1 | /* Multiple versions of strcmp |
2 | Copyright (C) 2009-2017 Free Software Foundation, Inc. |
3 | Contributed by Intel Corporation. |
4 | This file is part of the GNU C Library. |
5 | |
6 | The GNU C Library is free software; you can redistribute it and/or |
7 | modify it under the terms of the GNU Lesser General Public |
8 | License as published by the Free Software Foundation; either |
9 | version 2.1 of the License, or (at your option) any later version. |
10 | |
11 | The GNU C Library is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | Lesser General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU Lesser General Public |
17 | License along with the GNU C Library; if not, see |
18 | <http://www.gnu.org/licenses/>. */ |
19 | |
20 | #include <sysdep.h> |
21 | #include <init-arch.h> |
22 | |
23 | #ifdef USE_AS_STRNCMP |
24 | /* Since the counter, %r11, is unsigned, we branch to strcmp_exitz |
25 | if the new counter > the old one or is 0. */ |
26 | # define UPDATE_STRNCMP_COUNTER \ |
27 | /* calculate left number to compare */ \ |
28 | lea -16(%rcx, %r11), %r9; \ |
29 | cmp %r9, %r11; \ |
30 | jb LABEL(strcmp_exitz); \ |
31 | test %r9, %r9; \ |
32 | je LABEL(strcmp_exitz); \ |
33 | mov %r9, %r11 |
34 | |
35 | # define STRCMP_SSE42 __strncmp_sse42 |
36 | # define STRCMP_SSSE3 __strncmp_ssse3 |
37 | # define STRCMP_SSE2 __strncmp_sse2 |
38 | # define __GI_STRCMP __GI_strncmp |
39 | #elif defined USE_AS_STRCASECMP_L |
40 | # include "locale-defines.h" |
41 | |
42 | # define UPDATE_STRNCMP_COUNTER |
43 | |
44 | # define STRCMP_AVX __strcasecmp_l_avx |
45 | # define STRCMP_SSE42 __strcasecmp_l_sse42 |
46 | # define STRCMP_SSSE3 __strcasecmp_l_ssse3 |
47 | # define STRCMP_SSE2 __strcasecmp_l_sse2 |
48 | # define __GI_STRCMP __GI___strcasecmp_l |
49 | #elif defined USE_AS_STRNCASECMP_L |
50 | # include "locale-defines.h" |
51 | |
52 | /* Since the counter, %r11, is unsigned, we branch to strcmp_exitz |
53 | if the new counter > the old one or is 0. */ |
54 | # define UPDATE_STRNCMP_COUNTER \ |
55 | /* calculate left number to compare */ \ |
56 | lea -16(%rcx, %r11), %r9; \ |
57 | cmp %r9, %r11; \ |
58 | jb LABEL(strcmp_exitz); \ |
59 | test %r9, %r9; \ |
60 | je LABEL(strcmp_exitz); \ |
61 | mov %r9, %r11 |
62 | |
63 | # define STRCMP_AVX __strncasecmp_l_avx |
64 | # define STRCMP_SSE42 __strncasecmp_l_sse42 |
65 | # define STRCMP_SSSE3 __strncasecmp_l_ssse3 |
66 | # define STRCMP_SSE2 __strncasecmp_l_sse2 |
67 | # define __GI_STRCMP __GI___strncasecmp_l |
68 | #else |
69 | # define USE_AS_STRCMP |
70 | # define UPDATE_STRNCMP_COUNTER |
71 | # ifndef STRCMP |
72 | # define STRCMP strcmp |
73 | # define STRCMP_SSE42 __strcmp_sse42 |
74 | # define STRCMP_SSSE3 __strcmp_ssse3 |
75 | # define STRCMP_SSE2 __strcmp_sse2 |
76 | # define __GI_STRCMP __GI_strcmp |
77 | # endif |
78 | #endif |
79 | |
80 | /* Define multiple versions only for the definition in libc. Don't |
81 | define multiple versions for strncmp in static library since we |
82 | need strncmp before the initialization happened. */ |
83 | #if (defined SHARED || !defined USE_AS_STRNCMP) && IS_IN (libc) |
84 | .text |
85 | ENTRY(STRCMP) |
86 | .type STRCMP, @gnu_indirect_function |
87 | LOAD_RTLD_GLOBAL_RO_RDX |
88 | #ifdef USE_AS_STRCMP |
89 | leaq __strcmp_sse2_unaligned(%rip), %rax |
90 | HAS_ARCH_FEATURE (Fast_Unaligned_Load) |
91 | jnz 3f |
92 | #else |
93 | HAS_ARCH_FEATURE (Slow_SSE4_2) |
94 | jnz 2f |
95 | leaq STRCMP_SSE42(%rip), %rax |
96 | HAS_CPU_FEATURE (SSE4_2) |
97 | jnz 3f |
