shithub: riscv

ref: 985b2457cda207c2c65edeb647aedbb6e92dac46
dir: /sys/src/cmd/gs/src/gxclimag.c/

View raw version
/* Copyright (C) 1996, 1997, 1998, 1999 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: gxclimag.c,v 1.13 2005/10/10 18:58:18 leonardo Exp $ */
/* Higher-level image operations for band lists */
#include "math_.h"
#include "memory_.h"
#include "gx.h"
#include "gserrors.h"
#include "gscspace.h"
#include "gscdefs.h"		/* for image type table */
#include "gxarith.h"
#include "gxcspace.h"
#include "gxdevice.h"
#include "gxdevmem.h"		/* must precede gxcldev.h */
#include "gxcldev.h"
#include "gxclpath.h"
#include "gxfmap.h"
#include "gxiparam.h"
#include "gxpath.h"
#include "stream.h"
#include "strimpl.h"		/* for sisparam.h */
#include "sisparam.h"
#include "gxcomp.h"
#include "gsserial.h"
#include "gxdhtserial.h"

extern_gx_image_type_table();

/* Define whether we should use high-level images. */
/* (See below for additional restrictions.) */
static const bool USE_HL_IMAGES = true;

/* Forward references */
private int cmd_put_set_data_x(gx_device_clist_writer * cldev,
			       gx_clist_state * pcls, int data_x);
private bool check_rect_for_trivial_clip(
    const gx_clip_path *pcpath,  /* May be NULL, clip to evaluate */
    int px, int py, int qx, int qy  /* corners of box to test */
);

/* ------ Driver procedures ------ */

int
clist_fill_mask(gx_device * dev,
		const byte * data, int data_x, int raster, gx_bitmap_id id,
		int x, int y, int width, int height,
		const gx_drawing_color * pdcolor, int depth,
		gs_logical_operation_t lop, const gx_clip_path * pcpath)
{
    gx_device_clist_writer * const cdev =
	&((gx_device_clist *)dev)->writer;
    const byte *orig_data = data;	/* for writing tile */
    int orig_data_x = data_x;	/* ditto */
    int orig_x = x;		/* ditto */
    int orig_width = width;	/* ditto */
    int orig_height = height;	/* ditto */
    int log2_depth = ilog2(depth);
    int y0;
    int data_x_bit;
    byte copy_op =
	(depth > 1 ? cmd_op_copy_color_alpha :
	 cmd_op_copy_mono + cmd_copy_ht_color);
    bool slow_rop =
	cmd_slow_rop(dev, lop_know_S_0(lop), pdcolor) ||
	cmd_slow_rop(dev, lop_know_S_1(lop), pdcolor);

    /* If depth > 1, this call will be translated to a copy_alpha call. */
    /* if the target device can't perform copy_alpha, exit now. */
    if (depth > 1 && (cdev->disable_mask & clist_disable_copy_alpha) != 0)
	return_error(gs_error_unknownerror);

    fit_copy(dev, data, data_x, raster, id, x, y, width, height);
    y0 = y;			/* must do after fit_copy */

    /* If non-trivial clipping & complex clipping disabled, default */
    /* Also default for uncached bitmap or non-defaul lop; */
    /* We could handle more RasterOp cases here directly, but it */
    /* doesn't seem worth the trouble right now. */
    /* Lastly, the command list will translate calls with depth > 1 to */
    /* copy_alpha calls, so the device color must be pure */
    if (((cdev->disable_mask & clist_disable_complex_clip) &&
	 !check_rect_for_trivial_clip(pcpath, x, y, x + width, y + height)) ||
	gs_debug_c('`') || id == gx_no_bitmap_id || lop != lop_default ||
	(depth > 1 && !color_writes_pure(pdcolor, lop))
	)
  copy:
	return gx_default_fill_mask(dev, data, data_x, raster, id,
				    x, y, width, height, pdcolor, depth,
				    lop, pcpath);

    if (cmd_check_clip_path(cdev, pcpath))
	cmd_clear_known(cdev, clip_path_known);
    data_x_bit = data_x << log2_depth;
    FOR_RECTS {
	int code;
        ulong offset_temp;

	TRY_RECT {
	    code = cmd_update_lop(cdev, pcls, lop);
	} HANDLE_RECT(code);
	if (depth > 1 && !pcls->color_is_alpha) {
	    byte *dp;

	    TRY_RECT {
		code =
		    set_cmd_put_op(dp, cdev, pcls, cmd_opv_set_copy_alpha, 1);
	    } HANDLE_RECT(code);
	    pcls->color_is_alpha = 1;
	}
	TRY_RECT {
	    code = cmd_do_write_unknown(cdev, pcls, clip_path_known);
	    if (code >= 0)
		code = cmd_do_enable_clip(cdev, pcls, pcpath != NULL);
	} HANDLE_RECT(code);
	TRY_RECT {
	    code = cmd_put_drawing_color(cdev, pcls, pdcolor);
	    if (depth > 1 && code >= 0)
		code = cmd_set_color1(cdev, pcls, pdcolor->colors.pure);
	} HANDLE_RECT(code);
	pcls->colors_used.slow_rop |= slow_rop;

	/* Put it in the cache if possible. */
	if (!cls_has_tile_id(cdev, pcls, id, offset_temp)) {
	    gx_strip_bitmap tile;

	    tile.data = (byte *) orig_data;	/* actually const */
	    tile.raster = raster;
	    tile.size.x = tile.rep_width = orig_width;
	    tile.size.y = tile.rep_height = orig_height;
	    tile.rep_shift = tile.shift = 0;
	    tile.id = id;
	    TRY_RECT {
	        code = clist_change_bits(cdev, pcls, &tile, depth);
	    } HANDLE_RECT_UNLESS(code,
	        (code != gs_error_VMerror || !cdev->error_is_retryable) );
	    if (code < 0) {
	        /* Something went wrong; just copy the bits. */
	        goto copy;
	    }
	}
	{
	    gx_cmd_rect rect;
	    int rsize;
	    byte op = copy_op + cmd_copy_use_tile;

	    /* Output a command to copy the entire character. */
	    /* It will be truncated properly per band. */
	    rect.x = orig_x, rect.y = y0;
	    rect.width = orig_width, rect.height = yend - y0;
	    rsize = 1 + cmd_sizexy(rect);
	    TRY_RECT {
	        code = (orig_data_x ?
	      		cmd_put_set_data_x(cdev, pcls, orig_data_x) : 0);
		if (code >= 0) {
		    byte *dp;

		    code = set_cmd_put_op(dp, cdev, pcls, op, rsize);
		    /*
		     * The following conditional is unnecessary: the two
		     * statements inside it should go outside the
		     * HANDLE_RECT.  They are here solely to pacify
		     * stupid compilers that don't understand that dp
		     * will always be set if control gets past the
		     * HANDLE_RECT.
		     */
		    if (code >= 0) {
		        dp++;
		        cmd_putxy(rect, dp);
		    }
		}
	    } HANDLE_RECT(code);
	    pcls->rect = rect;
	    goto end;
	}
end:
	;
    } END_RECTS;
	return 0;
}

