| 1 | /* strcmp with unaligned loads |
| 2 | Copyright (C) 2013-2020 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 | #if IS_IN (libc) |
| 20 | |
| 21 | #include "sysdep.h" |
| 22 | |
| 23 | ENTRY ( __strcmp_sse2_unaligned) |
| 24 | movl %edi, %eax |
| 25 | xorl %edx, %edx |
| 26 | pxor %xmm7, %xmm7 |
| 27 | orl %esi, %eax |
| 28 | andl $4095, %eax |
| 29 | cmpl $4032, %eax |
| 30 | jg L(cross_page) |
| 31 | movdqu (%rdi), %xmm1 |
| 32 | movdqu (%rsi), %xmm0 |
| 33 | pcmpeqb %xmm1, %xmm0 |
| 34 | pminub %xmm1, %xmm0 |
| 35 | pxor %xmm1, %xmm1 |
| 36 | pcmpeqb %xmm1, %xmm0 |
| 37 | pmovmskb %xmm0, %eax |
| 38 | testq %rax, %rax |
| 39 | je L(next_48_bytes) |
| 40 | L(return): |
| 41 | bsfq %rax, %rdx |
| 42 | movzbl (%rdi, %rdx), %eax |
| 43 | movzbl (%rsi, %rdx), %edx |
| 44 | subl %edx, %eax |
| 45 | ret |
| 46 | |
| 47 | .p2align 4 |
| 48 | L(next_48_bytes): |
| 49 | movdqu 16(%rdi), %xmm6 |
| 50 | movdqu 16(%rsi), %xmm3 |
| 51 | movdqu 32(%rdi), %xmm5 |
| 52 | pcmpeqb %xmm6, %xmm3 |
| 53 | movdqu 32(%rsi), %xmm2 |
| 54 | pminub %xmm6, %xmm3 |
| 55 | pcmpeqb %xmm1, %xmm3 |
| 56 | movdqu 48(%rdi), %xmm4 |
| 57 | pcmpeqb %xmm5, %xmm2 |
| 58 | pmovmskb %xmm3, %edx |
| 59 | movdqu 48(%rsi), %xmm0 |
| 60 | pminub %xmm5, %xmm2 |
| 61 | pcmpeqb %xmm1, %xmm2 |
| 62 | pcmpeqb %xmm4, %xmm0 |
| 63 | pmovmskb %xmm2, %eax |
| 64 | salq $16, %rdx |
| 65 | pminub %xmm4, %xmm0 |
| 66 | pcmpeqb %xmm1, %xmm0 |
| 67 | salq $32, %rax |
| 68 | orq %rdx, %rax |
| 69 | pmovmskb %xmm0, %ecx |
| 70 | movq %rcx, %rdx |
| 71 | salq $48, %rdx |
| 72 | orq %rdx, %rax |
| 73 | jne L(return) |
| 74 | L(main_loop_header): |
| 75 | leaq 64(%rdi), %rdx |
| 76 | movl $4096, %ecx |
| 77 | pxor %xmm9, %xmm9 |
| 78 | andq $-64, %rdx |
| 79 | subq %rdi, %rdx |
| 80 | leaq (%rdi, %rdx), %rax |
| 81 | addq %rsi, %rdx |
| 82 | movq %rdx, %rsi |
| 83 | andl $4095, %esi |
| 84 | subq %rsi, %rcx |
| 85 | shrq $6, %rcx |
| 86 | movq %rcx, %rsi |
| 87 | jmp L(loop_start) |
| 88 | |
| 89 | .p2align 4 |
| 90 | L(loop): |
| 91 | addq $64, %rax |
| 92 | addq $64, %rdx |
| 93 | L(loop_start): |
| 94 | testq %rsi, %rsi |
| 95 | leaq -1(%rsi), %rsi |
| 96 | je L(loop_cross_page) |
| 97 | L(back_to_loop): |
| 98 | movdqu (%rdx), %xmm0 |
| 99 | movdqu 16(%rdx), %xmm1 |
| 100 | movdqa (%rax), %xmm2 |
| 101 | movdqa 16(%rax), %xmm3 |
| 102 | pcmpeqb %xmm2, %xmm0 |
| 103 | movdqu 32(%rdx), %xmm5 |
| 104 | pcmpeqb %xmm3, %xmm1 |
| 105 | pminub %xmm2, %xmm0 |
| 106 | movdqu 48(%rdx), %xmm6 |
| 107 | pminub %xmm3, %xmm1 |
| 108 | movdqa 32(%rax), %xmm2 |
| 109 | pminub %xmm1, %xmm0 |
| 110 | movdqa 48(%rax), %xmm3 |
| 111 | pcmpeqb %xmm2, %xmm5 |
| 112 | pcmpeqb %xmm3, %xmm6 |
| 113 | pminub %xmm2, %xmm5 |
| 114 | pminub %xmm3, %xmm6 |
