yuezonghe | 824eb0c | 2024-06-27 02:32:26 -0700 | [diff] [blame] | 1 | /* Determine the length of a string. For SPARC v9. |
| 2 | Copyright (C) 1998, 1999, 2003 Free Software Foundation, Inc. |
| 3 | This file is part of the GNU C Library. |
| 4 | Contributed by Jan Vondrak <jvon4518@ss1000.ms.mff.cuni.cz> and |
| 5 | Jakub Jelinek <jj@ultra.linux.cz>. |
| 6 | |
| 7 | The GNU C Library is free software; you can redistribute it and/or |
| 8 | modify it under the terms of the GNU Lesser General Public |
| 9 | License as published by the Free Software Foundation; either |
| 10 | version 2.1 of the License, or (at your option) any later version. |
| 11 | |
| 12 | The GNU C Library is distributed in the hope that it will be useful, |
| 13 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | Lesser General Public License for more details. |
| 16 | |
| 17 | You should have received a copy of the GNU Lesser General Public |
| 18 | License along with the GNU C Library; if not, write to the Free |
| 19 | Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA |
| 20 | 02111-1307 USA. */ |
| 21 | |
| 22 | #include <asm/asi.h> |
| 23 | |
| 24 | /* Normally, this uses |
| 25 | ((xword - 0x0101010101010101) & 0x8080808080808080) test |
| 26 | to find out if any byte in xword could be zero. This is fast, but |
| 27 | also gives false alarm for any byte in range 0x81-0xff. It does |
| 28 | not matter for correctness, as if this test tells us there could |
| 29 | be some zero byte, we check it byte by byte, but if bytes with |
| 30 | high bits set are common in the strings, then this will give poor |
| 31 | performance. You can #define EIGHTBIT_NOT_RARE and the algorithm |
| 32 | will use one tick slower, but more precise test |
| 33 | ((xword - 0x0101010101010101) & (~xword) & 0x8080808080808080), |
| 34 | which does not give any false alarms (but if some bits are set, |
| 35 | one cannot assume from it which bytes are zero and which are not). |
| 36 | It is yet to be measured, what is the correct default for glibc |
| 37 | in these days for an average user. |
| 38 | */ |
| 39 | |
| 40 | .text |
| 41 | .align 32 |
| 42 | ENTRY(strlen) |
| 43 | sethi %hi(0x01010101), %g1 /* IEU0 Group */ |
| 44 | ldub [%o0], %o3 /* Load */ |
| 45 | or %g1, %lo(0x01010101), %g1 /* IEU0 Group */ |
| 46 | mov %o0, %o1 /* IEU1 */ |
| 47 | |
| 48 | sllx %g1, 32, %g4 /* IEU0 Group */ |
| 49 | andcc %o0, 7, %g0 /* IEU1 */ |
| 50 | or %g1, %g4, %g1 /* IEU0 Group */ |
| 51 | brz,pn %o3, 13f /* CTI+IEU1 */ |
| 52 | |
| 53 | sllx %g1, 7, %g4 /* IEU0 Group */ |
| 54 | bne,a,pn %icc, 15f /* CTI */ |
| 55 | add %o0, 1, %o0 /* IEU1 */ |
| 56 | /* %g1 = 0x0101010101010101 * |
| 57 | * %g4 = 0x8080808080808080 * |
| 58 | * %o0 = string pointer * |
| 59 | * %o1 = start of string */ |
| 60 | 1: ldx [%o0], %o3 /* Load Group */ |
| 61 | |
| 62 | add %o0, 8, %o0 /* IEU1 */ |
| 63 | 2: sub %o3, %g1, %o2 /* IEU0 Group */ |
| 64 | #ifdef EIGHTBIT_NOT_RARE |
| 65 | andn %o2, %o3, %o5 /* IEU0 Group */ |
| 66 | ldxa [%o0] ASI_PNF, %o3 /* Load */ |
| 67 | andcc %o5, %g4, %g0 /* IEU1 Group */ |
| 68 | #else |
| 69 | ldxa [%o0] ASI_PNF, %o3 /* Load */ |
| 70 | andcc %o2, %g4, %g0 /* IEU1 Group */ |
| 71 | #endif |
| 72 | |
| 73 | be,pt %xcc, 2b /* CTI */ |
| 74 | add %o0, 8, %o0 /* IEU0 */ |
