xref: /plan9/sys/src/cmd/gs/src/gdevm1.c (revision 593dc095aefb2a85c828727bbfa9da139a49bdf4)

/* Copyright (C) 1989, 1995, 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: gdevm1.c,v 1.6 2002/10/07 08:28:56 ghostgum Exp $ */
/* Monobit "memory" (stored bitmap) device */
#include "memory_.h"
#include "gx.h"
#include "gxdevice.h"
#include "gxdevmem.h"		/* semi-public definitions */
#include "gdevmem.h"		/* private definitions */

/* Optionally, use the slow RasterOp implementations for testing. */
/*#define USE_COPY_ROP */

#ifdef USE_COPY_ROP
#include "gsrop.h"
#endif

/* ================ Standard (byte-oriented) device ================ */

/* Procedures */
private dev_proc_map_rgb_color(mem_mono_map_rgb_color);
private dev_proc_map_color_rgb(mem_mono_map_color_rgb);
private dev_proc_copy_mono(mem_mono_copy_mono);
private dev_proc_fill_rectangle(mem_mono_fill_rectangle);
private dev_proc_strip_tile_rectangle(mem_mono_strip_tile_rectangle);

/* The device descriptor. */
/* The instance is public. */
const gx_device_memory mem_mono_device =
mem_full_alpha_device("image1", 0, 1, mem_open,
		      mem_mono_map_rgb_color, mem_mono_map_color_rgb,
	 mem_mono_copy_mono, gx_default_copy_color, mem_mono_fill_rectangle,
		      gx_default_map_cmyk_color, gx_no_copy_alpha,
		      mem_mono_strip_tile_rectangle, mem_mono_strip_copy_rop,
		      mem_get_bits_rectangle);

/* Map color to/from RGB.  This may be inverted. */
private gx_color_index
mem_mono_map_rgb_color(gx_device * dev, const gx_color_value cv[])
{
    gx_device_memory * const mdev = (gx_device_memory *)dev;
    return (gx_default_w_b_map_rgb_color(dev, cv) ^ mdev->palette.data[0]) & 1;
}

private int
mem_mono_map_color_rgb(gx_device * dev, gx_color_index color,
		       gx_color_value prgb[3])
{
    gx_device_memory * const mdev = (gx_device_memory *)dev;
    /* NB code doesn't make sense... map_color_rgb procedures return an error code */
    return (gx_default_w_b_map_color_rgb(dev, color, prgb) ^ mdev->palette.data[0]) & 1;
}

/* Fill a rectangle with a color. */
private int
mem_mono_fill_rectangle(gx_device * dev, int x, int y, int w, int h,
			gx_color_index color)
{
    gx_device_memory * const mdev = (gx_device_memory *)dev;

#ifdef USE_COPY_ROP
    return mem_mono_copy_rop(dev, NULL, 0, 0, gx_no_bitmap_id, NULL,
			     NULL, NULL,
			     x, y, w, h, 0, 0,
			     (color ? rop3_1 : rop3_0));
#else
    fit_fill(dev, x, y, w, h);
    bits_fill_rectangle(scan_line_base(mdev, y), x, mdev->raster,
			-(int)(mono_fill_chunk) color, w, h);
    return 0;
#endif
}

/* Convert x coordinate to byte offset in scan line. */
#define x_to_byte(x) ((x) >> 3)

/* Copy a monochrome bitmap. */
#undef mono_masks
#define mono_masks mono_copy_masks

/*
 * Fetch a chunk from the source.
 *
 * Since source and destination are both always big-endian,
 * fetching an aligned chunk never requires byte swapping.
 */
#define CFETCH_ALIGNED(cptr)\
  (*(const chunk *)(cptr))

