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
  29 #define __HAVE_ARCH_STRCPY
  30 static inline char * strcpy(char * dest,const char *src)
  31 {
  32 int d0, d1, d2;
  33 __asm__ __volatile__(
  34         "1:\tlodsb\n\t"
  35         "stosb\n\t"
  36         "testb %%al,%%al\n\t"
  37         "jne 1b"
  38         : "=&S" (d0), "=&D" (d1), "=&a" (d2)
  39         :"0" (src),"1" (dest) : "memory");
  40 return dest;
  41 }
  42
  43 #define __HAVE_ARCH_STRNCPY
  44 static inline char * strncpy(char * dest,const char *src,size_t count)
  45 {
  46 int d0, d1, d2, d3;
  47 __asm__ __volatile__(
  48         "1:\tdecl %2\n\t"
  49         "js 2f\n\t"
  50         "lodsb\n\t"
  51         "stosb\n\t"
  52         "testb %%al,%%al\n\t"
  53         "jne 1b\n\t"
  54         "rep\n\t"
  55         "stosb\n"
  56         "2:"
  57         : "=&S" (d0), "=&D" (d1), "=&c" (d2), "=&a" (d3)
  58         :"0" (src),"1" (dest),"2" (count) : "memory");
  59 return dest;
  60 }
  61
  62 #define __HAVE_ARCH_STRCAT
  63 static inline char * strcat(char * dest,const char * src)
  64 {
  65 int d0, d1, d2, d3;
  66 __asm__ __volatile__(
  67         "repne\n\t"
  68         "scasb\n\t"
  69         "decl %1\n"
  70         "1:\tlodsb\n\t"
  71         "stosb\n\t"
  72         "testb %%al,%%al\n\t"
  73         "jne 1b"
  74         : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
  75         : "0" (src), "1" (dest), "2" (0), "3" (0xffffffffu):"memory");
  76 return dest;
  77 }
  78
  79 #define __HAVE_ARCH_STRNCAT
  80 static inline char * strncat(char * dest,const char * src,size_t count)
  81 {
  82 int d0, d1, d2, d3;
  83 __asm__ __volatile__(
  84         "repne\n\t"
  85         "scasb\n\t"
  86         "decl %1\n\t"
  87         "movl %8,%3\n"
  88         "1:\tdecl %3\n\t"
  89         "js 2f\n\t"
  90         "lodsb\n\t"
  91         "stosb\n\t"
  92         "testb %%al,%%al\n\t"
  93         "jne 1b\n"
  94         "2:\txorl %2,%2\n\t"
  95         "stosb"
  96         : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
  97         : "0" (src),"1" (dest),"2" (0),"3" (0xffffffffu), "g" (count)
  98         : "memory");
  99 return dest;
 100 }
 101
 102 #define __HAVE_ARCH_STRCMP
 103 static inline int strcmp(const char * cs,const char * ct)
 104 {
 105 int d0, d1;
 106 register int __res;
 107 __asm__ __volatile__(
 108         "1:\tlodsb\n\t"
 109         "scasb\n\t"
 110         "jne 2f\n\t"
 111         "testb %%al,%%al\n\t"
 112         "jne 1b\n\t"
 113         "xorl %%eax,%%eax\n\t"
 114         "jmp 3f\n"
 115         "2:\tsbbl %%eax,%%eax\n\t"
 116         "orb $1,%%al\n"
 117         "3:"
 118         :"=a" (__res), "=&S" (d0), "=&D" (d1)
 119         :"1" (cs),"2" (ct)
 120         :"memory");
 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         :"memory");
 145 return __res;
 146 }
 147
 148 #define __HAVE_ARCH_STRCHR
 149 static inline char * strchr(const char * s, int c)
 150 {
 151 int d0;
 152 register char * __res;
 153 __asm__ __volatile__(
 154         "movb %%al,%%ah\n"
 155         "1:\tlodsb\n\t"
 156         "cmpb %%ah,%%al\n\t"
 157         "je 2f\n\t"
 158         "testb %%al,%%al\n\t"
 159         "jne 1b\n\t"
 160         "movl $1,%1\n"
 161         "2:\tmovl %1,%0\n\t"
 162         "decl %0"
 163         :"=a" (__res), "=&S" (d0)
 164         :"1" (s),"0" (c)
 165         :"memory");