/* ------ Bitmap image driver procedures ------ */

/* Define the structure for keeping track of progress through an image. */
typedef struct clist_image_enum_s {
    gx_image_enum_common;
    /* Arguments of begin_image */
    gs_memory_t *memory;
    gs_pixel_image_t image;	/* only uses Width, Height, Interpolate */
    gx_drawing_color dcolor;	/* only pure right now */
    gs_int_rect rect;
    const gs_imager_state *pis;
    const gx_clip_path *pcpath;
    /* Set at creation time */
    gs_image_format_t format;
    gs_int_point support;	/* extra source pixels for interpolation */
    int bits_per_plane;		/* bits per pixel per plane */
    gs_matrix matrix;		/* image space -> device space */
    bool uses_color;
    clist_color_space_t color_space;
    int ymin, ymax;
    gx_colors_used_t colors_used;
    /* begin_image command prepared & ready to output */
    /****** SIZE COMPUTATION IS WRONG, TIED TO gximage.c, gsmatrix.c ******/
    byte begin_image_command[3 +
			    /* Width, Height */
			    2 * cmd_sizew_max +
			    /* ImageMatrix */
			    1 + 6 * sizeof(float) +
			    /* Decode */
			    (GS_IMAGE_MAX_COMPONENTS + 3) / 4 +
			      GS_IMAGE_MAX_COMPONENTS * 2 * sizeof(float) +
			    /* MaskColors */
			    GS_IMAGE_MAX_COMPONENTS * cmd_sizew_max +
			    /* rect */
			    4 * cmd_sizew_max];
    int begin_image_command_length;
    /* Updated dynamically */
    int y;
    bool color_map_is_known;
} clist_image_enum;
gs_private_st_suffix_add3(st_clist_image_enum, clist_image_enum,
			  "clist_image_enum", clist_image_enum_enum_ptrs,
			  clist_image_enum_reloc_ptrs, st_gx_image_enum_common,
			  pis, pcpath, color_space.space);

private image_enum_proc_plane_data(clist_image_plane_data);
private image_enum_proc_end_image(clist_image_end_image);
private const gx_image_enum_procs_t clist_image_enum_procs =
{
    clist_image_plane_data, clist_image_end_image
};

/* Forward declarations */
private bool image_band_box(gx_device * dev, const clist_image_enum * pie,
			    int y, int h, gs_int_rect * pbox);
private int begin_image_command(byte *buf, uint buf_size,
				const gs_image_common_t *pic);
private int cmd_image_plane_data(gx_device_clist_writer * cldev,
				 gx_clist_state * pcls,
				 const gx_image_plane_t * planes,
				 const gx_image_enum_common_t * pie,
				 uint bytes_per_plane,
				 const uint * offsets, int dx, int h);
private uint clist_image_unknowns(gx_device *dev,
				  const clist_image_enum *pie);
private int write_image_end_all(gx_device *dev,
				const clist_image_enum *pie);

/*
 * Since currently we are limited to writing a single subrectangle of the
 * image for each band, images that are rotated by angles other than
 * multiples of 90 degrees may wind up writing many copies of the data.
 * Eventually we will fix this by breaking up the image into multiple
 * subrectangles, but for now, don't use the high-level approach if it would
 * cause the data to explode because of this.
 */
private bool
image_matrix_ok_to_band(const gs_matrix * pmat)
{
    double t;

    /* Don't band if the matrix is (nearly) singular. */
    if (fabs(pmat->xx * pmat->yy - pmat->xy * pmat->yx) < 0.001)
	return false;
    if (is_xxyy(pmat) || is_xyyx(pmat))
	return true;
    t = (fabs(pmat->xx) + fabs(pmat->yy)) /
	(fabs(pmat->xy) + fabs(pmat->yx));
    return (t < 0.2 || t > 5);
}

/* Start processing an image. */
int
clist_begin_typed_image(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)
{
    const gs_pixel_image_t * const pim = (const gs_pixel_image_t *)pic;
    gx_device_clist_writer * const cdev =
	&((gx_device_clist *)dev)->writer;
    clist_image_enum *pie = 0;
    int base_index;
    bool indexed;
    bool masked = false;
    bool has_alpha = false;
    int num_components;
    int bits_per_pixel;
    bool uses_color;
    bool varying_depths = false;
    gs_matrix mat;
    gs_rect sbox, dbox;
    gs_image_format_t format;
    gx_color_index colors_used = 0;
    int code;

    /* We can only handle a limited set of image types. */
    switch ((gs_debug_c('`') ? -1 : pic->type->index)) {
    case 1:
	masked = ((const gs_image1_t *)pim)->ImageMask;
	has_alpha = ((const gs_image1_t *)pim)->Alpha != 0;
    case 4:
	if (pmat == 0)
	    break;
    default:
	goto use_default;
    }
    format = pim->format;
    /* See above for why we allocate the enumerator as immovable. */
    pie = gs_alloc_struct_immovable(mem, clist_image_enum,
				    &st_clist_image_enum,
				    "clist_begin_typed_image");
    if (pie == 0)
	return_error(gs_error_VMerror);
    pie->memory = mem;
    *pinfo = (gx_image_enum_common_t *) pie;
    /* num_planes and plane_depths[] are set later, */
    /* by gx_image_enum_common_init. */
    if (masked) {
	base_index = gs_color_space_index_DeviceGray;	/* arbitrary */
	indexed = false;
	num_components = 1;
	uses_color = true;
	/* cmd_put_drawing_color handles colors_used */
    } else {
	const gs_color_space *pcs = pim->ColorSpace;

	base_index = gs_color_space_get_index(pcs);
	if (base_index == gs_color_space_index_Indexed) {
	    const gs_color_space *pbcs =
		gs_color_space_indexed_base_space(pcs);

	    indexed = true;
	    base_index = gs_color_space_get_index(pbcs);
	    num_components = 1;
	} else {
	    indexed = false;
	    num_components = gs_color_space_num_components(pcs);
	}
	uses_color = pim->CombineWithColor && rop3_uses_T(pis->log_op);
    }
    code = gx_image_enum_common_init((gx_image_enum_common_t *) pie,
				     (const gs_data_image_t *) pim,
				     &clist_image_enum_procs, dev,
				     num_components, format);
    {
	int i;

	for (i = 1; i < pie->num_planes; ++i)
	    varying_depths |= pie->plane_depths[i] != pie->plane_depths[0];
    }
    if (code < 0 ||
	!USE_HL_IMAGES ||	/* Always use the default. */
	(cdev->disable_mask & clist_disable_hl_image) || 
	cdev->image_enum_id != gs_no_id ||  /* Can't handle nested images */
	/****** CAN'T HANDLE CIE COLOR YET ******/
	base_index > gs_color_space_index_DeviceCMYK ||
	/****** CAN'T HANDLE NON-PURE COLORS YET ******/
	(uses_color && !gx_dc_is_pure(pdcolor)) ||
	/****** CAN'T HANDLE IMAGES WITH ALPHA YET ******/
	has_alpha ||
	/****** CAN'T HANDLE IMAGES WITH IRREGULAR DEPTHS ******/
	varying_depths ||
	(code = gs_matrix_invert(&pim->ImageMatrix, &mat)) < 0 ||
	(code = gs_matrix_multiply(&mat, &ctm_only(pis), &mat)) < 0 ||
	!(cdev->disable_mask & clist_disable_nonrect_hl_image ?
	  (is_xxyy(&mat) || is_xyyx(&mat)) :
	  image_matrix_ok_to_band(&mat))
	)
	goto use_default;
    {
	int bytes_per_plane, bytes_per_row;

