1 /* 2 * Copyright (c) 2013 ARM Ltd 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. The name of the company may not be used to endorse or promote 14 * products derived from this software without specific prior written 15 * permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED 22 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 23 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 24 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 25 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 26 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include <machine/cpu-features.h> 30 #include <machine/asm.h> 31 32 #ifdef __ARMEB__ 33 #define S2LOMEM lsl 34 #define S2LOMEMEQ lsleq 35 #define S2HIMEM lsr 36 #define MSB 0x000000ff 37 #define LSB 0xff000000 38 #define BYTE0_OFFSET 24 39 #define BYTE1_OFFSET 16 40 #define BYTE2_OFFSET 8 41 #define BYTE3_OFFSET 0 42 #else /* not __ARMEB__ */ 43 #define S2LOMEM lsr 44 #define S2LOMEMEQ lsreq 45 #define S2HIMEM lsl 46 #define BYTE0_OFFSET 0 47 #define BYTE1_OFFSET 8 48 #define BYTE2_OFFSET 16 49 #define BYTE3_OFFSET 24 50 #define MSB 0xff000000 51 #define LSB 0x000000ff 52 #endif /* not __ARMEB__ */ 53 54 .syntax unified 55 56 #if defined (__thumb__) 57 .thumb 58 .thumb_func 59 #endif 60 61 ENTRY(strcmp) 62 /* Use LDRD whenever possible. */ 63 64 /* The main thing to look out for when comparing large blocks is that 65 the loads do not cross a page boundary when loading past the index 66 of the byte with the first difference or the first string-terminator. 67 68 For example, if the strings are identical and the string-terminator 69 is at index k, byte by byte comparison will not load beyond address 70 s1+k and s2+k; word by word comparison may load up to 3 bytes beyond 71 k; double word - up to 7 bytes. If the load of these bytes crosses 72 a page boundary, it might cause a memory fault (if the page is not mapped) 73 that would not have happened in byte by byte comparison. 74 75 If an address is (double) word aligned, then a load of a (double) word 76 from that address will not cross a page boundary. 77 Therefore, the algorithm below considers word and double-word alignment 78 of strings separately. */ 79 80 /* High-level description of the algorithm. 81 82 * The fast path: if both strings are double-word aligned, 83 use LDRD to load two words from each string in every loop iteration. 84 * If the strings have the same offset from a word boundary, 85 use LDRB to load and compare byte by byte until 86 the first string is aligned to a word boundary (at most 3 bytes). 87 This is optimized for quick return on short unaligned strings. 88 * If the strings have the same offset from a double-word boundary, 89 use LDRD to load two words from each string in every loop iteration, as in the fast path. 90 * If the strings do not have the same offset from a double-word boundary, 91 load a word from the second string before the loop to initialize the queue. 92 Use LDRD to load two words from every string in every loop iteration. 93 Inside the loop, load the second word from the second string only after comparing 94 the first word, using the queued value, to guarantee safety across page boundaries. 95 * If the strings do not have the same offset from a word boundary, 96 use LDR and a shift queue. Order of loads and comparisons matters, 97 similarly to the previous case. 98 99 * Use UADD8 and SEL to compare words, and use REV and CLZ to compute the return value. 100 * The only difference between ARM and Thumb modes is the use of CBZ instruction. 