1 /* 2 * Copyright (c) 2011 The Android Open Source Project 3 * Copyright (c) 2008 ARM Ltd 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. The name of the company may not be used to endorse or promote 15 * products derived from this software without specific prior written 16 * permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED 19 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 20 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED 23 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 24 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 25 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 26 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 27 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include <private/bionic_asm.h> 31 32 .text 33 34 #ifdef __ARMEB__ 35 #define SHFT2LSB lsl 36 #define SHFT2LSBEQ lsleq 37 #define SHFT2MSB lsr 38 #define SHFT2MSBEQ lsreq 39 #define MSB 0x000000ff 40 #define LSB 0xff000000 41 #else 42 #define SHFT2LSB lsr 43 #define SHFT2LSBEQ lsreq 44 #define SHFT2MSB lsl 45 #define SHFT2MSBEQ lsleq 46 #define MSB 0xff000000 47 #define LSB 0x000000ff 48 #endif 49 50 #define magic1(REG) REG 51 #define magic2(REG) REG, lsl #7 52 53 ENTRY(strcmp) 54 pld [r0, #0] 55 pld [r1, #0] 56 eor r2, r0, r1 57 tst r2, #3 58 59 /* Strings not at same byte offset from a word boundary. */ 60 bne .Lstrcmp_unaligned 61 ands r2, r0, #3 62 bic r0, r0, #3 63 bic r1, r1, #3 64 ldr ip, [r0], #4 65 it eq 66 ldreq r3, [r1], #4 67 beq 1f 68 69 /* Although s1 and s2 have identical initial alignment, they are 70 * not currently word aligned. Rather than comparing bytes, 71 * make sure that any bytes fetched from before the addressed 72 * bytes are forced to 0xff. Then they will always compare 73 * equal. 74 */ 75 eor r2, r2, #3 76 lsl r2, r2, #3 77 mvn r3, #MSB 78 SHFT2LSB r2, r3, r2 79 ldr r3, [r1], #4 80 orr ip, ip, r2 81 orr r3, r3, r2 82 1: 83 /* Load the 'magic' constant 0x01010101. */ 84 str r4, [sp, #-4]! 85 mov r4, #1 86 orr r4, r4, r4, lsl #8 87 orr r4, r4, r4, lsl #16 88 .p2align 2 89 4: 90 pld [r0, #8] 91 pld [r1, #8] 92 sub r2, ip, magic1(r4) 93 cmp ip, r3 94 itttt eq 95 96 /* check for any zero bytes in first word */ 97 biceq r2, r2, ip 98 tsteq r2, magic2(r4) 99 ldreq ip, [r0], #4 100 ldreq r3, [r1], #4 101 beq 4b 102 2: 103 /* There's a zero or a different byte in the word */ 104 SHFT2MSB r0, ip, #24 105 SHFT2LSB ip, ip, #8 106 cmp r0, #1 107 it cs 108 cmpcs r0, r3, SHFT2MSB #24 109 it eq 110 SHFT2LSBEQ r3, r3, #8 111 beq 2b 112 /* On a big-endian machine, r0 contains the desired byte in bits 113 * 0-7; on a little-endian machine they are in bits 24-31. In 114 * both cases the other bits in r0 are all zero. For r3 the 115 * interesting byte is at the other end of the word, but the 116 * other bits are not necessarily zero. We need a signed result 117 * representing the differnece in the unsigned bytes, so for the 118 * little-endian case we can't just shift the interesting bits up. 119 */ 120 #ifdef __ARMEB__ 121 sub r0, r0, r3, lsr #24 122 #else 123 and r3, r3, #255 124 /* No RSB instruction in Thumb2 */ 125 #ifdef __thumb2__ 126 lsr r0, r0, #24 127 sub r0, r0, r3 128 #else 129 rsb r0, r3, r0, lsr #24 130 #endif 131 #endif 132 ldr r4, [sp], #4 133 bx lr 134 135 .Lstrcmp_unaligned: 136 wp1 .req r0 137 wp2 .req r1 138 b1 .req r2 139 w1 .req r4 140 w2 .req r5 141 t1 .req ip 142 @ r3 is scratch 143 144 /* First of all, compare bytes until wp1(sp1) is word-aligned. */ 145 1: 146 tst wp1, #3 147 beq 2f 148 ldrb r2, [wp1], #1 149 ldrb r3, [wp2], #1 150 cmp r2, #1 151 it cs 152 cmpcs r2, r3 153 beq 1b 154 sub r0, r2, r3 155 bx lr 156 157 2: 158 str r5, [sp, #-4]! 