1*412f47f9SXin Li/* 2*412f47f9SXin Li * memcmp - compare memory 3*412f47f9SXin Li * 4*412f47f9SXin Li * Copyright (c) 2018-2022, Arm Limited. 5*412f47f9SXin Li * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception 6*412f47f9SXin Li */ 7*412f47f9SXin Li 8*412f47f9SXin Li#include "asmdefs.h" 9*412f47f9SXin Li 10*412f47f9SXin Li#if __ARM_FEATURE_SVE 11*412f47f9SXin Li/* Assumptions: 12*412f47f9SXin Li * 13*412f47f9SXin Li * ARMv8-a, AArch64 14*412f47f9SXin Li * SVE Available. 15*412f47f9SXin Li */ 16*412f47f9SXin Li 17*412f47f9SXin LiENTRY (__memcmp_aarch64_sve) 18*412f47f9SXin Li PTR_ARG (0) 19*412f47f9SXin Li PTR_ARG (1) 20*412f47f9SXin Li SIZE_ARG (2) 21*412f47f9SXin Li mov x3, 0 /* initialize off */ 22*412f47f9SXin Li 23*412f47f9SXin Li0: whilelo p0.b, x3, x2 /* while off < max */ 24*412f47f9SXin Li b.none 9f 25*412f47f9SXin Li 26*412f47f9SXin Li ld1b z0.b, p0/z, [x0, x3] /* read vectors bounded by max. */ 27*412f47f9SXin Li ld1b z1.b, p0/z, [x1, x3] 28*412f47f9SXin Li 29*412f47f9SXin Li /* Increment for a whole vector, even if we've only read a partial. 30*412f47f9SXin Li This is significantly cheaper than INCP, and since OFF is not 31*412f47f9SXin Li used after the loop it is ok to increment OFF past MAX. */ 32*412f47f9SXin Li incb x3 33*412f47f9SXin Li 34*412f47f9SXin Li cmpne p1.b, p0/z, z0.b, z1.b /* while no inequalities */ 35*412f47f9SXin Li b.none 0b 36*412f47f9SXin Li 37*412f47f9SXin Li /* Found inequality. */ 38*412f47f9SXin Li1: brkb p1.b, p0/z, p1.b /* find first such */ 39*412f47f9SXin Li lasta w0, p1, z0.b /* extract each byte */ 40*412f47f9SXin Li lasta w1, p1, z1.b 41*412f47f9SXin Li sub x0, x0, x1 /* return comparison */ 42*412f47f9SXin Li ret 43*412f47f9SXin Li 44*412f47f9SXin Li /* Found end-of-count. */ 45*412f47f9SXin Li9: mov x0, 0 /* return equality */ 46*412f47f9SXin Li ret 47*412f47f9SXin Li 48*412f47f9SXin LiEND (__memcmp_aarch64_sve) 49*412f47f9SXin Li 50*412f47f9SXin Li#endif 51*412f47f9SXin Li 52