sysdeps/x86_64/multiarch/strcmp.S

   1 /* Multiple versions of strcmp
   2    Copyright (C) 2009-2014 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) && !defined NOT_IN_libc
  84         .text
  85 ENTRY(STRCMP)
  86         .type   STRCMP, @gnu_indirect_function
  87         /* Manually inlined call to __get_cpu_features.  */
  88         cmpl    $0, __cpu_features+KIND_OFFSET(%rip)
  89         jne     1f
  90         call    __init_cpu_features
  91 1:
  92 #ifdef USE_AS_STRCMP
  93         leaq    __strcmp_sse2_unaligned(%rip), %rax
  94         testl   $bit_Fast_Unaligned_Load, __cpu_features+CPUID_OFFSET+index_Fast_Unaligned_Load(%rip)
  95         jnz     3f
  96 #else
  97         testl   $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
  98         jnz     2f
  99         leaq    STRCMP_SSE42(%rip), %rax
 100         testl   $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
 101         jnz     3f
 102 #endif
 103 2:      leaq    STRCMP_SSSE3(%rip), %rax
 104         testl   $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip)
 105         jnz     3f
 106         leaq    STRCMP_SSE2(%rip), %rax
 107 3:      ret
 108 END(STRCMP)
 109
 110 # ifdef USE_AS_STRCASECMP_L
 111 ENTRY(__strcasecmp)
 112         .type   __strcasecmp, @gnu_indirect_function
 113         /* Manually inlined call to __get_cpu_features.  */
 114         cmpl    $0, __cpu_features+KIND_OFFSET(%rip)
 115         jne     1f
 116         call    __init_cpu_features
 117 1:
 118 #  ifdef HAVE_AVX_SUPPORT
 119         leaq    __strcasecmp_avx(%rip), %rax
 120         testl   $bit_AVX_Usable, __cpu_features+FEATURE_OFFSET+index_AVX_Usable(%rip)
 121         jnz     3f
 122 #  endif
 123         testl   $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
 124         jnz     2f
 125         leaq    __strcasecmp_sse42(%rip), %rax
 126         testl   $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
 127         jnz     3f
 128 2:      leaq    __strcasecmp_ssse3(%rip), %rax
 129         testl   $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip)
 130         jnz     3f
 131         leaq    __strcasecmp_sse2(%rip), %rax
 132 3:      ret
 133 END(__strcasecmp)
 134 weak_alias (__strcasecmp, strcasecmp)
 135 # endif
 136 # ifdef USE_AS_STRNCASECMP_L
 137 ENTRY(__strncasecmp)
 138         .type   __strncasecmp, @gnu_indirect_function
 139         /* Manually inlined call to __get_cpu_features.  */
 140         cmpl    $0, __cpu_features+KIND_OFFSET(%rip)
 141         jne     1f
 142         call    __init_cpu_features
 143 1:
 144 #  ifdef HAVE_AVX_SUPPORT
 145         leaq    __strncasecmp_avx(%rip), %rax
 146         testl   $bit_AVX_Usable, __cpu_features+FEATURE_OFFSET+index_AVX_Usable(%rip)
 147         jnz     3f
 148 #  endif
 149         testl   $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
 150         jnz     2f
 151         leaq    __strncasecmp_sse42(%rip), %rax
 152         testl   $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
 153         jnz     3f
 154 2:      leaq    __strncasecmp_ssse3(%rip), %rax
 155         testl   $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip)
 156         jnz     3f
 157         leaq    __strncasecmp_sse2(%rip), %rax
 158 3:      ret
 159 END(__strncasecmp)
 160 weak_alias (__strncasecmp, strncasecmp)
 161 # endif
 162
 163 # undef LABEL
 164 # define LABEL(l) .L##l##_sse42
 165 # define GLABEL(l) l##_sse42
 166 # define SECTION sse4.2
 167 # include "strcmp-sse42.S"
 168
 169
 170 # ifdef HAVE_AVX_SUPPORT
 171 #  if defined USE_AS_STRCASECMP_L || defined USE_AS_STRNCASECMP_L
 172 #   define LABEL(l) .L##l##_avx
 173 #   define GLABEL(l) l##_avx
 174 #   define USE_AVX 1
 175 #   undef STRCMP_SSE42
 176 #   define STRCMP_SSE42 STRCMP_AVX
 177 #   define SECTION avx
 178 #   include "strcmp-sse42.S"
 179 #  endif
 180 # endif
 181
 182
 183 # undef ENTRY
 184 # define ENTRY(name) \
 185         .type STRCMP_SSE2, @function; \
 186         .align 16; \
 187         .globl STRCMP_SSE2; \
 188         .hidden STRCMP_SSE2; \
 189         STRCMP_SSE2: cfi_startproc; \
 190         CALL_MCOUNT
 191 # undef END
 192 # define END(name) \
 193         cfi_endproc; .size STRCMP_SSE2, .-STRCMP_SSE2
 194
 195 # ifdef USE_AS_STRCASECMP_L
 196 #  define ENTRY2(name) \
 197         .type __strcasecmp_sse2, @function; \
 198         .align 16; \
 199         .globl __strcasecmp_sse2; \
 200         .hidden __strcasecmp_sse2; \
 201         __strcasecmp_sse2: cfi_startproc; \
 202         CALL_MCOUNT
 203 #  define END2(name) \
 204         cfi_endproc; .size __strcasecmp_sse2, .-__strcasecmp_sse2
 205 # endif
 206
 207 # ifdef USE_AS_STRNCASECMP_L
 208 #  define ENTRY2(name) \
 209         .type __strncasecmp_sse2, @function; \
 210         .align 16; \
 211         .globl __strncasecmp_sse2; \
 212         .hidden __strncasecmp_sse2; \
 213         __strncasecmp_sse2: cfi_startproc; \
 214         CALL_MCOUNT
 215 #  define END2(name) \
 216         cfi_endproc; .size __strncasecmp_sse2, .-__strncasecmp_sse2
 217 # endif
 218
 219 # undef libc_hidden_builtin_def
 220 /* It doesn't make sense to send libc-internal strcmp calls through a PLT.
 221    The speedup we get from using SSE4.2 instruction is likely eaten away
 222    by the indirect call in the PLT.  */
 223 # define libc_hidden_builtin_def(name) \
 224         .globl __GI_STRCMP; __GI_STRCMP = STRCMP_SSE2
 225 #endif
 226
 227 #include "../strcmp.S"