| 75 | addcc %o2, %g1, %g5 /* IEU1 Group */ |
| 76 | #ifdef EIGHTBIT_NOT_RARE |
| 77 | srlx %o5, 32, %o5 /* IEU0 */ |
| 78 | |
| 79 | 3: andcc %o5, %g4, %g0 /* IEU1 Group */ |
| 80 | #else |
| 81 | srlx %o2, 32, %o2 /* IEU0 */ |
| 82 | |
| 83 | 3: andcc %o2, %g4, %g0 /* IEU1 Group */ |
| 84 | #endif |
| 85 | be,pn %xcc, 4f /* CTI */ |
| 86 | srlx %g5, 56, %o2 /* IEU0 */ |
| 87 | andcc %o2, 0xff, %g0 /* IEU1 Group */ |
| 88 | |
| 89 | be,pn %icc, 12f /* CTI */ |
| 90 | srlx %g5, 48, %o2 /* IEU0 */ |
| 91 | andcc %o2, 0xff, %g0 /* IEU1 Group */ |
| 92 | be,pn %icc, 11f /* CTI */ |
| 93 | |
| 94 | srlx %g5, 40, %o2 /* IEU0 */ |
| 95 | andcc %o2, 0xff, %g0 /* IEU1 Group */ |
| 96 | be,pn %icc, 10f /* CTI */ |
| 97 | srlx %g5, 32, %o2 /* IEU0 */ |
| 98 | |
| 99 | andcc %o2, 0xff, %g0 /* IEU1 Group */ |
| 100 | be,pn %icc, 9f /* CTI */ |
| 101 | 4: srlx %g5, 24, %o2 /* IEU0 */ |
| 102 | andcc %o2, 0xff, %g0 /* IEU1 Group */ |
| 103 | |
| 104 | be,pn %icc, 8f /* CTI */ |
| 105 | srlx %g5, 16, %o2 /* IEU0 */ |
| 106 | andcc %o2, 0xff, %g0 /* IEU1 Group */ |
| 107 | be,pn %icc, 7f /* CTI */ |
| 108 | |
| 109 | srlx %g5, 8, %o2 /* IEU0 */ |
| 110 | andcc %o2, 0xff, %g0 /* IEU1 Group */ |
| 111 | be,pn %icc, 6f /* CTI */ |
| 112 | sub %o3, %g1, %o2 /* IEU0 */ |
| 113 | |
| 114 | andcc %g5, 0xff, %g0 /* IEU1 Group */ |
| 115 | be,pn %icc, 5f /* CTI */ |
| 116 | ldxa [%o0] ASI_PNF, %o3 /* Load */ |
| 117 | andcc %o2, %g4, %g0 /* IEU1 Group */ |
| 118 | |
| 119 | be,pt %xcc, 2b /* CTI */ |
| 120 | add %o0, 8, %o0 /* IEU0 */ |
| 121 | addcc %o2, %g1, %g5 /* IEU1 Group */ |
| 122 | ba,pt %xcc, 3b /* CTI */ |
| 123 | |
| 124 | srlx %o2, 32, %o2 /* IEU0 */ |
| 125 | 5: add %o0, -9, %o0 /* IEU0 Group */ |
| 126 | retl /* CTI+IEU1 Group */ |
| 127 | sub %o0, %o1, %o0 /* IEU0 */ |
| 128 | |
| 129 | 6: add %o0, -10, %o0 /* IEU0 Group */ |
| 130 | retl /* CTI+IEU1 Group */ |
| 131 | sub %o0, %o1, %o0 /* IEU0 */ |
| 132 | 7: add %o0, -11, %o0 /* IEU0 Group */ |
| 133 | |
| 134 | retl /* CTI+IEU1 Group */ |
| 135 | sub %o0, %o1, %o0 /* IEU0 */ |
| 136 | 8: add %o0, -12, %o0 /* IEU0 Group */ |
| 137 | retl /* CTI+IEU1 Group */ |
| 138 | |
| 139 | sub %o0, %o1, %o0 /* IEU0 */ |
| 140 | 9: add %o0, -13, %o0 /* IEU0 Group */ |
| 141 | retl /* CTI+IEU1 Group */ |
| 142 | sub %o0, %o1, %o0 /* IEU0 */ |
| 143 | |
| 144 | 10: add %o0, -14, %o0 /* IEU0 Group */ |
| 145 | retl /* CTI+IEU1 Group */ |
| 146 | sub %o0, %o1, %o0 /* IEU0 */ |
| 147 | 11: add %o0, -15, %o0 /* IEU0 Group */ |
| 148 | |
| 149 | retl /* CTI+IEU1 Group */ |
| 150 | sub %o0, %o1, %o0 /* IEU0 */ |
| 151 | 12: add %o0, -16, %o0 /* IEU0 Group */ |
| 152 | retl /* CTI+IEU1 Group */ |
| 153 | |
| 154 | sub %o0, %o1, %o0 /* IEU0 */ |
| 155 | 13: retl /* CTI+IEU1 Group */ |
| 156 | mov 0, %o0 /* IEU0 */ |
| 157 | nop |
| 158 | |
| 159 | 15: ldub [%o0], %o3 /* Load Group */ |
| 160 | 16: andcc %o0, 7, %g0 /* IEU1 */ |
| 161 | be,pn %icc, 1b /* CTI */ |
| 162 | nop /* IEU0 Group */ |
| 163 | |
| 164 | add %o0, 1, %o0 /* IEU1 */ |
| 165 | andcc %o3, 0xff, %g0 /* IEU1 Group */ |
| 166 | bne,a,pt %icc, 16b /* CTI */ |
| 167 | lduba [%o0] ASI_PNF, %o3 /* Load */ |
| 168 | |
| 169 | add %o0, -1, %o0 /* IEU0 Group */ |
| 170 | retl /* CTI+IEU1 Group */ |
| 171 | sub %o0, %o1, %o0 /* IEU0 */ |
| 172 | END(strlen) |
| 173 | libc_hidden_def(strlen) |