| 1 | /* |
| 2 | * strrchr - find last position of a character in a string. |
| 3 | * |
| 4 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 5 | * See https://llvm.org/LICENSE.txt for license information. |
| 6 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 7 | */ |
| 8 | |
| 9 | /* Assumptions: |
| 10 | * |
| 11 | * ARMv8-a, AArch64 |
| 12 | * Neon Available. |
| 13 | */ |
| 14 | |
| 15 | #include "../asmdefs.h" |
| 16 | |
| 17 | /* Arguments and results. */ |
| 18 | #define srcin x0 |
| 19 | #define chrin w1 |
| 20 | |
| 21 | #define result x0 |
| 22 | |
| 23 | #define src x2 |
| 24 | #define tmp1 x3 |
| 25 | #define wtmp2 w4 |
| 26 | #define tmp3 x5 |
| 27 | #define src_match x6 |
| 28 | #define src_offset x7 |
| 29 | #define const_m1 x8 |
| 30 | #define tmp4 x9 |
| 31 | #define nul_match x10 |
| 32 | #define chr_match x11 |
| 33 | |
| 34 | #define vrepchr v0 |
| 35 | #define vdata1 v1 |
| 36 | #define vdata2 v2 |
| 37 | #define vhas_nul1 v3 |
| 38 | #define vhas_nul2 v4 |
| 39 | #define vhas_chr1 v5 |
| 40 | #define vhas_chr2 v6 |
| 41 | #define vrepmask_0 v7 |
| 42 | #define vrepmask_c v16 |
| 43 | #define vend1 v17 |
| 44 | #define vend2 v18 |
| 45 | |
| 46 | /* Core algorithm. |
| 47 | |
| 48 | For each 32-byte hunk we calculate a 64-bit syndrome value, with |
| 49 | two bits per byte (LSB is always in bits 0 and 1, for both big |
| 50 | and little-endian systems). For each tuple, bit 0 is set iff |
| 51 | the relevant byte matched the requested character; bit 1 is set |
| 52 | iff the relevant byte matched the NUL end of string (we trigger |
| 53 | off bit0 for the special case of looking for NUL). Since the bits |
| 54 | in the syndrome reflect exactly the order in which things occur |
| 55 | in the original string a count_trailing_zeros() operation will |
| 56 | identify exactly which byte is causing the termination, and why. */ |
| 57 | |
| 58 | ENTRY (__strrchr_aarch64) |
| 59 | /* Magic constant 0x40100401 to allow us to identify which lane |
| 60 | matches the requested byte. Magic constant 0x80200802 used |
| 61 | similarly for NUL termination. */ |
| 62 | mov wtmp2, #0x0401 |
| 63 | movk wtmp2, #0x4010, lsl #16 |
| 64 | dup vrepchr.16b, chrin |
| 65 | bic src, srcin, #31 /* Work with aligned 32-byte hunks. */ |
| 66 | dup vrepmask_c.4s, wtmp2 |
| 67 | mov src_offset, #0 |
| 68 | ands tmp1, srcin, #31 |
| 69 | add vrepmask_0.4s, vrepmask_c.4s, vrepmask_c.4s /* equiv: lsl #1 */ |
| 70 | b.eq L(aligned) |
| 71 | |
| 72 | /* Input string is not 32-byte aligned. Rather than forcing |
| 73 | the padding bytes to a safe value, we calculate the syndrome |
| 74 | for all the bytes, but then mask off those bits of the |
| 75 | syndrome that are related to the padding. */ |
| 76 | ld1 {vdata1.16b, vdata2.16b}, [src], #32 |
| 77 | neg tmp1, tmp1 |
| 78 | cmeq vhas_nul1.16b, vdata1.16b, #0 |
| 79 | cmeq vhas_chr1.16b, vdata1.16b, vrepchr.16b |
| 80 | cmeq vhas_nul2.16b, vdata2.16b, #0 |