	bits_per_pixel = pim->BitsPerComponent * num_components;
	pie->image = *pim;
	pie->dcolor = *pdcolor;
	if (prect)
	    pie->rect = *prect;
	else {
	    pie->rect.p.x = 0, pie->rect.p.y = 0;
	    pie->rect.q.x = pim->Width, pie->rect.q.y = pim->Height;
	}
	pie->pis = pis;
	pie->pcpath = pcpath;
	pie->format = format;
	pie->bits_per_plane = bits_per_pixel / pie->num_planes;
	pie->matrix = mat;
	pie->uses_color = uses_color;
	if (masked) {
	    pie->color_space.byte1 = 0;  /* arbitrary */
	    pie->color_space.space = 0;
	    pie->color_space.id = gs_no_id;
	} else {
	    pie->color_space.byte1 = (base_index << 4) |
		(indexed ? (pim->ColorSpace->params.indexed.use_proc ? 12 : 8) : 0);
	    pie->color_space.id =
		(pie->color_space.space = pim->ColorSpace)->id;
	}
	pie->y = pie->rect.p.y;

	/* Image row has to fit in cmd writer's buffer */
	bytes_per_plane =
	    (pim->Width * pie->bits_per_plane + 7) >> 3;
	bytes_per_row = bytes_per_plane * pie->num_planes;
	bytes_per_row = max(bytes_per_row, 1);
	if (cmd_largest_size + bytes_per_row > cdev->cend - cdev->cbuf)
	    goto use_default;
    }
    if (pim->Interpolate)
	pie->support.x = pie->support.y = MAX_ISCALE_SUPPORT + 1;
    else
	pie->support.x = pie->support.y = 0;
    sbox.p.x = pie->rect.p.x - pie->support.x;
    sbox.p.y = pie->rect.p.y - pie->support.y;
    sbox.q.x = pie->rect.q.x + pie->support.x;
    sbox.q.y = pie->rect.q.y + pie->support.y;
    gs_bbox_transform(&sbox, &mat, &dbox);

    if (cdev->disable_mask & clist_disable_complex_clip)
	if (!check_rect_for_trivial_clip(pcpath,
				(int)floor(dbox.p.x), (int)floor(dbox.p.y),
				(int)ceil(dbox.q.x), (int)ceil(dbox.q.y)))
	    goto use_default;
    /* Create the begin_image command. */
    if ((pie->begin_image_command_length =
	 begin_image_command(pie->begin_image_command,
			     sizeof(pie->begin_image_command), pic)) < 0)
	goto use_default;
    if (!masked) {
	/*
	 * Calculate (conservatively) the set of colors that this image
	 * might generate.  For single-component images with up to 4 bits
	 * per pixel, standard Decode values, and no Interpolate, we
	 * generate all the possible colors now; otherwise, we assume that
	 * any color might be generated.  It is possible to do better than
	 * this, but we won't bother unless there's evidence that it's
	 * worthwhile.
	 */
	gx_color_index all =
	    ((gx_color_index)1 << dev->color_info.depth) - 1;

	if (bits_per_pixel > 4 || pim->Interpolate || num_components > 1)
	    colors_used = all;
	else {
	    int max_value = (1 << bits_per_pixel) - 1;
	    float dmin = pim->Decode[0], dmax = pim->Decode[1];
	    float dtemp;

	    if (dmax < dmin)
		dtemp = dmax, dmax = dmin, dmin = dtemp;
	    if (dmin != 0 ||
		dmax != (indexed ? max_value : 1)
		) {
		colors_used = all;
	    } else {
		/* Enumerate the possible pixel values. */
		const gs_color_space *pcs = pim->ColorSpace;
		cs_proc_remap_color((*remap_color)) = pcs->type->remap_color;
		gs_client_color cc;
		gx_drawing_color dcolor;
		int i;
		double denom = (indexed ? 1 : max_value);

		for (i = 0; i <= max_value; ++i) {
		    cc.paint.values[0] = (double)i / denom;
		    remap_color(&cc, pcs, &dcolor, pis, dev,
				gs_color_select_source);
		    colors_used |= cmd_drawing_colors_used(cdev, &dcolor);
		}
	    }
	}
    }

    pie->colors_used.or = colors_used;
    pie->colors_used.slow_rop =
	cmd_slow_rop(dev, pis->log_op, (uses_color ? pdcolor : NULL));
    pie->color_map_is_known = false;
    /*
     * Calculate a (slightly conservative) Y bounding interval for the image
     * in device space.
     */
    {
	int y0 = (int)floor(dbox.p.y - 0.51);	/* adjust + rounding slop */
	int y1 = (int)ceil(dbox.q.y + 0.51);	/* ditto */

	pie->ymin = max(y0, 0);
	pie->ymax = min(y1, dev->height);
    }

    /*
     * Make sure the CTM, color space, and clipping region (and, for
     * masked images or images with CombineWithColor, the current color)
     * are known at the time of the begin_image command.
     */
    cmd_clear_known(cdev, clist_image_unknowns(dev, pie) | begin_image_known);

    cdev->image_enum_id = pie->id;
    return 0;

    /*
     * We couldn't handle the image.  Use the default algorithms, which
     * break the image up into rectangles or small pixmaps.
     */
use_default:
    gs_free_object(mem, pie, "clist_begin_typed_image");
    return gx_default_begin_typed_image(dev, pis, pmat, pic, prect,
					pdcolor, pcpath, mem, pinfo);
}

/* Process the next piece of an image. */
private int
clist_image_plane_data(gx_image_enum_common_t * info,
		       const gx_image_plane_t * planes, int yh,
		       int *rows_used)
{
    gx_device *dev = info->dev;
    gx_device_clist_writer * const cdev =
	&((gx_device_clist *)dev)->writer;
    clist_image_enum *pie = (clist_image_enum *) info;
    gs_rect sbox, dbox;
    int y_orig = pie->y;
    int yh_used = min(yh, pie->rect.q.y - y_orig);
    int y0, y1;
    int y, height;		/* for BEGIN/END_RECT */
    int code;

#ifdef DEBUG
    if (pie->id != cdev->image_enum_id) {
	lprintf2("end_image id = %lu != clist image id = %lu!\n",
		 (ulong) pie->id, (ulong) cdev->image_enum_id);
	*rows_used = 0;
	return_error(gs_error_Fatal);
    }
#endif
    /****** CAN'T HANDLE VARYING data_x VALUES YET ******/
    {
	int i;

	for (i = 1; i < info->num_planes; ++i)
	    if (planes[i].data_x != planes[0].data_x) {
		*rows_used = 0;
		return_error(gs_error_rangecheck);
	    }
    }
    sbox.p.x = pie->rect.p.x - pie->support.x;
    sbox.p.y = (y0 = y_orig) - pie->support.y;
    sbox.q.x = pie->rect.q.x + pie->support.x;
    sbox.q.y = (y1 = pie->y += yh_used) + pie->support.y;
    gs_bbox_transform(&sbox, &pie->matrix, &dbox);
    /*
     * In order to keep the band list consistent, we must write out
     * the image data in precisely those bands whose begin_image
     * Y range includes the respective image scan lines.  Because of
     * rounding, we must expand the dbox by a little extra, and then
     * use image_band_box to calculate the precise range for each band.
     * This is slow, but we don't see any faster way to do it in the
     * general case.
     */
    {
	int ry0 = (int)floor(dbox.p.y) - 2;
	int ry1 = (int)ceil(dbox.q.y) + 2;
	int band_height = cdev->page_band_height;

	/*
	 * Make sure we don't go into any bands beyond the Y range
	 * determined at begin_image time.
	 */
	if (ry0 < pie->ymin)
	    ry0 = pie->ymin;
	if (ry1 > pie->ymax)
	    ry1 = pie->ymax;
	/*
	 * If the image extends off the page in the Y direction,
	 * we may have ry0 > ry1.  Check for this here.
	 */
	if (ry0 >= ry1)
	    goto done;
	/* Expand the range out to band boundaries. */
	y = ry0 / band_height * band_height;
	height = min(ROUND_UP(ry1, band_height), dev->height) - y;
    }

    FOR_RECTS {
	/*
	 * Just transmit the subset of the data that intersects this band.
	 * Note that y and height always define a complete band.
	 */
	gs_int_rect ibox;
	gs_int_rect entire_box;

	if (!image_band_box(dev, pie, y, height, &ibox))
	    continue;
	/*
	 * The transmitted subrectangle has to be computed at the time
	 * we write the begin_image command; this in turn controls how
	 * much of each scan line we write out.
	 */
	{
	    int band_ymax = min(band_end, pie->ymax);
	    int band_ymin = max(band_end - band_height, pie->ymin);

	    if (!image_band_box(dev, pie, band_ymin,
				band_ymax - band_ymin, &entire_box))
		continue;
	}

	pcls->colors_used.or |= pie->colors_used.or;
	pcls->colors_used.slow_rop |= pie->colors_used.slow_rop;

	/* Write out begin_image & its preamble for this band */
	if (!(pcls->known & begin_image_known)) {
	    gs_logical_operation_t lop = pie->pis->log_op;
	    byte *dp;
	    byte *bp = pie->begin_image_command +
		pie->begin_image_command_length;
	    uint len;
	    byte image_op = cmd_opv_begin_image;

	    /* Make sure the imager state is up to date. */
	    TRY_RECT {
	        code = (pie->color_map_is_known ? 0 :
			cmd_put_color_mapping(cdev, pie->pis));
		pie->color_map_is_known = true;
		if (code >= 0) {
		    uint want_known = ctm_known | clip_path_known |
				op_bm_tk_known | opacity_alpha_known |
				shape_alpha_known | alpha_known |
				(pie->color_space.id == gs_no_id ? 0 :
							 color_space_known);

		    code = cmd_do_write_unknown(cdev, pcls, want_known);
		}
		if (code >= 0)
		    code = cmd_do_enable_clip(cdev, pcls, pie->pcpath != NULL);
		if (code >= 0)
		    code = cmd_update_lop(cdev, pcls, lop);
	    } HANDLE_RECT(code);
	    if (pie->uses_color) {
 	        TRY_RECT {
		    code = cmd_put_drawing_color(cdev, pcls, &pie->dcolor);
		} HANDLE_RECT(code);
	    }
	    if (entire_box.p.x != 0 || entire_box.p.y != 0 ||
		entire_box.q.x != pie->image.Width ||
		entire_box.q.y != pie->image.Height
		) {
		image_op = cmd_opv_begin_image_rect;
		cmd_put2w(entire_box.p.x, entire_box.p.y, bp);
		cmd_put2w(pie->image.Width - entire_box.q.x,
			  pie->image.Height - entire_box.q.y, bp);
 	        }
	    len = bp - pie->begin_image_command;
	    TRY_RECT {
		code =
		    set_cmd_put_op(dp, cdev, pcls, image_op, 1 + len);
	    } HANDLE_RECT(code);
	    memcpy(dp + 1, pie->begin_image_command, len);
 
	    /* Mark band's begin_image as known */
	    pcls->known |= begin_image_known;
	}

	/*
	 * The data that we write out must use the X values set by
	 * begin_image, which may cover a larger interval than the ones
	 * actually needed for these particular scan lines if the image is
	 * rotated.
	 */
	{
	    /*
	     * image_band_box ensures that b{x,y}{0,1} fall within 
	     * pie->rect.
	     */
	    int bx0 = entire_box.p.x, bx1 = entire_box.q.x;
	    int by0 = ibox.p.y, by1 = ibox.q.y;
	    int bpp = pie->bits_per_plane;
	    int num_planes = pie->num_planes;
	    uint offsets[gs_image_max_planes];
	    int i, iy, ih, xskip, xoff, nrows;
	    uint bytes_per_plane, bytes_per_row, rows_per_cmd;

	    if (by0 < y0)
		by0 = y0;
	    if (by1 > y1)
		by1 = y1;
	    /*
	     * Make sure we're skipping an integral number of pixels, by
	     * truncating the initial X coordinate to the next lower
	     * value that is an exact multiple of a byte.
	     */
	    xoff = bx0 - pie->rect.p.x;
	    xskip = xoff & -(int)"\001\010\004\010\002\010\004\010"[bpp & 7];
	    for (i = 0; i < num_planes; ++i)
		offsets[i] =
		    (by0 - y0) * planes[i].raster + ((xskip * bpp) >> 3);
	    bytes_per_plane = ((bx1 - (pie->rect.p.x + xskip)) * bpp + 7) >> 3;
	    bytes_per_row = bytes_per_plane * pie->num_planes;
	    rows_per_cmd =
		(cbuf_size - cmd_largest_size) / max(bytes_per_row, 1);

	    if (rows_per_cmd == 0) {
		/* The reader will have to buffer a row separately. */
		rows_per_cmd = 1;
	    }
	    for (iy = by0, ih = by1 - by0; ih > 0; iy += nrows, ih -= nrows) {
		nrows = min(ih, rows_per_cmd);
		TRY_RECT {
		    code = cmd_image_plane_data(cdev, pcls, planes, info,
						bytes_per_plane, offsets,
						xoff - xskip, nrows);
		} HANDLE_RECT(code);
		for (i = 0; i < num_planes; ++i)
		    offsets[i] += planes[i].raster * nrows;
	    }
	}
    } END_RECTS_ON_ERROR(
	BEGIN
	    ++cdev->ignore_lo_mem_warnings;
	    NEST_RECT {
		code = write_image_end_all(dev, pie);
	    } UNNEST_RECT;
	    --cdev->ignore_lo_mem_warnings;
	    /* Update sub-rect */
	    if (!pie->image.Interpolate)
	        pie->rect.p.y += yh_used;  /* interpolate & mem recovery currently incompat */
	END,
	(code < 0 ? (band_code = code) : code) >= 0,
	(cmd_clear_known(cdev,
			 clist_image_unknowns(dev, pie) | begin_image_known),
	 pie->color_map_is_known = false,
	 cdev->image_enum_id = pie->id, true)
	);
 done:
    *rows_used = pie->y - y_orig;
    return pie->y >= pie->rect.q.y;
}

