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 wether 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_STRRCHR
  30 extern char *strrchr(const char *s, int c);
  31
  32 #define __HAVE_ARCH_STRLEN
  33 extern size_t strlen(const char *s);
  34
  35 static __always_inline void * __memcpy(void * to, const void * from, size_t n)
  36 {
  37 int d0, d1, d2;
  38 __asm__ __volatile__(
  39         "rep ; movsl\n\t"
  40         "movl %4,%%ecx\n\t"
  41         "andl $3,%%ecx\n\t"
  42         "jz 1f\n\t"
  43         "rep ; movsb\n\t"
  44         "1:"
  45         : "=&c" (d0), "=&D" (d1), "=&S" (d2)
  46         : "0" (n/4), "g" (n), "1" ((long) to), "2" ((long) from)
  47         : "memory");
  48 return (to);
  49 }
  50
  51 /*
  52  * This looks ugly, but the compiler can optimize it totally,
  53  * as the count is constant.
  54  */
  55 static __always_inline void * __constant_memcpy(void * to, const void * from, size_t n)
  56 {
  57         long esi, edi;
  58         if (!n) return to;
  59 #if 1   /* want to do small copies with non-string ops? */
  60         switch (n) {
  61                 case 1: *(char*)to = *(char*)from; return to;
  62                 case 2: *(short*)to = *(short*)from; return to;
  63                 case 4: *(int*)to = *(int*)from; return to;
  64 #if 1   /* including those doable with two moves? */
  65                 case 3: *(short*)to = *(short*)from;
  66                         *((char*)to+2) = *((char*)from+2); return to;
  67                 case 5: *(int*)to = *(int*)from;
  68                         *((char*)to+4) = *((char*)from+4); return to;
  69                 case 6: *(int*)to = *(int*)from;
  70                         *((short*)to+2) = *((short*)from+2); return to;
  71                 case 8: *(int*)to = *(int*)from;
  72                         *((int*)to+1) = *((int*)from+1); return to;
  73 #endif
  74         }
  75 #endif
  76         esi = (long) from;
  77         edi = (long) to;
  78         if (n >= 5*4) {
  79                 /* large block: use rep prefix */
  80                 int ecx;
  81                 __asm__ __volatile__(
  82                         "rep ; movsl"
  83                         : "=&c" (ecx), "=&D" (edi), "=&S" (esi)
  84                         : "0" (n/4), "1" (edi),"2" (esi)
  85                         : "memory"
  86                 );
  87         } else {
  88                 /* small block: don't clobber ecx + smaller code */
  89                 if (n >= 4*4) __asm__ __volatile__("movsl"
  90                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  91                 if (n >= 3*4) __asm__ __volatile__("movsl"
  92                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  93                 if (n >= 2*4) __asm__ __volatile__("movsl"
  94                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  95                 if (n >= 1*4) __asm__ __volatile__("movsl"
  96                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
  97         }
  98         switch (n % 4) {
  99                 /* tail */
 100                 case 0: return to;
 101                 case 1: __asm__ __volatile__("movsb"
 102                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 103                         return to;
 104                 case 2: __asm__ __volatile__("movsw"
 105                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 106                         return to;
 107                 default: __asm__ __volatile__("movsw\n\tmovsb"
 108                         :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
 109                         return to;
 110         }
 111 }
 112
 113 #define __HAVE_ARCH_MEMCPY
 114
 115 #ifdef CONFIG_X86_USE_3DNOW
 116
 117 #include <asm/mmx.h>
 118
 119 /*
 120  *      This CPU favours 3DNow strongly (eg AMD Athlon)
 121  */
 122
 123 static inline void * __constant_memcpy3d(void * to, const void * from, size_t len)
 124 {
 125         if (len < 512)
 126                 return __constant_memcpy(to, from, len);
 127         return _mmx_memcpy(to, from, len);
 128 }
 129
 130 static __inline__ void *__memcpy3d(void *to, const void *from, size_t len)
 131 {
 132         if (len < 512)
 133                 return __memcpy(to, from, len);
 134         return _mmx_memcpy(to, from, len);
 135 }
 136
 137 #define memcpy(t, f, n) \
 138 (__builtin_constant_p(n) ? \
 139  __constant_memcpy3d((t),(f),(n)) : \
 140  __memcpy3d((t),(f),(n)))
 141
 142 #else
 143
 144 /*
 145  *      No 3D Now!
