1 /*
2  * Copyright (c) 2022 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 #if defined(__ARM_FEATURE_SVE)
26 
27 template <>
interleave_block(uint8_t * & out,const uint8_t * const * in,size_t width,size_t height,size_t row_offset,bool first)28 void interleave_block<2, 4, VLType::SME, true>(
29   uint8_t * &out, const uint8_t * const *in,
30   size_t width, size_t height, size_t row_offset, bool first
31 )
32 {
33   __asm__ __volatile__(
34       ".inst 0xd503477f  // SMSTART ZA\n"
35       "mov z20.b, #0x1\n"
36       "mov z19.s, #0x0\n"
37       "cntb x20\n"
38       "mov z18.s, #0x0\n"
39       "cntw x16\n"
40       "cntw x15, ALL, MUL #2\n"
41       "cntw x14, ALL, MUL #3\n"
42       "ptrue p2.b\n"
43       "mov x19, %x[width]\n"
44       "incb x19\n"
45       "sub x19, x19, #0x1\n"
46       "udiv x19, x19, x20\n" // n_passes = ceildiv(width, VL<T>)
47       "sub x13, x19, #0x1\n"
48       "lsr x13, x13, #0x1\n" // n_loops = (n_passes - 1) / 2
49       "and x11, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1)
50       "mov x19, %x[width]\n"
51       "sub x10, x20, #0x1\n"
52       "ands x10, x19, x10\n"
53       "csel x10, x10, x20, NE\n"
54       "add x10, x10, #0x3\n"
55       "lsr x10, x10, #0x2\n"
56       "sub x9, x16, #0x2\n"
57       "ptrue p11.s\n"
58       "lsl x20, %x[height], #0x1\n" // height * 2
59       "lsl x19, x16, #0x1\n"
60       "whilelt p9.b, XZR, x20\n"
61       "whilelt p8.b, x19, x20\n"
62       "zip1 p10.b, p9.b, p8.b\n"
63       "mov x28, %x[row_offset]\n"
64       "mov x27, %x[out]\n"
65       "mov x26, #0x0\n"
66       "whilelt p9.b, x26, %x[width]\n"
67       "whilelt p8.b, x26, %x[width]\n"
68       "cbnz %x[first], 1f\n"
69       "addvl x27, x27, #-2\n"
70       "ld1w { z19.s }, p2/Z, [x27]\n"
71       "ld1w { z18.s }, p2/Z, [x27, #1, MUL VL]\n"
72       "1:"  // K loop: Load row sums: End
73       "mov x25, %x[in]\n"
74       "add x24, %x[in], x16, LSL #3\n"
75       "ldr x23, [x25, #0x0]\n"
76       "ldr x22, [x24, #0x0]\n"
77       "ldr x21, [x25, #0x8]\n"
78       "ldr x20, [x24, #0x8]\n"
79       "add x25, x25, #0x10\n"
80       "add x24, x24, #0x10\n"
81       "mov x12, #0x0\n"
82       "cbz x9, 3f\n"
83       "2:"  // K loop: Charge: Loop
84       ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
85       ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
86       ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
87       ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
88       ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
89       ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
90       "ldr x23, [x25, #0x0]\n"
91       ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
92       "ldr x22, [x24, #0x0]\n"
93       "ldr x21, [x25, #0x8]\n"
94       ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
95       "ldr x20, [x24, #0x8]\n"
96       "add x25, x25, #0x10\n"
97       "add x24, x24, #0x10\n"
98       "add x12, x12, #0x8\n"
99       "cmp x12, x9, LSL #2\n"
100       "blt 2b\n"
101       "3:"  // K loop: Charge: End
102       ".inst 0x25246140  // dup p0.b, p8.b/Z, p10.b[w12]\n"
103       ".inst 0xe01c02e0  // ld1b { za0h.b[x12] }, p0/Z, [x23, x28]\n"
104       ".inst 0x252c6140  // dup p0.b, p8.b/Z, p10.b[w12, #1]\n"
105       ".inst 0x25646141  // dup p1.b, p8.b/Z, p10.b[w12, #4]\n"
106       ".inst 0xe01c02c1  // ld1b { za0h.b[x12, #1] }, p0/Z, [x22, x28]\n"
107       ".inst 0x256c6140  // dup p0.b, p8.b/Z, p10.b[w12, #5]\n"
108       "mov x25, %x[in]\n"
109       ".inst 0xe01c06a4  // ld1b { za0h.b[x12, #4] }, p1/Z, [x21, x28]\n"
110       "add x24, %x[in], x16, LSL #3\n"
111       "ldr x23, [x25, #0x0]\n"
112       ".inst 0xe01c0285  // ld1b { za0h.b[x12, #5] }, p0/Z, [x20, x28]\n"
113       "ldr x22, [x24, #0x0]\n"
114       "ldr x21, [x25, #0x8]\n"
115       "ldr x20, [x24, #0x8]\n"
116       "add x25, x25, #0x10\n"