98 | #endif |
99 | 2: leaq STRCMP_SSSE3(%rip), %rax |
100 | HAS_CPU_FEATURE (SSSE3) |
101 | jnz 3f |
102 | leaq STRCMP_SSE2(%rip), %rax |
103 | 3: ret |
104 | END(STRCMP) |
105 | |
106 | # ifdef USE_AS_STRCASECMP_L |
107 | ENTRY(__strcasecmp) |
108 | .type __strcasecmp, @gnu_indirect_function |
109 | LOAD_RTLD_GLOBAL_RO_RDX |
110 | leaq __strcasecmp_avx(%rip), %rax |
111 | HAS_ARCH_FEATURE (AVX_Usable) |
112 | jnz 3f |
113 | HAS_ARCH_FEATURE (Slow_SSE4_2) |
114 | jnz 2f |
115 | leaq __strcasecmp_sse42(%rip), %rax |
116 | HAS_CPU_FEATURE (SSE4_2) |
117 | jnz 3f |
118 | 2: leaq __strcasecmp_ssse3(%rip), %rax |
119 | HAS_CPU_FEATURE (SSSE3) |
120 | jnz 3f |
121 | leaq __strcasecmp_sse2(%rip), %rax |
122 | 3: ret |
123 | END(__strcasecmp) |
124 | weak_alias (__strcasecmp, strcasecmp) |
125 | # endif |
126 | # ifdef USE_AS_STRNCASECMP_L |
127 | ENTRY(__strncasecmp) |
128 | .type __strncasecmp, @gnu_indirect_function |
129 | LOAD_RTLD_GLOBAL_RO_RDX |
130 | leaq __strncasecmp_avx(%rip), %rax |
131 | HAS_ARCH_FEATURE (AVX_Usable) |
132 | jnz 3f |
133 | HAS_ARCH_FEATURE (Slow_SSE4_2) |
134 | jnz 2f |
135 | leaq __strncasecmp_sse42(%rip), %rax |
136 | HAS_CPU_FEATURE (SSE4_2) |
137 | jnz 3f |
138 | 2: leaq __strncasecmp_ssse3(%rip), %rax |
139 | HAS_CPU_FEATURE (SSSE3) |
140 | jnz 3f |
141 | leaq __strncasecmp_sse2(%rip), %rax |
142 | 3: ret |
143 | END(__strncasecmp) |
144 | weak_alias (__strncasecmp, strncasecmp) |
145 | # endif |
146 | |
147 | # undef LABEL |
148 | # define LABEL(l) .L##l##_sse42 |
149 | # define GLABEL(l) l##_sse42 |
150 | # define SECTION sse4.2 |
151 | # include "strcmp-sse42.S" |
152 | |
153 | |
154 | # if defined USE_AS_STRCASECMP_L || defined USE_AS_STRNCASECMP_L |
155 | # define LABEL(l) .L##l##_avx |
156 | # define GLABEL(l) l##_avx |
157 | # define USE_AVX 1 |
158 | # undef STRCMP_SSE42 |
159 | # define STRCMP_SSE42 STRCMP_AVX |
160 | # define SECTION avx |
161 | # include "strcmp-sse42.S" |
162 | # endif |
163 | |
164 | |
165 | # undef ENTRY |
166 | # define ENTRY(name) \ |
167 | .type STRCMP_SSE2, @function; \ |
168 | .align 16; \ |
169 | .globl STRCMP_SSE2; \ |
170 | .hidden STRCMP_SSE2; \ |
171 | STRCMP_SSE2: cfi_startproc; \ |
172 | CALL_MCOUNT |
173 | # undef END |
174 | # define END(name) \ |
175 | cfi_endproc; .size STRCMP_SSE2, .-STRCMP_SSE2 |
176 | |
177 | # ifdef USE_AS_STRCASECMP_L |
178 | # define ENTRY2(name) \ |
179 | .type __strcasecmp_sse2, @function; \ |
180 | .align 16; \ |
181 | .globl __strcasecmp_sse2; \ |
182 | .hidden __strcasecmp_sse2; \ |
183 | __strcasecmp_sse2: cfi_startproc; \ |
184 | CALL_MCOUNT |
185 | # define END2(name) \ |
186 | cfi_endproc; .size __strcasecmp_sse2, .-__strcasecmp_sse2 |
187 | # endif |
188 | |
189 | # ifdef USE_AS_STRNCASECMP_L |
190 | # define ENTRY2(name) \ |
191 | .type __strncasecmp_sse2, @function; \ |
192 | .align 16; \ |
193 | .globl __strncasecmp_sse2; \ |
194 | .hidden __strncasecmp_sse2; \ |
195 | __strncasecmp_sse2: cfi_startproc; \ |
196 | CALL_MCOUNT |
197 | # define END2(name) \ |
198 | cfi_endproc; .size __strncasecmp_sse2, .-__strncasecmp_sse2 |
199 | # endif |
200 | |
201 | # undef libc_hidden_builtin_def |
202 | /* It doesn't make sense to send libc-internal strcmp calls through a PLT. |
203 | The speedup we get from using SSE4.2 instruction is likely eaten away |
204 | by the indirect call in the PLT. */ |
205 | # define libc_hidden_builtin_def(name) \ |
206 | .globl __GI_STRCMP; __GI_STRCMP = STRCMP_SSE2 |
207 | #endif |
208 | |
209 | #include "../strcmp.S" |
210 | |