shrink2 now done

big mem saving, still a bit slower though

TODO

@@ -14,17 +14,15 @@
   we would then only need to demand the source a line at a time, and the
   intermediate image would be much smaller ... possible cache savings as well
 
-  added shrink2 for the new code
+  currently a bit slower than old shrink, much less mem though
 
-  current state on work machine, debug build
+  try:
 
-	  $ time vips shrink wtc.jpg x.v 10 10
+	* special path for RGB and mono
-	  memory: high-water mark 238.09 MB
-	  real	0m1.178s
-	  user	0m1.384s
-	  sys	0m0.064s
 
-  done h/v, get shrink2 done and bench against plain shrink
+	* for shrinkh, prepare input in batches?
+
+	* change loop indexing? don't use *p++ etc., it'll break vectorization
 
 
 - add bandjoinconst ... append a band (or bands?) from a constant

doc/gtk-doc.make

@@ -25,6 +25,7 @@ TARGET_DIR=$(HTML_DIR)/$(DOC_MODULE)
 
 SETUP_FILES = \
 	$(content_files)		\
+	$(expand_content_files)		\
 	$(DOC_MAIN_SGML_FILE)		\
 	$(DOC_MODULE)-sections.txt	\
 	$(DOC_MODULE)-overrides.txt
@@ -86,7 +87,7 @@ GTK_DOC_V_SETUP_0=@echo "  DOC   Preparing build";
 
 setup-build.stamp:
 	-$(GTK_DOC_V_SETUP)if test "$(abs_srcdir)" != "$(abs_builddir)" ; then \
-	    files=`echo $(SETUP_FILES) $(expand_content_files) $(DOC_MODULE).types`; \
+	    files=`echo $(SETUP_FILES) $(DOC_MODULE).types`; \
 	    if test "x$$files" != "x" ; then \
 	        for file in $$files ; do \
 	            destdir=`dirname $(abs_builddir)/$$file`; \
@@ -118,7 +119,7 @@ scan-build.stamp: setup-build.stamp $(HFILE_GLOB) $(CFILE_GLOB)
 	$(GTK_DOC_V_INTROSPECT)if grep -l '^..*$$' $(DOC_MODULE).types > /dev/null 2>&1 ; then \
 	    scanobj_options=""; \
 	    gtkdoc-scangobj 2>&1 --help | grep  >/dev/null "\-\-verbose"; \
-	    if test "$(?)" = "0"; then \
+	    if test "$$?" = "0"; then \
 	        if test "x$(V)" = "x1"; then \
 	            scanobj_options="--verbose"; \
 	        fi; \
@@ -162,17 +163,17 @@ GTK_DOC_V_XREF=$(GTK_DOC_V_XREF_$(V))
 GTK_DOC_V_XREF_=$(GTK_DOC_V_XREF_$(AM_DEFAULT_VERBOSITY))
 GTK_DOC_V_XREF_0=@echo "  DOC   Fixing cross-references";
 
-html-build.stamp: sgml.stamp $(DOC_MAIN_SGML_FILE) $(content_files)
+html-build.stamp: sgml.stamp $(DOC_MAIN_SGML_FILE) $(content_files) $(expand_content_files)
 	$(GTK_DOC_V_HTML)rm -rf html && mkdir html && \
 	mkhtml_options=""; \
 	gtkdoc-mkhtml 2>&1 --help | grep  >/dev/null "\-\-verbose"; \
-	if test "$(?)" = "0"; then \
+	if test "$$?" = "0"; then \
 	  if test "x$(V)" = "x1"; then \
 	    mkhtml_options="$$mkhtml_options --verbose"; \
 	  fi; \
 	fi; \
 	gtkdoc-mkhtml 2>&1 --help | grep  >/dev/null "\-\-path"; \
-	if test "$(?)" = "0"; then \
+	if test "$$?" = "0"; then \
 	  mkhtml_options="$$mkhtml_options --path=\"$(abs_srcdir)\""; \
 	fi; \
 	cd html && gtkdoc-mkhtml $$mkhtml_options $(MKHTML_OPTIONS) $(DOC_MODULE) ../$(DOC_MAIN_SGML_FILE)
@@ -194,11 +195,11 @@ GTK_DOC_V_PDF=$(GTK_DOC_V_PDF_$(V))
 GTK_DOC_V_PDF_=$(GTK_DOC_V_PDF_$(AM_DEFAULT_VERBOSITY))
 GTK_DOC_V_PDF_0=@echo "  DOC   Building PDF";
 
