URL https://opencores.org/ocsvn/or1k/or1k/trunk

Subversion Repositories or1k

[/] [or1k/] [trunk/] [linux/] [linux-2.4/] [include/] [asm-m68k/] [string.h] - Blame information for rev 1774

Go to most recent revision | Details | Compare with Previous | View Log


#ifndef _M68K_STRING_H_
#define _M68K_STRING_H_
 
#include <asm/setup.h>
#include <asm/page.h>
 
#define __HAVE_ARCH_STRCPY
static inline char * strcpy(char * dest,const char *src)
{
  char *xdest = dest;
 
  __asm__ __volatile__
       ("1:\tmoveb %1@+,%0@+\n\t"
        "jne 1b"
        : "=a" (dest), "=a" (src)
        : "0" (dest), "1" (src) : "memory");
  return xdest;
}
 
#define __HAVE_ARCH_STRNCPY
static inline char * strncpy(char *dest, const char *src, size_t n)
{
  char *xdest = dest;
 
  if (n == 0)
    return xdest;
 
  __asm__ __volatile__
       ("1:\tmoveb %1@+,%0@+\n\t"
        "jeq 2f\n\t"
        "subql #1,%2\n\t"
        "jne 1b\n\t"
        "2:"
        : "=a" (dest), "=a" (src), "=d" (n)
        : "0" (dest), "1" (src), "2" (n)
        : "memory");
  return xdest;
}
 
#define __HAVE_ARCH_STRCAT
static inline char * strcat(char * dest, const char * src)
{
        char *tmp = dest;
 
        while (*dest)
                dest++;
        while ((*dest++ = *src++))
                ;
 
        return tmp;
}
 
#define __HAVE_ARCH_STRNCAT
static inline char * strncat(char *dest, const char *src, size_t count)
{
        char *tmp = dest;
 
        if (count) {
                while (*dest)
                        dest++;
                while ((*dest++ = *src++)) {
                        if (--count == 0) {
                                *dest++='\0';
                                break;
                        }
                }
        }
 
        return tmp;
}
 
#define __HAVE_ARCH_STRCHR
static inline char * strchr(const char * s, int c)
{
  const char ch = c;
 
  for(; *s != ch; ++s)
    if (*s == '\0')
      return( NULL );
  return( (char *) s);
}
 
#if 0
#define __HAVE_ARCH_STRPBRK
static inline char *strpbrk(const char *cs,const char *ct)
{
  const char *sc1,*sc2;
 
  for( sc1 = cs; *sc1 != '\0'; ++sc1)
    for( sc2 = ct; *sc2 != '\0'; ++sc2)
      if (*sc1 == *sc2)
        return((char *) sc1);
  return( NULL );
}
#endif
 
#if 0
#define __HAVE_ARCH_STRSPN
static inline size_t strspn(const char *s, const char *accept)
{
  const char *p;
  const char *a;
  size_t count = 0;
 
  for (p = s; *p != '\0'; ++p)
    {
      for (a = accept; *a != '\0'; ++a)
        if (*p == *a)
          break;
      if (*a == '\0')
        return count;
      else
        ++count;
    }
 
  return count;
}
#endif
 
#if 0
#define __HAVE_ARCH_STRTOK
static inline char *strtok(char *s, const char *ct)
{
  char *sbegin, *send;
 
  sbegin  = s ? s : ___strtok;
  if (!sbegin) {
          return NULL;
  }
  sbegin += strspn(sbegin,ct);
  if (*sbegin == '\0') {
    ___strtok = NULL;
    return( NULL );
  }
  send = strpbrk( sbegin, ct);
  if (send && *send != '\0')
    *send++ = '\0';
  ___strtok = send;
  return (sbegin);
}
#endif
 
/* strstr !! */
 
#define __HAVE_ARCH_STRLEN
static inline size_t strlen(const char * s)
{
  const char *sc;
  for (sc = s; *sc != '\0'; ++sc) ;
  return(sc - s);
}
 
/* strnlen !! */
 
#define __HAVE_ARCH_STRCMP
static inline int strcmp(const char * cs,const char * ct)
{
  char __res;
 
  __asm__
       ("1:\tmoveb %0@+,%2\n\t" /* get *cs */
        "cmpb %1@+,%2\n\t"      /* compare a byte */
        "jne  2f\n\t"           /* not equal, break out */
        "tstb %2\n\t"           /* at end of cs? */
        "jne  1b\n\t"           /* no, keep going */
        "jra  3f\n\t"           /* strings are equal */
        "2:\tsubb %1@-,%2\n\t"  /* *cs - *ct */
        "3:"
        : "=a" (cs), "=a" (ct), "=d" (__res)
        : "0" (cs), "1" (ct));
  return __res;
}
 
