Merge branch 'master' of git://localhost/acess2

[tpg/acess2.git] / Usermode / Libraries / libc.so_src / string.c

diff --git a/Usermode/Libraries/libc.so_src/string.c b/Usermode/Libraries/libc.so_src/string.c
index ceafd99..d8d0872 100644 (file)
--- a/Usermode/Libraries/libc.so_src/string.c
+++ b/Usermode/Libraries/libc.so_src/string.c
@@ -14,7 +14,7 @@
  */
 EXPORT int strcmp(const char *s1, const char *s2)
 {
-       while(*s1 == *s2 && *s1 != '\0' && *s2 != '\0') {
+       while(*s1 && *s1 == *s2) {
                s1++; s2++;
        }
        return (int)*s1 - (int)*s2;
@@ -26,7 +26,7 @@ EXPORT int strcmp(const char *s1, const char *s2)
  */
 EXPORT int strncmp(const char *s1, const char *s2, size_t n)
 {
-       while(n && *s1 == *s2 && *s1 != '\0' && *s2 != '\0')
+       while(n && *s1 && *s1 == *s2)
        {
                s1++; s2++;
                n --;
@@ -217,12 +217,51 @@ EXPORT void *memset(void *dest, int val, size_t num)
  * \fn EXPORT void *memcpy(void *dest, const void *src, size_t count)
  * \brief Copy one memory area to another
  */
-EXPORT void *memcpy(void *dest, const void *src, size_t count)
+EXPORT void *memcpy(void *__dest, const void *__src, size_t count)
 {
-    char *sp = (char *)src;
-    char *dp = (char *)dest;
-    for(;count--;) *dp++ = *sp++;
-    return dest;
+       const int       wordmask = sizeof(void*)-1;
+       uintptr_t       src = (uintptr_t)__src;
+       uintptr_t       dst = (uintptr_t)__dest;
+
+       if( count < sizeof(void*)*2 || (dst & wordmask) != (src & wordmask) )
+       {
+               char    *dp = __dest;
+               const char      *sp = __src;
+               while(count--) *dp++ = *sp ++;
+       }
+       // TODO: Bulk aligned copies
+       #if 0
+       else if(count > 128 && (dst & 15) == (src & 15) )
+       {
+               // SSE/bulk copy
+               for( ; dst & 15; count -- )
+                       *(char*)dst++ = *(char*)src++;
+               memcpy_16byte(dst, src, count / 16);
+               dst += count & ~15;
+               src += count & ~15;
+               count &= 15;
+               while(count --)
+                       *(char*)dst++ = *(char*)src++;
+       }
+       #endif
+       else
+       {
+               void    **dp, **sp;
+               for( ; count && (dst & wordmask) != 0; count -- )
+                       *(char*)dst++ = *(char*)src++;
+
+               dp = (void*)dst; sp = (void*)src;
+               while( count >= sizeof(void*) )
+               {
+                       *dp++ = *sp++;
+                       count -= sizeof(void*);
+               }
+               dst = (uintptr_t)dp; src = (uintptr_t)sp;
+               for( ; count; count -- )
+                       *(char*)dst++ = *(char*)src++;
+       }
+
+       return __dest;
 }
 
 /**
@@ -231,14 +270,15 @@ EXPORT void *memcpy(void *dest, const void *src, size_t count)
  */
 EXPORT void *memmove(void *dest, const void *src, size_t count)
 {
-    char *sp = (char *)src;
-    char *dp = (char *)dest;
-       // Check if corruption will happen
-       if( (intptr_t)dest > (intptr_t)src && (intptr_t)dest < (intptr_t)src+count )
-               for(;count--;) dp[count] = sp[count];
+       char *sp = (char *)src;
+       char *dp = (char *)dest;
+       // Check if the areas overlap
+       if( (uintptr_t)src < (uintptr_t)dest && (uintptr_t)dest < (uintptr_t)src+count )
+               for(;count--;)
+                       dp[count] = sp[count];
        else
-       for(;count--;) *dp++ = *sp++;
-    return dest;
+               memcpy(dest, src, count);
+       return dest;
 }
 
 /**
@@ -250,12 +290,13 @@ EXPORT void *memmove(void *dest, const void *src, size_t count)
  */
 EXPORT int memcmp(const void *mem1, const void *mem2, size_t count)
 {
+       const unsigned char     *p1 = mem1, *p2 = mem2;
        while(count--)
        {
-               if( *(unsigned char*)mem1 != *(unsigned char*)mem2 )
-                       return *(unsigned char*)mem1 - *(unsigned char*)mem2;
-               mem1 ++;
-               mem2 ++;
+               if( *p1 != *p2 )
+                       return *p1 - *p2;
+               p1 ++;
+               p2 ++;
        }
        return 0;
 }