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

Subversion Repositories or1k

[/] [or1k/] [trunk/] [rc203soc/] [sw/] [uClinux/] [include/] [asm-m68knommu/] [string.h] - Blame information for rev 1777

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


#ifndef _M68K_STRING_H_
#define _M68K_STRING_H_
 
#include <linux/config.h>
#include <asm/page.h>
 
#if 0
 
#define __HAVE_ARCH_STRCPY
extern 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
extern 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
extern inline char * strcat(char * dest, const char * src)
{
        char *tmp = dest;
 
        while (*dest)
                dest++;
        while ((*dest++ = *src++))
                ;
 
        return tmp;
}
 
#define __HAVE_ARCH_STRNCAT
extern 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
extern inline char * strchr(const char * s, int c)
{
  const char ch = c;
 
  for(; *s != ch; ++s)
    if (*s == '\0')
      return( NULL );
  return( (char *) s);
}
 
#define __HAVE_ARCH_STRPBRK
extern 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 );
}
 
#define __HAVE_ARCH_STRSPN
extern 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;
}
 
#define __HAVE_ARCH_STRTOK
extern 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);
}
 
/* strstr !! */
 
#define __HAVE_ARCH_STRLEN
extern inline size_t strlen(const char * s)
{
  const char *sc;
  for (sc = s; *sc != '\0'; ++sc) ;
  return(sc - s);
}
 
/* strnlen !! */
 
#define __HAVE_ARCH_STRCMP
extern 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
extern 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
 */
extern 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
 */
extern 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;
 
#if defined(CONFIG_OPTIMIZE_040) || defined(CONFIG_OPTIMIZE_060)
 
  if (((unsigned long) s) & 0x0f)
          memset(s, c, count);
  else{
          *((unsigned long *)(s))++ = data;
          *((unsigned long *)(s))++ = data;
          *((unsigned long *)(s))++ = data;
          *((unsigned long *)(s))++ = data;
 
          __asm__ __volatile__("1:\t"
                               "move16 %2@+,%0@+\n\t"
                               "subqw  #8,%2\n\t"
                               "subqw  #8,%2\n\t"
                               "subql  #1,%1\n\t"
                               "bne    1b\n\t"
                               : "=a" (s), "=d" (tmp)
                               : "a" (sp), "0" (s), "1" ((count - 16) / 16)
                               );
  }
 
#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"
                       "subql #1,%1\n\t"
                       "bne   1b\n\t"
                       : "=a" (s), "=d" (tmp)
                       : "d" (data), "0" (s), "1" (count / 32)
                       );
#endif
 
  return xs;
}
 
#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((s),(c),(count)))
 
#define __HAVE_ARCH_MEMCPY
/*
 * __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
 */
extern inline void * __memcpy_page(void * to, const void * from, size_t count)
{
  unsigned long tmp;
  void *xto = to;
 
#if defined(CONFIG_OPTIMIZE_040) || defined(CONFIG_OPTIMIZE_060)
 
  if (((unsigned long) to | (unsigned long) from) & 0x0f)
          return memcpy(to, from, count);
 
  __asm__ __volatile__("1:\t"
                       "move16 %1@+,%0@+\n\t"
                       "move16 %1@+,%0@+\n\t"
                       "subql  #1,%2\n\t"
                       "bne    1b\n\t"
                       : "=a" (to), "=a" (from), "=d" (tmp)
                       : "0" (to), "1" (from) , "2" (count / 32)
                       );
#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"
                       "subql #1,%2\n\t"
                       "bne   1b\n\t"
                       : "=a" (to), "=a" (from), "=d" (tmp)
                       : "0" (to), "1" (from) , "2" (count / 32)
                       );
#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
extern 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
#define memcmp(cs, ct, n) \
(__builtin_constant_p(n) ? \
 __builtin_memcmp((cs),(ct),(n)) : \
 memcmp((cs),(ct),(n)))
 
