zhichang.yuan | d875c9b | 2014-04-28 06:11:32 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2013 ARM Ltd. |
| 3 | * Copyright (C) 2013 Linaro. |
| 4 | * |
| 5 | * This code is based on glibc cortex strings work originally authored by Linaro |
| 6 | * and re-licensed under GPLv2 for the Linux kernel. The original code can |
| 7 | * be found @ |
| 8 | * |
| 9 | * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/ |
| 10 | * files/head:/src/aarch64/ |
| 11 | * |
| 12 | * This program is free software; you can redistribute it and/or modify |
| 13 | * it under the terms of the GNU General Public License version 2 as |
| 14 | * published by the Free Software Foundation. |
| 15 | * |
| 16 | * This program is distributed in the hope that it will be useful, |
| 17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 19 | * GNU General Public License for more details. |
| 20 | * |
| 21 | * You should have received a copy of the GNU General Public License |
| 22 | * along with this program. If not, see <http://www.gnu.org/licenses/>. |
| 23 | */ |
| 24 | |
| 25 | #include <linux/linkage.h> |
| 26 | #include <asm/assembler.h> |
| 27 | |
| 28 | /* |
| 29 | * compare memory areas(when two memory areas' offset are different, |
| 30 | * alignment handled by the hardware) |
| 31 | * |
| 32 | * Parameters: |
| 33 | * x0 - const memory area 1 pointer |
| 34 | * x1 - const memory area 2 pointer |
| 35 | * x2 - the maximal compare byte length |
| 36 | * Returns: |
| 37 | * x0 - a compare result, maybe less than, equal to, or greater than ZERO |
| 38 | */ |
| 39 | |
| 40 | /* Parameters and result. */ |
| 41 | src1 .req x0 |
| 42 | src2 .req x1 |
| 43 | limit .req x2 |
| 44 | result .req x0 |
| 45 | |
| 46 | /* Internal variables. */ |
| 47 | data1 .req x3 |
| 48 | data1w .req w3 |
| 49 | data2 .req x4 |
| 50 | data2w .req w4 |
| 51 | has_nul .req x5 |
| 52 | diff .req x6 |
| 53 | endloop .req x7 |
| 54 | tmp1 .req x8 |
| 55 | tmp2 .req x9 |
| 56 | tmp3 .req x10 |
| 57 | pos .req x11 |
| 58 | limit_wd .req x12 |
| 59 | mask .req x13 |
| 60 | |
| 61 | ENTRY(memcmp) |
| 62 | cbz limit, .Lret0 |
| 63 | eor tmp1, src1, src2 |
| 64 | tst tmp1, #7 |
| 65 | b.ne .Lmisaligned8 |
| 66 | ands tmp1, src1, #7 |
| 67 | b.ne .Lmutual_align |
| 68 | sub limit_wd, limit, #1 /* limit != 0, so no underflow. */ |
| 69 | lsr limit_wd, limit_wd, #3 /* Convert to Dwords. */ |
| 70 | /* |
| 71 | * The input source addresses are at alignment boundary. |
| 72 | * Directly compare eight bytes each time. |
| 73 | */ |
| 74 | .Lloop_aligned: |
| 75 | ldr data1, [src1], #8 |
| 76 | ldr data2, [src2], #8 |
| 77 | .Lstart_realigned: |
| 78 | subs limit_wd, limit_wd, #1 |
| 79 | eor diff, data1, data2 /* Non-zero if differences found. */ |
| 80 | csinv endloop, diff, xzr, cs /* Last Dword or differences. */ |
| 81 | cbz endloop, .Lloop_aligned |
| 82 | |
| 83 | /* Not reached the limit, must have found a diff. */ |
| 84 | tbz limit_wd, #63, .Lnot_limit |
| 85 | |
| 86 | /* Limit % 8 == 0 => the diff is in the last 8 bytes. */ |
| 87 | ands limit, limit, #7 |
| 88 | b.eq .Lnot_limit |
| 89 | /* |
