sdl_rleaccel.c

视频音频编码程序 视频音频编码程序 视频音频编码程序

/*
    SDL - Simple DirectMedia Layer
    Copyright (C) 1997, 1998, 1999, 2000, 2001, 2002  Sam Lantinga

    This library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Library General Public
    License as published by the Free Software Foundation; either
    version 2 of the License, or (at your option) any later version.

    This library is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    Library General Public License for more details.

    You should have received a copy of the GNU Library General Public
    License along with this library; if not, write to the Free
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

    Sam Lantinga
    [e-mail]
*/

#ifdef SAVE_RCSID
static char rcsid =
 "@(#) $Id: SDL_RLEaccel.c,v 1.4 2002/04/22 21:38:03 wmay Exp $";
#endif

/*
 * RLE encoding for software colorkey and alpha-channel acceleration
 *
 * Original version by Sam Lantinga
 *
 * Mattias Engdeg錼d (Yorick): Rewrite. New encoding format, encoder and
 * decoder. Added per-surface alpha blitter. Added per-pixel alpha
 * format, encoder and blitter.
 *
 * Many thanks to Xark and johns for hints, benchmarks and useful comments
 * leading to this code.
 *
 * Welcome to Macro Mayhem.
 */

/*
 * The encoding translates the image data to a stream of segments of the form
 *
 * <skip> <run> <data>
 *
 * where <skip> is the number of transparent pixels to skip,
 *       <run>  is the number of opaque pixels to blit,
 * and   <data> are the pixels themselves.
 *
 * This basic structure is used both for colorkeyed surfaces, used for simple
 * binary transparency and for per-surface alpha blending, and for surfaces
 * with per-pixel alpha. The details differ, however:
 *
 * Encoding of colorkeyed surfaces:
 *
 *   Encoded pixels always have the same format as the target surface.
 *   <skip> and <run> are unsigned 8 bit integers, except for 32 bit depth
 *   where they are 16 bit. This makes the pixel data aligned at all times.
 *   Segments never wrap around from one scan line to the next.
 *
 *   The end of the sequence is marked by a zero <skip>,<run> pair at the *
 *   beginning of a line.
 *
 * Encoding of surfaces with per-pixel alpha:
 *
 *   The sequence begins with a struct RLEDestFormat describing the target
 *   pixel format, to provide reliable un-encoding.
 *
 *   Each scan line is encoded twice: First all completely opaque pixels,
 *   encoded in the target format as described above, and then all
 *   partially transparent (translucent) pixels (where 1 <= alpha <= 254),
 *   in the following 32-bit format:
 *
 *   For 32-bit targets, each pixel has the target RGB format but with
 *   the alpha value occupying the highest 8 bits. The <skip> and <run>
 *   counts are 16 bit.
 * 
 *   For 16-bit targets, each pixel has the target RGB format, but with
 *   the middle component (usually green) shifted 16 steps to the left,
 *   and the hole filled with the 5 most significant bits of the alpha value.
 *   i.e. if the target has the format         rrrrrggggggbbbbb,
 *   the encoded pixel will be 00000gggggg00000rrrrr0aaaaabbbbb.
 *   The <skip> and <run> counts are 8 bit for the opaque lines, 16 bit
 *   for the translucent lines. Two padding bytes may be inserted
 *   before each translucent line to keep them 32-bit aligned.
 *
 *   The end of the sequence is marked by a zero <skip>,<run> pair at the
 *   beginning of an opaque line.
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "SDL_types.h"
#include "SDL_video.h"
#include "SDL_error.h"
#include "SDL_sysvideo.h"
#include "SDL_blit.h"
#include "SDL_memops.h"
#include "SDL_RLEaccel_c.h"

#ifndef MAX
#define MAX(a, b) ((a) > (b) ? (a) : (b))
#endif
#ifndef MIN
#define MIN(a, b) ((a) < (b) ? (a) : (b))
#endif

#define PIXEL_COPY(to, from, len, bpp)			\
do {							\
    if(bpp == 4) {					\
	SDL_memcpy4(to, from, (unsigned)(len));		\
    } else {						\
	SDL_memcpy(to, from, (unsigned)(len) * (bpp));	\
    }							\
} while(0)

/*
 * Various colorkey blit methods, for opaque and per-surface alpha
 */

#define OPAQUE_BLIT(to, from, length, bpp, alpha)	\
    PIXEL_COPY(to, from, length, bpp)