/*
 * Note that the macros always cast cptr,
 * so it doesn't matter what the type of cptr is.
 */
/* cshift = chunk_bits - shift. */
#undef chunk
#if arch_is_big_endian
#  define chunk uint
#  define CFETCH_RIGHT(cptr, shift, cshift)\
	(CFETCH_ALIGNED(cptr) >> shift)
#  define CFETCH_LEFT(cptr, shift, cshift)\
	(CFETCH_ALIGNED(cptr) << shift)
#  define CFETCH_USES_CSKEW 0
/* Fetch a chunk that straddles a chunk boundary. */
#  define CFETCH2(cptr, cskew, skew)\
    (CFETCH_LEFT(cptr, cskew, skew) +\
     CFETCH_RIGHT((const chunk *)(cptr) + 1, skew, cskew))
#else /* little-endian */
#  define chunk bits16
private const bits16 right_masks2[9] =
{
    0xffff, 0x7f7f, 0x3f3f, 0x1f1f, 0x0f0f, 0x0707, 0x0303, 0x0101, 0x0000
};
private const bits16 left_masks2[9] =
{
    0xffff, 0xfefe, 0xfcfc, 0xf8f8, 0xf0f0, 0xe0e0, 0xc0c0, 0x8080, 0x0000
};

#  define CCONT(cptr, off) (((const chunk *)(cptr))[off])
#  define CFETCH_RIGHT(cptr, shift, cshift)\
	((shift) < 8 ?\
	 ((CCONT(cptr, 0) >> (shift)) & right_masks2[shift]) +\
	  (CCONT(cptr, 0) << (cshift)) :\
	 ((chunk)*(const byte *)(cptr) << (cshift)) & 0xff00)
#  define CFETCH_LEFT(cptr, shift, cshift)\
	((shift) < 8 ?\
	 ((CCONT(cptr, 0) << (shift)) & left_masks2[shift]) +\
	  (CCONT(cptr, 0) >> (cshift)) :\
	 ((CCONT(cptr, 0) & 0xff00) >> (cshift)) & 0xff)
#  define CFETCH_USES_CSKEW 1
/* Fetch a chunk that straddles a chunk boundary. */
/* We can avoid testing the shift amount twice */
/* by expanding the CFETCH_LEFT/right macros in-line. */
#  define CFETCH2(cptr, cskew, skew)\
	((cskew) < 8 ?\
	 ((CCONT(cptr, 0) << (cskew)) & left_masks2[cskew]) +\
	  (CCONT(cptr, 0) >> (skew)) +\
	  (((chunk)(((const byte *)(cptr))[2]) << (cskew)) & 0xff00) :\
	 (((CCONT(cptr, 0) & 0xff00) >> (skew)) & 0xff) +\
	  ((CCONT(cptr, 1) >> (skew)) & right_masks2[skew]) +\
	   (CCONT(cptr, 1) << (cskew)))
#endif

typedef enum {
    COPY_OR = 0, COPY_STORE, COPY_AND, COPY_FUNNY
} copy_function;
typedef struct {
    int invert;
    copy_function op;
} copy_mode;

/*
 * Map from <color0,color1> to copy_mode.
 * Logically, this is a 2-D array.
 * The indexing is (transparent, 0, 1, unused). */
private const copy_mode copy_modes[16] = {
    {~0, COPY_FUNNY},		/* NN */
    {~0, COPY_AND},		/* N0 */
    {0, COPY_OR},		/* N1 */
    {0, 0},			/* unused */
    {0, COPY_AND},		/* 0N */
    {0, COPY_FUNNY},		/* 00 */
    {0, COPY_STORE},		/* 01 */
    {0, 0},			/* unused */
    {~0, COPY_OR},		/* 1N */
    {~0, COPY_STORE},		/* 10 */
    {0, COPY_FUNNY},		/* 11 */
    {0, 0},			/* unused */
    {0, 0},			/* unused */
    {0, 0},			/* unused */
    {0, 0},			/* unused */
    {0, 0},			/* unused */
};

/* Handle the funny cases that aren't supposed to happen. */
#define FUNNY_CASE()\
  (invert ? gs_note_error(-1) :\
   mem_mono_fill_rectangle(dev, x, y, w, h, color0))

private int
mem_mono_copy_mono(gx_device * dev,
 const byte * source_data, int source_x, int source_raster, gx_bitmap_id id,
   int x, int y, int w, int h, gx_color_index color0, gx_color_index color1)
{
    gx_device_memory * const mdev = (gx_device_memory *)dev;

