Skip to content

Commit b520b85

Browse files
Andi KleenLinus Torvalds
Andi Kleen
authored and
Linus Torvalds
committed
i386: Move all simple string operations out of line
The compiler generally generates reasonable inline code for the simple cases and for the rest it's better for code size for them to be out of line. Also there they can be potentially optimized more in the future. In fact they probably should be in a .S file because they're all pure assembly, but that's for another day. Also some code style cleanup on them while I was on it (this seems to be the last untouched really early Linux code) This saves ~12k text for a defconfig kernel with gcc 4.1. Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1 parent aac57f8 commit b520b85

File tree

3 files changed

+271
-231
lines changed

3 files changed

+271
-231
lines changed

‎arch/i386/lib/Makefile

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4,7 +4,7 @@
44

55

66
lib-y = checksum.o delay.o usercopy.o getuser.o putuser.o memcpy.o strstr.o \
7-
bitops.o semaphore.o
7+
bitops.o semaphore.o string.o
88

99
lib-$(CONFIG_X86_USE_3DNOW) += mmx.o
1010

‎arch/i386/lib/string.c

Lines changed: 257 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,257 @@
1+
/*
2+
* Most of the string-functions are rather heavily hand-optimized,
3+
* see especially strsep,strstr,str[c]spn. They should work, but are not
4+
* very easy to understand. Everything is done entirely within the register
5+
* set, making the functions fast and clean. String instructions have been
6+
* used through-out, making for "slightly" unclear code :-)
7+
*
8+
* AK: On P4 and K7 using non string instruction implementations might be faster
9+
* for large memory blocks. But most of them are unlikely to be used on large
10+
* strings.
11+
*/
12+
13+
#include <linux/string.h>
14+
#include <linux/module.h>
15+
16+
#ifdef __HAVE_ARCH_STRCPY
17+
char *strcpy(char * dest,const char *src)
18+
{
19+
int d0, d1, d2;
20+
asm volatile( "1:\tlodsb\n\t"
21+
"stosb\n\t"
22+
"testb %%al,%%al\n\t"
23+
"jne 1b"
24+
: "=&S" (d0), "=&D" (d1), "=&a" (d2)
25+
:"0" (src),"1" (dest) : "memory");
26+
return dest;
27+
}
28+
EXPORT_SYMBOL(strcpy);
29+
#endif
30+
31+
#ifdef __HAVE_ARCH_STRNCPY
32+
char *strncpy(char * dest,const char *src,size_t count)
33+
{
34+
int d0, d1, d2, d3;
35+
asm volatile( "1:\tdecl %2\n\t"
36+
"js 2f\n\t"
37+
"lodsb\n\t"
38+
"stosb\n\t"
39+
"testb %%al,%%al\n\t"
40+
"jne 1b\n\t"
41+
"rep\n\t"
42+
"stosb\n"
43+
"2:"
44+
: "=&S" (d0), "=&D" (d1), "=&c" (d2), "=&a" (d3)
45+
:"0" (src),"1" (dest),"2" (count) : "memory");
46+
return dest;
47+
}
48+
EXPORT_SYMBOL(strncpy);
49+
#endif
50+
51+
#ifdef __HAVE_ARCH_STRCAT
52+
char *strcat(char * dest,const char * src)
53+
{
54+
int d0, d1, d2, d3;
55+
asm volatile( "repne\n\t"
56+
"scasb\n\t"
57+
"decl %1\n"
58+
"1:\tlodsb\n\t"
59+
"stosb\n\t"
60+
"testb %%al,%%al\n\t"
61+
"jne 1b"
62+
: "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
63+
: "0" (src), "1" (dest), "2" (0), "3" (0xffffffffu): "memory");
64+
return dest;
65+
}
66+
EXPORT_SYMBOL(strcat);
67+
#endif
68+
69+
#ifdef __HAVE_ARCH_STRNCAT
70+
char *strncat(char * dest,const char * src,size_t count)
71+
{
72+
int d0, d1, d2, d3;
73+
asm volatile( "repne\n\t"
74+
"scasb\n\t"
75+
"decl %1\n\t"
76+
"movl %8,%3\n"
77+
"1:\tdecl %3\n\t"
78+
"js 2f\n\t"
79+
"lodsb\n\t"
80+
"stosb\n\t"
81+
"testb %%al,%%al\n\t"
82+
"jne 1b\n"
83+
"2:\txorl %2,%2\n\t"
84+
"stosb"
85+
: "=&S" (d0), "=&D" (d1), "=&a" (d2), "=&c" (d3)
86+
: "0" (src),"1" (dest),"2" (0),"3" (0xffffffffu), "g" (count)
87+
: "memory");
88+
return dest;
89+
