![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to string.h CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GnuMach] / Gnu-Mach / linux / dev / include / asm-i386|
Return to string.h CVS log | Up to [GnuMach] / Gnu-Mach / linux / dev / include / asm-i386 |
1.1 root 1: #ifndef _I386_STRING_H_
2: #define _I386_STRING_H_
3:
4: /*
5: * On a 486 or Pentium, we are better off not using the
6: * byte string operations. But on a 386 or a PPro the
7: * byte string ops are faster than doing it by hand
8: * (MUCH faster on a Pentium).
9: *
10: * Also, the byte strings actually work correctly. Forget
11: * the i486 routines for now as they may be broken..
12: */
13: #if FIXED_486_STRING && (CPU == 486 || CPU == 586)
14: #include <asm/string-486.h>
15: #else
16:
17: /*
18: * This string-include defines all string functions as inline
19: * functions. Use gcc. It also assumes ds=es=data space, this should be
20: * normal. Most of the string-functions are rather heavily hand-optimized,
21: * see especially strtok,strstr,str[c]spn. They should work, but are not
22: * very easy to understand. Everything is done entirely within the register
23: * set, making the functions fast and clean. String instructions have been
24: * used through-out, making for "slightly" unclear code :-)
25: *
1.1.1.2 ! root 26: * NO Copyright (C) 1991, 1992 Linus Torvalds,
! 27: * consider these trivial functions to be PD.
1.1 root 28: */
29:
30: #define __HAVE_ARCH_STRCPY
31: extern inline char * strcpy(char * dest,const char *src)
32: {
1.1.1.2 ! root 33: int d0, d1, d2;
1.1 root 34: __asm__ __volatile__(
35: "cld\n"
36: "1:\tlodsb\n\t"
37: "stosb\n\t"
38: "testb %%al,%%al\n\t"
39: "jne 1b"
1.1.1.2 ! root 40: : "=&S" (d0), "=&D" (d1), "=&a" (d2)
! 41: :"0" (src),"1" (dest) : "memory");
1.1 root 42: return dest;
43: }
44:
45: #define __HAVE_ARCH_STRNCPY
46: extern inline char * strncpy(char * dest,const char *src,size_t count)
47: {
1.1.1.2 ! root 48: int d0, d1, d2, d3;
1.1 root 49: __asm__ __volatile__(
50: "cld\n"
51: "1:\tdecl %2\n\t"
52: "js 2f\n\t"
53: "lodsb\n\t"
54: "stosb\n\t"
55: "testb %%al,%%al\n\t"
56: "jne 1b\n\t"
57: "rep\n\t"
58: "stosb\n"
59: "2:"
1.1.1.2 ! root 60: : "=&S" (d0), "=&D" (d1), "=&c" (d2), "=&a" (d3)
! 61: :"0" (src),"1" (dest),"2" (count) : "memory");
1.1 root 62: return dest;
63: }
64:
65: #define __HAVE_ARCH_STRCAT
66: extern inline char * strcat(char * dest,const char * src)
67: {
1.1.1.2 ! root 68: int d0, d1, d2, d3;
1.1 root 69: __asm__ __volatile__(
70: "cld\n\t"
71: "repne\n\t"
72: "scasb\n\t"
73: "decl %1\n"
74: "1:\tlodsb\n\t"
75: "stosb\n\t"
76: "testb %%al,%%al\n\t"
77: "jne 1b"
1.1.1.2 ! root 78: : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
! 79: : "0" (src), "1" (dest), "2" (0), "3" (0xffffffff):"memory");
