view src/video/SDL_blit_A.c @ 5284:96a22141cf86 tip

Changed output directory of Universal libSDL.a for iOS to respect build configurations. Template generator was updated to reflect these changes as well.
author Eric Wing <ewing . public |-at-| gmail . com>
date Sat, 12 Feb 2011 21:52:30 -0800
parents b530ef003506
children
line wrap: on
line source

/*
    SDL - Simple DirectMedia Layer
    Copyright (C) 1997-2011 Sam Lantinga

    This library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
    License as published by the Free Software Foundation; either
    version 2.1 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
    Lesser General Public License for more details.

    You should have received a copy of the GNU Lesser General Public
    License along with this library; if not, write to the Free Software
    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA

    Sam Lantinga
    slouken@libsdl.org
*/
#include "SDL_config.h"

#include "SDL_video.h"
#include "SDL_blit.h"

/* Functions to perform alpha blended blitting */

/* N->1 blending with per-surface alpha */
static void
BlitNto1SurfaceAlpha(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint8 *src = info->src;
    int srcskip = info->src_skip;
    Uint8 *dst = info->dst;
    int dstskip = info->dst_skip;
    Uint8 *palmap = info->table;
    SDL_PixelFormat *srcfmt = info->src_fmt;
    SDL_PixelFormat *dstfmt = info->dst_fmt;
    int srcbpp = srcfmt->BytesPerPixel;

    const unsigned A = info->a;

    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4(
	    {
		Uint32 Pixel;
		unsigned sR;
		unsigned sG;
		unsigned sB;
		unsigned dR;
		unsigned dG;
		unsigned dB;
		DISEMBLE_RGB(src, srcbpp, srcfmt, Pixel, sR, sG, sB);
		dR = dstfmt->palette->colors[*dst].r;
		dG = dstfmt->palette->colors[*dst].g;
		dB = dstfmt->palette->colors[*dst].b;
		ALPHA_BLEND(sR, sG, sB, A, dR, dG, dB);
		dR &= 0xff;
		dG &= 0xff;
		dB &= 0xff;
		/* Pack RGB into 8bit pixel */
		if ( palmap == NULL ) {
		    *dst =((dR>>5)<<(3+2))|
			  ((dG>>5)<<(2))|
			  ((dB>>6)<<(0));
		} else {
		    *dst = palmap[((dR>>5)<<(3+2))|
				  ((dG>>5)<<(2))  |
				  ((dB>>6)<<(0))];
		}
		dst++;
		src += srcbpp;
	    },
	    width);
	    /* *INDENT-ON* */
        src += srcskip;
        dst += dstskip;
    }
}

/* N->1 blending with pixel alpha */
static void
BlitNto1PixelAlpha(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint8 *src = info->src;
    int srcskip = info->src_skip;
    Uint8 *dst = info->dst;
    int dstskip = info->dst_skip;
    Uint8 *palmap = info->table;
    SDL_PixelFormat *srcfmt = info->src_fmt;
    SDL_PixelFormat *dstfmt = info->dst_fmt;
    int srcbpp = srcfmt->BytesPerPixel;

    /* FIXME: fix alpha bit field expansion here too? */
    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4(
	    {
		Uint32 Pixel;
		unsigned sR;
		unsigned sG;
		unsigned sB;
		unsigned sA;
		unsigned dR;
		unsigned dG;
		unsigned dB;
		DISEMBLE_RGBA(src,srcbpp,srcfmt,Pixel,sR,sG,sB,sA);
		dR = dstfmt->palette->colors[*dst].r;
		dG = dstfmt->palette->colors[*dst].g;
		dB = dstfmt->palette->colors[*dst].b;
		ALPHA_BLEND(sR, sG, sB, sA, dR, dG, dB);
		dR &= 0xff;
		dG &= 0xff;
		dB &= 0xff;
		/* Pack RGB into 8bit pixel */
		if ( palmap == NULL ) {
		    *dst =((dR>>5)<<(3+2))|
			  ((dG>>5)<<(2))|
			  ((dB>>6)<<(0));
		} else {
		    *dst = palmap[((dR>>5)<<(3+2))|
				  ((dG>>5)<<(2))  |
				  ((dB>>6)<<(0))  ];
		}
		dst++;
		src += srcbpp;
	    },
	    width);
	    /* *INDENT-ON* */
        src += srcskip;
        dst += dstskip;
    }
}

/* colorkeyed N->1 blending with per-surface alpha */
static void
BlitNto1SurfaceAlphaKey(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint8 *src = info->src;
    int srcskip = info->src_skip;
    Uint8 *dst = info->dst;
    int dstskip = info->dst_skip;
    Uint8 *palmap = info->table;
    SDL_PixelFormat *srcfmt = info->src_fmt;
    SDL_PixelFormat *dstfmt = info->dst_fmt;
    int srcbpp = srcfmt->BytesPerPixel;
    Uint32 ckey = info->colorkey;

    const int A = info->a;

    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP(
	    {
		Uint32 Pixel;
		unsigned sR;
		unsigned sG;
		unsigned sB;
		unsigned dR;
		unsigned dG;
		unsigned dB;
		DISEMBLE_RGB(src, srcbpp, srcfmt, Pixel, sR, sG, sB);
		if ( Pixel != ckey ) {
		    dR = dstfmt->palette->colors[*dst].r;
		    dG = dstfmt->palette->colors[*dst].g;
		    dB = dstfmt->palette->colors[*dst].b;
		    ALPHA_BLEND(sR, sG, sB, A, dR, dG, dB);
		    dR &= 0xff;
		    dG &= 0xff;
		    dB &= 0xff;
		    /* Pack RGB into 8bit pixel */
		    if ( palmap == NULL ) {
			*dst =((dR>>5)<<(3+2))|
			      ((dG>>5)<<(2)) |
			      ((dB>>6)<<(0));
		    } else {
			*dst = palmap[((dR>>5)<<(3+2))|
				      ((dG>>5)<<(2))  |
				      ((dB>>6)<<(0))  ];
		    }
		}
		dst++;
		src += srcbpp;
	    },
	    width);
	    /* *INDENT-ON* */
        src += srcskip;
        dst += dstskip;
    }
}

