[PATCH] s390: spin lock retry
[linux/fpc-iii.git] / include / asm-m68k / string.h
blob44def078132a756ec2790e7073d0f33eedf7407a
1 #ifndef _M68K_STRING_H_
2 #define _M68K_STRING_H_
4 #include <asm/setup.h>
5 #include <asm/page.h>
7 #define __HAVE_ARCH_STRCPY
8 static inline char * strcpy(char * dest,const char *src)
10 char *xdest = dest;
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;
20 #define __HAVE_ARCH_STRNCPY
21 static inline char * strncpy(char *dest, const char *src, size_t n)
23 char *xdest = dest;
25 if (n == 0)
26 return xdest;
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;
40 #define __HAVE_ARCH_STRCAT
41 static inline char * strcat(char * dest, const char * src)
43 char *tmp = dest;
45 while (*dest)
46 dest++;
47 while ((*dest++ = *src++))
50 return tmp;
53 #define __HAVE_ARCH_STRNCAT
54 static inline char * strncat(char *dest, const char *src, size_t count)
56 char *tmp = dest;
58 if (count) {
59 while (*dest)
60 dest++;
61 while ((*dest++ = *src++)) {
62 if (--count == 0) {
63 *dest++='\0';
64 break;
69 return tmp;
72 #define __HAVE_ARCH_STRCHR
73 static inline char * strchr(const char * s, int c)
75 const char ch = c;
77 for(; *s != ch; ++s)
78 if (*s == '\0')
79 return( NULL );
80 return( (char *) s);
83 #if 0
84 #define __HAVE_ARCH_STRPBRK
85 static inline char *strpbrk(const char *cs,const char *ct)
87 const char *sc1,*sc2;
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 );
95 #endif
97 #if 0
98 #define __HAVE_ARCH_STRSPN
99 static inline size_t strspn(const char *s, const char *accept)
101 const char *p;
102 const char *a;
103 size_t count = 0;
105 for (p = s; *p != '\0'; ++p)
107 for (a = accept; *a != '\0'; ++a)
108 if (*p == *a)
109 break;
110 if (*a == '\0')
111 return count;
112 else
113 ++count;
116 return count;
118 #endif
120 /* strstr !! */
122 #define __HAVE_ARCH_STRLEN
123 static inline size_t strlen(const char * s)
125 const char *sc;
126 for (sc = s; *sc != '\0'; ++sc) ;
127 return(sc - s);
130 /* strnlen !! */
132 #define __HAVE_ARCH_STRCMP
133 static inline int strcmp(const char * cs,const char * ct)
135 char __res;
137 __asm__
138 ("1:\tmoveb %0@+,%2\n\t" /* get *cs */
139 "cmpb %1@+,%2\n\t" /* compare a byte */
140 "jne 2f\n\t" /* not equal, break out */
141 "tstb %2\n\t" /* at end of cs? */
142 "jne 1b\n\t" /* no, keep going */
143 "jra 3f\n\t" /* strings are equal */
144 "2:\tsubb %1@-,%2\n\t" /* *cs - *ct */
145 "3:"
146 : "=a" (cs), "=a" (ct), "=d" (__res)
147 : "0" (cs), "1" (ct));
148 return __res;
151 #define __HAVE_ARCH_STRNCMP
152 static inline int strncmp(const char * cs,const char * ct,size_t count)
154 char __res;
156 if (!count)
157 return 0;
158 __asm__
159 ("1:\tmovb %0@+,%3\n\t" /* get *cs */
160 "cmpb %1@+,%3\n\t" /* compare a byte */
161 "jne 3f\n\t" /* not equal, break out */
162 "tstb %3\n\t" /* at end of cs? */
163 "jeq 4f\n\t" /* yes, all done */
164 "subql #1,%2\n\t" /* no, adjust count */
165 "jne 1b\n\t" /* more to do, keep going */
166 "2:\tmoveq #0,%3\n\t" /* strings are equal */
167 "jra 4f\n\t"
168 "3:\tsubb %1@-,%3\n\t" /* *cs - *ct */
169 "4:"
170 : "=a" (cs), "=a" (ct), "=d" (count), "=d" (__res)
171 : "0" (cs), "1" (ct), "2" (count));
172 return __res;
175 #define __HAVE_ARCH_MEMSET
177 * This is really ugly, but its highly optimizatiable by the
178 * compiler and is meant as compensation for gcc's missing
179 * __builtin_memset(). For the 680[23]0 it might be worth considering
180 * the optimal number of misaligned writes compared to the number of
181 * tests'n'branches needed to align the destination address. The
182 * 680[46]0 doesn't really care due to their copy-back caches.