#endif
 
#endif /* _M68K_STRING_H_ */

Browse

Tools

Subversion Repositories or1k

[/] [or1k/] [trunk/] [rc203soc/] [sw/] [uClinux/] [include/] [asm-m68knommu/] [string.h] - Blame information for rev 1777

Line No.	Rev	Author	Line
1	1633	jcastillo	`#ifndef _M68K_STRING_H_`
2			`#define _M68K_STRING_H_`
3
4			`#include <linux/config.h>`
5			`#include <asm/page.h>`
6
7			`#if 0`
8
9			`#define __HAVE_ARCH_STRCPY`
10			`extern inline char * strcpy(char * dest,const char *src)`
11			`{`
12			`char *xdest = dest;`
13
14			`__asm__ __volatile__`
15			`("1:\tmoveb %1@+,%0@+\n\t"`
16			`"jne 1b"`
17			`: "=a" (dest), "=a" (src)`
18			`: "0" (dest), "1" (src) : "memory");`
19			`return xdest;`
20			`}`
21
22			`#define __HAVE_ARCH_STRNCPY`
23			`extern inline char * strncpy(char dest, const char src, size_t n)`
24			`{`
25			`char *xdest = dest;`
26
27			`if (n == 0)`
28			`return xdest;`
29
30			`__asm__ __volatile__`
31			`("1:\tmoveb %1@+,%0@+\n\t"`
32			`"jeq 2f\n\t"`
33			`"subql #1,%2\n\t"`
34			`"jne 1b\n\t"`
35			`"2:"`
36			`: "=a" (dest), "=a" (src), "=d" (n)`
37			`: "0" (dest), "1" (src), "2" (n)`
38			`: "memory");`
39			`return xdest;`
40			`}`
41
42			`#define __HAVE_ARCH_STRCAT`
43			`extern inline char * strcat(char * dest, const char * src)`
44			`{`
45			`char *tmp = dest;`
46
47			`while (*dest)`
48			`dest++;`
49			`while ((dest++ = src++))`
50			`;`
51
52			`return tmp;`
53			`}`
54
55			`#define __HAVE_ARCH_STRNCAT`
56			`extern inline char * strncat(char dest, const char src, size_t count)`
57			`{`
58			`char *tmp = dest;`
59
60			`if (count) {`
61			`while (*dest)`
62			`dest++;`
63			`while ((dest++ = src++)) {`
64			`if (--count == 0) {`
65			`*dest++='\0';`
66			`break;`
67			`}`
68			`}`
69			`}`
70
71			`return tmp;`
72			`}`
73
74			`#define __HAVE_ARCH_STRCHR`
75			`extern inline char * strchr(const char * s, int c)`
76			`{`
77			`const char ch = c;`
78
79			`for(; *s != ch; ++s)`
80			`if (*s == '\0')`
81			`return( NULL );`
82			`return( (char *) s);`
83			`}`
84
85			`#define __HAVE_ARCH_STRPBRK`
86			`extern inline char * strpbrk(const char * cs,const char * ct)`
87			`{`
88			`const char sc1,sc2;`
89
90			`for( sc1 = cs; *sc1 != '\0'; ++sc1)`
91			`for( sc2 = ct; *sc2 != '\0'; ++sc2)`
92			`if (sc1 == sc2)`
93			`return((char *) sc1);`
94			`return( NULL );`
95			`}`
96
97			`#define __HAVE_ARCH_STRSPN`
98			`extern inline size_t strspn(const char s, const char accept)`
99			`{`
100			`const char *p;`
101			`const char *a;`
102			`size_t count = 0;`
103
104			`for (p = s; *p != '\0'; ++p)`
105			`{`
106			`for (a = accept; *a != '\0'; ++a)`
107			`if (p == a)`
108			`break;`
109			`if (*a == '\0')`
110			`return count;`
111			`else`
112			`++count;`
113			`}`
114
115			`return count;`
116			`}`
117
118			`#define __HAVE_ARCH_STRTOK`
119			`extern inline char * strtok(char * s,const char * ct)`
120			`{`
121			`char sbegin, send;`
