fbmmx.c

按照官方的说法：Cairo is a vector graphics library with cross-device output support. 翻译过来

        *dest = store8888(s);
        ++src;
        ++mask;
        ++dest;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineOverC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 sa = expand_alpha(s);

	*dest = store8888(in_over (s, sa, a, d));

        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineOverReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 da = expand_alpha(d);

	*dest = store8888(over (d, da, in (s, a)));

        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineInC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 da = expand_alpha(d);
        s = pix_multiply(s, a);
        s = pix_multiply(s, da);
        *dest = store8888(s);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineInReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 sa = expand_alpha(s);
        a = pix_multiply(a, sa);
        d = pix_multiply(d, a);
        *dest = store8888(d);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineOutC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 da = expand_alpha(d);
        da = negate(da);
        s = pix_multiply(s, a);
        s = pix_multiply(s, da);
        *dest = store8888(s);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineOutReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 sa = expand_alpha(s);
        a = pix_multiply(a, sa);
        a = negate(a);
        d = pix_multiply(d, a);
        *dest = store8888(d);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineAtopC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 da = expand_alpha(d);
        __m64 sa = expand_alpha(s);
        s = pix_multiply(s, a);
        a = pix_multiply(a, sa);
        a = negate(a);
	d = pix_add_mul (d, a, s, da);
        *dest = store8888(d);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineAtopReverseC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 da = expand_alpha(d);
        __m64 sa = expand_alpha(s);
        s = pix_multiply(s, a);
        a = pix_multiply(a, sa);
        da = negate(da);
	d = pix_add_mul (d, a, s, da);
        *dest = store8888(d);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineXorC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        __m64 da = expand_alpha(d);
        __m64 sa = expand_alpha(s);
        s = pix_multiply(s, a);
        a = pix_multiply(a, sa);
        da = negate(da);
        a = negate(a);
	d = pix_add_mul (d, a, s, da);
        *dest = store8888(d);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

static FASTCALL void
mmxCombineAddC (CARD32 *dest, CARD32 *src, CARD32 *mask, int width)
{
    const CARD32 *end = src + width;
    while (src < end) {
        __m64 a = load8888(*mask);
        __m64 s = load8888(*src);
        __m64 d = load8888(*dest);
        s = pix_multiply(s, a);
        d = pix_add(s, d);
        *dest = store8888(d);
        ++src;
        ++dest;
        ++mask;
    }
    _mm_empty();
}

extern FbComposeFunctions composeFunctions;

void fbComposeSetupMMX(void)
{
    /* check if we have MMX support and initialize accordingly */
    if (fbHaveMMX()) {
        composeFunctions.combineU[PIXMAN_OPERATOR_OVER] = mmxCombineOverU;
        composeFunctions.combineU[PIXMAN_OPERATOR_OVER_REVERSE] = mmxCombineOverReverseU;
        composeFunctions.combineU[PIXMAN_OPERATOR_IN] = mmxCombineInU;
        composeFunctions.combineU[PIXMAN_OPERATOR_IN_REVERSE] = mmxCombineInReverseU;
        composeFunctions.combineU[PIXMAN_OPERATOR_OUT] = mmxCombineOutU;
        composeFunctions.combineU[PIXMAN_OPERATOR_OUT_REVERSE] = mmxCombineOutReverseU;
        composeFunctions.combineU[PIXMAN_OPERATOR_ATOP] = mmxCombineAtopU;
        composeFunctions.combineU[PIXMAN_OPERATOR_ATOP_REVERSE] = mmxCombineAtopReverseU;
        composeFunctions.combineU[PIXMAN_OPERATOR_XOR] = mmxCombineXorU;
        composeFunctions.combineU[PIXMAN_OPERATOR_ADD] = mmxCombineAddU;
        composeFunctions.combineU[PIXMAN_OPERATOR_SATURATE] = mmxCombineSaturateU;

