include/asm-i386/string.h

   1 #ifndef _I386_STRING_H_
   2 #define _I386_STRING_H_
   3
   4 #ifdef __KERNEL__
   5 #include <linux/config.h>
   6 /*
   7  * On a 486 or Pentium, we are better off not using the
   8  * byte string operations. But on a 386 or a PPro the
   9  * byte string ops are faster than doing it by hand
  10  * (MUCH faster on a Pentium).
  11  */
  12
  13 /*
  14  * This string-include defines all string functions as inline
  15  * functions. Use gcc. It also assumes ds=es=data space, this should be
  16  * normal. Most of the string-functions are rather heavily hand-optimized,
  17  * see especially strsep,strstr,str[c]spn. They should work, but are not
  18  * very easy to understand. Everything is done entirely within the register
  19  * set, making the functions fast and clean. String instructions have been
  20  * used through-out, making for "slightly" unclear code :-)
  21  *
  22  *              NO Copyright (C) 1991, 1992 Linus Torvalds,
  23  *              consider these trivial functions to be PD.
  24  */
  25
  26 /* AK: in fact I bet it would be better to move this stuff all out of line.
  27  */
  28 #if !defined(IN_STRING_C)
  29
  30 #define __HAVE_ARCH_STRCPY
  31 static inline char * strcpy(char * dest,const char *src)
  32 {
  33 int d0, d1, d2;
  34 __asm__ __volatile__(
  35         "1:\tlodsb\n\t"
  36         "stosb\n\t"
  37         "testb %%al,%%al\n\t"
  38         "jne 1b"
  39         : "=&S" (d0), "=&D" (d1), "=&a" (d2)
  40         :"0" (src),"1" (dest) : "memory");
  41 return dest;
  42 }
  43
  44 #define __HAVE_ARCH_STRNCPY
  45 static inline char * strncpy(char * dest,const char *src,size_t count)
  46 {
  47 int d0, d1, d2, d3;
  48 __asm__ __volatile__(
  49         "1:\tdecl %2\n\t"
  50         "js 2f\n\t"
  51         "lodsb\n\t"
  52         "stosb\n\t"
  53         "testb %%al,%%al\n\t"
  54         "jne 1b\n\t"
  55         "rep\n\t"
  56         "stosb\n"
  57         "2:"
  58         : "=&S" (d0), "=&D" (d1), "=&c" (d2), "=&a" (d3)
  59         :"0" (src),"1" (dest),"2" (count) : "memory");
  60 return dest;
  61 }
  62
  63 #define __HAVE_ARCH_STRCAT
  64 static inline char * strcat(char * dest,const char * src)
  65 {
  66 int d0, d1, d2, d3;
  67 __asm__ __volatile__(
  68         "repne\n\t"
  69         "scasb\n\t"
  70         "decl %1\n"
  71         "1:\tlodsb\n\t"
  72         "stosb\n\t"
  73         "testb %%al,%%al\n\t"
  74         "jne 1b"
  75         : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
  76         : "0" (src), "1" (dest), "2" (0), "3" (0xffffffffu):"memory");
  77 return dest;
  78 }
  79
  80 #define __HAVE_ARCH_STRNCAT
  81 static inline char * strncat(char * dest,const char * src,size_t count)
  82 {
  83 int d0, d1, d2, d3;
  84 __asm__ __volatile__(
  85         "repne\n\t"
  86         "scasb\n\t"
  87         "decl %1\n\t"
  88         "movl %8,%3\n"
  89         "1:\tdecl %3\n\t"
  90         "js 2f\n\t"
  91         "lodsb\n\t"
  92         "stosb\n\t"
  93         "testb %%al,%%al\n\t"
  94         "jne 1b\n"
  95         "2:\txorl %2,%2\n\t"
  96         "stosb"
  97         : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
  98         : "0" (src),"1" (dest),"2" (0),"3" (0xffffffffu), "g" (count)
  99         : "memory");
 100 return dest;