117       "add x24, x24, #0x10\n"
118       "incb x28\n"
119       "incb x26\n"
120       "cbz x13, 9f\n"
121       "mov x19, x13\n"
122       "4:"  // K loop: Main loop
123       "whilelt p8.b, x26, %x[width]\n"
124       "mov x13, #0x0\n"
125       "mov x12, #0x0\n"
126       "cbz x9, 6f\n"
127       "5:"  // K loop: Main loop: First: Loop
128       ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
129       ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
130       ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
131       ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
132       ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
133       ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
134       "ldr x23, [x25, #0x0]\n"
135       ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
136       "ldr x22, [x24, #0x0]\n"
137       "ldr x21, [x25, #0x8]\n"
138       ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
139       "ldr x20, [x24, #0x8]\n"
140       "add x25, x25, #0x10\n"
141       ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
142       ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
143       "add x24, x24, #0x10\n"
144       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
145       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
146       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
147       "udot z19.s, z17.b, z20.b\n"
148       ".inst 0xc0828831  // mova z17.s, p2/M, za0v.s[x12, #1]\n"
149       "udot z18.s, z16.b, z20.b\n"
150       ".inst 0xc08288b0  // mova z16.s, p2/M, za1v.s[x12, #1]\n"
151       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
152       ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
153       ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
154       "add x13, x13, #0x8\n"
155       ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
156       "udot z19.s, z17.b, z20.b\n"
157       "udot z18.s, z16.b, z20.b\n"
158       ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
159       "addvl x27, x27, #4\n"
160       "add x12, x12, #0x2\n"
161       "cmp x12, x9\n"
162       "blt 5b\n"
163       "6:"  // K loop: Main loop: First: Tail
164       "mov x25, %x[in]\n"
165       "add x24, %x[in], x16, LSL #3\n"
166       ".inst 0x25356140  // dup p0.b, p8.b/Z, p10.b[w13, #2]\n"
167       ".inst 0xe01c22e2  // ld1b { za0h.b[x13, #2] }, p0/Z, [x23, x28]\n"
168       ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
169       ".inst 0x25756141  // dup p1.b, p8.b/Z, p10.b[w13, #6]\n"
170       ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
171       ".inst 0x257d6140  // dup p0.b, p8.b/Z, p10.b[w13, #7]\n"
172       "ldr x23, [x25, #0x0]\n"
173       ".inst 0xe01c26a6  // ld1b { za0h.b[x13, #6] }, p1/Z, [x21, x28]\n"
174       "ldr x22, [x24, #0x0]\n"
175       "ldr x21, [x25, #0x8]\n"
176       ".inst 0xe01c2287  // ld1b { za0h.b[x13, #7] }, p0/Z, [x20, x28]\n"
177       "ldr x20, [x24, #0x8]\n"
178       "add x25, x25, #0x10\n"
179       ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
180       ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
181       "add x24, x24, #0x10\n"
182       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
183       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
184       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
185       "udot z19.s, z17.b, z20.b\n"
186       ".inst 0xc0828831  // mova z17.s, p2/M, za0v.s[x12, #1]\n"
187       "udot z18.s, z16.b, z20.b\n"
188       ".inst 0xc08288b0  // mova z16.s, p2/M, za1v.s[x12, #1]\n"
189       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
190       ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
191       ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
192       "whilelt p9.b, x26, %x[width]\n"
193       ".inst 0xe0af8761  // st1w { za0v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