122
123			`sbegin = s ? s : ___strtok;`
124			`if (!sbegin) {`
125			`return NULL;`
126			`}`
127			`sbegin += strspn(sbegin,ct);`
128			`if (*sbegin == '\0') {`
129			`___strtok = NULL;`
130			`return( NULL );`
131			`}`
132			`send = strpbrk( sbegin, ct);`
133			`if (send && *send != '\0')`
134			`*send++ = '\0';`
135			`___strtok = send;`
136			`return (sbegin);`
137			`}`
138
139			`/* strstr !! */`
140
141			`#define __HAVE_ARCH_STRLEN`
142			`extern inline size_t strlen(const char * s)`
143			`{`
144			`const char *sc;`
145			`for (sc = s; *sc != '\0'; ++sc) ;`
146			`return(sc - s);`
147			`}`
148
149			`/* strnlen !! */`
150
151			`#define __HAVE_ARCH_STRCMP`
152			`extern inline int strcmp(const char * cs,const char * ct)`
153			`{`
154			`char __res;`
155
156			`__asm__`
157			`("1:\tmoveb %0@+,%2\n\t" /* get cs /`
158			`"cmpb %1@+,%2\n\t" /* compare a byte */`
159			`"jne 2f\n\t" /* not equal, break out */`
160			`"tstb %2\n\t" /* at end of cs? */`
161			`"jne 1b\n\t" /* no, keep going */`
162			`"jra 3f\n\t" /* strings are equal */`
163			`"2:\tsubb %1@-,%2\n\t" /* cs - ct */`
164			`"3:"`
165			`: "=a" (cs), "=a" (ct), "=d" (__res)`
166			`: "0" (cs), "1" (ct));`
167			`return __res;`
168			`}`
169
170			`#define __HAVE_ARCH_STRNCMP`
171			`extern inline int strncmp(const char * cs,const char * ct,size_t count)`
172			`{`
173			`char __res;`
174
175			`if (!count)`
176			`return 0;`
177			`__asm__`
178			`("1:\tmovb %0@+,%3\n\t" /* get cs /`
179			`"cmpb %1@+,%3\n\t" /* compare a byte */`
180			`"jne 3f\n\t" /* not equal, break out */`
181			`"tstb %3\n\t" /* at end of cs? */`
182			`"jeq 4f\n\t" /* yes, all done */`
183			`"subql #1,%2\n\t" /* no, adjust count */`
184			`"jne 1b\n\t" /* more to do, keep going */`
185			`"2:\tmoveq #0,%3\n\t" /* strings are equal */`
186			`"jra 4f\n\t"`
187			`"3:\tsubb %1@-,%3\n\t" /* cs - ct */`
188			`"4:"`
189			`: "=a" (cs), "=a" (ct), "=d" (count), "=d" (__res)`
190			`: "0" (cs), "1" (ct), "2" (count));`
191			`return __res;`
192			`}`
193
194			`#define __HAVE_ARCH_MEMSET`
195			`/*`
196			`* This is really ugly, but its highly optimizatiable by the`
197			`* compiler and is meant as compensation for gcc's missing`
198			`* __builtin_memset(). For the 680[23]0 it might be worth considering`
199			`* the optimal number of misaligned writes compared to the number of`
200			`* tests'n'branches needed to align the destination address. The`
201			`* 680[46]0 doesn't really care due to their copy-back caches.`
202			`* 10/09/96 - Jes Sorensen`
203			`*/`
204			`extern inline void * __memset_g(void * s, int c, size_t count)`
205			`{`
206			`void *xs = s;`
207			`size_t temp;`
208
209			`if (!count)`
210			`return xs;`
211
212			`c &= 0xff;`
213			`c \|= c << 8;`
214			`c \|= c << 16;`
215
216			`if (count < 36){`
217			`long *ls = s;`
218
219			`switch(count){`
220			`case 32: case 33: case 34: case 35:`
