diff -buHrN linux-1.3.15/arch/i386/config.in linux/arch/i386/config.in
--- linux-1.3.15/arch/i386/config.in	Wed Aug  2 13:27:01 1995
+++ linux/arch/i386/config.in	Wed Aug  2 13:22:24 1995
@@ -38,6 +38,10 @@
 bool 'Use -m486 flag for 486-specific optimizations' CONFIG_M486 y
 #fi
 
+comment 'Using of assembler optimized library (experimental)'
+comment 'combined with 486-specific optimizations only on 486 or 586'
+bool 'Assembler optimized library' CONFIG_ASM_OPTIMIZE y
+
 comment 'Loadable module support'
 bool 'Set version information on all symbols for modules' CONFIG_MODVERSIONS n
 
diff -buHrN linux-1.3.15/include/asm-i386/byteorder.h linux/include/asm-i386/byteorder.h
--- linux-1.3.15/include/asm-i386/byteorder.h	Sat Apr 29 09:19:32 1995
+++ linux/include/asm-i386/byteorder.h	Tue Jul 18 21:47:04 1995
@@ -27,9 +27,14 @@
 extern __inline__ unsigned long int
 __ntohl(unsigned long int x)
 {
+#if (defined(CONFIG_M486) || defined(CONFIG_M586)) && \
+     defined(CONFIG_ASM_OPTIMIZE)
+	__asm__("bswap %0"
+#else /* (CONFIG_M486 || CONFIG_M586) && CONFIG_ASM_OPTIMIZE */
 	__asm__("xchgb %b0,%h0\n\t"	/* swap lower bytes	*/
 		"rorl $16,%0\n\t"	/* swap words		*/
 		"xchgb %b0,%h0"		/* swap higher bytes	*/
+#endif /* (CONFIG_M486 || CONFIG_M586) && CONFIG_ASM_OPTIMIZE */
 		:"=q" (x)
 		: "0" (x));
 	return x;
diff -buHrN linux-1.3.15/include/asm-i386/string.h linux/include/asm-i386/string.h
--- linux-1.3.15/include/asm-i386/string.h	Thu Jul 20 08:50:29 1995
+++ linux/include/asm-i386/string.h	Tue Jul 25 20:12:07 1995
@@ -10,12 +10,30 @@
  * set, making the functions fast and clean. String instructions have been
  * used through-out, making for "slightly" unclear code :-)
  *
- *		Copyright (C) 1991, 1992 Linus Torvalds
+ *		Copyright (C) 1991,1992,1993,1994 Linus Torvalds
+ *		Revised and optimized for i486/pentium
+ *		1994-1995 a.vignani@crf.it, d.parodi@crf.it
  */
 
 #define __HAVE_ARCH_STRCPY
 extern inline char * strcpy(char * dest,const char *src)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register char *tmp= (char *)dest;
+  register char dummy;
+  __asm__ __volatile__
+    ("\n1:\t"
+     "movb (%0),%2\n\t"
+     "incl %0\n\t"
+     "movb %2,(%1)\n\t"
+     "incl %1\n\t"
+     "testb %2,%2\n\t"
+     "jne 1b"
+     :"=r" (src), "=r" (tmp), "=q" (dummy)
+     :"0" (src), "1" (tmp)
+     :"memory");
+  return dest;
+#else /* CONFIG_M486 || CONFIG_M586 */
 __asm__ __volatile__(
 	"cld\n"
 	"1:\tlodsb\n\t"
@@ -25,11 +43,37 @@
 	: /* no output */
 	:"S" (src),"D" (dest):"si","di","ax","memory");
 return dest;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRNCPY
 extern inline char * strncpy(char * dest,const char *src,size_t count)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register char *tmp= (char *)dest;
+  register char dummy;
+  if (count) {
+  __asm__ __volatile__ (
+        "\n1:\t"
+        "movb (%0),%2\n\t"
+        "incl %0\n\t"
+        "movb %2,(%1)\n\t"
+        "incl %1\n\t"
+        "decl %3\n\t"
+        "je 3f\n\t"
+        "testb %2,%2\n\t"
+        "jne 1b\n\t"
+        "2:\tmovb %2,(%1)\n\t"
+        "incl %1\n\t"
+        "decl %3\n\t"
+        "jne 2b\n\t"
+        "3:"
+        :"=r" (src), "=r" (tmp), "=q" (dummy), "=r" (count)
+        :"0" (src), "1" (tmp), "3" (count)
+        :"memory");
+  }
+  return dest;
+#else /* CONFIG_M486 || CONFIG_M586 */
 __asm__ __volatile__(
 	"cld\n"
 	"1:\tdecl %2\n\t"
@@ -44,11 +88,30 @@
 	: /* no output */
 	:"S" (src),"D" (dest),"c" (count):"si","di","ax","cx","memory");
 return dest;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRCAT
 extern inline char * strcat(char * dest,const char * src)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register char *tmp = (char *)(dest-1);
+  register char dummy;
+  __asm__ __volatile__
+    ("\n1:\tincl %1\n\t"
+     "cmpb $0,(%1)\n\t"
+     "jne 1b\n"
+     "2:\tmovb (%2),%b0\n\t"
+     "incl %2\n\t"
+     "movb %b0,(%1)\n\t"
+     "incl %1\n\t"
+     "testb %b0,%b0\n\t"
+     "jne 2b\n"
+     :"=q" (dummy), "=r" (tmp), "=r" (src)
+     :"1"  (tmp), "2"  (src)
+     :"memory");
+  return dest;
+#else /* CONFIG_M486 || CONFIG_M586 */
 __asm__ __volatile__(
 	"cld\n\t"
 	"repne\n\t"
@@ -61,11 +124,34 @@
 	: /* no output */
 	:"S" (src),"D" (dest),"a" (0),"c" (0xffffffff):"si","di","ax","cx");
 return dest;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRNCAT
 extern inline char * strncat(char * dest,const char * src,size_t count)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register char *tmp = (char *)(dest-1);
+  register char dummy;
+  __asm__ __volatile__
+    ("\n1:\tincl %1\n\t"
+     "cmpb $0,(%1)\n\t"
+     "jne 1b\n"
+     "2:\tdecl %3\n\t"
+     "js 3f\n\t"
+     "movb (%2),%b0\n\t"
+     "incl %2\n\t"
+     "movb %b0,(%1)\n\t"
+     "incl %1\n\t"
+     "testb %b0,%b0\n\t"
+     "jne 2b\n"
+     "3:\txorl %0,%0\n\t"
+     "movb %b0,(%1)\n\t"
+     :"=q" (dummy), "=r" (tmp), "=r" (src), "=r" (count)
+     :"1"  (tmp), "2"  (src), "3"  (count)
+     :"memory");
+  return dest;
+#else /* CONFIG_M486 || CONFIG_M586 */
 __asm__ __volatile__(
 	"cld\n\t"
 	"repne\n\t"
@@ -84,11 +170,33 @@
 	:"S" (src),"D" (dest),"a" (0),"c" (0xffffffff),"g" (count)
 	:"si","di","ax","cx","memory");
 return dest;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRCMP
 extern inline int strcmp(const char * cs,const char * ct)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register int __res;
+  __asm__ __volatile__
+    ("\n1:\tmovb (%1),%b0\n\t"
+     "incl %1\n\t"