#ifdef __MMX__

/* fast RGB888->(A)RGB888 blending with surface alpha=128 special case */
static void
BlitRGBtoRGBSurfaceAlpha128MMX(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint32 *srcp = (Uint32 *) info->src;
    int srcskip = info->src_skip >> 2;
    Uint32 *dstp = (Uint32 *) info->dst;
    int dstskip = info->dst_skip >> 2;
    Uint32 dalpha = info->dst_fmt->Amask;

    __m64 src1, src2, dst1, dst2, lmask, hmask, dsta;

    hmask = _mm_set_pi32(0x00fefefe, 0x00fefefe);       /* alpha128 mask -> hmask */
    lmask = _mm_set_pi32(0x00010101, 0x00010101);       /* !alpha128 mask -> lmask */
    dsta = _mm_set_pi32(dalpha, dalpha);        /* dst alpha mask -> dsta */

    while (height--) {
        int n = width;
        if (n & 1) {
            Uint32 s = *srcp++;
            Uint32 d = *dstp;
            *dstp++ = ((((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)
                       + (s & d & 0x00010101)) | dalpha;
            n--;
        }

        for (n >>= 1; n > 0; --n) {
            dst1 = *(__m64 *) dstp;     /* 2 x dst -> dst1(ARGBARGB) */
            dst2 = dst1;        /* 2 x dst -> dst2(ARGBARGB) */

            src1 = *(__m64 *) srcp;     /* 2 x src -> src1(ARGBARGB) */
            src2 = src1;        /* 2 x src -> src2(ARGBARGB) */

            dst2 = _mm_and_si64(dst2, hmask);   /* dst & mask -> dst2 */
            src2 = _mm_and_si64(src2, hmask);   /* src & mask -> src2 */
            src2 = _mm_add_pi32(src2, dst2);    /* dst2 + src2 -> src2 */
            src2 = _mm_srli_pi32(src2, 1);      /* src2 >> 1 -> src2 */

            dst1 = _mm_and_si64(dst1, src1);    /* src & dst -> dst1 */
            dst1 = _mm_and_si64(dst1, lmask);   /* dst1 & !mask -> dst1 */
            dst1 = _mm_add_pi32(dst1, src2);    /* src2 + dst1 -> dst1 */
            dst1 = _mm_or_si64(dst1, dsta);     /* dsta(full alpha) | dst1 -> dst1 */

            *(__m64 *) dstp = dst1;     /* dst1 -> 2 x dst pixels */
            dstp += 2;
            srcp += 2;
        }

        srcp += srcskip;
        dstp += dstskip;
    }
    _mm_empty();
}

/* fast RGB888->(A)RGB888 blending with surface alpha */
static void
BlitRGBtoRGBSurfaceAlphaMMX(SDL_BlitInfo * info)
{
    SDL_PixelFormat *df = info->dst_fmt;
    Uint32 chanmask = df->Rmask | df->Gmask | df->Bmask;
    unsigned alpha = info->a;

    if (alpha == 128 && (df->Rmask | df->Gmask | df->Bmask) == 0x00FFFFFF) {
        /* only call a128 version when R,G,B occupy lower bits */
        BlitRGBtoRGBSurfaceAlpha128MMX(info);
    } else {
        int width = info->dst_w;
        int height = info->dst_h;
        Uint32 *srcp = (Uint32 *) info->src;
        int srcskip = info->src_skip >> 2;
        Uint32 *dstp = (Uint32 *) info->dst;
        int dstskip = info->dst_skip >> 2;
        Uint32 dalpha = df->Amask;
        Uint32 amult;

        __m64 src1, src2, dst1, dst2, mm_alpha, mm_zero, dsta;

        mm_zero = _mm_setzero_si64();   /* 0 -> mm_zero */
        /* form the alpha mult */
        amult = alpha | (alpha << 8);
        amult = amult | (amult << 16);
        chanmask =
            (0xff << df->Rshift) | (0xff << df->
                                    Gshift) | (0xff << df->Bshift);
        mm_alpha = _mm_set_pi32(0, amult & chanmask);   /* 0000AAAA -> mm_alpha, minus 1 chan */
        mm_alpha = _mm_unpacklo_pi8(mm_alpha, mm_zero); /* 0A0A0A0A -> mm_alpha, minus 1 chan */
        /* at this point mm_alpha can be 000A0A0A or 0A0A0A00 or another combo */
        dsta = _mm_set_pi32(dalpha, dalpha);    /* dst alpha mask -> dsta */

        while (height--) {
            int n = width;
            if (n & 1) {
                /* One Pixel Blend */
                src2 = _mm_cvtsi32_si64(*srcp); /* src(ARGB) -> src2 (0000ARGB) */
                src2 = _mm_unpacklo_pi8(src2, mm_zero); /* 0A0R0G0B -> src2 */

                dst1 = _mm_cvtsi32_si64(*dstp); /* dst(ARGB) -> dst1 (0000ARGB) */
                dst1 = _mm_unpacklo_pi8(dst1, mm_zero); /* 0A0R0G0B -> dst1 */

                src2 = _mm_sub_pi16(src2, dst1);        /* src2 - dst2 -> src2 */
                src2 = _mm_mullo_pi16(src2, mm_alpha);  /* src2 * alpha -> src2 */
                src2 = _mm_srli_pi16(src2, 8);  /* src2 >> 8 -> src2 */
                dst1 = _mm_add_pi8(src2, dst1); /* src2 + dst1 -> dst1 */

                dst1 = _mm_packs_pu16(dst1, mm_zero);   /* 0000ARGB -> dst1 */
                dst1 = _mm_or_si64(dst1, dsta); /* dsta | dst1 -> dst1 */
                *dstp = _mm_cvtsi64_si32(dst1); /* dst1 -> pixel */

                ++srcp;
                ++dstp;

                n--;
            }

            for (n >>= 1; n > 0; --n) {
                /* Two Pixels Blend */
                src1 = *(__m64 *) srcp; /* 2 x src -> src1(ARGBARGB) */
                src2 = src1;    /* 2 x src -> src2(ARGBARGB) */
                src1 = _mm_unpacklo_pi8(src1, mm_zero); /* low - 0A0R0G0B -> src1 */
                src2 = _mm_unpackhi_pi8(src2, mm_zero); /* high - 0A0R0G0B -> src2 */

                dst1 = *(__m64 *) dstp; /* 2 x dst -> dst1(ARGBARGB) */
                dst2 = dst1;    /* 2 x dst -> dst2(ARGBARGB) */
                dst1 = _mm_unpacklo_pi8(dst1, mm_zero); /* low - 0A0R0G0B -> dst1 */
                dst2 = _mm_unpackhi_pi8(dst2, mm_zero); /* high - 0A0R0G0B -> dst2 */

                src1 = _mm_sub_pi16(src1, dst1);        /* src1 - dst1 -> src1 */
                src1 = _mm_mullo_pi16(src1, mm_alpha);  /* src1 * alpha -> src1 */
                src1 = _mm_srli_pi16(src1, 8);  /* src1 >> 8 -> src1 */
                dst1 = _mm_add_pi8(src1, dst1); /* src1 + dst1(dst1) -> dst1 */

                src2 = _mm_sub_pi16(src2, dst2);        /* src2 - dst2 -> src2 */
                src2 = _mm_mullo_pi16(src2, mm_alpha);  /* src2 * alpha -> src2 */
                src2 = _mm_srli_pi16(src2, 8);  /* src2 >> 8 -> src2 */
                dst2 = _mm_add_pi8(src2, dst2); /* src2 + dst2(dst2) -> dst2 */

                dst1 = _mm_packs_pu16(dst1, dst2);      /* 0A0R0G0B(res1), 0A0R0G0B(res2) -> dst1(ARGBARGB) */
                dst1 = _mm_or_si64(dst1, dsta); /* dsta | dst1 -> dst1 */

                *(__m64 *) dstp = dst1; /* dst1 -> 2 x pixel */

                srcp += 2;
                dstp += 2;
            }
            srcp += srcskip;
            dstp += dstskip;
        }
        _mm_empty();
    }
}

/* fast ARGB888->(A)RGB888 blending with pixel alpha */
static void
BlitRGBtoRGBPixelAlphaMMX(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint32 *srcp = (Uint32 *) info->src;
    int srcskip = info->src_skip >> 2;
    Uint32 *dstp = (Uint32 *) info->dst;
    int dstskip = info->dst_skip >> 2;
    SDL_PixelFormat *sf = info->src_fmt;
    Uint32 chanmask = sf->Rmask | sf->Gmask | sf->Bmask;
    Uint32 amask = sf->Amask;
    Uint32 ashift = sf->Ashift;
    Uint64 multmask;

    __m64 src1, dst1, mm_alpha, mm_zero, dmask;

    mm_zero = _mm_setzero_si64();       /* 0 -> mm_zero */
    multmask = 0xFFFF;
    multmask <<= (ashift * 2);
    multmask = ~multmask;
    dmask = *(__m64 *) & multmask;      /* dst alpha mask -> dmask */

    while (height--) {
		/* *INDENT-OFF* */
		DUFFS_LOOP4({
		Uint32 alpha = *srcp & amask;
		if (alpha == 0) {
			/* do nothing */
		} else if (alpha == amask) {
			/* opaque alpha -- copy RGB, keep dst alpha */
			*dstp = (*srcp & chanmask) | (*dstp & ~chanmask);
		} else {
			src1 = _mm_cvtsi32_si64(*srcp); /* src(ARGB) -> src1 (0000ARGB)*/
			src1 = _mm_unpacklo_pi8(src1, mm_zero); /* 0A0R0G0B -> src1 */

			dst1 = _mm_cvtsi32_si64(*dstp); /* dst(ARGB) -> dst1 (0000ARGB)*/
			dst1 = _mm_unpacklo_pi8(dst1, mm_zero); /* 0A0R0G0B -> dst1 */

			mm_alpha = _mm_cvtsi32_si64(alpha); /* alpha -> mm_alpha (0000000A) */
			mm_alpha = _mm_srli_si64(mm_alpha, ashift); /* mm_alpha >> ashift -> mm_alpha(0000000A) */
			mm_alpha = _mm_unpacklo_pi16(mm_alpha, mm_alpha); /* 00000A0A -> mm_alpha */
			mm_alpha = _mm_unpacklo_pi32(mm_alpha, mm_alpha); /* 0A0A0A0A -> mm_alpha */
			mm_alpha = _mm_and_si64(mm_alpha, dmask); /* 000A0A0A -> mm_alpha, preserve dst alpha on add */

			/* blend */		    
			src1 = _mm_sub_pi16(src1, dst1);/* src1 - dst1 -> src1 */
			src1 = _mm_mullo_pi16(src1, mm_alpha); /* (src1 - dst1) * alpha -> src1 */
			src1 = _mm_srli_pi16(src1, 8); /* src1 >> 8 -> src1(000R0G0B) */
			dst1 = _mm_add_pi8(src1, dst1); /* src1 + dst1 -> dst1(0A0R0G0B) */
			dst1 = _mm_packs_pu16(dst1, mm_zero);  /* 0000ARGB -> dst1 */
			
			*dstp = _mm_cvtsi64_si32(dst1); /* dst1 -> pixel */
		}
		++srcp;
		++dstp;
	    }, width);
		/* *INDENT-ON* */
        srcp += srcskip;
        dstp += dstskip;
    }
    _mm_empty();
}

#endif /* __MMX__ */

/* fast RGB888->(A)RGB888 blending with surface alpha=128 special case */
static void
BlitRGBtoRGBSurfaceAlpha128(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint32 *srcp = (Uint32 *) info->src;
    int srcskip = info->src_skip >> 2;
    Uint32 *dstp = (Uint32 *) info->dst;
    int dstskip = info->dst_skip >> 2;

    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4({
		    Uint32 s = *srcp++;
		    Uint32 d = *dstp;
		    *dstp++ = ((((s & 0x00fefefe) + (d & 0x00fefefe)) >> 1)
			       + (s & d & 0x00010101)) | 0xff000000;
	    }, width);
	    /* *INDENT-ON* */
        srcp += srcskip;
        dstp += dstskip;
    }
}

/* fast RGB888->(A)RGB888 blending with surface alpha */
static void
BlitRGBtoRGBSurfaceAlpha(SDL_BlitInfo * info)
{
    unsigned alpha = info->a;
    if (alpha == 128) {
        BlitRGBtoRGBSurfaceAlpha128(info);
    } else {
        int width = info->dst_w;
        int height = info->dst_h;
        Uint32 *srcp = (Uint32 *) info->src;
        int srcskip = info->src_skip >> 2;
        Uint32 *dstp = (Uint32 *) info->dst;
        int dstskip = info->dst_skip >> 2;
        Uint32 s;
        Uint32 d;
        Uint32 s1;
        Uint32 d1;

        while (height--) {
			/* *INDENT-OFF* */
			DUFFS_LOOP4({
				s = *srcp;
				d = *dstp;
				s1 = s & 0xff00ff;
				d1 = d & 0xff00ff;
				d1 = (d1 + ((s1 - d1) * alpha >> 8))
				     & 0xff00ff;
				s &= 0xff00;
				d &= 0xff00;
				d = (d + ((s - d) * alpha >> 8)) & 0xff00;
				*dstp = d1 | d | 0xff000000;
				++srcp;
				++dstp;
			}, width);
			/* *INDENT-ON* */
            srcp += srcskip;
            dstp += dstskip;
        }
    }
}

/* fast ARGB888->(A)RGB888 blending with pixel alpha */
static void
BlitRGBtoRGBPixelAlpha(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint32 *srcp = (Uint32 *) info->src;
    int srcskip = info->src_skip >> 2;
    Uint32 *dstp = (Uint32 *) info->dst;
    int dstskip = info->dst_skip >> 2;

    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4({
		Uint32 dalpha;
		Uint32 d;
		Uint32 s1;
		Uint32 d1;
		Uint32 s = *srcp;
		Uint32 alpha = s >> 24;
		/* FIXME: Here we special-case opaque alpha since the
		   compositioning used (>>8 instead of /255) doesn't handle
		   it correctly. Also special-case alpha=0 for speed?
		   Benchmark this! */
		if(alpha) {   
		  if(alpha == SDL_ALPHA_OPAQUE) {
		    *dstp = (s & 0x00ffffff) | (*dstp & 0xff000000);
		  } else {
		    /*
		     * take out the middle component (green), and process
		     * the other two in parallel. One multiply less.
		     */
		    d = *dstp;
		    dalpha = d & 0xff000000;
		    s1 = s & 0xff00ff;
		    d1 = d & 0xff00ff;
		    d1 = (d1 + ((s1 - d1) * alpha >> 8)) & 0xff00ff;
		    s &= 0xff00;
		    d &= 0xff00;
		    d = (d + ((s - d) * alpha >> 8)) & 0xff00;
		    *dstp = d1 | d | dalpha;
		  }
		}
		++srcp;
		++dstp;
	    }, width);
	    /* *INDENT-ON* */
        srcp += srcskip;
        dstp += dstskip;
    }
}

/* 16bpp special case for per-surface alpha=50%: blend 2 pixels in parallel */

/* blend a single 16 bit pixel at 50% */
#define BLEND16_50(d, s, mask)						\
	((((s & mask) + (d & mask)) >> 1) + (s & d & (~mask & 0xffff)))

/* blend two 16 bit pixels at 50% */
#define BLEND2x16_50(d, s, mask)					     \
	(((s & (mask | mask << 16)) >> 1) + ((d & (mask | mask << 16)) >> 1) \
	 + (s & d & (~(mask | mask << 16))))

static void
Blit16to16SurfaceAlpha128(SDL_BlitInfo * info, Uint16 mask)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint16 *srcp = (Uint16 *) info->src;
    int srcskip = info->src_skip >> 1;
    Uint16 *dstp = (Uint16 *) info->dst;
    int dstskip = info->dst_skip >> 1;

    while (height--) {
        if (((uintptr_t) srcp ^ (uintptr_t) dstp) & 2) {
            /*
             * Source and destination not aligned, pipeline it.
             * This is mostly a win for big blits but no loss for
             * small ones
             */
            Uint32 prev_sw;
            int w = width;

            /* handle odd destination */
            if ((uintptr_t) dstp & 2) {
                Uint16 d = *dstp, s = *srcp;
                *dstp = BLEND16_50(d, s, mask);
                dstp++;
                srcp++;
                w--;
            }
            srcp++;             /* srcp is now 32-bit aligned */

            /* bootstrap pipeline with first halfword */
            prev_sw = ((Uint32 *) srcp)[-1];

            while (w > 1) {
                Uint32 sw, dw, s;
                sw = *(Uint32 *) srcp;
                dw = *(Uint32 *) dstp;
#if SDL_BYTEORDER == SDL_BIG_ENDIAN
                s = (prev_sw << 16) + (sw >> 16);
#else
                s = (prev_sw >> 16) + (sw << 16);
#endif
                prev_sw = sw;
                *(Uint32 *) dstp = BLEND2x16_50(dw, s, mask);
                dstp += 2;
                srcp += 2;
                w -= 2;
            }

            /* final pixel if any */
            if (w) {
                Uint16 d = *dstp, s;
#if SDL_BYTEORDER == SDL_BIG_ENDIAN
                s = (Uint16) prev_sw;
#else
                s = (Uint16) (prev_sw >> 16);
#endif
                *dstp = BLEND16_50(d, s, mask);
                srcp++;
                dstp++;
            }
            srcp += srcskip - 1;
            dstp += dstskip;
        } else {
            /* source and destination are aligned */
            int w = width;

            /* first odd pixel? */
            if ((uintptr_t) srcp & 2) {
                Uint16 d = *dstp, s = *srcp;
                *dstp = BLEND16_50(d, s, mask);
                srcp++;
                dstp++;
                w--;
            }
            /* srcp and dstp are now 32-bit aligned */

            while (w > 1) {
                Uint32 sw = *(Uint32 *) srcp;
                Uint32 dw = *(Uint32 *) dstp;
                *(Uint32 *) dstp = BLEND2x16_50(dw, sw, mask);
                srcp += 2;
                dstp += 2;
                w -= 2;
            }

            /* last odd pixel? */
            if (w) {
                Uint16 d = *dstp, s = *srcp;
                *dstp = BLEND16_50(d, s, mask);
                srcp++;
                dstp++;
            }
            srcp += srcskip;
            dstp += dstskip;
        }
    }
}

#ifdef __MMX__

/* fast RGB565->RGB565 blending with surface alpha */
static void
Blit565to565SurfaceAlphaMMX(SDL_BlitInfo * info)
{
    unsigned alpha = info->a;
    if (alpha == 128) {
        Blit16to16SurfaceAlpha128(info, 0xf7de);
    } else {
        int width = info->dst_w;
        int height = info->dst_h;
        Uint16 *srcp = (Uint16 *) info->src;
        int srcskip = info->src_skip >> 1;
        Uint16 *dstp = (Uint16 *) info->dst;
        int dstskip = info->dst_skip >> 1;
        Uint32 s, d;

