1 | /* |
2 | * memcmp - compare memory |
3 | * |
4 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
5 | * See https://llvm.org/LICENSE.txt for license information. |
6 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
7 | */ |
8 | |
9 | #if __ARM_FEATURE_SVE |
10 | /* Assumptions: |
11 | * |
12 | * ARMv8-a, AArch64 |
13 | * SVE Available. |
14 | */ |
15 | |
16 | .arch armv8-a+sve |
17 | .text |
18 | |
19 | .globl __memcmp_aarch64_sve |
20 | .type __memcmp_aarch64_sve, %function |
21 | .p2align 4 |
22 | __memcmp_aarch64_sve: |
23 | mov x3, 0 /* initialize off */ |
24 | |
25 | 0: whilelo p0.b, x3, x2 /* while off < max */ |
26 | b.none 9f |
27 | |
28 | ld1b z0.b, p0/z, [x0, x3] /* read vectors bounded by max. */ |
29 | ld1b z1.b, p0/z, [x1, x3] |
30 | |
31 | /* Increment for a whole vector, even if we've only read a partial. |
32 | This is significantly cheaper than INCP, and since OFF is not |
33 | used after the loop it is ok to increment OFF past MAX. */ |
34 | incb x3 |
35 | |
36 | cmpne p1.b, p0/z, z0.b, z1.b /* while no inequalities */ |
37 | b.none 0b |
38 | |
39 | /* Found inequality. */ |
40 | 1: brkb p1.b, p0/z, p1.b /* find first such */ |
41 | lasta w0, p1, z0.b /* extract each byte */ |
42 | lasta w1, p1, z1.b |
43 | sub x0, x0, x1 /* return comparison */ |
44 | ret |
45 | |
46 | /* Found end-of-count. */ |
47 | 9: mov x0, 0 /* return equality */ |
48 | ret |
49 | |
50 | .size __memcmp_aarch64_sve, . - __memcmp_aarch64_sve |
51 | #endif |
52 | |