/* Copyright (C) 2000 Aladdin Enterprises.  All rights reserved.
  
  This software is provided AS-IS with no warranty, either express or
  implied.
  
  This software is distributed under license and may not be copied,
  modified or distributed except as expressly authorized under the terms
  of the license contained in the file LICENSE in this distribution.
  
  For more information about licensing, please refer to
  http://www.ghostscript.com/licensing/. For information on
  commercial licensing, go to http://www.artifex.com/licensing/ or
  contact Artifex Software, Inc., 101 Lucas Valley Road #110,
  San Rafael, CA  94903, U.S.A., +1(415)492-9861.
*/

/* $Id: gximag3x.c,v 1.20 2004/09/16 08:03:56 igor Exp $ */
/* ImageType 3x image implementation */
/****** THE REAL WORK IS NYI ******/
#include "math_.h"		/* for ceil, floor */
#include "memory_.h"
#include "gx.h"
#include "gserrors.h"
#include "gsbitops.h"
#include "gscspace.h"
#include "gscpixel.h"
#include "gsstruct.h"
#include "gxdevice.h"
#include "gxdevmem.h"
#include "gximag3x.h"
#include "gxistate.h"
#include "gdevbbox.h"

extern_st(st_color_space);

/* Forward references */
private dev_proc_begin_typed_image(gx_begin_image3x);
private image_enum_proc_plane_data(gx_image3x_plane_data);
private image_enum_proc_end_image(gx_image3x_end_image);
private image_enum_proc_flush(gx_image3x_flush);
private image_enum_proc_planes_wanted(gx_image3x_planes_wanted);

/* GC descriptor */
private_st_gs_image3x();

/* Define the image type for ImageType 3x images. */
const gx_image_type_t gs_image_type_3x = {
    &st_gs_image3x, gx_begin_image3x, gx_data_image_source_size,
    gx_image_no_sput, gx_image_no_sget, gx_image_default_release,
    IMAGE3X_IMAGETYPE
};
private const gx_image_enum_procs_t image3x_enum_procs = {
    gx_image3x_plane_data, gx_image3x_end_image,
    gx_image3x_flush, gx_image3x_planes_wanted
};

/* Initialize an ImageType 3x image. */
private void
gs_image3x_mask_init(gs_image3x_mask_t *pimm)
{
    pimm->InterleaveType = 0;	/* not a valid type */
    pimm->has_Matte = false;
    gs_data_image_t_init(&pimm->MaskDict, 1);
    pimm->MaskDict.BitsPerComponent = 0;	/* not supplied */
}
void
gs_image3x_t_init(gs_image3x_t * pim, const gs_color_space * color_space)
{
    gs_pixel_image_t_init((gs_pixel_image_t *) pim, color_space);
    pim->type = &gs_image_type_3x;
    gs_image3x_mask_init(&pim->Opacity);
    gs_image3x_mask_init(&pim->Shape);
}

/*
 * We implement ImageType 3 images by interposing a mask clipper in
 * front of an ordinary ImageType 1 image.  Note that we build up the
 * mask row-by-row as we are processing the image.
 *
 * We export a generalized form of the begin_image procedure for use by
 * the PDF and PostScript writers.
 */

typedef struct image3x_channel_state_s {
    gx_image_enum_common_t *info;
    gx_device *mdev;		/* gx_device_memory in default impl. */
				/* (only for masks) */
    gs_image3_interleave_type_t InterleaveType;
    int width, height, full_height, depth;
    byte *data;			/* (if chunky) */
    /* Only the following change dynamically. */
    int y;
    int skip;			/* only for masks, # of rows to skip, */
				/* see below */
} image3x_channel_state_t;
typedef struct gx_image3x_enum_s {
    gx_image_enum_common;
    gx_device *pcdev;		/* gx_device_mask_clip in default impl. */
    int num_components;		/* (not counting masks) */
    int bpc;			/* pixel BitsPerComponent */
    gs_memory_t *memory;
#define NUM_MASKS 2		/* opacity, shape */
    image3x_channel_state_t mask[NUM_MASKS], pixel;
} gx_image3x_enum_t;

extern_st(st_gx_image_enum_common);
gs_private_st_suffix_add9(st_image3x_enum, gx_image3x_enum_t,
  "gx_image3x_enum_t", image3x_enum_enum_ptrs, image3x_enum_reloc_ptrs,
  st_gx_image_enum_common, pcdev, mask[0].info, mask[0].mdev, mask[0].data,
  mask[1].info, mask[1].mdev, mask[1].data, pixel.info, pixel.data);

