stuff

ChangeLog

@@ -13,6 +13,10 @@
 - performance improvements for morphology ops, esp. when zooming out
 - oop, im_render() was broken for mask == NULL
 - better support for multiple Python installs (thanks Jay)
+- better IM_SETSTR() stops some warnings
+- im_histcum() works for signed histograms
+- better rounding for im_conv(), im_convsep()
+- tiny speedup for im_conv()
 
 25/1/08 started 7.14.0
 - bump all version numbers for new stable

TODO

@@ -1,3 +1,10 @@
+- try
+
+	libsrc/convolution$ grep -l offsets *.c
+
+  could we do the don't calc offsets thing unless bpl; changes thing in more
+  places?
+
 - should check for gettext in configure? see
 
 https://sourceforge.net/tracker/index.php?func=detail&aid=1836080&group_id=100050&atid=626186

benchmark/benchmarkn.sh

@@ -27,7 +27,7 @@ echo " by" `header -f Ysize temp.v` "pixels"
 echo "starting benchmark ..."
 echo "chain=$chain"
 
-for cpus in 1 2 3 ; do
+for cpus in 1 2 ; do
   export IM_CONCURRENCY=$cpus
 
   echo IM_CONCURRENCY=$IM_CONCURRENCY

include/vips/util.h

@@ -161,10 +161,10 @@ extern "C" {
 } while( 0 )
 #define IM_FREE( A ) IM_FREEF( im_free, A )
 #define IM_SETSTR( S, V ) do { \
-        if( (S) != (V) ) { \
-                if( !(S) || !(V) || strcmp( (S), (V) ) != 0 ) { \
-                        const char *sst = (V); \
-                        \
+        const char *sst = (V); \
+	\
+        if( (S) != sst ) { \
+                if( !(S) || !sst || strcmp( (S), sst ) != 0 ) { \
                         IM_FREE( S ); \
                         if( sst ) \
                                 (S) = im_strdup( NULL, sst ); \

libsrc/convolution/im_conv.c

@@ -61,6 +61,9 @@
  * 	- simpler inner loop avoids gcc4 bug 
  * 7/11/07
  * 	- new evalstart/end callbacks
+ * 12/5/08
+ * 	- int rounding was +1 too much, argh
+ * 	- only rebuild the buffer offsets if bpl changes
  */
 
 /*
@@ -200,6 +203,8 @@ typedef struct {
 
 	int underflow;		/* Underflow/overflow counts */
 	int overflow;
+
+	int last_bpl;		/* Avoid recalcing offsets, if we can */
 } ConvSequence;
 
 /* Free a sequence value.
@@ -239,6 +244,7 @@ conv_start( IMAGE *out, void *a, void *b )
 	seq->pts = NULL;
 	seq->underflow = 0;
 	seq->overflow = 0;
+	seq->last_bpl = -1;
 
 	/* Attach region and arrays.
 	 */
@@ -303,7 +309,11 @@ conv_gen( REGION *or, void *vseq, void *a, void *b )
 	Conv *conv = (Conv *) b;
 	REGION *ir = seq->ir;
 	INTMASK *mask = conv->mask;
-	int rounding = (mask->scale + 1)/2;
+
+	/* You might think this should be (scale+1)/2, but then we'd be adding
+	 * one for scale == 1.
+	 */
+	int rounding = mask->scale / 2;
 
 	Rect *r = &or->valid;
 	Rect s;
@@ -323,15 +333,21 @@ conv_gen( REGION *or, void *vseq, void *a, void *b )
 	if( im_prepare( ir, &s ) )
 		return( -1 );
 
-        /* Fill offset array.
-         */
-	z = 0;
-        for( i = 0, y = 0; y < mask->ysize; y++ )
-                for( x = 0; x < mask->xsize; x++, i++ )
-                        if( mask->coeff[i] )
-                                seq->offsets[z++] = 
-					IM_REGION_ADDR( ir, x + le, y + to ) -
-                                        IM_REGION_ADDR( ir, le, to );
+        /* Fill offset array. Only do this if the bpl has changed since the 
+	 * previous im_prepare().
+	 */
+	if( seq->last_bpl != IM_REGION_LSKIP( ir ) ) {
+		seq->last_bpl = IM_REGION_LSKIP( ir );
+
+		z = 0;
+		for( i = 0, y = 0; y < mask->ysize; y++ )
+			for( x = 0; x < mask->xsize; x++, i++ )
+				if( mask->coeff[i] )
+					seq->offsets[z++] = 
+						IM_REGION_ADDR( ir, 
+							x + le, y + to ) -
+						IM_REGION_ADDR( ir, le, to );
+	}
 
 	for( y = to; y < bo; y++ ) { 
 		/* Init pts for this line of PELs.

libsrc/convolution/im_convsep.c

@@ -27,6 +27,8 @@
  * 21/4/04
  *	- scale down int convolves at 1/2 way mark, much less likely to integer
  *	  overflow on intermediates
+ * 12/5/08
+ * 	- int rounding was +1 too much, argh
  */
 
 /*
@@ -304,7 +306,12 @@ conv_gen( REGION *or, void *vseq, void *a, void *b )
 	Conv *conv = (Conv *) b;
 	REGION *ir = seq->ir;
 	INTMASK *mask = conv->mask;
-	int rounding = (mask->scale + 1)/2;
+
+	/* You might think this should be (scale+1)/2, but then we'd be adding
+	 * one for scale == 1.
+	 */
+	int rounding = mask->scale / 2;
+
 	int bands = in->Bands;
 	int *coeff = conv->mask->coeff; 
 

libsrc/histograms_lut/im_histeq.c

@@ -25,6 +25,8 @@
  * 	- neater im_histnorm()
  * 23/7/07
  * 	- eek, off by 1 for more than 1 band hists
+ * 12/5/08
+ * 	- histcum works for signed hists now as well
  */
 
 /*
@@ -104,14 +106,14 @@ im_histcum( IMAGE *in, IMAGE *out )
 	if( im_incheck( in ) )
 		return( -1 );
 
-	/* int types -> uint, float/double stay as they are.
-	 */
 	if( im_cp_desc( out, in ) )
 		return( -1 );
 	out->Xsize = px;
 	out->Ysize = 1;
-	if( im_isint( in ) )
+	if( im_isuint( in ) )
 		out->BandFmt = IM_BANDFMT_UINT;
+	else if( im_isint( in ) )
+		out->BandFmt = IM_BANDFMT_INT;
 	out->Bbits = im_bits_of_fmt( out->BandFmt );
 
 	if( !(outbuf = im_malloc( out, IM_IMAGE_SIZEOF_LINE( out ))) )
@@ -119,15 +121,15 @@ im_histcum( IMAGE *in, IMAGE *out )
 
         switch( in->BandFmt ) {
         case IM_BANDFMT_CHAR: 		
-		ACCUMULATE( signed char, unsigned int ); break; 
+		ACCUMULATE( signed char, signed int ); break; 
         case IM_BANDFMT_UCHAR: 		
 		ACCUMULATE( unsigned char, unsigned int ); break; 
         case IM_BANDFMT_SHORT: 		
-		ACCUMULATE( signed short, unsigned int ); break; 
+		ACCUMULATE( signed short, signed int ); break; 
         case IM_BANDFMT_USHORT: 	
 		ACCUMULATE( unsigned short, unsigned int ); break; 
         case IM_BANDFMT_INT: 		
-		ACCUMULATE( signed int, unsigned int ); break; 
+		ACCUMULATE( signed int, signed int ); break; 
         case IM_BANDFMT_UINT: 		
 		ACCUMULATE( unsigned int, unsigned int ); break;