1 /*
2 * Copyright (c) 2019-2021 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25 #ifdef __aarch64__
26
27 template<>
interleave_block(int16_t * & out_ptr,const int8_t * const * in,size_t width,size_t height,size_t row_offset,bool)28 void interleave_block<8, 1, VLType::None, false>(
29 int16_t * &out_ptr, const int8_t * const * in, size_t width, size_t height,
30 size_t row_offset, bool
31 )
32 {
33 __asm__ __volatile__(
34 "ldr x27, [%x[in], #0x0]\n"
35 "cmp %x[height], #0x8\n"
36 "ldr x26, [%x[in], #0x8]\n"
37 "add x27, x27, %x[row_offset]\n"
38 "ldr x25, [%x[in], #0x10]\n"
39 "ldr x24, [%x[in], #0x18]\n"
40 "add x26, x26, %x[row_offset]\n"
41 "ldr x23, [%x[in], #0x20]\n"
42 "add x25, x25, %x[row_offset]\n"
43 "ldr x22, [%x[in], #0x28]\n"
44 "ldr x21, [%x[in], #0x30]\n"
45 "add x24, x24, %x[row_offset]\n"
46 "ldr x20, [%x[in], #0x38]\n"
47 "add x23, x23, %x[row_offset]\n"
48 "add x22, x22, %x[row_offset]\n"
49 "add x21, x21, %x[row_offset]\n"
50 "add x20, x20, %x[row_offset]\n"
51 "beq 1f\n"
52 "mov x20, x27\n"
53 "cmp %x[height], #0x2\n"
54 "csel x26, x26, x27, GE\n"
55 "csel x25, x25, x27, GT\n"
56 "cmp %x[height], #0x4\n"
57 "csel x24, x24, x27, GE\n"
58 "csel x23, x23, x27, GT\n"
59 "cmp %x[height], #0x6\n"
60 "csel x22, x22, x27, GE\n"
61 "csel x21, x21, x27, GT\n"
62 "1:" // no_pointer_adj
63 "prfm pldl1keep, [x27, #0x0]\n"
64 "cmp %x[width], #0x8\n"
65 "prfm pldl1keep, [x26, #0x0]\n"
66 "prfm pldl1keep, [x25, #0x0]\n"
67 "prfm pldl1keep, [x24, #0x0]\n"
68 "prfm pldl1keep, [x23, #0x0]\n"
69 "prfm pldl1keep, [x22, #0x0]\n"
70 "prfm pldl1keep, [x21, #0x0]\n"
71 "prfm pldl1keep, [x20, #0x0]\n"
72 "prfm pldl1keep, [x27, #0x40]\n"
73 "prfm pldl1keep, [x26, #0x40]\n"
74 "prfm pldl1keep, [x25, #0x40]\n"
75 "prfm pldl1keep, [x24, #0x40]\n"
76 "prfm pldl1keep, [x23, #0x40]\n"
77 "prfm pldl1keep, [x22, #0x40]\n"
78 "prfm pldl1keep, [x21, #0x40]\n"
79 "prfm pldl1keep, [x20, #0x40]\n"
80 "blt 3f\n"
81 "2:" // Main loop head
82 "ldr d31, [x27], #0x8\n"
83 "sshll v31.8h, v31.8b, #0x0\n"
84 "ldr d30, [x26], #0x8\n"
85 "subs %x[width], %x[width], #0x8\n"
86 "sshll v30.8h, v30.8b, #0x0\n"
87 "ldr d29, [x25], #0x8\n"
88 "cmp %x[width], #0x8\n"
89 "sshll v29.8h, v29.8b, #0x0\n"
90 "ldr d28, [x24], #0x8\n"
91 "ldr d25, [x23], #0x8\n"
92 "sshll v28.8h, v28.8b, #0x0\n"
93 "ldr d23, [x22], #0x8\n"
94 "sshll v25.8h, v25.8b, #0x0\n"
95 "ldr d27, [x21], #0x8\n"
96 "zip1 v20.8h, v31.8h, v25.8h\n"
97 "ldr d26, [x20], #0x8\n"
98 "zip2 v25.8h, v31.8h, v25.8h\n"
99 "prfm pldl1keep, [x27, #0x70]\n"
100 "prfm pldl1keep, [x26, #0x70]\n"
101 "sshll v23.8h, v23.8b, #0x0\n"
102 "prfm pldl1keep, [x25, #0x70]\n"
103 "zip1 v24.8h, v30.8h, v23.8h\n"
104 "zip2 v23.8h, v30.8h, v23.8h\n"
105 "prfm pldl1keep, [x24, #0x70]\n"
106 "sshll v27.8h, v27.8b, #0x0\n"
107 "prfm pldl1keep, [x23, #0x70]\n"
108 "zip1 v19.8h, v29.8h, v27.8h\n"