/* Clean up by releasing the buffers. */
private int
clist_image_end_image(gx_image_enum_common_t * info, bool draw_last)
{
    gx_device *dev = info->dev;
    gx_device_clist_writer * const cdev =
	&((gx_device_clist *)dev)->writer;
    clist_image_enum *pie = (clist_image_enum *) info;
    int code;

#ifdef DEBUG
    if (pie->id != cdev->image_enum_id) {
	lprintf2("end_image id = %lu != clist image id = %lu!\n",
		 (ulong) pie->id, (ulong) cdev->image_enum_id);
	return_error(gs_error_Fatal);
    }
#endif
    NEST_RECT {
	do {
	    code = write_image_end_all(dev, pie);
	} while (code < 0 && cdev->error_is_retryable &&
		 (code = clist_VMerror_recover(cdev, code)) >= 0
		 );
	/* if couldn't write successsfully, do a hard flush */
	if (code < 0 && cdev->error_is_retryable) {
	    int retry_code;
	    ++cdev->ignore_lo_mem_warnings;
	    retry_code = write_image_end_all(dev, pie); /* force it out */
	    --cdev->ignore_lo_mem_warnings;
	    if (retry_code >= 0 && cdev->driver_call_nesting == 0)
		code = clist_VMerror_recover_flush(cdev, code);
	}
    } UNNEST_RECT;
    cdev->image_enum_id = gs_no_id;
    gs_free_object(pie->memory, pie, "clist_image_end_image");
    return code;
}

/* Create a compositor device. */
int
clist_create_compositor(gx_device * dev,
			gx_device ** pcdev, const gs_composite_t * pcte,
			gs_imager_state * pis, gs_memory_t * mem)
{
    byte * dp;
    uint size = 0;
    int code = pcte->type->procs.write(pcte, 0, &size);

    /* determine the amount of space required */
    if (code < 0 && code != gs_error_rangecheck)
        return code;
    size += 2 + 1;      /* 2 bytes for the command code, one for the id */

    /* Create a compositor device for clist writing (if needed) */
    code = pcte->type->procs.clist_compositor_write_update(pcte, dev,
		    					pcdev, pis, mem);
    if (code < 0)
        return code;

    if (pcte->type->comp_id == GX_COMPOSITOR_PDF14_TRANS) {
	gx_device_clist_writer * const cldev =
			&((gx_device_clist *)dev)->writer;
	int len = cmd_write_ctm_return_length(cldev, &ctm_only(pis));

	code = set_cmd_put_all_op(dp, (gx_device_clist_writer *)dev,
                           cmd_opv_set_ctm, len + 1);
	if (code < 0)
	    return code;
	/* fixme: would like to set pcls->known for covered bands. */
	code = cmd_write_ctm(&ctm_only(pis), dp, len);
	if (code < 0)
	    return code;
    }

    /* overprint applies to all bands */
    /* fixme: optimize: the pdf14 compositor could be applied 
       only to bands covered by the pcte->params.bbox. */
    code = set_cmd_put_all_op( dp,
                               (gx_device_clist_writer *)dev,
                               cmd_opv_extend,
                               size );
    if (code < 0)
        return code;

    /* insert the command and compositor identifier */
    dp[1] = cmd_opv_ext_create_compositor;
    dp[2] = pcte->type->comp_id;

    /* serialize the remainder of the compositor */
    if ((code = pcte->type->procs.write(pcte, dp + 3, &size)) < 0)
        ((gx_device_clist_writer *)dev)->cnext = dp;
    return code;
}

/* ------ Utilities ------ */

/* Add a command to set data_x. */
private int
cmd_put_set_data_x(gx_device_clist_writer * cldev, gx_clist_state * pcls,
		   int data_x)
{
    byte *dp;
    int code;

    if (data_x > 0x1f) {
	int dx_msb = data_x >> 5;

	code = set_cmd_put_op(dp, cldev, pcls, cmd_opv_set_misc,
			      2 + cmd_size_w(dx_msb));
	if (code >= 0) {
	    dp[1] = cmd_set_misc_data_x + 0x20 + (data_x & 0x1f);
	    cmd_put_w(dx_msb, dp + 2);
	}
    } else {
	code = set_cmd_put_op(dp, cldev, pcls, cmd_opv_set_misc, 2);
	if (code >= 0)
	    dp[1] = cmd_set_misc_data_x + data_x;
    }
    return code;
}

/* Add commands to represent a full (device) halftone. */
int
cmd_put_halftone(gx_device_clist_writer * cldev, const gx_device_halftone * pdht)
{
    uint    ht_size = 0, req_size;
    byte *  dp;
    byte *  dp0 = 0;
    byte *  pht_buff = 0;
    int     code = gx_ht_write(pdht, (gx_device *)cldev, 0, &ht_size);

    /*
     * Determine the required size, and if necessary allocate a buffer.
     *
     * The full serialized representation consists of:
     *  command code (2 bytes)
     *  length of serialized halftone (enc_u_sizew(ht_size)
     *  one or more halfton segments, which consist of:
     *    command code (2 bytes)
     *    segment size (enc_u_sizew(seg_size) (seg_size < cbuf_ht_seg_max_size)
     *    the serialized halftone segment (seg_size)
     *
     * Serialized halftones may be larger than the command buffer, so it
     * is sent in segments. The cmd_opv_extend/cmd_opv_ext_put_halftone
     * combination indicates that a device halftone is being sent, and
     * provides the length of the entire halftone. This is followed by
     * one or more cmd_opv_extend/cmd_opv_ext_ht_seg commands, which
     * convey the segments of the serialized hafltone. The reader can
     * identify the final segment by adding segment lengths.
     *
     * This complexity is hidden from the serialization code. If the
     * halftone is larger than a single halftone buffer, we allocate a
     * buffer to hold the entire representation, and divided into
     * segments in this routine.
     */
    if (code < 0 && code != gs_error_rangecheck)
        return code;
    req_size = 2 + enc_u_sizew(ht_size);

    /* output the "put halftone" command */
    if ((code = set_cmd_put_all_op(dp, cldev, cmd_opv_extend, req_size)) < 0)
        return code;
    dp[1] = cmd_opv_ext_put_halftone;
    dp += 2;
    enc_u_putw(ht_size, dp);

