now compiles with new add.c

compiles, but not tested

libvips/arithmetic/Makefile.am

@@ -27,6 +27,7 @@ libarithmetic_la_SOURCES = \
 	math.c \
 	arithmetic.c \
 	binary.c \
+	add.c \
 	power.c \
 	round.c 
 

libvips/arithmetic/add.c

@@ -34,7 +34,7 @@
 #ifdef HAVE_CONFIG_H
 #include <config.h>
 #endif /*HAVE_CONFIG_H*/
-#include <vips8/intl.h>
+#include <vips/intl.h>
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -73,33 +73,33 @@ typedef VipsBinaryClass VipsAddClass;
 G_DEFINE_TYPE( VipsAdd, vips_add, VIPS_TYPE_BINARY );
 
 #define LOOP( IN, OUT ) { \
-	IN *p1 = (IN *) in[0]; \
+	IN *p1 = (IN *) left; \
-	IN *p2 = (IN *) in[1]; \
+	IN *p2 = (IN *) right; \
 	OUT *q = (OUT *) out; \
 	\
 	for( x = 0; x < sz; x++ ) \
 		q[x] = p1[x] + p2[x]; \
 }
 
-static VipsVector *add_vectors[VIPS_FORMAT_LAST] = { NULL };
-
 static void
-add_buffer( PEL **in, PEL *out, int width, IMAGE *im )
+add_buffer( VipsBinary *binary, PEL *out, PEL *left, PEL *right, int width )
 {
+	VipsArithmeticClass *class = VIPS_ARITHMETIC_GET_CLASS( binary );
+	VipsImage *im = binary->left_processed;
+
 	/* Complex just doubles the size.
 	 */
 	const int sz = width * im->Bands * 
-		(vips_bandfmt_iscomplex( im->BandFmt ) ? 2 : 1);
+		(vips_band_format_iscomplex( im->BandFmt ) ? 2 : 1);
 
-	if( vips_vector_get_enabled() && 
+	VipsVector *v;
-		add_vectors[im->BandFmt] ) {
-		VipsVector *vector = add_vectors[im->BandFmt];
 
+	if( (v = vips_arithmetic_get_vector( class, im->BandFmt )) ) {
 		VipsExecutor ex;
 
-		vips_executor_set_program( &ex, vector, sz );
+		vips_executor_set_program( &ex, v, sz );
-		vips_executor_set_array( &ex, vector->s[0], in[0] );
+		vips_executor_set_array( &ex, v->s[0], left );
-		vips_executor_set_array( &ex, vector->s[1], in[1] );
+		vips_executor_set_array( &ex, v->s[1], right );
 		vips_executor_set_destination( &ex, out );
 
 		vips_executor_run( &ex );
@@ -151,47 +151,6 @@ add_buffer( PEL **in, PEL *out, int width, IMAGE *im )
 #define D IM_BANDFMT_DOUBLE
 #define DX IM_BANDFMT_DPCOMPLEX
 
-VipsVector *
-im__init_program( VipsVector *vectors[IM_BANDFMT_LAST], 
-	VipsBandFmt format_table[IM_BANDFMT_LAST], VipsBandFmt fmt )
-{
-	int isize = im__sizeof_bandfmt[fmt];
-	int osize = im__sizeof_bandfmt[format_table[fmt]];
-
-	VipsVector *v;
-
-	v = vips_vector_new( "binary arith", osize );
-
-	vips_vector_source_name( v, "s1", isize );
-	vips_vector_source_name( v, "s2", isize );
-	vips_vector_temporary( v, "t1", osize );
-	vips_vector_temporary( v, "t2", osize );
-
-	vectors[fmt] = v;
-
-	return( v );
-}
-
-void
-im__compile_programs( VipsVector *vectors[IM_BANDFMT_LAST] )
-{
-	int fmt;
-
-	for( fmt = 0; fmt < IM_BANDFMT_LAST; fmt++ ) {
-		if( vectors[fmt] &&
-			!vips_vector_compile( vectors[fmt] ) )
-			IM_FREEF( vips_vector_free, vectors[fmt] );
-	}
-
-#ifdef DEBUG
-	printf( "im__compile_programs: " );
-	for( fmt = 0; fmt < IM_BANDFMT_LAST; fmt++ ) 
-		if( vectors[fmt] )
-			printf( "%s ", im_BandFmt2char( fmt ) );
-	printf( "\n" );
-#endif /*DEBUG*/
-}
-
 /* Type promotion for addition. Sign and value preserving. Make sure these
  * match the case statement in add_buffer() above.
  */
@@ -201,32 +160,30 @@ static int bandfmt_add[10] = {
 };
 