/*
 * For 32bpp pixels on the form 0x00rrggbb:
 * If we treat the middle component separately, we can process the two
 * remaining in parallel. This is safe to do because of the gap to the left
 * of each component, so the bits from the multiplication don't collide.
 * This can be used for any RGB permutation of course.
 */
#define ALPHA_BLIT32_888(to, from, length, bpp, alpha)		\
    do {							\
        int i;							\
	Uint32 *src = (Uint32 *)(from);				\
	Uint32 *dst = (Uint32 *)(to);				\
	for(i = 0; i < (int)(length); i++) {			\
	    Uint32 s = *src++;					\
	    Uint32 d = *dst;					\
	    Uint32 s1 = s & 0xff00ff;				\
	    Uint32 d1 = d & 0xff00ff;				\
	    d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;	\
	    s &= 0xff00;					\
	    d &= 0xff00;					\
	    d = (d + ((s - d) * alpha >> 8)) & 0xff00;		\
	    *dst++ = d1 | d;					\
	}							\
    } while(0)

/*
 * For 16bpp pixels we can go a step further: put the middle component
 * in the high 16 bits of a 32 bit word, and process all three RGB
 * components at the same time. Since the smallest gap is here just
 * 5 bits, we have to scale alpha down to 5 bits as well.
 */
#define ALPHA_BLIT16_565(to, from, length, bpp, alpha)	\
    do {						\
        int i;						\
	Uint16 *src = (Uint16 *)(from);			\
	Uint16 *dst = (Uint16 *)(to);			\
	for(i = 0; i < (int)(length); i++) {		\
	    Uint32 s = *src++;				\
	    Uint32 d = *dst;				\
	    s = (s | s << 16) & 0x07e0f81f;		\
	    d = (d | d << 16) & 0x07e0f81f;		\
	    d += (s - d) * alpha >> 5;			\
	    d &= 0x07e0f81f;				\
	    *dst++ = d | d >> 16;			\
	}						\
    } while(0)

#define ALPHA_BLIT16_555(to, from, length, bpp, alpha)	\
    do {						\
        int i;						\
	Uint16 *src = (Uint16 *)(from);			\
	Uint16 *dst = (Uint16 *)(to);			\
	for(i = 0; i < (int)(length); i++) {		\
	    Uint32 s = *src++;				\
	    Uint32 d = *dst;				\
	    s = (s | s << 16) & 0x03e07c1f;		\
	    d = (d | d << 16) & 0x03e07c1f;		\
	    d += (s - d) * alpha >> 5;			\
	    d &= 0x03e07c1f;				\
	    *dst++ = d | d >> 16;			\
	}						\
    } while(0)

/*
 * The general slow catch-all function, for remaining depths and formats
 */
#define ALPHA_BLIT_ANY(to, from, length, bpp, alpha)			\
    do {								\
        int i;								\
	Uint8 *src = from;						\
	Uint8 *dst = to;						\
	for(i = 0; i < (int)(length); i++) {				\
	    Uint32 s, d;						\
	    unsigned rs, gs, bs, rd, gd, bd;				\
	    switch(bpp) {						\
	    case 2:							\
		s = *(Uint16 *)src;					\
		d = *(Uint16 *)dst;					\
		break;							\
	    case 3:							\
		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
		    s = (src[0] << 16) | (src[1] << 8) | src[2];	\
		    d = (dst[0] << 16) | (dst[1] << 8) | dst[2];	\
		} else {						\
		    s = (src[2] << 16) | (src[1] << 8) | src[0];	\
		    d = (dst[2] << 16) | (dst[1] << 8) | dst[0];	\
		}							\
		break;							\
	    case 4:							\
		s = *(Uint32 *)src;					\
		d = *(Uint32 *)dst;					\
		break;							\
	    }								\
	    RGB_FROM_PIXEL(s, fmt, rs, gs, bs);				\
	    RGB_FROM_PIXEL(d, fmt, rd, gd, bd);				\
	    rd += (rs - rd) * alpha >> 8;				\
	    gd += (gs - gd) * alpha >> 8;				\
	    bd += (bs - bd) * alpha >> 8;				\
	    PIXEL_FROM_RGB(d, fmt, rd, gd, bd);				\
	    switch(bpp) {						\
	    case 2:							\
		*(Uint16 *)dst = d;					\
		break;							\
	    case 3:							\
		if(SDL_BYTEORDER == SDL_BIG_ENDIAN) {			\
		    dst[0] = d >> 16;					\
		    dst[1] = d >> 8;					\
		    dst[2] = d;						\
		} else {						\
		    dst[0] = d;						\
		    dst[1] = d >> 8;					\
		    dst[2] = d >> 16;					\
		}							\
		break;							\
	    case 4:							\
		*(Uint32 *)dst = d;					\
		break;							\
	    }								\
	    src += bpp;							\
	    dst += bpp;							\
	}								\
    } while(0)


