projects / thirdparty / glibc.git / blame
| Commit | Line | Data |
|---|---|---|
| 14d941e4 MS |
1 | /* memcmp - compare memory |
| 2 | ||
| d4697bc9 | 3 | Copyright (C) 2013-2014 Free Software Foundation, Inc. |
| 14d941e4 MS |
4 | |
| 5 | This file is part of the GNU C Library. | |
| 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, see | |
| 19 | <http://www.gnu.org/licenses/>. */ | |
| 20 | ||
| 21 | #include <sysdep.h> | |
| 22 | ||
| 23 | /* Assumptions: | |
| 24 | * | |
| 25 | * ARMv8-a, AArch64 | |
| 26 | */ | |
| 27 | ||
| 28 | /* Parameters and result. */ | |
| 29 | #define src1 x0 | |
| 30 | #define src2 x1 | |
| 31 | #define limit x2 | |
| 32 | #define result x0 | |
| 33 | ||
| 34 | /* Internal variables. */ | |
| 35 | #define data1 x3 | |
| 36 | #define data1w w3 | |
| 37 | #define data2 x4 | |
| 38 | #define data2w w4 | |
| 39 | #define has_nul x5 | |
| 40 | #define diff x6 | |
| 41 | #define endloop x7 | |
| 42 | #define tmp1 x8 | |
| 43 | #define tmp2 x9 | |
| 44 | #define tmp3 x10 | |
| 45 | #define pos x11 | |
| 46 | #define limit_wd x12 | |
| 47 | #define mask x13 | |
| 48 | ||
| 49 | ENTRY_ALIGN (memcmp, 6) | |
| 50 | cbz limit, L(ret0) | |
| 51 | eor tmp1, src1, src2 | |
| 52 | tst tmp1, #7 | |
| 53 | b.ne L(misaligned8) | |
| 54 | ands tmp1, src1, #7 | |
| 55 | b.ne L(mutual_align) | |
| 56 | add limit_wd, limit, #7 | |
| 57 | lsr limit_wd, limit_wd, #3 | |
| 58 | /* Start of performance-critical section -- one 64B cache line. */ | |
| 59 | L(loop_aligned): | |
| 60 | ldr data1, [src1], #8 | |
| 61 | ldr data2, [src2], #8 | |
| 62 | L(start_realigned): | |
| 63 | subs limit_wd, limit_wd, #1 | |
| 64 | eor diff, data1, data2 /* Non-zero if differences found. */ | |
| 65 | csinv endloop, diff, xzr, ne /* Last Dword or differences. */ | |
| 66 | cbz endloop, L(loop_aligned) | |
| 67 | /* End of performance-critical section -- one 64B cache line. */ | |
| 68 | ||
| 69 | /* Not reached the limit, must have found a diff. */ | |
| 70 | cbnz limit_wd, L(not_limit) | |
| 71 | ||
| 72 | /* Limit % 8 == 0 => all bytes significant. */ | |
| 73 | ands limit, limit, #7 | |
| 74 | b.eq L(not_limit) | |
| 75 | ||
| 76 | lsl limit, limit, #3 /* Bits -> bytes. */ | |
| 77 | mov mask, #~0 | |
| 78 | #ifdef __AARCH64EB__ | |
| 79 | lsr mask, mask, limit | |
| 80 | #else | |
| 81 | lsl mask, mask, limit | |
| 82 | #endif | |
| 83 | bic data1, data1, mask | |
| 84 | bic data2, data2, mask | |
| 85 | ||
| 86 | orr diff, diff, mask | |
| 87 | L(not_limit): | |
| 88 | ||
| 89 | #ifndef __AARCH64EB__ | |
| 90 | rev diff, diff | |
| 91 | rev data1, data1 | |
| 92 | rev data2, data2 | |
| 93 | #endif | |
| 94 | /* The MS-non-zero bit of DIFF marks either the first bit | |
| 95 | that is different, or the end of the significant data. | |
| 96 | Shifting left now will bring the critical information into the | |
| 97 | top bits. */ | |
| 98 | clz pos, diff | |
| 99 | lsl data1, data1, pos | |
| 100 | lsl data2, data2, pos | |
| 101 | /* But we need to zero-extend (char is unsigned) the value and then | |
| 102 | perform a signed 32-bit subtraction. */ | |
| 103 | lsr data1, data1, #56 | |
| 104 | sub result, data1, data2, lsr #56 | |
| 105 | RET | |
| 106 | ||
| 107 | L(mutual_align): | |
| 108 | /* Sources are mutually aligned, but are not currently at an | |
| 109 | alignment boundary. Round down the addresses and then mask off | |
| 110 | the bytes that precede the start point. */ | |
| 111 | bic src1, src1, #7 | |
| 112 | bic src2, src2, #7 | |
| 113 | add limit, limit, tmp1 /* Adjust the limit for the extra. */ | |
| 114 | lsl tmp1, tmp1, #3 /* Bytes beyond alignment -> bits. */ | |
| 115 | ldr data1, [src1], #8 | |
| 116 | neg tmp1, tmp1 /* Bits to alignment -64. */ | |
| 117 | ldr data2, [src2], #8 | |
| 118 | mov tmp2, #~0 | |
| 119 | #ifdef __AARCH64EB__ | |
| 120 | /* Big-endian. Early bytes are at MSB. */ | |
| 121 | lsl tmp2, tmp2, tmp1 /* Shift (tmp1 & 63). */ | |
| 122 | #else | |
| 123 | /* Little-endian. Early bytes are at LSB. */ | |
| 124 | lsr tmp2, tmp2, tmp1 /* Shift (tmp1 & 63). */ | |
| 125 | #endif | |
| 126 | add limit_wd, limit, #7 | |
| 127 | orr data1, data1, tmp2 | |
| 128 | orr data2, data2, tmp2 | |
| 129 | lsr limit_wd, limit_wd, #3 | |
| 130 | b L(start_realigned) | |
| 131 | ||
| 132 | L(ret0): | |
| 133 | mov result, #0 | |
| 134 | RET | |
| 135 | ||
| 136 | .p2align 6 | |
| 137 | L(misaligned8): | |
| 138 | sub limit, limit, #1 | |
| 139 | 1: | |
| 140 | /* Perhaps we can do better than this. */ | |
| 141 | ldrb data1w, [src1], #1 | |
| 142 | ldrb data2w, [src2], #1 | |
| 143 | subs limit, limit, #1 | |
| 144 | ccmp data1w, data2w, #0, cs /* NZCV = 0b0000. */ | |
| 145 | b.eq 1b | |
| 146 | sub result, data1, data2 | |
| 147 | RET | |
| 148 | END (memcmp) | |
| 149 | #undef bcmp | |
| 150 | weak_alias (memcmp, bcmp) | |
| 151 | libc_hidden_builtin_def (memcmp) |