 101 }
 102
 103 #define __HAVE_ARCH_STRCMP
 104 static inline int strcmp(const char * cs,const char * ct)
 105 {
 106 int d0, d1;
 107 register int __res;
 108 __asm__ __volatile__(
 109         "1:\tlodsb\n\t"
 110         "scasb\n\t"
 111         "jne 2f\n\t"
 112         "testb %%al,%%al\n\t"
 113         "jne 1b\n\t"
 114         "xorl %%eax,%%eax\n\t"
 115         "jmp 3f\n"
 116         "2:\tsbbl %%eax,%%eax\n\t"
 117         "orb $1,%%al\n"
 118         "3:"
 119         :"=a" (__res), "=&S" (d0), "=&D" (d1)
 120                      :"1" (cs),"2" (ct));
 121 return __res;
 122 }
 123
 124 #define __HAVE_ARCH_STRNCMP
 125 static inline int strncmp(const char * cs,const char * ct,size_t count)
 126 {
 127 register int __res;
 128 int d0, d1, d2;
 129 __asm__ __volatile__(
 130         "1:\tdecl %3\n\t"
 131         "js 2f\n\t"
 132         "lodsb\n\t"
 133         "scasb\n\t"
 134         "jne 3f\n\t"
 135         "testb %%al,%%al\n\t"
 136         "jne 1b\n"
 137         "2:\txorl %%eax,%%eax\n\t"
 138         "jmp 4f\n"
 139         "3:\tsbbl %%eax,%%eax\n\t"
 140         "orb $1,%%al\n"
 141         "4:"
 142                      :"=a" (__res), "=&S" (d0), "=&D" (d1), "=&c" (d2)
 143                      :"1" (cs),"2" (ct),"3" (count));
 144 return __res;
 145 }
 146
 147 #define __HAVE_ARCH_STRCHR
 148 static inline char * strchr(const char * s, int c)
 149 {
 150 int d0;
 151 register char * __res;
 152 __asm__ __volatile__(
 153         "movb %%al,%%ah\n"
 154         "1:\tlodsb\n\t"
 155         "cmpb %%ah,%%al\n\t"
 156         "je 2f\n\t"
 157         "testb %%al,%%al\n\t"
 158         "jne 1b\n\t"
 159         "movl $1,%1\n"
 160         "2:\tmovl %1,%0\n\t"
 161         "decl %0"
 162         :"=a" (__res), "=&S" (d0) : "1" (s),"0" (c));
 163 return __res;
 164 }
 165
 166 #define __HAVE_ARCH_STRRCHR
 167 static inline char * strrchr(const char * s, int c)
 168 {
 169 int d0, d1;
 170 register char * __res;
 171 __asm__ __volatile__(
 172         "movb %%al,%%ah\n"
 173         "1:\tlodsb\n\t"
 174         "cmpb %%ah,%%al\n\t"
 175         "jne 2f\n\t"
 176         "leal -1(%%esi),%0\n"
 177         "2:\ttestb %%al,%%al\n\t"
 178         "jne 1b"
 179         :"=g" (__res), "=&S" (d0), "=&a" (d1) :"0" (0),"1" (s),"2" (c));
 180 return __res;
 181 }
 182
 183 #endif
 184
 185 #define __HAVE_ARCH_STRLEN
 186 static inline size_t strlen(const char * s)
 187 {
 188 int d0;
 189 register int __res;
 190 __asm__ __volatile__(
 191         "repne\n\t"
 192         "scasb\n\t"
 193         "notl %0\n\t"
 194         "decl %0"
 195         :"=c" (__res), "=&D" (d0) :"1" (s),"a" (0), "0" (0xffffffffu));
 196 return __res;
 197 }
 198
 199 static inline void * __memcpy(void * to, const void * from, size_t n)
 200 {
 201 int d0, d1, d2;
 202 __asm__ __volatile__(
 203         "rep ; movsl\n\t"
 204         "testb $2,%b4\n\t"
 205         "je 1f\n\t"
 206         "movsw\n"
 207         "1:\ttestb $1,%b4\n\t"
 208         "je 2f\n\t"
 209         "movsb\n"
 210         "2:"
 211         : "=&c" (d0), "=&D" (d1), "=&S" (d2)
 212         :"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from)
 213         : "memory");
 214 return (to);
 215 }
 216
 217 /*
 218  * This looks horribly ugly, but the compiler can optimize it totally,
 219  * as the count is constant.