#ifdef USE_COPY_ROP
    return mem_mono_copy_rop(dev, source_data, source_x, source_raster,
			     id, NULL, NULL, NULL,
			     x, y, w, h, 0, 0,
			     ((color0 == gx_no_color_index ? rop3_D :
			       color0 == 0 ? rop3_0 : rop3_1) & ~rop3_S) |
			     ((color1 == gx_no_color_index ? rop3_D :
			       color1 == 0 ? rop3_0 : rop3_1) & rop3_S));
#else /* !USE_COPY_ROP */
    register const byte *bptr;	/* actually chunk * */
    int dbit, wleft;
    uint mask;
    copy_mode mode;

    DECLARE_SCAN_PTR_VARS(dbptr, byte *, dest_raster);
#define optr ((chunk *)dbptr)
    register int skew;
    register uint invert;

    fit_copy(dev, source_data, source_x, source_raster, id, x, y, w, h);
#if gx_no_color_index_value != -1	/* hokey! */
    if (color0 == gx_no_color_index)
	color0 = -1;
    if (color1 == gx_no_color_index)
	color1 = -1;
#endif
    mode = copy_modes[((int)color0 << 2) + (int)color1 + 5];
    invert = (uint)mode.invert;	/* load register */
    SETUP_RECT_VARS(dbptr, byte *, dest_raster);
    bptr = source_data + ((source_x & ~chunk_align_bit_mask) >> 3);
    dbit = x & chunk_align_bit_mask;
    skew = dbit - (source_x & chunk_align_bit_mask);

/* Macros for writing partial chunks. */
/* The destination pointer is always named optr, */
/* and must be declared as chunk *. */
/* CINVERT may be temporarily redefined. */
#define CINVERT(bits) ((bits) ^ invert)
#define WRITE_OR_MASKED(bits, mask, off)\
  optr[off] |= (CINVERT(bits) & mask)
#define WRITE_STORE_MASKED(bits, mask, off)\
  optr[off] = ((optr[off] & ~mask) | (CINVERT(bits) & mask))
#define WRITE_AND_MASKED(bits, mask, off)\
  optr[off] &= (CINVERT(bits) | ~mask)
/* Macros for writing full chunks. */
#define WRITE_OR(bits)  *optr |= CINVERT(bits)
#define WRITE_STORE(bits) *optr = CINVERT(bits)
#define WRITE_AND(bits) *optr &= CINVERT(bits)
/* Macro for incrementing to next chunk. */
#define NEXT_X_CHUNK()\
  bptr += chunk_bytes; dbptr += chunk_bytes
/* Common macro for the end of each scan line. */
#define END_Y_LOOP(sdelta, ddelta)\
  bptr += sdelta; dbptr += ddelta

    if ((wleft = w + dbit - chunk_bits) <= 0) {		/* The entire operation fits in one (destination) chunk. */
	set_mono_thin_mask(mask, w, dbit);

#define WRITE_SINGLE(wr_op, src)\
  for ( ; ; )\
   { wr_op(src, mask, 0);\
     if ( --h == 0 ) break;\
     END_Y_LOOP(source_raster, dest_raster);\
   }

#define WRITE1_LOOP(src)\
  switch ( mode.op ) {\
    case COPY_OR: WRITE_SINGLE(WRITE_OR_MASKED, src); break;\
    case COPY_STORE: WRITE_SINGLE(WRITE_STORE_MASKED, src); break;\
    case COPY_AND: WRITE_SINGLE(WRITE_AND_MASKED, src); break;\
    default: return FUNNY_CASE();\
  }

	if (skew >= 0) {	/* single -> single, right/no shift */
	    if (skew == 0) {	/* no shift */
		WRITE1_LOOP(CFETCH_ALIGNED(bptr));
	    } else {		/* right shift */
#if CFETCH_USES_CSKEW
		int cskew = chunk_bits - skew;
#endif

		WRITE1_LOOP(CFETCH_RIGHT(bptr, skew, cskew));
	    }
	} else if (wleft <= skew) {	/* single -> single, left shift */
#if CFETCH_USES_CSKEW
	    int cskew = chunk_bits + skew;
#endif

	    skew = -skew;
	    WRITE1_LOOP(CFETCH_LEFT(bptr, skew, cskew));
	} else {		/* double -> single */
	    int cskew = -skew;

	    skew += chunk_bits;
	    WRITE1_LOOP(CFETCH2(bptr, cskew, skew));
	}
#undef WRITE1_LOOP
#undef WRITE_SINGLE
    } else if (wleft <= skew) {	/* 1 source chunk -> 2 destination chunks. */
	/* This is an important special case for */
	/* both characters and halftone tiles. */
	uint rmask;
	int cskew = chunk_bits - skew;

	set_mono_left_mask(mask, dbit);
	set_mono_right_mask(rmask, wleft);
#undef CINVERT
#define CINVERT(bits) (bits)	/* pre-inverted here */