 166 return __res;
 167 }
 168
 169 #define __HAVE_ARCH_STRRCHR
 170 static inline char * strrchr(const char * s, int c)
 171 {
 172 int d0, d1;
 173 register char * __res;
 174 __asm__ __volatile__(
 175         "movb %%al,%%ah\n"
 176         "1:\tlodsb\n\t"
 177         "cmpb %%ah,%%al\n\t"
 178         "jne 2f\n\t"
 179         "leal -1(%%esi),%0\n"
 180         "2:\ttestb %%al,%%al\n\t"
 181         "jne 1b"
 182         :"=g" (__res), "=&S" (d0), "=&a" (d1)
 183         :"0" (0),"1" (s),"2" (c)
 184         :"memory");
 185 return __res;
 186 }
 187
 188 #define __HAVE_ARCH_STRLEN
 189 static inline size_t strlen(const char * s)
 190 {
 191 int d0;
 192 register int __res;
 193 __asm__ __volatile__(
 194         "repne\n\t"
 195         "scasb\n\t"
 196         "notl %0\n\t"
 197         "decl %0"
 198         :"=c" (__res), "=&D" (d0)
 199         :"1" (s),"a" (0), "0" (0xffffffffu)
 200         :"memory");
 201 return __res;
 202 }
 203
 204 static inline void * __memcpy(void * to, const void * from, size_t n)
 205 {
 206 int d0, d1, d2;
 207 __asm__ __volatile__(
 208         "rep ; movsl\n\t"
 209         "movl %4,%%ecx\n\t"
 210         "andl $3,%%ecx\n\t"
 211 #if 1   /* want to pay 2 byte penalty for a chance to skip microcoded rep? */
 212         "jz 1f\n\t"
 213 #endif
 214         "rep ; movsb\n\t"
 215         "1:"
 216         : "=&c" (d0), "=&D" (d1), "=&S" (d2)
 217         : "0" (n/4), "g" (n), "1" ((long) to), "2" ((long) from)
 218         : "memory");
 219 return (to);
 220 }
 221
 222 /*
 223  * This looks ugly, but the compiler can optimize it totally,
 224  * as the count is constant.
 225  */
 226 static inline void * __constant_memcpy(void * to, const void * from, size_t n)
 227 {
 228         long esi, edi;
 229         if (!n) return to;
 230 #if 1   /* want to do small copies with non-string ops? */
 231         switch (n) {
 232                 case 1: *(char*)to = *(char*)from; return to;
 233                 case 2: *(short*)to = *(short*)from; return to;
 234                 case 4: *(int*)to = *(int*)from; return to;
 235 #if 1   /* including those doable with two moves? */
 236                 case 3: *(short*)to = *(short*)from;
 237                         *((char*)to+2) = *((char*)from+2); return to;
 238                 case 5: *(int*)to = *(int*)from;
 239                         *((char*)to+4) = *((char*)from+4); return to;
 240                 case 6: *(int*)to = *(int*)from;
 241                         *((short*)to+2) = *((short*)from+2); return to;
 242                 case 8: *(int*)to = *(int*)from;
 243                         *((int*)to+1) = *((int*)from+1); return to;
 244 #endif
 245         }
 246 #endif
 247         esi = (long) from;
 248         edi = (long) to;
 249         if (n >= 5*4) {
 250                 /* large block: use rep prefix */
 251                 int ecx;
 252                 __asm__ __volatile__(
 253                         "rep ; movsl"
 254                         : "=&c" (ecx), "=&D" (edi), "=&S" (esi)
 255                         : "0" (n/4), "1" (edi),"2" (esi)
 256                         : "memory"
 257                 );
 258         } else {
 259                 /* small block: don't clobber ecx + smaller code */
 260                 if (n >= 4*4) __asm__ __volatile__("movsl"