109 "prfm pldl1keep, [x22, #0x70]\n"
110 "zip1 v22.8h, v20.8h, v19.8h\n"
111 "prfm pldl1keep, [x21, #0x70]\n"
112 "zip2 v21.8h, v20.8h, v19.8h\n"
113 "prfm pldl1keep, [x20, #0x70]\n"
114 "zip2 v19.8h, v29.8h, v27.8h\n"
115 "zip1 v20.8h, v25.8h, v19.8h\n"
116 "zip2 v19.8h, v25.8h, v19.8h\n"
117 "sshll v26.8h, v26.8b, #0x0\n"
118 "zip1 v18.8h, v28.8h, v26.8h\n"
119 "zip1 v17.8h, v24.8h, v18.8h\n"
120 "zip1 v16.8h, v22.8h, v17.8h\n"
121 "str q16, [%x[out_ptr], #0x0]\n"
122 "zip2 v16.8h, v22.8h, v17.8h\n"
123 "str q16, [%x[out_ptr], #0x10]\n"
124 "zip2 v17.8h, v24.8h, v18.8h\n"
125 "zip1 v16.8h, v21.8h, v17.8h\n"
126 "str q16, [%x[out_ptr], #0x20]\n"
127 "zip2 v16.8h, v21.8h, v17.8h\n"
128 "str q16, [%x[out_ptr], #0x30]\n"
129 "zip2 v18.8h, v28.8h, v26.8h\n"
130 "zip1 v17.8h, v23.8h, v18.8h\n"
131 "zip1 v16.8h, v20.8h, v17.8h\n"
132 "str q16, [%x[out_ptr], #0x40]\n"
133 "zip2 v16.8h, v20.8h, v17.8h\n"
134 "str q16, [%x[out_ptr], #0x50]\n"
135 "zip2 v17.8h, v23.8h, v18.8h\n"
136 "zip1 v16.8h, v19.8h, v17.8h\n"
137 "str q16, [%x[out_ptr], #0x60]\n"
138 "zip2 v16.8h, v19.8h, v17.8h\n"
139 "str q16, [%x[out_ptr], #0x70]\n"
140 "add %x[out_ptr], %x[out_ptr], #0x80\n"
141 "bge 2b\n"
142 "3:" // Main loop skip
143 "cbz %x[width], 8f\n"
144 "tbz %x[width], #2, 5f\n"
145 "ldr s31, [x27], #0x4\n"
146 "ldr s30, [x26], #0x4\n"
147 "ldr s29, [x25], #0x4\n"
148 "ldr s28, [x24], #0x4\n"
149 "ldr s25, [x23], #0x4\n"
150 "ldr s23, [x22], #0x4\n"
151 "ldr s27, [x21], #0x4\n"
152 "ldr s26, [x20], #0x4\n"
153 "tbz %x[width], #1, 4f\n"
154 "ld1 { v31.h }[2], [x27], #0x2\n"
155 "mov x19, #0x6\n"
156 "ld1 { v30.h }[2], [x26], #0x2\n"
157 "ld1 { v29.h }[2], [x25], #0x2\n"
158 "ld1 { v28.h }[2], [x24], #0x2\n"
159 "ld1 { v25.h }[2], [x23], #0x2\n"
160 "ld1 { v23.h }[2], [x22], #0x2\n"
161 "ld1 { v27.h }[2], [x21], #0x2\n"
162 "ld1 { v26.h }[2], [x20], #0x2\n"
163 "tbz %x[width], #0, 7f\n"
164 "ld1 { v31.b }[6], [x27]\n"
165 "mov x19, #0x7\n"
166 "ld1 { v30.b }[6], [x26]\n"
167 "ld1 { v29.b }[6], [x25]\n"
168 "ld1 { v28.b }[6], [x24]\n"
169 "ld1 { v25.b }[6], [x23]\n"
170 "ld1 { v23.b }[6], [x22]\n"
171 "ld1 { v27.b }[6], [x21]\n"
172 "ld1 { v26.b }[6], [x20]\n"
173 "b 7f\n"
174 "4:" // odd_loads_1_4
175 "mov x19, #0x4\n"
176 "tbz %x[width], #0, 7f\n"
177 "ld1 { v31.b }[4], [x27]\n"
178 "ld1 { v30.b }[4], [x26]\n"
179 "mov x19, #0x5\n"
180 "ld1 { v29.b }[4], [x25]\n"
181 "ld1 { v28.b }[4], [x24]\n"
182 "ld1 { v25.b }[4], [x23]\n"
183 "ld1 { v23.b }[4], [x22]\n"
184 "ld1 { v27.b }[4], [x21]\n"
185 "ld1 { v26.b }[4], [x20]\n"
186 "b 7f\n"
187 "5:" // odd_loads_2_0
188 "tbz %x[width], #1, 6f\n"
189 "ldr h31, [x27], #0x2\n"
190 "ldr h30, [x26], #0x2\n"
191 "mov x19, #0x2\n"
192 "ldr h29, [x25], #0x2\n"
193 "ldr h28, [x24], #0x2\n"
194 "ldr h25, [x23], #0x2\n"
195 "ldr h23, [x22], #0x2\n"
196 "ldr h27, [x21], #0x2\n"
