| 1 | /* Optimized rawmemchr implementation using basic LoongArch instructions. |
|---|
| 2 | Copyright (C) 2023-2024 Free Software Foundation, Inc. |
|---|
| 3 | This file is part of the GNU C Library. |
|---|
| 4 | |
|---|
| 5 | The GNU C Library is free software; you can redistribute it and/or |
|---|
| 6 | modify it under the terms of the GNU Lesser General Public |
|---|
| 7 | License as published by the Free Software Foundation; either |
|---|
| 8 | version 2.1 of the License, or (at your option) any later version. |
|---|
| 9 | |
|---|
| 10 | The GNU C Library is distributed in the hope that it will be useful, |
|---|
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
|---|
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|---|
| 13 | Lesser General Public License for more details. |
|---|
| 14 | |
|---|
| 15 | You should have received a copy of the GNU Lesser General Public |
|---|
| 16 | License along with the GNU C Library. If not, see |
|---|
| 17 | <https://www.gnu.org/licenses/>. */ |
|---|
| 18 | |
|---|
| 19 | #include <sysdep.h> |
|---|
| 20 | #include <sys/regdef.h> |
|---|
| 21 | #include <sys/asm.h> |
|---|
| 22 | |
|---|
| 23 | #if IS_IN (libc) |
|---|
| 24 | # define RAWMEMCHR_NAME __rawmemchr_aligned |
|---|
| 25 | #else |
|---|
| 26 | # define RAWMEMCHR_NAME __rawmemchr |
|---|
| 27 | #endif |
|---|
| 28 | |
|---|
| 29 | LEAF(RAWMEMCHR_NAME, 6) |
|---|
| 30 | andi t1, a0, 0x7 |
|---|
| 31 | bstrins.d a0, zero, 2, 0 |
|---|
| 32 | lu12i.w a2, 0x01010 |
|---|
| 33 | bstrins.d a1, a1, 15, 8 |
|---|
| 34 | |
|---|
| 35 | ld.d t0, a0, 0 |
|---|
| 36 | slli.d t1, t1, 3 |
|---|
| 37 | ori a2, a2, 0x101 |
|---|
| 38 | bstrins.d a1, a1, 31, 16 |
|---|
| 39 | |
|---|
| 40 | li.w t8, -1 |
|---|
| 41 | bstrins.d a1, a1, 63, 32 |
|---|
| 42 | bstrins.d a2, a2, 63, 32 |
|---|
| 43 | sll.d t2, t8, t1 |
|---|
| 44 | |
|---|
| 45 | sll.d t3, a1, t1 |
|---|
| 46 | orn t0, t0, t2 |
|---|
| 47 | slli.d a3, a2, 7 |
|---|
| 48 | beqz a1, L(find_zero) |
|---|
| 49 | |
|---|
| 50 | xor t0, t0, t3 |
|---|
| 51 | sub.d t1, t0, a2 |
|---|
| 52 | andn t2, a3, t0 |
|---|
| 53 | and t3, t1, t2 |
|---|
| 54 | |
|---|
| 55 | bnez t3, L(count_pos) |
|---|
| 56 | addi.d a0, a0, 8 |
|---|
| 57 | |
|---|
| 58 | L(loop): |
|---|
| 59 | ld.d t0, a0, 0 |
|---|
| 60 | xor t0, t0, a1 |
|---|
| 61 | |
|---|
| 62 | sub.d t1, t0, a2 |
|---|
| 63 | andn t2, a3, t0 |
|---|
| 64 | and t3, t1, t2 |
|---|
| 65 | bnez t3, L(count_pos) |
|---|
| 66 | |
|---|
| 67 | ld.d t0, a0, 8 |
|---|
| 68 | addi.d a0, a0, 16 |
|---|
| 69 | xor t0, t0, a1 |
|---|
| 70 | sub.d t1, t0, a2 |
|---|
| 71 | |
|---|
| 72 | andn t2, a3, t0 |
|---|
| 73 | and t3, t1, t2 |
|---|
| 74 | beqz t3, L(loop) |
|---|
| 75 | addi.d a0, a0, -8 |
|---|
| 76 | L(count_pos): |
|---|
| 77 | ctz.d t0, t3 |
|---|
| 78 | srli.d t0, t0, 3 |
|---|
| 79 | add.d a0, a0, t0 |
|---|
| 80 | jr ra |
|---|
| 81 | |
|---|
| 82 | L(loop_7bit): |
|---|
| 83 | ld.d t0, a0, 0 |
|---|
| 84 | L(find_zero): |
|---|
| 85 | sub.d t1, t0, a2 |
|---|
| 86 | and t2, t1, a3 |
|---|
| 87 | bnez t2, L(more_check) |
|---|
| 88 | |
|---|
| 89 | ld.d t0, a0, 8 |
|---|
| 90 | addi.d a0, a0, 16 |
|---|
| 91 | sub.d t1, t0, a2 |
|---|
| 92 | and t2, t1, a3 |
|---|
| 93 | |
|---|
| 94 | beqz t2, L(loop_7bit) |
|---|
| 95 | addi.d a0, a0, -8 |
|---|
| 96 | |
|---|
| 97 | L(more_check): |
|---|
| 98 | andn t2, a3, t0 |
|---|
| 99 | and t3, t1, t2 |
|---|
| 100 | bnez t3, L(count_pos) |
|---|
| 101 | addi.d a0, a0, 8 |
|---|
| 102 | |
|---|
| 103 | L(loop_8bit): |
|---|
| 104 | ld.d t0, a0, 0 |
|---|
| 105 | |
|---|
| 106 | sub.d t1, t0, a2 |
|---|
| 107 | andn t2, a3, t0 |
|---|
| 108 | and t3, t1, t2 |
|---|
| 109 | bnez t3, L(count_pos) |
|---|
| 110 | |
|---|
| 111 | ld.d t0, a0, 8 |
|---|
| 112 | addi.d a0, a0, 16 |
|---|
| 113 | sub.d t1, t0, a2 |
|---|
| 114 | |
|---|
| 115 | andn t2, a3, t0 |
|---|
| 116 | and t3, t1, t2 |
|---|
| 117 | beqz t3, L(loop_8bit) |
|---|
| 118 | |
|---|
| 119 | addi.d a0, a0, -8 |
|---|
| 120 | b L(count_pos) |
|---|
| 121 | |
|---|
| 122 | END(RAWMEMCHR_NAME) |
|---|
| 123 | |
|---|
| 124 | libc_hidden_builtin_def (__rawmemchr) |
|---|
| 125 | |
|---|