101 * The only difference between big and little endian is the use of REV in little endian 102 to compute the return value, instead of MOV. 103 */ 104 105 .macro m_cbz reg label 106 #ifdef __thumb2__ 107 cbz \reg, \label 108 #else /* not defined __thumb2__ */ 109 cmp \reg, #0 110 beq \label 111 #endif /* not defined __thumb2__ */ 112 .endm /* m_cbz */ 113 114 .macro m_cbnz reg label 115 #ifdef __thumb2__ 116 cbnz \reg, \label 117 #else /* not defined __thumb2__ */ 118 cmp \reg, #0 119 bne \label 120 #endif /* not defined __thumb2__ */ 121 .endm /* m_cbnz */ 122 123 .macro init 124 /* Macro to save temporary registers and prepare magic values. */ 125 subs sp, sp, #16 126 strd r4, r5, [sp, #8] 127 strd r6, r7, [sp] 128 mvn r6, #0 /* all F */ 129 mov r7, #0 /* all 0 */ 130 .endm /* init */ 131 132 .macro magic_compare_and_branch w1 w2 label 133 /* Macro to compare registers w1 and w2 and conditionally branch to label. */ 134 cmp \w1, \w2 /* Are w1 and w2 the same? */ 135 magic_find_zero_bytes \w1 136 it eq 137 cmpeq ip, #0 /* Is there a zero byte in w1? */ 138 bne \label 139 .endm /* magic_compare_and_branch */ 140 141 .macro magic_find_zero_bytes w1 142 /* Macro to find all-zero bytes in w1, result is in ip. */ 143 #if (defined (__ARM_FEATURE_DSP)) 144 uadd8 ip, \w1, r6 145 sel ip, r7, r6 146 #else /* not defined (__ARM_FEATURE_DSP) */ 147 /* __ARM_FEATURE_DSP is not defined for some Cortex-M processors. 148 Coincidently, these processors only have Thumb-2 mode, where we can use the 149 the (large) magic constant available directly as an immediate in instructions. 150 Note that we cannot use the magic constant in ARM mode, where we need 151 to create the constant in a register. */ 152 sub ip, \w1, #0x01010101 153 bic ip, ip, \w1 154 and ip, ip, #0x80808080 155 #endif /* not defined (__ARM_FEATURE_DSP) */ 156 .endm /* magic_find_zero_bytes */ 157 158 .macro setup_return w1 w2 159 #ifdef __ARMEB__ 160 mov r1, \w1 161 mov r2, \w2 162 #else /* not __ARMEB__ */ 163 rev r1, \w1 164 rev r2, \w2 165 #endif /* not __ARMEB__ */ 166 .endm /* setup_return */ 167 168 pld [r0, #0] 169 pld [r1, #0] 170 171 /* Are both strings double-word aligned? */ 172 orr ip, r0, r1 173 tst ip, #7 174 bne do_align 175 176 /* Fast path. */ 177 init 178 179 doubleword_aligned: 180 181 /* Get here when the strings to compare are double-word aligned. */ 182 /* Compare two words in every iteration. */ 183 .p2align 2 184 2: 185 pld [r0, #16] 186 pld [r1, #16] 187 188 /* Load the next double-word from each string. */ 189 ldrd r2, r3, [r0], #8 190 ldrd r4, r5, [r1], #8 191 192 magic_compare_and_branch w1=r2, w2=r4, label=return_24 193 magic_compare_and_branch w1=r3, w2=r5, label=return_35 194 b 2b 195 196 do_align: 197 /* Is the first string word-aligned? */ 198 ands ip, r0, #3 199 beq word_aligned_r0 200 201 /* Fast compare byte by byte until the first string is word-aligned. */ 202 /* The offset of r0 from a word boundary is in ip. Thus, the number of bytes 203 to read until the next word boundary is 4-ip. */ 204 bic r0, r0, #3 205 ldr r2, [r0], #4 206 lsls ip, ip, #31 207 beq byte2 208 bcs byte3 209 210 byte1: 211 ldrb ip, [r1], #1 212 uxtb r3, r2, ror #BYTE1_OFFSET 213 subs ip, r3, ip 214 bne fast_return 215 m_cbz reg=r3, label=fast_return 216 217 byte2: 218 ldrb ip, [r1], #1 219 uxtb r3, r2, ror #BYTE2_OFFSET 220 subs ip, r3, ip 221 bne fast_return 222 m_cbz reg=r3, label=fast_return 223 224 byte3: 225 ldrb ip, [r1], #1 226 uxtb r3, r2, ror #BYTE3_OFFSET 227 subs ip, r3, ip 228 bne fast_return 229 m_cbnz reg=r3, label=word_aligned_r0 230 231 fast_return: 232 mov r0, ip 233 bx lr 234 235 word_aligned_r0: 236 init 237 /* The first string is word-aligned. */ 238 /* Is the second string word-aligned? */ 239 ands ip, r1, #3 240 bne strcmp_unaligned 241 242 word_aligned: 243 /* The strings are word-aligned. */ 244 /* Is the first string double-word aligned? */ 245 tst r0, #4 246 beq doubleword_aligned_r0 247 248 /* If r0 is not double-word aligned yet, align it by loading 249 and comparing the next word from each string. */ 250 ldr r2, [r0], #4 