183 * 10/09/96 - Jes Sorensen
185 static inline void * __memset_g(void * s, int c, size_t count)
187 void *xs = s;
188 size_t temp;
190 if (!count)
191 return xs;
193 c &= 0xff;
194 c |= c << 8;
195 c |= c << 16;
197 if (count < 36){
198 long *ls = s;
200 switch(count){
201 case 32: case 33: case 34: case 35:
202 *ls++ = c;
203 case 28: case 29: case 30: case 31:
204 *ls++ = c;
205 case 24: case 25: case 26: case 27:
206 *ls++ = c;
207 case 20: case 21: case 22: case 23:
208 *ls++ = c;
209 case 16: case 17: case 18: case 19:
210 *ls++ = c;
211 case 12: case 13: case 14: case 15:
212 *ls++ = c;
213 case 8: case 9: case 10: case 11:
214 *ls++ = c;
215 case 4: case 5: case 6: case 7:
216 *ls++ = c;
217 break;
218 default:
219 break;
221 s = ls;
222 if (count & 0x02){
223 short *ss = s;
224 *ss++ = c;
225 s = ss;
227 if (count & 0x01){
228 char *cs = s;
229 *cs++ = c;
230 s = cs;
232 return xs;
235 if ((long) s & 1)
237 char *cs = s;
238 *cs++ = c;
239 s = cs;
240 count--;
242 if (count > 2 && (long) s & 2)
244 short *ss = s;
245 *ss++ = c;
246 s = ss;
247 count -= 2;
249 temp = count >> 2;
250 if (temp)
252 long *ls = s;
253 temp--;
255 *ls++ = c;
256 while (temp--);
257 s = ls;
259 if (count & 2)
261 short *ss = s;
262 *ss++ = c;
263 s = ss;
265 if (count & 1)
267 char *cs = s;
268 *cs = c;
270 return xs;
274 * __memset_page assumes that data is longword aligned. Most, if not
275 * all, of these page sized memsets are performed on page aligned
276 * areas, thus we do not need to check if the destination is longword
277 * aligned. Of course we suffer a serious performance loss if this is
278 * not the case but I think the risk of this ever happening is
279 * extremely small. We spend a lot of time clearing pages in
280 * get_empty_page() so I think it is worth it anyway. Besides, the
281 * 680[46]0 do not really care about misaligned writes due to their
282 * copy-back cache.
284 * The optimized case for the 680[46]0 is implemented using the move16
285 * instruction. My tests showed that this implementation is 35-45%
286 * faster than the original implementation using movel, the only
287 * caveat is that the destination address must be 16-byte aligned.
288 * 01/09/96 - Jes Sorensen
290 static inline void * __memset_page(void * s,int c,size_t count)
292 unsigned long data, tmp;
293 void *xs = s;
295 c = c & 255;
296 data = c | (c << 8);
297 data |= data << 16;
299 #ifdef CPU_M68040_OR_M68060_ONLY
301 if (((unsigned long) s) & 0x0f)
302 __memset_g(s, c, count);
303 else{
304 unsigned long *sp = s;
305 *sp++ = data;
306 *sp++ = data;
307 *sp++ = data;
308 *sp++ = data;
310 __asm__ __volatile__("1:\t"
311 ".chip 68040\n\t"
312 "move16 %2@+,%0@+\n\t"
313 ".chip 68k\n\t"
314 "subqw #8,%2\n\t"
315 "subqw #8,%2\n\t"
316 "dbra %1,1b\n\t"
317 : "=a" (sp), "=d" (tmp)
318 : "a" (s), "0" (sp), "1" ((count - 16) / 16 - 1)
322 #else
323 __asm__ __volatile__("1:\t"
324 "movel %2,%0@+\n\t"
325 "movel %2,%0@+\n\t"
326 "movel %2,%0@+\n\t"
327 "movel %2,%0@+\n\t"
328 "movel %2,%0@+\n\t"
329 "movel %2,%0@+\n\t"
330 "movel %2,%0@+\n\t"
331 "movel %2,%0@+\n\t"
332 "dbra %1,1b\n\t"
333 : "=a" (s), "=d" (tmp)
334 : "d" (data), "0" (s), "1" (count / 32 - 1)
336 #endif
338 return xs;
341 extern void *memset(void *,int,__kernel_size_t);
343 #define __memset_const(s,c,count) \
344 ((count==PAGE_SIZE) ? \
345 __memset_page((s),(c),(count)) : \
346 __memset_g((s),(c),(count)))
348 #define memset(s, c, count) \
349 (__builtin_constant_p(count) ? \
350 __memset_const((s),(c),(count)) : \
351 __memset_g((s),(c),(count)))
353 #define __HAVE_ARCH_MEMCPY
354 extern void * memcpy(void *, const void *, size_t );
356 * __builtin_memcpy() does not handle page-sized memcpys very well,
357 * thus following the same assumptions as for page-sized memsets, this
358 * function copies page-sized areas using an unrolled loop, without
359 * considering alignment.
