1*5d9d9091SRichard Lowe/* 2*5d9d9091SRichard Lowe * CDDL HEADER START 3*5d9d9091SRichard Lowe * 4*5d9d9091SRichard Lowe * The contents of this file are subject to the terms of the 5*5d9d9091SRichard Lowe * Common Development and Distribution License (the "License"). 6*5d9d9091SRichard Lowe * You may not use this file except in compliance with the License. 7*5d9d9091SRichard Lowe * 8*5d9d9091SRichard Lowe * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9*5d9d9091SRichard Lowe * or http://www.opensolaris.org/os/licensing. 10*5d9d9091SRichard Lowe * See the License for the specific language governing permissions 11*5d9d9091SRichard Lowe * and limitations under the License. 12*5d9d9091SRichard Lowe * 13*5d9d9091SRichard Lowe * When distributing Covered Code, include this CDDL HEADER in each 14*5d9d9091SRichard Lowe * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15*5d9d9091SRichard Lowe * If applicable, add the following below this CDDL HEADER, with the 16*5d9d9091SRichard Lowe * fields enclosed by brackets "[]" replaced with your own identifying 17*5d9d9091SRichard Lowe * information: Portions Copyright [yyyy] [name of copyright owner] 18*5d9d9091SRichard Lowe * 19*5d9d9091SRichard Lowe * CDDL HEADER END 20*5d9d9091SRichard Lowe */ 21*5d9d9091SRichard Lowe/* 22*5d9d9091SRichard Lowe * Copyright 2004 Sun Microsystems, Inc. All rights reserved. 23*5d9d9091SRichard Lowe * Use is subject to license terms. 24*5d9d9091SRichard Lowe */ 25*5d9d9091SRichard Lowe 26*5d9d9091SRichard Lowe .file "strcat.s" 27*5d9d9091SRichard Lowe 28*5d9d9091SRichard Lowe/ 29*5d9d9091SRichard Lowe/ strcat(s1, s2) 30*5d9d9091SRichard Lowe/ 31*5d9d9091SRichard Lowe/ Concatenates s2 on the end of s1. s1's space must be large enough. 32*5d9d9091SRichard Lowe/ Returns s1. 33*5d9d9091SRichard Lowe/ 34*5d9d9091SRichard Lowe/ Fast assembly language version of the following C-program strcat 35*5d9d9091SRichard Lowe/ which represents the `standard' for the C-library. 36*5d9d9091SRichard Lowe/ 37*5d9d9091SRichard Lowe/ char * 38*5d9d9091SRichard Lowe/ strcat(char *s1, const char *s2) 39*5d9d9091SRichard Lowe/ { 40*5d9d9091SRichard Lowe/ char *os1 = s1; 41*5d9d9091SRichard Lowe/ 42*5d9d9091SRichard Lowe/ while (*s1++) 43*5d9d9091SRichard Lowe/ ; 44*5d9d9091SRichard Lowe/ --s1; 45*5d9d9091SRichard Lowe/ while (*s1++ = *s2++) 46*5d9d9091SRichard Lowe/ ; 47*5d9d9091SRichard Lowe/ return (os1); 48*5d9d9091SRichard Lowe/ } 49*5d9d9091SRichard Lowe/ 50*5d9d9091SRichard Lowe/ In this assembly language version, the following expression is used 51*5d9d9091SRichard Lowe/ to check if a 32-bit word data contains a null byte or not: 52*5d9d9091SRichard Lowe/ (((A & 0x7f7f7f7f) + 0x7f7f7f7f) | A) & 0x80808080 53*5d9d9091SRichard Lowe/ If the above expression geneates a value other than 0x80808080, 54*5d9d9091SRichard Lowe/ that means the 32-bit word data contains a null byte. 55*5d9d9091SRichard Lowe/ 56*5d9d9091SRichard Lowe 57*5d9d9091SRichard Lowe#include "SYS.h" 58*5d9d9091SRichard Lowe 59*5d9d9091SRichard Lowe ENTRY(strcat) 60*5d9d9091SRichard Lowe pushl %edi / save register variable 61*5d9d9091SRichard Lowe / find a null byte in destination string 62*5d9d9091SRichard Lowe movl 8(%esp), %edi / %edi = destination string address 