-pdf-build.stamp: sgml.stamp $(DOC_MAIN_SGML_FILE) $(content_files)
+pdf-build.stamp: sgml.stamp $(DOC_MAIN_SGML_FILE) $(content_files) $(expand_content_files)
 	$(GTK_DOC_V_PDF)rm -f $(DOC_MODULE).pdf && \
 	mkpdf_options=""; \
 	gtkdoc-mkpdf 2>&1 --help | grep  >/dev/null "\-\-verbose"; \
-	if test "$(?)" = "0"; then \
+	if test "$$?" = "0"; then \
 	  if test "x$(V)" = "x1"; then \
 	    mkpdf_options="$$mkpdf_options --verbose"; \
 	  fi; \
@@ -223,12 +224,15 @@ clean-local:
 	@if echo $(SCAN_OPTIONS) | grep -q "\-\-rebuild-types" ; then \
 	  rm -f $(DOC_MODULE).types; \
 	fi
+	@if echo $(SCAN_OPTIONS) | grep -q "\-\-rebuild-sections" ; then \
+	  rm -f $(DOC_MODULE)-sections.txt; \
+	fi
 
 distclean-local:
 	@rm -rf xml html $(REPORT_FILES) $(DOC_MODULE).pdf \
 	    $(DOC_MODULE)-decl-list.txt $(DOC_MODULE)-decl.txt
 	@if test "$(abs_srcdir)" != "$(abs_builddir)" ; then \
-	    rm -f $(SETUP_FILES) $(expand_content_files) $(DOC_MODULE).types; \
+	    rm -f $(SETUP_FILES) $(DOC_MODULE).types; \
 	fi
 
 maintainer-clean-local:

libvips/include/vips/resample.h

@@ -43,6 +43,8 @@ int vips_shrink( VipsImage *in, VipsImage **out,
 	__attribute__((sentinel));
 int vips_shrinkh( VipsImage *in, VipsImage **out, int xshrink, ... );
 int vips_shrinkv( VipsImage *in, VipsImage **out, int yshrink, ... );
+int vips_shrink2( VipsImage *in, VipsImage **out, 
+	int xshrink, int yshrink, ... );
 int vips_similarity( VipsImage *in, VipsImage **out, ... )
 	__attribute__((sentinel));
 int vips_resize( VipsImage *in, VipsImage **out, double scale, ... )

libvips/resample/Makefile.am

@@ -8,6 +8,7 @@ libresample_la_SOURCES = \
 	shrink.c \
 	shrinkh.c \
 	shrinkv.c \
+	shrink2.c \
 	interpolate.c \
 	transform.c \
 	bicubic.cpp \

libvips/resample/resample.c

@@ -114,6 +114,7 @@ vips_resample_operation_init( void )
 	extern GType vips_shrink_get_type( void ); 
 	extern GType vips_shrinkh_get_type( void ); 
 	extern GType vips_shrinkv_get_type( void ); 
+	extern GType vips_shrink2_get_type( void ); 
 	extern GType vips_quadratic_get_type( void ); 
 	extern GType vips_affine_get_type( void ); 
 	extern GType vips_similarity_get_type( void ); 
@@ -122,6 +123,7 @@ vips_resample_operation_init( void )
 	vips_shrink_get_type(); 
 	vips_shrinkh_get_type(); 
 	vips_shrinkv_get_type(); 
+	vips_shrink2_get_type(); 
 	vips_quadratic_get_type(); 
 	vips_affine_get_type(); 
 	vips_similarity_get_type(); 

libvips/resample/shrink2.c

@@ -69,19 +69,15 @@ vips_shrink2_build( VipsObject *object )
 	VipsResample *resample = VIPS_RESAMPLE( object );
 	VipsShrink2 *shrink = (VipsShrink2 *) object;
 	VipsImage **t = (VipsImage **) 
-		vips_object_local_array( object, 1 );
+		vips_object_local_array( object, 2 );
-
-	VipsImage *in;
 
 	if( VIPS_OBJECT_CLASS( vips_shrink2_parent_class )->build( object ) )
 		return( -1 );
 
-	in = resample->in; 
+	if( vips_shrinkh( resample->in, &t[0], shrink->xshrink, NULL ) ||
-
+		vips_shrinkv( t[0], &t[1], shrink->yshrink, NULL ) ||
-	if( vips_shrinkh( in, &t[0], shrink->xshrink, NULL ) ||
+		vips_image_write( t[1], resample->out ) )
-		vips_shrinkv( t[0], &t[1], shrink->yshrink, NULL ) )
 		return( -1 );
-	in = t[1];
 
 	return( 0 );
 }
@@ -102,7 +98,7 @@ vips_shrink2_class_init( VipsShrink2Class *class )
 	vobject_class->description = _( "shrink an image" );
 	vobject_class->build = vips_shrink2_build;
 