 static void
-build_programs( void )
+vips_add_class_init( VipsAddClass *class )
 {
-	static gboolean done = FALSE;
+	VipsArithmeticClass *aclass = VIPS_ARITHMETIC_CLASS( class );
-
+	VipsBinaryClass *bclass = VIPS_BINARY_CLASS( class );
 	VipsVector *v;
 
-	if( done )
+	vips_arithmetic_set_format_table( aclass, bandfmt_add );
-		return;
-	done = TRUE;
 
-	v = im__init_program( add_vectors, bandfmt_add, IM_BANDFMT_UCHAR );
+	v = vips_arithmetic_get_program( aclass, VIPS_FORMAT_UCHAR );
 	vips_vector_asm2( v, "convubw", "t1", "s1" );
 	vips_vector_asm2( v, "convubw", "t2", "s2" );
 	vips_vector_asm3( v, "addw", "d1", "t1", "t2" ); 
 
-	v = im__init_program( add_vectors, bandfmt_add, IM_BANDFMT_CHAR );
+	v = vips_arithmetic_get_program( aclass, VIPS_FORMAT_CHAR );
 	vips_vector_asm2( v, "convsbw", "t1", "s1" );
 	vips_vector_asm2( v, "convsbw", "t2", "s2" );
 	vips_vector_asm3( v, "addw", "d1", "t1", "t2" ); 
 
-	v = im__init_program( add_vectors, bandfmt_add, IM_BANDFMT_USHORT );
+	v = vips_arithmetic_get_program( aclass, VIPS_FORMAT_USHORT );
 	vips_vector_asm2( v, "convuwl", "t1", "s1" );
 	vips_vector_asm2( v, "convuwl", "t2", "s2" );
 	vips_vector_asm3( v, "addl", "d1", "t1", "t2" );
 
-	v = im__init_program( add_vectors, bandfmt_add, IM_BANDFMT_SHORT );
+	v = vips_arithmetic_get_program( aclass, VIPS_FORMAT_SHORT );
 	vips_vector_asm2( v, "convswl", "t1", "s1" );
 	vips_vector_asm2( v, "convswl", "t2", "s2" );
 	vips_vector_asm3( v, "addl", "d1", "t1", "t2" );
@@ -237,14 +194,20 @@ build_programs( void )
 
 	   float/double/complex are not handled well
 
-	v = im__init_program( add_vectors, IM_BANDFMT_UINT );
+	v = vips_arithmetic_get_vector( aclass, VIPS_FORMAT_UINT );
 	vips_vector_asm3( v, "addl", "d1", "s1", "s2" );
 
-	v = im__init_program( add_vectors, IM_BANDFMT_INT );
+	v = vips_arithmetic_get_vector( aclass, VIPS_FORMAT_INT );
 	vips_vector_asm3( v, "addl", "d1", "s1", "s2" );
 
 	 */
 
-	im__compile_programs( add_vectors );
+	vips_arithmetic_compile( aclass );
+
+	bclass->process_line = add_buffer;
 }
 