#define __HAVE_ARCH_STRNCMP
static inline int strncmp(const char * cs,const char * ct,size_t count)
{
  char __res;
 
  if (!count)
    return 0;
  __asm__
       ("1:\tmovb %0@+,%3\n\t"          /* get *cs */
        "cmpb   %1@+,%3\n\t"            /* compare a byte */
        "jne    3f\n\t"                 /* not equal, break out */
        "tstb   %3\n\t"                 /* at end of cs? */
        "jeq    4f\n\t"                 /* yes, all done */
        "subql  #1,%2\n\t"              /* no, adjust count */
        "jne    1b\n\t"                 /* more to do, keep going */
        "2:\tmoveq #0,%3\n\t"           /* strings are equal */
        "jra    4f\n\t"
        "3:\tsubb %1@-,%3\n\t"          /* *cs - *ct */
        "4:"
        : "=a" (cs), "=a" (ct), "=d" (count), "=d" (__res)
        : "0" (cs), "1" (ct), "2" (count));
  return __res;
}
 
#define __HAVE_ARCH_MEMSET
/*
 * This is really ugly, but its highly optimizatiable by the
 * compiler and is meant as compensation for gcc's missing
 * __builtin_memset(). For the 680[23]0 it might be worth considering
 * the optimal number of misaligned writes compared to the number of
 * tests'n'branches needed to align the destination address. The
 * 680[46]0 doesn't really care due to their copy-back caches.
 *                                              10/09/96 - Jes Sorensen
 */
static inline void * __memset_g(void * s, int c, size_t count)
{
  void *xs = s;
  size_t temp;
 
  if (!count)
    return xs;
 
  c &= 0xff;
  c |= c << 8;
  c |= c << 16;
 
  if (count < 36){
          long *ls = s;
 
          switch(count){
          case 32: case 33: case 34: case 35:
                  *ls++ = c;
          case 28: case 29: case 30: case 31:
                  *ls++ = c;
          case 24: case 25: case 26: case 27:
                  *ls++ = c;
          case 20: case 21: case 22: case 23:
                  *ls++ = c;
          case 16: case 17: case 18: case 19:
                  *ls++ = c;
          case 12: case 13: case 14: case 15:
                  *ls++ = c;
          case 8: case 9: case 10: case 11:
                  *ls++ = c;
          case 4: case 5: case 6: case 7:
                  *ls++ = c;
                  break;
          default:
                  break;
          }
          s = ls;
          if (count & 0x02){
                  short *ss = s;
                  *ss++ = c;
                  s = ss;
          }
          if (count & 0x01){
                  char *cs = s;
                  *cs++ = c;
                  s = cs;
          }
          return xs;
  }
 
  if ((long) s & 1)
    {
      char *cs = s;
      *cs++ = c;
      s = cs;
      count--;
    }
  if (count > 2 && (long) s & 2)
    {
      short *ss = s;
      *ss++ = c;
      s = ss;
      count -= 2;
    }
  temp = count >> 2;
  if (temp)
    {
      long *ls = s;
      temp--;
      do
        *ls++ = c;
      while (temp--);
      s = ls;
    }
  if (count & 2)
    {
      short *ss = s;
      *ss++ = c;
      s = ss;
    }
  if (count & 1)
    {
      char *cs = s;
      *cs = c;
    }
  return xs;
}
 
/*
 * __memset_page assumes that data is longword aligned. Most, if not
 * all, of these page sized memsets are performed on page aligned
 * areas, thus we do not need to check if the destination is longword
 * aligned. Of course we suffer a serious performance loss if this is
 * not the case but I think the risk of this ever happening is
 * extremely small. We spend a lot of time clearing pages in
 * get_empty_page() so I think it is worth it anyway. Besides, the
 * 680[46]0 do not really care about misaligned writes due to their
 * copy-back cache.
 *
 * The optimized case for the 680[46]0 is implemented using the move16
 * instruction. My tests showed that this implementation is 35-45%
 * faster than the original implementation using movel, the only
 * caveat is that the destination address must be 16-byte aligned.
 *                                            01/09/96 - Jes Sorensen
 */
static inline void * __memset_page(void * s,int c,size_t count)
{
  unsigned long data, tmp;
  void *xs, *sp;
 
  xs = sp = s;
 
  c = c & 255;
  data = c | (c << 8);
  data |= data << 16;
 