#if arch_is_big_endian		/* no byte swapping */
#  define WRITE_1TO2(wr_op)\
  for ( ; ; )\
   { register uint bits = CFETCH_ALIGNED(bptr) ^ invert;\
     wr_op(bits >> skew, mask, 0);\
     wr_op(bits << cskew, rmask, 1);\
     if ( --h == 0 ) break;\
     END_Y_LOOP(source_raster, dest_raster);\
   }
#else /* byte swapping */
#  define WRITE_1TO2(wr_op)\
  for ( ; ; )\
   { wr_op(CFETCH_RIGHT(bptr, skew, cskew) ^ invert, mask, 0);\
     wr_op(CFETCH_LEFT(bptr, cskew, skew) ^ invert, rmask, 1);\
     if ( --h == 0 ) break;\
     END_Y_LOOP(source_raster, dest_raster);\
   }
#endif

	switch (mode.op) {
	    case COPY_OR:
		WRITE_1TO2(WRITE_OR_MASKED);
		break;
	    case COPY_STORE:
		WRITE_1TO2(WRITE_STORE_MASKED);
		break;
	    case COPY_AND:
		WRITE_1TO2(WRITE_AND_MASKED);
		break;
	    default:
		return FUNNY_CASE();
	}
#undef CINVERT
#define CINVERT(bits) ((bits) ^ invert)
#undef WRITE_1TO2
    } else {			/* More than one source chunk and more than one */
	/* destination chunk are involved. */
	uint rmask;
	int words = (wleft & ~chunk_bit_mask) >> 3;
	uint sskip = source_raster - words;
	uint dskip = dest_raster - words;
	register uint bits;

	set_mono_left_mask(mask, dbit);
	set_mono_right_mask(rmask, wleft & chunk_bit_mask);
	if (skew == 0) {	/* optimize the aligned case */

#define WRITE_ALIGNED(wr_op, wr_op_masked)\
  for ( ; ; )\
   { int count = wleft;\
     /* Do first partial chunk. */\
     wr_op_masked(CFETCH_ALIGNED(bptr), mask, 0);\
     /* Do full chunks. */\
     while ( (count -= chunk_bits) >= 0 )\
      { NEXT_X_CHUNK(); wr_op(CFETCH_ALIGNED(bptr)); }\
     /* Do last chunk */\
     if ( count > -chunk_bits )\
      { wr_op_masked(CFETCH_ALIGNED(bptr + chunk_bytes), rmask, 1); }\
     if ( --h == 0 ) break;\
     END_Y_LOOP(sskip, dskip);\
   }

	    switch (mode.op) {
		case COPY_OR:
		    WRITE_ALIGNED(WRITE_OR, WRITE_OR_MASKED);
		    break;
		case COPY_STORE:
		    WRITE_ALIGNED(WRITE_STORE, WRITE_STORE_MASKED);
		    break;
		case COPY_AND:
		    WRITE_ALIGNED(WRITE_AND, WRITE_AND_MASKED);
		    break;
		default:
		    return FUNNY_CASE();
	    }
#undef WRITE_ALIGNED
	} else {		/* not aligned */
	    int cskew = -skew & chunk_bit_mask;
	    bool case_right =
	    (skew >= 0 ? true :
	     ((bptr += chunk_bytes), false));

	    skew &= chunk_bit_mask;