 146  */
 147
 148 #define memcpy(t, f, n) \
 149 (__builtin_constant_p(n) ? \
 150  __constant_memcpy((t),(f),(n)) : \
 151  __memcpy((t),(f),(n)))
 152
 153 #endif
 154
 155 #define __HAVE_ARCH_MEMMOVE
 156 void *memmove(void * dest,const void * src, size_t n);
 157
 158 #define memcmp __builtin_memcmp
 159
 160 #define __HAVE_ARCH_MEMCHR
 161 extern void *memchr(const void * cs,int c,size_t count);
 162
 163 static inline void * __memset_generic(void * s, char c,size_t count)
 164 {
 165 int d0, d1;
 166 __asm__ __volatile__(
 167         "rep\n\t"
 168         "stosb"
 169         : "=&c" (d0), "=&D" (d1)
 170         :"a" (c),"1" (s),"0" (count)
 171         :"memory");
 172 return s;
 173 }
 174
 175 /* we might want to write optimized versions of these later */
 176 #define __constant_count_memset(s,c,count) __memset_generic((s),(c),(count))
 177
 178 /*
 179  * memset(x,0,y) is a reasonably common thing to do, so we want to fill
 180  * things 32 bits at a time even when we don't know the size of the
 181  * area at compile-time..
 182  */
 183 static __always_inline void * __constant_c_memset(void * s, unsigned long c, size_t count)
 184 {
 185 int d0, d1;
 186 __asm__ __volatile__(
 187         "rep ; stosl\n\t"
 188         "testb $2,%b3\n\t"
 189         "je 1f\n\t"
 190         "stosw\n"
 191         "1:\ttestb $1,%b3\n\t"
 192         "je 2f\n\t"
 193         "stosb\n"
 194         "2:"
 195         :"=&c" (d0), "=&D" (d1)
 196         :"a" (c), "q" (count), "0" (count/4), "1" ((long) s)
 197         :"memory");
 198 return (s);
 199 }
 200
 201 /* Added by Gertjan van Wingerde to make minix and sysv module work */
 202 #define __HAVE_ARCH_STRNLEN
 203 extern size_t strnlen(const char * s, size_t count);
 204 /* end of additional stuff */
 205
 206 #define __HAVE_ARCH_STRSTR
 207 extern char *strstr(const char *cs, const char *ct);
 208
 209 /*
 210  * This looks horribly ugly, but the compiler can optimize it totally,
 211  * as we by now know that both pattern and count is constant..
 212  */
 213 static __always_inline void * __constant_c_and_count_memset(void * s, unsigned long pattern, size_t count)
 214 {
 215         switch (count) {
 216                 case 0:
 217                         return s;
 218                 case 1:
 219                         *(unsigned char *)s = pattern;
 220                         return s;
 221                 case 2:
 222                         *(unsigned short *)s = pattern;
 223                         return s;
 224                 case 3:
 225                         *(unsigned short *)s = pattern;
 226                         *(2+(unsigned char *)s) = pattern;
 227                         return s;
 228                 case 4:
 229                         *(unsigned long *)s = pattern;
 230                         return s;
 231         }
 232 #define COMMON(x) \
 233 __asm__  __volatile__( \
 234         "rep ; stosl" \
 235         x \
 236         : "=&c" (d0), "=&D" (d1) \
 237         : "a" (pattern),"0" (count/4),"1" ((long) s) \
 238         : "memory")
 239 {
 240         int d0, d1;
 241         switch (count % 4) {
 242                 case 0: COMMON(""); return s;
 243                 case 1: COMMON("\n\tstosb"); return s;
 244                 case 2: COMMON("\n\tstosw"); return s;
 245                 default: COMMON("\n\tstosw\n\tstosb"); return s;
 246         }
 247 }
 248
 249 #undef COMMON
 250 }
 251
 252 #define __constant_c_x_memset(s, c, count) \
 253 (__builtin_constant_p(count) ? \
 254  __constant_c_and_count_memset((s),(c),(count)) : \
 255  __constant_c_memset((s),(c),(count)))
 256
 257 #define __memset(s, c, count) \
 258 (__builtin_constant_p(count) ? \
 259  __constant_count_memset((s),(c),(count)) : \
 260  __memset_generic((s),(c),(count)))
 261
 262 #define __HAVE_ARCH_MEMSET
 263 #define memset(s, c, count) \
 264 (__builtin_constant_p(c) ? \
 265  __constant_c_x_memset((s),(0x01010101UL*(unsigned char)(c)),(count)) : \
 266  __memset((s),(c),(count)))
 267
 268 /*
 269  * find the first occurrence of byte 'c', or 1 past the area if none
 270  */
 271 #define __HAVE_ARCH_MEMSCAN
 272 extern void *memscan(void * addr, int c, size_t size);
 273
 274 #endif /* __KERNEL__ */
 275
 276 #endif