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 2003 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 * strcpy(s1, s2) 33 * 34 * Copy string s2 to s1. s1 must be large enough. Return s1. 35 * 36 * Fast assembler language version of the following C-program strcpy 37 * which represents the `standard' for the C-library. 38 * 39 * char * 40 * strcpy(s1, s2) 41 * register char *s1; 42 * register const char *s2; 43 * { 44 * char *os1 = s1; 45 * 46 * while(*s1++ = *s2++) 47 * ; 48 * return(os1); 49 * } 50 * 51 */ 52 53#include <sys/asm_linkage.h> 54#include "synonyms.h" 55 56 ! This implementation of strcpy works by first checking the 57 ! source alignment and copying byte, half byte, or word 58 ! quantities until the source ptr is aligned at an extended 59 ! word boundary. Once this has occurred, the string is copied, 60 ! checking for zero bytes, depending upon its dst ptr alignment. 61 ! (methods for xword, word, half-word, and byte copies are present) 62 63 ENTRY(strcpy) 64 65 .align 32 66 67 sub %o1, %o0, %o3 ! src - dst 68 andcc %o1, 7, %o4 ! dword aligned ? 69 bz,pn %ncc, .srcaligned ! yup 70 mov %o0, %o2 ! save dst 71 72.chkbyte: 73 andcc %o1, 1, %g0 ! need to copy byte ? 74 bz,pn %ncc, .chkhalfword ! nope, maybe halfword 75 sub %g0, %o1, %g1 ! %g1<2:0> = # of unaligned bytes 76 ldub [%o2 + %o3], %o5 ! src[0] 77 tst %o5 ! src[0] == 0 ? 78 stb %o5, [%o2] ! dst[0] = src[0] 79 bz,pn %ncc, .done ! yup, done 80 inc %o2 ! src++, dst++ 81 82.chkhalfword: 83 andcc %g1, 2, %g0 ! need to copy half-word ? 84 bz,pn %ncc, .chkword ! nope, maybe word 85 nop ! 86 lduh [%o2 + %o3], %o5 ! load src halfword 87 srl %o5, 8, %o4 ! extract first byte 88 tst %o4 ! first byte == 0 ? 89 bz,pn %ncc, .done ! yup, done 90 stb %o4, [%o2] ! store first byte 91 andcc %o5, 0xff, %g0 ! extract second byte 92 stb %o5, [%o2 + 1] ! store second byte 93 bz,pn %ncc, .done ! yup, 2nd byte zero, done 94 add %o2, 2, %o2 ! src += 2 95 96.chkword: 97 andcc %g1, 4, %g0 ! need to copy word ? 98 bz,pn %ncc, .srcaligned ! nope 99 nop ! 100 lduw [%o2 + %o3], %o5 ! load src word 101 srl %o5, 24, %o4 ! extract first byte 102 tst %o4 ! is first byte zero ? 103 bz,pn %ncc, .done ! yup, done 104 stb %o4, [%o2] ! store first byte 105 srl %o5, 16, %o4 ! extract second byte 106 andcc %o4, 0xff, %g0 ! is second byte zero ? 107 bz,pn %ncc, .done ! yup, done 108 stb %o4, [%o2 + 1] ! store second byte 109 srl %o5, 8, %o4 ! extract third byte 110 andcc %o4, 0xff, %g0 ! third byte zero ? 111 bz,pn %ncc, .done ! yup, done 112 stb %o4, [%o2 + 2] ! store third byte 113 andcc %o5, 0xff, %g0 ! fourth byte zero ? 