194       "udot z19.s, z17.b, z20.b\n"
195       "incb x26\n"
196       "udot z18.s, z16.b, z20.b\n"
197       "incb x28\n"
198       ".inst 0xe0ae8365  // st1w { za1v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
199       "addvl x27, x27, #4\n"
200       "whilelt p8.b, x26, %x[width]\n"
201       "mov x13, #0x0\n"
202       "mov x12, #0x0\n"
203       "cbz x9, 8f\n"
204       "7:"  // K loop: Main loop: Second: Loop
205       ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
206       ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
207       ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
208       ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
209       ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
210       ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
211       "ldr x23, [x25, #0x0]\n"
212       ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
213       "ldr x22, [x24, #0x0]\n"
214       "ldr x21, [x25, #0x8]\n"
215       ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
216       "ldr x20, [x24, #0x8]\n"
217       "add x25, x25, #0x10\n"
218       ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
219       ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
220       "add x24, x24, #0x10\n"
221       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
222       ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
223       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
224       "udot z19.s, z17.b, z20.b\n"
225       ".inst 0xc0828931  // mova z17.s, p2/M, za2v.s[x12, #1]\n"
226       "udot z18.s, z16.b, z20.b\n"
227       ".inst 0xc08289b0  // mova z16.s, p2/M, za3v.s[x12, #1]\n"
228       ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
229       ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
230       ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
231       "add x13, x13, #0x8\n"
232       ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
233       "udot z19.s, z17.b, z20.b\n"
234       "udot z18.s, z16.b, z20.b\n"
235       ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
236       "addvl x27, x27, #4\n"
237       "add x12, x12, #0x2\n"
238       "cmp x12, x9\n"
239       "blt 7b\n"
240       "8:"  // K loop: Main loop: Second: Tail
241       "mov x25, %x[in]\n"
242       "add x24, %x[in], x16, LSL #3\n"
243       ".inst 0x25256140  // dup p0.b, p8.b/Z, p10.b[w13]\n"
244       ".inst 0xe01c22e0  // ld1b { za0h.b[x13] }, p0/Z, [x23, x28]\n"
245       ".inst 0x252d6140  // dup p0.b, p8.b/Z, p10.b[w13, #1]\n"
246       ".inst 0x25656141  // dup p1.b, p8.b/Z, p10.b[w13, #4]\n"
247       ".inst 0xe01c22c1  // ld1b { za0h.b[x13, #1] }, p0/Z, [x22, x28]\n"
248       ".inst 0x256d6140  // dup p0.b, p8.b/Z, p10.b[w13, #5]\n"
249       "ldr x23, [x25, #0x0]\n"
250       ".inst 0xe01c26a4  // ld1b { za0h.b[x13, #4] }, p1/Z, [x21, x28]\n"
251       "ldr x22, [x24, #0x0]\n"
252       "ldr x21, [x25, #0x8]\n"
253       ".inst 0xe01c2285  // ld1b { za0h.b[x13, #5] }, p0/Z, [x20, x28]\n"
254       "ldr x20, [x24, #0x8]\n"
255       "add x25, x25, #0x10\n"
256       ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
257       ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
258       "add x24, x24, #0x10\n"
259       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
260       ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
261       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
262       "udot z19.s, z17.b, z20.b\n"
263       ".inst 0xc0828931  // mova z17.s, p2/M, za2v.s[x12, #1]\n"
264       "udot z18.s, z16.b, z20.b\n"
265       ".inst 0xc08289b0  // mova z16.s, p2/M, za3v.s[x12, #1]\n"
266       ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
267       ".inst 0x25706d21  // dup p1.s, p11.s/Z, p9.s[w12, #1]\n"
268       ".inst 0x25706d20  // dup p0.s, p11.s/Z, p9.s[w12, #1]\n"