1.1 root 80: return dest;
81: }
82:
83: #define __HAVE_ARCH_STRNCAT
84: extern inline char * strncat(char * dest,const char * src,size_t count)
85: {
1.1.1.2 ! root 86: int d0, d1, d2, d3;
1.1 root 87: __asm__ __volatile__(
88: "cld\n\t"
89: "repne\n\t"
90: "scasb\n\t"
91: "decl %1\n\t"
1.1.1.2 ! root 92: "movl %8,%3\n"
1.1 root 93: "1:\tdecl %3\n\t"
94: "js 2f\n\t"
95: "lodsb\n\t"
96: "stosb\n\t"
97: "testb %%al,%%al\n\t"
98: "jne 1b\n"
99: "2:\txorl %2,%2\n\t"
100: "stosb"
1.1.1.2 ! root 101: : "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
! 102: : "0" (src),"1" (dest),"2" (0),"3" (0xffffffff), "g" (count)
! 103: : "memory");
1.1 root 104: return dest;
105: }
106:
107: #define __HAVE_ARCH_STRCMP
108: extern inline int strcmp(const char * cs,const char * ct)
109: {
1.1.1.2 ! root 110: int d0, d1;
1.1 root 111: register int __res;
112: __asm__ __volatile__(
113: "cld\n"
114: "1:\tlodsb\n\t"
115: "scasb\n\t"
116: "jne 2f\n\t"
117: "testb %%al,%%al\n\t"
118: "jne 1b\n\t"
119: "xorl %%eax,%%eax\n\t"
120: "jmp 3f\n"
121: "2:\tsbbl %%eax,%%eax\n\t"
1.1.1.2 ! root 122: "orb $1,%%al\n"
1.1 root 123: "3:"
1.1.1.2 ! root 124: :"=a" (__res), "=&S" (d0), "=&D" (d1)
! 125: :"1" (cs),"2" (ct));
1.1 root 126: return __res;
127: }
128:
129: #define __HAVE_ARCH_STRNCMP
130: extern inline int strncmp(const char * cs,const char * ct,size_t count)
131: {
132: register int __res;
1.1.1.2 ! root 133: int d0, d1, d2;
1.1 root 134: __asm__ __volatile__(
135: "cld\n"
136: "1:\tdecl %3\n\t"
137: "js 2f\n\t"
138: "lodsb\n\t"
139: "scasb\n\t"
140: "jne 3f\n\t"
141: "testb %%al,%%al\n\t"
142: "jne 1b\n"
143: "2:\txorl %%eax,%%eax\n\t"
144: "jmp 4f\n"
145: "3:\tsbbl %%eax,%%eax\n\t"
146: "orb $1,%%al\n"
147: "4:"
1.1.1.2 ! root 148: :"=a" (__res), "=&S" (d0), "=&D" (d1), "=&c" (d2)
! 149: :"1" (cs),"2" (ct),"3" (count));
1.1 root 150: return __res;
151: }
152:
153: #define __HAVE_ARCH_STRCHR
154: extern inline char * strchr(const char * s, int c)
155: {
1.1.1.2 ! root 156: int d0;
1.1 root 157: register char * __res;
158: __asm__ __volatile__(
159: "cld\n\t"
160: "movb %%al,%%ah\n"
161: "1:\tlodsb\n\t"
162: "cmpb %%ah,%%al\n\t"
163: "je 2f\n\t"
164: "testb %%al,%%al\n\t"
165: "jne 1b\n\t"
166: "movl $1,%1\n"
167: "2:\tmovl %1,%0\n\t"
168: "decl %0"
1.1.1.2 ! root 169: :"=a" (__res), "=&S" (d0) : "1" (s),"0" (c));
1.1 root 170: return __res;
171: }
172:
173: #define __HAVE_ARCH_STRRCHR
174: extern inline char * strrchr(const char * s, int c)
175: {
1.1.1.2 ! root 176: int d0, d1;
1.1 root 177: register char * __res;
178: __asm__ __volatile__(
179: "cld\n\t"
180: "movb %%al,%%ah\n"
181: "1:\tlodsb\n\t"
182: "cmpb %%ah,%%al\n\t"
183: "jne 2f\n\t"
184: "leal -1(%%esi),%0\n"
185: "2:\ttestb %%al,%%al\n\t"
186: "jne 1b"