63*5d9d9091SRichard Lowe testl $3, %edi / if %edi not word aligned 64*5d9d9091SRichard Lowe jnz .L1 / goto .L1 65*5d9d9091SRichard Lowe .align 4 66*5d9d9091SRichard Lowe.L2: 67*5d9d9091SRichard Lowe movl (%edi), %edx / move 1 word from (%edi) to %edx 68*5d9d9091SRichard Lowe movl $0x7f7f7f7f, %ecx 69*5d9d9091SRichard Lowe andl %edx, %ecx / %ecx = %edx & 0x7f7f7f7f 70*5d9d9091SRichard Lowe addl $4, %edi / next word 71*5d9d9091SRichard Lowe addl $0x7f7f7f7f, %ecx / %ecx += 0x7f7f7f7f 72*5d9d9091SRichard Lowe orl %edx, %ecx / %ecx |= %edx 73*5d9d9091SRichard Lowe andl $0x80808080, %ecx / %ecx &= 0x80808080 74*5d9d9091SRichard Lowe cmpl $0x80808080, %ecx / if no null byte in this word 75*5d9d9091SRichard Lowe je .L2 / goto .L2 76*5d9d9091SRichard Lowe subl $4, %edi / post-incremented 77*5d9d9091SRichard Lowe.L1: 78*5d9d9091SRichard Lowe cmpb $0, (%edi) / if a byte in (%edi) is null 79*5d9d9091SRichard Lowe je .L3 / goto .L3 80*5d9d9091SRichard Lowe incl %edi / next byte 81*5d9d9091SRichard Lowe testl $3, %edi / if %edi not word aligned 82*5d9d9091SRichard Lowe jnz .L1 / goto .L1 83*5d9d9091SRichard Lowe jmp .L2 / goto .L2 (%edi word aligned) 84*5d9d9091SRichard Lowe .align 4 85*5d9d9091SRichard Lowe.L3: 86*5d9d9091SRichard Lowe / %edi points to a null byte in destination string 87*5d9d9091SRichard Lowe movl 12(%esp), %eax / %eax = source string address 88*5d9d9091SRichard Lowe testl $3, %eax / if %eax not word aligned 89*5d9d9091SRichard Lowe jnz .L4 / goto .L4 90*5d9d9091SRichard Lowe .align 4 91*5d9d9091SRichard Lowe.L5: 92*5d9d9091SRichard Lowe movl (%eax), %edx / move 1 word from (%eax) to %edx 93*5d9d9091SRichard Lowe movl $0x7f7f7f7f, %ecx 94*5d9d9091SRichard Lowe andl %edx, %ecx / %ecx = %edx & 0x7f7f7f7f 95*5d9d9091SRichard Lowe addl $4, %eax / next word 96*5d9d9091SRichard Lowe addl $0x7f7f7f7f, %ecx / %ecx += 0x7f7f7f7f 97*5d9d9091SRichard Lowe orl %edx, %ecx / %ecx |= %edx 98*5d9d9091SRichard Lowe andl $0x80808080, %ecx / %ecx &= 0x80808080 99*5d9d9091SRichard Lowe cmpl $0x80808080, %ecx / if null byte in this word 100*5d9d9091SRichard Lowe jne .L7 / goto .L7 101*5d9d9091SRichard Lowe movl %edx, (%edi) / copy this word to (%edi) 102*5d9d9091SRichard Lowe addl $4, %edi / next word 103*5d9d9091SRichard Lowe jmp .L5 / goto .L5 104*5d9d9091SRichard Lowe.L7: 105*5d9d9091SRichard Lowe subl $4, %eax / post-incremented 106*5d9d9091SRichard Lowe .align 4 107*5d9d9091SRichard Lowe.L4: 108*5d9d9091SRichard Lowe movb (%eax), %dl / %dl = a byte in (%eax) 109*5d9d9091SRichard Lowe cmpb $0, %dl / compare %dl with a null byte 110*5d9d9091SRichard Lowe movb %dl, (%edi) / copy %dl to (%edi) 111*5d9d9091SRichard Lowe je .L6 / if %dl is a null, goto .L6 112*5d9d9091SRichard Lowe incl %eax / next byte 113*5d9d9091SRichard Lowe incl %edi / next byte 114*5d9d9091SRichard Lowe testl $3, %eax / if %eax not word aligned 115*5d9d9091SRichard Lowe jnz .L4 / goto .L4 116*5d9d9091SRichard Lowe jmp .L5 / goto .L5 (%eax word aligned) 117*5d9d9091SRichard Lowe .align 4 118*5d9d9091SRichard Lowe.L6: 119*5d9d9091SRichard Lowe movl 8(%esp), %eax / return the destination address 120*5d9d9091SRichard Lowe popl %edi / restore register variable 121*5d9d9091SRichard Lowe ret 122*5d9d9091SRichard Lowe SET_SIZE(strcat) 123