        composeFunctions.combineC[PIXMAN_OPERATOR_SRC] = mmxCombineSrcC;
        composeFunctions.combineC[PIXMAN_OPERATOR_OVER] = mmxCombineOverC;
        composeFunctions.combineC[PIXMAN_OPERATOR_OVER_REVERSE] = mmxCombineOverReverseC;
        composeFunctions.combineC[PIXMAN_OPERATOR_IN] = mmxCombineInC;
        composeFunctions.combineC[PIXMAN_OPERATOR_IN_REVERSE] = mmxCombineInReverseC;
        composeFunctions.combineC[PIXMAN_OPERATOR_OUT] = mmxCombineOutC;
        composeFunctions.combineC[PIXMAN_OPERATOR_OUT_REVERSE] = mmxCombineOutReverseC;
        composeFunctions.combineC[PIXMAN_OPERATOR_ATOP] = mmxCombineAtopC;
        composeFunctions.combineC[PIXMAN_OPERATOR_ATOP_REVERSE] = mmxCombineAtopReverseC;
        composeFunctions.combineC[PIXMAN_OPERATOR_XOR] = mmxCombineXorC;
        composeFunctions.combineC[PIXMAN_OPERATOR_ADD] = mmxCombineAddC;

        composeFunctions.combineMaskU = mmxCombineMaskU;
    }
}

/* ------------------ MMX code paths called from fbpict.c ----------------------- */

void
fbCompositeSolid_nx8888mmx (pixman_operator_t	op,
			    PicturePtr pSrc,
			    PicturePtr pMask,
			    PicturePtr pDst,
			    INT16	xSrc,
			    INT16	ySrc,
			    INT16	xMask,
			    INT16	yMask,
			    INT16	xDst,
			    INT16	yDst,
			    CARD16	width,
			    CARD16	height)
{
    CARD32	src;
    CARD32	*dstLine, *dst;
    CARD16	w;
    FbStride	dstStride;
    __m64	vsrc, vsrca;

    CHECKPOINT();

    fbComposeGetSolid(pSrc, pDst, src);

    if (src >> 24 == 0)
	return;

    fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);

    vsrc = load8888 (src);
    vsrca = expand_alpha (vsrc);

    while (height--)
    {
	dst = dstLine;
	dstLine += dstStride;
	w = width;

	CHECKPOINT();

	while (w && (unsigned long)dst & 7)
	{
	    *dst = store8888(over(vsrc, vsrca, load8888(*dst)));

	    w--;
	    dst++;
	}

	while (w >= 2)
	{
	    __m64 vdest;
	    __m64 dest0, dest1;

	    vdest = *(__m64 *)dst;

	    dest0 = over(vsrc, vsrca, expand8888(vdest, 0));
	    dest1 = over(vsrc, vsrca, expand8888(vdest, 1));

	    *(__m64 *)dst = pack8888(dest0, dest1);

	    dst += 2;
	    w -= 2;
	}

	CHECKPOINT();

	while (w)
	{
	    *dst = store8888(over(vsrc, vsrca, load8888(*dst)));

	    w--;
	    dst++;
	}
    }

    _mm_empty();
}

void
fbCompositeSolid_nx0565mmx (pixman_operator_t	op,
			    PicturePtr pSrc,
			    PicturePtr pMask,
			    PicturePtr pDst,
			    INT16	xSrc,
			    INT16	ySrc,
			    INT16	xMask,
			    INT16	yMask,
			    INT16	xDst,
			    INT16	yDst,
			    CARD16	width,
			    CARD16	height)
{
    CARD32	src;
    CARD16	*dstLine, *dst;
    CARD16	w;
    FbStride	dstStride;
    __m64	vsrc, vsrca;

    CHECKPOINT();

    fbComposeGetSolid(pSrc, pDst, src);

    if (src >> 24 == 0)
	return;

    fbComposeGetStart (pDst, xDst, yDst, CARD16, dstStride, dstLine, 1);

    vsrc = load8888 (src);
    vsrca = expand_alpha (vsrc);

    while (height--)
    {
	dst = dstLine;
	dstLine += dstStride;
	w = width;

	CHECKPOINT();

	while (w && (unsigned long)dst & 7)
	{
	    ullong d = *dst;
	    __m64 vdest = expand565 ((__m64)d, 0);
	    vdest = pack565(over(vsrc, vsrca, vdest), vdest, 0);
	    *dst = (ullong)vdest;

	    w--;
	    dst++;
	}

	while (w >= 4)
	{
	    __m64 vdest;

	    vdest = *(__m64 *)dst;

	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 0)), vdest, 0);
	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 1)), vdest, 1);
	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 2)), vdest, 2);
	    vdest = pack565 (over(vsrc, vsrca, expand565(vdest, 3)), vdest, 3);