        __m64 src1, dst1, src2, dst2, gmask, bmask, mm_res, mm_alpha;

        alpha &= ~(1 + 2 + 4);  /* cut alpha to get the exact same behaviour */
        mm_alpha = _mm_set_pi32(0, alpha);      /* 0000000A -> mm_alpha */
        alpha >>= 3;            /* downscale alpha to 5 bits */

        mm_alpha = _mm_unpacklo_pi16(mm_alpha, mm_alpha);       /* 00000A0A -> mm_alpha */
        mm_alpha = _mm_unpacklo_pi32(mm_alpha, mm_alpha);       /* 0A0A0A0A -> mm_alpha */
        /* position alpha to allow for mullo and mulhi on diff channels
           to reduce the number of operations */
        mm_alpha = _mm_slli_si64(mm_alpha, 3);

        /* Setup the 565 color channel masks */
        gmask = _mm_set_pi32(0x07E007E0, 0x07E007E0);   /* MASKGREEN -> gmask */
        bmask = _mm_set_pi32(0x001F001F, 0x001F001F);   /* MASKBLUE -> bmask */

        while (height--) {
			/* *INDENT-OFF* */
			DUFFS_LOOP_124(
			{
				s = *srcp++;
				d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x07e0f81f;
				d = (d | d << 16) & 0x07e0f81f;
				d += (s - d) * alpha >> 5;
				d &= 0x07e0f81f;
				*dstp++ = (Uint16)(d | d >> 16);
			},{
				s = *srcp++;
				d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x07e0f81f;
				d = (d | d << 16) & 0x07e0f81f;
				d += (s - d) * alpha >> 5;
				d &= 0x07e0f81f;
				*dstp++ = (Uint16)(d | d >> 16);
				s = *srcp++;
				d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x07e0f81f;
				d = (d | d << 16) & 0x07e0f81f;
				d += (s - d) * alpha >> 5;
				d &= 0x07e0f81f;
				*dstp++ = (Uint16)(d | d >> 16);
			},{
				src1 = *(__m64*)srcp; /* 4 src pixels -> src1 */
				dst1 = *(__m64*)dstp; /* 4 dst pixels -> dst1 */

				/* red */
				src2 = src1;
				src2 = _mm_srli_pi16(src2, 11); /* src2 >> 11 -> src2 [000r 000r 000r 000r] */

				dst2 = dst1;
				dst2 = _mm_srli_pi16(dst2, 11); /* dst2 >> 11 -> dst2 [000r 000r 000r 000r] */

				/* blend */
				src2 = _mm_sub_pi16(src2, dst2);/* src - dst -> src2 */
				src2 = _mm_mullo_pi16(src2, mm_alpha); /* src2 * alpha -> src2 */
				src2 = _mm_srli_pi16(src2, 11); /* src2 >> 11 -> src2 */
				dst2 = _mm_add_pi16(src2, dst2); /* src2 + dst2 -> dst2 */
				dst2 = _mm_slli_pi16(dst2, 11); /* dst2 << 11 -> dst2 */

				mm_res = dst2; /* RED -> mm_res */

				/* green -- process the bits in place */
				src2 = src1;
				src2 = _mm_and_si64(src2, gmask); /* src & MASKGREEN -> src2 */

				dst2 = dst1;
				dst2 = _mm_and_si64(dst2, gmask); /* dst & MASKGREEN -> dst2 */

				/* blend */
				src2 = _mm_sub_pi16(src2, dst2);/* src - dst -> src2 */
				src2 = _mm_mulhi_pi16(src2, mm_alpha); /* src2 * alpha -> src2 */
				src2 = _mm_slli_pi16(src2, 5); /* src2 << 5 -> src2 */
				dst2 = _mm_add_pi16(src2, dst2); /* src2 + dst2 -> dst2 */

				mm_res = _mm_or_si64(mm_res, dst2); /* RED | GREEN -> mm_res */

				/* blue */
				src2 = src1;
				src2 = _mm_and_si64(src2, bmask); /* src & MASKBLUE -> src2[000b 000b 000b 000b] */

				dst2 = dst1;
				dst2 = _mm_and_si64(dst2, bmask); /* dst & MASKBLUE -> dst2[000b 000b 000b 000b] */

				/* blend */
				src2 = _mm_sub_pi16(src2, dst2);/* src - dst -> src2 */
				src2 = _mm_mullo_pi16(src2, mm_alpha); /* src2 * alpha -> src2 */
				src2 = _mm_srli_pi16(src2, 11); /* src2 >> 11 -> src2 */
				dst2 = _mm_add_pi16(src2, dst2); /* src2 + dst2 -> dst2 */
				dst2 = _mm_and_si64(dst2, bmask); /* dst2 & MASKBLUE -> dst2 */

				mm_res = _mm_or_si64(mm_res, dst2); /* RED | GREEN | BLUE -> mm_res */

				*(__m64*)dstp = mm_res; /* mm_res -> 4 dst pixels */

				srcp += 4;
				dstp += 4;
			}, width);
			/* *INDENT-ON* */
            srcp += srcskip;
            dstp += dstskip;
        }
        _mm_empty();
    }
}

/* fast RGB555->RGB555 blending with surface alpha */
static void
Blit555to555SurfaceAlphaMMX(SDL_BlitInfo * info)
{
    unsigned alpha = info->a;
    if (alpha == 128) {
        Blit16to16SurfaceAlpha128(info, 0xfbde);
    } else {
        int width = info->dst_w;
        int height = info->dst_h;
        Uint16 *srcp = (Uint16 *) info->src;
        int srcskip = info->src_skip >> 1;
        Uint16 *dstp = (Uint16 *) info->dst;
        int dstskip = info->dst_skip >> 1;
        Uint32 s, d;

        __m64 src1, dst1, src2, dst2, rmask, gmask, bmask, mm_res, mm_alpha;

        alpha &= ~(1 + 2 + 4);  /* cut alpha to get the exact same behaviour */
        mm_alpha = _mm_set_pi32(0, alpha);      /* 0000000A -> mm_alpha */
        alpha >>= 3;            /* downscale alpha to 5 bits */

        mm_alpha = _mm_unpacklo_pi16(mm_alpha, mm_alpha);       /* 00000A0A -> mm_alpha */
        mm_alpha = _mm_unpacklo_pi32(mm_alpha, mm_alpha);       /* 0A0A0A0A -> mm_alpha */
        /* position alpha to allow for mullo and mulhi on diff channels
           to reduce the number of operations */
        mm_alpha = _mm_slli_si64(mm_alpha, 3);