114 stb %o5, [%o2 + 3] ! store fourth byte 115 bz,pn %ncc, .done ! yup, fourth byte zero, done 116 add %o2, 4, %o2 ! src += 2 117 118.srcaligned: 119 sethi %hi(0x01010101), %o4 ! Alan Mycroft's magic1 120 or %o4, %lo(0x01010101),%o4! finish loading magic1 121 sllx %o4, 32, %o1 ! spread magic1 122 and %o2, 3, %g4 ! dst<1:0> to examine offset 123 or %o4, %o1, %o4 ! to all 64 bits 124 cmp %g4, 1 ! dst offset of 1 or 5 125 sllx %o4, 7, %o5 ! Alan Mycroft's magic2 126 be,pn %ncc, .storebyte1241 ! store 1, 2, 4, 1 bytes 127 cmp %g4, 3 ! dst offset of 3 or 7 128 be,pn %ncc, .storebyte1421 ! store 1, 4, 2, 1 bytes 129 cmp %g4, 2 ! dst halfword aligned ? 130 be,pn %ncc, .storehalfword ! yup, store half-word wise 131 andcc %o2, 7, %g0 ! dst word aligned ? 132 bnz,pn %ncc, .storeword2 ! yup, store word wise 133 .empty 134 135.storedword: 136 ldx [%o2 + %o3], %o1 ! src dword 137 add %o2, 8, %o2 ! src += 8, dst += 8 138 andn %o5, %o1, %g1 ! ~dword & 0x8080808080808080 139 sub %o1, %o4, %g4 ! dword - 0x0101010101010101 140 andcc %g4, %g1, %g0 ! ((dword - 0x0101010101010101) & ~dword & 0x8080808080808080) 141 bz,a,pt %ncc, .storedword ! no zero byte if magic expression == 0 142 stx %o1, [%o2 - 8] ! store word to dst (address pre-incremented) 143 144.zerobyte: 145 orn %o4, %g0, %o4 ! 0xffffffffffffffff 146 sllx %o4, 56, %o4 ! 0xff00000000000000 147 srlx %o1, 56, %o3 ! %o3<7:0> = first byte 148 andcc %o1, %o4, %g0 ! first byte zero? 149 bz,pn %ncc, .done ! yup, done 150 stb %o3, [%o2 - 8] ! store first byte 151 srlx %o4, 8, %o4 ! 0x00ff000000000000 152 srlx %o1, 48, %o3 ! %o3<7:0> = second byte 153 andcc %o1, %o4, %g0 ! second byte zero? 154 bz,pn %ncc, .done ! yup, done 155 stb %o3, [%o2 - 7] ! store second byte 156 srlx %o4, 8, %o4 ! 0x0000ff0000000000 157 srlx %o1, 40, %o3 ! %o3<7:0> = third byte 158 andcc %o1, %o4, %g0 ! third byte zero? 159 bz,pn %ncc, .done ! yup, done 160 stb %o3, [%o2 - 6] ! store third byte 161 srlx %o4, 8, %o4 ! 0x000000ff00000000 162 srlx %o1, 32, %o3 ! %o3<7:0> = fourth byte 163 andcc %o1, %o4, %g0 ! fourth byte zero? 164 bz,pn %ncc, .done ! yup, done 165 stb %o3, [%o2 - 5] ! store fourth byte 166 srlx %o4, 8, %o4 ! 0x00000000ff000000 167 srlx %o1, 24, %o3 ! %o3<7:0> = fifth byte 168 andcc %o1, %o4, %g0 ! fifth byte zero? 169 bz,pn %ncc, .done ! yup, done 170 stb %o3, [%o2 - 4] ! store fifth byte 171 srlx %o4, 8, %o4 ! 0x0000000000ff0000 172 srlx %o1, 16, %o3 ! %o3<7:0> = sixth byte 173 andcc %o1, %o4, %g0 ! sixth byte zero? 174 bz,pn %ncc, .done ! yup, done 175 stb %o3, [%o2 - 3] ! store sixth byte 176 srlx %o4, 8, %o4 ! 0x000000000000ff00 177 andcc %o1, %o4, %g0 ! seventh byte zero? 178 srlx %o1, 8, %o3 ! %o3<7:0> = seventh byte 179 bz,pn %ncc, .done ! yup, done 180 stb %o3, [%o2 - 2] ! store seventh byte 181 stb %o1, [%o2 - 1] ! store eigth byte 182.done: 183 retl ! done with leaf function 184 185 nop ! ensure following loop 16-byte aligned 186 187.storebyte1421: 188 ldx [%o2 + %o3], %o1 ! x = src[] 189 add %o2, 8, %o2 ! src += 8, dst += 8 190 andn %o5, %o1, %g1 ! ~x & 0x8080808080808080 191 sub %o1, %o4, %g4 ! x - 0x0101010101010101 192 andcc %g4, %g1, %g0 ! ((x - 0x0101010101010101) & ~x & 0x8080808080808080) 193 bnz,pn %ncc, .zerobyte ! x has zero byte, handle end cases 194 srlx %o1, 56, %g1 ! %g1<7:0> = first byte; word aligned now 195 stb %g1, [%o2 - 8] ! store first byte 196 srlx %o1, 24, %g1 ! %g1<31:0> = bytes 2, 3, 4, 5 197 stw %g1, [%o2 - 7] ! store bytes 2, 3, 4, 5 198 srlx %o1, 8, %g1 ! %g1<15:0> = bytes 6, 7 199 sth %g1, [%o2 - 3] ! store bytes 6, 7 200 ba .storebyte1421 ! next dword 201 stb %o1, [%o2 - 1] ! store eigth byte 202 203 nop ! ensure following loop 16-byte aligned 204 nop ! ensure following loop 16-byte aligned 205 206.storebyte1241: 207 ldx [%o2 + %o3], %o1 ! x = src[] 208 add %o2, 8, %o2 ! src += 8, dst += 8 209 andn %o5, %o1, %g1 ! ~x & 0x8080808080808080 210 sub %o1, %o4, %g4 ! x - 0x0101010101010101 211 andcc %g4, %g1, %g0 ! ((x - 0x0101010101010101) & ~x & 0x8080808080808080) 212 bnz,pn %ncc, .zerobyte ! x has zero byte, handle end cases 213 srlx %o1, 56, %g1 ! %g1<7:0> = first byte; word aligned now 214 stb %g1, [%o2 - 8] ! store first byte 215 srlx %o1, 40, %g1 ! %g1<15:0> = bytes 2, 3 216 sth %g1, [%o2 - 7] ! store bytes 2, 3 217 srlx %o1, 8, %g1 ! %g1<31:0> = bytes 4, 5, 6, 7 218 stw %g1, [%o2 - 5] ! store bytes 4, 5, 6, 7 219 ba .storebyte1241 ! next dword 220 stb %o1, [%o2 - 1] ! store eigth byte 221 222 nop ! ensure following loop 16-byte aligned 223 nop ! ensure following loop 16-byte aligned 224 225.storehalfword: 226 ldx [%o2 + %o3], %o1 ! x = src[] 227 add %o2, 8, %o2 ! src += 8, dst += 8 228 andn %o5, %o1, %g1 ! ~x & 0x8080808080808080 229 sub %o1, %o4, %g4 ! x - 0x0101010101010101 230 andcc %g4, %g1, %g0 ! ((x - 0x0101010101010101) & ~x & 0x8080808080808080) 231 bnz,pn %ncc, .zerobyte ! x has zero byte, handle end cases 232 srlx %o1, 48, %g1 ! get first and second byte 233 sth %g1, [%o2 - 8] ! store first and second byte; word aligned now 234 srlx %o1, 16, %g1 ! %g1<31:0> = bytes 3, 4, 5, 6 235 stw %g1, [%o2 - 6] ! store bytes 3, 4, 5, 6 236 ba .storehalfword ! next word 237 sth %o1, [%o2 - 2] ! store seventh and eigth byte 238 239.storeword: 240 ldx [%o2 + %o3], %o1 ! x = src[] 241.storeword2: 242 add %o2, 8, %o2 ! src += 8, dst += 8 243 andn %o5, %o1, %g1 ! ~x & 0x0x8080808080808080 244 sub %o1, %o4, %g4 ! x - 0x0101010101010101 245 andcc %g4, %g1, %g0 ! ((x - 0x0101010101010101) & ~x & 0x8080808080808080) 246 bnz,pn %ncc, .zerobyte ! x has zero byte, handle end cases 247 srlx %o1, 32, %g1 ! get bytes 1,2,3,4 248 stw %g1, [%o2 - 8] ! store bytes 1,2,3,4 (address is pre-incremented) 249 ba .storeword ! no zero byte if magic expression == 0 250 stw %o1, [%o2 - 4] ! store bytes 5,6,7,8 251 252 nop ! padding, do not remove!!! 253 nop ! padding, do not remove!!! 254 SET_SIZE(strcpy) 255 256