197 "ldr h26, [x20], #0x2\n"
198 "tbz %x[width], #0, 7f\n"
199 "ld1 { v31.b }[2], [x27]\n"
200 "mov x19, #0x3\n"
201 "ld1 { v30.b }[2], [x26]\n"
202 "ld1 { v29.b }[2], [x25]\n"
203 "ld1 { v28.b }[2], [x24]\n"
204 "ld1 { v25.b }[2], [x23]\n"
205 "ld1 { v23.b }[2], [x22]\n"
206 "ld1 { v27.b }[2], [x21]\n"
207 "ld1 { v26.b }[2], [x20]\n"
208 "b 7f\n"
209 "6:" // odd_loads_1_0
210 "ldr b31, [x27, #0x0]\n"
211 "mov x19, #0x1\n"
212 "ldr b30, [x26, #0x0]\n"
213 "ldr b29, [x25, #0x0]\n"
214 "ldr b28, [x24, #0x0]\n"
215 "ldr b25, [x23, #0x0]\n"
216 "ldr b23, [x22, #0x0]\n"
217 "ldr b27, [x21, #0x0]\n"
218 "ldr b26, [x20, #0x0]\n"
219 "7:" // Odd load end
220 "sshll v31.8h, v31.8b, #0x0\n"
221 "subs x19, x19, #0x1\n"
222 "sshll v30.8h, v30.8b, #0x0\n"
223 "sshll v29.8h, v29.8b, #0x0\n"
224 "sshll v28.8h, v28.8b, #0x0\n"
225 "sshll v25.8h, v25.8b, #0x0\n"
226 "zip1 v20.8h, v31.8h, v25.8h\n"
227 "sshll v23.8h, v23.8b, #0x0\n"
228 "zip1 v24.8h, v30.8h, v23.8h\n"
229 "sshll v27.8h, v27.8b, #0x0\n"
230 "zip1 v19.8h, v29.8h, v27.8h\n"
231 "zip1 v22.8h, v20.8h, v19.8h\n"
232 "sshll v26.8h, v26.8b, #0x0\n"
233 "zip1 v18.8h, v28.8h, v26.8h\n"
234 "zip1 v17.8h, v24.8h, v18.8h\n"
235 "zip1 v16.8h, v22.8h, v17.8h\n"
236 "str q16, [%x[out_ptr], #0x0]\n"
237 "add %x[out_ptr], %x[out_ptr], #0x10\n"
238 "beq 8f\n"
239 "zip2 v16.8h, v22.8h, v17.8h\n"
240 "str q16, [%x[out_ptr], #0x0]\n"
241 "subs x19, x19, #0x1\n"
242 "add %x[out_ptr], %x[out_ptr], #0x10\n"
243 "beq 8f\n"
244 "zip2 v21.8h, v20.8h, v19.8h\n"
245 "zip2 v17.8h, v24.8h, v18.8h\n"
246 "subs x19, x19, #0x1\n"
247 "zip1 v16.8h, v21.8h, v17.8h\n"
248 "str q16, [%x[out_ptr], #0x0]\n"
249 "add %x[out_ptr], %x[out_ptr], #0x10\n"
250 "beq 8f\n"
251 "zip2 v16.8h, v21.8h, v17.8h\n"
252 "str q16, [%x[out_ptr], #0x0]\n"
253 "subs x19, x19, #0x1\n"
254 "add %x[out_ptr], %x[out_ptr], #0x10\n"
255 "beq 8f\n"
256 "zip2 v25.8h, v31.8h, v25.8h\n"
257 "zip2 v19.8h, v29.8h, v27.8h\n"
258 "subs x19, x19, #0x1\n"
259 "zip1 v20.8h, v25.8h, v19.8h\n"
260 "zip2 v23.8h, v30.8h, v23.8h\n"
261 "zip2 v18.8h, v28.8h, v26.8h\n"
262 "zip1 v17.8h, v23.8h, v18.8h\n"
263 "zip1 v16.8h, v20.8h, v17.8h\n"
264 "str q16, [%x[out_ptr], #0x0]\n"
265 "add %x[out_ptr], %x[out_ptr], #0x10\n"
266 "beq 8f\n"
267 "zip2 v16.8h, v20.8h, v17.8h\n"
268 "str q16, [%x[out_ptr], #0x0]\n"
269 "subs x19, x19, #0x1\n"
270 "add %x[out_ptr], %x[out_ptr], #0x10\n"
271 "beq 8f\n"
272 "zip2 v19.8h, v25.8h, v19.8h\n"
273 "zip2 v17.8h, v23.8h, v18.8h\n"
274 "zip1 v16.8h, v19.8h, v17.8h\n"
275 "str q16, [%x[out_ptr], #0x0]\n"
276 "add %x[out_ptr], %x[out_ptr], #0x10\n"
277 "8:" // Odds skip
278
279 : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width)
280 : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset)
281 : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27"
282 );
283 }
284
285
286 #endif // __aarch64__
287