 261                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 262                 if (n >= 3*4) __asm__ __volatile__("movsl"
 263                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 264                 if (n >= 2*4) __asm__ __volatile__("movsl"
 265                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 266                 if (n >= 1*4) __asm__ __volatile__("movsl"
 267                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 268         }
 269         switch (n % 4) {
 270                 /* tail */
 271                 case 0: return to;
 272                 case 1: __asm__ __volatile__("movsb"
 273                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 274                         return to;
 275                 case 2: __asm__ __volatile__("movsw"
 276                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 277                         return to;
 278                 default: __asm__ __volatile__("movsw\n\tmovsb"
 279                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 280                         return to;
 281         }
 282 }
 283
 284 #define __HAVE_ARCH_MEMCPY
 285
 286 #ifdef CONFIG_X86_USE_3DNOW
 287
 288 #include <asm/mmx.h>
 289
 290 /*
 291  *      This CPU favours 3DNow strongly (eg AMD Athlon)
 292  */
 293
 294 static inline void * __constant_memcpy3d(void * to, const void * from, size_t len)
 295 {
 296         if (len < 512)
 297                 return __constant_memcpy(to, from, len);
 298         return _mmx_memcpy(to, from, len);
 299 }
 300
 301 static __inline__ void *__memcpy3d(void *to, const void *from, size_t len)
 302 {
 303         if (len < 512)
 304                 return __memcpy(to, from, len);
 305         return _mmx_memcpy(to, from, len);
 306 }
 307
 308 #define memcpy(t, f, n) \
 309 (__builtin_constant_p(n) ? \
 310  __constant_memcpy3d((t),(f),(n)) : \
 311  __memcpy3d((t),(f),(n)))
 312
 313 #else
 314
 315 /*
 316  *      No 3D Now!
 317  */
 318
 319 #define memcpy(t, f, n) \
 320 (__builtin_constant_p(n) ? \
 321  __constant_memcpy((t),(f),(n)) : \
 322  __memcpy((t),(f),(n)))
 323
 324 #endif
 325
 326 #define __HAVE_ARCH_MEMMOVE
 327 void *memmove(void * dest,const void * src, size_t n);
 328
 329 #define memcmp __builtin_memcmp
 330
 331 #define __HAVE_ARCH_MEMCHR
 332 static inline void * memchr(const void * cs,int c,size_t count)
 333 {
 334 int d0;
 335 register void * __res;
 336 if (!count)
 337         return NULL;
 338 __asm__ __volatile__(
 339         "repne\n\t"
 340         "scasb\n\t"
 341         "je 1f\n\t"
 342         "movl $1,%0\n"
 343         "1:\tdecl %0"
 344         :"=D" (__res), "=&c" (d0)
 345         :"a" (c),"0" (cs),"1" (count)
 346         :"memory");
 347 return __res;
 348 }
 349
 350 static inline void * __memset_generic(void * s, char c,size_t count)
 351 {
 352 int d0, d1;
 353 __asm__ __volatile__(
 354         "rep\n\t"
 355         "stosb"
 356         : "=&c" (d0), "=&D" (d1)
 357         :"a" (c),"1" (s),"0" (count)
 358         :"memory");
 359 return s;
 360 }
 361
 362 /* we might want to write optimized versions of these later */
 363 #define __constant_count_memset(s,c,count) __memset_generic((s),(c),(count))
 364
 365 /*
 366  * memset(x,0,y) is a reasonably common thing to do, so we want to fill
 367  * things 32 bits at a time even when we don't know the size of the
 368  * area at compile-time..
