memcmp.S revision 1.1
1 /* $NetBSD: memcmp.S,v 1.1 2008/02/21 17:35:47 garbled Exp $ */ 2 3 /* stropt/memcmp.S, pl_string_common, pl_linux 10/11/04 11:45:35 4 * ========================================================================== 5 * Optimized memcmp implementation for IBM PowerPC 405/440. 6 * 7 * Copyright (c) 2003, IBM Corporation 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * * Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * * Redistributions in binary form must reproduce the above 18 * copyright notice, this list of conditions and the following 19 * disclaimer in the documentation and/or other materials 20 * provided with the distribution. 21 * * Neither the name of IBM nor the names of its contributors 22 * may be used to endorse or promote products derived from this 23 * software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND 26 * CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, 27 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 28 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 29 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, 31 * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 32 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 33 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY 34 * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 35 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE 36 * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 37 * 38 * ========================================================================== 39 * 40 * Function: Compare two character strings (up to n characters) 41 * 42 * int memcmp(const char *s1, const char *s2, int n) 43 * 44 * Input: r3 - buffer 1 address 45 * r4 - buffer 2 address 46 * r5 - maximum characters to compare 47 * Output: r3 <0 (less), 0 (equal), >0 (greater) 48 * 49 * ========================================================================== 50 */ 51 52 #define _NOREGNAMES 53 #include <machine/asm.h> 54 #ifdef _KERNEL 55 #include <assym.h> 56 #endif 57 58 .text 59 .align 4 60 /* LINTSTUB: Func: void *memcmp(const void *, const void *, size_t) */ 61 ENTRY(memcmp) 62 63 /* 64 * Check count passed in R5. If zero, return 0; otherwise continue. 65 */ 66 cmpwi %r5,0 67 beq- ret_0; 68 69 /* 70 * Most of the time the difference is found in the first 71 * several bytes. The following code minimizes the number 72 * of load operations for short compares. 73 */ 74 75 mr %r11, %r3 /* Save buffer 1 */ 76 77 again: 78 79 not %r10, %r4 /* buffer 2: bytes to page bdy */ 80 rlwinm. %r10, %r10,29,23,31 /* buffer 2: dwords to page bdy */ 81 beq- bytebybyte /* If < 8 bytes to the page bdy */ 82 /* do byte by byte */ 83 lwz %r8, 0(%r4) /* load 1st buffer 2 word */ 84 85 not %r12, %r11 /* buffer 1: bytes to page bdy */ 86 rlwinm. %r12, %r12,29,23,31 /* buffer 1: dwords to page bdy */ 87 beq- bytebybyte /* If < 8 bytes to the page bdy */ 88 /* do byte by byte */ 89 lwz %r6, 0(%r11) /* load 1st buffer 1 word */ 90 91 cmpwi %r5, 4 /* If remaining count <= 4 */ 92 ble+ first4 /* handle specially. DWG */ 93 94 cmplw %r8, %r6 /* compare buffer 2 and buffer 1*/ 95 bne+ all_done /* different => we're done */ 96 97 lwzu %r9, 4(%r4) /* load 2nd buffer 2 word */ 98 lwzu %r7, 4(%r11) /* load 2nd buffer 1 word */ 99 100 cmpwi %r5, 8 /* If remaining count <= 8 */ 101 ble+ last4 /* handle specially. DWG */ 102 103 cmplw %r9, %r7 /* compare buffer 2 and buffer 1*/ 104 bne+ all_done /* different => we're done */ 105 106 addi %r5, %r5, -8 /* Update character counter DWG */ 107 addi %r10, %r4, 0x0004 /* DWG*/ 108 not %r10, %r10 /* buffer 2: bytes to page bdy DWG */ 109 rlwinm. %r10, %r10,29,23,31 /* buffer 2: dwords to page bdy DWG */ 110 addi %r12, %r11, 0x0004 /* DWG */ 111 not %r12, %r12 /* buffer 1: bytes to page bdy DWG */ 112 