/*
 * Begin a generic ImageType 3x image, with client handling the creation of
 * the mask image and mask clip devices.
 */
typedef struct image3x_channel_values_s {
    gs_matrix matrix;
    gs_point corner;
    gs_int_rect rect;
    gs_image_t image;
} image3x_channel_values_t;
private int check_image3x_mask(const gs_image3x_t *pim,
			       const gs_image3x_mask_t *pimm,
			       const image3x_channel_values_t *ppcv,
			       image3x_channel_values_t *pmcv,
			       image3x_channel_state_t *pmcs,
			       gs_memory_t *mem);
int
gx_begin_image3x_generic(gx_device * dev,
			const gs_imager_state *pis, const gs_matrix *pmat,
			const gs_image_common_t *pic, const gs_int_rect *prect,
			const gx_drawing_color *pdcolor,
			const gx_clip_path *pcpath, gs_memory_t *mem,
			image3x_make_mid_proc_t make_mid,
			image3x_make_mcde_proc_t make_mcde,
			gx_image_enum_common_t **pinfo)
{
    const gs_image3x_t *pim = (const gs_image3x_t *)pic;
    gx_image3x_enum_t *penum;
    gx_device *pcdev = 0;
    image3x_channel_values_t mask[2], pixel;
    gs_matrix mat;
    gx_device *midev[2];
    gx_image_enum_common_t *minfo[2];
    gs_int_point origin[2];
    int code;
    int i;

    /* Validate the parameters. */
    if (pim->Height <= 0)
	return_error(gs_error_rangecheck);
    penum = gs_alloc_struct(mem, gx_image3x_enum_t, &st_image3x_enum,
			    "gx_begin_image3x");
    if (penum == 0)
	return_error(gs_error_VMerror);
    /* Initialize pointers now in case we bail out. */
    penum->mask[0].info = 0, penum->mask[0].mdev = 0, penum->mask[0].data = 0;
    penum->mask[1].info = 0, penum->mask[1].mdev = 0, penum->mask[1].data = 0;
    penum->pixel.info = 0, penum->pixel.data = 0;
    if (prect)
	pixel.rect = *prect;
    else {
	pixel.rect.p.x = pixel.rect.p.y = 0;
	pixel.rect.q.x = pim->Width;
	pixel.rect.q.y = pim->Height;
    }
    if ((code = gs_matrix_invert(&pim->ImageMatrix, &pixel.matrix)) < 0 ||
	(code = gs_point_transform(pim->Width, pim->Height, &pixel.matrix,
				   &pixel.corner)) < 0 ||
	(code = check_image3x_mask(pim, &pim->Opacity, &pixel, &mask[0],
				   &penum->mask[0], mem)) < 0 ||
	(code = check_image3x_mask(pim, &pim->Shape, &pixel, &mask[1],
				   &penum->mask[1], mem)) < 0
	) {
	goto out0;
    }
    penum->num_components =
	gs_color_space_num_components(pim->ColorSpace);
    gx_image_enum_common_init((gx_image_enum_common_t *) penum,
			      (const gs_data_image_t *)pim,
			      &image3x_enum_procs, dev,
			      1 + penum->num_components,
			      pim->format);
    penum->pixel.width = pixel.rect.q.x - pixel.rect.p.x;
    penum->pixel.height = pixel.rect.q.y - pixel.rect.p.y;
    penum->pixel.full_height = pim->Height;
    penum->pixel.y = 0;
    if (penum->mask[0].data || penum->mask[1].data) {
	/* Also allocate a row buffer for the pixel data. */
	penum->pixel.data =
	    gs_alloc_bytes(mem,
			   (penum->pixel.width * pim->BitsPerComponent *
			    penum->num_components + 7) >> 3,
			   "gx_begin_image3x(pixel.data)");
	if (penum->pixel.data == 0) {
	    code = gs_note_error(gs_error_VMerror);
	    goto out1;
	}
    }
    penum->bpc = pim->BitsPerComponent;
    penum->memory = mem;
    if (pmat == 0)
	pmat = &ctm_only(pis);
    for (i = 0; i < NUM_MASKS; ++i) {
	gs_rect mrect;
	gx_device *mdev;
	/*
	 * The mask data has to be defined in a DevicePixel color space
	 * of the correct depth so that no color mapping will occur.
	 */
	/****** FREE COLOR SPACE ON ERROR OR AT END ******/
	gs_color_space *pmcs;

