include/asm-x86/string_32.h

   1 #ifndef _I386_STRING_H_
   2 #define _I386_STRING_H_
   3
   4 #ifdef __KERNEL__
   5
   6 /* Let gcc decide whether to inline or use the out of line functions */
   7
   8 #define __HAVE_ARCH_STRCPY
   9 extern char *strcpy(char *dest, const char *src);
  10
  11 #define __HAVE_ARCH_STRNCPY
  12 extern char *strncpy(char *dest, const char *src, size_t count);
  13
  14 #define __HAVE_ARCH_STRCAT
  15 extern char *strcat(char *dest, const char *src);
  16
  17 #define __HAVE_ARCH_STRNCAT
  18 extern char *strncat(char *dest, const char *src, size_t count);
  19
  20 #define __HAVE_ARCH_STRCMP
  21 extern int strcmp(const char *cs, const char *ct);
  22
  23 #define __HAVE_ARCH_STRNCMP
  24 extern int strncmp(const char *cs, const char *ct, size_t count);
  25
  26 #define __HAVE_ARCH_STRCHR
  27 extern char *strchr(const char *s, int c);
  28
  29 #define __HAVE_ARCH_STRLEN
  30 extern size_t strlen(const char *s);
  31
  32 static __always_inline void * __memcpy(void * to, const void * from, size_t n)
  33 {
  34 int d0, d1, d2;
  35 __asm__ __volatile__(
  36         "rep ; movsl\n\t"
  37         "movl %4,%%ecx\n\t"
  38         "andl $3,%%ecx\n\t"
  39         "jz 1f\n\t"
  40         "rep ; movsb\n\t"
  41         "1:"
  42         : "=&c" (d0), "=&D" (d1), "=&S" (d2)
  43         : "0" (n/4), "g" (n), "1" ((long) to), "2" ((long) from)
  44         : "memory");
  45 return (to);
  46 }
  47
  48 /*
  49  * This looks ugly, but the compiler can optimize it totally,
  50  * as the count is constant.
  51  */
  52 static __always_inline void * __constant_memcpy(void * to, const void * from, size_t n)
  53 {
  54         long esi, edi;
  55         if (!n) return to;
  56 #if 1   /* want to do small copies with non-string ops? */
  57         switch (n) {
  58                 case 1: *(char*)to = *(char*)from; return to;
  59                 case 2: *(short*)to = *(short*)from; return to;
  60                 case 4: *(int*)to = *(int*)from; return to;
  61 #if 1   /* including those doable with two moves? */
  62                 case 3: *(short*)to = *(short*)from;
  63                         *((char*)to+2) = *((char*)from+2); return to;
  64                 case 5: *(int*)to = *(int*)from;
  65                         *((char*)to+4) = *((char*)from+4); return to;
  66                 case 6: *(int*)to = *(int*)from;
  67                         *((short*)to+2) = *((short*)from+2); return to;
  68                 case 8: *(int*)to = *(int*)from;
  69                         *((int*)to+1) = *((int*)from+1); return to;
  70 #endif
  71         }
  72 #endif
  73         esi = (long) from;
  74         edi = (long) to;
  75         if (n >= 5*4) {
  76                 /* large block: use rep prefix */
  77                 int ecx;
  78                 __asm__ __volatile__(
  79                         "rep ; movsl"
  80                         : "=&c" (ecx), "=&D" (edi), "=&S" (esi)
  81                         : "0" (n/4), "1" (edi),"2" (esi)
  82                         : "memory"
  83                 );
  84         } else {
  85                 /* small block: don't clobber ecx + smaller code */
  86                 if (n >= 4*4) __asm__ __volatile__("movsl"
  87                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  88                 if (n >= 3*4) __asm__ __volatile__("movsl"
  89                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  90                 if (n >= 2*4) __asm__ __volatile__("movsl"
  91                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  92                 if (n >= 1*4) __asm__ __volatile__("movsl"
  93                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  94         }
  95         switch (n % 4) {
  96                 /* tail */
  97                 case 0: return to;
  98                 case 1: __asm__ __volatile__("movsb"
  99                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 100                         return to;
 101                 case 2: __asm__ __volatile__("movsw"
 102                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 103                         return to;
 104                 default: __asm__ __volatile__("movsw\n\tmovsb"
 105                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 106                         return to;
 107         }
 108 }
 109
 110 #define __HAVE_ARCH_MEMCPY
 111
 112 #ifdef CONFIG_X86_USE_3DNOW
 113
 114 #include <asm/mmx.h>
 115
 116 /*
 117  *      This CPU favours 3DNow strongly (eg AMD Athlon)
 118  */
 119
 120 static inline void * __constant_memcpy3d(void * to, const void * from, size_t len)
 121 {
 122         if (len < 512)
 123                 return __constant_memcpy(to, from, len);
 124         return _mmx_memcpy(to, from, len);
 125 }
 126
 127 static __inline__ void *__memcpy3d(void *to, const void *from, size_t len)
 128 {
 129         if (len < 512)
 130                 return __memcpy(to, from, len);
 131         return _mmx_memcpy(to, from, len);
 132 }
 133
 134 #define memcpy(t, f, n) \
 135 (__builtin_constant_p(n) ? \
 136  __constant_memcpy3d((t),(f),(n)) : \
 137  __memcpy3d((t),(f),(n)))
 138
 139 #else
 140
 141 /*
 142  *      No 3D Now!