221			`*ls++ = c;`
222			`case 28: case 29: case 30: case 31:`
223			`*ls++ = c;`
224			`case 24: case 25: case 26: case 27:`
225			`*ls++ = c;`
226			`case 20: case 21: case 22: case 23:`
227			`*ls++ = c;`
228			`case 16: case 17: case 18: case 19:`
229			`*ls++ = c;`
230			`case 12: case 13: case 14: case 15:`
231			`*ls++ = c;`
232			`case 8: case 9: case 10: case 11:`
233			`*ls++ = c;`
234			`case 4: case 5: case 6: case 7:`
235			`*ls++ = c;`
236			`break;`
237			`default:`
238			`break;`
239			`}`
240			`s = ls;`
241			`if (count & 0x02){`
242			`short *ss = s;`
243			`*ss++ = c;`
244			`s = ss;`
245			`}`
246			`if (count & 0x01){`
247			`char *cs = s;`
248			`*cs++ = c;`
249			`s = cs;`
250			`}`
251			`return xs;`
252			`}`
253
254			`if ((long) s & 1)`
255			`{`
256			`char *cs = s;`
257			`*cs++ = c;`
258			`s = cs;`
259			`count--;`
260			`}`
261			`if (count > 2 && (long) s & 2)`
262			`{`
263			`short *ss = s;`
264			`*ss++ = c;`
265			`s = ss;`
266			`count -= 2;`
267			`}`
268			`temp = count >> 2;`
269			`if (temp)`
270			`{`
271			`long *ls = s;`
272			`temp--;`
273			`do`
274			`*ls++ = c;`
275			`while (temp--);`
276			`s = ls;`
277			`}`
278			`if (count & 2)`
279			`{`
280			`short *ss = s;`
281			`*ss++ = c;`
282			`s = ss;`
283			`}`
284			`if (count & 1)`
285			`{`
286			`char *cs = s;`
287			`*cs = c;`
288			`}`
289			`return xs;`
290			`}`
291
292			`/*`
293			`* __memset_page assumes that data is longword aligned. Most, if not`
294			`* all, of these page sized memsets are performed on page aligned`
295			`* areas, thus we do not need to check if the destination is longword`
296			`* aligned. Of course we suffer a serious performance loss if this is`
297			`* not the case but I think the risk of this ever happening is`
298			`* extremely small. We spend a lot of time clearing pages in`
299			`* get_empty_page() so I think it is worth it anyway. Besides, the`
300			`* 680[46]0 do not really care about misaligned writes due to their`
301			`* copy-back cache.`
302			`*`
303			`* The optimized case for the 680[46]0 is implemented using the move16`
304			`* instruction. My tests showed that this implementation is 35-45%`
305			`* faster than the original implementation using movel, the only`
306			`* caveat is that the destination address must be 16-byte aligned.`
307			`* 01/09/96 - Jes Sorensen`
308			`*/`
309			`extern inline void * __memset_page(void * s,int c,size_t count)`
310			`{`
311			`unsigned long data, tmp;`
312			`void xs, sp;`
313
314			`xs = sp = s;`
315
316			`c = c & 255;`
317			`data = c \| (c << 8);`
318			`data \|= data << 16;`
319
320			`#if defined(CONFIG_OPTIMIZE_040) \|\| defined(CONFIG_OPTIMIZE_060)`
321
322			`if (((unsigned long) s) & 0x0f)`
323			`memset(s, c, count);`
324			`else{`
325			`((unsigned long )(s))++ = data;`
326			`((unsigned long )(s))++ = data;`
327			`((unsigned long )(s))++ = data;`