	if (penum->mask[i].depth == 0) {	/* mask not supplied */
	    midev[i] = 0;
	    minfo[i] = 0;
	    continue;
	}
	pmcs =  gs_alloc_struct(mem, gs_color_space, &st_color_space,
				"gx_begin_image3x_generic");
	if (pmcs == 0)
	    return_error(gs_error_VMerror);
	gs_cspace_init_DevicePixel(mem, pmcs, penum->mask[i].depth);
	mrect.p.x = mrect.p.y = 0;
	mrect.q.x = penum->mask[i].width;
	mrect.q.y = penum->mask[i].height;
	if ((code = gs_matrix_multiply(&mask[i].matrix, pmat, &mat)) < 0 ||
	    (code = gs_bbox_transform(&mrect, &mat, &mrect)) < 0
	    )
	    return code;
	origin[i].x = (int)floor(mrect.p.x);
	origin[i].y = (int)floor(mrect.p.y);
	code = make_mid(&mdev, dev,
			(int)ceil(mrect.q.x) - origin[i].x,
			(int)ceil(mrect.q.y) - origin[i].y,
			penum->mask[i].depth, mem);
	if (code < 0)
	    goto out1;
	penum->mask[i].mdev = mdev;
        gs_image_t_init(&mask[i].image, pmcs);
	mask[i].image.ColorSpace = pmcs;
	mask[i].image.adjust = false;
	{
	    const gx_image_type_t *type1 = mask[i].image.type;
	    const gs_image3x_mask_t *pixm =
		(i == 0 ? &pim->Opacity : &pim->Shape);

	    *(gs_data_image_t *)&mask[i].image = pixm->MaskDict;
	    mask[i].image.type = type1;
	    mask[i].image.BitsPerComponent = pixm->MaskDict.BitsPerComponent;
	}
	{
	    gs_matrix m_mat;

	    /*
	     * Adjust the translation for rendering the mask to include a
	     * negative translation by origin.{x,y} in device space.
	     */
	    m_mat = *pmat;
	    m_mat.tx -= origin[i].x;
	    m_mat.ty -= origin[i].y;
	    /*
	     * Note that pis = NULL here, since we don't want to have to
	     * create another imager state with default log_op, etc.
	     * dcolor = NULL is OK because this is an opaque image with
	     * CombineWithColor = false.
	     */
	    code = gx_device_begin_typed_image(mdev, NULL, &m_mat,
			       (const gs_image_common_t *)&mask[i].image,
					       &mask[i].rect, NULL, NULL,
					       mem, &penum->mask[i].info);
	    if (code < 0)
		goto out2;
	}
	midev[i] = mdev;
	minfo[i] = penum->mask[i].info;
    }
    gs_image_t_init(&pixel.image, pim->ColorSpace);
    {
	const gx_image_type_t *type1 = pixel.image.type;

	*(gs_pixel_image_t *)&pixel.image = *(const gs_pixel_image_t *)pim;
	pixel.image.type = type1;
    }
    code = make_mcde(dev, pis, pmat, (const gs_image_common_t *)&pixel.image,
		     prect, pdcolor, pcpath, mem, &penum->pixel.info,
		     &pcdev, midev, minfo, origin, pim);
    if (code < 0)
	goto out3;
    penum->pcdev = pcdev;
    /*
     * Set num_planes, plane_widths, and plane_depths from the values in the
     * enumerators for the mask(s) and the image data.
     */
    {
	int added_depth = 0;
	int pi = 0;