/*
 * Special case: 50% alpha (alpha=128)
 * This is treated specially because it can be optimized very well, and
 * since it is good for many cases of semi-translucency.
 * The theory is to do all three components at the same time:
 * First zero the lowest bit of each component, which gives us room to
 * add them. Then shift right and add the sum of the lowest bits.
 */
#define ALPHA_BLIT32_888_50(to, from, length, bpp, alpha)		\
    do {								\
        int i;								\
	Uint32 *src = (Uint32 *)(from);					\
	Uint32 *dst = (Uint32 *)(to);					\
	for(i = 0; i < (int)(length); i++) {				\
	    Uint32 s = *src++;						\
	    Uint32 d = *dst;						\
	    *dst++ = (((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)	\
		     + (s & d & 0x00010101);				\
	}								\
    } while(0)

/*
 * For 16bpp, we can actually blend two pixels in parallel, if we take
 * care to shift before we add, not after.
 */

/* helper: blend a single 16 bit pixel at 50% */
#define BLEND16_50(dst, src, mask)			\
    do {						\
        Uint32 s = *src++;				\
	Uint32 d = *dst;				\
	*dst++ = (((s & mask) + (d & mask)) >> 1)	\
	         + (s & d & (~mask & 0xffff));		\
    } while(0)

/* basic 16bpp blender. mask is the pixels to keep when adding. */
#define ALPHA_BLIT16_50(to, from, length, bpp, alpha, mask)		\
    do {								\
	unsigned n = (length);						\
	Uint16 *src = (Uint16 *)(from);					\
	Uint16 *dst = (Uint16 *)(to);					\
	if(((unsigned long)src ^ (unsigned long)dst) & 3) {		\
	    /* source and destination not in phase, blit one by one */	\
	    while(n--)							\
		BLEND16_50(dst, src, mask);				\
	} else {							\
	    if((unsigned long)src & 3) {				\
		/* first odd pixel */					\
		BLEND16_50(dst, src, mask);				\
		n--;							\
	    }								\
	    for(; n > 1; n -= 2) {					\
		Uint32 s = *(Uint32 *)src;				\
		Uint32 d = *(Uint32 *)dst;				\
		*(Uint32 *)dst = ((s & (mask | mask << 16)) >> 1)	\
		               + ((d & (mask | mask << 16)) >> 1)	\
		               + (s & d & (~(mask | mask << 16)));	\
		src += 2;						\
		dst += 2;						\
	    }								\
	    if(n)							\
		BLEND16_50(dst, src, mask); /* last odd pixel */	\
	}								\
    } while(0)

#define ALPHA_BLIT16_565_50(to, from, length, bpp, alpha)	\
    ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xf7de)

#define ALPHA_BLIT16_555_50(to, from, length, bpp, alpha)	\
    ALPHA_BLIT16_50(to, from, length, bpp, alpha, 0xfbde)


#define CHOOSE_BLIT(blitter, alpha, fmt)				\
    do {								\
        if(alpha == 255) {						\
	    switch(fmt->BytesPerPixel) {				\
	    case 1: blitter(1, Uint8, OPAQUE_BLIT); break;		\
	    case 2: blitter(2, Uint8, OPAQUE_BLIT); break;		\
	    case 3: blitter(3, Uint8, OPAQUE_BLIT); break;		\
	    case 4: blitter(4, Uint16, OPAQUE_BLIT); break;		\
	    }								\
	} else {							\
	    switch(fmt->BytesPerPixel) {				\
	    case 1:							\
		/* No 8bpp alpha blitting */				\
		break;							\
									\
	    case 2:							\
		switch(fmt->Rmask | fmt->Gmask | fmt->Bmask) {		\
		case 0xffff:						\
		    if(fmt->Gmask == 0x07e0				\
		       || fmt->Rmask == 0x07e0				\
		       || fmt->Bmask == 0x07e0) {			\
			if(alpha == 128)				\
			    blitter(2, Uint8, ALPHA_BLIT16_565_50);	\
			else {						\
			    alpha >>= 3; /* use 5 bit alpha */		\
			    blitter(2, Uint8, ALPHA_BLIT16_565);	\
			}						\
		    } else						\
			goto general16;					\
		    break;						\
									\
		case 0x7fff:						\
		    if(fmt->Gmask == 0x03e0				\
		       || fmt->Rmask == 0x03e0				\
		       || fmt->Bmask == 0x03e0) {			\
			if(alpha == 128)				\
			    blitter(2, Uint8, ALPHA_BLIT16_555_50);	\
			else {						\
			    alpha >>= 3; /* use 5 bit alpha */		\
			    blitter(2, Uint8, ALPHA_BLIT16_555);	\
			}						\
			break;						\
		    }							\
		    /* fallthrough */					\
									\
		default:						\
		general16:						\
		    blitter(2, Uint8, ALPHA_BLIT_ANY);			\
		}							\
		break;							\
									\
	    case 3:							\
		blitter(3, Uint8, ALPHA_BLIT_ANY);			\
		break;							\
									\
	    case 4:							\
		if((fmt->Rmask | fmt->Gmask | fmt->Bmask) == 0x00ffffff	\
		   && (fmt->Gmask == 0xff00 || fmt->Rmask == 0xff00	\
		       || fmt->Bmask == 0xff00)) {			\
		    if(alpha == 128)					\
			blitter(4, Uint16, ALPHA_BLIT32_888_50);	\
		    else						\
			blitter(4, Uint16, ALPHA_BLIT32_888);		\
		} else							\
		    blitter(4, Uint16, ALPHA_BLIT_ANY);			\
		break;							\
	    }								\
	}								\
    } while(0)


