1/******************************************************************************* 2 * 3 * Copyright (c) 1993 Intel Corporation 4 * 5 * Intel hereby grants you permission to copy, modify, and distribute this 6 * software and its documentation. Intel grants this permission provided 7 * that the above copyright notice appears in all copies and that both the 8 * copyright notice and this permission notice appear in supporting 9 * documentation. In addition, Intel grants this permission provided that 10 * you prominently mark as "not part of the original" any modifications 11 * made to this software or documentation, and that the name of Intel 12 * Corporation not be used in advertising or publicity pertaining to 13 * distribution of the software or the documentation without specific, 14 * written prior permission. 15 * 16 * Intel Corporation provides this AS IS, WITHOUT ANY WARRANTY, EXPRESS OR 17 * IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTY OF MERCHANTABILITY 18 * OR FITNESS FOR A PARTICULAR PURPOSE. Intel makes no guarantee or 19 * representations regarding the use of, or the results of the use of, 20 * the software and documentation in terms of correctness, accuracy, 21 * reliability, currentness, or otherwise; and you rely on the software, 22 * documentation and results solely at your own risk. 23 * 24 * IN NO EVENT SHALL INTEL BE LIABLE FOR ANY LOSS OF USE, LOSS OF BUSINESS, 25 * LOSS OF PROFITS, INDIRECT, INCIDENTAL, SPECIAL OR CONSEQUENTIAL DAMAGES 26 * OF ANY KIND. IN NO EVENT SHALL INTEL'S TOTAL LIABILITY EXCEED THE SUM 27 * PAID TO INTEL FOR THE PRODUCT LICENSED HEREUNDER. 28 * 29 ******************************************************************************/ 30 31 .file "strcm_ca.s" 32#ifdef __PIC 33 .pic 34#endif 35#ifdef __PID 36 .pid 37#endif 38/* 39 * (c) copyright 1988,1993 Intel Corp., all rights reserved 40 */ 41 42/* 43 procedure strcmp (optimized assembler version for the CA) 44 45 result = strcmp (src1_addr, src2_addr) 46 47 compare the null terminated string pointed to by src1_addr to 48 the string space pointed to by src2_addr. Return 0 iff the strings 49 are equal, -1 if src1_addr is lexicly less than src2_addr, and 1 50 if it is lexicly greater. 51 52 Undefined behavior will occur if the end of either source string 53 (i.e. the terminating null byte) is in the last word of the program's 54 allocated memory space. This is so because, in several cases, strcmp 55 will fetch ahead one word. Disallowing the fetch ahead would impose 56 a severe performance penalty. 57 58 This program handles five cases: 59 60 1) both arguments start on a word boundary 61 2) neither are word aligned, but they are offset by the same amount 62 3) source1 is word aligned, source2 is not 63 4) source2 is word aligned, source1 is not 64 5) neither is word aligned, and they are offset by differing amounts 65 66 At the time of this writing, only g0 thru g7 and g14 are available 67 for use in this leafproc; other registers would have to be saved and 68 restored. These nine registers are sufficient to implement the routine. 69 The registers are used as follows: 70 71 g0 original src1 ptr; return result 72 g1 src2 ptr; 0xff -- byte extraction mask 73 g2 src1 word ptr 74 g3 src2 word ptr 75 Little endian: 76 g4 lsw of src1 77 g5 msw of src1 78 g6 src2 word 79 g7 extracted src1 80 Big endian: 81 g4 msw of src1 82 g5 lsw of src1 83 g6 extracted src1 84 g7 src2 word 85 g13 return address 86 g14 shift count 87*/ 88 89#if __i960_BIG_ENDIAN__ 90#define MSW g4 91#define LSW g5 92#define SRC1 g6 93#define SRC2 g7 94#else 95#define LSW g4 96#define MSW g5 97#define SRC2 g6 98#define SRC1 g7 99#endif 100 101 .globl _strcmp 102 .globl __strcmp 103 .leafproc _strcmp, __strcmp 104 .align 2 105_strcmp: 106#ifndef __PIC 107 lda Lrett,g14 108#else 109 lda Lrett-(.