+     "cmpb %b0,(%2)\n\t"
+     "jne 2f\n\t"
+     "incl %2\n\t"
+     "testb %b0,%b0\n\t"
+     "jne 1b\n\t"
+     "xorl %0,%0\n\t"
+     "jmp 3f\n"
+     "2:\tmovl $1,%0\n\t"
+     "jb 3f\n\t"
+     "negl %0\n"
+     "3:"
+     :"=q" (__res), "=r" (cs), "=r" (ct)
+     :"1" (cs), "2" (ct)
+     : "memory" );
+  return __res;
+#else /* CONFIG_M486 || CONFIG_M586 */
 register int __res;
 __asm__ __volatile__(
 	"cld\n"
@@ -104,11 +212,34 @@
 	"3:"
 	:"=a" (__res):"S" (cs),"D" (ct):"si","di");
 return __res;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRNCMP
 extern inline int strncmp(const char * cs,const char * ct,size_t count)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register int __res;
+  __asm__ __volatile__
+    ("\n1:\tdecl %3\n\t"
+     "js 2f\n\t"
+     "movb (%1),%b0\n\t"
+     "incl %1\n\t"
+     "cmpb %b0,(%2)\n\t"
+     "jne 3f\n\t"
+     "incl %2\n\t"
+     "testb %b0,%b0\n\t"
+     "jne 1b\n"
+     "2:\txorl %0,%0\n\t"
+     "jmp 4f\n"
+     "3:\tmovl $1,%0\n\t"
+     "jb 4f\n\t"
+     "negl %0\n"
+     "4:"
+     :"=q" (__res), "=r" (cs), "=r" (ct), "=r" (count)
+     :"1"  (cs), "2"  (ct),  "3" (count));
+  return __res;
+#else /* CONFIG_M486 || CONFIG_M586 */
 register int __res;
 __asm__ __volatile__(
 	"cld\n"
@@ -126,11 +257,28 @@
 	"4:"
 	:"=a" (__res):"S" (cs),"D" (ct),"c" (count):"si","di","cx");
 return __res;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRCHR
 extern inline char * strchr(const char * s, int c)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register char * __res;
+  __asm__ __volatile__
+    ("movb %%al,%%ah\n"
+     "1:\tmovb (%1),%%al\n\t"
+     "cmpb %%ah,%%al\n\t"
+     "je 2f\n\t"
+     "incl %1\n\t"
+     "testb %%al,%%al\n\t"
+     "jne 1b\n\t"
+     "xorl %1,%1\n"
+     "2:\tmovl %1,%0\n\t"
+     :"=a" (__res), "=r" (s)
+     :"0" (c),      "1"  (s));
+  return __res;
+#else /* CONFIG_M486 || CONFIG_M586 */
 register char * __res;
 __asm__ __volatile__(
 	"cld\n\t"
@@ -145,11 +293,27 @@
 	"decl %0"
 	:"=a" (__res):"S" (s),"0" (c):"si");
 return __res;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRRCHR
 extern inline char * strrchr(const char * s, int c)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register char * __res;
+  __asm__ __volatile__
+    ("movb %b2,%h2\n"
+     "1:\tmovb (%1),%b2\n\t"
+     "cmpb %h2,%b2\n\t"
+     "jne 2f\n\t"
+     "movl %1,%0\n"
+     "2:\tincl %1\n\t"
+     "testb %b2,%h2\n\t"
+     "jne 1b"
+     :"=r" (__res), "=r" (s), "=q" (c)
+     :"0" (0), "1"  (s), "2" (c));
+  return __res;
+#else /* CONFIG_M486 || CONFIG_M586 */
 register char * __res;
 __asm__ __volatile__(
 	"cld\n\t"
@@ -162,6 +326,7 @@
 	"jne 1b"
 	:"=d" (__res):"0" (0),"S" (s),"a" (c):"ax","si");
 return __res;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define __HAVE_ARCH_STRSPN
@@ -277,6 +442,19 @@
 #define __HAVE_ARCH_STRLEN
 extern inline size_t strlen(const char * s)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  /* slightly slower on a 486, but with better chances of register allocation */
+  register char dummy, *tmp= (char *)s;
+  __asm__ __volatile__("\n1:\t"
+    "movb\t(%0),%1\n\t"
+    "incl\t%0\n\t"
+    "testb\t%1,%1\n\t"
+    "jne\t1b"
+    :"=r" (tmp),"=q" (dummy)
+    :"0" (s)
+    : "memory" );
+    return (tmp-s-1);
+#else /* CONFIG_M486 || CONFIG_M586 */
 register int __res;
 __asm__ __volatile__(
 	"cld\n\t"
@@ -286,6 +464,7 @@
 	"decl %0"
 	:"=c" (__res):"D" (s),"a" (0),"0" (0xffffffff):"di");
 return __res;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 extern char * ___strtok;
@@ -351,6 +530,79 @@
 return __res;
 }
 
