[xserver-commit] xserver ChangeLog,3.77,3.78

[Jaymz Julian]

Committed by: jaymz

Update of /cvs/xserver/xserver
In directory pdx:/tmp/cvs-serv14894

Modified Files:
	ChangeLog 
Log Message:
Fix fompositeSrcSrc_nxn()'s repeating code to reset the to the start of the
pixmap on repeat - d'oh!



Index: ChangeLog
===================================================================
RCS file: /cvs/xserver/xserver/ChangeLog,v
retrieving revision 3.77
retrieving revision 3.78
diff -u -d -r3.77 -r3.78
--- ChangeLog	8 Dec 2003 01:55:06 -0000	3.77
+++ ChangeLog	21 Dec 2003 02:51:53 -0000	3.78
@@ -1,3 +1,56 @@
+2003-12-21  Jaymz Julian <jaymz@artificial-stupidity.net>
+	* fb/fbpict.c: 
+	Fix the fbCompositeSrcSrc_nxn() looping code to loop to zero, not 
+	to xSrc.  This should fix the "my pixmap is too small" issue some 
+	where experiencing.
+
+	Updated the changelog for the stuff I forgot to add for the past few
+	days.
+
+2003-12-20  Jaymz Julian <jaymz@artifical-stupidity.net>
+	* fb/fbpict.c:
+	* fb/fbblt.c:
+	Add a special case to the fbBlt for the very common case of just blitting
+	directly (GXcopy) and not applying any ops.  Profiles show this to be
+	significnatly faster, although I'd like to one day change it to be able
+	to choose the fastest memcpy() for this system ala xine when it	benchmarks
+	on startup (does it still do that?), but i'm reluctant to do that right now, 
+	since it means that it can't be inlined, and there'd be a function call
+	per line... eew!  (the correct solution is probably 5 blitters or something,
+	heh)
+
+	... which was rolled back five minutes later, due to being off by one
+	when compiled with optimizations turned on.
+
+2003-12-19  Jaymz Julian <jaymz@artificial-stupidity.net>
+	* fb/fbpict.c:
+	Changed fbCompositeTrans_0565xnx0565 to have an aligned path similar
+	to the 24bit version of same, as well as making it do it's alpha blend 
+	in a single pass for 2x16bit values rather than a seperate pass for the
+	high/low 16 bits
+	
+	Optimize fbCompositeTrans_0888xnx0888()'s unaligned path by doing the
+	unpacking on source rather than dest (which reuiqres half the amount
+	of variables to do so), which also has the effect of making the code 
+	look slightly less on crack (it isn't, but it looks less on crack :-p)
+
+2003-12-18  Jaymz Julian <jaymz@artificial-stupidity.net>
+	* configure.ac: 
+	Check if unaligned word accesses work as you'd expect
+	
+	* fb/fbcomposite.c: 
+	If they do, make fetch_r8g8b8() use them
+	
+	* fb/fbpict.h:
+	* fb/fbpict.c: 
+	Optimize 24bit (and 16bit colour less so) mode significantly, by 
+	making all reads aligned if easily possible, and also using a 
+	different composite function (see fastcombine32).
+
+	Make fbCompositeSrcSrc_nxn() know about repeating sources, and start
+	a whitelist check for functions that know this so they can be 
+	incrementally implented.
+
 2003-12-07  Eric Anholt  <anholt@FreeBSD.org>
 
 	* configure.ac: