135 lines · plain
1/* memcmp - compare memory2 *3 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.4 * See https://llvm.org/LICENSE.txt for license information.5 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception6 */7 8/* Assumptions:9 *10 * ARMv8-a, AArch64, unaligned accesses.11 */12 13#include "../asmdefs.h"14 15/* Parameters and result. */16#define src1 x017#define src2 x118#define limit x219#define result w020 21/* Internal variables. */22#define data1 x323#define data1w w324#define data1h x425#define data2 x526#define data2w w527#define data2h x628#define tmp1 x729#define tmp2 x830 31ENTRY (__memcmp_aarch64)32 subs limit, limit, 833 b.lo L(less8)34 35 ldr data1, [src1], 836 ldr data2, [src2], 837 cmp data1, data238 b.ne L(return)39 40 subs limit, limit, 841 b.gt L(more16)42 43 ldr data1, [src1, limit]44 ldr data2, [src2, limit]45 b L(return)46 47L(more16):48 ldr data1, [src1], 849 ldr data2, [src2], 850 cmp data1, data251 bne L(return)52 53 /* Jump directly to comparing the last 16 bytes for 32 byte (or less)54 strings. */55 subs limit, limit, 1656 b.ls L(last_bytes)57 58 /* We overlap loads between 0-32 bytes at either side of SRC1 when we59 try to align, so limit it only to strings larger than 128 bytes. */60 cmp limit, 9661 b.ls L(loop16)62 63 /* Align src1 and adjust src2 with bytes not yet done. */64 and tmp1, src1, 1565 add limit, limit, tmp166 sub src1, src1, tmp167 sub src2, src2, tmp168 69 /* Loop performing 16 bytes per iteration using aligned src1.70 Limit is pre-decremented by 16 and must be larger than zero.71 Exit if <= 16 bytes left to do or if the data is not equal. */72 .p2align 473L(loop16):74 ldp data1, data1h, [src1], 1675 ldp data2, data2h, [src2], 1676 subs limit, limit, 1677 ccmp data1, data2, 0, hi78 ccmp data1h, data2h, 0, eq79 b.eq L(loop16)80 81 cmp data1, data282 bne L(return)83 mov data1, data1h84 mov data2, data2h85 cmp data1, data286 bne L(return)87 88 /* Compare last 1-16 bytes using unaligned access. */89L(last_bytes):90 add src1, src1, limit91 add src2, src2, limit92 ldp data1, data1h, [src1]93 ldp data2, data2h, [src2]94 cmp data1, data295 bne L(return)96 mov data1, data1h97 mov data2, data2h98 cmp data1, data299 100 /* Compare data bytes and set return value to 0, -1 or 1. */101L(return):102#ifndef __AARCH64EB__103 rev data1, data1104 rev data2, data2105#endif106 cmp data1, data2107L(ret_eq):108 cset result, ne109 cneg result, result, lo110 ret111 112 .p2align 4113 /* Compare up to 8 bytes. Limit is [-8..-1]. */114L(less8):115 adds limit, limit, 4116 b.lo L(less4)117 ldr data1w, [src1], 4118 ldr data2w, [src2], 4119 cmp data1w, data2w120 b.ne L(return)121 sub limit, limit, 4122L(less4):123 adds limit, limit, 4124 beq L(ret_eq)125L(byte_loop):126 ldrb data1w, [src1], 1127 ldrb data2w, [src2], 1128 subs limit, limit, 1129 ccmp data1w, data2w, 0, ne /* NZCV = 0b0000. */130 b.eq L(byte_loop)131 sub result, data1w, data2w132 ret133 134END (__memcmp_aarch64)135