#ifdef CPU_M68040_OR_M68060_ONLY
 
  if (((unsigned long) s) & 0x0f)
          __memset_g(s, c, count);
  else{
          *((unsigned long *)(s))++ = data;
          *((unsigned long *)(s))++ = data;
          *((unsigned long *)(s))++ = data;
          *((unsigned long *)(s))++ = data;
 
          __asm__ __volatile__("1:\t"
                               ".chip 68040\n\t"
                               "move16 %2@+,%0@+\n\t"
                               ".chip 68k\n\t"
                               "subqw  #8,%2\n\t"
                               "subqw  #8,%2\n\t"
                               "dbra   %1,1b\n\t"
                               : "=a" (s), "=d" (tmp)
                               : "a" (sp), "0" (s), "1" ((count - 16) / 16 - 1)
                               );
  }
 
#else
  __asm__ __volatile__("1:\t"
                       "movel %2,%0@+\n\t"
                       "movel %2,%0@+\n\t"
                       "movel %2,%0@+\n\t"
                       "movel %2,%0@+\n\t"
                       "movel %2,%0@+\n\t"
                       "movel %2,%0@+\n\t"
                       "movel %2,%0@+\n\t"
                       "movel %2,%0@+\n\t"
                       "dbra  %1,1b\n\t"
                       : "=a" (s), "=d" (tmp)
                       : "d" (data), "0" (s), "1" (count / 32 - 1)
                       );
#endif
 
  return xs;
}
 
extern void *memset(void *,int,__kernel_size_t);
 
#define __memset_const(s,c,count) \
((count==PAGE_SIZE) ? \
  __memset_page((s),(c),(count)) : \
  __memset_g((s),(c),(count)))
 
#define memset(s, c, count) \
(__builtin_constant_p(count) ? \
 __memset_const((s),(c),(count)) : \
 __memset_g((s),(c),(count)))
 
#define __HAVE_ARCH_MEMCPY
extern void * memcpy(void *, const void *, size_t );
/*
 * __builtin_memcpy() does not handle page-sized memcpys very well,
 * thus following the same assumptions as for page-sized memsets, this
 * function copies page-sized areas using an unrolled loop, without
 * considering alignment.
 *
 * For the 680[46]0 only kernels we use the move16 instruction instead
 * as it writes through the data-cache, invalidating the cache-lines
 * touched. In this way we do not use up the entire data-cache (well,
 * half of it on the 68060) by copying a page. An unrolled loop of two
 * move16 instructions seem to the fastest. The only caveat is that
 * both source and destination must be 16-byte aligned, if not we fall
 * back to the generic memcpy function.  - Jes
 */
static inline void * __memcpy_page(void * to, const void * from, size_t count)
{
  unsigned long tmp;
  void *xto = to;
 
#ifdef CPU_M68040_OR_M68060_ONLY
 
  if (((unsigned long) to | (unsigned long) from) & 0x0f)
          return memcpy(to, from, count);
 
  __asm__ __volatile__("1:\t"
                       ".chip 68040\n\t"
                       "move16 %1@+,%0@+\n\t"
                       "move16 %1@+,%0@+\n\t"
                       ".chip 68k\n\t"
                       "dbra  %2,1b\n\t"
                       : "=a" (to), "=a" (from), "=d" (tmp)
                       : "0" (to), "1" (from) , "2" (count / 32 - 1)
                       );
#else
  __asm__ __volatile__("1:\t"
                       "movel %1@+,%0@+\n\t"
                       "movel %1@+,%0@+\n\t"
                       "movel %1@+,%0@+\n\t"
                       "movel %1@+,%0@+\n\t"
                       "movel %1@+,%0@+\n\t"
                       "movel %1@+,%0@+\n\t"
                       "movel %1@+,%0@+\n\t"
                       "movel %1@+,%0@+\n\t"
                       "dbra  %2,1b\n\t"
                       : "=a" (to), "=a" (from), "=d" (tmp)
                       : "0" (to), "1" (from) , "2" (count / 32 - 1)
                       );
#endif
  return xto;
}
 
#define __memcpy_const(to, from, n) \
((n==PAGE_SIZE) ? \
  __memcpy_page((to),(from),(n)) : \
  __builtin_memcpy((to),(from),(n)))
 
#define memcpy(to, from, n) \
(__builtin_constant_p(n) ? \
 __memcpy_const((to),(from),(n)) : \
 memcpy((to),(from),(n)))
 