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+
+#define __memcpy_c(d,s,count) \
+((count%4==0) ? \
+ __memcpy_by4((d),(s),(count)) : \
+ ((count%2==0) ? \
+  __memcpy_by2((d),(s),(count)) : \
+  __memcpy_g((d),(s),(count))))
+
+#define memcpy(d,s,count) \
+(__builtin_constant_p(count) ? \
+ __memcpy_c((d),(s),(count)) : \
+ __memcpy_g((d),(s),(count)))
+
+extern inline void * __memcpy_by4(void * to, const void * from, size_t n)
+{
+  register void *tmp = (void *)to;
+  register int dummy1,dummy2;
+  __asm__ __volatile__
+    ("\n1:\tmovl (%2),%0\n\t"
+     "addl $4,%2\n\t"
+     "movl %0,(%1)\n\t"
+     "addl $4,%1\n\t"
+     "decl %3\n\t"
+     "jnz 1b"
+     :"=r" (dummy1), "=r" (tmp), "=r" (from), "=r" (dummy2) 
+     :"1" (tmp), "2" (from), "3" (n/4)
+     :"memory");
+  return (to);
+}
+
+extern inline void * __memcpy_by2(void * to, const void * from, size_t n)
+{
+  register void *tmp = (void *)to;
+  register int dummy1,dummy2;
+  __asm__ __volatile__
+    ("shrl $1,%3\n\t"
+     "jz 2f\n"                 /* only a word */
+     "1:\tmovl (%2),%0\n\t"
+     "addl $4,%2\n\t"
+     "movl %0,(%1)\n\t"
+     "addl $4,%1\n\t"
+     "decl %3\n\t"
+     "jnz 1b\n"
+     "2:\tmovw (%2),%w0\n\t"
+     "movw %w0,(%1)"
+     :"=r" (dummy1), "=r" (tmp), "=r" (from), "=r" (dummy2) 
+     :"1" (tmp), "2" (from), "3" (n/2)
+     :"memory");
+  return (to);
+}
+
+extern inline void * __memcpy_g(void * to, const void * from, size_t n)
+{
+  register void *tmp = (void *)to;
+  __asm__ __volatile__
+    ("cld\n\t"
+     "shrl $1,%%ecx\n\t"
+     "jnc 1f\n\t"
+     "movsb\n"
+     "1:\tshrl $1,%%ecx\n\t"
+     "jnc 2f\n\t"
+     "movsw\n"
+     "2:\trep\n\t"
+     "movsl"
+     : /* no output */
+     :"c" (n),"D" ((long) tmp),"S" ((long) from)
+     :"cx","di","si","memory");
+  return (to);
+}
+
+#else /* CONFIG_M486 || CONFIG_M586 */
+
 extern inline void * __memcpy(void * to, const void * from, size_t n)
 {
 __asm__ __volatile__(
@@ -415,9 +667,31 @@
  __constant_memcpy((t),(f),(n)) : \
  __memcpy((t),(f),(n)))
 
