1/* 2 strchr - find a character in a string 3 4 Copyright (c) 2014, ARM Limited 5 All rights Reserved. 6 7 Redistribution and use in source and binary forms, with or without 8 modification, are permitted provided that the following conditions are met: 9 * Redistributions of source code must retain the above copyright 10 notice, this list of conditions and the following disclaimer. 11 * Redistributions in binary form must reproduce the above copyright 12 notice, this list of conditions and the following disclaimer in the 13 documentation and/or other materials provided with the distribution. 14 * Neither the name of the company nor the names of its contributors 15 may be used to endorse or promote products derived from this 16 software without specific prior written permission. 17 18 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 20 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 21 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 22 HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 23 SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 24 LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 28 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ 29 30/* Assumptions: 31 * 32 * ARMv8-a, AArch64 33 * Neon Available. 34 */ 35 36/* Arguments and results. */ 37#define srcin x0 38#define chrin w1 39 40#define result x0 41 42#define src x2 43#define tmp1 x3 44#define wtmp2 w4 45#define tmp3 x5 46 47#define vrepchr v0 48#define vdata1 v1 49#define vdata2 v2 50#define vhas_nul1 v3 51#define vhas_nul2 v4 52#define vhas_chr1 v5 53#define vhas_chr2 v6 54#define vrepmask_0 v7 55#define vrepmask_c v16 56#define vend1 v17 57#define vend2 v18 58 59/* Core algorithm. 60 61 For each 32-byte hunk we calculate a 64-bit syndrome value, with 62 two bits per byte (LSB is always in bits 0 and 1, for both big 63 and little-endian systems). For each tuple, bit 0 is set iff 64 the relevant byte matched the requested character; bit 1 is set 65 iff the relevant byte matched the NUL end of string (we trigger 66 off bit0 for the special case of looking for NUL). Since the bits 67 in the syndrome reflect exactly the order in which things occur 68 in the original string a count_trailing_zeros() operation will 69 identify exactly which byte is causing the termination, and why. */ 70 71/* Locals and temporaries. */ 72 73 .macro def_fn f p2align=0 74 .text 75 .p2align \p2align 76 .global \f 77 .type \f, %function 78\f: 79 .endm 80 81 .macro def_alias f a 82 .weak \a 83 .set \a,\f 84 .endm 85 86def_fn strchr 87def_alias strchr index 88 /* Magic constant 0x40100401 to allow us to identify which lane 89 matches the requested byte. Magic constant 0x80200802 used 90 similarly for NUL termination. */ 91 mov wtmp2, #0x0401 92 movk wtmp2, #0x4010, lsl #16 93 dup vrepchr.16b, chrin 94 bic src, srcin, #31 /* Work with aligned 32-byte hunks. */ 95 dup vrepmask_c.4s, wtmp2 96 ands tmp1, srcin, #31 97 add vrepmask_0.4s, vrepmask_c.4s, vrepmask_c.4s /* equiv: lsl #1 */ 98 b.eq .Lloop 99 100 /* Input string is not 32-byte aligned. Rather than forcing 101 the padding bytes to a safe value, we calculate the syndrome 102 for all the bytes, but then mask off those bits of the 103 syndrome that are related to the padding. */ 104 ld1 {vdata1.16b, vdata2.16b}, [src], #32 105 neg tmp1, tmp1 106 cmeq vhas_nul1.16b, vdata1.16b, #0 107 cmeq vhas_chr1.16b, vdata1.16b, vrepchr.16b 108 cmeq vhas_nul2.16b, vdata2.16b, #0 109 cmeq vhas_chr2.16b, vdata2.16b, vrepchr.16b 110 and vhas_nul1.16b, vhas_nul1.16b, vrepmask_0.16b 111 and vhas_nul2.16b, vhas_nul2.16b, vrepmask_0.16b 112 and vhas_chr1.16b, vhas_chr1.16b, vrepmask_c.16b 113 and vhas_chr2.16b, vhas_chr2.16b, vrepmask_c.16b 114 orr vend1.16b, vhas_nul1.16b, vhas_chr1.16b 115 orr vend2.16b, vhas_nul2.16b, vhas_chr2.16b 116 lsl tmp1, tmp1, #1 117 addp vend1.16b, vend1.16b, vend2.16b // 256->128 118 mov tmp3, #~0 119 addp vend1.16b, vend1.16b, vend2.16b // 128->64 120 lsr tmp1, tmp3, tmp1 121 122 mov tmp3, vend1.d[0] 123 bic tmp1, tmp3, tmp1 // Mask padding bits. 124 cbnz tmp1, .Ltail 125 126.Lloop: 127 ld1 {vdata1.16b, vdata2.16b}, [src], #32 128 cmeq vhas_nul1.16b, vdata1.16b, #0 129 cmeq vhas_chr1.16b, vdata1.16b, vrepchr.16b 130 cmeq vhas_nul2.16b, vdata2.16b, #0 131 cmeq vhas_chr2.16b, vdata2.16b, vrepchr.16b 132 /* Use a fast check for the termination condition. */ 133 orr vend1.16b, vhas_nul1.16b, vhas_chr1.16b 134 orr vend2.16b, vhas_nul2.16b, vhas_chr2.16b 135 orr vend1.16b, vend1.16b, vend2.16b 136 addp vend1.2d, vend1.2d, vend1.2d 137 mov tmp1, vend1.d[0] 138 cbz tmp1, .Lloop 139 140 /* Termination condition found. Now need to establish exactly why 141 we terminated. */ 142 and vhas_nul1.16b, vhas_nul1.16b, vrepmask_0.16b 143 and vhas_nul2.16b, vhas_nul2.16b, vrepmask_0.16b 144 and vhas_chr1.16b, vhas_chr1.16b, vrepmask_c.16b 145 and vhas_chr2.16b, vhas_chr2.16b, vrepmask_c.16b 146 orr vend1.16b, vhas_nul1.16b, vhas_chr1.16b 147 orr vend2.16b, vhas_nul2.16b, vhas_chr2.16b 148 addp vend1.16b, vend1.16b, vend2.16b // 256->128 149 addp vend1.16b, vend1.16b, vend2.16b // 128->64 150 151 mov tmp1, vend1.d[0] 152.Ltail: 153 /* Count the trailing zeros, by bit reversing... */ 154 rbit tmp1, tmp1 155 /* Re-bias source. */ 156 sub src, src, #32 157 clz tmp1, tmp1 /* And counting the leading zeros. */ 158 /* Tmp1 is even if the target charager was found first. Otherwise 159 we've found the end of string and we weren't looking for NUL. */ 160 tst tmp1, #1 161 add result, src, tmp1, lsr #1 162 csel result, result, xzr, eq 163 ret 164 165 .size strchr, . - strchr 166