        /* Setup the 555 color channel masks */
        rmask = _mm_set_pi32(0x7C007C00, 0x7C007C00);   /* MASKRED -> rmask */
        gmask = _mm_set_pi32(0x03E003E0, 0x03E003E0);   /* MASKGREEN -> gmask */
        bmask = _mm_set_pi32(0x001F001F, 0x001F001F);   /* MASKBLUE -> bmask */

        while (height--) {
			/* *INDENT-OFF* */
			DUFFS_LOOP_124(
			{
				s = *srcp++;
				d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x03e07c1f;
				d = (d | d << 16) & 0x03e07c1f;
				d += (s - d) * alpha >> 5;
				d &= 0x03e07c1f;
				*dstp++ = (Uint16)(d | d >> 16);
			},{
				s = *srcp++;
				d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x03e07c1f;
				d = (d | d << 16) & 0x03e07c1f;
				d += (s - d) * alpha >> 5;
				d &= 0x03e07c1f;
				*dstp++ = (Uint16)(d | d >> 16);
			        s = *srcp++;
				d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x03e07c1f;
				d = (d | d << 16) & 0x03e07c1f;
				d += (s - d) * alpha >> 5;
				d &= 0x03e07c1f;
				*dstp++ = (Uint16)(d | d >> 16);
			},{
				src1 = *(__m64*)srcp; /* 4 src pixels -> src1 */
				dst1 = *(__m64*)dstp; /* 4 dst pixels -> dst1 */

				/* red -- process the bits in place */
				src2 = src1;
				src2 = _mm_and_si64(src2, rmask); /* src & MASKRED -> src2 */

				dst2 = dst1;
				dst2 = _mm_and_si64(dst2, rmask); /* dst & MASKRED -> dst2 */

				/* blend */
				src2 = _mm_sub_pi16(src2, dst2);/* src - dst -> src2 */
				src2 = _mm_mulhi_pi16(src2, mm_alpha); /* src2 * alpha -> src2 */
				src2 = _mm_slli_pi16(src2, 5); /* src2 << 5 -> src2 */
				dst2 = _mm_add_pi16(src2, dst2); /* src2 + dst2 -> dst2 */
				dst2 = _mm_and_si64(dst2, rmask); /* dst2 & MASKRED -> dst2 */

				mm_res = dst2; /* RED -> mm_res */
				
				/* green -- process the bits in place */
				src2 = src1;
				src2 = _mm_and_si64(src2, gmask); /* src & MASKGREEN -> src2 */

				dst2 = dst1;
				dst2 = _mm_and_si64(dst2, gmask); /* dst & MASKGREEN -> dst2 */

				/* blend */
				src2 = _mm_sub_pi16(src2, dst2);/* src - dst -> src2 */
				src2 = _mm_mulhi_pi16(src2, mm_alpha); /* src2 * alpha -> src2 */
				src2 = _mm_slli_pi16(src2, 5); /* src2 << 5 -> src2 */
				dst2 = _mm_add_pi16(src2, dst2); /* src2 + dst2 -> dst2 */

				mm_res = _mm_or_si64(mm_res, dst2); /* RED | GREEN -> mm_res */

				/* blue */
				src2 = src1; /* src -> src2 */
				src2 = _mm_and_si64(src2, bmask); /* src & MASKBLUE -> src2[000b 000b 000b 000b] */

				dst2 = dst1; /* dst -> dst2 */
				dst2 = _mm_and_si64(dst2, bmask); /* dst & MASKBLUE -> dst2[000b 000b 000b 000b] */

				/* blend */
				src2 = _mm_sub_pi16(src2, dst2);/* src - dst -> src2 */
				src2 = _mm_mullo_pi16(src2, mm_alpha); /* src2 * alpha -> src2 */
				src2 = _mm_srli_pi16(src2, 11); /* src2 >> 11 -> src2 */
				dst2 = _mm_add_pi16(src2, dst2); /* src2 + dst2 -> dst2 */
				dst2 = _mm_and_si64(dst2, bmask); /* dst2 & MASKBLUE -> dst2 */

				mm_res = _mm_or_si64(mm_res, dst2); /* RED | GREEN | BLUE -> mm_res */

				*(__m64*)dstp = mm_res; /* mm_res -> 4 dst pixels */

				srcp += 4;
				dstp += 4;
			}, width);
			/* *INDENT-ON* */
            srcp += srcskip;
            dstp += dstskip;
        }
        _mm_empty();
    }
}

#endif /* __MMX__ */

/* fast RGB565->RGB565 blending with surface alpha */
static void
Blit565to565SurfaceAlpha(SDL_BlitInfo * info)
{
    unsigned alpha = info->a;
    if (alpha == 128) {
        Blit16to16SurfaceAlpha128(info, 0xf7de);
    } else {
        int width = info->dst_w;
        int height = info->dst_h;
        Uint16 *srcp = (Uint16 *) info->src;
        int srcskip = info->src_skip >> 1;
        Uint16 *dstp = (Uint16 *) info->dst;
        int dstskip = info->dst_skip >> 1;
        alpha >>= 3;            /* downscale alpha to 5 bits */

        while (height--) {
			/* *INDENT-OFF* */
			DUFFS_LOOP4({
				Uint32 s = *srcp++;
				Uint32 d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x07e0f81f;
				d = (d | d << 16) & 0x07e0f81f;
				d += (s - d) * alpha >> 5;
				d &= 0x07e0f81f;
				*dstp++ = (Uint16)(d | d >> 16);
			}, width);
			/* *INDENT-ON* */
            srcp += srcskip;
            dstp += dstskip;
        }
    }
}