    /* see if a spearate allocated buffer is required */
    if (ht_size > cbuf_ht_seg_max_size) {
        pht_buff = gs_alloc_bytes( cldev->bandlist_memory,
                                   ht_size,
                                   "cmd_put_halftone" );
        if (pht_buff == 0)
            return_error(gs_error_VMerror);
    } else {
        /* send the only segment command */
        req_size += ht_size;
        code = set_cmd_put_all_op(dp, cldev, cmd_opv_extend, req_size);
        if (code < 0)
            return code;
        dp0 = dp;
        dp[1] = cmd_opv_ext_put_ht_seg;
        dp += 2;
        enc_u_putw(ht_size, dp);
        pht_buff = dp;
    }

    /* serialize the halftone */
    code = gx_ht_write(pdht, (gx_device *)cldev, pht_buff, &ht_size);
    if (code < 0) {
        if (ht_size > cbuf_ht_seg_max_size)
            gs_free_object( cldev->bandlist_memory,
                            pht_buff,
                            "cmd_put_halftone" );
        else
            cldev->cnext = dp0;
        return code;
    }

    /*
     * If the halftone fit into a single command buffer, we are done.
     * Otherwise, process the individual segments.
     *
     * If bandlist memory is exhausted while processing the segments,
     * we do not make any attempt to recover the partially submitted
     * halftone. The reader will discard any partially sent hafltone
     * when it receives the next cmd_opv_extend/
     * cmd_opv_ext_put_halftone combination.
     */
    if (ht_size > cbuf_ht_seg_max_size) {
        byte *  pbuff = pht_buff;

        while (ht_size > 0 && code >= 0) {
            int     seg_size, tmp_size;

            seg_size = ( ht_size > cbuf_ht_seg_max_size ? cbuf_ht_seg_max_size
                                                        : ht_size );
            tmp_size = 2 + enc_u_sizew(seg_size) + seg_size;
            code = set_cmd_put_all_op(dp, cldev, cmd_opv_extend, tmp_size);
            if (code >= 0) {
                dp[1] = cmd_opv_ext_put_ht_seg;
                dp += 2;
                enc_u_putw(seg_size, dp);
                memcpy(dp, pbuff, seg_size);
                ht_size -= seg_size;
                pbuff += seg_size;
            }
        }
        gs_free_object( cldev->bandlist_memory, pht_buff, "cmd_put_halftone");
        pht_buff = 0;
    }

    if (code >= 0)
        cldev->device_halftone_id = pdht->id;

    return code;
}

/* Write out any necessary color mapping data. */
int
cmd_put_color_mapping(gx_device_clist_writer * cldev,
		      const gs_imager_state * pis)
{
    int code;
    const gx_device_halftone *pdht = pis->dev_ht;

    /* Put out the halftone. */
    if (pdht->id != cldev->device_halftone_id) {
	code = cmd_put_halftone(cldev, pdht);
	if (code < 0)
	    return code;
	cldev->device_halftone_id = pdht->id;
    }
    /* Put the under color removal and black generation functions */
    code = cmd_put_color_map(cldev, cmd_map_black_generation,
				 0, pis->black_generation,
				 &cldev->black_generation_id);
    if (code < 0)
	return code;
    code = cmd_put_color_map(cldev, cmd_map_undercolor_removal,
				 0, pis->undercolor_removal,
				 &cldev->undercolor_removal_id);
    if (code < 0)
	return code;
    /* Now put out the transfer functions. */
    {
	uint which = 0;
	bool all_same = true;
	bool send_default_comp = false;
	int i;
	gs_id default_comp_id, xfer_ids[4];

	/*
	 * Determine the ids for the transfer functions that we currently
	 * have in the set_transfer structure.  The halftone xfer funcs
	 * are sent in cmd_put_halftone.
	 */
#define get_id(pis, color, color_num) \
    ((pis->set_transfer.color != NULL && pis->set_transfer.color_num >= 0) \
	? pis->set_transfer.color->id\
	: pis->set_transfer.gray->id)

        xfer_ids[0] = get_id(pis, red, red_component_num);
        xfer_ids[1] = get_id(pis, green, green_component_num);
        xfer_ids[2] = get_id(pis, blue, blue_component_num);
	xfer_ids[3] = default_comp_id = pis->set_transfer.gray->id;
#undef get_id

	for (i = 0; i < countof(cldev->transfer_ids); ++i) {
	    if (xfer_ids[i] != cldev->transfer_ids[i])
		which |= 1 << i;
	    if (xfer_ids[i] != default_comp_id)
		all_same = false;
	    if (xfer_ids[i] == default_comp_id &&
		cldev->transfer_ids[i] != default_comp_id)
		send_default_comp = true;
	}
	/* There are 3 cases for transfer functions: nothing to write, */
	/* a single function, and multiple functions. */
	if (which == 0)
	    return 0;
	/*
	 * Send default transfer function if changed or we need it for a
	 * component
	 */
	if (send_default_comp || cldev->transfer_ids[0] != default_comp_id) {
	    gs_id dummy = gs_no_id;

	    code = cmd_put_color_map(cldev, cmd_map_transfer, 0,
		pis->set_transfer.gray, &dummy);
	    if (code < 0)
		return code;
	    /* Sending a default will force all xfers to default */
	    for (i = 0; i < countof(cldev->transfer_ids); ++i)
		cldev->transfer_ids[i] = default_comp_id;
	}
	/* Send any transfer functions which have changed */
	if (cldev->transfer_ids[0] != xfer_ids[0]) {
	    code = cmd_put_color_map(cldev, cmd_map_transfer_0,
			pis->set_transfer.red_component_num,
			pis->set_transfer.red, &cldev->transfer_ids[0]);
	    if (code < 0)
		return code;
	}
	if (cldev->transfer_ids[1] != xfer_ids[1]) {
	    code = cmd_put_color_map(cldev, cmd_map_transfer_1,
			pis->set_transfer.green_component_num,
			pis->set_transfer.green, &cldev->transfer_ids[1]);
	    if (code < 0)
		return code;
	}
	if (cldev->transfer_ids[2] != xfer_ids[2]) {
	    code = cmd_put_color_map(cldev, cmd_map_transfer_2,
			pis->set_transfer.blue_component_num,
			pis->set_transfer.blue, &cldev->transfer_ids[2]);
	    if (code < 0)
		return code;
	}
    }

    return 0;
}

/*
 * Compute the subrectangle of an image that intersects a band;
 * return false if it is empty.
 * It is OK for this to be too large; in fact, with the present
 * algorithm, it will be quite a bit too large if the transformation isn't
 * well-behaved ("well-behaved" meaning either xy = yx = 0 or xx = yy = 0).
 */
#define I_FLOOR(x) ((int)floor(x))
#define I_CEIL(x) ((int)ceil(x))
private void
box_merge_point(gs_int_rect * pbox, floatp x, floatp y)
{
    int t;

    if ((t = I_FLOOR(x)) < pbox->p.x)
	pbox->p.x = t;
    if ((t = I_CEIL(x)) > pbox->q.x)
	pbox->q.x = t;
    if ((t = I_FLOOR(y)) < pbox->p.y)
	pbox->p.y = t;
    if ((t = I_CEIL(y)) > pbox->q.y)
	pbox->q.y = t;
}
private bool
image_band_box(gx_device * dev, const clist_image_enum * pie, int y, int h,
	       gs_int_rect * pbox)
{
    fixed by0 = int2fixed(y);
    fixed by1 = int2fixed(y + h);
    int
        px = pie->rect.p.x, py = pie->rect.p.y,
	qx = pie->rect.q.x, qy = pie->rect.q.y;
    gs_fixed_rect cbox;		/* device clipping box */
    gs_rect bbox;		/* cbox intersected with band */