 369  */
 370 static inline void * __constant_c_memset(void * s, unsigned long c, size_t count)
 371 {
 372 int d0, d1;
 373 __asm__ __volatile__(
 374         "rep ; stosl\n\t"
 375         "testb $2,%b3\n\t"
 376         "je 1f\n\t"
 377         "stosw\n"
 378         "1:\ttestb $1,%b3\n\t"
 379         "je 2f\n\t"
 380         "stosb\n"
 381         "2:"
 382         :"=&c" (d0), "=&D" (d1)
 383         :"a" (c), "q" (count), "0" (count/4), "1" ((long) s)
 384         :"memory");
 385 return (s);
 386 }
 387
 388 /* Added by Gertjan van Wingerde to make minix and sysv module work */
 389 #define __HAVE_ARCH_STRNLEN
 390 static inline size_t strnlen(const char * s, size_t count)
 391 {
 392 int d0;
 393 register int __res;
 394 __asm__ __volatile__(
 395         "movl %2,%0\n\t"
 396         "jmp 2f\n"
 397         "1:\tcmpb $0,(%0)\n\t"
 398         "je 3f\n\t"
 399         "incl %0\n"
 400         "2:\tdecl %1\n\t"
 401         "cmpl $-1,%1\n\t"
 402         "jne 1b\n"
 403         "3:\tsubl %2,%0"
 404         :"=a" (__res), "=&d" (d0)
 405         :"c" (s),"1" (count)
 406         :"memory");
 407 return __res;
 408 }
 409 /* end of additional stuff */
 410
 411 #define __HAVE_ARCH_STRSTR
 412
 413 extern char *strstr(const char *cs, const char *ct);
 414
 415 /*
 416  * This looks horribly ugly, but the compiler can optimize it totally,
 417  * as we by now know that both pattern and count is constant..
 418  */
 419 static inline void * __constant_c_and_count_memset(void * s, unsigned long pattern, size_t count)
 420 {
 421         switch (count) {
 422                 case 0:
 423                         return s;
 424                 case 1:
 425                         *(unsigned char *)s = pattern;
 426                         return s;
 427                 case 2:
 428                         *(unsigned short *)s = pattern;
 429                         return s;
 430                 case 3:
 431                         *(unsigned short *)s = pattern;
 432                         *(2+(unsigned char *)s) = pattern;
 433                         return s;
 434                 case 4:
 435                         *(unsigned long *)s = pattern;
 436                         return s;
 437         }
 438 #define COMMON(x) \
 439 __asm__  __volatile__( \
 440         "rep ; stosl" \
 441         x \
 442         : "=&c" (d0), "=&D" (d1) \
 443         : "a" (pattern),"0" (count/4),"1" ((long) s) \
 444         : "memory")
 445 {
 446         int d0, d1;
 447         switch (count % 4) {
 448                 case 0: COMMON(""); return s;
 449                 case 1: COMMON("\n\tstosb"); return s;
 450                 case 2: COMMON("\n\tstosw"); return s;
 451                 default: COMMON("\n\tstosw\n\tstosb"); return s;
 452         }
 453 }
 454
 455 #undef COMMON
 456 }
 457
 458 #define __constant_c_x_memset(s, c, count) \
 459 (__builtin_constant_p(count) ? \
 460  __constant_c_and_count_memset((s),(c),(count)) : \
 461  __constant_c_memset((s),(c),(count)))
 462
 463 #define __memset(s, c, count) \
 464 (__builtin_constant_p(count) ? \
 465  __constant_count_memset((s),(c),(count)) : \
 466  __memset_generic((s),(c),(count)))
 467
 468 #define __HAVE_ARCH_MEMSET
 469 #define memset(s, c, count) \
 470 (__builtin_constant_p(c) ? \
 471  __constant_c_x_memset((s),(0x01010101UL*(unsigned char)(c)),(count)) : \
 472  __memset((s),(c),(count)))
 473
 474 /*
 475  * find the first occurrence of byte 'c', or 1 past the area if none
 476  */
 477 #define __HAVE_ARCH_MEMSCAN
 478 static inline void * memscan(void * addr, int c, size_t size)
 479 {
 480         if (!size)
 481                 return addr;
 482         __asm__("repnz; scasb\n\t"
 483                 "jnz 1f\n\t"
 484                 "dec %%edi\n"
 485                 "1:"
 486                 : "=D" (addr), "=c" (size)
 487                 : "0" (addr), "1" (size), "a" (c)
 488                 : "memory");
 489         return addr;
 490 }
 491
 492 #endif /* __KERNEL__ */
 493
 494 #endif