159 str r4, [sp, #-4]! 160 mov b1, #1 161 orr b1, b1, b1, lsl #8 162 orr b1, b1, b1, lsl #16 163 164 and t1, wp2, #3 165 bic wp2, wp2, #3 166 ldr w1, [wp1], #4 167 ldr w2, [wp2], #4 168 cmp t1, #2 169 beq 2f 170 bhi 3f 171 172 /* Critical inner Loop: Block with 3 bytes initial overlap */ 173 .p2align 2 174 1: 175 bic t1, w1, #MSB 176 cmp t1, w2, SHFT2LSB #8 177 sub r3, w1, b1 178 bic r3, r3, w1 179 bne 4f 180 ands r3, r3, b1, lsl #7 181 it eq 182 ldreq w2, [wp2], #4 183 bne 5f 184 eor t1, t1, w1 185 cmp t1, w2, SHFT2MSB #24 186 bne 6f 187 ldr w1, [wp1], #4 188 b 1b 189 4: 190 SHFT2LSB w2, w2, #8 191 b 8f 192 193 5: 194 #ifdef __ARMEB__ 195 /* The syndrome value may contain false ones if the string ends 196 * with the bytes 0x01 0x00 197 */ 198 tst w1, #0xff000000 199 itt ne 200 tstne w1, #0x00ff0000 201 tstne w1, #0x0000ff00 202 beq 7f 203 #else 204 bics r3, r3, #0xff000000 205 bne 7f 206 #endif 207 ldrb w2, [wp2] 208 SHFT2LSB t1, w1, #24 209 #ifdef __ARMEB__ 210 lsl w2, w2, #24 211 #endif 212 b 8f 213 214 6: 215 SHFT2LSB t1, w1, #24 216 and w2, w2, #LSB 217 b 8f 218 219 /* Critical inner Loop: Block with 2 bytes initial overlap */ 220 .p2align 2 221 2: 222 SHFT2MSB t1, w1, #16 223 sub r3, w1, b1 224 SHFT2LSB t1, t1, #16 225 bic r3, r3, w1 226 cmp t1, w2, SHFT2LSB #16 227 bne 4f 228 ands r3, r3, b1, lsl #7 229 it eq 230 ldreq w2, [wp2], #4 231 bne 5f 232 eor t1, t1, w1 233 cmp t1, w2, SHFT2MSB #16 234 bne 6f 235 ldr w1, [wp1], #4 236 b 2b 237 238 5: 239 #ifdef __ARMEB__ 240 /* The syndrome value may contain false ones if the string ends 241 * with the bytes 0x01 0x00 242 */ 243 tst w1, #0xff000000 244 it ne 245 tstne w1, #0x00ff0000 246 beq 7f 247 #else 248 lsls r3, r3, #16 249 bne 7f 250 #endif 251 ldrh w2, [wp2] 252 SHFT2LSB t1, w1, #16 253 #ifdef __ARMEB__ 254 lsl w2, w2, #16 255 #endif 256 b 8f 257 258 6: 259 SHFT2MSB w2, w2, #16 260 SHFT2LSB t1, w1, #16 261 4: 262 SHFT2LSB w2, w2, #16 263 b 8f 264 265 /* Critical inner Loop: Block with 1 byte initial overlap */ 266 .p2align 2 267 3: 268 and t1, w1, #LSB 269 cmp t1, w2, SHFT2LSB #24 270 sub r3, w1, b1 271 bic r3, r3, w1 272 bne 4f 273 ands r3, r3, b1, lsl #7 274 it eq 275 ldreq w2, [wp2], #4 276 bne 5f 277 eor t1, t1, w1 278 cmp t1, w2, SHFT2MSB #8 279 bne 6f 280 ldr w1, [wp1], #4 281 b 3b 282 4: 283 SHFT2LSB w2, w2, #24 284 b 8f 285 5: 286 /* The syndrome value may contain false ones if the string ends 287 * with the bytes 0x01 0x00 288 */ 289 tst w1, #LSB 290 beq 7f 291 ldr w2, [wp2], #4 292 6: 293 SHFT2LSB t1, w1, #8 294 bic w2, w2, #MSB 295 b 8f 296 7: 297 mov r0, #0 298 ldr r4, [sp], #4 299 ldr r5, [sp], #4 300 bx lr 301 302 8: 303 and r2, t1, #LSB 304 and r0, w2, #LSB 305 cmp r0, #1 306 it cs 307 cmpcs r0, r2 308 itt eq 309 SHFT2LSBEQ t1, t1, #8 310 SHFT2LSBEQ w2, w2, #8 311 beq 8b 312 sub r0, r2, r0 313 ldr r4, [sp], #4 314 ldr r5, [sp], #4 315 bx lr 316 END(strcmp) 317