rlwinm. %r12, %r12,29,23,31 /* buffer 1: dwords to page bdy DWG */ 113 114 /* The following section prior to loop: figures out whether */ 115 /* the buffer 1 or buffer 2 is closer to the page boundary. */ 116 /* The main loop count is then set up to reflect the number of */ 117 /* double words of the buffer that is closest */ 118 119 cmpw %r10, %r12 /* Find closest */ 120 blt lt 121 122 mr %r10, %r12 123 124 lt: 125 126 srwi %r12, %r5, 3 /* Double check the total count */ 127 cmpw %r10, %r12 /* limitation */ 128 blt lt2 129 130 mr %r10, %r12 /* DWG */ 131 lt2: /* DWG */ 132 cmpwi %r10, 0 /* DWG */ 133 bne lt3 /* DWG */ 134 addi %r4, %r4, 0x0004 /* DWG */ 135 addi %r11,%r11,0x0004 /* DWG */ 136 b again /* DWG */ 137 lt3: /* DWG */ 138 mtctr %r10 /* dword count for loop */ 139 lwzu %r6, 4(%r11) /* pre-load buffer 1 word */ 140 141 b in /* To the loop */ 142 143 loop: /* main loop */ 144 145 cmplw %r8, %r6 /* Compare first buffer 2 word */ 146 bne- all_done /* with first buffer 1 word */ 147 /* If different, we're done */ 148 cmplw %r9, %r7 /* Compare second buffer 2 word */ 149 /* with second buffer 1 word */ 150 lwzu %r6, 4(%r11) /* pre-load buffer 1 word */ 151 152 bne- all_done /* If different, we're done */ 153 154 in: 155 156 lwzu %r7, 4(%r11) /* pre-load buffer 1 word */ 157 lwzu %r8, 4(%r4) /* pre-load buffer 2 word */ 158 lwzu %r9, 4(%r4) /* pre-load buffer 2 word */ 159 160 bdnz+ loop /* Do more DW's if cnt > 0 */ 161 162 /*mfctr %r12*/ /*DWG*/ /* number of dwords left */ 163 /*subf %r10, %r12, %r10*/ /*DWG*//* number of dwords compared */ 164 slwi %r10, %r10, 3 165 subf %r5, %r10, %r5 /* adjust byte counter */ 166 /*bne+ partial*/ /*DWG*/ /* If less than 8 bytes, handle */ 167 /* specially */ 168 /*cmpwi %r5, 8*/ /* Removed. DWG */ 169 /*blt partial*/ /* Removed. DWG */ 170 171 /*addic %r5, %r5, -8*/ /*DWG*/ /* Subtract two words from count*/ 172 173 cmplw %r8, %r6 /* compare last dword */ 174 addi %r4, %r4, 4 175 bne- all_done 176 177 cmplw %r9, %r7 178 addi %r11, %r11, 4 179 bne- all_done 180 181 bytebybyte: 182 183 /* We've gotten close to a page boundary: do a byte-byte-byte 184 * compare for the following 8 bytes, and then go back to 185 * the full-word compare loop. 186 */ 187 188 li %r3, 8 /* loop count */ 189 cmpw %r3, %r5 /* take min(8, counter) */ 190 ble f2 191 192 mr. %r3, %r5 193 194 beqlr 195 196 f2: 197 198 mtctr %r3 199 subf %r5, %r3, %r5 /* adjust counter */ 200 201 bbb: 202 203 lbz %r6, 0(%r11) /* byte copy loop */ 204 205 addi %r11, %r11, 1 206 207 lbz %r8, 0(%r4) 208 209 addi %r4, %r4, 1 210 211 cmplw %r8, %r6 212 213 bdnzt+ eq, bbb 214 215 bne all_done 216 217 cmpwi %r5, 0 218 bgt again /* handle the rest */ 219 220 xor %r3,%r3,%r3 221 222 blr 223 224 #if 0 /* Removed code section. DWG */ 225 partial: 226 227 mr. %r3, %r5 228 229 beqlr /* If count -> 0, we're done */ 230 231 f1: 232 233 subfic %r3, %r3, 4 /* zero/end in first word? */ 234 cmpwi %r3, 0 235 blt last4 236 #endif /* DWG */ 237 238 first4: 239 subfic %r3, %r5, 4 /* If count <= 4, handle */ 240 rlwinm %r3, %r3, 3, 0, 31 /* count *= 8 */ 241 srw %r6, %r6, %r3 /* align 1st buffer 1 word */ 242 srw %r8, %r8, %r3 /* align 1st buffer 2 word */ 243 244 cmplw %r8, %r6 /* get result */ 245 bne all_done 246 xor %r3,%r3,%r3 247 blr 248 249 last4: 250 subfic %r10, %r5, 8 /*DWG*/ 251 rlwinm %r10, %r10, 3, 0, 31 /* count *= 8 */ 252 srw %r7, %r7, %r10 /* align 2nd buffer 1 word */ 253 srw %r9, %r9, %r10 /* align 2nd buffer 2 word */ 254 255 cmplw %r9, %r7 /* get result */ 256 bne all_done 257 ret_0: 258 xor %r3,%r3,%r3 /* Equal result */ 259 blr 260 261 all_done: 262 263 blt finish_lt 264 265 addi %r3,0,-1 /* Less than result */ 266 267 blr 268 269 finish_lt: 270 271 addi %r3,0,1 /* Greater than result */ 272 273 blr 274
Indexes created Sun Sep 21 20:09:37 GMT 2025