 143  */
 144
 145 #define memcpy(t, f, n) \
 146 (__builtin_constant_p(n) ? \
 147  __constant_memcpy((t),(f),(n)) : \
 148  __memcpy((t),(f),(n)))
 149
 150 #endif
 151
 152 #define __HAVE_ARCH_MEMMOVE
 153 void *memmove(void * dest,const void * src, size_t n);
 154
 155 #define memcmp __builtin_memcmp
 156
 157 #define __HAVE_ARCH_MEMCHR
 158 extern void *memchr(const void * cs,int c,size_t count);
 159
 160 static inline void * __memset_generic(void * s, char c,size_t count)
 161 {
 162 int d0, d1;
 163 __asm__ __volatile__(
 164         "rep\n\t"
 165         "stosb"
 166         : "=&c" (d0), "=&D" (d1)
 167         :"a" (c),"1" (s),"0" (count)
 168         :"memory");
 169 return s;
 170 }
 171
 172 /* we might want to write optimized versions of these later */
 173 #define __constant_count_memset(s,c,count) __memset_generic((s),(c),(count))
 174
 175 /*
 176  * memset(x,0,y) is a reasonably common thing to do, so we want to fill
 177  * things 32 bits at a time even when we don't know the size of the
 178  * area at compile-time..
 179  */
 180 static __always_inline void * __constant_c_memset(void * s, unsigned long c, size_t count)
 181 {
 182 int d0, d1;
 183 __asm__ __volatile__(
 184         "rep ; stosl\n\t"
 185         "testb $2,%b3\n\t"
 186         "je 1f\n\t"
 187         "stosw\n"
 188         "1:\ttestb $1,%b3\n\t"
 189         "je 2f\n\t"
 190         "stosb\n"
 191         "2:"
 192         :"=&c" (d0), "=&D" (d1)
 193         :"a" (c), "q" (count), "0" (count/4), "1" ((long) s)
 194         :"memory");
 195 return (s);
 196 }
 197
 198 /* Added by Gertjan van Wingerde to make minix and sysv module work */
 199 #define __HAVE_ARCH_STRNLEN
 200 extern size_t strnlen(const char * s, size_t count);
 201 /* end of additional stuff */
 202
 203 #define __HAVE_ARCH_STRSTR
 204 extern char *strstr(const char *cs, const char *ct);
 205
 206 /*
 207  * This looks horribly ugly, but the compiler can optimize it totally,
 208  * as we by now know that both pattern and count is constant..
 209  */
 210 static __always_inline void * __constant_c_and_count_memset(void * s, unsigned long pattern, size_t count)
 211 {
 212         switch (count) {
 213                 case 0:
 214                         return s;
 215                 case 1:
 216                         *(unsigned char *)s = pattern & 0xff;
 217                         return s;
 218                 case 2:
 219                         *(unsigned short *)s = pattern & 0xffff;
 220                         return s;
 221                 case 3:
 222                         *(unsigned short *)s = pattern & 0xffff;
 223                         *(2+(unsigned char *)s) = pattern & 0xff;
 224                         return s;
 225                 case 4:
 226                         *(unsigned long *)s = pattern;
 227                         return s;
 228         }
 229 #define COMMON(x) \
 230 __asm__  __volatile__( \
 231         "rep ; stosl" \
 232         x \
 233         : "=&c" (d0), "=&D" (d1) \
 234         : "a" (pattern),"0" (count/4),"1" ((long) s) \
 235         : "memory")
 236 {
 237         int d0, d1;
 238         switch (count % 4) {
 239                 case 0: COMMON(""); return s;
 240                 case 1: COMMON("\n\tstosb"); return s;
 241                 case 2: COMMON("\n\tstosw"); return s;
 242                 default: COMMON("\n\tstosw\n\tstosb"); return s;
 243         }
 244 }
 245
 246 #undef COMMON
 247 }
 248
 249 #define __constant_c_x_memset(s, c, count) \
 250 (__builtin_constant_p(count) ? \
 251  __constant_c_and_count_memset((s),(c),(count)) : \
 252  __constant_c_memset((s),(c),(count)))
 253
 254 #define __memset(s, c, count) \
 255 (__builtin_constant_p(count) ? \
 256  __constant_count_memset((s),(c),(count)) : \
 257  __memset_generic((s),(c),(count)))
 258
 259 #define __HAVE_ARCH_MEMSET
 260 #define memset(s, c, count) \
 261 (__builtin_constant_p(c) ? \
 262  __constant_c_x_memset((s),(0x01010101UL*(unsigned char)(c)),(count)) : \
 263  __memset((s),(c),(count)))
 264
 265 /*
 266  * find the first occurrence of byte 'c', or 1 past the area if none
 267  */
 268 #define __HAVE_ARCH_MEMSCAN
 269 extern void *memscan(void * addr, int c, size_t size);
 270
 271 #endif /* __KERNEL__ */
 272
 273 #endif