Another attempt to handle overlapping buffers in WideCharToMultiByte

[Dmitry Timoshkov]

Hello,

new assembly looks quite similar to the old one, so I assume that
it's as fast as it was before. But getting rid of 2 comparisons
per each WideCharToMultiByte call for testing whether the buffers
overlap should slightly improve performance, taking into account
amount of A<->W translations in a usual application run.

Changelog:
    Dmitry Timoshkov <dmitry at codeweavers.com>
    Another attempt to handle overlapping buffers in WideCharToMultiByte.

diff -up cvs/hq/wine/dlls/kernel/tests/codepage.c wine/dlls/kernel/tests/codepage.c
--- cvs/hq/wine/dlls/kernel/tests/codepage.c	Sat Jul 16 14:45:25 2005
+++ wine/dlls/kernel/tests/codepage.c	Sat Jul 16 15:55:02 2005
@@ -58,8 +58,8 @@ static void test_negative_source_length(
 
 static void test_overlapped_buffers(void)
 {
-    static const WCHAR strW[] = {'j','u','s','t',' ','a',' ','t','e','s','t',0};
-    static const char strA[] = "just a test";
+    static const WCHAR strW[] = {'j','u','s','t',' ','a',' ','t','e','s','t',' ','s','t','r','i','n','g',' ','l','o','n','g','e','r',' ','1','6',' ','c','h','a','r','a','c','t','e','r','s',0};
+    static const char strA[] = "just a test string longer 16 characters";
     char buf[256];
     int ret;
 
diff -up cvs/hq/wine/libs/unicode/wctomb.c wine/libs/unicode/wctomb.c
--- cvs/hq/wine/libs/unicode/wctomb.c	Thu Dec  2 14:04:52 2004
+++ wine/libs/unicode/wctomb.c	Sat Jul 16 19:42:26 2005
@@ -133,7 +133,7 @@ static inline int wcstombs_sbcs( const s
 {
     const unsigned char  * const uni2cp_low = table->uni2cp_low;
     const unsigned short * const uni2cp_high = table->uni2cp_high;
-    int ret = srclen;
+    int ret = srclen, inc;
 
     if (dstlen < srclen)
     {
@@ -142,44 +142,39 @@ static inline int wcstombs_sbcs( const s
         ret = -1;
     }
 
-    if (dst <= (const char *)src && dst + 16 > (const char *)src)
-    {
-        /* overlapping buffers, do it char by char */
-        while (srclen--)
-        {
-            *dst++ = uni2cp_low[uni2cp_high[*src >> 8] + (*src & 0xff)];
-            src++;
-        }
-        return ret;
-    }
+    /* handle remainder first */
+    inc = srclen & 0xf;
 
     for (;;)
     {
-        switch(srclen)
+        dst += inc;
+        src += inc;
+
+        switch(inc)
         {
+        /* do it char by char in order to handle overlapping buffers */
         default:
-        case 16: dst[15] = uni2cp_low[uni2cp_high[src[15] >> 8] + (src[15] & 0xff)];
-        case 15: dst[14] = uni2cp_low[uni2cp_high[src[14] >> 8] + (src[14] & 0xff)];
-        case 14: dst[13] = uni2cp_low[uni2cp_high[src[13] >> 8] + (src[13] & 0xff)];
-        case 13: dst[12] = uni2cp_low[uni2cp_high[src[12] >> 8] + (src[12] & 0xff)];
-        case 12: dst[11] = uni2cp_low[uni2cp_high[src[11] >> 8] + (src[11] & 0xff)];
-        case 11: dst[10] = uni2cp_low[uni2cp_high[src[10] >> 8] + (src[10] & 0xff)];
-        case 10: dst[9]  = uni2cp_low[uni2cp_high[src[9]  >> 8] + (src[9]  & 0xff)];
-        case 9:  dst[8]  = uni2cp_low[uni2cp_high[src[8]  >> 8] + (src[8]  & 0xff)];
-        case 8:  dst[7]  = uni2cp_low[uni2cp_high[src[7]  >> 8] + (src[7]  & 0xff)];
-        case 7:  dst[6]  = uni2cp_low[uni2cp_high[src[6]  >> 8] + (src[6]  & 0xff)];
-        case 6:  dst[5]  = uni2cp_low[uni2cp_high[src[5]  >> 8] + (src[5]  & 0xff)];
-        case 5:  dst[4]  = uni2cp_low[uni2cp_high[src[4]  >> 8] + (src[4]  & 0xff)];
-        case 4:  dst[3]  = uni2cp_low[uni2cp_high[src[3]  >> 8] + (src[3]  & 0xff)];
-        case 3:  dst[2]  = uni2cp_low[uni2cp_high[src[2]  >> 8] + (src[2]  & 0xff)];
-        case 2:  dst[1]  = uni2cp_low[uni2cp_high[src[1]  >> 8] + (src[1]  & 0xff)];
-        case 1:  dst[0]  = uni2cp_low[uni2cp_high[src[0]  >> 8] + (src[0]  & 0xff)];
+        case 16: dst[-16] = uni2cp_low[uni2cp_high[src[-16] >> 8] + (src[-16] & 0xff)];
+        case 15: dst[-15] = uni2cp_low[uni2cp_high[src[-15] >> 8] + (src[-15] & 0xff)];
+        case 14: dst[-14] = uni2cp_low[uni2cp_high[src[-14] >> 8] + (src[-14] & 0xff)];
+        case 13: dst[-13] = uni2cp_low[uni2cp_high[src[-13] >> 8] + (src[-13] & 0xff)];
+        case 12: dst[-12] = uni2cp_low[uni2cp_high[src[-12] >> 8] + (src[-12] & 0xff)];
+        case 11: dst[-11] = uni2cp_low[uni2cp_high[src[-11] >> 8] + (src[-11] & 0xff)];
+        case 10: dst[-10] = uni2cp_low[uni2cp_high[src[-10] >> 8] + (src[-10] & 0xff)];
+        case 9:  dst[-9]  = uni2cp_low[uni2cp_high[src[-9]  >> 8] + (src[-9]  & 0xff)];
+        case 8:  dst[-8]  = uni2cp_low[uni2cp_high[src[-8]  >> 8] + (src[-8]  & 0xff)];
+        case 7:  dst[-7]  = uni2cp_low[uni2cp_high[src[-7]  >> 8] + (src[-7]  & 0xff)];
+        case 6:  dst[-6]  = uni2cp_low[uni2cp_high[src[-6]  >> 8] + (src[-6]  & 0xff)];
+        case 5:  dst[-5]  = uni2cp_low[uni2cp_high[src[-5]  >> 8] + (src[-5]  & 0xff)];
+        case 4:  dst[-4]  = uni2cp_low[uni2cp_high[src[-4]  >> 8] + (src[-4]  & 0xff)];
+        case 3:  dst[-3]  = uni2cp_low[uni2cp_high[src[-3]  >> 8] + (src[-3]  & 0xff)];
+        case 2:  dst[-2]  = uni2cp_low[uni2cp_high[src[-2]  >> 8] + (src[-2]  & 0xff)];
+        case 1:  dst[-1]  = uni2cp_low[uni2cp_high[src[-1]  >> 8] + (src[-1]  & 0xff)];
         case 0: break;
         }
         if (srclen < 16) return ret;
-        dst += 16;
-        src += 16;
         srclen -= 16;
+        inc = 16; /* from now on srclen is aligned to 16 */
     }
 }