/*    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    slouken@libsdl.org*/#ifdef SAVE_RCSIDstatic char rcsid = "@(#) $Id: SDL_blit_N.c,v 1.7 2002/08/01 23:06:39 slouken Exp $";#endif#include <stdio.h>#include "SDL_types.h"#include "SDL_video.h"#include "SDL_blit.h"#include "SDL_byteorder.h"/* Function to check the CPU flags */#define MMX_CPU		0x800000#ifdef USE_ASMBLIT#define CPU_Flags()	Hermes_X86_CPU()#else#define CPU_Flags()	0L#endif/* Functions to blit from N-bit surfaces to other surfaces */#ifdef USE_ASMBLIT/* Heheheh, we coerce Hermes into using SDL blit information */#define X86_ASSEMBLER#define HermesConverterInterface	SDL_BlitInfo#define HermesClearInterface		void#define STACKCALL#include "HeadMMX.h"#include "HeadX86.h"#else/* This is now endian dependent */#if ( SDL_BYTEORDER == SDL_LIL_ENDIAN )#define HI	1#define LO	0#else /* ( SDL_BYTEORDER == SDL_BIG_ENDIAN ) */#define HI	0#define LO	1#endif/* Special optimized blit for RGB 8-8-8 --> RGB 3-3-2 */#define RGB888_RGB332(dst, src) { \	dst = (((src)&0x00E00000)>>16)| \	      (((src)&0x0000E000)>>11)| \	      (((src)&0x000000C0)>>6); \}static void Blit_RGB888_index8(SDL_BlitInfo *info){#ifndef USE_DUFFS_LOOP	int c;#endif	int width, height;	Uint32 *src;	const Uint8 *map;	Uint8 *dst;	int srcskip, dstskip;	/* Set up some basic variables */	width = info->d_width;	height = info->d_height;	src = (Uint32 *)info->s_pixels;	srcskip = info->s_skip/4;	dst = info->d_pixels;	dstskip = info->d_skip;	map = info->table;	if ( map == NULL ) {		while ( height-- ) {#ifdef USE_DUFFS_LOOP			DUFFS_LOOP(				RGB888_RGB332(*dst++, *src);			, width);#else			for ( c=width/4; c; --c ) {				/* Pack RGB into 8bit pixel */				++src;				RGB888_RGB332(*dst++, *src);				++src;				RGB888_RGB332(*dst++, *src);				++src;				RGB888_RGB332(*dst++, *src);				++src;			}			switch ( width & 3 ) {				case 3:					RGB888_RGB332(*dst++, *src);					++src;				case 2:					RGB888_RGB332(*dst++, *src);					++src;				case 1:					RGB888_RGB332(*dst++, *src);					++src;			}#endif /* USE_DUFFS_LOOP */			src += srcskip;			dst += dstskip;		}	} else {		int pixel;		while ( height-- ) {#ifdef USE_DUFFS_LOOP			DUFFS_LOOP(				RGB888_RGB332(pixel, *src);				*dst++ = map[pixel];				++src;			, width);#else			for ( c=width/4; c; --c ) {				/* Pack RGB into 8bit pixel */				RGB888_RGB332(pixel, *src);				*dst++ = map[pixel];				++src;				RGB888_RGB332(pixel, *src);				*dst++ = map[pixel];				++src;				RGB888_RGB332(pixel, *src);				*dst++ = map[pixel];				++src;				RGB888_RGB332(pixel, *src);				*dst++ = map[pixel];				++src;			}			switch ( width & 3 ) {				case 3:					RGB888_RGB332(pixel, *src);					*dst++ = map[pixel];					++src;				case 2:					RGB888_RGB332(pixel, *src);					*dst++ = map[pixel];					++src;				case 1:					RGB888_RGB332(pixel, *src);					*dst++ = map[pixel];					++src;			}#endif /* USE_DUFFS_LOOP */			src += srcskip;			dst += dstskip;		}	}}/* Special optimized blit for RGB 8-8-8 --> RGB 5-5-5 */#define RGB888_RGB555(dst, src) { \	*(Uint16 *)(dst) = (((*src)&0x00F80000)>>9)| \	                   (((*src)&0x0000F800)>>6)| \	                   (((*src)&0x000000F8)>>3); \}#define RGB888_RGB555_TWO(dst, src) { \	*(Uint32 *)(dst) = (((((src[HI])&0x00F80000)>>9)| \	                     (((src[HI])&0x0000F800)>>6)| \	                     (((src[HI])&0x000000F8)>>3))<<16)| \	                     (((src[LO])&0x00F80000)>>9)| \	                     (((src[LO])&0x0000F800)>>6)| \	                     (((src[LO])&0x000000F8)>>3); \}static void Blit_RGB888_RGB555(SDL_BlitInfo *info){#ifndef USE_DUFFS_LOOP	int c;#endif	int width, height;	Uint32 *src;	Uint16 *dst;	int srcskip, dstskip;	/* Set up some basic variables */	width = info->d_width;	height = info->d_height;	src = (Uint32 *)info->s_pixels;	srcskip = info->s_skip/4;	dst = (Uint16 *)info->d_pixels;	dstskip = info->d_skip/2;#ifdef USE_DUFFS_LOOP	while ( height-- ) {		DUFFS_LOOP(			RGB888_RGB555(dst, src);			++src;			++dst;		, width);		src += srcskip;		dst += dstskip;	}#else	/* Memory align at 4-byte boundary, if necessary */	if ( (long)dst & 0x03 ) {		/* Don't do anything if width is 0 */		if ( width == 0 ) {			return;		}		--width;		while ( height-- ) {			/* Perform copy alignment */			RGB888_RGB555(dst, src);			++src;			++dst;			/* Copy in 4 pixel chunks */			for ( c=width/4; c; --c ) {				RGB888_RGB555_TWO(dst, src);				src += 2;				dst += 2;				RGB888_RGB555_TWO(dst, src);				src += 2;				dst += 2;			}			/* Get any leftovers */			switch (width & 3) {				case 3:					RGB888_RGB555(dst, src);					++src;					++dst;				case 2:					RGB888_RGB555_TWO(dst, src);					src += 2;					dst += 2;					break;				case 1:					RGB888_RGB555(dst, src);					++src;					++dst;					break;			}			src += srcskip;			dst += dstskip;		}	} else { 		while ( height-- ) {			/* Copy in 4 pixel chunks */			for ( c=width/4; c; --c ) {				RGB888_RGB555_TWO(dst, src);				src += 2;				dst += 2;				RGB888_RGB555_TWO(dst, src);				src += 2;				dst += 2;			}			/* Get any leftovers */			switch (width & 3) {				case 3:					RGB888_RGB555(dst, src);					++src;					++dst;				case 2:					RGB888_RGB555_TWO(dst, src);					src += 2;					dst += 2;					break;				case 1:					RGB888_RGB555(dst, src);					++src;					++dst;					break;			}			src += srcskip;			dst += dstskip;		}	}#endif /* USE_DUFFS_LOOP */}/* Special optimized blit for RGB 8-8-8 --> RGB 5-6-5 */#define RGB888_RGB565(dst, src) { \	*(Uint16 *)(dst) = (((*src)&0x00F80000)>>8)| \	                   (((*src)&0x0000FC00)>>5)| \	                   (((*src)&0x000000F8)>>3); \}#define RGB888_RGB565_TWO(dst, src) { \	*(Uint32 *)(dst) = (((((src[HI])&0x00F80000)>>8)| \	                     (((src[HI])&0x0000FC00)>>5)| \	                     (((src[HI])&0x000000F8)>>3))<<16)| \	                     (((src[LO])&0x00F80000)>>8)| \	                     (((src[LO])&0x0000FC00)>>5)| \	                     (((src[LO])&0x000000F8)>>3); \}static void Blit_RGB888_RGB565(SDL_BlitInfo *info){#ifndef USE_DUFFS_LOOP	int c;#endif	int width, height;	Uint32 *src;	Uint16 *dst;	int srcskip, dstskip;	/* Set up some basic variables */	width = info->d_width;	height = info->d_height;	src = (Uint32 *)info->s_pixels;	srcskip = info->s_skip/4;	dst = (Uint16 *)info->d_pixels;	dstskip = info->d_skip/2;#ifdef USE_DUFFS_LOOP	while ( height-- ) {		DUFFS_LOOP(			RGB888_RGB565(dst, src);			++src;			++dst;		, width);		src += srcskip;		dst += dstskip;	}#else	/* Memory align at 4-byte boundary, if necessary */	if ( (long)dst & 0x03 ) {		/* Don't do anything if width is 0 */		if ( width == 0 ) {			return;		}		--width;		while ( height-- ) {			/* Perform copy alignment */			RGB888_RGB565(dst, src);			++src;			++dst;			/* Copy in 4 pixel chunks */			for ( c=width/4; c; --c ) {				RGB888_RGB565_TWO(dst, src);				src += 2;				dst += 2;				RGB888_RGB565_TWO(dst, src);				src += 2;				dst += 2;			}			/* Get any leftovers */			switch (width & 3) {				case 3:					RGB888_RGB565(dst, src);					++src;					++dst;				case 2:					RGB888_RGB565_TWO(dst, src);					src += 2;					dst += 2;					break;				case 1:					RGB888_RGB565(dst, src);					++src;					++dst;					break;			}			src += srcskip;			dst += dstskip;		}	} else { 		while ( height-- ) {			/* Copy in 4 pixel chunks */			for ( c=width/4; c; --c ) {				RGB888_RGB565_TWO(dst, src);				src += 2;				dst += 2;				RGB888_RGB565_TWO(dst, src);				src += 2;				dst += 2;			}			/* Get any leftovers */			switch (width & 3) {				case 3:					RGB888_RGB565(dst, src);					++src;					++dst;				case 2:					RGB888_RGB565_TWO(dst, src);					src += 2;					dst += 2;					break;				case 1:					RGB888_RGB565(dst, src);					++src;					++dst;					break;