| 90 | * The remained bytes less than 8. It is needed to extract valid data |
| 91 | * from last eight bytes of the intended memory range. |
| 92 | */ |
| 93 | lsl limit, limit, #3 /* bytes-> bits. */ |
| 94 | mov mask, #~0 |
| 95 | CPU_BE( lsr mask, mask, limit ) |
| 96 | CPU_LE( lsl mask, mask, limit ) |
| 97 | bic data1, data1, mask |
| 98 | bic data2, data2, mask |
| 99 | |
| 100 | orr diff, diff, mask |
| 101 | b .Lnot_limit |
| 102 | |
| 103 | .Lmutual_align: |
| 104 | /* |
| 105 | * Sources are mutually aligned, but are not currently at an |
| 106 | * alignment boundary. Round down the addresses and then mask off |
| 107 | * the bytes that precede the start point. |
| 108 | */ |
| 109 | bic src1, src1, #7 |
| 110 | bic src2, src2, #7 |
| 111 | ldr data1, [src1], #8 |
| 112 | ldr data2, [src2], #8 |
| 113 | /* |
| 114 | * We can not add limit with alignment offset(tmp1) here. Since the |
| 115 | * addition probably make the limit overflown. |
| 116 | */ |
| 117 | sub limit_wd, limit, #1/*limit != 0, so no underflow.*/ |
| 118 | and tmp3, limit_wd, #7 |
| 119 | lsr limit_wd, limit_wd, #3 |
| 120 | add tmp3, tmp3, tmp1 |
| 121 | add limit_wd, limit_wd, tmp3, lsr #3 |
| 122 | add limit, limit, tmp1/* Adjust the limit for the extra. */ |
| 123 | |
| 124 | lsl tmp1, tmp1, #3/* Bytes beyond alignment -> bits.*/ |
| 125 | neg tmp1, tmp1/* Bits to alignment -64. */ |
| 126 | mov tmp2, #~0 |
| 127 | /*mask off the non-intended bytes before the start address.*/ |
| 128 | CPU_BE( lsl tmp2, tmp2, tmp1 )/*Big-endian.Early bytes are at MSB*/ |
| 129 | /* Little-endian. Early bytes are at LSB. */ |
| 130 | CPU_LE( lsr tmp2, tmp2, tmp1 ) |
| 131 | |
| 132 | orr data1, data1, tmp2 |
| 133 | orr data2, data2, tmp2 |
| 134 | b .Lstart_realigned |
| 135 | |
| 136 | /*src1 and src2 have different alignment offset.*/ |
| 137 | .Lmisaligned8: |
| 138 | cmp limit, #8 |
| 139 | b.lo .Ltiny8proc /*limit < 8: compare byte by byte*/ |
| 140 | |
| 141 | and tmp1, src1, #7 |
| 142 | neg tmp1, tmp1 |
| 143 | add tmp1, tmp1, #8/*valid length in the first 8 bytes of src1*/ |
| 144 | and tmp2, src2, #7 |
| 145 | neg tmp2, tmp2 |
| 146 | add tmp2, tmp2, #8/*valid length in the first 8 bytes of src2*/ |
| 147 | subs tmp3, tmp1, tmp2 |
| 148 | csel pos, tmp1, tmp2, hi /*Choose the maximum.*/ |
| 149 | |
| 150 | sub limit, limit, pos |
| 151 | /*compare the proceeding bytes in the first 8 byte segment.*/ |
| 152 | .Ltinycmp: |
| 153 | ldrb data1w, [src1], #1 |
| 154 | ldrb data2w, [src2], #1 |
| 155 | subs pos, pos, #1 |
| 156 | ccmp data1w, data2w, #0, ne /* NZCV = 0b0000. */ |
| 157 | b.eq .Ltinycmp |
| 158 | cbnz pos, 1f /*diff occurred before the last byte.*/ |
| 159 | cmp data1w, data2w |
| 160 | b.eq .Lstart_align |
| 161 | 1: |
| 162 | sub result, data1, data2 |
| 163 | ret |
| 164 | |
| 165 | .Lstart_align: |
| 166 | lsr limit_wd, limit, #3 |
| 167 | cbz limit_wd, .Lremain8 |
| 168 | |
| 169 | ands xzr, src1, #7 |
| 170 | b.eq .Lrecal_offset |
| 171 | /*process more leading bytes to make src1 aligned...*/ |
| 172 | add src1, src1, tmp3 /*backwards src1 to alignment boundary*/ |
| 173 | add src2, src2, tmp3 |
| 174 | sub limit, limit, tmp3 |