    /* Intersect the device clipping box and the band. */
    (*dev_proc(dev, get_clipping_box)) (dev, &cbox);
    /* The fixed_half here is to allow for adjustment. */
    bbox.p.x = fixed2float(cbox.p.x - fixed_half);
    bbox.q.x = fixed2float(cbox.q.x + fixed_half);
    bbox.p.y = fixed2float(max(cbox.p.y, by0) - fixed_half);
    bbox.q.y = fixed2float(min(cbox.q.y, by1) + fixed_half);
#ifdef DEBUG
    if (gs_debug_c('b')) {
	dlprintf6("[b]band box for (%d,%d),(%d,%d), band (%d,%d) =>\n",
		  px, py, qx, qy, y, y + h);
	dlprintf10("      (%g,%g),(%g,%g), matrix=[%g %g %g %g %g %g]\n",
		   bbox.p.x, bbox.p.y, bbox.q.x, bbox.q.y,
		   pie->matrix.xx, pie->matrix.xy, pie->matrix.yx,
		   pie->matrix.yy, pie->matrix.tx, pie->matrix.ty);
    }
#endif
    if (is_xxyy(&pie->matrix) || is_xyyx(&pie->matrix)) {
	/*
	 * The inverse transform of the band is a rectangle aligned with
	 * the coordinate axes, so we can just intersect it with the
	 * image subrectangle.
	 */
	gs_rect ibox;		/* bbox transformed back to image space */

	if (gs_bbox_transform_inverse(&bbox, &pie->matrix, &ibox) < 0)
	    return false;
	pbox->p.x = max(px, I_FLOOR(ibox.p.x));
	pbox->q.x = min(qx, I_CEIL(ibox.q.x));
	pbox->p.y = max(py, I_FLOOR(ibox.p.y));
	pbox->q.y = min(qy, I_CEIL(ibox.q.y));
    } else {
	/*
	 * The inverse transform of the band is not aligned with the
	 * axes, i.e., is a general parallelogram.  To compute an exact
	 * bounding box, we need to find the intersections of this
	 * parallelogram with the image subrectangle.
	 *
	 * There is probably a much more efficient way to do this
	 * computation, but we don't know what it is.
	 */
	gs_point rect[4];
	gs_point corners[5];
	int i;

	/* Store the corners of the image rectangle. */
	rect[0].x = rect[3].x = px;
	rect[1].x = rect[2].x = qx;
	rect[0].y = rect[1].y = py;
	rect[2].y = rect[3].y = qy;
	/*
	 * Compute the corners of the clipped band in image space.  If
	 * the matrix is singular or an overflow occurs, the result will
	 * be nonsense: in this case, there isn't anything useful we
	 * can do, so return an empty intersection.
	 */
	if (gs_point_transform_inverse(bbox.p.x, bbox.p.y, &pie->matrix,
				       &corners[0]) < 0 ||
	    gs_point_transform_inverse(bbox.q.x, bbox.p.y, &pie->matrix,
				       &corners[1]) < 0 ||
	    gs_point_transform_inverse(bbox.q.x, bbox.q.y, &pie->matrix,
				       &corners[2]) < 0 ||
	    gs_point_transform_inverse(bbox.p.x, bbox.q.y, &pie->matrix,
				       &corners[3]) < 0
	    ) {
	    if_debug0('b', "[b]can't inverse-transform a band corner!\n");
	    return false;
	}
	corners[4] = corners[0];
	pbox->p.x = qx, pbox->p.y = qy;
	pbox->q.x = px, pbox->q.y = py;
	/*
	 * We iterate over both the image rectangle and the band
	 * parallelogram in a single loop for convenience, even though
	 * there is no coupling between the two.
	 */
	for (i = 0; i < 4; ++i) {
	    gs_point pa, pt;
	    double dx, dy;

	    /* Check the image corner for being inside the band. */
	    pa = rect[i];
	    gs_point_transform(pa.x, pa.y, &pie->matrix, &pt);
	    if (pt.x >= bbox.p.x && pt.x <= bbox.q.x &&
		pt.y >= bbox.p.y && pt.y <= bbox.q.y
		)
		box_merge_point(pbox, pa.x, pa.y);
	    /* Check the band corner for being inside the image. */
	    pa = corners[i];
	    if (pa.x >= px && pa.x <= qx && pa.y >= py && pa.y <= qy)
		box_merge_point(pbox, pa.x, pa.y);
	    /* Check for intersections of band edges with image edges. */
	    dx = corners[i + 1].x - pa.x;
	    dy = corners[i + 1].y - pa.y;
#define in_range(t, tc, p, q)\
  (0 <= t && t <= 1 && (t = tc) >= p && t <= q)
	    if (dx != 0) {
		double t = (px - pa.x) / dx;

		if_debug3('b', "   (px) t=%g => (%d,%g)\n",
			  t, px, pa.y + t * dy);
		if (in_range(t, pa.y + t * dy, py, qy))
		    box_merge_point(pbox, (floatp) px, t);
		t = (qx - pa.x) / dx;
		if_debug3('b', "   (qx) t=%g => (%d,%g)\n",
			  t, qx, pa.y + t * dy);
		if (in_range(t, pa.y + t * dy, py, qy))
		    box_merge_point(pbox, (floatp) qx, t);
	    }
	    if (dy != 0) {
		double t = (py - pa.y) / dy;

		if_debug3('b', "   (py) t=%g => (%g,%d)\n",
			  t, pa.x + t * dx, py);
		if (in_range(t, pa.x + t * dx, px, qx))
		    box_merge_point(pbox, t, (floatp) py);
		t = (qy - pa.y) / dy;
		if_debug3('b', "   (qy) t=%g => (%g,%d)\n",
			  t, pa.x + t * dx, qy);
		if (in_range(t, pa.x + t * dx, px, qx))
		    box_merge_point(pbox, t, (floatp) qy);
	    }
#undef in_range
	}
    }
    if_debug4('b', "    => (%d,%d),(%d,%d)\n", pbox->p.x, pbox->p.y,
	      pbox->q.x, pbox->q.y);
    /*
     * If necessary, add pixels around the edges so we will have
     * enough information to do interpolation.
     */
    if ((pbox->p.x -= pie->support.x) < pie->rect.p.x)
	pbox->p.x = pie->rect.p.x;
    if ((pbox->p.y -= pie->support.y) < pie->rect.p.y)
	pbox->p.y = pie->rect.p.y;
    if ((pbox->q.x += pie->support.x) > pie->rect.q.x)
	pbox->q.x = pie->rect.q.x;
    if ((pbox->q.y += pie->support.y) > pie->rect.q.y)
	pbox->q.y = pie->rect.q.y;
    return (pbox->p.x < pbox->q.x && pbox->p.y < pbox->q.y);
}