#define WRITE_UNALIGNED(wr_op, wr_op_masked)\
  /* Prefetch partial word. */\
  bits =\
    (case_right ? CFETCH_RIGHT(bptr, skew, cskew) :\
     CFETCH2(bptr - chunk_bytes, cskew, skew));\
  wr_op_masked(bits, mask, 0);\
  /* Do full chunks. */\
  while ( count >= chunk_bits )\
    { bits = CFETCH2(bptr, cskew, skew);\
      NEXT_X_CHUNK(); wr_op(bits); count -= chunk_bits;\
    }\
  /* Do last chunk */\
  if ( count > 0 )\
    { bits = CFETCH_LEFT(bptr, cskew, skew);\
      if ( count > skew ) bits += CFETCH_RIGHT(bptr + chunk_bytes, skew, cskew);\
      wr_op_masked(bits, rmask, 1);\
    }

	    switch (mode.op) {
		case COPY_OR:
		    for (;;) {
			int count = wleft;

			WRITE_UNALIGNED(WRITE_OR, WRITE_OR_MASKED);
			if (--h == 0)
			    break;
			END_Y_LOOP(sskip, dskip);
		    }
		    break;
		case COPY_STORE:
		    for (;;) {
			int count = wleft;

			WRITE_UNALIGNED(WRITE_STORE, WRITE_STORE_MASKED);
			if (--h == 0)
			    break;
			END_Y_LOOP(sskip, dskip);
		    }
		    break;
		case COPY_AND:
		    for (;;) {
			int count = wleft;

			WRITE_UNALIGNED(WRITE_AND, WRITE_AND_MASKED);
			if (--h == 0)
			    break;
			END_Y_LOOP(sskip, dskip);
		    }
		    break;
		default /*case COPY_FUNNY */ :
		    return FUNNY_CASE();
	    }
#undef WRITE_UNALIGNED
	}
    }
#undef END_Y_LOOP
#undef NEXT_X_CHUNK
    return 0;
#undef optr
#endif /* !USE_COPY_ROP */
}

/* Strip-tile with a monochrome halftone. */
/* This is a performance bottleneck for monochrome devices, */
/* so we re-implement it, even though it takes a lot of code. */
private int
mem_mono_strip_tile_rectangle(gx_device * dev,
			      register const gx_strip_bitmap * tiles,
int tx, int y, int tw, int th, gx_color_index color0, gx_color_index color1,
			      int px, int py)
{
    gx_device_memory * const mdev = (gx_device_memory *)dev;

#ifdef USE_COPY_ROP
    return mem_mono_strip_copy_rop(dev, NULL, 0, 0, tile->id, NULL,
				   tiles, NULL,
				   tx, y, tw, th, px, py,
				   ((color0 == gx_no_color_index ? rop3_D :
				 color0 == 0 ? rop3_0 : rop3_1) & ~rop3_T) |
				   ((color1 == gx_no_color_index ? rop3_D :
				  color1 == 0 ? rop3_0 : rop3_1) & rop3_T));
#else /* !USE_COPY_ROP */
    register uint invert;
    int source_raster;
    uint tile_bits_size;
    const byte *source_data;
    const byte *end;
    int x, rw, w, h;
    register const byte *bptr;	/* actually chunk * */
    int dbit, wleft;
    uint mask;
    byte *dbase;

    DECLARE_SCAN_PTR_VARS(dbptr, byte *, dest_raster);
#define optr ((chunk *)dbptr)
    register int skew;

    /* This implementation doesn't handle strips yet. */
    if (color0 != (color1 ^ 1) || tiles->shift != 0)
	return gx_default_strip_tile_rectangle(dev, tiles, tx, y, tw, th,
					       color0, color1, px, py);
    fit_fill(dev, tx, y, tw, th);
    invert = (uint)(-(int) color0);
    source_raster = tiles->raster;
    source_data = tiles->data + ((y + py) % tiles->rep_height) * source_raster;
    tile_bits_size = tiles->size.y * source_raster;
    end = tiles->data + tile_bits_size;
#undef END_Y_LOOP
#define END_Y_LOOP(sdelta, ddelta)\
  if ( end - bptr <= sdelta )	/* wrap around */\
    bptr -= tile_bits_size;\
  bptr += sdelta; dbptr += ddelta
    dest_raster = mdev->raster;
    dbase = scan_line_base(mdev, y);
    x = tx;
    rw = tw;
    /*
     * The outermost loop here works horizontally, one iteration per
     * copy of the tile.  Note that all iterations except the first
     * have source_x = 0.
     */
    {
	int source_x = (x + px) % tiles->rep_width;