| 115 | pminub %xmm5, %xmm0 |
| 116 | pminub %xmm6, %xmm0 |
| 117 | pcmpeqb %xmm7, %xmm0 |
| 118 | pmovmskb %xmm0, %ecx |
| 119 | testl %ecx, %ecx |
| 120 | je L(loop) |
| 121 | pcmpeqb %xmm7, %xmm5 |
| 122 | movdqu (%rdx), %xmm0 |
| 123 | pcmpeqb %xmm7, %xmm1 |
| 124 | movdqa (%rax), %xmm2 |
| 125 | pcmpeqb %xmm2, %xmm0 |
| 126 | pminub %xmm2, %xmm0 |
| 127 | pcmpeqb %xmm7, %xmm6 |
| 128 | pcmpeqb %xmm7, %xmm0 |
| 129 | pmovmskb %xmm1, %ecx |
| 130 | pmovmskb %xmm5, %r8d |
| 131 | pmovmskb %xmm0, %edi |
| 132 | salq $16, %rcx |
| 133 | salq $32, %r8 |
| 134 | pmovmskb %xmm6, %esi |
| 135 | orq %r8, %rcx |
| 136 | orq %rdi, %rcx |
| 137 | salq $48, %rsi |
| 138 | orq %rsi, %rcx |
| 139 | bsfq %rcx, %rcx |
| 140 | movzbl (%rax, %rcx), %eax |
| 141 | movzbl (%rdx, %rcx), %edx |
| 142 | subl %edx, %eax |
| 143 | ret |
| 144 | |
| 145 | .p2align 4 |
| 146 | L(loop_cross_page): |
| 147 | xor %r10, %r10 |
| 148 | movq %rdx, %r9 |
| 149 | and $63, %r9 |
| 150 | subq %r9, %r10 |
| 151 | |
| 152 | movdqa (%rdx, %r10), %xmm0 |
| 153 | movdqa 16(%rdx, %r10), %xmm1 |
| 154 | movdqu (%rax, %r10), %xmm2 |
| 155 | movdqu 16(%rax, %r10), %xmm3 |
| 156 | pcmpeqb %xmm2, %xmm0 |
| 157 | movdqa 32(%rdx, %r10), %xmm5 |
| 158 | pcmpeqb %xmm3, %xmm1 |
| 159 | pminub %xmm2, %xmm0 |
| 160 | movdqa 48(%rdx, %r10), %xmm6 |
| 161 | pminub %xmm3, %xmm1 |
| 162 | movdqu 32(%rax, %r10), %xmm2 |
| 163 | movdqu 48(%rax, %r10), %xmm3 |
| 164 | pcmpeqb %xmm2, %xmm5 |
| 165 | pcmpeqb %xmm3, %xmm6 |
| 166 | pminub %xmm2, %xmm5 |
| 167 | pminub %xmm3, %xmm6 |
| 168 | |
| 169 | pcmpeqb %xmm7, %xmm0 |
| 170 | pcmpeqb %xmm7, %xmm1 |
| 171 | pcmpeqb %xmm7, %xmm5 |
| 172 | pcmpeqb %xmm7, %xmm6 |
| 173 | |
| 174 | pmovmskb %xmm1, %ecx |
| 175 | pmovmskb %xmm5, %r8d |
| 176 | pmovmskb %xmm0, %edi |
| 177 | salq $16, %rcx |
| 178 | salq $32, %r8 |
| 179 | pmovmskb %xmm6, %esi |
| 180 | orq %r8, %rdi |
| 181 | orq %rcx, %rdi |
| 182 | salq $48, %rsi |
| 183 | orq %rsi, %rdi |
| 184 | movq %r9, %rcx |
| 185 | movq $63, %rsi |
| 186 | shrq %cl, %rdi |
| 187 | test %rdi, %rdi |
| 188 | je L(back_to_loop) |
| 189 | bsfq %rdi, %rcx |
| 190 | movzbl (%rax, %rcx), %eax |
| 191 | movzbl (%rdx, %rcx), %edx |
| 192 | subl %edx, %eax |
| 193 | ret |
| 194 | |
| 195 | .p2align 4 |
| 196 | L(cross_page_loop): |
| 197 | cmpb %cl, %al |
| 198 | jne L(different) |
| 199 | addq $1, %rdx |
| 200 | cmpq $64, %rdx |
| 201 | je L(main_loop_header) |
| 202 | L(cross_page): |
| 203 | movzbl (%rdi, %rdx), %eax |
| 204 | movzbl (%rsi, %rdx), %ecx |
| 205 | testb %al, %al |
| 206 | jne L(cross_page_loop) |
| 207 | xorl %eax, %eax |
| 208 | L(different): |
| 209 | subl %ecx, %eax |
| 210 | ret |
| 211 | END (__strcmp_sse2_unaligned) |
| 212 | |
| 213 | #endif |
| 214 | |