#define __HAVE_ARCH_MEMMOVE
static inline void * memmove(void * dest,const void * src, size_t n)
{
  void *xdest = dest;
  size_t temp;
 
  if (!n)
    return xdest;
 
  if (dest < src)
    {
      if ((long) dest & 1)
        {
          char *cdest = dest;
          const char *csrc = src;
          *cdest++ = *csrc++;
          dest = cdest;
          src = csrc;
          n--;
        }
      if (n > 2 && (long) dest & 2)
        {
          short *sdest = dest;
          const short *ssrc = src;
          *sdest++ = *ssrc++;
          dest = sdest;
          src = ssrc;
          n -= 2;
        }
      temp = n >> 2;
      if (temp)
        {
          long *ldest = dest;
          const long *lsrc = src;
          temp--;
          do
            *ldest++ = *lsrc++;
          while (temp--);
          dest = ldest;
          src = lsrc;
        }
      if (n & 2)
        {
          short *sdest = dest;
          const short *ssrc = src;
          *sdest++ = *ssrc++;
          dest = sdest;
          src = ssrc;
        }
      if (n & 1)
        {
          char *cdest = dest;
          const char *csrc = src;
          *cdest = *csrc;
        }
    }
  else
    {
      dest = (char *) dest + n;
      src = (const char *) src + n;
      if ((long) dest & 1)
        {
          char *cdest = dest;
          const char *csrc = src;
          *--cdest = *--csrc;
          dest = cdest;
          src = csrc;
          n--;
        }
      if (n > 2 && (long) dest & 2)
        {
          short *sdest = dest;
          const short *ssrc = src;
          *--sdest = *--ssrc;
          dest = sdest;
          src = ssrc;
          n -= 2;
        }
      temp = n >> 2;
      if (temp)
        {
          long *ldest = dest;
          const long *lsrc = src;
          temp--;
          do
            *--ldest = *--lsrc;
          while (temp--);
          dest = ldest;
          src = lsrc;
        }
      if (n & 2)
        {
          short *sdest = dest;
          const short *ssrc = src;
          *--sdest = *--ssrc;
          dest = sdest;
          src = ssrc;
        }
      if (n & 1)
        {
          char *cdest = dest;
          const char *csrc = src;
          *--cdest = *--csrc;
        }
    }
  return xdest;
}
 
#define __HAVE_ARCH_MEMCMP
extern int memcmp(const void * ,const void * ,size_t );
#define memcmp(cs, ct, n) \
(__builtin_constant_p(n) ? \
 __builtin_memcmp((cs),(ct),(n)) : \
 memcmp((cs),(ct),(n)))
 
#define __HAVE_ARCH_MEMCHR
static inline void *memchr(const void *cs, int c, size_t count)
{
        /* Someone else can optimize this, I don't care - tonym@mac.linux-m68k.org */
        unsigned char *ret = (unsigned char *)cs;
        for(;count>0;count--,ret++)
                if(*ret == c) return ret;
 
        return NULL;
}
 
#endif /* _M68K_STRING_H_ */

Browse

Tools

Subversion Repositories or1k

[/] [or1k/] [trunk/] [linux/] [linux-2.4/] [include/] [asm-m68k/] [string.h] - Blame information for rev 1774

