| lh | 9ed821d | 2023-04-07 01:36:19 -0700 | [diff] [blame^] | 1 | /* Optimized memset for Xtensa. |
| 2 | Copyright (C) 2001, 2007 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, write to the Free |
| 17 | Software Foundation, Inc., 51 Franklin Street - Fifth Floor, |
| 18 | Boston, MA 02110-1301, USA. */ |
| 19 | |
| 20 | #include "../../sysdeps/linux/xtensa/sysdep.h" |
| 21 | #include <bits/xtensa-config.h> |
| 22 | |
| 23 | /* Do not use .literal_position in the ENTRY macro. */ |
| 24 | #undef LITERAL_POSITION |
| 25 | #define LITERAL_POSITION |
| 26 | |
| 27 | /* void *memset (void *dst, int c, size_t length) |
| 28 | |
| 29 | The algorithm is as follows: |
| 30 | |
| 31 | Create a word with c in all byte positions. |
| 32 | |
| 33 | If the destination is aligned, set 16B chunks with a loop, and then |
| 34 | finish up with 8B, 4B, 2B, and 1B stores conditional on the length. |
| 35 | |
| 36 | If the destination is unaligned, align it by conditionally |
| 37 | setting 1B and/or 2B and then go to aligned case. |
| 38 | |
| 39 | This code tries to use fall-through branches for the common |
| 40 | case of an aligned destination (except for the branches to |
| 41 | the alignment labels). */ |
| 42 | |
| 43 | |
| 44 | /* Byte-by-byte set. */ |
| 45 | |
| 46 | .text |
| 47 | .align 4 |
| 48 | .literal_position |
| 49 | __memset_aux: |
| 50 | |
| 51 | /* Skip a byte to get 1 mod 4 alignment for LOOPNEZ |
| 52 | (0 mod 4 alignment for LBEG). */ |
| 53 | .byte 0 |
| 54 | |
| 55 | .Lbyteset: |
| 56 | #if XCHAL_HAVE_LOOPS |
| 57 | loopnez a4, 2f |
| 58 | #else |
| 59 | beqz a4, 2f |
| 60 | add a6, a5, a4 /* a6 = ending address */ |
| 61 | #endif |
| 62 | 1: s8i a3, a5, 0 |
| 63 | addi a5, a5, 1 |
| 64 | #if !XCHAL_HAVE_LOOPS |
| 65 | blt a5, a6, 1b |
| 66 | #endif |
| 67 | 2: retw |
| 68 | |
| 69 | |
| 70 | /* Destination is unaligned. */ |
| 71 | |
| 72 | .align 4 |
| 73 | |
| 74 | .Ldst1mod2: /* dst is only byte aligned */ |
| 75 | |
| 76 | /* Do short sizes byte-by-byte. */ |
| 77 | bltui a4, 8, .Lbyteset |
| 78 | |
| 79 | /* Set 1 byte. */ |
| 80 | s8i a3, a5, 0 |
| 81 | addi a5, a5, 1 |
| 82 | addi a4, a4, -1 |
| 83 | |
| 84 | /* Now retest if dst is aligned. */ |
| 85 | _bbci.l a5, 1, .Ldstaligned |
| 86 | |
| 87 | .Ldst2mod4: /* dst has 16-bit alignment */ |
| 88 | |
| 89 | /* Do short sizes byte-by-byte. */ |
| 90 | bltui a4, 8, .Lbyteset |
| 91 | |
| 92 | /* Set 2 bytes. */ |
| 93 | s16i a3, a5, 0 |
| 94 | addi a5, a5, 2 |
| 95 | addi a4, a4, -2 |
| 96 | |
| 97 | /* dst is now aligned; return to main algorithm */ |
| 98 | j .Ldstaligned |
| 99 | |
| 100 | |
| 101 | ENTRY (memset) |
| 102 | /* a2 = dst, a3 = c, a4 = length */ |
| 103 | |
| 104 | /* Duplicate character into all bytes of word. */ |
| 105 | extui a3, a3, 0, 8 |
| 106 | slli a7, a3, 8 |
| 107 | or a3, a3, a7 |
| 108 | slli a7, a3, 16 |
| 109 | or a3, a3, a7 |
| 110 | |
| 111 | mov a5, a2 /* copy dst so that a2 is return value */ |
| 112 | |
| 113 | /* Check if dst is unaligned. */ |
| 114 | _bbsi.l a2, 0, .Ldst1mod2 |
| 115 | _bbsi.l a2, 1, .Ldst2mod4 |
| 116 | .Ldstaligned: |
| 117 | |
| 118 | /* Get number of loop iterations with 16B per iteration. */ |
| 119 | srli a7, a4, 4 |
| 120 | |
| 121 | /* Destination is word-aligned. */ |
| 122 | #if XCHAL_HAVE_LOOPS |
| 123 | loopnez a7, 2f |
| 124 | #else |
| 125 | beqz a7, 2f |
| 126 | slli a6, a7, 4 |
| 127 | add a6, a6, a5 /* a6 = end of last 16B chunk */ |
| 128 | #endif |
| 129 | /* Set 16 bytes per iteration. */ |
| 130 | 1: s32i a3, a5, 0 |
| 131 | s32i a3, a5, 4 |
| 132 | s32i a3, a5, 8 |
| 133 | s32i a3, a5, 12 |
| 134 | addi a5, a5, 16 |
| 135 | #if !XCHAL_HAVE_LOOPS |
| 136 | blt a5, a6, 1b |
| 137 | #endif |
| 138 | |
| 139 | /* Set any leftover pieces smaller than 16B. */ |
| 140 | 2: bbci.l a4, 3, 3f |
| 141 | |
| 142 | /* Set 8 bytes. */ |
| 143 | s32i a3, a5, 0 |
| 144 | s32i a3, a5, 4 |
| 145 | addi a5, a5, 8 |
| 146 | |
| 147 | 3: bbci.l a4, 2, 4f |
| 148 | |
| 149 | /* Set 4 bytes. */ |
| 150 | s32i a3, a5, 0 |
| 151 | addi a5, a5, 4 |
| 152 | |
| 153 | 4: bbci.l a4, 1, 5f |
| 154 | |
| 155 | /* Set 2 bytes. */ |
| 156 | s16i a3, a5, 0 |
| 157 | addi a5, a5, 2 |
| 158 | |
| 159 | 5: bbci.l a4, 0, 6f |
| 160 | |
| 161 | /* Set 1 byte. */ |
| 162 | s8i a3, a5, 0 |
| 163 | 6: retw |
| 164 | |
| 165 | libc_hidden_def (memset) |