+static void
+vips_add_init( VipsAdd *add )
+{
+}

libvips/arithmetic/arithmetic.c

@@ -43,7 +43,7 @@
 #ifdef HAVE_CONFIG_H
 #include <config.h>
 #endif /*HAVE_CONFIG_H*/
-#include <vips8/intl.h>
+#include <vips/intl.h>
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -69,8 +69,6 @@ G_DEFINE_ABSTRACT_TYPE( VipsArithmetic, vips_arithmetic, VIPS_TYPE_OPERATION );
 static int
 vips_arithmetic_build( VipsObject *object )
 {
-	VipsArithmetic *arithmetic = VIPS_ARITHMETIC (object);
-
 	if( VIPS_OBJECT_CLASS( vips_arithmetic_parent_class )->build( object ) )
 		return( -1 );
 
@@ -110,6 +108,85 @@ vips_arithmetic_init( VipsArithmetic *arithmetic )
 {
 }
 
+void 
+vips_arithmetic_set_format_table( VipsArithmeticClass *class, 
+	VipsBandFormat *format_table )
+{
+	int i;
+
+	g_assert( !class->format_table );
+
+	class->format_table = format_table;
+
+	for( i = 0; i < VIPS_FORMAT_LAST; i++ ) {
+		int isize = vips_format_sizeof( i );
+		int osize = vips_format_sizeof( (int) format_table[i] );
+
+		VipsVector *v;
+
+		v = vips_vector_new( "arithmetic", osize );
+
+		vips_vector_source_name( v, "s1", isize );
+		vips_vector_source_name( v, "s2", isize );
+		vips_vector_temporary( v, "t1", osize );
+		vips_vector_temporary( v, "t2", osize );
+
+		class->vectors[i] = v;
+	}
+}
+
+/* Get the stub for this program ... use _get_vector() to get the compiled
+ * code.
+ */
+VipsVector *
+vips_arithmetic_get_program( VipsArithmeticClass *class, VipsBandFormat fmt )
+{
+	g_assert( (int) fmt >= 0 && (int) fmt < VIPS_FORMAT_LAST );
+	g_assert( !class->vector_program[fmt] );
+
+	class->vector_program[fmt] = TRUE;
+
+	return( class->vectors[fmt] );
+}
+
+/* Get the compiled code for this type, if available.
+ */
+VipsVector *
+vips_arithmetic_get_vector( VipsArithmeticClass *class, VipsBandFormat fmt )
+{
+	g_assert( fmt >= 0 && fmt < VIPS_FORMAT_LAST );
+
+	if( !vips_vector_get_enabled() ||
+		!class->vector_program[fmt] )
+		return( NULL );
+
+	return( class->vectors[fmt] );
+}
+
+void
+vips_arithmetic_compile( VipsArithmeticClass *class ) 
+{
+	int i;
+
+	g_assert( class->format_table );
+
+	for( i = 0; i < VIPS_FORMAT_LAST; i++ ) 
+		if( class->vector_program[i] &&
+			!vips_vector_compile( class->vectors[i] ) )
+			/* If compilation fails, turn off the vector for this
+			 * type.
+			 */
+			class->vector_program[i] = FALSE;
+
+#ifdef DEBUG
+	printf( "vips_arithmetic_compile: " );
+	for( i = 0; i < IM_BANDFMT_LAST; i++ ) 
+		if( class->vector_program[i] )
+			printf( "%s ", VIPS_ENUM_NICK( VIPS_TYPE_FORMAT, i ) );
+	printf( "\n" );
+#endif /*DEBUG*/
+}
+
 /* Called from iofuncs to init all operations in this dir. Use a plugin system
  * instead?
  */
@@ -120,3 +197,4 @@ vips_arithmetic_operation_init( void )
 
 	vips_add_get_type();
 }
+

libvips/arithmetic/arithmetic.h

@@ -34,6 +34,8 @@
 extern "C" {
 #endif /*__cplusplus*/
 
+#include <vips/vector.h>
+
 #define VIPS_TYPE_ARITHMETIC (vips_arithmetic_get_type())
 #define VIPS_ARITHMETIC( obj ) \
 	(G_TYPE_CHECK_INSTANCE_CAST( (obj), \
@@ -56,19 +58,35 @@ typedef struct _VipsArithmetic {
 	 */
 	VipsImage *output;
 