/*
 * This takes care of the case when the surface is clipped on the left and/or
 * right. Top clipping has already been taken care of.
 */
static void RLEClipBlit(int w, Uint8 *srcbuf, SDL_Surface *dst,
			Uint8 *dstbuf, SDL_Rect *srcrect, unsigned alpha)
{
    SDL_PixelFormat *fmt = dst->format;

#define RLECLIPBLIT(bpp, Type, do_blit)					   \
    do {								   \
	int linecount = srcrect->h;					   \
	int ofs = 0;							   \
	int left = srcrect->x;						   \
	int right = left + srcrect->w;					   \
	dstbuf -= left * bpp;						   \
	for(;;) {							   \
	    int run;							   \
	    ofs += *(Type *)srcbuf;					   \
	    run = ((Type *)srcbuf)[1];					   \
	    srcbuf += 2 * sizeof(Type);					   \
	    if(run) {							   \
		/* clip to left and right borders */			   \
		if(ofs < right) {					   \
		    int start = 0;					   \
		    int len = run;					   \
		    int startcol;					   \
		    if(left - ofs > 0) {				   \
			start = left - ofs;				   \
			len -= start;					   \
			if(len <= 0)					   \
			    goto nocopy ## bpp ## do_blit;		   \
		    }							   \
		    startcol = ofs + start;				   \
		    if(len > right - startcol)				   \
			len = right - startcol;				   \
		    do_blit(dstbuf + startcol * bpp, srcbuf + start * bpp, \
			    len, bpp, alpha);				   \
		}							   \
	    nocopy ## bpp ## do_blit:					   \
		srcbuf += run * bpp;					   \
		ofs += run;						   \
	    } else if(!ofs)						   \
		break;							   \
	    if(ofs == w) {						   \
		ofs = 0;						   \
		dstbuf += dst->pitch;					   \
		if(!--linecount)					   \
		    break;						   \
	    }								   \
	}								   \
    } while(0)

    CHOOSE_BLIT(RLECLIPBLIT, alpha, fmt);

#undef RLECLIPBLIT

}


/* blit a colorkeyed RLE surface */
int SDL_RLEBlit(SDL_Surface *src, SDL_Rect *srcrect,
		SDL_Surface *dst, SDL_Rect *dstrect)
{
	Uint8 *dstbuf;
	Uint8 *srcbuf;
	int x, y;
	int w = src->w;
	unsigned alpha;

	/* Lock the destination if necessary */
	if ( dst->flags & (SDL_HWSURFACE|SDL_ASYNCBLIT) ) {
		SDL_VideoDevice *video = current_video;
		SDL_VideoDevice *this  = current_video;
		if ( video->LockHWSurface(this, dst) < 0 ) {
			return(-1);
		}
	}

	/* Set up the source and destination pointers */
	x = dstrect->x;
	y = dstrect->y;
	dstbuf = (Uint8 *)dst->pixels + dst->offset
	         + y * dst->pitch + x * src->format->BytesPerPixel;
	srcbuf = (Uint8 *)src->map->sw_data->aux_data;

	{
	    /* skip lines at the top if neccessary */
	    int vskip = srcrect->y;
	    int ofs = 0;
	    if(vskip) {

#define RLESKIP(bpp, Type)			\
		for(;;) {			\
		    int run;			\
		    ofs += *(Type *)srcbuf;	\
		    run = ((Type *)srcbuf)[1];	\
		    srcbuf += sizeof(Type) * 2;	\
		    if(run) {			\
			srcbuf += run * bpp;	\
			ofs += run;		\
		    } else if(!ofs)		\
			goto done;		\
		    if(ofs == w) {		\
			ofs = 0;		\
			if(!--vskip)		\
			    break;		\
		    }				\
		}

		switch(src->format->BytesPerPixel) {
		case 1: RLESKIP(1, Uint8); break;
		case 2: RLESKIP(2, Uint8); break;
		case 3: RLESKIP(3, Uint8); break;
		case 4: RLESKIP(4, Uint16); break;
		}

#undef RLESKIP

	    }
	}

	alpha = (src->flags & SDL_SRCALPHA) == SDL_SRCALPHA
	        ? src->format->alpha : 255;
	/* if left or right edge clipping needed, call clip blit */