/* Determine which image-related properties are unknown */
private uint	/* mask of unknown properties(see pcls->known) */
clist_image_unknowns(gx_device *dev, const clist_image_enum *pie)
{
    gx_device_clist_writer * const cdev =
	&((gx_device_clist *)dev)->writer;
    const gs_imager_state *const pis = pie->pis;
    uint unknown = 0;

    /*
     * Determine if the CTM, color space, and clipping region (and, for
     * masked images or images with CombineWithColor, the current color)
     * are unknown. Set the device state in anticipation of the values
     * becoming known.
     */
    if (cdev->imager_state.ctm.xx != pis->ctm.xx ||
	cdev->imager_state.ctm.xy != pis->ctm.xy ||
	cdev->imager_state.ctm.yx != pis->ctm.yx ||
	cdev->imager_state.ctm.yy != pis->ctm.yy ||
	cdev->imager_state.ctm.tx != pis->ctm.tx ||
	cdev->imager_state.ctm.ty != pis->ctm.ty
	) {
	unknown |= ctm_known;
	cdev->imager_state.ctm = pis->ctm;
    }
    if (pie->color_space.id == gs_no_id) { /* masked image */
	cdev->color_space.space = 0; /* for GC */
    } else {			/* not masked */
	if (cdev->color_space.id == pie->color_space.id) {
	    /* The color space pointer might not be valid: update it. */
	    cdev->color_space.space = pie->color_space.space;
	} else {
	    unknown |= color_space_known;
	    cdev->color_space = pie->color_space;
	}
    }
    if (cmd_check_clip_path(cdev, pie->pcpath))
	unknown |= clip_path_known;
    /*
     * Note: overprint and overprint_mode are implemented via a compositor
     * device, which is passed separately through the command list. Hence,
     * though both parameters are passed in the state as well, this usually
     * has no effect.
     */
    if (cdev->imager_state.overprint != pis->overprint ||
        cdev->imager_state.overprint_mode != pis->overprint_mode ||
        cdev->imager_state.blend_mode != pis->blend_mode ||
        cdev->imager_state.text_knockout != pis->text_knockout) {
	unknown |= op_bm_tk_known;
        cdev->imager_state.overprint = pis->overprint;
        cdev->imager_state.overprint_mode = pis->overprint_mode;
        cdev->imager_state.blend_mode = pis->blend_mode;
        cdev->imager_state.text_knockout = pis->text_knockout;
    }
    if (cdev->imager_state.opacity.alpha != pis->opacity.alpha) {
	unknown |= opacity_alpha_known;
        cdev->imager_state.opacity.alpha = pis->opacity.alpha;
    }
    if (cdev->imager_state.shape.alpha != pis->shape.alpha) {
	unknown |= shape_alpha_known;
        cdev->imager_state.shape.alpha = pis->shape.alpha;
    }
    if (cdev->imager_state.alpha != pis->alpha) {
	unknown |= alpha_known;
        cdev->imager_state.alpha = pis->alpha;
    }
    return unknown;
}

/* Construct the begin_image command. */
private int
begin_image_command(byte *buf, uint buf_size, const gs_image_common_t *pic)
{
    int i;
    stream s;
    const gs_color_space *ignore_pcs;
    int code;

    for (i = 0; i < gx_image_type_table_count; ++i)
	if (gx_image_type_table[i] == pic->type)
	    break;
    if (i >= gx_image_type_table_count)
	return_error(gs_error_rangecheck);
    s_init(&s, NULL);
    swrite_string(&s, buf, buf_size);
    sputc(&s, (byte)i);
    code = pic->type->sput(pic, &s, &ignore_pcs);
    return (code < 0 ? code : stell(&s));
}

/* Write data for a partial image. */
private int
cmd_image_plane_data(gx_device_clist_writer * cldev, gx_clist_state * pcls,
		     const gx_image_plane_t * planes,
		     const gx_image_enum_common_t * pie,
		     uint bytes_per_plane, const uint * offsets,
		     int dx, int h)
{
    int data_x = planes[0].data_x + dx;
    uint nbytes = bytes_per_plane * pie->num_planes * h;
    uint len = 1 + cmd_size2w(h, bytes_per_plane) + nbytes;
    byte *dp;
    uint offset = 0;
    int plane, i;
    int code;

    if (data_x) {
	code = cmd_put_set_data_x(cldev, pcls, data_x);
	if (code < 0)
	    return code;
	offset = ((data_x & ~7) * cldev->color_info.depth) >> 3;
    }
    code = set_cmd_put_op(dp, cldev, pcls, cmd_opv_image_data, len);
    if (code < 0)
	return code;
    dp++;
    cmd_put2w(h, bytes_per_plane, dp);
    for (plane = 0; plane < pie->num_planes; ++plane)
	for (i = 0; i < h; ++i) {
	    memcpy(dp,
		   planes[plane].data + i * planes[plane].raster +
		   offsets[plane] + offset,
		   bytes_per_plane);
	    dp += bytes_per_plane;
	}
    return 0;
}

/* Write image_end commands into all bands */
private int	/* ret 0 ok, else -ve error status */
write_image_end_all(gx_device *dev, const clist_image_enum *pie)
{
    gx_device_clist_writer * const cdev =
	&((gx_device_clist *)dev)->writer;
    int code;
    int y = pie->ymin;
    int height = pie->ymax - y;

    /*
     * We need to check specially for images lying entirely outside the
     * page, since FOR_RECTS doesn't do this.
     */
    if (height <= 0)
	return 0;
    FOR_RECTS {
	byte *dp;

	if (!(pcls->known & begin_image_known))
	    continue;
	TRY_RECT {
	    if_debug1('L', "[L]image_end for band %d\n", band);
	    code = set_cmd_put_op(dp, cdev, pcls, cmd_opv_image_data, 2);
	} HANDLE_RECT(code);
	dp[1] = 0;	    /* EOD */
	pcls->known ^= begin_image_known;
    } END_RECTS;
    return 0;
}

/*
 * Compare a rectangle vs. clip path.  Return true if there is no clipping
 * path, if the rectangle is unclipped, or if the clipping path is a
 * rectangle and intersects the given rectangle.
 */
private bool
check_rect_for_trivial_clip(
    const gx_clip_path *pcpath,	/* May be NULL, clip to evaluate */
    int px, int py, int qx, int qy	/* corners of box to test */
)
{
    gs_fixed_rect obox;
    gs_fixed_rect imgbox;

    if (!pcpath)
	return true;

    imgbox.p.x = int2fixed(px);
    imgbox.p.y = int2fixed(py);
    imgbox.q.x = int2fixed(qx);
    imgbox.q.y = int2fixed(qy);
    if (gx_cpath_includes_rectangle(pcpath,
				    imgbox.p.x, imgbox.p.y,
				    imgbox.q.x, imgbox.q.y))
	return true;

    return (gx_cpath_outer_box(pcpath, &obox) /* cpath is rectangle */ &&
	    obox.p.x <= imgbox.q.x && obox.q.x >= imgbox.p.x &&
	    obox.p.y <= imgbox.q.y && obox.q.y >= imgbox.p.y );
}