}
90+
EXPORT_SYMBOL(strncat);
91+
#endif
92+
93+
#ifdef __HAVE_ARCH_STRCMP
94+
int strcmp(const char * cs,const char * ct)
95+
{
96+
int d0, d1;
97+
int res;
98+
asm volatile( "1:\tlodsb\n\t"
99+
"scasb\n\t"
100+
"jne 2f\n\t"
101+
"testb %%al,%%al\n\t"
102+
"jne 1b\n\t"
103+
"xorl %%eax,%%eax\n\t"
104+
"jmp 3f\n"
105+
"2:\tsbbl %%eax,%%eax\n\t"
106+
"orb $1,%%al\n"
107+
"3:"
108+
:"=a" (res), "=&S" (d0), "=&D" (d1)
109+
:"1" (cs),"2" (ct)
110+
:"memory");
111+
return res;
112+
}
113+
EXPORT_SYMBOL(strcmp);
114+
#endif
115+
116+
#ifdef __HAVE_ARCH_STRNCMP
117+
int strncmp(const char * cs,const char * ct,size_t count)
118+
{
119+
int res;
120+
int d0, d1, d2;
121+
asm volatile( "1:\tdecl %3\n\t"
122+
"js 2f\n\t"
123+
"lodsb\n\t"
124+
"scasb\n\t"
125+
"jne 3f\n\t"
126+
"testb %%al,%%al\n\t"
127+
"jne 1b\n"
128+
"2:\txorl %%eax,%%eax\n\t"
129+
"jmp 4f\n"
130+
"3:\tsbbl %%eax,%%eax\n\t"
131+
"orb $1,%%al\n"
132+
"4:"
133+
:"=a" (res), "=&S" (d0), "=&D" (d1), "=&c" (d2)
134+
:"1" (cs),"2" (ct),"3" (count)
135+
:"memory");
136+
return res;
137+
}
138+
EXPORT_SYMBOL(strncmp);
139+
#endif
140+
141+
#ifdef __HAVE_ARCH_STRCHR
142+
char *strchr(const char * s, int c)
143+
{
144+
int d0;
145+
char * res;
146+
asm volatile( "movb %%al,%%ah\n"
147+
"1:\tlodsb\n\t"
148+
"cmpb %%ah,%%al\n\t"
149+
"je 2f\n\t"
150+
"testb %%al,%%al\n\t"
151+
"jne 1b\n\t"
152+
"movl $1,%1\n"
153+
"2:\tmovl %1,%0\n\t"
154+
"decl %0"
155+
:"=a" (res), "=&S" (d0)
156+
:"1" (s),"0" (c)
157+
:"memory");
158+
return res;
159+
}
160+
EXPORT_SYMBOL(strchr);
161+
#endif
162+
163+
#ifdef __HAVE_ARCH_STRRCHR
164+
char *strrchr(const char * s, int c)
165+
{
166+
int d0, d1;
167+
char * res;
168+
asm volatile( "movb %%al,%%ah\n"
169+
"1:\tlodsb\n\t"
170+
"cmpb %%ah,%%al\n\t"
171+
"jne 2f\n\t"
172+
"leal -1(%%esi),%0\n"
173+
"2:\ttestb %%al,%%al\n\t"
174+
"jne 1b"
175+
:"=g" (res), "=&S" (d0), "=&a" (d1)
176+
:"0" (0),"1" (s),"2" (c)
177+
:"memory");
178+
return res;
179+
}
180+
EXPORT_SYMBOL(strrchr);
181+
#endif
182+
183+
#ifdef __HAVE_ARCH_STRLEN
184+
size_t strlen(const char * s)
185+
{
186+
int d0;
187+
int res;
188+
asm volatile( "repne\n\t"
189+
"scasb\n\t"
190+
"notl %0\n\t"
191+
"decl %0"
192+
:"=c" (res), "=&D" (d0)
193+
:"1" (s),"a" (0), "0" (0xffffffffu)
194+
:"memory");
195+
return res;
196+
}
197+
EXPORT_SYMBOL(strlen);
198+
#endif
199+
200+
#ifdef __HAVE_ARCH_MEMCHR
201+
void *memchr(const void *cs,int c,size_t count)
202+
{
203+
int d0;
204+
void *res;
205+
if (!count)
206+
return NULL;
207+
asm volatile( "repne\n\t"
208+
"scasb\n\t"
209+
"je 1f\n\t"
210+
"movl $1,%0\n"
211+
"1:\tdecl %0"
212+
:"=D" (res), "=&c" (d0)
213+
:"a" (c),"0" (cs),"1" (count)
214+
:"memory");
215+
return res;
216+
}
217+
EXPORT_SYMBOL(memchr);
218+
#endif
219+
220+
#ifdef __HAVE_ARCH_MEMSCAN
221+
void *memscan(void * addr, int c, size_t size)
222+
{
223+
if (!size)
224+
return addr;
225+
asm volatile("repnz; scasb\n\t"
226+
"jnz 1f\n\t"
227+
"dec %%edi\n"
228+
"1:"
229+
: "=D" (addr), "=c" (size)
230+
: "0" (addr), "1" (size), "a" (c)
231+
: "memory");
232+
return addr;
233+
}
234+
EXPORT_SYMBOL(memscan);
235+
#endif
236+
237+
#ifdef __HAVE_ARCH_STRNLEN
238+
size_t strnlen(const char *s, size_t count)
239+
{
240+
int d0;
241+
int res;
242+
asm volatile( "movl %2,%0\n\t"
243+
"jmp 2f\n"
244+
"1:\tcmpb $0,(%0)\n\t"
245+
"je 3f\n\t"
246+
"incl %0\n"
247+
"2:\tdecl %1\n\t"
248+
"cmpl $-1,%1\n\t"
249+
"jne 1b\n"
250+
"3:\tsubl %2,%0"
251+
:"=a" (res), "=&d" (d0)
252+
:"c" (s),"1" (count)
253+
:"memory");
254+
return res;
255+
}
256+
EXPORT_SYMBOL(strnlen);
257+
#endif

0 commit comments

Comments
 (0)