/* fast RGB555->RGB555 blending with surface alpha */
static void
Blit555to555SurfaceAlpha(SDL_BlitInfo * info)
{
    unsigned alpha = info->a;   /* downscale alpha to 5 bits */
    if (alpha == 128) {
        Blit16to16SurfaceAlpha128(info, 0xfbde);
    } else {
        int width = info->dst_w;
        int height = info->dst_h;
        Uint16 *srcp = (Uint16 *) info->src;
        int srcskip = info->src_skip >> 1;
        Uint16 *dstp = (Uint16 *) info->dst;
        int dstskip = info->dst_skip >> 1;
        alpha >>= 3;            /* downscale alpha to 5 bits */

        while (height--) {
			/* *INDENT-OFF* */
			DUFFS_LOOP4({
				Uint32 s = *srcp++;
				Uint32 d = *dstp;
				/*
				 * shift out the middle component (green) to
				 * the high 16 bits, and process all three RGB
				 * components at the same time.
				 */
				s = (s | s << 16) & 0x03e07c1f;
				d = (d | d << 16) & 0x03e07c1f;
				d += (s - d) * alpha >> 5;
				d &= 0x03e07c1f;
				*dstp++ = (Uint16)(d | d >> 16);
			}, width);
			/* *INDENT-ON* */
            srcp += srcskip;
            dstp += dstskip;
        }
    }
}

/* fast ARGB8888->RGB565 blending with pixel alpha */
static void
BlitARGBto565PixelAlpha(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint32 *srcp = (Uint32 *) info->src;
    int srcskip = info->src_skip >> 2;
    Uint16 *dstp = (Uint16 *) info->dst;
    int dstskip = info->dst_skip >> 1;

    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4({
		Uint32 s = *srcp;
		unsigned alpha = s >> 27; /* downscale alpha to 5 bits */
		/* FIXME: Here we special-case opaque alpha since the
		   compositioning used (>>8 instead of /255) doesn't handle
		   it correctly. Also special-case alpha=0 for speed?
		   Benchmark this! */
		if(alpha) {   
		  if(alpha == (SDL_ALPHA_OPAQUE >> 3)) {
		    *dstp = (Uint16)((s >> 8 & 0xf800) + (s >> 5 & 0x7e0) + (s >> 3  & 0x1f));
		  } else {
		    Uint32 d = *dstp;
		    /*
		     * convert source and destination to G0RAB65565
		     * and blend all components at the same time
		     */
		    s = ((s & 0xfc00) << 11) + (s >> 8 & 0xf800)
		      + (s >> 3 & 0x1f);
		    d = (d | d << 16) & 0x07e0f81f;
		    d += (s - d) * alpha >> 5;
		    d &= 0x07e0f81f;
		    *dstp = (Uint16)(d | d >> 16);
		  }
		}
		srcp++;
		dstp++;
	    }, width);
	    /* *INDENT-ON* */
        srcp += srcskip;
        dstp += dstskip;
    }
}

/* fast ARGB8888->RGB555 blending with pixel alpha */
static void
BlitARGBto555PixelAlpha(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint32 *srcp = (Uint32 *) info->src;
    int srcskip = info->src_skip >> 2;
    Uint16 *dstp = (Uint16 *) info->dst;
    int dstskip = info->dst_skip >> 1;

    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4({
		unsigned alpha;
		Uint32 s = *srcp;
		alpha = s >> 27; /* downscale alpha to 5 bits */
		/* FIXME: Here we special-case opaque alpha since the
		   compositioning used (>>8 instead of /255) doesn't handle
		   it correctly. Also special-case alpha=0 for speed?
		   Benchmark this! */
		if(alpha) {   
		  if(alpha == (SDL_ALPHA_OPAQUE >> 3)) {
		    *dstp = (Uint16)((s >> 9 & 0x7c00) + (s >> 6 & 0x3e0) + (s >> 3  & 0x1f));
		  } else {
		    Uint32 d = *dstp;
		    /*
		     * convert source and destination to G0RAB65565
		     * and blend all components at the same time
		     */
		    s = ((s & 0xf800) << 10) + (s >> 9 & 0x7c00)
		      + (s >> 3 & 0x1f);
		    d = (d | d << 16) & 0x03e07c1f;
		    d += (s - d) * alpha >> 5;
		    d &= 0x03e07c1f;
		    *dstp = (Uint16)(d | d >> 16);
		  }
		}
		srcp++;
		dstp++;
	    }, width);
	    /* *INDENT-ON* */
        srcp += srcskip;
        dstp += dstskip;
    }
}

/* General (slow) N->N blending with per-surface alpha */
static void
BlitNtoNSurfaceAlpha(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint8 *src = info->src;
    int srcskip = info->src_skip;
    Uint8 *dst = info->dst;
    int dstskip = info->dst_skip;
    SDL_PixelFormat *srcfmt = info->src_fmt;
    SDL_PixelFormat *dstfmt = info->dst_fmt;
    int srcbpp = srcfmt->BytesPerPixel;
    int dstbpp = dstfmt->BytesPerPixel;
    unsigned sA = info->a;
    unsigned dA = dstfmt->Amask ? SDL_ALPHA_OPAQUE : 0;

    if (sA) {
        while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4(
	    {
		Uint32 Pixel;
		unsigned sR;
		unsigned sG;
		unsigned sB;
		unsigned dR;
		unsigned dG;
		unsigned dB;
		DISEMBLE_RGB(src, srcbpp, srcfmt, Pixel, sR, sG, sB);
		DISEMBLE_RGB(dst, dstbpp, dstfmt, Pixel, dR, dG, dB);
		ALPHA_BLEND(sR, sG, sB, sA, dR, dG, dB);
		ASSEMBLE_RGBA(dst, dstbpp, dstfmt, dR, dG, dB, dA);
		src += srcbpp;
		dst += dstbpp;
	    },
	    width);
	    /* *INDENT-ON* */
            src += srcskip;
            dst += dstskip;
        }
    }
}

/* General (slow) colorkeyed N->N blending with per-surface alpha */
static void
BlitNtoNSurfaceAlphaKey(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint8 *src = info->src;
    int srcskip = info->src_skip;
    Uint8 *dst = info->dst;
    int dstskip = info->dst_skip;
    SDL_PixelFormat *srcfmt = info->src_fmt;
    SDL_PixelFormat *dstfmt = info->dst_fmt;
    Uint32 ckey = info->colorkey;
    int srcbpp = srcfmt->BytesPerPixel;
    int dstbpp = dstfmt->BytesPerPixel;
    unsigned sA = info->a;
    unsigned dA = dstfmt->Amask ? SDL_ALPHA_OPAQUE : 0;