 220  */
 221 static inline void * __constant_memcpy(void * to, const void * from, size_t n)
 222 {
 223         if (n <= 128)
 224                 return __builtin_memcpy(to, from, n);
 225
 226 #define COMMON(x) \
 227 __asm__ __volatile__( \
 228         "rep ; movsl" \
 229         x \
 230         : "=&c" (d0), "=&D" (d1), "=&S" (d2) \
 231         : "0" (n/4),"1" ((long) to),"2" ((long) from) \
 232         : "memory");
 233 {
 234         int d0, d1, d2;
 235         switch (n % 4) {
 236                 case 0: COMMON(""); return to;
 237                 case 1: COMMON("\n\tmovsb"); return to;
 238                 case 2: COMMON("\n\tmovsw"); return to;
 239                 default: COMMON("\n\tmovsw\n\tmovsb"); return to;
 240         }
 241 }
 242
 243 #undef COMMON
 244 }
 245
 246 #define __HAVE_ARCH_MEMCPY
 247
 248 #ifdef CONFIG_X86_USE_3DNOW
 249
 250 #include <asm/mmx.h>
 251
 252 /*
 253  *      This CPU favours 3DNow strongly (eg AMD Athlon)
 254  */
 255
 256 static inline void * __constant_memcpy3d(void * to, const void * from, size_t len)
 257 {
 258         if (len < 512)
 259                 return __constant_memcpy(to, from, len);
 260         return _mmx_memcpy(to, from, len);
 261 }
 262
 263 static __inline__ void *__memcpy3d(void *to, const void *from, size_t len)
 264 {
 265         if (len < 512)
 266                 return __memcpy(to, from, len);
 267         return _mmx_memcpy(to, from, len);
 268 }
 269
 270 #define memcpy(t, f, n) \
 271 (__builtin_constant_p(n) ? \
 272  __constant_memcpy3d((t),(f),(n)) : \
 273  __memcpy3d((t),(f),(n)))
 274
 275 #else
 276
 277 /*
 278  *      No 3D Now!
 279  */
 280
 281 #define memcpy(t, f, n) \
 282 (__builtin_constant_p(n) ? \
 283  __constant_memcpy((t),(f),(n)) : \
 284  __memcpy((t),(f),(n)))
 285
 286 #endif
 287
 288 #define __HAVE_ARCH_MEMMOVE
 289 void *memmove(void * dest,const void * src, size_t n);
 290
 291 #define memcmp __builtin_memcmp
 292
 293 #define __HAVE_ARCH_MEMCHR
 294 static inline void * memchr(const void * cs,int c,size_t count)
 295 {
 296 int d0;
 297 register void * __res;
 298 if (!count)
 299         return NULL;
 300 __asm__ __volatile__(
 301         "repne\n\t"
 302         "scasb\n\t"
 303         "je 1f\n\t"
 304         "movl $1,%0\n"
 305         "1:\tdecl %0"
 306         :"=D" (__res), "=&c" (d0) : "a" (c),"0" (cs),"1" (count));
 307 return __res;
 308 }
 309
 310 static inline void * __memset_generic(void * s, char c,size_t count)
 311 {
 312 int d0, d1;
 313 __asm__ __volatile__(
 314         "rep\n\t"
 315         "stosb"
 316         : "=&c" (d0), "=&D" (d1)
 317         :"a" (c),"1" (s),"0" (count)
 318         :"memory");
 319 return s;
 320 }
 321
 322 /* we might want to write optimized versions of these later */
 323 #define __constant_count_memset(s,c,count) __memset_generic((s),(c),(count))
 324
 325 /*
 326  * memset(x,0,y) is a reasonably common thing to do, so we want to fill
 327  * things 32 bits at a time even when we don't know the size of the
 328  * area at compile-time..