	for (i = 0; i < NUM_MASKS; ++i) {
	    if (penum->mask[i].depth == 0)	/* no mask */
		continue;
	    switch (penum->mask[i].InterleaveType) {
	    case interleave_chunky:
		/* Add the mask data to the depth of the image data. */
		added_depth += pim->BitsPerComponent;
		break;
	    case interleave_separate_source:
		/* Insert the mask as a separate plane. */
		penum->plane_widths[pi] = penum->mask[i].width;
		penum->plane_depths[pi] = penum->mask[i].depth;
		++pi;
		break;
	    default:		/* can't happen */
		code = gs_note_error(gs_error_Fatal);
		goto out3;
	    }
	}
	memcpy(&penum->plane_widths[pi], &penum->pixel.info->plane_widths[0],
	       penum->pixel.info->num_planes * sizeof(penum->plane_widths[0]));
	memcpy(&penum->plane_depths[pi], &penum->pixel.info->plane_depths[0],
	       penum->pixel.info->num_planes * sizeof(penum->plane_depths[0]));
	penum->plane_depths[pi] += added_depth;
	penum->num_planes = pi + penum->pixel.info->num_planes;
    }
    if (midev[0])
	gx_device_retain(midev[0], true); /* will free explicitly */
    if (midev[1])
	gx_device_retain(midev[1], true); /* ditto */
    gx_device_retain(pcdev, true); /* ditto */
    *pinfo = (gx_image_enum_common_t *) penum;
    return 0;
  out3:
    if (penum->mask[1].info)
	gx_image_end(penum->mask[1].info, false);
    if (penum->mask[0].info)
	gx_image_end(penum->mask[0].info, false);
  out2:
    if (penum->mask[1].mdev) {
	gs_closedevice(penum->mask[1].mdev);
	gs_free_object(mem, penum->mask[1].mdev,
		       "gx_begin_image3x(mask[1].mdev)");
    }
    if (penum->mask[0].mdev) {
	gs_closedevice(penum->mask[0].mdev);
	gs_free_object(mem, penum->mask[0].mdev,
		       "gx_begin_image3x(mask[0].mdev)");
    }
  out1:
    gs_free_object(mem, penum->mask[0].data, "gx_begin_image3x(mask[0].data)");
    gs_free_object(mem, penum->mask[1].data, "gx_begin_image3x(mask[1].data)");
    gs_free_object(mem, penum->pixel.data, "gx_begin_image3x(pixel.data)");
  out0:
    gs_free_object(mem, penum, "gx_begin_image3x");
    return code;
}
private bool
check_image3x_extent(floatp mask_coeff, floatp data_coeff)
{
    if (mask_coeff == 0)
	return data_coeff == 0;
    if (data_coeff == 0 || (mask_coeff > 0) != (data_coeff > 0))
	return false;
    return true;
}
/*
 * Check mask parameters.
 * Reads ppcv->{matrix,corner,rect}, sets pmcv->{matrix,corner,rect} and
 * pmcs->{InterleaveType,width,height,full_height,depth,data,y,skip}.
 * If the mask is omitted, sets pmcs->depth = 0 and returns normally.
 */
private bool
check_image3x_mask(const gs_image3x_t *pim, const gs_image3x_mask_t *pimm,
		   const image3x_channel_values_t *ppcv,
		   image3x_channel_values_t *pmcv,
		   image3x_channel_state_t *pmcs, gs_memory_t *mem)
{
    int mask_width = pimm->MaskDict.Width, mask_height = pimm->MaskDict.Height;
    int code;

    if (pimm->MaskDict.BitsPerComponent == 0) { /* mask missing */
	pmcs->depth = 0;
        pmcs->InterleaveType = 0;	/* not a valid type */
	return 0;
    }
    if (mask_height <= 0)
	return_error(gs_error_rangecheck);
    switch (pimm->InterleaveType) {
	/*case interleave_scan_lines:*/	/* not supported */
	default:
	    return_error(gs_error_rangecheck);
	case interleave_chunky:
	    if (mask_width != pim->Width ||
		mask_height != pim->Height ||
		pimm->MaskDict.BitsPerComponent != pim->BitsPerComponent ||
		pim->format != gs_image_format_chunky
		)
		return_error(gs_error_rangecheck);
	    break;
	case interleave_separate_source:
	    switch (pimm->MaskDict.BitsPerComponent) {
	    case 1: case 2: case 4: case 8:
		break;
	    default:
		return_error(gs_error_rangecheck);
	    }
    }
    if (!check_image3x_extent(pim->ImageMatrix.xx,
			      pimm->MaskDict.ImageMatrix.xx) ||
	!check_image3x_extent(pim->ImageMatrix.xy,
			      pimm->MaskDict.ImageMatrix.xy) ||
	!check_image3x_extent(pim->ImageMatrix.yx,
			      pimm->MaskDict.ImageMatrix.yx) ||
	!check_image3x_extent(pim->ImageMatrix.yy,
			      pimm->MaskDict.ImageMatrix.yy)
	)
	return_error(gs_error_rangecheck);
    if ((code = gs_matrix_invert(&pimm->MaskDict.ImageMatrix, &pmcv->matrix)) < 0 ||
	(code = gs_point_transform(mask_width, mask_height,
				   &pmcv->matrix, &pmcv->corner)) < 0
	)
	return code;
    if (fabs(ppcv->matrix.tx - pmcv->matrix.tx) >= 0.5 ||
	fabs(ppcv->matrix.ty - pmcv->matrix.ty) >= 0.5 ||
	fabs(ppcv->corner.x - pmcv->corner.x) >= 0.5 ||
	fabs(ppcv->corner.y - pmcv->corner.y) >= 0.5
	)
	return_error(gs_error_rangecheck);
    pmcv->rect.p.x = ppcv->rect.p.x * mask_width / pim->Width;
    pmcv->rect.p.y = ppcv->rect.p.y * mask_height / pim->Height;
    pmcv->rect.q.x = (ppcv->rect.q.x * mask_width + pim->Width - 1) /
	pim->Width;
    pmcv->rect.q.y = (ppcv->rect.q.y * mask_height + pim->Height - 1) /
	pim->Height;
    /* Initialize the channel state in the enumerator. */
    pmcs->InterleaveType = pimm->InterleaveType;
    pmcs->width = pmcv->rect.q.x - pmcv->rect.p.x;
    pmcs->height = pmcv->rect.q.y - pmcv->rect.p.y;
    pmcs->full_height = pimm->MaskDict.Height;
    pmcs->depth = pimm->MaskDict.BitsPerComponent;
    if (pmcs->InterleaveType == interleave_chunky) {
	/* Allocate a buffer for the data. */
	pmcs->data =
	    gs_alloc_bytes(mem,
			   (pmcs->width * pimm->MaskDict.BitsPerComponent + 7) >> 3,
			   "gx_begin_image3x(mask data)");
	if (pmcs->data == 0)
	    return_error(gs_error_VMerror);
    }
    pmcs->y = pmcs->skip = 0;
    return 0;
}