    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4(
	    {
		Uint32 Pixel;
		unsigned sR;
		unsigned sG;
		unsigned sB;
		unsigned dR;
		unsigned dG;
		unsigned dB;
		RETRIEVE_RGB_PIXEL(src, srcbpp, Pixel);
		if(sA && Pixel != ckey) {
		    RGB_FROM_PIXEL(Pixel, srcfmt, sR, sG, sB);
		    DISEMBLE_RGB(dst, dstbpp, dstfmt, Pixel, dR, dG, dB);
		    ALPHA_BLEND(sR, sG, sB, sA, dR, dG, dB);
		    ASSEMBLE_RGBA(dst, dstbpp, dstfmt, dR, dG, dB, dA);
		}
		src += srcbpp;
		dst += dstbpp;
	    },
	    width);
	    /* *INDENT-ON* */
        src += srcskip;
        dst += dstskip;
    }
}

/* General (slow) N->N blending with pixel alpha */
static void
BlitNtoNPixelAlpha(SDL_BlitInfo * info)
{
    int width = info->dst_w;
    int height = info->dst_h;
    Uint8 *src = info->src;
    int srcskip = info->src_skip;
    Uint8 *dst = info->dst;
    int dstskip = info->dst_skip;
    SDL_PixelFormat *srcfmt = info->src_fmt;
    SDL_PixelFormat *dstfmt = info->dst_fmt;

    int srcbpp;
    int dstbpp;

    /* Set up some basic variables */
    srcbpp = srcfmt->BytesPerPixel;
    dstbpp = dstfmt->BytesPerPixel;

    /* FIXME: for 8bpp source alpha, this doesn't get opaque values
       quite right. for <8bpp source alpha, it gets them very wrong
       (check all macros!)
       It is unclear whether there is a good general solution that doesn't
       need a branch (or a divide). */
    while (height--) {
	    /* *INDENT-OFF* */
	    DUFFS_LOOP4(
	    {
		Uint32 Pixel;
		unsigned sR;
		unsigned sG;
		unsigned sB;
		unsigned dR;
		unsigned dG;
		unsigned dB;
		unsigned sA;
		unsigned dA;
		DISEMBLE_RGBA(src, srcbpp, srcfmt, Pixel, sR, sG, sB, sA);
		if(sA) {
		  DISEMBLE_RGBA(dst, dstbpp, dstfmt, Pixel, dR, dG, dB, dA);
		  ALPHA_BLEND(sR, sG, sB, sA, dR, dG, dB);
		  ASSEMBLE_RGBA(dst, dstbpp, dstfmt, dR, dG, dB, dA);
		}
		src += srcbpp;
		dst += dstbpp;
	    },
	    width);
	    /* *INDENT-ON* */
        src += srcskip;
        dst += dstskip;
    }
}


SDL_BlitFunc
SDL_CalculateBlitA(SDL_Surface * surface)
{
    SDL_PixelFormat *sf = surface->format;
    SDL_PixelFormat *df = surface->map->dst->format;

    switch (surface->map->info.flags & ~SDL_COPY_RLE_MASK) {
    case SDL_COPY_BLEND:
        /* Per-pixel alpha blits */
        switch (df->BytesPerPixel) {
        case 1:
            return BlitNto1PixelAlpha;

        case 2:
            if (sf->BytesPerPixel == 4 && sf->Amask == 0xff000000
                && sf->Gmask == 0xff00
                && ((sf->Rmask == 0xff && df->Rmask == 0x1f)
                    || (sf->Bmask == 0xff && df->Bmask == 0x1f))) {
                if (df->Gmask == 0x7e0)
                    return BlitARGBto565PixelAlpha;
                else if (df->Gmask == 0x3e0)
                    return BlitARGBto555PixelAlpha;
            }
            return BlitNtoNPixelAlpha;

        case 4:
            if (sf->Rmask == df->Rmask
                && sf->Gmask == df->Gmask
                && sf->Bmask == df->Bmask && sf->BytesPerPixel == 4) {
#if defined(__MMX__)
                if (sf->Rshift % 8 == 0
                    && sf->Gshift % 8 == 0
                    && sf->Bshift % 8 == 0
                    && sf->Ashift % 8 == 0 && sf->Aloss == 0) {
                    if (SDL_HasMMX())
                        return BlitRGBtoRGBPixelAlphaMMX;
                }
#endif /* __MMX__ */
                if (sf->Amask == 0xff000000) {
                    return BlitRGBtoRGBPixelAlpha;
                }
            }
            return BlitNtoNPixelAlpha;

        case 3:
        default:
            return BlitNtoNPixelAlpha;
        }
        break;

    case SDL_COPY_MODULATE_ALPHA | SDL_COPY_BLEND:
        if (sf->Amask == 0) {
            /* Per-surface alpha blits */
            switch (df->BytesPerPixel) {
            case 1:
                return BlitNto1SurfaceAlpha;

            case 2:
                if (surface->map->identity) {
                    if (df->Gmask == 0x7e0) {
#ifdef __MMX__
                        if (SDL_HasMMX())
                            return Blit565to565SurfaceAlphaMMX;
                        else
#endif
                            return Blit565to565SurfaceAlpha;
                    } else if (df->Gmask == 0x3e0) {
#ifdef __MMX__
                        if (SDL_HasMMX())
                            return Blit555to555SurfaceAlphaMMX;
                        else
#endif
                            return Blit555to555SurfaceAlpha;
                    }
                }
                return BlitNtoNSurfaceAlpha;

            case 4:
                if (sf->Rmask == df->Rmask
                    && sf->Gmask == df->Gmask
                    && sf->Bmask == df->Bmask && sf->BytesPerPixel == 4) {
#ifdef __MMX__
                    if (sf->Rshift % 8 == 0
                        && sf->Gshift % 8 == 0
                        && sf->Bshift % 8 == 0 && SDL_HasMMX())
                        return BlitRGBtoRGBSurfaceAlphaMMX;
#endif
                    if ((sf->Rmask | sf->Gmask | sf->Bmask) == 0xffffff) {
                        return BlitRGBtoRGBSurfaceAlpha;
                    }
                }
                return BlitNtoNSurfaceAlpha;

            case 3:
            default:
                return BlitNtoNSurfaceAlpha;
            }
        }
        break;

    case SDL_COPY_COLORKEY | SDL_COPY_MODULATE_ALPHA | SDL_COPY_BLEND:
        if (sf->Amask == 0) {
            if (df->BytesPerPixel == 1)
                return BlitNto1SurfaceAlphaKey;
            else
                return BlitNtoNSurfaceAlphaKey;
        }
        break;
    }

    return NULL;
}

/* vi: set ts=4 sw=4 expandtab: */