| 81 | cmeq vhas_chr2.16b, vdata2.16b, vrepchr.16b |
| 82 | and vhas_nul1.16b, vhas_nul1.16b, vrepmask_0.16b |
| 83 | and vhas_chr1.16b, vhas_chr1.16b, vrepmask_c.16b |
| 84 | and vhas_nul2.16b, vhas_nul2.16b, vrepmask_0.16b |
| 85 | and vhas_chr2.16b, vhas_chr2.16b, vrepmask_c.16b |
| 86 | addp vhas_nul1.16b, vhas_nul1.16b, vhas_nul2.16b // 256->128 |
| 87 | addp vhas_chr1.16b, vhas_chr1.16b, vhas_chr2.16b // 256->128 |
| 88 | addp vhas_nul1.16b, vhas_nul1.16b, vhas_nul1.16b // 128->64 |
| 89 | addp vhas_chr1.16b, vhas_chr1.16b, vhas_chr1.16b // 128->64 |
| 90 | mov nul_match, vhas_nul1.d[0] |
| 91 | lsl tmp1, tmp1, #1 |
| 92 | mov const_m1, #~0 |
| 93 | mov chr_match, vhas_chr1.d[0] |
| 94 | lsr tmp3, const_m1, tmp1 |
| 95 | |
| 96 | bic nul_match, nul_match, tmp3 // Mask padding bits. |
| 97 | bic chr_match, chr_match, tmp3 // Mask padding bits. |
| 98 | cbnz nul_match, L(tail) |
| 99 | |
| 100 | L(loop): |
| 101 | cmp chr_match, #0 |
| 102 | csel src_match, src, src_match, ne |
| 103 | csel src_offset, chr_match, src_offset, ne |
| 104 | L(aligned): |
| 105 | ld1 {vdata1.16b, vdata2.16b}, [src], #32 |
| 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 | addp vend1.16b, vhas_nul1.16b, vhas_nul2.16b // 256->128 |
| 111 | and vhas_chr1.16b, vhas_chr1.16b, vrepmask_c.16b |
| 112 | and vhas_chr2.16b, vhas_chr2.16b, vrepmask_c.16b |
| 113 | addp vhas_chr1.16b, vhas_chr1.16b, vhas_chr2.16b // 256->128 |
| 114 | addp vend1.16b, vend1.16b, vend1.16b // 128->64 |
| 115 | addp vhas_chr1.16b, vhas_chr1.16b, vhas_chr1.16b // 128->64 |
| 116 | mov nul_match, vend1.d[0] |
| 117 | mov chr_match, vhas_chr1.d[0] |
| 118 | cbz nul_match, L(loop) |
| 119 | |
| 120 | and vhas_nul1.16b, vhas_nul1.16b, vrepmask_0.16b |
| 121 | and vhas_nul2.16b, vhas_nul2.16b, vrepmask_0.16b |
| 122 | addp vhas_nul1.16b, vhas_nul1.16b, vhas_nul2.16b |
| 123 | addp vhas_nul1.16b, vhas_nul1.16b, vhas_nul1.16b |
| 124 | mov nul_match, vhas_nul1.d[0] |
| 125 | |
| 126 | L(tail): |
| 127 | /* Work out exactly where the string ends. */ |
| 128 | sub tmp4, nul_match, #1 |
| 129 | eor tmp4, tmp4, nul_match |
| 130 | ands chr_match, chr_match, tmp4 |
| 131 | /* And pick the values corresponding to the last match. */ |
| 132 | csel src_match, src, src_match, ne |
| 133 | csel src_offset, chr_match, src_offset, ne |
| 134 | |
| 135 | /* Count down from the top of the syndrome to find the last match. */ |
| 136 | clz tmp3, src_offset |
| 137 | /* Src_match points beyond the word containing the match, so we can |
| 138 | simply subtract half the bit-offset into the syndrome. Because |
| 139 | we are counting down, we need to go back one more character. */ |
| 140 | add tmp3, tmp3, #2 |
| 141 | sub result, src_match, tmp3, lsr #1 |
| 142 | /* But if the syndrome shows no match was found, then return NULL. */ |
| 143 | cmp src_offset, #0 |
| 144 | csel result, result, xzr, ne |
| 145 | |
| 146 | ret |
| 147 | |
| 148 | END (__strrchr_aarch64) |
| 149 | |