From 83fc52c491cb863de2b01d5aaa62870ce8b8a7f5 Mon Sep 17 00:00:00 2001
From: John Cupitt <jcupitt@gmail.com>
Date: Tue, 30 Nov 2010 11:53:53 +0000
Subject: [PATCH] smaller norm factor for d->i mask

---
 ChangeLog                     |  5 ++++-
 TODO                          |  5 +++++
 libvips/convolution/im_conv.c | 16 +++++++++++++---
 libvips/mask/rw_mask.c        |  9 ++++++---
 4 files changed, 28 insertions(+), 7 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index d46e5290..14651d7e 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -39,7 +39,8 @@
 - mask gtk-doc done
 - add cfitsio dependancy
 - add FITS reader
-- land the vector branmch ... we have SSE erode/dilate/add/conv
+- land the vector branch and the orc dependancy ... we have SSE 
+  erode/dilate/add/conv
 - add IM_SWAP
 - dilate/erode do (!=0) on non-uchar images
 - add multipass Orc to im_conv(), 3.5x faster for 5x5 mask
@@ -51,6 +52,8 @@
 - oop, bool constants are always (int) now, so (^-1) works for unsigned types,
   thanks Nicolas
 - much lower memuse for im_cache() in complex pipelines
+- im_scale_dmask() normalises to 20, not 100 ... we hit the fast 
+  conv path more often
 
 12/5/10 started 7.22.2
 - the conditional image of ifthenelse can be any format, a (!=0) is added if
diff --git a/TODO b/TODO
index 9d306689..515e7ba6 100644
--- a/TODO
+++ b/TODO
@@ -1,6 +1,11 @@
+
+
 - im_conv()/im_morph() could have more than 10 programs? try 20 and see if we
   still have a speedup
 
+
+
+
 - fits >3d load
 
   fits save
diff --git a/libvips/convolution/im_conv.c b/libvips/convolution/im_conv.c
index 0d329db9..ba931df3 100644
--- a/libvips/convolution/im_conv.c
+++ b/libvips/convolution/im_conv.c
@@ -1037,14 +1037,24 @@ im_conv_raw( IMAGE *in, IMAGE *out, INTMASK *mask )
 		generate = convvec_gen;
 
 #ifdef DEBUG
-		printf( "im_conv_raw: using vector path\n" );
 #endif /*DEBUG*/
+		printf( "im_conv_raw: using vector path\n" );
 	}
-	else if( mask->xsize == 3 && mask->ysize == 3 )
+	else if( mask->xsize == 3 && mask->ysize == 3 ) {
 		generate = conv3x3_gen;
-	else
+
+#ifdef DEBUG
+#endif /*DEBUG*/
+		printf( "im_conv_raw: using 3x3 path\n" );
+	}
+	else {
 		generate = conv_gen;
 
+#ifdef DEBUG
+#endif /*DEBUG*/
+		printf( "im_conv_raw: using general path\n" );
+	}
+
 	/* Set demand hints. FATSTRIP is good for us, as THINSTRIP will cause
 	 * too many recalculations on overlaps.
 	 */
diff --git a/libvips/mask/rw_mask.c b/libvips/mask/rw_mask.c
index 49a1167e..02ffc5c4 100644
--- a/libvips/mask/rw_mask.c
+++ b/libvips/mask/rw_mask.c
@@ -49,6 +49,9 @@
  * 	- gtk-doc
  * 	- you can use commas to separate eader fields
  * 	- small cleanups
+ * 30/11/10
+ * 	- im_scale_dmask() normalises to 20, not 100 ... we hit the fast
+ * 	  conv path more often
  */
 
 /*
@@ -575,7 +578,7 @@ im_read_imask( const char *filename )
  * @in: mask to scale
  * @filename: filename for returned mask
  *
- * Scale the dmask to make an imask with a maximum value of 100.
+ * Scale the dmask to make an imask with a maximum value of 20.
  *
  * See also: im_norm_dmask().
  *
@@ -605,10 +608,10 @@ im_scale_dmask( DOUBLEMASK *in, const char *filename )
 		if( in->coeff[i] > maxval )
 			maxval = in->coeff[i];
 
-	/* Copy and scale, setting max to 100.
+	/* Copy and scale, setting max to 20.
 	 */
 	for( i = 0; i < size; i++ ) 
-		out->coeff[i] = IM_RINT( in->coeff[i] * 100.0 / maxval );
+		out->coeff[i] = IM_RINT( in->coeff[i] * 20.0 / maxval );
 	out->offset = in->offset;
 
 	/* Set the scale to match the adjustment to max.