fblin32alpha.c

the embedded GUI for SamSung s3c2410 cpu based board.is microwindows0.90

				register int d;
				a = (int) (unsigned char)
					src8[MWI_BYTE_OFFSET_ALPHA];
				if (a == 255) {
					dst8[0] = src8[0];
					dst8[1] = src8[1];
					dst8[2] = src8[2];
					dst8[MWI_BYTE_OFFSET_ALPHA] = a;
				} else if (a != 0) {
					s = (int) (unsigned char)
						src8[MWI_BYTE_OFFSET_R];
					d = (int) (unsigned char)
						dst8[MWI_BYTE_OFFSET_R];
					dst8[MWI_BYTE_OFFSET_R] =
						(unsigned
						 char) ((((s - d) * a) >> 8) +
							d);

					s = (int) (unsigned char)
						src8[MWI_BYTE_OFFSET_G];
					d = (int) (unsigned char)
						dst8[MWI_BYTE_OFFSET_G];
					dst8[MWI_BYTE_OFFSET_G] =
						(unsigned
						 char) ((((s - d) * a) >> 8) +
							d);

					s = (int) (unsigned char)
						src8[MWI_BYTE_OFFSET_B];
					d = (int) (unsigned char)
						dst8[MWI_BYTE_OFFSET_B];
					dst8[MWI_BYTE_OFFSET_B] =
						(unsigned
						 char) ((((s - d) * a) >> 8) +
							d);

					d = (int) (unsigned char)
						dst8[MWI_BYTE_OFFSET_ALPHA];
					dst8[MWI_BYTE_OFFSET_ALPHA] =
						(a + ((d * (256 - a)) >> 8));
				}
				src8 += 4;
				dst8 += 4;
			}
			dst8 += (dlinelen - w) * 4;
			src8 += (slinelen - w) * 4;
		}
	} else {
		/* FIXME: Implement other alpha blend modes */
	}
	DRAWOFF;
}

/* srccopy stretchblt*/
static void
linear32a_stretchblit(PSD dstpsd, MWCOORD dstx, MWCOORD dsty, MWCOORD dstw,
		      MWCOORD dsth, PSD srcpsd, MWCOORD srcx, MWCOORD srcy,
		      MWCOORD srcw, MWCOORD srch, long op)
{
	ADDR32 dst;
	ADDR32 src;
	int dlinelen = dstpsd->linelen;
	int slinelen = srcpsd->linelen;
	int i, ymax;
	int row_pos, row_inc;
	int col_pos, col_inc;
	unsigned long pixel = 0;

	assert(dstpsd->addr != 0);
	assert(dstx >= 0 && dstx < dstpsd->xres);
	assert(dsty >= 0 && dsty < dstpsd->yres);
	assert(dstw > 0);
	assert(dsth > 0);
	assert(srcpsd->addr != 0);
	assert(srcx >= 0 && srcx < srcpsd->xres);
	assert(srcy >= 0 && srcy < srcpsd->yres);
	assert(srcw > 0);
	assert(srch > 0);
	assert(dstx + dstw <= dstpsd->xres);
	assert(dsty + dsth <= dstpsd->yres);
	assert(srcx + srcw <= srcpsd->xres);
	assert(srcy + srch <= srcpsd->yres);

	DRAWON;
	row_pos = 0x10000;
	row_inc = (srch << 16) / dsth;

	/* stretch blit using integer ratio between src/dst height/width */
	for (ymax = dsty + dsth; dsty < ymax; ++dsty) {

		/* find source y position */
		while (row_pos >= 0x10000L) {
			++srcy;
			row_pos -= 0x10000L;
		}

		dst = (ADDR32) dstpsd->addr + dstx + dsty * dlinelen;
		src = (ADDR32) srcpsd->addr + srcx + (srcy - 1) * slinelen;

		/* copy a row of pixels */
		col_pos = 0x10000;
		col_inc = (srcw << 16) / dstw;
		for (i = 0; i < dstw; ++i) {
			/* get source x pixel */
			while (col_pos >= 0x10000L) {
				pixel = *src++;
				col_pos -= 0x10000L;
			}
			*dst++ = pixel;
			col_pos += col_inc;
		}

		row_pos += row_inc;
	}
	DRAWOFF;
}

/*
 * This stretchblit code was originally written for the TriMedia
 * VLIW CPU.  Therefore it uses RESTRICT pointers, and the special
 * one-assembler-opcode pseudo-functions SIGN and ABS.
 *
 * (The 'restrict' extension is in C99, so for a C99 compiler you
 * could "#define RESTRICT restrict" or put
 * "CFLAGS += -DRESTRICT=restrict" in the makefile).
 *
 * Compatibility definitions:
 */
#ifndef RESTRICT
#define RESTRICT
#endif
#ifndef SIGN
#define SIGN(x) (((x) > 0) ? 1 : (((x) == 0) ? 0 : -1))
#endif
#ifndef ABS
#define ABS(x) (((x) >= 0) ? (x) : -(x))
#endif

/* Blit a 32-bit image.
 * Can stretch the image by any X and/or Y scale factor.
 * Can flip the image in the X and/or Y axis.
 *
 * This is the faster version with no per-pixel multiply and a single
 * decision tree for the inner loop, by Jon.  Based on Alex's original
 * all-integer version.
 *
 * Paramaters:
 * srf              - Dest surface
 * dest_x_start
 * dest_y_start    - Top left corner of dest rectangle
 * width, height   - Size in dest co-ordinates.
 * x_denominator   - Denominator for source X value fractions.  Note that
 *                   this must be even, and all the numerators must also be
 *                   even, so we can easily divide by 2.
 * y_denominator   - Denominator for source Y value fractions.  Note that
 *                   this must be even, and all the numerators must also be
 *                   even, so we can easily divide by 2.
 * src_x_fraction  -
 * src_y_fraction  - Point in source that corresponds to the top left corner
 *                   of the pixel (dest_x_start, dest_y_start).  This is a
 *                   fraction - to get a float, divide by y_denominator.
 * x_step_fraction - X step in src for an "x++" step in dest.  May be negative
 *                   (for a flip).  Expressed as a fraction - divide it by
 *                   x_denominator for a float.
 * y_step_fraction - Y step in src for a  "y++" step in dest.  May be negative
 *                   (for a flip).  Expressed as a fraction - divide it by
 *                   y_denominator for a float.
 * image           - Source image.
 * op              - Raster operation, currently ignored.
 */
static void
linear32a_stretchblitex(PSD dstpsd,
			PSD srcpsd,
			MWCOORD dest_x_start,
			MWCOORD dest_y_start,
			MWCOORD width,
			MWCOORD height,
			int x_denominator,
			int y_denominator,
			int src_x_fraction,
			int src_y_fraction,
			int x_step_fraction, int y_step_fraction, long op)
{
	/* Pointer to the current pixel in the source image */
	unsigned long *RESTRICT src_ptr;

	/* Pointer to x=xs1 on the next line in the source image */
	unsigned long *RESTRICT next_src_ptr;

	/* Pointer to the current pixel in the dest image */
	unsigned long *RESTRICT dest_ptr;

	/* Pointer to x=xd1 on the next line in the dest image */
	unsigned long *next_dest_ptr;

	/* Keep track of error in the source co-ordinates */
	int x_error;
	int y_error;

	/* 1-unit steps "forward" through the source image, as steps in the image
	 * byte array.
	 */
	int src_x_step_one;
	int src_y_step_one;

	/* normal steps "forward" through the source image, as steps in the image
	 * byte array.
	 */
	int src_x_step_normal;
	int src_y_step_normal;

	/* 1-unit steps "forward" through the source image, as steps in the image
	 * byte array.
	 */
	int x_error_step_normal;
	int y_error_step_normal;

	/* Countdown to the end of the destination image */
	int x_count;
	int y_count;

	/* Start position in source, in whole pixels */
	int src_x_start;
	int src_y_start;

	/* Error values for start X position in source */
	int x_error_start;

	/* 1-unit step down dest, in bytes. */
	int dest_y_step;

	/*DPRINTF("Nano-X: linear32_stretchflipblit( dest=(%d,%d) %dx%d )\n",
	   dest_x_start, dest_y_start, width, height); */

	/* We add half a dest pixel here so we're sampling from the middle of
	 * the dest pixel, not the top left corner.
	 */
	src_x_fraction += (x_step_fraction >> 1);
	src_y_fraction += (y_step_fraction >> 1);

	/* Seperate the whole part from the fractions.
	 *
	 * Also, We need to do lots of comparisons to see if error values are
	 * >= x_denominator.  So subtract an extra x_denominator for speed - then
	 * we can just check if it's >= 0.
	 */
	src_x_start = src_x_fraction / x_denominator;
	src_y_start = src_y_fraction / y_denominator;
	x_error_start = src_x_fraction - (src_x_start + 1) * x_denominator;
	y_error = src_y_fraction - (src_y_start + 1) * y_denominator;