269       "whilelt p9.b, x26, %x[width]\n"
270       ".inst 0xe0af8769  // st1w { za2v.s[x12, #1] }, p1/Z, [x27, x15, LSL #2]\n"
271       "udot z19.s, z17.b, z20.b\n"
272       "incb x26\n"
273       "udot z18.s, z16.b, z20.b\n"
274       "incb x28\n"
275       ".inst 0xe0ae836d  // st1w { za3v.s[x12, #1] }, p0/Z, [x27, x14, LSL #2]\n"
276       "addvl x27, x27, #4\n"
277       "subs x19, x19, #0x1\n"
278       "bgt 4b\n"
279       "9:"  // K loop: Tails
280       "cbnz x11, 12f\n"
281       "mov x25, %x[in]\n"
282       "whilelt p8.b, x26, %x[width]\n"
283       "mov x13, #0x0\n"
284       "mov x12, #0x0\n"
285       "10:"  // K loop: Tails: Even: First
286       ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
287       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
288       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
289       ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
290       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
291       ".inst 0x25356141  // dup p1.b, p8.b/Z, p10.b[w13, #2]\n"
292       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
293       ".inst 0x253d6140  // dup p0.b, p8.b/Z, p10.b[w13, #3]\n"
294       "udot z19.s, z17.b, z20.b\n"
295       "udot z18.s, z16.b, z20.b\n"
296       "ldr x23, [x25, #0x0]\n"
297       ".inst 0xe01c26e2  // ld1b { za0h.b[x13, #2] }, p1/Z, [x23, x28]\n"
298       "ldr x22, [x25, x16, LSL #0x3]\n"
299       "addvl x27, x27, #2\n"
300       ".inst 0xe01c22c3  // ld1b { za0h.b[x13, #3] }, p0/Z, [x22, x28]\n"
301       "add x25, x25, #0x8\n"
302       "add x13, x13, #0x4\n"
303       "add x12, x12, #0x1\n"
304       "cmp x12, x16\n"
305       "blt 10b\n"
306       "whilelt p9.b, x26, %x[width]\n"
307       "whilelt p8.b, x26, %x[width]\n"
308       "mov x19, #0x0\n"
309       "mov x12, #0x0\n"
310       "11:"  // K loop: Tails: Even: Second
311       ".inst 0xc0828911  // mova z17.s, p2/M, za2v.s[x12]\n"
312       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
313       ".inst 0xe0bf8368  // st1w { za2v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
314       ".inst 0xc0828990  // mova z16.s, p2/M, za3v.s[x12]\n"
315       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
316       "add x19, x19, #0x4\n"
317       ".inst 0xe0b0836c  // st1w { za3v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
318       "addvl x27, x27, #2\n"
319       "udot z19.s, z17.b, z20.b\n"
320       "udot z18.s, z16.b, z20.b\n"
321       "add x12, x12, #0x1\n"
322       "cmp x12, x10\n"
323       "blt 11b\n"
324       "whilelt p9.b, x26, %x[width]\n"
325       "b 14f\n"
326       "12:"  // K loop: Tails: Odd
327       "mov x12, #0x0\n"
328       "13:"  // K loop: Tails: Odd: Loop
329       ".inst 0xc0828811  // mova z17.s, p2/M, za0v.s[x12]\n"
330       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
331       ".inst 0xe0bf8360  // st1w { za0v.s[x12] }, p0/Z, [x27, XZR, LSL #2]\n"
332       ".inst 0xc0828890  // mova z16.s, p2/M, za1v.s[x12]\n"
333       ".inst 0x25306d20  // dup p0.s, p11.s/Z, p9.s[w12]\n"
334       ".inst 0xe0b08364  // st1w { za1v.s[x12] }, p0/Z, [x27, x16, LSL #2]\n"
335       "addvl x27, x27, #2\n"
336       "udot z19.s, z17.b, z20.b\n"
337       "udot z18.s, z16.b, z20.b\n"
338       "add x12, x12, #0x1\n"
339       "cmp x12, x10\n"
340       "blt 13b\n"
341       "14:"  // K loop: End
342       "st1w { z19.s }, p2, [x27]\n"
343       "st1w { z18.s }, p2, [x27, #1, MUL VL]\n"
344       "addvl x27, x27, #2\n"
345       "mov %x[out], x27\n"
346       ".inst 0xd503467f  // SMSTOP\n"
347       : [out] "+&r" (out)
348       : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width)
349       : "cc", "memory", "p0", "p1", "p2", "p8", "p9", "p10", "p11", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
350     );
351 }
352 
353 #endif  // defined(__ARM_FEATURE_SVE)
354