1.1.1.2 ! root 187: :"=g" (__res), "=&S" (d0), "=&a" (d1) :"0" (0),"1" (s),"2" (c));
1.1 root 188: return __res;
189: }
190:
191: #define __HAVE_ARCH_STRLEN
192: extern inline size_t strlen(const char * s)
193: {
1.1.1.2 ! root 194: int d0;
1.1 root 195: register int __res;
196: __asm__ __volatile__(
197: "cld\n\t"
198: "repne\n\t"
199: "scasb\n\t"
200: "notl %0\n\t"
201: "decl %0"
1.1.1.2 ! root 202: :"=c" (__res), "=&D" (d0) :"1" (s),"a" (0), "0" (0xffffffff));
1.1 root 203: return __res;
204: }
205:
206: extern inline void * __memcpy(void * to, const void * from, size_t n)
207: {
1.1.1.2 ! root 208: int d0, d1, d2;
1.1 root 209: __asm__ __volatile__(
210: "cld\n\t"
211: "rep ; movsl\n\t"
1.1.1.2 ! root 212: "testb $2,%b4\n\t"
1.1 root 213: "je 1f\n\t"
214: "movsw\n"
1.1.1.2 ! root 215: "1:\ttestb $1,%b4\n\t"
1.1 root 216: "je 2f\n\t"
217: "movsb\n"
218: "2:"
1.1.1.2 ! root 219: : "=&c" (d0), "=&D" (d1), "=&S" (d2)
! 220: :"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from)
! 221: : "memory");
1.1 root 222: return (to);
223: }
224:
225: /*
226: * This looks horribly ugly, but the compiler can optimize it totally,
227: * as the count is constant.
228: */
229: extern inline void * __constant_memcpy(void * to, const void * from, size_t n)
230: {
231: switch (n) {
232: case 0:
233: return to;
234: case 1:
235: *(unsigned char *)to = *(const unsigned char *)from;
236: return to;
237: case 2:
238: *(unsigned short *)to = *(const unsigned short *)from;
239: return to;
240: case 3:
241: *(unsigned short *)to = *(const unsigned short *)from;
242: *(2+(unsigned char *)to) = *(2+(const unsigned char *)from);
243: return to;
244: case 4:
245: *(unsigned long *)to = *(const unsigned long *)from;
246: return to;
1.1.1.2 ! root 247: case 6: /* for Ethernet addresses */
! 248: *(unsigned long *)to = *(const unsigned long *)from;
! 249: *(2+(unsigned short *)to) = *(2+(const unsigned short *)from);
! 250: return to;
1.1 root 251: case 8:
252: *(unsigned long *)to = *(const unsigned long *)from;
253: *(1+(unsigned long *)to) = *(1+(const unsigned long *)from);
254: return to;
255: case 12:
256: *(unsigned long *)to = *(const unsigned long *)from;
257: *(1+(unsigned long *)to) = *(1+(const unsigned long *)from);
258: *(2+(unsigned long *)to) = *(2+(const unsigned long *)from);
259: return to;
260: case 16:
261: *(unsigned long *)to = *(const unsigned long *)from;
262: *(1+(unsigned long *)to) = *(1+(const unsigned long *)from);
263: *(2+(unsigned long *)to) = *(2+(const unsigned long *)from);
264: *(3+(unsigned long *)to) = *(3+(const unsigned long *)from);
265: return to;
266: case 20:
267: *(unsigned long *)to = *(const unsigned long *)from;
268: *(1+(unsigned long *)to) = *(1+(const unsigned long *)from);
269: *(2+(unsigned long *)to) = *(2+(const unsigned long *)from);