/*
 * Return > 0 if we want more data from channel 1 now, < 0 if we want more
 * from channel 2 now, 0 if we want both.
 */
private int
channel_next(const image3x_channel_state_t *pics1,
	     const image3x_channel_state_t *pics2)
{
    /*
     * The invariant we need to maintain is that we always have at least as
     * much channel N as channel N+1 data, where N = 0 = opacity, 1 = shape,
     * and 2 = pixel.  I.e., for any two consecutive channels c1 and c2, we
     * require c1.y / c1.full_height >= c2.y / c2.full_height, or, to avoid
     * floating point, c1.y * c2.full_height >= c2.y * c1.full_height.  We
     * know this condition is true now; return a value that indicates how to
     * maintain it.
     */
    int h1 = pics1->full_height;
    int h2 = pics2->full_height;
    long current = pics1->y * (long)h2 - pics2->y * (long)h1;

#ifdef DEBUG
    if (current < 0)
	lprintf4("channel_next invariant fails: %d/%d < %d/%d\n",
		 pics1->y, pics1->full_height,
		 pics2->y, pics2->full_height);
#endif
    return ((current -= h1) >= 0 ? -1 :
	    current + h2 >= 0 ? 0 : 1);
}

/* Define the default implementation of ImageType 3 processing. */
private IMAGE3X_MAKE_MID_PROC(make_midx_default); /* check prototype */
private int
make_midx_default(gx_device **pmidev, gx_device *dev, int width, int height,
		 int depth, gs_memory_t *mem)
{
    const gx_device_memory *mdproto = gdev_mem_device_for_bits(depth);
    gx_device_memory *midev;
    int code;

    if (mdproto == 0)
	return_error(gs_error_rangecheck);
    midev = gs_alloc_struct(mem, gx_device_memory, &st_device_memory,
			    "make_mid_default");
    if (midev == 0)
	return_error(gs_error_VMerror);
    gs_make_mem_device(midev, mdproto, mem, 0, NULL);
    midev->bitmap_memory = mem;
    midev->width = width;
    midev->height = height;
    check_device_separable((gx_device *)midev);
    gx_device_fill_in_procs((gx_device *)midev);
    code = dev_proc(midev, open_device)((gx_device *)midev);
    if (code < 0) {
	gs_free_object(mem, midev, "make_midx_default");
	return code;
    }
    midev->is_open = true;
    dev_proc(midev, fill_rectangle)
	((gx_device *)midev, 0, 0, width, height, (gx_color_index)0);
    *pmidev = (gx_device *)midev;
    return 0;
}
private IMAGE3X_MAKE_MCDE_PROC(make_mcdex_default);  /* check prototype */
private int
make_mcdex_default(gx_device *dev, const gs_imager_state *pis,
		   const gs_matrix *pmat, const gs_image_common_t *pic,
		   const gs_int_rect *prect, const gx_drawing_color *pdcolor,
		   const gx_clip_path *pcpath, gs_memory_t *mem,
		   gx_image_enum_common_t **pinfo,
		   gx_device **pmcdev, gx_device *midev[2],
		   gx_image_enum_common_t *pminfo[2],
		   const gs_int_point origin[2],
		   const gs_image3x_t *pim)
{
    /**************** NYI ****************/
    /*
     * There is no soft-mask analogue of make_mcde_default, because
     * soft-mask clipping is a more complicated operation, implemented
     * by the general transparency code.  As a default, we simply ignore
     * the soft mask.  However, we have to create an intermediate device
     * that can be freed at the end and that simply forwards all calls.
     * The most convenient device for this purpose is the bbox device.
     */
    gx_device_bbox *bbdev =
	gs_alloc_struct_immovable(mem, gx_device_bbox, &st_device_bbox,
				  "make_mcdex_default");
    int code;

