[xserver-commit] xserver/fb fbblt.c,1.9,1.10 fbpict.c,1.22,1.23

[Jaymz Julian]

Committed by: jaymz

Update of /cvs/xserver/xserver/fb
In directory pdx:/tmp/cvs-serv22511/fb

Modified Files:
	fbblt.c fbpict.c 
Log Message:

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)



Index: fbblt.c
===================================================================
RCS file: /cvs/xserver/xserver/fb/fbblt.c,v
retrieving revision 1.9
retrieving revision 1.10
diff -u -d -r1.9 -r1.10
--- fbblt.c	11 Sep 2003 05:12:50 -0000	1.9
+++ fbblt.c	20 Dec 2003 14:36:41 -0000	1.10
@@ -40,6 +40,7 @@
     } \
 }
 
+
 void
 fbBlt (FbBits   *srcLine,
        FbStride	srcStride,
@@ -66,6 +67,23 @@
     int	    n, nmiddle;
     Bool    destInvarient;
     int	    startbyte, endbyte;
+
+    if((pm==FB_ALLONES) && (alu==GXcopy) && !reverse && !upsidedown)
+    {
+	CARD8 *isrc=(CARD8 *)srcLine;
+	CARD8 *idst=(CARD8 *)dstLine;
+	width>>=3;
+	int sstride=srcStride*sizeof(FbBits);
+	int dstride=dstStride*sizeof(FbBits);
+	isrc+=(srcX>>3);
+	idst+=(dstX>>3);
+	int j;
+	for(j=0;j<height;j++)
+		memcpy(idst+j*dstride, isrc+j*sstride, width);
+	
+	return;
+    }
+    
     FbDeclareMergeRop ();
 
 #ifdef FB_24BIT

Index: fbpict.c
===================================================================
RCS file: /cvs/xserver/xserver/fb/fbpict.c,v
retrieving revision 1.22
retrieving revision 1.23
diff -u -d -r1.22 -r1.23
--- fbpict.c	19 Dec 2003 13:08:01 -0000	1.22
+++ fbpict.c	20 Dec 2003 14:36:41 -0000	1.23
@@ -1131,7 +1131,7 @@
 
 		// are xSrc and xDst at the same alignment?  if not, we need to be complicated :)
 		//if(0==0)
-		if( (((xSrc*3)&3)!=((xDst*3)&3)) || (srcStride&3)!=0 || (dstStride&3)!=0)
+		if( (((xSrc*3)&3)!=((xDst*3)&3)) || ((srcStride&3)!=(dstStride&3)))
 		{
 			while (height--)
 			{
@@ -1283,32 +1283,40 @@
 {
     FbBits	*dst;
     FbBits	*src;
-    FbStride	dstStride, srcStride;
-    int		srcXoff, srcYoff;
-    int		dstXoff, dstYoff;
-    int		srcBpp;
-    int		dstBpp;
+	CARD8 *isrc;
+	CARD8 *idst;
+    int	dstStride, srcStride;
+    int	srcXoff, srcYoff;
+    int	dstXoff, dstYoff;
+    int	srcBpp;
+    int	dstBpp;
     // these need to be signed now!
-    int 	iwidth=width;
-    int 	iheight=height;
-    Bool	reverse = FALSE;
-    Bool	upsidedown = FALSE;
+    int	iwidth=width;
+    int	iheight=height;
+	int	bytesPerPixel;
 	int initialWidth=width;
 	int initialX=xDst;
 
-	// FIXME: this is possibly the worst piece of code I've ever written.
-	// My main objection to it, is that it is incrfedibly slow in a few cases, due to the
-	// call-per-repeat structure of it - the *correct* solution is to implement
-	// repeat into fbBlt(), but that's a nontrivial job, and it's far more 
-	// important to get the "requireRepeat" stuff implented functionally
-	// first, *then* make it fast.
+	// FIXME: this is possibly the second worst piece of code I've ever written.
+	// (the worst being the previous imlementation of this)
 	//  -- jj
+	
 	Bool srcRepeat=pSrc->repeat;
 	CARD32 srcHeight=pSrc->pDrawable->height;
 	CARD32 srcWidth=pSrc->pDrawable->width;
 
 	fbGetDrawable(pSrc->pDrawable,src,srcStride,srcBpp,srcXoff,srcYoff);
 	fbGetDrawable(pDst->pDrawable,dst,dstStride,dstBpp,dstXoff,dstYoff);
+	
+	src+=(srcXoff+(srcYoff*srcStride));
+	dst+=(dstXoff+(dstYoff*dstStride));
+	
+    isrc=(CARD8 *)src;
+	idst=(CARD8 *)dst;
+
+	bytesPerPixel=(srcBpp>>3);
+	srcStride*=sizeof(FbBits);
+	dstStride*=sizeof(FbBits);
 
 	if(srcRepeat)
 	{
@@ -1326,26 +1334,14 @@
 		while(iwidth>0)
 		{
 			int wwidth=iwidth;
+			int j;
 			if(wwidth>(srcWidth-xSrc))
 				wwidth=(srcWidth-xSrc);
 
-			fbBlt (src + (ySrc + srcYoff) * srcStride,
-					srcStride,
-					(xSrc + srcXoff) * srcBpp,
-
-					dst + (yDst + dstYoff) * dstStride,
-					dstStride,
-					(xDst + dstXoff) * dstBpp,
-
-					(wwidth) * dstBpp,
-					(wheight),
-
-					GXcopy,
-					FB_ALLONES,
-					dstBpp,
-
-					reverse,
-					upsidedown);
+			for(j=0;j<wheight;j++)
+				memcpy( idst + (yDst + j) * dstStride + ((xDst * bytesPerPixel)), 
+						isrc + (ySrc + j) * srcStride + ((xSrc * bytesPerPixel)), 
+						(wwidth*bytesPerPixel));
 			if(!srcRepeat)
 				iwidth=0;
 			else