Line No.	Rev	Author	Line
1	1276	phoenix	`#ifndef _M68K_STRING_H_`
2			`#define _M68K_STRING_H_`
3
4			`#include <asm/setup.h>`
5			`#include <asm/page.h>`
6
7			`#define __HAVE_ARCH_STRCPY`
8			`static inline char * strcpy(char * dest,const char *src)`
9			`{`
10			`char *xdest = dest;`
11
12			`__asm__ __volatile__`
13			`("1:\tmoveb %1@+,%0@+\n\t"`
14			`"jne 1b"`
15			`: "=a" (dest), "=a" (src)`
16			`: "0" (dest), "1" (src) : "memory");`
17			`return xdest;`
18			`}`
19
20			`#define __HAVE_ARCH_STRNCPY`
21			`static inline char * strncpy(char dest, const char src, size_t n)`
22			`{`
23			`char *xdest = dest;`
24
25			`if (n == 0)`
26			`return xdest;`
27
28			`__asm__ __volatile__`
29			`("1:\tmoveb %1@+,%0@+\n\t"`
30			`"jeq 2f\n\t"`
31			`"subql #1,%2\n\t"`
32			`"jne 1b\n\t"`
33			`"2:"`
34			`: "=a" (dest), "=a" (src), "=d" (n)`
35			`: "0" (dest), "1" (src), "2" (n)`
36			`: "memory");`
37			`return xdest;`
38			`}`
39
40			`#define __HAVE_ARCH_STRCAT`
41			`static inline char * strcat(char * dest, const char * src)`
42			`{`
43			`char *tmp = dest;`
44
45			`while (*dest)`
46			`dest++;`
47			`while ((dest++ = src++))`
48			`;`
49
50			`return tmp;`
51			`}`
52
53			`#define __HAVE_ARCH_STRNCAT`
54			`static inline char * strncat(char dest, const char src, size_t count)`
55			`{`
56			`char *tmp = dest;`
57
58			`if (count) {`
59			`while (*dest)`
60			`dest++;`
61			`while ((dest++ = src++)) {`
62			`if (--count == 0) {`
63			`*dest++='\0';`
64			`break;`
65			`}`
66			`}`
67			`}`
68
69			`return tmp;`
70			`}`
71
72			`#define __HAVE_ARCH_STRCHR`
73			`static inline char * strchr(const char * s, int c)`
74			`{`
75			`const char ch = c;`
76
77			`for(; *s != ch; ++s)`
78			`if (*s == '\0')`
79			`return( NULL );`
80			`return( (char *) s);`
81			`}`
82
83			`#if 0`
84			`#define __HAVE_ARCH_STRPBRK`
85			`static inline char strpbrk(const char cs,const char *ct)`
86			`{`
87			`const char sc1,sc2;`
88
89			`for( sc1 = cs; *sc1 != '\0'; ++sc1)`
90			`for( sc2 = ct; *sc2 != '\0'; ++sc2)`
91			`if (sc1 == sc2)`
92			`return((char *) sc1);`
93			`return( NULL );`
94			`}`
95			`#endif`
96
97			`#if 0`
98			`#define __HAVE_ARCH_STRSPN`
99			`static inline size_t strspn(const char s, const char accept)`
100			`{`
101			`const char *p;`
102			`const char *a;`
103			`size_t count = 0;`
104
105			`for (p = s; *p != '\0'; ++p)`
106			`{`
107			`for (a = accept; *a != '\0'; ++a)`
108			`if (p == a)`
109			`break;`
110			`if (*a == '\0')`
111			`return count;`
112			`else`
113			`++count;`
114			`}`
115
116			`return count;`
117			`}`
118			`#endif`
119
120			`#if 0`
121			`#define __HAVE_ARCH_STRTOK`
122			`static inline char strtok(char s, const char *ct)`
123			`{`
124			`char sbegin, send;`
125
126			`sbegin = s ? s : ___strtok;`
127			`if (!sbegin) {`
128			`return NULL;`
129			`}`
130			`sbegin += strspn(sbegin,ct);`
131			`if (*sbegin == '\0') {`
132			`___strtok = NULL;`
133			`return( NULL );`
134			`}`
135			`send = strpbrk( sbegin, ct);`
136			`if (send && *send != '\0')`
137			`*send++ = '\0';`
138			`___strtok = send;`
139			`return (sbegin);`
140			`}`
141			`#endif`
142
143			`/* strstr !! */`
144
145			`#define __HAVE_ARCH_STRLEN`
146			`static inline size_t strlen(const char * s)`
147			`{`
148			`const char *sc;`
149			`for (sc = s; *sc != '\0'; ++sc) ;`
150			`return(sc - s);`
151			`}`
152
153			`/* strnlen !! */`
154
155			`#define __HAVE_ARCH_STRCMP`
156			`static inline int strcmp(const char * cs,const char * ct)`
157			`{`
158			`char __res;`
159
160			`__asm__`
161			`("1:\tmoveb %0@+,%2\n\t" /* get cs /`
162			`"cmpb %1@+,%2\n\t" /* compare a byte */`
163			`"jne 2f\n\t" /* not equal, break out */`
164			`"tstb %2\n\t" /* at end of cs? */`
165			`"jne 1b\n\t" /* no, keep going */`
166			`"jra 3f\n\t" /* strings are equal */`
167			`"2:\tsubb %1@-,%2\n\t" /* cs - ct */`
168			`"3:"`
169			`: "=a" (cs), "=a" (ct), "=d" (__res)`
170			`: "0" (cs), "1" (ct));`
171			`return __res;`
172			`}`
173
174			`#define __HAVE_ARCH_STRNCMP`
175			`static inline int strncmp(const char * cs,const char * ct,size_t count)`
176			`{`
177			`char __res;`
178
179			`if (!count)`
180			`return 0;`
181			`__asm__`
182			`("1:\tmovb %0@+,%3\n\t" /* get cs /`
183			`"cmpb %1@+,%3\n\t" /* compare a byte */`
184			`"jne 3f\n\t" /* not equal, break out */`
185			`"tstb %3\n\t" /* at end of cs? */`
186			`"jeq 4f\n\t" /* yes, all done */`
187			`"subql #1,%2\n\t" /* no, adjust count */`
188			`"jne 1b\n\t" /* more to do, keep going */`
189			`"2:\tmoveq #0,%3\n\t" /* strings are equal */`
190			`"jra 4f\n\t"`
191			`"3:\tsubb %1@-,%3\n\t" /* cs - ct */`
192			`"4:"`
193			`: "=a" (cs), "=a" (ct), "=d" (count), "=d" (__res)`
194			`: "0" (cs), "1" (ct), "2" (count));`
195			`return __res;`
196			`}`
197
198			`#define __HAVE_ARCH_MEMSET`
199			`/*`
200			`* This is really ugly, but its highly optimizatiable by the`
201			`* compiler and is meant as compensation for gcc's missing`
202			`* __builtin_memset(). For the 680[23]0 it might be worth considering`
203			`* the optimal number of misaligned writes compared to the number of`
204			`* tests'n'branches needed to align the destination address. The`
205			`* 680[46]0 doesn't really care due to their copy-back caches.`
206			`* 10/09/96 - Jes Sorensen`
207			`*/`
208			`static inline void * __memset_g(void * s, int c, size_t count)`
209			`{`
210			`void *xs = s;`
211			`size_t temp;`
212
213			`if (!count)`
214			`return xs;`
215
216			`c &= 0xff;`
217			`c \|= c << 8;`
218			`c \|= c << 16;`
219
220			`if (count < 36){`
221			`long *ls = s;`
222
223			`switch(count){`
224			`case 32: case 33: case 34: case 35:`
225			`*ls++ = c;`
226			`case 28: case 29: case 30: case 31:`
227			`*ls++ = c;`
228			`case 24: case 25: case 26: case 27:`
229			`*ls++ = c;`
230			`case 20: case 21: case 22: case 23:`
231			`*ls++ = c;`
232			`case 16: case 17: case 18: case 19:`
233			`*ls++ = c;`
234			`case 12: case 13: case 14: case 15:`
235			`*ls++ = c;`
236			`case 8: case 9: case 10: case 11:`
237			`*ls++ = c;`
238			`case 4: case 5: case 6: case 7:`
239			`*ls++ = c;`
240			`break;`
241			`default:`
242			`break;`
243			`}`
244			`s = ls;`
245			`if (count & 0x02){`
246			`short *ss = s;`
247			`*ss++ = c;`
248			`s = ss;`
249			`}`
250			`if (count & 0x01){`
251			`char *cs = s;`
252			`*cs++ = c;`
253			`s = cs;`
254			`}`
255			`return xs;`
256			`}`
257
258			`if ((long) s & 1)`
259			`{`
260			`char *cs = s;`
261			`*cs++ = c;`
262			`s = cs;`
263			`count--;`
264			`}`
265			`if (count > 2 && (long) s & 2)`
266			`{`
267			`short *ss = s;`
268			`*ss++ = c;`
269			`s = ss;`
270			`count -= 2;`
271			`}`
272			`temp = count >> 2;`
273			`if (temp)`
274			`{`
275			`long *ls = s;`
276			`temp--;`
277			`do`
278			`*ls++ = c;`
279			`while (temp--);`
280			`s = ls;`
281			`}`
282			`if (count & 2)`
283			`{`
284			`short *ss = s;`
285			`*ss++ = c;`
286			`s = ss;`
287			`}`
288			`if (count & 1)`
289			`{`
290			`char *cs = s;`
291			`*cs = c;`
292			`}`
293			`return xs;`
294			`}`
295
296			`/*`
297			`* __memset_page assumes that data is longword aligned. Most, if not`
298			`* all, of these page sized memsets are performed on page aligned`
299			`* areas, thus we do not need to check if the destination is longword`
300			`* aligned. Of course we suffer a serious performance loss if this is`
301			`* not the case but I think the risk of this ever happening is`