    if (bbdev == 0)
	return_error(gs_error_VMerror);
    gx_device_bbox_init(bbdev, dev, mem);
    gx_device_bbox_fwd_open_close(bbdev, false);
    code = dev_proc(bbdev, begin_typed_image)
	((gx_device *)bbdev, pis, pmat, pic, prect, pdcolor, pcpath, mem,
	 pinfo);
    if (code < 0) {
	gs_free_object(mem, bbdev, "make_mcdex_default");
	return code;
    }
    *pmcdev = (gx_device *)bbdev;
    return 0;
}
private int
gx_begin_image3x(gx_device * dev,
		const gs_imager_state * pis, const gs_matrix * pmat,
		const gs_image_common_t * pic, const gs_int_rect * prect,
		const gx_drawing_color * pdcolor, const gx_clip_path * pcpath,
		gs_memory_t * mem, gx_image_enum_common_t ** pinfo)
{
    return gx_begin_image3x_generic(dev, pis, pmat, pic, prect, pdcolor,
				    pcpath, mem, make_midx_default,
				    make_mcdex_default, pinfo);
}

/* Process the next piece of an ImageType 3 image. */
private int
gx_image3x_plane_data(gx_image_enum_common_t * info,
		     const gx_image_plane_t * planes, int height,
		     int *rows_used)
{
    gx_image3x_enum_t *penum = (gx_image3x_enum_t *) info;
    int pixel_height = penum->pixel.height;
    int pixel_used = 0;
    int mask_height[2];
    int mask_used[2];
    int h1 = pixel_height - penum->pixel.y;
    int h;
    const gx_image_plane_t *pixel_planes;
    gx_image_plane_t pixel_plane, mask_plane[2];
    int code = 0;
    int i, pi = 0;
    int num_chunky = 0;

    for (i = 0; i < NUM_MASKS; ++i) {
	int mh = mask_height[i] = penum->mask[i].height;

	mask_plane[i].data = 0;
	mask_used[i] = 0;
	if (!penum->mask[i].depth)
	    continue;
	h1 = min(h1, mh - penum->mask[i].y);
	if (penum->mask[i].InterleaveType == interleave_chunky)
	    ++num_chunky;
    }
    h = min(height, h1);
    /* Initialized rows_used in case we get an error. */
    *rows_used = 0;
    if (h <= 0)
	return 0;

    /* Handle masks from separate sources. */
    for (i = 0; i < NUM_MASKS; ++i)
	if (penum->mask[i].InterleaveType == interleave_separate_source) {
	    /*
	     * In order to be able to recover from interruptions, we must
	     * limit separate-source processing to 1 scan line at a time.
	     */
	    if (h > 1)
		h = 1;
	    mask_plane[i] = planes[pi++];
	}
    pixel_planes = &planes[pi];

    /* Handle chunky masks. */
    if (num_chunky) {
	int bpc = penum->bpc;
	int num_components = penum->num_components;
	int width = penum->pixel.width;
	/* Pull apart the source data and the mask data. */
	/* We do this in the simplest (not fastest) way for now. */
	uint bit_x = bpc * (num_components + num_chunky) * planes[pi].data_x;
	sample_load_declare_setup(sptr, sbit, planes[0].data + (bit_x >> 3),
				  bit_x & 7, bpc);
	sample_store_declare_setup(pptr, pbit, pbbyte,
				   penum->pixel.data, 0, bpc);
	sample_store_declare(dptr[NUM_MASKS], dbit[NUM_MASKS],
			     dbbyte[NUM_MASKS]);
	int depth[NUM_MASKS];
	int x;

