summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorDenis Vlasenko <vda@port.imtp.ilyichevsk.odessa.ua>2005-05-01 08:58:48 -0700
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-05-01 08:58:48 -0700
commitd5b63d78f1e75f6c6f04862dfb2f2a4aeffafd4c (patch)
treee208151fd10b19c774ff51478bf3e857b99d63f9
parentd637413f3f05b41f678f8004225b33b62274183f (diff)
[PATCH] fix i386 memcpy
This patch shortens non-constant memcpy() by two bytes and fixes spurious out-of-line constant memcpy(). # size vmlinux.org vmlinux text data bss dec hex filename 3954591 1553426 236544 5744561 57a7b1 vmlinux.org 3952615 1553426 236544 5742585 579ff9 vmlinux Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
-rw-r--r--include/asm-i386/string.h89
1 files changed, 61 insertions, 28 deletions
diff --git a/include/asm-i386/string.h b/include/asm-i386/string.h
index 1679983d053f..6a78ac58c194 100644
--- a/include/asm-i386/string.h
+++ b/include/asm-i386/string.h
@@ -198,47 +198,80 @@ static inline void * __memcpy(void * to, const void * from, size_t n)
int d0, d1, d2;
__asm__ __volatile__(
"rep ; movsl\n\t"
- "testb $2,%b4\n\t"
- "je 1f\n\t"
- "movsw\n"
- "1:\ttestb $1,%b4\n\t"
- "je 2f\n\t"
- "movsb\n"
- "2:"
+ "movl %4,%%ecx\n\t"
+ "andl $3,%%ecx\n\t"
+#if 1 /* want to pay 2 byte penalty for a chance to skip microcoded rep? */
+ "jz 1f\n\t"
+#endif
+ "rep ; movsb\n\t"
+ "1:"
: "=&c" (d0), "=&D" (d1), "=&S" (d2)
- :"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from)
+ : "0" (n/4), "g" (n), "1" ((long) to), "2" ((long) from)
: "memory");
return (to);
}
/*
- * This looks horribly ugly, but the compiler can optimize it totally,
+ * This looks ugly, but the compiler can optimize it totally,
* as the count is constant.
*/
static inline void * __constant_memcpy(void * to, const void * from, size_t n)
{
- if (n <= 128)
- return __builtin_memcpy(to, from, n);
-
-#define COMMON(x) \
-__asm__ __volatile__( \
- "rep ; movsl" \
- x \
- : "=&c" (d0), "=&D" (d1), "=&S" (d2) \
- : "0" (n/4),"1" ((long) to),"2" ((long) from) \
- : "memory");
-{
- int d0, d1, d2;
+ long esi, edi;
+ if (!n) return to;
+#if 1 /* want to do small copies with non-string ops? */
+ switch (n) {
+ case 1: *(char*)to = *(char*)from; return to;
+ case 2: *(short*)to = *(short*)from; return to;
+ case 4: *(int*)to = *(int*)from; return to;
+#if 1 /* including those doable with two moves? */
+ case 3: *(short*)to = *(short*)from;
+ *((char*)to+2) = *((char*)from+2); return to;
+ case 5: *(int*)to = *(int*)from;
+ *((char*)to+4) = *((char*)from+4); return to;
+ case 6: *(int*)to = *(int*)from;
+ *((short*)to+2) = *((short*)from+2); return to;
+ case 8: *(int*)to = *(int*)from;
+ *((int*)to+1) = *((int*)from+1); return to;
+#endif
+ }
+#endif
+ esi = (long) from;
+ edi = (long) to;
+ if (n >= 5*4) {
+ /* large block: use rep prefix */
+ int ecx;
+ __asm__ __volatile__(
+ "rep ; movsl"
+ : "=&c" (ecx), "=&D" (edi), "=&S" (esi)
+ : "0" (n/4), "1" (edi),"2" (esi)
+ : "memory"
+ );
+ } else {
+ /* small block: don't clobber ecx + smaller code */
+ if (n >= 4*4) __asm__ __volatile__("movsl"
+ :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
+ if (n >= 3*4) __asm__ __volatile__("movsl"
+ :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
+ if (n >= 2*4) __asm__ __volatile__("movsl"
+ :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
+ if (n >= 1*4) __asm__ __volatile__("movsl"
+ :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
+ }
switch (n % 4) {
- case 0: COMMON(""); return to;
- case 1: COMMON("\n\tmovsb"); return to;
- case 2: COMMON("\n\tmovsw"); return to;
- default: COMMON("\n\tmovsw\n\tmovsb"); return to;
+ /* tail */
+ case 0: return to;
+ case 1: __asm__ __volatile__("movsb"
+ :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
+ return to;
+ case 2: __asm__ __volatile__("movsw"
+ :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
+ return to;
+ default: __asm__ __volatile__("movsw\n\tmovsb"
+ :"=&D"(edi),"=&S"(esi):"0"(edi),"1"(esi):"memory");
+ return to;
}
}
-
-#undef COMMON
-}
#define __HAVE_ARCH_MEMCPY