"
"incb x9\n"
"whilelt p4.b, x9, %x[n_channels]\n"
"b.any 8b\n"
"14:" // End
:
: [inptrs] "r" (inptrs), [n_channels] "r" (n_channels), [n_valid_cells] "r" (n_valid_cells), [outptr] "r" (outptr)
: "cc", "memory", "p0", "p1", "p2", "p3", "p4", "x9", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
);
}
} // namespace pooling
} // namespace arm_conv
#endif // defined(ARM_COMPUTE_ENABLE_SVE)
|
