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
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
|
/*
* 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.
*/
#pragma once
#if defined(__aarch64__)
namespace {
void a64_transpose_interleave_128(uint32_t *out, const uint32_t *in, size_t width, size_t in_stride, size_t height)
{
size_t out_stride = 32 * height * sizeof(uint32_t);
__asm__ __volatile__(
"cmp %x[height], #0x4\n"
"blt 10f\n"
"1:" // Main row loop: Head
"mov x25, %x[in]\n"
"mov x24, %x[width]\n"
"add x23, x25, %x[in_stride]\n"
"add x22, x23, %x[in_stride]\n"
"add x20, x22, %x[in_stride]\n"
"cmp x24, #0x20\n"
"add %x[in], x20, %x[in_stride]\n"
"mov x21, %x[out]\n"
"sub %x[height], %x[height], #0x4\n"
"blt 3f\n"
"2:" // Main row loop: Column loop
"ldr q15, [x25], #0x10\n"
"ldr q14, [x23], #0x10\n"
"sub x24, x24, #0x20\n"
"cmp x24, #0x20\n"
"ldr q13, [x22], #0x10\n"
"ldr q12, [x20], #0x10\n"
"ldr q11, [x25], #0x10\n"
"ldr q10, [x23], #0x10\n"
"ldr q9, [x22], #0x10\n"
"ldr q8, [x20], #0x10\n"
"ldr q7, [x25], #0x10\n"
"ldr q6, [x23], #0x10\n"
"ldr q5, [x22], #0x10\n"
"ldr q4, [x20], #0x10\n"
"ldr q3, [x25], #0x10\n"
"ldr q2, [x23], #0x10\n"
"ldr q1, [x22], #0x10\n"
"ldr q0, [x20], #0x10\n"
"ldr q31, [x25], #0x10\n"
"ldr q30, [x23], #0x10\n"
"ldr q29, [x22], #0x10\n"
"ldr q28, [x20], #0x10\n"
"ldr q27, [x25], #0x10\n"
"ldr q26, [x23], #0x10\n"
"ldr q25, [x22], #0x10\n"
"ldr q24, [x20], #0x10\n"
"ldr q23, [x25], #0x10\n"
"ldr q22, [x23], #0x10\n"
"ldr q21, [x22], #0x10\n"
"ldr q20, [x20], #0x10\n"
"ldr q19, [x25], #0x10\n"
"ldr q18, [x23], #0x10\n"
"ldr q17, [x22], #0x10\n"
"ldr q16, [x20], #0x10\n"
"str q15, [x21, #0x0]\n"
"str q11, [x21, #0x10]\n"
"str q7, [x21, #0x20]\n"
"str q3, [x21, #0x30]\n"
"str q31, [x21, #0x40]\n"
"str q27, [x21, #0x50]\n"
"str q23, [x21, #0x60]\n"
"str q19, [x21, #0x70]\n"
"str q14, [x21, #0x80]\n"
"str q10, [x21, #0x90]\n"
"str q6, [x21, #0xa0]\n"
"str q2, [x21, #0xb0]\n"
"str q30, [x21, #0xc0]\n"
"str q26, [x21, #0xd0]\n"
"str q22, [x21, #0xe0]\n"
"str q18, [x21, #0xf0]\n"
"str q13, [x21, #0x100]\n"
"str q9, [x21, #0x110]\n"
"str q5, [x21, #0x120]\n"
"str q1, [x21, #0x130]\n"
"str q29, [x21, #0x140]\n"
"str q25, [x21, #0x150]\n"
"str q21, [x21, #0x160]\n"
"str q17, [x21, #0x170]\n"
"str q12, [x21, #0x180]\n"
"str q8, [x21, #0x190]\n"
"str q4, [x21, #0x1a0]\n"
"str q0, [x21, #0x1b0]\n"
"str q28, [x21, #0x1c0]\n"
"str q24, [x21, #0x1d0]\n"
"str q20, [x21, #0x1e0]\n"
"str q16, [x21, #0x1f0]\n"
"add x21, x21, %x[out_stride]\n"
"bge 2b\n"
"3:" // Main row loop: Column loop skip
"cmp x24, #0x10\n"
"blt 5f\n"
"4:" // Main row loop: width 16 loop: loop
"ldr q31, [x25], #0x10\n"
"ldr q30, [x23], #0x10\n"
"sub x24, x24, #0x10\n"
"cmp x24, #0x10\n"
"ldr q29, [x22], #0x10\n"
"ldr q28, [x20], #0x10\n"
"ldr q27, [x25], #0x10\n"
"ldr q26, [x23], #0x10\n"
"ldr q25, [x22], #0x10\n"
"ldr q24, [x20], #0x10\n"
"ldr q23, [x25], #0x10\n"
"ldr q22, [x23], #0x10\n"
"ldr q21, [x22], #0x10\n"
"ldr q20, [x20], #0x10\n"
"ldr q19, [x25], #0x10\n"
"ldr q18, [x23], #0x10\n"
"ldr q17, [x22], #0x10\n"
"ldr q16, [x20], #0x10\n"
"str q31, [x21, #0x0]\n"
"str q27, [x21, #0x10]\n"
"str q23, [x21, #0x20]\n"
"str q19, [x21, #0x30]\n"
"str q30, [x21, #0x80]\n"
"str q26, [x21, #0x90]\n"
"str q22, [x21, #0xa0]\n"
"str q18, [x21, #0xb0]\n"
"str q29, [x21, #0x100]\n"
"str q25, [x21, #0x110]\n"
"str q21, [x21, #0x120]\n"
"str q17, [x21, #0x130]\n"