302			`* extremely small. We spend a lot of time clearing pages in`
303			`* get_empty_page() so I think it is worth it anyway. Besides, the`
304			`* 680[46]0 do not really care about misaligned writes due to their`
305			`* copy-back cache.`
306			`*`
307			`* The optimized case for the 680[46]0 is implemented using the move16`
308			`* instruction. My tests showed that this implementation is 35-45%`
309			`* faster than the original implementation using movel, the only`
310			`* caveat is that the destination address must be 16-byte aligned.`
311			`* 01/09/96 - Jes Sorensen`
312			`*/`
313			`static inline void * __memset_page(void * s,int c,size_t count)`
314			`{`
315			`unsigned long data, tmp;`
316			`void xs, sp;`
317
318			`xs = sp = s;`
319
320			`c = c & 255;`
321			`data = c \| (c << 8);`
322			`data \|= data << 16;`
323
324			`#ifdef CPU_M68040_OR_M68060_ONLY`
325
326			`if (((unsigned long) s) & 0x0f)`
327			`__memset_g(s, c, count);`
328			`else{`
329			`((unsigned long )(s))++ = data;`
330			`((unsigned long )(s))++ = data;`
331			`((unsigned long )(s))++ = data;`
332			`((unsigned long )(s))++ = data;`
333
334			`__asm__ __volatile__("1:\t"`
335			`".chip 68040\n\t"`
336			`"move16 %2@+,%0@+\n\t"`
337			`".chip 68k\n\t"`
338			`"subqw #8,%2\n\t"`
339			`"subqw #8,%2\n\t"`
340			`"dbra %1,1b\n\t"`
341			`: "=a" (s), "=d" (tmp)`
342			`: "a" (sp), "0" (s), "1" ((count - 16) / 16 - 1)`
343			`);`
344			`}`
345
346			`#else`
347			`__asm__ __volatile__("1:\t"`
348			`"movel %2,%0@+\n\t"`
349			`"movel %2,%0@+\n\t"`
350			`"movel %2,%0@+\n\t"`
351			`"movel %2,%0@+\n\t"`
352			`"movel %2,%0@+\n\t"`
353			`"movel %2,%0@+\n\t"`
354			`"movel %2,%0@+\n\t"`
355			`"movel %2,%0@+\n\t"`
356			`"dbra %1,1b\n\t"`
357			`: "=a" (s), "=d" (tmp)`
358			`: "d" (data), "0" (s), "1" (count / 32 - 1)`
359			`);`
360			`#endif`
361
362			`return xs;`
363			`}`
364
365			`extern void memset(void ,int,__kernel_size_t);`
366
367			`#define __memset_const(s,c,count) \`
368			`((count==PAGE_SIZE) ? \`
369			`__memset_page((s),(c),(count)) : \`
370			`__memset_g((s),(c),(count)))`
371
372			`#define memset(s, c, count) \`
373			`(__builtin_constant_p(count) ? \`
374			`__memset_const((s),(c),(count)) : \`
375			`__memset_g((s),(c),(count)))`
376
377			`#define __HAVE_ARCH_MEMCPY`
378			`extern void * memcpy(void , const void , size_t );`
379			`/*`
380			`* __builtin_memcpy() does not handle page-sized memcpys very well,`
381			`* thus following the same assumptions as for page-sized memsets, this`
382			`* function copies page-sized areas using an unrolled loop, without`
383			`* considering alignment.`
384			`*`
385			`* For the 680[46]0 only kernels we use the move16 instruction instead`
386			`* as it writes through the data-cache, invalidating the cache-lines`
387			`* touched. In this way we do not use up the entire data-cache (well,`
388			`* half of it on the 68060) by copying a page. An unrolled loop of two`
389			`* move16 instructions seem to the fastest. The only caveat is that`
390			`* both source and destination must be 16-byte aligned, if not we fall`
391			`* back to the generic memcpy function. - Jes`
392			`*/`
393			`static inline void * __memcpy_page(void * to, const void * from, size_t count)`
394			`{`
395			`unsigned long tmp;`
396			`void *xto = to;`
397
398			`#ifdef CPU_M68040_OR_M68060_ONLY`
399
400			`if (((unsigned long) to \| (unsigned long) from) & 0x0f)`
401			`return memcpy(to, from, count);`
402
403			`__asm__ __volatile__("1:\t"`
404			`".chip 68040\n\t"`
405			`"move16 %1@+,%0@+\n\t"`
406			`"move16 %1@+,%0@+\n\t"`
407			`".chip 68k\n\t"`
408			`"dbra %2,1b\n\t"`
409			`: "=a" (to), "=a" (from), "=d" (tmp)`
410			`: "0" (to), "1" (from) , "2" (count / 32 - 1)`