	w = tiles->size.x - source_x;
	bptr = source_data + ((source_x & ~chunk_align_bit_mask) >> 3);
	dbit = x & chunk_align_bit_mask;
	skew = dbit - (source_x & chunk_align_bit_mask);
    }
  outer:if (w > rw)
	w = rw;
    h = th;
    dbptr = dbase + ((x >> 3) & -chunk_align_bytes);
    if ((wleft = w + dbit - chunk_bits) <= 0) {		/* The entire operation fits in one (destination) chunk. */
	set_mono_thin_mask(mask, w, dbit);
#define WRITE1_LOOP(src)\
  for ( ; ; )\
   { WRITE_STORE_MASKED(src, mask, 0);\
     if ( --h == 0 ) break;\
     END_Y_LOOP(source_raster, dest_raster);\
   }
	if (skew >= 0) {	/* single -> single, right/no shift */
	    if (skew == 0) {	/* no shift */
		WRITE1_LOOP(CFETCH_ALIGNED(bptr));
	    } else {		/* right shift */
#if CFETCH_USES_CSKEW
		int cskew = chunk_bits - skew;
#endif

		WRITE1_LOOP(CFETCH_RIGHT(bptr, skew, cskew));
	    }
	} else if (wleft <= skew) {	/* single -> single, left shift */
#if CFETCH_USES_CSKEW
	    int cskew = chunk_bits + skew;
#endif

	    skew = -skew;
	    WRITE1_LOOP(CFETCH_LEFT(bptr, skew, cskew));
	} else {		/* double -> single */
	    int cskew = -skew;

	    skew += chunk_bits;
	    WRITE1_LOOP(CFETCH2(bptr, cskew, skew));
	}
#undef WRITE1_LOOP
    } else if (wleft <= skew) {	/* 1 source chunk -> 2 destination chunks. */
	/* This is an important special case for */
	/* both characters and halftone tiles. */
	uint rmask;
	int cskew = chunk_bits - skew;

	set_mono_left_mask(mask, dbit);
	set_mono_right_mask(rmask, wleft);
#if arch_is_big_endian		/* no byte swapping */
#undef CINVERT
#define CINVERT(bits) (bits)	/* pre-inverted here */
	for (;;) {
	    register uint bits = CFETCH_ALIGNED(bptr) ^ invert;

	    WRITE_STORE_MASKED(bits >> skew, mask, 0);
	    WRITE_STORE_MASKED(bits << cskew, rmask, 1);
	    if (--h == 0)
		break;
	    END_Y_LOOP(source_raster, dest_raster);
	}
#undef CINVERT
#define CINVERT(bits) ((bits) ^ invert)
#else /* byte swapping */
	for (;;) {
	    WRITE_STORE_MASKED(CFETCH_RIGHT(bptr, skew, cskew), mask, 0);
	    WRITE_STORE_MASKED(CFETCH_LEFT(bptr, cskew, skew), rmask, 1);
	    if (--h == 0)
		break;
	    END_Y_LOOP(source_raster, dest_raster);
	}
#endif
    } else {			/* More than one source chunk and more than one */
	/* destination chunk are involved. */
	uint rmask;
	int words = (wleft & ~chunk_bit_mask) >> 3;
	uint sskip = source_raster - words;
	uint dskip = dest_raster - words;
	register uint bits;

#define NEXT_X_CHUNK()\
  bptr += chunk_bytes; dbptr += chunk_bytes

	set_mono_right_mask(rmask, wleft & chunk_bit_mask);
	if (skew == 0) {	/* optimize the aligned case */
	    if (dbit == 0)
		mask = 0;
	    else
		set_mono_left_mask(mask, dbit);
	    for (;;) {
		int count = wleft;

		/* Do first partial chunk. */
		if (mask)
		    WRITE_STORE_MASKED(CFETCH_ALIGNED(bptr), mask, 0);
		else
		    WRITE_STORE(CFETCH_ALIGNED(bptr));
		/* Do full chunks. */
		while ((count -= chunk_bits) >= 0) {
		    NEXT_X_CHUNK();
		    WRITE_STORE(CFETCH_ALIGNED(bptr));
		}
		/* Do last chunk */
		if (count > -chunk_bits) {
		    WRITE_STORE_MASKED(CFETCH_ALIGNED(bptr + chunk_bytes), rmask, 1);
		}
		if (--h == 0)
		    break;
		END_Y_LOOP(sskip, dskip);
	    }
	} else {		/* not aligned */
	    bool case_right =
	    (skew >= 0 ? true :
	     ((bptr += chunk_bytes), false));
	    int cskew = -skew & chunk_bit_mask;