-	/* For each input format, what output format. Used for arithmetic
-	 * too, since we cast inputs to match.
-	 */
-	VipsFormat *cast_table;
 } VipsArithmetic;
 
 typedef struct _VipsArithmeticClass {
 	VipsOperationClass parent_class;
 
+	/* For each input format, what output format. Used for arithmetic
+	 * too, since we cast inputs to match.
+	 */
+	VipsBandFormat *format_table;
+
+	/* A vector program for each input type.
+	 */
+	VipsVector *vectors[VIPS_FORMAT_LAST];
+
+	/* ... and if we've set a program for this format.
+	 */
+	gboolean vector_program[VIPS_FORMAT_LAST];
 } VipsArithmeticClass;
 
 GType vips_arithmetic_get_type( void );
 
+void vips_arithmetic_set_format_table( VipsArithmeticClass *klass, 
+	VipsBandFormat *format_table );
+VipsVector *vips_arithmetic_get_vector( VipsArithmeticClass *klass, 
+	VipsBandFormat fmt );
+void vips_arithmetic_compile( VipsArithmeticClass *klass ); 
+VipsVector *vips_arithmetic_get_program( VipsArithmeticClass *klass, 
+	VipsBandFormat fmt );
+
 #ifdef __cplusplus
 }
 #endif /*__cplusplus*/

libvips/arithmetic/binary.c

@@ -34,7 +34,7 @@
 #ifdef HAVE_CONFIG_H
 #include <config.h>
 #endif /*HAVE_CONFIG_H*/
-#include <vips8/intl.h>
+#include <vips/intl.h>
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -75,7 +75,7 @@ G_DEFINE_ABSTRACT_TYPE( VipsBinary, vips_binary, VIPS_TYPE_ARITHMETIC );
 /* For two integer types, the "largest", ie. one which can represent the
  * full range of both.
  */
-static VipsFormat format_largest[6][6] = {
+static VipsBandFormat format_largest[6][6] = {
         /* UC  C   US  S   UI  I */
 /* UC */ { UC, S,  US, S,  UI, I },
 /* C */  { S,  C,  I,  S,  I,  I },
@@ -87,11 +87,11 @@ static VipsFormat format_largest[6][6] = {
 
 /* For two formats, find one which can represent the full range of both.
  */
-static VipsFormat
+static VipsBandFormat
-vips_format_common( VipsFormat a, VipsFormat b )
+vips_format_common( VipsBandFormat a, VipsBandFormat b )
 {
-	if( vips_format_iscomplex( a ) || 
+	if( vips_band_format_iscomplex( a ) || 
-		vips_format_iscomplex( b ) ) {
+		vips_band_format_iscomplex( b ) ) {
 		if( a == VIPS_FORMAT_DPCOMPLEX || 
 			b == VIPS_FORMAT_DPCOMPLEX )
 			return( VIPS_FORMAT_DPCOMPLEX );
@@ -99,8 +99,8 @@ vips_format_common( VipsFormat a, VipsFormat b )
 			return( VIPS_FORMAT_COMPLEX );
 
 	}
-	else if( vips_format_isfloat( a ) || 
+	else if( vips_band_format_isfloat( a ) || 
-		vips_format_isfloat( b ) ) {
+		vips_band_format_isfloat( b ) ) {
 		if( a == VIPS_FORMAT_DOUBLE || 
 			b == VIPS_FORMAT_DOUBLE )
 			return( VIPS_FORMAT_DOUBLE );
@@ -115,13 +115,13 @@ int
 vips__formatalike_vec( VipsImage **in, VipsImage **out, int n )
 {
 	int i;
-	VipsFormat format;
+	VipsBandFormat format;
 
 	g_assert( n >= 1 );
 
 	format = in[0]->BandFmt;
 	for( i = 1; i < n; i++ )
-		fmt = vips_format_common( format, in[i]->BandFmt );
+		format = vips_format_common( format, in[i]->BandFmt );
 
 	for( i = 0; i < n; i++ )
 		if( im_clip2fmt( in[i], out[i], format ) )
@@ -219,6 +219,7 @@ vips_binary_process_region( VipsRegion *or, void *seq, void *a, void *b )
 {
 	VipsRegion **ir = (VipsRegion **) seq;
 	VipsBinary *binary = VIPS_BINARY( b ); 
+	VipsBinaryClass *class = VIPS_BINARY_GET_CLASS( binary ); 
 