+8)(ip),g14 110#endif 111 112__strcmp: 113Lrestart: 114 notand g0,3,g2 # extract word addr of start of src1 115 lda (g14),g13 # preserve return address 116#if __i960_BIG_ENDIAN__ 117 cmpo g0,g2 # check alignment of src1 118#endif 119 ld (g2),LSW # fetch word with at least first byte of src1 120 notand g1,3,g3 # extract word addr of start of src2 121 ld 4(g2),MSW # fetch second word of src1 122#if __i960_BIG_ENDIAN__ 123 bne Lsrc1_unaligned # branch if src1 is unaligned 124 cmpo g3,g1 # check alignment of src2 125 ld (g3),SRC2 # fetch word with at least first byte of src2 126 mov LSW,SRC1 # extract word of src1 127 lda 8(g2),g2 # advance src1 word addr 128 bne.f Lsrc2_unaligned # branch if src2 is NOT word aligned 129 130 /* src2 is word aligned */ 131 132Lwloop2: # word comparing loop 133 cmpo SRC2,SRC1 # compare src1 and src2 words 134 lda 0xff000000,g1 # byte extraction mask 135 mov MSW,LSW # move msw of src1 to lsw 136 ld (g2),MSW # pre-fetch next msw of src1 137 addo 4,g2,g2 # post-increment src1 addr 138 lda 4(g3),g3 # pre-increment src2 addr 139 bne.f Lcloop # branch if src1 and src2 unequal 140 scanbyte 0,SRC1 # check for null byte in src1 word 141 ld (g3),SRC2 # pre-fetch next word of src2 142 mov LSW,SRC1 # extract word of src1 143 lda 0,g0 # prepare to return zero, indicating equality 144 bno.t Lwloop2 # branch if null byte not encountered 145 146 /* words were equal and contained null byte */ 147 148 mov 0,g14 # conform to register conventions 149 bx (g13) # return 150 151 152Lsrc1_unaligned: 153#endif 154 cmpo g3,g1 # check alignment of src2 155 ld (g3),SRC2 # fetch word with at least first byte of src2 156 shlo 3,g0,g14 # compute shift count for src1 157#if __i960_BIG_ENDIAN__ 158 subo g14,0,g14 # 32 - shift count for big endian. 159#endif 160 eshro g14,g4,SRC1 # extract word of src1 161 lda 8(g2),g2 # advance src1 word addr 162 bne.f Lsrc2_unaligned # branch if src2 is NOT word aligned 163 164 /* at least src2 is word aligned */ 165 166Lwloop: # word comparing loop 167 cmpo SRC2,SRC1 # compare src1 and src2 words 168#if __i960_BIG_ENDIAN__ 169 lda 0xff000000,g1 # byte extraction mask 170#else 171 lda 0xff,g1 # byte extraction mask 172#endif 173 mov MSW,LSW # move msw of src1 to lsw 174 ld (g2),MSW # pre-fetch next msw of src1 175 addo 4,g2,g2 # post-increment src1 addr 176 lda 4(g3),g3 # pre-increment src2 addr 177 bne.f Lcloop # branch if src1 and src2 unequal 178 scanbyte 0,SRC1 # check for null byte in src1 word 179 ld (g3),SRC2 # pre-fetch next word of src2 180 eshro g14,g4,SRC1 # extract word of src1 181 lda 0,g0 # prepare to return zero, indicating equality 182 bno.t Lwloop # branch if null byte not encountered 183 184 /* words were equal and contained null byte */ 185 186 mov 0,g14 # conform to register conventions 187 bx (g13) # return 188 189Lcloop_setup: # setup for coming from Lsrc2_unaligned 190 mov LSW,SRC1 # restore extracted src1 word 191#if __i960_BIG_ENDIAN__ 192 lda 0xff000000,g1 # byte extraction mask 193#else 194 lda 0xff,g1 # byte extraction mask 195#endif 196 197Lcloop: # character comparing loop 198 and SRC2,g1,g3 # extract next char of src2 199 and SRC1,g1,g0 # extract next char of src1 200 cmpobne.f g0,g3,.diff # check for equality 201 cmpo 0,g0 # check for null byte 202#if __i960_BIG_ENDIAN__ 203 shro 8,g1,g1 # shift mask for next byte 204#else 205 shlo 8,g1,g1 # shift mask for next byte 206#endif 207 bne.t Lcloop # branch if null not reached 208 209 /* words are equal up thru null byte */ 210 211 mov 0,g14 212 bx (g13) # g0 = 0 (src1 == src2) 213Lrett: 214 ret 215 216.diff: 217 mov 0,g14 218 bl Lless_than_exit 219Lgreater_than_exit: 220 mov 1,g0 221 bx (g13) # g0 = 1 (src1 > src2) 222Lless_than_exit: 223 subi 1,0,g0 224 bx (g13) # g0 = -1 (src1 < src2) 225 226Lsrc2_unaligned: 227 mov SRC1,LSW # retain src1 extracted word 228 ld 4(g3),SRC1 # fetch second word of src2 229 shlo 3,g1,MSW # compute shift count for src2 230#if __i960_BIG_ENDIAN__ 231 subo MSW,0,MSW # 32 - shift count for big endian. 232#endif 233 eshro MSW,g6,SRC2 # extract word of src2 234 cmpo LSW,SRC2 # compare src1 and src2 words 235 notor g1,3,MSW # first step in computing new src1 ptr 236 lda 4(g3),g1 # set new src2 ptr 237 bne.f Lcloop_setup # first four bytes differ 238 scanbyte 0,LSW # check for null byte 239 lda (g13),g14 # prepare return pointer for Lrestart 240 subo MSW,g0,g0 # second (final) step in computing new src1 ptr 241 bno.t Lrestart # if null byte not encountered, continue 242 /* with both string fetches shifted such that */ 243 /* src2 is now word aligned. */ 244 mov 0,g14 # conform to register conventions. 245 lda 0,g0 # return indicator of equality. 246 bx (g13) 247