361 * For the 680[46]0 only kernels we use the move16 instruction instead
362 * as it writes through the data-cache, invalidating the cache-lines
363 * touched. In this way we do not use up the entire data-cache (well,
364 * half of it on the 68060) by copying a page. An unrolled loop of two
365 * move16 instructions seem to the fastest. The only caveat is that
366 * both source and destination must be 16-byte aligned, if not we fall
367 * back to the generic memcpy function. - Jes
369 static inline void * __memcpy_page(void * to, const void * from, size_t count)
371 unsigned long tmp;
372 void *xto = to;
374 #ifdef CPU_M68040_OR_M68060_ONLY
376 if (((unsigned long) to | (unsigned long) from) & 0x0f)
377 return memcpy(to, from, count);
379 __asm__ __volatile__("1:\t"
380 ".chip 68040\n\t"
381 "move16 %1@+,%0@+\n\t"
382 "move16 %1@+,%0@+\n\t"
383 ".chip 68k\n\t"
384 "dbra %2,1b\n\t"
385 : "=a" (to), "=a" (from), "=d" (tmp)
386 : "0" (to), "1" (from) , "2" (count / 32 - 1)
388 #else
389 __asm__ __volatile__("1:\t"
390 "movel %1@+,%0@+\n\t"
391 "movel %1@+,%0@+\n\t"
392 "movel %1@+,%0@+\n\t"
393 "movel %1@+,%0@+\n\t"
394 "movel %1@+,%0@+\n\t"
395 "movel %1@+,%0@+\n\t"
396 "movel %1@+,%0@+\n\t"
397 "movel %1@+,%0@+\n\t"
398 "dbra %2,1b\n\t"
399 : "=a" (to), "=a" (from), "=d" (tmp)
400 : "0" (to), "1" (from) , "2" (count / 32 - 1)
402 #endif
403 return xto;
406 #define __memcpy_const(to, from, n) \
407 ((n==PAGE_SIZE) ? \
408 __memcpy_page((to),(from),(n)) : \
409 __builtin_memcpy((to),(from),(n)))
411 #define memcpy(to, from, n) \
412 (__builtin_constant_p(n) ? \
413 __memcpy_const((to),(from),(n)) : \
414 memcpy((to),(from),(n)))
416 #define __HAVE_ARCH_MEMMOVE
417 static inline void * memmove(void * dest,const void * src, size_t n)
419 void *xdest = dest;
420 size_t temp;
422 if (!n)
423 return xdest;
425 if (dest < src)
427 if ((long) dest & 1)
429 char *cdest = dest;
430 const char *csrc = src;
431 *cdest++ = *csrc++;
432 dest = cdest;
433 src = csrc;
434 n--;
436 if (n > 2 && (long) dest & 2)
438 short *sdest = dest;
439 const short *ssrc = src;
440 *sdest++ = *ssrc++;
441 dest = sdest;
442 src = ssrc;
443 n -= 2;
445 temp = n >> 2;
446 if (temp)
448 long *ldest = dest;
449 const long *lsrc = src;
450 temp--;
452 *ldest++ = *lsrc++;
453 while (temp--);
454 dest = ldest;
455 src = lsrc;
457 if (n & 2)
459 short *sdest = dest;
460 const short *ssrc = src;
461 *sdest++ = *ssrc++;
462 dest = sdest;
463 src = ssrc;
465 if (n & 1)
467 char *cdest = dest;
468 const char *csrc = src;
469 *cdest = *csrc;
472 else
474 dest = (char *) dest + n;
475 src = (const char *) src + n;
476 if ((long) dest & 1)
478 char *cdest = dest;
479 const char *csrc = src;
480 *--cdest = *--csrc;
481 dest = cdest;
482 src = csrc;
483 n--;
485 if (n > 2 && (long) dest & 2)
487 short *sdest = dest;
488 const short *ssrc = src;
489 *--sdest = *--ssrc;
490 dest = sdest;
491 src = ssrc;
492 n -= 2;
494 temp = n >> 2;
495 if (temp)
497 long *ldest = dest;
498 const long *lsrc = src;
499 temp--;
501 *--ldest = *--lsrc;
502 while (temp--);
503 dest = ldest;
504 src = lsrc;
506 if (n & 2)
508 short *sdest = dest;
509 const short *ssrc = src;
510 *--sdest = *--ssrc;
511 dest = sdest;
512 src = ssrc;
514 if (n & 1)
516 char *cdest = dest;
517 const char *csrc = src;
518 *--cdest = *--csrc;
521 return xdest;
524 #define __HAVE_ARCH_MEMCMP
525 extern int memcmp(const void * ,const void * ,size_t );
526 #define memcmp(cs, ct, n) \
527 (__builtin_constant_p(n) ? \
528 __builtin_memcmp((cs),(ct),(n)) : \
529 memcmp((cs),(ct),(n)))
531 #define __HAVE_ARCH_MEMCHR
532 static inline void *memchr(const void *cs, int c, size_t count)
534 /* Someone else can optimize this, I don't care - tonym@mac.linux-m68k.org */
535 unsigned char *ret = (unsigned char *)cs;
536 for(;count>0;count--,ret++)
537 if(*ret == c) return ret;
539 return NULL;
542 #endif /* _M68K_STRING_H_ */