im_generate(3) im_allocate_input_array,


#include <vips/vips.h>

void *im_start_one( out, in )
IMAGE *out, *in;

int im_stop_one( reg )
REGION *reg;

IMAGE **im_allocate_input_array( IMAGE *out, ... )

void *im_start_many( out, in )
IMAGE *out, **in;

int im_stop_many( REGION **out )
REGION **out;

int im_generate( im,
    start_fn, gen_fn, stop_fn, void *a, void *b )
IMAGE *im;
void *(*start_fn)();
int (*gen_fn)();
int (*stop_fn)();
void *a, void *b;

where, typically,

void *start_fn( im, a, b )
IMAGE *im;
void *a, *b;

int gen_fn( or, seq, a, b )
void *seq;
void *a, *b;

int stop_fn( seq, a, b )
void *seq;
void *a, *b;


im_generate(3), with its supporting convenience functions, is used for PIO image output. See also im_wrapone(3) for an easy alternative to im_generate(3) for simple image processing operations.

im_start_one(3) and im_stop_one(3) are convenience functions, useful for simple one-image-in, one-image-out operations. im_start_one(3) assumes the first of the two user arguments (a, above) is the input image. It creates a REGION on this image and returns a pointer to the region as a sequence value.

im_stop_one(3) assumes the sequence value is a REGION pointer, and frees it.

im_allocate_input_array(3) takes as arguments the output image and a list of input images, terminated with a NULL. It allocates a NULL-terminated array to hold the images, and attaches a close callback to the output image to free that array. Example:

    IMAGE *in, *in2, *in3, *in4;
    IMAGE **arry;

    if( !(arry = im_allocate_input_array( out, 
        in1, in2, in3, in4, NULL )) )
        return( -1 );

builds the structure

    IMAGE *arry[] = { in1, in2, in3, in4, NULL };

and makes sure it will be freed.

im_start_many(3) and im_stop_many(3) work exactly as im_start_one(3) and im_stop_one(3), but with NULL-terminated arrays of IMAGEs and REGIONs. They are useful for many-images-in, one-image-out operations. im_start_many(3) assumes that the first of the two user arguments is a pointer to a NULL-terminates array of IMAGEs. It builds and returns as the sequence value a NULL-terminated array of REGIONs.

im_stop_many(3) assumes the sequence value is a pointer to a NULL-terminated array of REGIONs. It frees all the regions in turn. See im_add(3) for an example of this pair of functions in action.

im_generate(3) looks at the type of im and acts accordingly:

    IM_PARTIAL: the start, process and stop functions are attached to the  region, and im_generate returns immediately. See im_prepare(3).

    IM_SETBUF: memory for the output image is created and sequences started to fill it. It is an error to write to the same buffer twice.

    IM_MMAPINRW: sequences are started, and asked to fill the image in patches.

    IM_OPENOUT: The output file is created and a header written to disc. A  buffer large enough to hold GENERATE_TILE_HEIGHT complete horizontal lines is created, and sequences started to fill this buffer. When the buffer has been filled, the whole set of lines are flushed to disc in a single write(2) operation, and work starts on the next set of lines.

Any other image type is an error. im_generate(3) returns 0 for complete success, and non-zero on failure.

    static int
    wombat_gen( or, ir, in )
    REGION *or, *ir;
    IMAGE *in;
        ... process!

        return( 0 );

    im_wombat( in, out )
    IMAGE *in, *out;
        if( im_iocheck( in, out ) )
            return( -1 );

        ... check parametersm check image descriptors 
        ... for type-compatibility, etc. etc.

        if( im_cp_desc( out, in ) )
            return( -1 );

        ... set fields in out for the type of image you
        ... wish to write

        if( im_generate( out,
            im_start_one, wombat_gen, im_stop_one, 
            in, NULL ) )
            return( -1 );

        return( 0 );

See also the source to im_invert(3), im_exptra(3), and, if you are brave, im_conv(3) or im_add(3).

On machines with several CPUs, im_generate(3) and im_iterate(3) automatically parallelise programs. You can set the desired concurrency level with the environment variable IM_CONCURRENCY, for example

    example% export IM_CONCURRENCY=2
    example% lintra 2.0 fred.v 0.0 fred2.v

will run lintra with enough concurrency to keep 2 CPUs fully occupied. If IM_CONCURRENCY is not set, then it defaults to 1. See also im_concurrency_set(3).

Most programs which use VIPS will also let you use the command-line argument --vips-concurrency to set parallelisation, see im_get_option_group(3).


National Gallery, 1993


J. Cupitt - 23/7/93