	if (h > 1) {
	    /* Do the operation one row at a time. */
	    h = 1;
	}
	for (i = 0; i < NUM_MASKS; ++i)
	    if (penum->mask[i].data) {
		depth[i] = penum->mask[i].depth;
		mask_plane[i].data = dptr[i] = penum->mask[i].data;
		mask_plane[i].data_x = 0;
		/* raster doesn't matter */
		sample_store_setup(dbit[i], 0, depth[i]);
		sample_store_preload(dbbyte[i], dptr[i], 0, depth[i]);
	    } else
		depth[i] = 0;
	pixel_plane.data = pptr;
	pixel_plane.data_x = 0;
	/* raster doesn't matter */
	pixel_planes = &pixel_plane;
	for (x = 0; x < width; ++x) {
	    uint value;

	    for (i = 0; i < NUM_MASKS; ++i)
		if (depth[i]) {
		    sample_load_next12(value, sptr, sbit, bpc);
		    sample_store_next12(value, dptr[i], dbit[i], depth[i],
					dbbyte[i]);
		}
	    for (i = 0; i < num_components; ++i) {
		sample_load_next12(value, sptr, sbit, bpc);
		sample_store_next12(value, pptr, pbit, bpc, pbbyte);
	    }
	}
	for (i = 0; i < NUM_MASKS; ++i)
	    if (penum->mask[i].data)
		sample_store_flush(dptr[i], dbit[i], depth[i], dbbyte[i]);
	sample_store_flush(pptr, pbit, bpc, pbbyte);
	}
    /*
     * Process the mask data first, so it will set up the mask
     * device for clipping the pixel data.
     */
    for (i = 0; i < NUM_MASKS; ++i)
	if (mask_plane[i].data) {
	    /*
	     * If, on the last call, we processed some mask rows
	     * successfully but processing the pixel rows was interrupted,
	     * we set rows_used to indicate the number of pixel rows
	     * processed (since there is no way to return two rows_used
	     * values).  If this happened, some mask rows may get presented
	     * again.  We must skip over them rather than processing them
	     * again.
	     */
	    int skip = penum->mask[i].skip;

	    if (skip >= h) {
		penum->mask[i].skip = skip - (mask_used[i] = h);
	    } else {
		int mask_h = h - skip;

		mask_plane[i].data += skip * mask_plane[i].raster;
		penum->mask[i].skip = 0;
		code = gx_image_plane_data_rows(penum->mask[i].info,
						&mask_plane[i],
						mask_h, &mask_used[i]);
		mask_used[i] += skip;
	    }
	    *rows_used = mask_used[i];
	    penum->mask[i].y += mask_used[i];
	    if (code < 0)
		return code;
	}
    if (pixel_planes[0].data) {
	/*
	 * If necessary, flush any buffered mask data to the mask clipping
	 * device.
	 */
	for (i = 0; i < NUM_MASKS; ++i)
	    if (penum->mask[i].info)
		gx_image_flush(penum->mask[i].info);
	code = gx_image_plane_data_rows(penum->pixel.info, pixel_planes, h,
					&pixel_used);
	/*
	 * There isn't any way to set rows_used if different amounts of
	 * the mask and pixel data were used.  Fake it.
	 */
	*rows_used = pixel_used;
	/*
	 * Don't return code yet: we must account for the fact that
	 * some mask data may have been processed.
	 */
	penum->pixel.y += pixel_used;
	if (code < 0) {
	    /*
	     * We must prevent the mask data from being processed again.
	     * We rely on the fact that h > 1 is only possible if the
	     * mask and pixel data have the same Y scaling.
	     */
	    for (i = 0; i < NUM_MASKS; ++i)
		if (mask_used[i] > pixel_used) {
		    int skip = mask_used[i] - pixel_used;

		    penum->mask[i].skip = skip;
		    penum->mask[i].y -= skip;
		    mask_used[i] = pixel_used;
		}
	}
    }
    if_debug7('b', "[b]image3x h=%d %sopacity.y=%d %sopacity.y=%d %spixel.y=%d\n",
	      h, (mask_plane[0].data ? "+" : ""), penum->mask[0].y,
	      (mask_plane[1].data ? "+" : ""), penum->mask[1].y,
	      (pixel_planes[0].data ? "+" : ""), penum->pixel.y);
    if (penum->mask[0].y >= penum->mask[0].height &&
	penum->mask[1].y >= penum->mask[1].height &&
	penum->pixel.y >= penum->pixel.height)
	return 1;
    /*
     * The mask may be complete (gx_image_plane_data_rows returned 1),
     * but there may still be pixel rows to go, so don't return 1 here.
     */
    return (code < 0 ? code : 0);
}