	    skew &= chunk_bit_mask;
	    set_mono_left_mask(mask, dbit);
	    for (;;) {
		int count = wleft;

		if (case_right)
		    bits = CFETCH_RIGHT(bptr, skew, cskew);
		else
		    bits = CFETCH2(bptr - chunk_bytes, cskew, skew);
		WRITE_STORE_MASKED(bits, mask, 0);
		/* Do full chunks. */
		while (count >= chunk_bits) {
		    bits = CFETCH2(bptr, cskew, skew);
		    NEXT_X_CHUNK();
		    WRITE_STORE(bits);
		    count -= chunk_bits;
		}
		/* Do last chunk */
		if (count > 0) {
		    bits = CFETCH_LEFT(bptr, cskew, skew);
		    if (count > skew)
			bits += CFETCH_RIGHT(bptr + chunk_bytes, skew, cskew);
		    WRITE_STORE_MASKED(bits, rmask, 1);
		}
		if (--h == 0)
		    break;
		END_Y_LOOP(sskip, dskip);
	    }
	}
    }
#undef END_Y_LOOP
#undef NEXT_X_CHUNK
#undef optr
    if ((rw -= w) > 0) {
	x += w;
	w = tiles->size.x;
	bptr = source_data;
	skew = dbit = x & chunk_align_bit_mask;
	goto outer;
    }
    return 0;
#endif /* !USE_COPY_ROP */
}

/* ================ "Word"-oriented device ================ */

/* Note that on a big-endian machine, this is the same as the */
/* standard byte-oriented-device. */

#if !arch_is_big_endian

/* Procedures */
private dev_proc_copy_mono(mem1_word_copy_mono);
private dev_proc_fill_rectangle(mem1_word_fill_rectangle);

#define mem1_word_strip_tile_rectangle gx_default_strip_tile_rectangle

/* Here is the device descriptor. */
const gx_device_memory mem_mono_word_device =
mem_full_alpha_device("image1w", 0, 1, mem_open,
		      mem_mono_map_rgb_color, mem_mono_map_color_rgb,
       mem1_word_copy_mono, gx_default_copy_color, mem1_word_fill_rectangle,
		      gx_default_map_cmyk_color, gx_no_copy_alpha,
		      mem1_word_strip_tile_rectangle, gx_no_strip_copy_rop,
		      mem_word_get_bits_rectangle);

/* Fill a rectangle with a color. */
private int
mem1_word_fill_rectangle(gx_device * dev, int x, int y, int w, int h,
			 gx_color_index color)
{
    gx_device_memory * const mdev = (gx_device_memory *)dev;
    byte *base;
    uint raster;

    fit_fill(dev, x, y, w, h);
    base = scan_line_base(mdev, y);
    raster = mdev->raster;
    mem_swap_byte_rect(base, raster, x, w, h, true);
    bits_fill_rectangle(base, x, raster, -(int)(mono_fill_chunk) color, w, h);
    mem_swap_byte_rect(base, raster, x, w, h, true);
    return 0;
}

/* Copy a bitmap. */
private int
mem1_word_copy_mono(gx_device * dev,
 const byte * source_data, int source_x, int source_raster, gx_bitmap_id id,
   int x, int y, int w, int h, gx_color_index color0, gx_color_index color1)
{
    gx_device_memory * const mdev = (gx_device_memory *)dev;
    byte *row;
    uint raster;
    bool store;

    fit_copy(dev, source_data, source_x, source_raster, id, x, y, w, h);
    row = scan_line_base(mdev, y);
    raster = mdev->raster;
    store = (color0 != gx_no_color_index && color1 != gx_no_color_index);
    mem_swap_byte_rect(row, raster, x, w, h, store);
    mem_mono_copy_mono(dev, source_data, source_x, source_raster, id,
		       x, y, w, h, color0, color1);
    mem_swap_byte_rect(row, raster, x, w, h, false);
    return 0;
}

#endif /* !arch_is_big_endian */