-	operation_class->flags = VIPS_OPERATION_SEQUENTIAL;
+	operation_class->flags = VIPS_OPERATION_SEQUENTIAL_UNBUFFERED;
 
 	VIPS_ARG_INT( class, "xshrink", 8, 
 		_( "Xshrink" ), 

libvips/resample/shrinkh.c

@@ -106,9 +106,9 @@ vips_shrinkh_start( VipsImage *out, void *a, void *b )
 /* Integer shrink. 
  */
 #define ISHRINK( TYPE ) { \
-	int *sum = (int *) seq->sum; \
+	int * restrict sum = (int *) seq->sum; \
-	TYPE *p = (TYPE *) in; \
+	TYPE * restrict p = (TYPE *) in; \
-	TYPE *q = (TYPE *) out; \
+	TYPE * restrict q = (TYPE *) out; \
 	\
 	for( x = 0; x < width; x++ ) { \
 		for( b = 0; b < bands; b++ ) \
@@ -127,9 +127,9 @@ vips_shrinkh_start( VipsImage *out, void *a, void *b )
 /* Float shrink. 
  */
 #define FSHRINK( TYPE ) { \
-	double *sum = (double *) seq->sum; \
+	double * restrict sum = (double *) seq->sum; \
-	TYPE *p = (TYPE *) in; \
+	TYPE * restrict p = (TYPE *) in; \
-	TYPE *q = (TYPE *) out; \
+	TYPE * restrict q = (TYPE *) out; \
 	\
 	for( x = 0; x < width; x++ ) { \
 		for( b = 0; b < bands; b++ ) \
@@ -324,7 +324,7 @@ vips_shrinkh_class_init( VipsShrinkhClass *class )
 	vobject_class->description = _( "shrink an image horizontally" );
 	vobject_class->build = vips_shrinkh_build;
 
-	operation_class->flags = VIPS_OPERATION_SEQUENTIAL;
+	operation_class->flags = VIPS_OPERATION_SEQUENTIAL_UNBUFFERED;
 
 	VIPS_ARG_INT( class, "xshrink", 8, 
 		_( "Xshrink" ), 

libvips/resample/shrinkv.c

@@ -144,8 +144,8 @@ vips_shrinkv_start( VipsImage *out, void *a, void *b )
 }
 
 #define ADD( ACC_TYPE, TYPE ) { \
-	ACC_TYPE *sum = (ACC_TYPE *) seq->sum; \
+	ACC_TYPE * restrict sum = (ACC_TYPE *) seq->sum; \
-	TYPE *p = (TYPE *) in; \
+	TYPE * restrict p = (TYPE *) in; \
 	\
 	for( x = 0; x < sz; x++ ) \
 		sum[x] += p[x]; \
@@ -196,8 +196,8 @@ vips_shrinkv_add_line( VipsShrinkv *shrink, VipsShrinkvSequence *seq,
 /* Integer average. 
  */
 #define IAVG( TYPE ) { \
-	int *sum = (int *) seq->sum; \
+	int * restrict sum = (int *) seq->sum; \
-	TYPE *q = (TYPE *) out; \
+	TYPE * restrict q = (TYPE *) out; \
 	\
 	for( x = 0; x < sz; x++ ) \
 		q[x] = (sum[x] + shrink->yshrink / 2) / shrink->yshrink; \
@@ -206,8 +206,8 @@ vips_shrinkv_add_line( VipsShrinkv *shrink, VipsShrinkvSequence *seq,
 /* Float average. 
  */
 #define FAVG( TYPE ) { \
-	double *sum = (double *) seq->sum; \
+	double * restrict sum = (double *) seq->sum; \
-	TYPE *q = (TYPE *) out; \
+	TYPE * restrict q = (TYPE *) out; \
 	\
 	for( x = 0; x < sz; x++ ) \
 		q[x] = sum[x] / shrink->yshrink; \
@@ -407,7 +407,7 @@ vips_shrinkv_class_init( VipsShrinkvClass *class )
 	vobject_class->description = _( "shrink an image vertically" );
 	vobject_class->build = vips_shrinkv_build;
 
-	operation_class->flags = VIPS_OPERATION_SEQUENTIAL;
+	operation_class->flags = VIPS_OPERATION_SEQUENTIAL_UNBUFFERED;
 
 	VIPS_ARG_INT( class, "yshrink", 9, 
 		_( "Yshrink" ),