251 ldr r4, [r1], #4 252 magic_compare_and_branch w1=r2 w2=r4 label=return_24 253 254 doubleword_aligned_r0: 255 /* Get here when r0 is double-word aligned. */ 256 /* Is r1 doubleword_aligned? */ 257 tst r1, #4 258 beq doubleword_aligned 259 260 /* Get here when the strings to compare are word-aligned, 261 r0 is double-word aligned, but r1 is not double-word aligned. */ 262 263 /* Initialize the queue. */ 264 ldr r5, [r1], #4 265 266 /* Compare two words in every iteration. */ 267 .p2align 2 268 3: 269 pld [r0, #16] 270 pld [r1, #16] 271 272 /* Load the next double-word from each string and compare. */ 273 ldrd r2, r3, [r0], #8 274 magic_compare_and_branch w1=r2 w2=r5 label=return_25 275 ldrd r4, r5, [r1], #8 276 magic_compare_and_branch w1=r3 w2=r4 label=return_34 277 b 3b 278 279 .macro miscmp_word offsetlo offsethi 280 /* Macro to compare misaligned strings. */ 281 /* r0, r1 are word-aligned, and at least one of the strings 282 is not double-word aligned. */ 283 /* Compare one word in every loop iteration. */ 284 /* OFFSETLO is the original bit-offset of r1 from a word-boundary, 285 OFFSETHI is 32 - OFFSETLO (i.e., offset from the next word). */ 286 287 /* Initialize the shift queue. */ 288 ldr r5, [r1], #4 289 290 /* Compare one word from each string in every loop iteration. */ 291 .p2align 2 292 7: 293 ldr r3, [r0], #4 294 S2LOMEM r5, r5, #\offsetlo 295 magic_find_zero_bytes w1=r3 296 cmp r7, ip, S2HIMEM #\offsetlo 297 and r2, r3, r6, S2LOMEM #\offsetlo 298 it eq 299 cmpeq r2, r5 300 bne return_25 301 ldr r5, [r1], #4 302 cmp ip, #0 303 eor r3, r2, r3 304 S2HIMEM r2, r5, #\offsethi 305 it eq 306 cmpeq r3, r2 307 bne return_32 308 b 7b 309 .endm /* miscmp_word */ 310 311 strcmp_unaligned: 312 /* r0 is word-aligned, r1 is at offset ip from a word. */ 313 /* Align r1 to the (previous) word-boundary. */ 314 bic r1, r1, #3 315 316 /* Unaligned comparison word by word using LDRs. */ 317 cmp ip, #2 318 beq miscmp_word_16 /* If ip == 2. */ 319 bge miscmp_word_24 /* If ip == 3. */ 320 miscmp_word offsetlo=8 offsethi=24 /* If ip == 1. */ 321 miscmp_word_16: miscmp_word offsetlo=16 offsethi=16 322 miscmp_word_24: miscmp_word offsetlo=24 offsethi=8 323 324 325 return_32: 326 setup_return w1=r3, w2=r2 327 b do_return 328 return_34: 329 setup_return w1=r3, w2=r4 330 b do_return 331 return_25: 332 setup_return w1=r2, w2=r5 333 b do_return 334 return_35: 335 setup_return w1=r3, w2=r5 336 b do_return 337 return_24: 338 setup_return w1=r2, w2=r4 339 340 do_return: 341 342 #ifdef __ARMEB__ 343 mov r0, ip 344 #else /* not __ARMEB__ */ 345 rev r0, ip 346 #endif /* not __ARMEB__ */ 347 348 /* Restore temporaries early, before computing the return value. */ 349 ldrd r6, r7, [sp] 350 ldrd r4, r5, [sp, #8] 351 adds sp, sp, #16 352 353 /* There is a zero or a different byte between r1 and r2. */ 354 /* r0 contains a mask of all-zero bytes in r1. */ 355 /* Using r0 and not ip here because cbz requires low register. */ 356 m_cbz reg=r0, label=compute_return_value 357 clz r0, r0 358 /* r0 contains the number of bits on the left of the first all-zero byte in r1. */ 359 rsb r0, r0, #24 360 /* Here, r0 contains the number of bits on the right of the first all-zero byte in r1. */ 361 lsr r1, r1, r0 362 lsr r2, r2, r0 363 364 compute_return_value: 365 movs r0, #1 366 cmp r1, r2 367 /* The return value is computed as follows. 368 If r1>r2 then (C==1 and Z==0) and LS doesn't hold and r0 is #1 at return. 369 If r1<r2 then (C==0 and Z==0) and we execute SBC with carry_in=0, 370 which means r0:=r0-r0-1 and r0 is #-1 at return. 371 If r1=r2 then (C==1 and Z==1) and we execute SBC with carry_in=1, 372 which means r0:=r0-r0 and r0 is #0 at return. 373 (C==0 and Z==1) cannot happen because the carry bit is "not borrow". */ 374 it ls 375 sbcls r0, r0, r0 376 bx lr 377 END(strcmp) 378