328			`((unsigned long )(s))++ = data;`
329
330			`__asm__ __volatile__("1:\t"`
331			`"move16 %2@+,%0@+\n\t"`
332			`"subqw #8,%2\n\t"`
333			`"subqw #8,%2\n\t"`
334			`"subql #1,%1\n\t"`
335			`"bne 1b\n\t"`
336			`: "=a" (s), "=d" (tmp)`
337			`: "a" (sp), "0" (s), "1" ((count - 16) / 16)`
338			`);`
339			`}`
340
341			`#else`
342			`__asm__ __volatile__("1:\t"`
343			`"movel %2,%0@+\n\t"`
344			`"movel %2,%0@+\n\t"`
345			`"movel %2,%0@+\n\t"`
346			`"movel %2,%0@+\n\t"`
347			`"movel %2,%0@+\n\t"`
348			`"movel %2,%0@+\n\t"`
349			`"movel %2,%0@+\n\t"`
350			`"movel %2,%0@+\n\t"`
351			`"subql #1,%1\n\t"`
352			`"bne 1b\n\t"`
353			`: "=a" (s), "=d" (tmp)`
354			`: "d" (data), "0" (s), "1" (count / 32)`
355			`);`
356			`#endif`
357
358			`return xs;`
359			`}`
360
361			`#define __memset_const(s,c,count) \`
362			`((count==PAGE_SIZE) ? \`
363			`__memset_page((s),(c),(count)) : \`
364			`__memset_g((s),(c),(count)))`
365
366			`#define memset(s, c, count) \`
367			`(__builtin_constant_p(count) ? \`
368			`__memset_const((s),(c),(count)) : \`
369			`memset((s),(c),(count)))`
370
371			`#define __HAVE_ARCH_MEMCPY`
372			`/*`
373			`* __builtin_memcpy() does not handle page-sized memcpys very well,`
374			`* thus following the same assumptions as for page-sized memsets, this`
375			`* function copies page-sized areas using an unrolled loop, without`
376			`* considering alignment.`
377			`*`
378			`* For the 680[46]0 only kernels we use the move16 instruction instead`
379			`* as it writes through the data-cache, invalidating the cache-lines`
380			`* touched. In this way we do not use up the entire data-cache (well,`
381			`* half of it on the 68060) by copying a page. An unrolled loop of two`
382			`* move16 instructions seem to the fastest. The only caveat is that`
383			`* both source and destination must be 16-byte aligned, if not we fall`
384			`* back to the generic memcpy function. - Jes`
385			`*/`
386			`extern inline void * __memcpy_page(void * to, const void * from, size_t count)`
387			`{`
388			`unsigned long tmp;`
389			`void *xto = to;`
390
391			`#if defined(CONFIG_OPTIMIZE_040) \|\| defined(CONFIG_OPTIMIZE_060)`
392
393			`if (((unsigned long) to \| (unsigned long) from) & 0x0f)`
394			`return memcpy(to, from, count);`
395
396			`__asm__ __volatile__("1:\t"`
397			`"move16 %1@+,%0@+\n\t"`
398			`"move16 %1@+,%0@+\n\t"`
399			`"subql #1,%2\n\t"`
400			`"bne 1b\n\t"`
401			`: "=a" (to), "=a" (from), "=d" (tmp)`
402			`: "0" (to), "1" (from) , "2" (count / 32)`
403			`);`
404			`#else`
405			`__asm__ __volatile__("1:\t"`
406			`"movel %1@+,%0@+\n\t"`
407			`"movel %1@+,%0@+\n\t"`
408			`"movel %1@+,%0@+\n\t"`
409			`"movel %1@+,%0@+\n\t"`
410			`"movel %1@+,%0@+\n\t"`
411			`"movel %1@+,%0@+\n\t"`
412			`"movel %1@+,%0@+\n\t"`
413			`"movel %1@+,%0@+\n\t"`
414			`"subql #1,%2\n\t"`
415			`"bne 1b\n\t"`
416			`: "=a" (to), "=a" (from), "=d" (tmp)`