270: *(3+(unsigned long *)to) = *(3+(const unsigned long *)from);
271: *(4+(unsigned long *)to) = *(4+(const unsigned long *)from);
272: return to;
273: }
274: #define COMMON(x) \
1.1.1.2 ! root 275: __asm__ __volatile__( \
! 276: "cld\n\t" \
1.1 root 277: "rep ; movsl" \
278: x \
1.1.1.2 ! root 279: : "=&c" (d0), "=&D" (d1), "=&S" (d2) \
! 280: : "0" (n/4),"1" ((long) to),"2" ((long) from) \
! 281: : "memory");
! 282: {
! 283: int d0, d1, d2;
1.1 root 284: switch (n % 4) {
285: case 0: COMMON(""); return to;
286: case 1: COMMON("\n\tmovsb"); return to;
287: case 2: COMMON("\n\tmovsw"); return to;
288: default: COMMON("\n\tmovsw\n\tmovsb"); return to;
289: }
1.1.1.2 ! root 290: }
! 291:
1.1 root 292: #undef COMMON
293: }
294:
295: #define __HAVE_ARCH_MEMCPY
296: #define memcpy(t, f, n) \
297: (__builtin_constant_p(n) ? \
298: __constant_memcpy((t),(f),(n)) : \
299: __memcpy((t),(f),(n)))
300:
301: #define __HAVE_ARCH_MEMMOVE
302: extern inline void * memmove(void * dest,const void * src, size_t n)
303: {
1.1.1.2 ! root 304: int d0, d1, d2;
1.1 root 305: if (dest<src)
306: __asm__ __volatile__(
307: "cld\n\t"
308: "rep\n\t"
309: "movsb"
1.1.1.2 ! root 310: : "=&c" (d0), "=&S" (d1), "=&D" (d2)
! 311: :"0" (n),"1" (src),"2" (dest)
! 312: : "memory");
1.1 root 313: else
314: __asm__ __volatile__(
315: "std\n\t"
316: "rep\n\t"
317: "movsb\n\t"
318: "cld"
1.1.1.2 ! root 319: : "=&c" (d0), "=&S" (d1), "=&D" (d2)
! 320: :"0" (n),
! 321: "1" (n-1+(const char *)src),
! 322: "2" (n-1+(char *)dest)
! 323: :"memory");
1.1 root 324: return dest;
325: }
326:
327: #define memcmp __builtin_memcmp
328:
329: #define __HAVE_ARCH_MEMCHR
330: extern inline void * memchr(const void * cs,int c,size_t count)
331: {
1.1.1.2 ! root 332: int d0;
1.1 root 333: register void * __res;
334: if (!count)
335: return NULL;
336: __asm__ __volatile__(
337: "cld\n\t"
338: "repne\n\t"
339: "scasb\n\t"
340: "je 1f\n\t"
341: "movl $1,%0\n"
342: "1:\tdecl %0"
1.1.1.2 ! root 343: :"=D" (__res), "=&c" (d0) : "a" (c),"0" (cs),"1" (count));
1.1 root 344: return __res;
345: }
346:
347: extern inline void * __memset_generic(void * s, char c,size_t count)
348: {
1.1.1.2 ! root 349: int d0, d1;
1.1 root 350: __asm__ __volatile__(
351: "cld\n\t"
352: "rep\n\t"
353: "stosb"
1.1.1.2 ! root 354: : "=&c" (d0), "=&D" (d1)
! 355: :"a" (c),"1" (s),"0" (count)
! 356: :"memory");
1.1 root 357: return s;
358: }
359:
360: /* we might want to write optimized versions of these later */
361: #define __constant_count_memset(s,c,count) __memset_generic((s),(c),(count))
362:
363: /*
364: * memset(x,0,y) is a reasonably common thing to do, so we want to fill
365: * things 32 bits at a time even when we don't know the size of the
366: * area at compile-time..
