From d1254419852d1816e204d98b081884fe31c6ea07 Mon Sep 17 00:00:00 2001
From: rfm <rfm@72102866-910b-0410-8b05-ffd578937521>
Date: Sat, 6 Jun 2009 15:52:11 +0000
Subject: [PATCH] Add some optimisation for converting from unicode to latin1
 or ascii ... move as much as possible outside the loop iterating over the
 characters.

git-svn-id: svn+ssh://svn.gna.org/svn/gnustep/libs/base/trunk@28332 72102866-910b-0410-8b05-ffd578937521
---
 ChangeLog                  |   5 ++
 Source/Additions/Unicode.m | 130 ++++++++++++++++++++++++++++---------
 2 files changed, 104 insertions(+), 31 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index a31b4b2d2..204eb4444 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,8 @@
+2009-06-06 Richard Frith-Macdonald <rfm@gnu.org>
+
+	* Source/Additions/Unicode.m: Optimise case where we are converting
+	from unicode to latin1 or ascii.
+
 2009-06-06  Wolfgang Lux  <wolfgang.lux@gmail.com>
 
 	* Source/GSFFIInvocation.m
diff --git a/Source/Additions/Unicode.m b/Source/Additions/Unicode.m
index c628d95c3..c16971391 100644
--- a/Source/Additions/Unicode.m
+++ b/Source/Additions/Unicode.m
@@ -1922,53 +1922,121 @@ GSFromUnicode(unsigned char **dst, unsigned int *size, const unichar *src,
 	goto bases;
 
 bases:
-	if (strict == NO)
+	if (dst == 0)
 	  {
-	    while (spos < slen)
+	    /* Just counting bytes, and we know there is exactly one
+	     * unicode codepoint needed for each character.
+	     */
+	    dpos = slen;
+	  }
+        else
+	  {
+	    /* Because we know that each ascii chartacter is exactly
+	     * one unicode character, we can check the destination
+	     * buffer size and allocate more space in one go, before
+	     * entering the loop where we deal with each character.
+	     */
+	    if (slen > bsize)
 	      {
-		unichar	u = src[spos++];
-
-		if (swapped == YES)
+		if (zone == 0)
 		  {
-		    u = (((u & 0xff00) >> 8) + ((u & 0x00ff) << 8));
-		  }
-		
-		if (dpos >= bsize)
-		  {
-		    GROW();
-		  }
-		if (u < base)
-		  {
-		    ptr[dpos++] = (unsigned char)u;
+		    result = NO; /* No buffer growth possible ... fail. */
+		    goto done;
 		  }
 		else
 		  {
-		    ptr[dpos++] =  '?';
+		    uint8_t	*tmp;
+
+#if	GS_WITH_GC
+		    tmp = NSAllocateCollectable(slen, 0);
+#else
+		    tmp = NSZoneMalloc(zone, slen);
+		    if (ptr != buf)
+		      {
+			NSZoneFree(zone, ptr);
+		      }
+#endif
+		    ptr = tmp;
+		    if (ptr == 0)
+		      {
+			result = NO;	/* Not enough memory */
+			break;
+		      }
+		    bsize = slen;
+		  }
+	      }
+	  }
+	if (strict == NO)
+	  {
+	    if (swapped == YES)
+	      {
+		while (spos < slen)
+		  {
+		    unichar	u = src[spos++];
+
+		    u = (((u & 0xff00) >> 8) + ((u & 0x00ff) << 8));
+		    if (u < base)
+		      {
+			ptr[dpos++] = (unsigned char)u;
+		      }
+		    else
+		      {
+			ptr[dpos++] =  '?';
+		      }
+		  }
+	      }
+	    else
+	      {
+		while (spos < slen)
+		  {
+		    unichar	u = src[spos++];
+
+		    if (u < base)
+		      {
+			ptr[dpos++] = (unsigned char)u;
+		      }
+		    else
+		      {
+			ptr[dpos++] =  '?';
+		      }
 		  }
 	      }
 	  }
 	else
 	  {
-	    while (spos < slen)
+	    if (swapped == YES)
 	      {
-		unichar	u = src[spos++];
+		while (spos < slen)
+		  {
+		    unichar	u = src[spos++];
 
-		if (swapped == YES)
-		  {
 		    u = (((u & 0xff00) >> 8) + ((u & 0x00ff) << 8));
+		    if (u < base)
+		      {
+			ptr[dpos++] = (unsigned char)u;
+		      }
+		    else
+		      {
+			result = NO;
+			goto done;
+		      }
 		  }
-		if (dpos >= bsize)
+	      }
+	    else
+	      {
+		while (spos < slen)
 		  {
-		    GROW();
-		  }
-		if (u < base)
-		  {
-		    ptr[dpos++] = (unsigned char)u;
-		  }
-		else
-		  {
-		    result = NO;
-		    goto done;
+		    unichar	u = src[spos++];
+
+		    if (u < base)
+		      {
+			ptr[dpos++] = (unsigned char)u;
+		      }
+		    else
+		      {
+			result = NO;
+			goto done;
+		      }
 		  }
 	      }
 	  }