	/* precalculate various deltas */

	src_x_step_normal = x_step_fraction / x_denominator;
	src_x_step_one = SIGN(x_step_fraction);
	x_error_step_normal =
		ABS(x_step_fraction) - ABS(src_x_step_normal) * x_denominator;

	src_y_step_normal = y_step_fraction / y_denominator;
	src_y_step_one = SIGN(y_step_fraction) * srcpsd->linelen;
	y_error_step_normal =
		ABS(y_step_fraction) - ABS(src_y_step_normal) * y_denominator;
	src_y_step_normal *= srcpsd->linelen;

	/* DPRINTF("ov_stretch_image8: X: One step=%d, err-=%d; normal step=%d, err+=%d\n                   Y: One step=%d, err-=%d; normal step=%d, err+=%d\n",
	   src_x_step_one, x_denominator, src_x_step_normal, x_error_step_normal,
	   src_y_step_one, y_denominator, src_y_step_normal, y_error_step_normal);
	 */

	/* Pointer to the first source pixel */
	next_src_ptr =
		((unsigned long *) srcpsd->addr) +
		src_y_start * srcpsd->linelen + src_x_start;

	/* Cache the width of a scanline in dest */
	dest_y_step = dstpsd->linelen;

	/* Pointer to the first dest pixel */
	next_dest_ptr =
		((unsigned long *) dstpsd->addr) +
		(dest_y_start * dest_y_step) + dest_x_start;

	/*
	 * Note: The MWROP_SRC and MWROP_XOR_FGBG cases below are simple
	 * expansions of the default case.  They can be removed without
	 * significant speed penalty if you need to reduce code size.
	 *
	 * The SRC_OVER case cannot be removed (since applyOp doesn't
	 * handle it correctly).
	 *
	 * The MWROP_CLEAR case could be removed.  But it is a large
	 * speed increase for a small quantity of code.
	 *
	 * FIXME Porter-Duff rules other than SRC_OVER not handled!!
	 */
	switch (op & MWROP_EXTENSION) {
	case MWROP_SRC:
		/* Benchmarking shows that this while loop is faster than the equivalent
		 * for loop: for(y_count=0; y_count<height; y_count++) { ... }
		 */
		y_count = height;
		while (y_count-- > 0) {
			src_ptr = next_src_ptr;
			dest_ptr = next_dest_ptr;

			x_error = x_error_start;

			x_count = width;
			while (x_count-- > 0) {
				*dest_ptr++ = *src_ptr;

				src_ptr += src_x_step_normal;
				x_error += x_error_step_normal;

				if (x_error >= 0) {
					src_ptr += src_x_step_one;
					x_error -= x_denominator;
				}
			}

			next_dest_ptr += dest_y_step;

			next_src_ptr += src_y_step_normal;
			y_error += y_error_step_normal;

			if (y_error >= 0) {
				next_src_ptr += src_y_step_one;
				y_error -= y_denominator;
			}
		}
		break;

	case MWROP_SRC_OVER:
		/* Benchmarking shows that this while loop is faster than the equivalent
		 * for loop: for(y_count=0; y_count<height; y_count++) { ... }
		 */
		y_count = height;
		while (y_count-- > 0) {
			src_ptr = next_src_ptr;
			dest_ptr = next_dest_ptr;

			x_error = x_error_start;

			x_count = width;
			while (x_count-- > 0) {
				int c;
				int a;
				int s;
				register int d;
				int orig;
				int result;
				c = *src_ptr;
				a = (c >> 24) & 0xFF;
				if (a == 255) {
					*dest_ptr = c;
				} else if (a != 0) {
					orig = *dest_ptr;
					d = (orig >> 24) & 0xFF;
					result = ((d + a -
						   ((d * a) >> 8)) << 24);
					s = ((c >> 16) & 0xFF);
					d = (orig >> 16) & 0xFF;
					result |=
						(((((s - d) * a) >> 8) +
						  d) << 16);
					s = ((c >> 8) & 0xFF);
					d = (orig >> 8) & 0xFF;
					result |=
						(((((s - d) * a) >> 8) +
						  d) << 8);
					s = (c & 0xFF);
					d = orig & 0xFF;
					result |= ((((s - d) * a) >> 8) + d);
					*dest_ptr = result;
				}

				dest_ptr++;

				src_ptr += src_x_step_normal;
				x_error += x_error_step_normal;

				if (x_error >= 0) {
					src_ptr += src_x_step_one;
					x_error -= x_denominator;
				}
			}

			next_dest_ptr += dest_y_step;

