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/* 23*5d9d9091SRichard Lowe * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24*5d9d9091SRichard Lowe * Use is subject to license terms. 25*5d9d9091SRichard Lowe */ 26*5d9d9091SRichard Lowe 27*5d9d9091SRichard Lowe .file "strlen.s" 28*5d9d9091SRichard Lowe 29*5d9d9091SRichard Lowe/* 30*5d9d9091SRichard Lowe * strlen(s) 31*5d9d9091SRichard Lowe * 32*5d9d9091SRichard Lowe * Given string s, return length (not including the terminating null). 33*5d9d9091SRichard Lowe * 34*5d9d9091SRichard Lowe * Fast assembler language version of the following C-program strlen 35*5d9d9091SRichard Lowe * which represents the `standard' for the C-library. 36*5d9d9091SRichard Lowe * 37*5d9d9091SRichard Lowe * size_t 38*5d9d9091SRichard Lowe * strlen(s) 39*5d9d9091SRichard Lowe * register const char *s; 40*5d9d9091SRichard Lowe * { 41*5d9d9091SRichard Lowe * register const char *s0 = s + 1; 42*5d9d9091SRichard Lowe * 43*5d9d9091SRichard Lowe * while (*s++ != '\0') 44*5d9d9091SRichard Lowe * ; 45*5d9d9091SRichard Lowe * return (s - s0); 46*5d9d9091SRichard Lowe * } 47*5d9d9091SRichard Lowe */ 48*5d9d9091SRichard Lowe 49*5d9d9091SRichard Lowe#include <sys/asm_linkage.h> 50*5d9d9091SRichard Lowe 51*5d9d9091SRichard Lowe ! The object of strlen is to, as quickly as possible, find the 52*5d9d9091SRichard Lowe ! null byte. To this end, we attempt to get our string aligned 53*5d9d9091SRichard Lowe ! and then blast across it using Alan Mycroft's algorithm for 54*5d9d9091SRichard Lowe ! finding null bytes. If we are not aligned, the string is 55*5d9d9091SRichard Lowe ! checked a byte at a time until it is. Once this occurs, 56*5d9d9091SRichard Lowe ! we can proceed word-wise across it. Once a word with a 57*5d9d9091SRichard Lowe ! zero byte has been found, we then check the word a byte 58*5d9d9091SRichard Lowe ! at a time until we've located the zero byte, and return 59*5d9d9091SRichard Lowe ! the proper length. 60*5d9d9091SRichard Lowe 61*5d9d9091SRichard Lowe .align 32 62*5d9d9091SRichard Lowe ENTRY(strlen) 63*5d9d9091SRichard Lowe andcc %o0, 3, %o4 ! is src word aligned 64*5d9d9091SRichard Lowe bz,pt %icc, .nowalgnd 65*5d9d9091SRichard Lowe mov %o0, %o2 66*5d9d9091SRichard Lowe 67*5d9d9091SRichard Lowe cmp %o4, 2 ! is src half-word aligned 68*5d9d9091SRichard Lowe be,a,pn %icc, .s2algn 69*5d9d9091SRichard Lowe lduh [%o2], %o1 70*5d9d9091SRichard Lowe 71*5d9d9091SRichard Lowe ldub [%o2], %o1 72*5d9d9091SRichard Lowe tst %o1 ! byte zero? 73*5d9d9091SRichard Lowe bz,pn %icc, .done 74*5d9d9091SRichard Lowe cmp %o4, 3 ! src is byte aligned 75*5d9d9091SRichard Lowe 76*5d9d9091SRichard Lowe be,pn %icc, .nowalgnd 77*5d9d9091SRichard Lowe inc 1, %o2 78*5d9d9091SRichard Lowe 79*5d9d9091SRichard Lowe lduh [%o2], %o1 80*5d9d9091SRichard Lowe 81*5d9d9091SRichard Lowe.s2algn: 82*5d9d9091SRichard Lowe srl %o1, 8, %o4 83*5d9d9091SRichard Lowe tst %o4 84*5d9d9091SRichard Lowe bz,pn %icc, .done 85*5d9d9091SRichard Lowe andcc %o1, 0xff, %g0 86*5d9d9091SRichard Lowe 87*5d9d9091SRichard Lowe bz,pn %icc, .done 88*5d9d9091SRichard Lowe inc 1, %o2 89*5d9d9091SRichard Lowe 90*5d9d9091SRichard Lowe inc 1, %o2 91*5d9d9091SRichard Lowe 92*5d9d9091SRichard Lowe.nowalgnd: 93*5d9d9091SRichard Lowe ld [%o2], %o1 94*5d9d9091SRichard Lowe sethi %hi(0x01010101), %o4 95*5d9d9091SRichard Lowe sethi %hi(0x80808080), %o5 96*5d9d9091SRichard Lowe or %o4, %lo(0x01010101), %o4 97*5d9d9091SRichard Lowe or %o5, %lo(0x80808080), %o5 98*5d9d9091SRichard Lowe 99*5d9d9091SRichard Lowe andn %o5, %o1, %o3 100*5d9d9091SRichard Lowe sub %o1, %o4, %g1 101*5d9d9091SRichard Lowe andcc %o3, %g1, %g0 102*5d9d9091SRichard Lowe bnz,a,pn %icc, .nullfound 103*5d9d9091SRichard Lowe sethi %hi(0xff000000), %o4 104*5d9d9091SRichard Lowe 105*5d9d9091SRichard Lowe ld [%o2+4], %o1 106*5d9d9091SRichard Lowe inc 4, %o2 107*5d9d9091SRichard Lowe 108*5d9d9091SRichard Lowe.loop: ! this should be aligned to 32 109*5d9d9091SRichard Lowe inc 4, %o2 110*5d9d9091SRichard Lowe andn %o5, %o1, %o3 ! %o5 = ~word & 0x80808080 111*5d9d9091SRichard Lowe sub %o1, %o4, %g1 ! %g1 = word - 0x01010101 112*5d9d9091SRichard Lowe andcc %o3, %g1, %g0 113*5d9d9091SRichard Lowe bz,a,pt %icc, .loop 114*5d9d9091SRichard Lowe ld [%o2], %o1 115*5d9d9091SRichard Lowe 116*5d9d9091SRichard Lowe dec 4, %o2 117*5d9d9091SRichard Lowe sethi %hi(0xff000000), %o4 118*5d9d9091SRichard Lowe.nullfound: 119*5d9d9091SRichard Lowe andcc %o1, %o4, %g0 120*5d9d9091SRichard Lowe bz,pn %icc, .done ! first byte zero 121*5d9d9091SRichard Lowe srl %o4, 8, %o4 122*5d9d9091SRichard Lowe 123*5d9d9091SRichard Lowe andcc %o1, %o4, %g0 124*5d9d9091SRichard Lowe bz,pn %icc, .done ! second byte zero 125*5d9d9091SRichard Lowe inc 1, %o2 126*5d9d9091SRichard Lowe 127*5d9d9091SRichard Lowe srl %o4, 8, %o4 128*5d9d9091SRichard Lowe andcc %o1, %o4, %g0 129*5d9d9091SRichard Lowe bz,pn %icc, .done ! thrid byte zero 130*5d9d9091SRichard Lowe inc 1, %o2 131*5d9d9091SRichard Lowe 132*5d9d9091SRichard Lowe inc 1, %o2 ! fourth byte zero 133*5d9d9091SRichard Lowe.done: 134*5d9d9091SRichard Lowe retl 135*5d9d9091SRichard Lowe sub %o2, %o0, %o0 136*5d9d9091SRichard Lowe SET_SIZE(strlen) 137*5d9d9091SRichard Lowe 138