| 1 | /* Optimized strrchr implementation using LoongArch LSX 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) && !defined __loongarch_soft_float |
|---|
| 24 | |
|---|
| 25 | #define STRRCHR __strrchr_lsx |
|---|
| 26 | |
|---|
| 27 | LEAF(STRRCHR, 6) |
|---|
| 28 | move a2, a0 |
|---|
| 29 | bstrins.d a0, zero, 4, 0 |
|---|
| 30 | vld vr0, a0, 0 |
|---|
| 31 | vld vr1, a0, 16 |
|---|
| 32 | |
|---|
| 33 | li.d t2, -1 |
|---|
| 34 | vreplgr2vr.b vr4, a1 |
|---|
| 35 | vmsknz.b vr2, vr0 |
|---|
| 36 | vmsknz.b vr3, vr1 |
|---|
| 37 | |
|---|
| 38 | vilvl.h vr2, vr3, vr2 |
|---|
| 39 | movfr2gr.s t0, fa2 |
|---|
| 40 | sra.w t0, t0, a2 |
|---|
| 41 | beq t0, t2, L(find_tail) |
|---|
| 42 | |
|---|
| 43 | vseq.b vr2, vr0, vr4 |
|---|
| 44 | vseq.b vr3, vr1, vr4 |
|---|
| 45 | vmsknz.b vr2, vr2 |
|---|
| 46 | vmsknz.b vr3, vr3 |
|---|
| 47 | |
|---|
| 48 | |
|---|
| 49 | vilvl.h vr1, vr3, vr2 |
|---|
| 50 | slli.d t3, t2, 1 |
|---|
| 51 | movfr2gr.s t1, fa1 |
|---|
| 52 | cto.w t0, t0 |
|---|
| 53 | |
|---|
| 54 | srl.w t1, t1, a2 |
|---|
| 55 | sll.d t3, t3, t0 |
|---|
| 56 | addi.d a0, a2, 31 |
|---|
| 57 | andn t1, t1, t3 |
|---|
| 58 | |
|---|
| 59 | clz.w t0, t1 |
|---|
| 60 | sub.d a0, a0, t0 |
|---|
| 61 | maskeqz a0, a0, t1 |
|---|
| 62 | jr ra |
|---|
| 63 | |
|---|
| 64 | .align 5 |
|---|
| 65 | L(find_tail): |
|---|
| 66 | addi.d a3, a0, 32 |
|---|
| 67 | L(loop): |
|---|
| 68 | vld vr2, a0, 32 |
|---|
| 69 | vld vr3, a0, 48 |
|---|
| 70 | addi.d a0, a0, 32 |
|---|
| 71 | |
|---|
| 72 | vmin.bu vr5, vr2, vr3 |
|---|
| 73 | vsetanyeqz.b fcc0, vr5 |
|---|
| 74 | bceqz fcc0, L(loop) |
|---|
| 75 | vmsknz.b vr5, vr2 |
|---|
| 76 | |
|---|
| 77 | vmsknz.b vr6, vr3 |
|---|
| 78 | vilvl.h vr5, vr6, vr5 |
|---|
| 79 | vseq.b vr2, vr2, vr4 |
|---|
| 80 | vseq.b vr3, vr3, vr4 |
|---|
| 81 | |
|---|
| 82 | vmsknz.b vr2, vr2 |
|---|
| 83 | vmsknz.b vr3, vr3 |
|---|
| 84 | vilvl.h vr2, vr3, vr2 |
|---|
| 85 | movfr2gr.s t0, fa5 |
|---|
| 86 | |
|---|
| 87 | |
|---|
| 88 | movfr2gr.s t1, fa2 |
|---|
| 89 | slli.d t3, t2, 1 |
|---|
| 90 | cto.w t0, t0 |
|---|
| 91 | sll.d t3, t3, t0 |
|---|
| 92 | |
|---|
| 93 | andn t1, t1, t3 |
|---|
| 94 | beqz t1, L(find_loop) |
|---|
| 95 | clz.w t0, t1 |
|---|
| 96 | addi.d a0, a0, 31 |
|---|
| 97 | |
|---|
| 98 | sub.d a0, a0, t0 |
|---|
| 99 | jr ra |
|---|
| 100 | L(find_loop): |
|---|
| 101 | beq a0, a3, L(find_end) |
|---|
| 102 | vld vr2, a0, -32 |
|---|
| 103 | |
|---|
| 104 | vld vr3, a0, -16 |
|---|
| 105 | addi.d a0, a0, -32 |
|---|
| 106 | vseq.b vr2, vr2, vr4 |
|---|
| 107 | vseq.b vr3, vr3, vr4 |
|---|
| 108 | |
|---|
| 109 | |
|---|
| 110 | vmax.bu vr5, vr2, vr3 |
|---|
| 111 | vseteqz.v fcc0, vr5 |
|---|
| 112 | bcnez fcc0, L(find_loop) |
|---|
| 113 | vmsknz.b vr0, vr2 |
|---|
| 114 | |
|---|
| 115 | vmsknz.b vr1, vr3 |
|---|
| 116 | vilvl.h vr0, vr1, vr0 |
|---|
| 117 | movfr2gr.s t0, fa0 |
|---|
| 118 | addi.d a0, a0, 31 |
|---|
| 119 | |
|---|
| 120 | clz.w t0, t0 |
|---|
| 121 | sub.d a0, a0, t0 |
|---|
| 122 | jr ra |
|---|
| 123 | nop |
|---|
| 124 | |
|---|
| 125 | L(find_end): |
|---|
| 126 | vseq.b vr2, vr0, vr4 |
|---|
| 127 | vseq.b vr3, vr1, vr4 |
|---|
| 128 | vmsknz.b vr2, vr2 |
|---|
| 129 | vmsknz.b vr3, vr3 |
|---|
| 130 | |
|---|
| 131 | |
|---|
| 132 | vilvl.h vr1, vr3, vr2 |
|---|
| 133 | movfr2gr.s t1, fa1 |
|---|
| 134 | addi.d a0, a2, 31 |
|---|
| 135 | srl.w t1, t1, a2 |
|---|
| 136 | |
|---|
| 137 | clz.w t0, t1 |
|---|
| 138 | sub.d a0, a0, t0 |
|---|
| 139 | maskeqz a0, a0, t1 |
|---|
| 140 | jr ra |
|---|
| 141 | END(STRRCHR) |
|---|
| 142 | |
|---|
| 143 | libc_hidden_builtin_def(STRRCHR) |
|---|
| 144 | #endif |
|---|
| 145 | |
|---|