411			`);`
412			`#else`
413			`__asm__ __volatile__("1:\t"`
414			`"movel %1@+,%0@+\n\t"`
415			`"movel %1@+,%0@+\n\t"`
416			`"movel %1@+,%0@+\n\t"`
417			`"movel %1@+,%0@+\n\t"`
418			`"movel %1@+,%0@+\n\t"`
419			`"movel %1@+,%0@+\n\t"`
420			`"movel %1@+,%0@+\n\t"`
421			`"movel %1@+,%0@+\n\t"`
422			`"dbra %2,1b\n\t"`
423			`: "=a" (to), "=a" (from), "=d" (tmp)`
424			`: "0" (to), "1" (from) , "2" (count / 32 - 1)`
425			`);`
426			`#endif`
427			`return xto;`
428			`}`
429
430			`#define __memcpy_const(to, from, n) \`
431			`((n==PAGE_SIZE) ? \`
432			`__memcpy_page((to),(from),(n)) : \`
433			`__builtin_memcpy((to),(from),(n)))`
434
435			`#define memcpy(to, from, n) \`
436			`(__builtin_constant_p(n) ? \`
437			`__memcpy_const((to),(from),(n)) : \`
438			`memcpy((to),(from),(n)))`
439
440			`#define __HAVE_ARCH_MEMMOVE`
441			`static inline void * memmove(void * dest,const void * src, size_t n)`
442			`{`
443			`void *xdest = dest;`
444			`size_t temp;`
445
446			`if (!n)`
447			`return xdest;`
448
449			`if (dest < src)`
450			`{`
451			`if ((long) dest & 1)`
452			`{`
453			`char *cdest = dest;`
454			`const char *csrc = src;`
455			`cdest++ = csrc++;`
456			`dest = cdest;`
457			`src = csrc;`
458			`n--;`
459			`}`
460			`if (n > 2 && (long) dest & 2)`
461			`{`
462			`short *sdest = dest;`
463			`const short *ssrc = src;`
464			`sdest++ = ssrc++;`
465			`dest = sdest;`
466			`src = ssrc;`
467			`n -= 2;`
468			`}`
469			`temp = n >> 2;`
470			`if (temp)`
471			`{`
472			`long *ldest = dest;`
473			`const long *lsrc = src;`
474			`temp--;`
475			`do`
476			`ldest++ = lsrc++;`
477			`while (temp--);`
478			`dest = ldest;`
479			`src = lsrc;`
480			`}`
481			`if (n & 2)`
482			`{`
483			`short *sdest = dest;`
484			`const short *ssrc = src;`
485			`sdest++ = ssrc++;`
486			`dest = sdest;`
487			`src = ssrc;`
488			`}`
489			`if (n & 1)`
490			`{`
491			`char *cdest = dest;`
492			`const char *csrc = src;`
493			`cdest = csrc;`
494			`}`
495			`}`
496			`else`
497			`{`
498			`dest = (char *) dest + n;`
499			`src = (const char *) src + n;`
500			`if ((long) dest & 1)`
501			`{`
502			`char *cdest = dest;`
503			`const char *csrc = src;`
504			`--cdest = --csrc;`
505			`dest = cdest;`
506			`src = csrc;`
507			`n--;`
508			`}`
509			`if (n > 2 && (long) dest & 2)`
510			`{`
511			`short *sdest = dest;`
512			`const short *ssrc = src;`
513			`--sdest = --ssrc;`
514			`dest = sdest;`
515			`src = ssrc;`
516			`n -= 2;`
517			`}`
518			`temp = n >> 2;`
519			`if (temp)`
520			`{`
521			`long *ldest = dest;`
522			`const long *lsrc = src;`
523			`temp--;`
524			`do`
525			`--ldest = --lsrc;`
526			`while (temp--);`
527			`dest = ldest;`
528			`src = lsrc;`
529			`}`
530			`if (n & 2)`
531			`{`
532			`short *sdest = dest;`
533			`const short *ssrc = src;`
534			`--sdest = --ssrc;`
535			`dest = sdest;`
536			`src = ssrc;`
537			`}`
538			`if (n & 1)`
539			`{`
540			`char *cdest = dest;`
541			`const char *csrc = src;`
542			`--cdest = --csrc;`
543			`}`
544			`}`
545			`return xdest;`
546			`}`
547
548			`#define __HAVE_ARCH_MEMCMP`
549			`extern int memcmp(const void * ,const void * ,size_t );`
550			`#define memcmp(cs, ct, n) \`
551			`(__builtin_constant_p(n) ? \`
552			`__builtin_memcmp((cs),(ct),(n)) : \`
553			`memcmp((cs),(ct),(n)))`
554
555			`#define __HAVE_ARCH_MEMCHR`
556			`static inline void memchr(const void cs, int c, size_t count)`
557			`{`
558			`/* Someone else can optimize this, I don't care - tonym@mac.linux-m68k.org */`
559			`unsigned char ret = (unsigned char )cs;`
560			`for(;count>0;count--,ret++)`
561			`if(*ret == c) return ret;`
562
563			`return NULL;`
564			`}`
565
566			`#endif /* _M68K_STRING_H_ */`