 329  */
 330 static inline void * __constant_c_memset(void * s, unsigned long c, size_t count)
 331 {
 332 int d0, d1;
 333 __asm__ __volatile__(
 334         "rep ; stosl\n\t"
 335         "testb $2,%b3\n\t"
 336         "je 1f\n\t"
 337         "stosw\n"
 338         "1:\ttestb $1,%b3\n\t"
 339         "je 2f\n\t"
 340         "stosb\n"
 341         "2:"
 342         : "=&c" (d0), "=&D" (d1)
 343         :"a" (c), "q" (count), "0" (count/4), "1" ((long) s)
 344         :"memory");
 345 return (s);
 346 }
 347
 348 /* Added by Gertjan van Wingerde to make minix and sysv module work */
 349 #define __HAVE_ARCH_STRNLEN
 350 static inline size_t strnlen(const char * s, size_t count)
 351 {
 352 int d0;
 353 register int __res;
 354 __asm__ __volatile__(
 355         "movl %2,%0\n\t"
 356         "jmp 2f\n"
 357         "1:\tcmpb $0,(%0)\n\t"
 358         "je 3f\n\t"
 359         "incl %0\n"
 360         "2:\tdecl %1\n\t"
 361         "cmpl $-1,%1\n\t"
 362         "jne 1b\n"
 363         "3:\tsubl %2,%0"
 364         :"=a" (__res), "=&d" (d0)
 365         :"c" (s),"1" (count));
 366 return __res;
 367 }
 368 /* end of additional stuff */
 369
 370 #define __HAVE_ARCH_STRSTR
 371
 372 extern char *strstr(const char *cs, const char *ct);
 373
 374 /*
 375  * This looks horribly ugly, but the compiler can optimize it totally,
 376  * as we by now know that both pattern and count is constant..
 377  */
 378 static inline void * __constant_c_and_count_memset(void * s, unsigned long pattern, size_t count)
 379 {
 380         switch (count) {
 381                 case 0:
 382                         return s;
 383                 case 1:
 384                         *(unsigned char *)s = pattern;
 385                         return s;
 386                 case 2:
 387                         *(unsigned short *)s = pattern;
 388                         return s;
 389                 case 3:
 390                         *(unsigned short *)s = pattern;
 391                         *(2+(unsigned char *)s) = pattern;
 392                         return s;
 393                 case 4:
 394                         *(unsigned long *)s = pattern;
 395                         return s;
 396         }
 397 #define COMMON(x) \
 398 __asm__  __volatile__( \
 399         "rep ; stosl" \
 400         x \
 401         : "=&c" (d0), "=&D" (d1) \
 402         : "a" (pattern),"0" (count/4),"1" ((long) s) \
 403         : "memory")
 404 {
 405         int d0, d1;
 406         switch (count % 4) {
 407                 case 0: COMMON(""); return s;
 408                 case 1: COMMON("\n\tstosb"); return s;
 409                 case 2: COMMON("\n\tstosw"); return s;
 410                 default: COMMON("\n\tstosw\n\tstosb"); return s;
 411         }
 412 }
 413
 414 #undef COMMON
 415 }
 416
 417 #define __constant_c_x_memset(s, c, count) \
 418 (__builtin_constant_p(count) ? \
 419  __constant_c_and_count_memset((s),(c),(count)) : \
 420  __constant_c_memset((s),(c),(count)))
 421
 422 #define __memset(s, c, count) \
 423 (__builtin_constant_p(count) ? \
 424  __constant_count_memset((s),(c),(count)) : \
 425  __memset_generic((s),(c),(count)))
 426
 427 #define __HAVE_ARCH_MEMSET
 428 #define memset(s, c, count) \
 429 (__builtin_constant_p(c) ? \
 430  __constant_c_x_memset((s),(0x01010101UL*(unsigned char)(c)),(count)) : \
 431  __memset((s),(c),(count)))
 432
 433 /*
 434  * find the first occurrence of byte 'c', or 1 past the area if none
 435  */
 436 #define __HAVE_ARCH_MEMSCAN
 437 static inline void * memscan(void * addr, int c, size_t size)
 438 {
 439         if (!size)
 440                 return addr;
 441         __asm__("repnz; scasb\n\t"
 442                 "jnz 1f\n\t"
 443                 "dec %%edi\n"
 444                 "1:"
 445                 : "=D" (addr), "=c" (size)
 446                 : "0" (addr), "1" (size), "a" (c));
 447         return addr;
 448 }
 449
 450 #endif /* __KERNEL__ */
 451
 452 #endif