			next_src_ptr += src_y_step_normal;
			y_error += y_error_step_normal;

			if (y_error >= 0) {
				next_src_ptr += src_y_step_one;
				y_error -= y_denominator;
			}
		}
		break;

	case MWROP_XOR_FGBG:
		y_count = height;
		while (y_count-- > 0) {
			src_ptr = next_src_ptr;
			dest_ptr = next_dest_ptr;

			x_error = x_error_start;

			x_count = width;
			while (x_count-- > 0) {
				*dest_ptr++ ^= *src_ptr ^ gr_background;

				src_ptr += src_x_step_normal;
				x_error += x_error_step_normal;

				if (x_error >= 0) {
					src_ptr += src_x_step_one;
					x_error -= x_denominator;
				}
			}

			next_dest_ptr += dest_y_step;

			next_src_ptr += src_y_step_normal;
			y_error += y_error_step_normal;

			if (y_error >= 0) {
				next_src_ptr += src_y_step_one;
				y_error -= y_denominator;
			}
		}
		break;

	case MWROP_CLEAR:
		y_count = height;
		while (y_count-- > 0) {
			dest_ptr = next_dest_ptr;
			x_count = width;
			while (x_count-- > 0) {
				*dest_ptr++ = 0;
			}
			next_dest_ptr += dest_y_step;
		}
		break;

	default:
		y_count = height;
		while (y_count-- > 0) {
			src_ptr = next_src_ptr;
			dest_ptr = next_dest_ptr;

			x_error = x_error_start;

			x_count = width;
			while (x_count-- > 0) {
				applyOp(MWROP_TO_MODE(op), *src_ptr, dest_ptr,
					ADDR32);
				dest_ptr++;

				src_ptr += src_x_step_normal;
				x_error += x_error_step_normal;

				if (x_error >= 0) {
					src_ptr += src_x_step_one;
					x_error -= x_denominator;
				}
			}

			next_dest_ptr += dest_y_step;

			next_src_ptr += src_y_step_normal;
			y_error += y_error_step_normal;

			if (y_error >= 0) {
				next_src_ptr += src_y_step_one;
				y_error -= y_denominator;
			}
		}
		break;

	}
}

#if MW_FEATURE_PSDOP_BITMAP_BYTES_LSB_FIRST
/* psd->DrawArea operation PSDOP_BITMAP_BYTES_LSB_FIRST which
 * takes a pixmap, each line is byte aligned, and copies it
 * to the screen using fg_color and bg_color to replace a 1
 * and 0 in the pixmap.  This pixmap is ordered the wrong
 * way around; it has the leftmost pixel (on the screen) in
 * LSB (Bit 0) of the bytes.
 *
 * The reason why this non-intuitive bit ordering is used is
 * to match the bit ordering used in the T1lib font rendering
 * library.
 *
 * Variables used in the gc:
 *       dstx, dsty, dsth, dstw   Destination rectangle
 *       srcx, srcy               Source rectangle
 *       src_linelen              Linesize in bytes of source
 *       pixels                   Pixmap data
 *       fg_color                 Color of a '1' bit
 *       bg_color                 Color of a '0' bit
 *       gr_usebg                 If set, bg_color is used.  If zero,
 *                                then '0' bits are transparent.
 */
static void
linear32a_drawarea_bitmap_bytes_lsb_first(PSD psd, driver_gc_t * gc)
{
/*
 * The difference between the MSB_FIRST and LSB_FIRST variants of
 * this function is simply the definition of these three #defines.
 *
 * MWI_IS_BIT_BEFORE_OR_EQUAL(A,B) returns true if bit A is before
 *     (i.e. to the left of) bit B.
 * MWI_ADVANCE_BIT(X) advances X on to the next bit to the right,
 *     and stores the result back in X.
 * MWI_BIT_NO(N), where 0<=n<=7, gives the Nth bit, where 0 is the
 *     leftmost bit and 7 is the rightmost bit.  This is a constant
 *     iff N is a constant.
 */
#define MWI_IS_BIT_BEFORE_OR_EQUAL(a,b) ((a) <= (b))
#define MWI_ADVANCE_BIT(target) ((target) <<= 1)
#define MWI_BIT_NO(n) (0x01 << (n))

/*
 * Two convenience defines - these are the same for MSB_FIRST and
 * LSB_FIRST.
 */
#define MWI_FIRST_BIT MWI_BIT_NO(0)
#define MWI_LAST_BIT  MWI_BIT_NO(7)

	unsigned char prefix_first_bit;
	unsigned char postfix_first_bit = MWI_FIRST_BIT;