| 175 | lsr limit_wd, limit, #3 |
| 176 | cbz limit_wd, .Lremain8 |
| 177 | /*load 8 bytes from aligned SRC1..*/ |
| 178 | ldr data1, [src1], #8 |
| 179 | ldr data2, [src2], #8 |
| 180 | |
| 181 | subs limit_wd, limit_wd, #1 |
| 182 | eor diff, data1, data2 /*Non-zero if differences found.*/ |
| 183 | csinv endloop, diff, xzr, ne |
| 184 | cbnz endloop, .Lunequal_proc |
| 185 | /*How far is the current SRC2 from the alignment boundary...*/ |
| 186 | and tmp3, tmp3, #7 |
| 187 | |
| 188 | .Lrecal_offset:/*src1 is aligned now..*/ |
| 189 | neg pos, tmp3 |
| 190 | .Lloopcmp_proc: |
| 191 | /* |
| 192 | * Divide the eight bytes into two parts. First,backwards the src2 |
| 193 | * to an alignment boundary,load eight bytes and compare from |
| 194 | * the SRC2 alignment boundary. If all 8 bytes are equal,then start |
| 195 | * the second part's comparison. Otherwise finish the comparison. |
| 196 | * This special handle can garantee all the accesses are in the |
| 197 | * thread/task space in avoid to overrange access. |
| 198 | */ |
| 199 | ldr data1, [src1,pos] |
| 200 | ldr data2, [src2,pos] |
| 201 | eor diff, data1, data2 /* Non-zero if differences found. */ |
| 202 | cbnz diff, .Lnot_limit |
| 203 | |
| 204 | /*The second part process*/ |
| 205 | ldr data1, [src1], #8 |
| 206 | ldr data2, [src2], #8 |
| 207 | eor diff, data1, data2 /* Non-zero if differences found. */ |
| 208 | subs limit_wd, limit_wd, #1 |
| 209 | csinv endloop, diff, xzr, ne/*if limit_wd is 0,will finish the cmp*/ |
| 210 | cbz endloop, .Lloopcmp_proc |
| 211 | .Lunequal_proc: |
| 212 | cbz diff, .Lremain8 |
| 213 | |
| 214 | /*There is differnence occured in the latest comparison.*/ |
| 215 | .Lnot_limit: |
| 216 | /* |
| 217 | * For little endian,reverse the low significant equal bits into MSB,then |
| 218 | * following CLZ can find how many equal bits exist. |
| 219 | */ |
| 220 | CPU_LE( rev diff, diff ) |
| 221 | CPU_LE( rev data1, data1 ) |
| 222 | CPU_LE( rev data2, data2 ) |
| 223 | |
| 224 | /* |
| 225 | * The MS-non-zero bit of DIFF marks either the first bit |
| 226 | * that is different, or the end of the significant data. |
| 227 | * Shifting left now will bring the critical information into the |
| 228 | * top bits. |
| 229 | */ |
| 230 | clz pos, diff |
| 231 | lsl data1, data1, pos |
| 232 | lsl data2, data2, pos |
| 233 | /* |
| 234 | * We need to zero-extend (char is unsigned) the value and then |
| 235 | * perform a signed subtraction. |
| 236 | */ |
| 237 | lsr data1, data1, #56 |
| 238 | sub result, data1, data2, lsr #56 |
| 239 | ret |
| 240 | |
| 241 | .Lremain8: |
| 242 | /* Limit % 8 == 0 =>. all data are equal.*/ |
| 243 | ands limit, limit, #7 |
| 244 | b.eq .Lret0 |
| 245 | |
| 246 | .Ltiny8proc: |
| 247 | ldrb data1w, [src1], #1 |
| 248 | ldrb data2w, [src2], #1 |
| 249 | subs limit, limit, #1 |
| 250 | |
| 251 | ccmp data1w, data2w, #0, ne /* NZCV = 0b0000. */ |
| 252 | b.eq .Ltiny8proc |
| 253 | sub result, data1, data2 |
| 254 | ret |
| 255 | .Lret0: |
| 256 | mov result, #0 |
| 257 | ret |
Ard Biesheuvel | 2079184 | 2015-10-08 20:02:03 +0100 | [diff] [blame] | 258 | ENDPIPROC(memcmp) |