+#endif /* CONFIG_M486 || CONFIG_M586 */
+
 #define __HAVE_ARCH_MEMMOVE
 extern inline void * memmove(void * dest,const void * src, size_t n)
 {
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+  register void *tmp = (void *)dest;
+  if (dest<src)
+    __asm__ __volatile__
+      ("cld\n\t"
+       "rep\n\t"
+       "movsb"
+       : /* no output */
+       :"c" (n),"S" (src),"D" (tmp)
+       :"cx","si","di");
+  else
+    __asm__ __volatile__
+       ("std\n\t"
+        "rep\n\t"
+        "movsb\n\t"
+        : /* no output */
+        :"c" (n), "S" (n-1+(const char *)src), "D" (n-1+(char *)tmp)
+        :"cx","si","di","memory");
+  return dest;
+#else /* CONFIG_M486 || CONFIG_M586 */
 if (dest<src)
 __asm__ __volatile__(
 	"cld\n\t"
@@ -438,6 +712,7 @@
 	 "D" (n-1+(char *)dest)
 	:"cx","si","di","memory");
 return dest;
+#endif /* CONFIG_M486 || CONFIG_M586 */
 }
 
 #define memcmp __builtin_memcmp
@@ -460,6 +735,143 @@
 return __res;
 }
 
+#if defined(CONFIG_M486) || defined(CONFIG_M586)
+
+#define __memset_cc(s,c,count) \
+((count%4==0) ? \
+ __memset_cc_by4((s),(c),(count)) : \
+ ((count%2==0) ? \
+  __memset_cc_by2((s),(c),(count)) : \
+  __memset_cg((s),(c),(count))))
+
+#define __memset_gc(s,c,count) \
+((count%4==0) ? \
+ __memset_gc_by4((s),(c),(count)) : \
+ ((count%2==0) ? \
+  __memset_gc_by2((s),(c),(count)) : \
+  __memset_gg((s),(c),(count))))
+
+#define memset(s,c,count) \
+(__builtin_constant_p(c) ? \
+ (__builtin_constant_p(count) ? \
+  __memset_cc((s),(c),(count)) : \
+  __memset_cg((s),(c),(count))) : \
+ (__builtin_constant_p(count) ? \
+  __memset_gc((s),(c),(count)) : \
+  __memset_gg((s),(c),(count))))
+
+extern inline void * __memset_cc_by4(void * s, char c, size_t count)
+{
+/*
+ *register char *tmp = s;
+ */
+  register void *tmp = (void *)s;
+  register int  dummy;
+  __asm__ __volatile__
+    ("\n1:\tmovl %2,(%0)\n\t"
+     "addl $4,%0\n\t"
+     "decl %1\n\t"
+     "jnz 1b"
+     :"=r" (tmp), "=r" (dummy)
+     :"q" (0x01010101UL * (unsigned char) c), "0" (tmp), "1" (count/4)
+     :"memory");
+  return s;
+}
+
+extern inline void * __memset_cc_by2(void * s, char c, size_t count)
+{
+  register void *tmp = (void *)s;
+  register int  dummy;
+  __asm__ __volatile__
+    ("shrl $1,%1\n\t"          /* may be divisible also by 4 */
+     "jz 2f\n"
+     "\n1:\tmovl %2,(%0)\n\t"
+     "addl $4,%0\n\t"
+     "decl %1\n\t"
+     "jnz 1b\n"
+     "2:\tmovw %w2,(%0)"
+     :"=r" (tmp), "=r" (dummy)
+     :"q" (0x01010101UL * (unsigned char) c), "0" (tmp), "1" (count/2)
+     :"memory");
+  return s;
+}
+
+extern inline void * __memset_gc_by4(void * s, char c, size_t count)
+{
+  register void *tmp = (void *)s;
+  register int dummy;
+  __asm__ __volatile__
+    ("movb %b0,%h0\n"
+     "pushw %w0\n\t"
+     "shll $16,%0\n\t"
+     "popw %w0\n"
+     "1:\tmovl %0,(%1)\n\t"
+     "addl $4,%1\n\t"
+     "decl %2\n\t"
+     "jnz 1b\n"
+     :"=q" (c), "=r" (tmp), "=r" (dummy)
+     :"0" ((unsigned) c),  "1"  (tmp), "2" (count/4)
+     :"memory");
+  return s;
+}
+
+extern inline void * __memset_gc_by2(void * s, char c, size_t count)
+{
+  register void *tmp = (void *)s;
+  register int dummy1,dummy2;
+  __asm__ __volatile__
+    ("movb %b0,%h0\n\t"
+     "shrl $1,%2\n\t"          /* may be divisible also by 4 */
+     "jz 2f\n\t"
+     "pushw %w0\n\t"
+     "shll $16,%0\n\t"
+     "popw %w0\n"
+     "1:\tmovl %0,(%1)\n\t"
+     "addl $4,%1\n\t"
+     "decl %2\n\t"
+     "jnz 1b\n"
+     "2:\tmovw %w0,(%1)"
+     :"=q" (dummy1), "=r" (tmp), "=r" (dummy2)
+     :"0" ((unsigned) c),  "1"  (tmp), "2" (count/2)
+     :"memory");
+  return s;
+}
+
+extern inline void * __memset_cg(void * s, char c, size_t count)
+{
+  register void *tmp = (void *)s;
+  __asm__ __volatile__
+    ("shrl $1,%%ecx\n\t"
+     "rep\n\t"
+     "stosw\n\t"
+     "jnc 1f\n\t"
+     "movb %%al,(%%edi)\n"
+     "1:"
+     : /* no output */
+     :"c" (count),"D" (tmp), "a" (0x0101U * (unsigned char) c)
+     :"cx","di","memory");
+  return s;
+}
+
+extern inline void * __memset_gg(void * s,char c,size_t count)
+{
+  register void *tmp = (void *)s;
+  __asm__ __volatile__
+    ("movb %%al,%%ah\n\t"
+     "shrl $1,%%ecx\n\t"
+     "rep\n\t"
+     "stosw\n\t"
+     "jnc 1f\n\t"
+     "movb %%al,(%%edi)\n"
+     "1:"
+     : /* no output */
+     :"c" (count),"D" (tmp), "a" (c)
+     :"cx","di","memory");
+  return s;
+}
+
+#else /* CONFIG_M486 || CONFIG_M586 */
+
 extern inline void * __memset_generic(void * s, char c,size_t count)
 {
 __asm__ __volatile__(
@@ -554,6 +966,8 @@
 (__builtin_constant_p(c) ? \
  __constant_c_x_memset((s),(0x01010101UL*(unsigned char)c),(count)) : \
  __memset((s),(c),(count)))
+
+#endif /* CONFIG_M486 || CONFIG_M586 */
 
 /*
  * find the first occurrence of byte 'c', or 1 past the area if none
diff -buHrN linux-1.3.15/lib/string.c linux/lib/string.c
--- linux-1.3.15/lib/string.c	Tue Jul 25 17:24:10 1995
+++ linux/lib/string.c	Thu Jul 27 21:20:58 1995
@@ -11,6 +11,14 @@
  * These are buggy as well..
  */
  
+#if defined(CONFIG_ASM_OPTIMIZE)
+
+#include <linux/string.h>
+
+char * ___strtok = NULL;
+
+#else /* CONFIG_ASM_OPTIMIZE */
+
 #include <linux/types.h>
 #include <linux/string.h>
 
@@ -297,3 +305,5 @@
 	return NULL;
 }
 #endif
+
+#endif /* CONFIG_ASM_OPTIMIZE */