417			`: "0" (to), "1" (from) , "2" (count / 32)`
418			`);`
419			`#endif`
420			`return xto;`
421			`}`
422
423			`#define __memcpy_const(to, from, n) \`
424			`((n==PAGE_SIZE) ? \`
425			`__memcpy_page((to),(from),(n)) : \`
426			`__builtin_memcpy((to),(from),(n)))`
427
428			`#define memcpy(to, from, n) \`
429			`(__builtin_constant_p(n) ? \`
430			`__memcpy_const((to),(from),(n)) : \`
431			`memcpy((to),(from),(n)))`
432
433			`#define __HAVE_ARCH_MEMMOVE`
434			`extern inline void * memmove(void * dest,const void * src, size_t n)`
435			`{`
436			`void *xdest = dest;`
437			`size_t temp;`
438
439			`if (!n)`
440			`return xdest;`
441
442			`if (dest < src)`
443			`{`
444			`if ((long) dest & 1)`
445			`{`
446			`char *cdest = dest;`
447			`const char *csrc = src;`
448			`cdest++ = csrc++;`
449			`dest = cdest;`
450			`src = csrc;`
451			`n--;`
452			`}`
453			`if (n > 2 && (long) dest & 2)`
454			`{`
455			`short *sdest = dest;`
456			`const short *ssrc = src;`
457			`sdest++ = ssrc++;`
458			`dest = sdest;`
459			`src = ssrc;`
460			`n -= 2;`
461			`}`
462			`temp = n >> 2;`
463			`if (temp)`
464			`{`
465			`long *ldest = dest;`
466			`const long *lsrc = src;`
467			`temp--;`
468			`do`
469			`ldest++ = lsrc++;`
470			`while (temp--);`
471			`dest = ldest;`
472			`src = lsrc;`
473			`}`
474			`if (n & 2)`
475			`{`
476			`short *sdest = dest;`
477			`const short *ssrc = src;`
478			`sdest++ = ssrc++;`
479			`dest = sdest;`
480			`src = ssrc;`
481			`}`
482			`if (n & 1)`
483			`{`
484			`char *cdest = dest;`
485			`const char *csrc = src;`
486			`cdest = csrc;`
487			`}`
488			`}`
489			`else`
490			`{`
491			`dest = (char *) dest + n;`
492			`src = (const char *) src + n;`
493			`if ((long) dest & 1)`
494			`{`
495			`char *cdest = dest;`
496			`const char *csrc = src;`
497			`--cdest = --csrc;`
498			`dest = cdest;`
499			`src = csrc;`
500			`n--;`
501			`}`
502			`if (n > 2 && (long) dest & 2)`
503			`{`
504			`short *sdest = dest;`
505			`const short *ssrc = src;`
506			`--sdest = --ssrc;`
507			`dest = sdest;`
508			`src = ssrc;`
509			`n -= 2;`
510			`}`
511			`temp = n >> 2;`
512			`if (temp)`
513			`{`
514			`long *ldest = dest;`
515			`const long *lsrc = src;`
516			`temp--;`
517			`do`
518			`--ldest = --lsrc;`
519			`while (temp--);`
520			`dest = ldest;`
521			`src = lsrc;`
522			`}`
523			`if (n & 2)`
524			`{`
525			`short *sdest = dest;`
526			`const short *ssrc = src;`
527			`--sdest = --ssrc;`
528			`dest = sdest;`
529			`src = ssrc;`
530			`}`
531			`if (n & 1)`
532			`{`
533			`char *cdest = dest;`
534			`const char *csrc = src;`
535			`--cdest = --csrc;`
536			`}`
537			`}`
538			`return xdest;`
539			`}`
540
541			`#define __HAVE_ARCH_MEMCMP`
542			`#define memcmp(cs, ct, n) \`
543			`(__builtin_constant_p(n) ? \`
544			`__builtin_memcmp((cs),(ct),(n)) : \`
545			`memcmp((cs),(ct),(n)))`
546
547			`#endif`
548
549			`#endif /* _M68K_STRING_H_ */`