/* Flush buffered data. */
private int
gx_image3x_flush(gx_image_enum_common_t * info)
{
    gx_image3x_enum_t * const penum = (gx_image3x_enum_t *) info;
    int code = gx_image_flush(penum->mask[0].info);

    if (code >= 0)
	code = gx_image_flush(penum->mask[1].info);
    if (code >= 0)
	code = gx_image_flush(penum->pixel.info);
    return code;
}

/* Determine which data planes are wanted. */
private bool
gx_image3x_planes_wanted(const gx_image_enum_common_t * info, byte *wanted)
{
    const gx_image3x_enum_t * const penum = (const gx_image3x_enum_t *) info;
    /*
     * We always want at least as much of the mask(s) to be filled as the
     * pixel data.
     */
    bool
	sso = penum->mask[0].InterleaveType == interleave_separate_source,
	sss = penum->mask[1].InterleaveType == interleave_separate_source;

    if (sso & sss) {
	/* Both masks have separate sources. */
	int mask_next = channel_next(&penum->mask[1], &penum->pixel);

	memset(wanted + 2, (mask_next <= 0 ? 0xff : 0), info->num_planes - 2);
	wanted[1] = (mask_next >= 0 ? 0xff : 0);
	if (wanted[1]) {
	    mask_next = channel_next(&penum->mask[0], &penum->mask[1]);
	    wanted[0] = mask_next >= 0;
	} else
	    wanted[0] = 0;
	return false;		/* see below */
    } else if (sso | sss) {
	/* Only one separate source. */
	const image3x_channel_state_t *pics =
	    (sso ? &penum->mask[0] : &penum->mask[1]);
	int mask_next = channel_next(pics, &penum->pixel);

	wanted[0] = (mask_next >= 0 ? 0xff : 0);
	memset(wanted + 1, (mask_next <= 0 ? 0xff : 0), info->num_planes - 1);
	/*
	 * In principle, wanted will always be true for both mask and pixel
	 * data if the full_heights are equal.  Unfortunately, even in this
	 * case, processing may be interrupted after a mask row has been
	 * passed to the underlying image processor but before the data row
	 * has been passed, in which case pixel data will be 'wanted', but
	 * not mask data, for the next call.  Therefore, we must return
	 * false.
	 */
	return false
	    /*(next == 0 &&
	      pics->full_height == penum->pixel.full_height)*/;
    } else {
	/* Everything is chunky, only 1 plane. */
	wanted[0] = 0xff;
	return true;
    }
}

/* Clean up after processing an ImageType 3x image. */
private int
gx_image3x_end_image(gx_image_enum_common_t * info, bool draw_last)
{
    gx_image3x_enum_t *penum = (gx_image3x_enum_t *) info;
    gs_memory_t *mem = penum->memory;
    gx_device *mdev0 = penum->mask[0].mdev;
    int ocode =
	(penum->mask[0].info ? gx_image_end(penum->mask[0].info, draw_last) :
	 0);
    gx_device *mdev1 = penum->mask[1].mdev;
    int scode =
	(penum->mask[1].info ? gx_image_end(penum->mask[1].info, draw_last) :
	 0);
    gx_device *pcdev = penum->pcdev;
    int pcode = gx_image_end(penum->pixel.info, draw_last);

    gs_closedevice(pcdev);
    if (mdev0)
	gs_closedevice(mdev0);
    if (mdev1)
	gs_closedevice(mdev1);
    gs_free_object(mem, penum->mask[0].data,
		   "gx_image3x_end_image(mask[0].data)");
    gs_free_object(mem, penum->mask[1].data,
		   "gx_image3x_end_image(mask[1].data)");
    gs_free_object(mem, penum->pixel.data,
		   "gx_image3x_end_image(pixel.data)");
    gs_free_object(mem, pcdev, "gx_image3x_end_image(pcdev)");
    gs_free_object(mem, mdev0, "gx_image3x_end_image(mask[0].mdev)");
    gs_free_object(mem, mdev1, "gx_image3x_end_image(mask[1].mdev)");
    gs_free_object(mem, penum, "gx_image3x_end_image");
    return (pcode < 0 ? pcode : scode < 0 ? scode : ocode);
}