1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License, Version 1.0 only 6 * (the "License"). You may not use this file except in compliance 7 * with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22/* 23 * Copyright 2004 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 .ident "%Z%%M% %I% %E% SMI" 28 29 .file "%M%" 30 31/ 32/ strcat(s1, s2) 33/ 34/ Concatenates s2 on the end of s1. s1's space must be large enough. 35/ Returns s1. 36/ 37/ Fast assembly language version of the following C-program strcat 38/ which represents the `standard' for the C-library. 39/ 40/ char * 41/ strcat(char *s1, const char *s2) 42/ { 43/ char *os1 = s1; 44/ 45/ while (*s1++) 46/ ; 47/ --s1; 48/ while (*s1++ = *s2++) 49/ ; 50/ return (os1); 51/ } 52/ 53/ In this assembly language version, the following expression is used 54/ to check if a 32-bit word data contains a null byte or not: 55/ (((A & 0x7f7f7f7f) + 0x7f7f7f7f) | A) & 0x80808080 56/ If the above expression geneates a value other than 0x80808080, 57/ that means the 32-bit word data contains a null byte. 58/ 59/ The above has been extended for 64-bit support. 60/ 61 62#include "SYS.h" 63 64 ENTRY(strcat) /* (char *s1, char *s2) */ 65 / find a null byte in destination string 66 movq %rdi,%rax / prepare return value 67 movabsq $0x7f7f7f7f7f7f7f7f, %r8 / %r8 = 0x7f... 68 movq %r8, %r9 69 notq %r9 / %r9 = 0x80... 70 testq $7, %rdi / if %rdi not quadword aligned 71 jnz .L1 / goto .L1 72 .align 4 73.L2: 74 movq (%rdi), %rdx / move 1 quadword from (%rdi) to %rdx 75 movq %r8, %rcx 76 andq %rdx, %rcx / %rcx = %rdx & 0x7f7f7f7f7f7f7f7f 77 addq $8, %rdi / next quadword 78 addq %r8, %rcx / %rcx += 0x7f7f7f7f7f7f7f7f 79 orq %rdx, %rcx / %rcx |= %rdx 80 andq %r9, %rcx / %rcx &= 0x8080808080808080 81 cmpq %r9, %rcx / if no null byte in this quadword 82 je .L2 / goto .L2 83 subq $8, %rdi / post-incremented 84.L1: 85 cmpb $0, (%rdi) / if a byte in (%rdi) is null 86 je .L3 / goto .L3 87 incq %rdi / next byte 88 testq $7, %rdi / if %rdi not quadword aligned 89 jnz .L1 / goto .L1 90 jmp .L2 / goto .L2 (%rdi quadword aligned) 91 .align 4 92.L3: 93 / %rdi points to a null byte in destination string 94 testq $7, %rsi / if %rsi not quadword aligned 95 jnz .L4 / goto .L4 96 .align 4 97.L5: 98 movq (%rsi), %rdx / move 1 quadword from (%rsi) to %rdx 99 movq %r8, %rcx 100 andq %rdx, %rcx / %rcx = %rdx & 0x7f7f7f7f7f7f7f7f 101 addq $8, %rsi / next quadword 102 addq %r8, %rcx / %rcx += 0x7f7f7f7f7f7f7f7f 103 orq %rdx, %rcx / %rcx |= %rdx 104 andq %r9, %rcx / %rcx &= 0x8080808080808080 105 cmpq %r9, %rcx / if null byte in this quadaword 106 jne .L7 / goto .L7 107 movq %rdx, (%rdi) / copy this quadword to (%rdi) 108 addq $8, %rdi / next quadword 109 jmp .L5 / goto .L5 110.L7: 111 subq $8, %rsi / post-incremented 112 .align 4 113.L4: 114 movb (%rsi), %dl / %dl = a byte in (%rsi) 115 cmpb $0, %dl / compare %dl with a null byte 116 movb %dl, (%rdi) / copy %dl to (%rdi) 117 je .L6 / if %dl is a null, goto .L6 118 incq %rsi / next byte 119 incq %rdi / next byte 120 testq $7, %rsi / if %rsi not word aligned 121 jnz .L4 / goto .L4 122 jmp .L5 / goto .L5 (%rsi word aligned) 123 .align 4 124.L6: 125 ret 126 SET_SIZE(strcat) 127