367: */
368: extern inline void * __constant_c_memset(void * s, unsigned long c, size_t count)
369: {
1.1.1.2 ! root 370: int d0, d1;
1.1 root 371: __asm__ __volatile__(
372: "cld\n\t"
373: "rep ; stosl\n\t"
1.1.1.2 ! root 374: "testb $2,%b3\n\t"
1.1 root 375: "je 1f\n\t"
376: "stosw\n"
1.1.1.2 ! root 377: "1:\ttestb $1,%b3\n\t"
1.1 root 378: "je 2f\n\t"
379: "stosb\n"
380: "2:"
1.1.1.2 ! root 381: : "=&c" (d0), "=&D" (d1)
! 382: :"a" (c), "q" (count), "0" (count/4), "1" ((long) s)
! 383: :"memory");
1.1 root 384: return (s);
385: }
386:
387: /* Added by Gertjan van Wingerde to make minix and sysv module work */
388: #define __HAVE_ARCH_STRNLEN
389: extern inline size_t strnlen(const char * s, size_t count)
390: {
1.1.1.2 ! root 391: int d0;
1.1 root 392: register int __res;
393: __asm__ __volatile__(
1.1.1.2 ! root 394: "movl %2,%0\n\t"
1.1 root 395: "jmp 2f\n"
396: "1:\tcmpb $0,(%0)\n\t"
397: "je 3f\n\t"
398: "incl %0\n"
1.1.1.2 ! root 399: "2:\tdecl %1\n\t"
! 400: "cmpl $-1,%1\n\t"
1.1 root 401: "jne 1b\n"
1.1.1.2 ! root 402: "3:\tsubl %2,%0"
! 403: :"=a" (__res), "=&d" (d0)
! 404: :"c" (s),"1" (count));
1.1 root 405: return __res;
406: }
407: /* end of additional stuff */
408:
409: /*
410: * This looks horribly ugly, but the compiler can optimize it totally,
411: * as we by now know that both pattern and count is constant..
412: */
413: extern inline void * __constant_c_and_count_memset(void * s, unsigned long pattern, size_t count)
414: {
415: switch (count) {
416: case 0:
417: return s;
418: case 1:
419: *(unsigned char *)s = pattern;
420: return s;
421: case 2:
422: *(unsigned short *)s = pattern;
423: return s;
424: case 3:
425: *(unsigned short *)s = pattern;
426: *(2+(unsigned char *)s) = pattern;
427: return s;
428: case 4:
429: *(unsigned long *)s = pattern;
430: return s;
431: }
432: #define COMMON(x) \
1.1.1.2 ! root 433: __asm__ __volatile__("cld\n\t" \
1.1 root 434: "rep ; stosl" \
435: x \
1.1.1.2 ! root 436: : "=&c" (d0), "=&D" (d1) \
! 437: : "a" (pattern),"0" (count/4),"1" ((long) s) \
! 438: : "memory")
! 439: {
! 440: int d0, d1;
1.1 root 441: switch (count % 4) {
442: case 0: COMMON(""); return s;
443: case 1: COMMON("\n\tstosb"); return s;
444: case 2: COMMON("\n\tstosw"); return s;
445: default: COMMON("\n\tstosw\n\tstosb"); return s;
446: }
1.1.1.2 ! root 447: }
! 448:
1.1 root 449: #undef COMMON
450: }
451:
452: #define __constant_c_x_memset(s, c, count) \
453: (__builtin_constant_p(count) ? \
454: __constant_c_and_count_memset((s),(c),(count)) : \
455: __constant_c_memset((s),(c),(count)))
456:
457: #define __memset(s, c, count) \
458: (__builtin_constant_p(count) ? \
459: __constant_count_memset((s),(c),(count)) : \
460: __memset_generic((s),(c),(count)))
461:
462: #define __HAVE_ARCH_MEMSET
463: #define memset(s, c, count) \
464: (__builtin_constant_p(c) ? \
1.1.1.2 ! root 465: __constant_c_x_memset((s),(0x01010101UL*(unsigned char)(c)),(count)) : \
1.1 root 466: __memset((s),(c),(count)))
467:
468: /*
469: * find the first occurrence of byte 'c', or 1 past the area if none
470: */
471: #define __HAVE_ARCH_MEMSCAN
472: extern inline void * memscan(void * addr, int c, size_t size)
473: {
474: if (!size)
475: return addr;
476: __asm__("cld
477: repnz; scasb
478: jnz 1f
479: dec %%edi
480: 1: "
481: : "=D" (addr), "=c" (size)
482: : "0" (addr), "1" (size), "a" (c));
483: return addr;
484: }
485:
486: #endif
487: #endif
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.