"str q28, [x21, #0x180]\n"
"str q24, [x21, #0x190]\n"
"str q20, [x21, #0x1a0]\n"
"str q16, [x21, #0x1b0]\n"
"add x21, x21, #0x40\n"
"bge 4b\n"
"5:" // Main row loop: width 16 loop: skip
"cmp x24, #0x4\n"
"blt 7f\n"
"6:" // Main row loop: width 4 loop: loop
"ldr q19, [x25], #0x10\n"
"ldr q18, [x23], #0x10\n"
"sub x24, x24, #0x4\n"
"cmp x24, #0x4\n"
"ldr q17, [x22], #0x10\n"
"ldr q16, [x20], #0x10\n"
"str q19, [x21, #0x0]\n"
"str q18, [x21, #0x80]\n"
"str q17, [x21, #0x100]\n"
"str q16, [x21, #0x180]\n"
"add x21, x21, #0x10\n"
"bge 6b\n"
"7:" // Main row loop: width 4 loop: skip
"cmp x24, #0x1\n"
"blt 9f\n"
"8:" // Main row loop: width 1 loop: loop
"ldr s19, [x25], #0x4\n"
"ldr s18, [x23], #0x4\n"
"sub x24, x24, #0x1\n"
"cmp x24, #0x1\n"
"ldr s17, [x22], #0x4\n"
"ldr s16, [x20], #0x4\n"
"str s19, [x21, #0x0]\n"
"str s18, [x21, #0x80]\n"
"str s17, [x21, #0x100]\n"
"str s16, [x21, #0x180]\n"
"add x21, x21, #0x4\n"
"bge 8b\n"
"9:" // Main row loop: width 1 loop: skip
"cmp %x[height], #0x4\n"
"add %x[out], %x[out], #0x200\n"
"bge 1b\n"
"cbz %x[height], 20f\n"
"10:" // Main loop skip
"11:" // Tail row loop: Head
"mov x20, %x[width]\n"
"mov x25, %x[in]\n"
"cmp x20, #0x20\n"
"add %x[in], x25, %x[in_stride]\n"
"mov x21, %x[out]\n"
"sub %x[height], %x[height], #0x1\n"
"blt 13f\n"
"12:" // Tail row loop: Column loop
"ldr q23, [x25], #0x10\n"
"ldr q22, [x25], #0x10\n"
"sub x20, x20, #0x20\n"
"cmp x20, #0x20\n"
"ldr q21, [x25], #0x10\n"
"ldr q20, [x25], #0x10\n"
"ldr q19, [x25], #0x10\n"
"ldr q18, [x25], #0x10\n"
"ldr q17, [x25], #0x10\n"
"ldr q16, [x25], #0x10\n"
"str q23, [x21, #0x0]\n"
"str q22, [x21, #0x10]\n"
"str q21, [x21, #0x20]\n"
"str q20, [x21, #0x30]\n"
"str q19, [x21, #0x40]\n"
"str q18, [x21, #0x50]\n"
"str q17, [x21, #0x60]\n"
"str q16, [x21, #0x70]\n"
"add x21, x21, %x[out_stride]\n"
"bge 12b\n"
"13:" // Tail row loop: Column loop skip
"cmp x20, #0x10\n"
"blt 15f\n"
"14:" // Tail row loop: width 16 loop: loop
"ldr q19, [x25], #0x10\n"
"ldr q18, [x25], #0x10\n"
"sub x20, x20, #0x10\n"
"cmp x20, #0x10\n"
"ldr q17, [x25], #0x10\n"
"ldr q16, [x25], #0x10\n"
"str q19, [x21, #0x0]\n"
"str q18, [x21, #0x10]\n"
"str q17, [x21, #0x20]\n"
"str q16, [x21, #0x30]\n"
"add x21, x21, #0x40\n"
"bge 14b\n"
"15:" // Tail row loop: width 16 loop: skip
"cmp x20, #0x4\n"
"blt 17f\n"
"16:" // Tail row loop: width 4 loop: loop
"ldr q16, [x25], #0x10\n"
"sub x20, x20, #0x4\n"
"cmp x20, #0x4\n"
"str q16, [x21, #0x0]\n"
"add x21, x21, #0x10\n"
"bge 16b\n"
"17:" // Tail row loop: width 4 loop: skip
"cmp x20, #0x1\n"
"blt 19f\n"
"18:" // Tail row loop: width 1 loop: loop
"ldr s16, [x25], #0x4\n"
"sub x20, x20, #0x1\n"
"cmp x20, #0x1\n"
"str s16, [x21, #0x0]\n"
"add x21, x21, #0x4\n"
"bge 18b\n"
"19:" // Tail row loop: width 1 loop: skip
"cmp %x[height], #0x1\n"
"add %x[out], %x[out], #0x80\n"
"bge 11b\n"
"20:" // Done
: [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out)
: [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [width] "r" (width)
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25"
);
}
} // anonymous namespace
template<>
void Transform<32, 1, true, VLType::None>(
float *out, const float *in, int stride, int x0, int xmax, int k0, int kmax)
{
a64_transpose_interleave_128(
reinterpret_cast<uint32_t *>(out),
reinterpret_cast<const uint32_t *>(in + k0 * stride + x0),
(xmax-x0) * sizeof(float) / 4,
stride * sizeof(float),
(kmax-k0)
);
}
#endif // defined(__aarch64__)
|