 	PEL *p[MAX_INPUT_IMAGES], *q;
 	int i, y;
@@ -237,7 +238,7 @@ vips_binary_process_region( VipsRegion *or, void *seq, void *a, void *b )
 	for( y = 0; y < or->valid.height; y++ ) {
 		/* Bizarre double-cast stops a bogus gcc 4.1 compiler warning.
 		 */
-		binary->process_line( binary, q, p[0], p[1], or->valid.width );
+		class->process_line( binary, q, p[0], p[1], or->valid.width );
 
 		for( i = 0; ir[i]; i++ )
 			p[i] += VIPS_REGION_LSKIP( ir[i] );
@@ -253,6 +254,7 @@ vips_binary_build( VipsObject *object )
 	VipsObjectClass *class = VIPS_OBJECT_GET_CLASS( object );
 	const char *domain = class->description;
 	VipsArithmetic *arithmetic = VIPS_ARITHMETIC( object );
+	VipsArithmeticClass *aclass = VIPS_ARITHMETIC_GET_CLASS( arithmetic ); 
 	VipsBinary *binary = VIPS_BINARY( object );
 
 	VipsImage *t[5];
@@ -284,15 +286,15 @@ vips_binary_build( VipsObject *object )
 	/* Hint demand style. Being a buffer processor, we are happiest with
 	 * thin strips.
 	 */
-        if( vips_demand_hint_array( binary->output, 
+        if( vips_demand_hint_array( arithmetic->output, 
 		VIPS_DEMAND_STYLE_THINSTRIP, t + 2 ) ||
-		vips_image_copy_fields_array( binary->output, t + 2 ) )
+		vips_image_copy_fields_array( arithmetic->output, t + 2 ) )
 		return( -1 );
 
-	binary->output->Bands = t[2]->Bands;
+	arithmetic->output->Bands = t[2]->Bands;
-	binary->output->BandFmt = arithmetic->cast_table[t[2]->BandFmt];
+	arithmetic->output->BandFmt = aclass->format_table[t[2]->BandFmt];
 
-	if( vips_image_generate( binary->output,
+	if( vips_image_generate( arithmetic->output,
 		vips_start_many, vips_binary_process_region, 
 			vips_stop_many, 
 		t + 2, binary ) )

libvips/arithmetic/binary.h

@@ -46,8 +46,9 @@ extern "C" {
 #define VIPS_BINARY_GET_CLASS( obj ) \
 	(G_TYPE_INSTANCE_GET_CLASS( (obj), VIPS_TYPE_BINARY, VipsBinaryClass ))
 
-typedef void (*VipsBinaryProcessFn)( VipsBinary *binary, 
+struct _VipsBinary;
-	void *out, void *left, void *right, int width );
+typedef void (*VipsBinaryProcessFn)( struct _VipsBinary *binary, 
+	PEL *out, PEL *left, PEL *right, int width );
 
 typedef struct _VipsBinary {
 	VipsArithmetic parent_instance;
@@ -66,9 +67,8 @@ typedef struct _VipsBinary {
 	VipsImage *left_processed;
 	VipsImage *right_processed;
 
-	/* The line processor, plus some client data.
+	/* Some client data for the line processor, if it wants it.
 	 */
-	VipsBinaryProcessFn process_line;
 	void *a;
 	void *b;
 } VipsBinary;
@@ -76,6 +76,10 @@ typedef struct _VipsBinary {
 typedef struct _VipsBinaryClass {
 	VipsArithmeticClass parent_class;
 
+	/* The line processor.
+	 */
+	VipsBinaryProcessFn process_line;
+
 } VipsBinaryClass;
 
 GType vips_binary_get_type( void );