70 lines · plain
1/*2 * strncmp - compare two strings with limit3 *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-exception7 */8 9#if __ARM_FEATURE_SVE10/* Assumptions:11 *12 * ARMv8-a, AArch6413 * SVE Available.14 */15 16 .arch armv8-a+sve17 .text18 19 .globl __strncmp_aarch64_sve20 .type __strncmp_aarch64_sve, %function21 .p2align 422__strncmp_aarch64_sve:23 setffr /* initialize FFR */24 mov x3, 0 /* initialize off */25 260: whilelo p0.b, x3, x2 /* while off < max */27 b.none 9f28 29 ldff1b z0.b, p0/z, [x0, x3]30 ldff1b z1.b, p0/z, [x1, x3]31 rdffrs p1.b, p0/z32 b.nlast 2f33 34 /* First fault did not fail: the vector up to max is valid.35 Avoid depending on the contents of FFR beyond the branch.36 Increment for a whole vector, even if we've only read a partial.37 This is significantly cheaper than INCP, and since OFF is not38 used after the loop it is ok to increment OFF past MAX. */39 incb x340 cmpeq p1.b, p0/z, z0.b, z1.b /* compare strings */41 cmpne p2.b, p0/z, z0.b, 0 /* search for ~zero */42 nands p2.b, p0/z, p1.b, p2.b /* ~(eq & ~zero) -> ne | zero */43 b.none 0b44 45 /* Found end-of-string or inequality. */461: brkb p2.b, p0/z, p2.b /* find first such */47 lasta w0, p2, z0.b /* extract each char */48 lasta w1, p2, z1.b49 sub x0, x0, x1 /* return comparison */50 ret51 52 /* First fault failed: only some of the vector is valid.53 Perform the comparison only on the valid bytes. */542: cmpeq p2.b, p1/z, z0.b, z1.b /* compare strings, as above */55 cmpne p3.b, p1/z, z0.b, 056 nands p2.b, p1/z, p2.b, p3.b57 b.any 1b58 59 /* No inequality or zero found. Re-init FFR, incr and loop. */60 setffr61 incp x3, p1.b62 b 0b63 64 /* Found end-of-count. */659: mov x0, 0 /* return equal */66 ret67 68 .size __strncmp_aarch64_sve, . - __strncmp_aarch64_sve69#endif70