	    *(__m64 *)dst = vdest;

	    dst += 4;
	    w -= 4;
	}

	CHECKPOINT();

	while (w)
	{
	    ullong d = *dst;
	    __m64 vdest = expand565 ((__m64)d, 0);
	    vdest = pack565(over(vsrc, vsrca, vdest), vdest, 0);
	    *dst = (ullong)vdest;

	    w--;
	    dst++;
	}
    }

    _mm_empty();
}

void
fbCompositeSolidMask_nx8888x8888Cmmx (pixman_operator_t	op,
				      PicturePtr pSrc,
				      PicturePtr pMask,
				      PicturePtr pDst,
				      INT16	xSrc,
				      INT16	ySrc,
				      INT16	xMask,
				      INT16	yMask,
				      INT16	xDst,
				      INT16	yDst,
				      CARD16	width,
				      CARD16	height)
{
    CARD32	src, srca;
    CARD32	*dstLine;
    CARD32	*maskLine;
    FbStride	dstStride, maskStride;
    __m64	vsrc, vsrca;

    CHECKPOINT();

    fbComposeGetSolid(pSrc, pDst, src);

    srca = src >> 24;
    if (srca == 0)
	return;

    fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
    fbComposeGetStart (pMask, xMask, yMask, CARD32, maskStride, maskLine, 1);

    vsrc = load8888(src);
    vsrca = expand_alpha(vsrc);

    while (height--)
    {
	int twidth = width;
	CARD32 *p = (CARD32 *)maskLine;
	CARD32 *q = (CARD32 *)dstLine;

	while (twidth && (unsigned long)q & 7)
	{
	    CARD32 m = *(CARD32 *)p;

	    if (m)
	    {
		__m64 vdest = load8888(*q);
		vdest = in_over(vsrc, vsrca, load8888(m), vdest);
		*q = store8888(vdest);
	    }

	    twidth--;
	    p++;
	    q++;
	}

	while (twidth >= 2)
	{
	    CARD32 m0, m1;
	    m0 = *p;
	    m1 = *(p + 1);

	    if (m0 | m1)
	    {
		__m64 dest0, dest1;
		__m64 vdest = *(__m64 *)q;

		dest0 = in_over(vsrc, vsrca, load8888(m0),
				expand8888 (vdest, 0));
		dest1 = in_over(vsrc, vsrca, load8888(m1),
				expand8888 (vdest, 1));

		*(__m64 *)q = pack8888(dest0, dest1);
	    }

	    p += 2;
	    q += 2;
	    twidth -= 2;
	}

	while (twidth)
	{
	    CARD32 m = *(CARD32 *)p;

	    if (m)
	    {
		__m64 vdest = load8888(*q);
		vdest = in_over(vsrc, vsrca, load8888(m), vdest);
		*q = store8888(vdest);
	    }

	    twidth--;
	    p++;
	    q++;
	}

	dstLine += dstStride;
	maskLine += maskStride;
    }

    _mm_empty();
}

void
fbCompositeSrc_8888x8x8888mmx (pixman_operator_t	op,
			       PicturePtr pSrc,
			       PicturePtr pMask,
			       PicturePtr pDst,
			       INT16	xSrc,
			       INT16	ySrc,
			       INT16      xMask,
			       INT16      yMask,
			       INT16      xDst,
			       INT16      yDst,
			       CARD16     width,
			       CARD16     height)
{
    CARD32	*dstLine, *dst;
    CARD32	*srcLine, *src;
    CARD8	*maskLine;
    CARD32	mask;
    __m64	vmask;
    FbStride	dstStride, srcStride, maskStride;
    CARD16	w;
    __m64  srca;

    CHECKPOINT();

    fbComposeGetStart (pDst, xDst, yDst, CARD32, dstStride, dstLine, 1);
    fbComposeGetStart (pSrc, xSrc, ySrc, CARD32, srcStride, srcLine, 1);
    fbComposeGetStart (pMask, xMask, yMask, CARD8, maskStride, maskLine, 1);

    mask = *maskLine << 24 | *maskLine << 16 | *maskLine << 8 | *maskLine;
    vmask = load8888 (mask);
    srca = MC(4x00ff);

    while (height--)
    {
	dst = dstLine;
	dstLine += dstStride;
	src = srcLine;
	srcLine += srcStride;
	w = width;

	while (w && (unsigned long)dst & 7)
	{
	    __m64 s = load8888 (*src);
	    __m64 d = load8888 (*dst);

	    *dst = store8888 (over (s, expand_alpha (s), d));

	    w--;
	    dst++;
	    src++;
	}

	while (w >= 2)
	{
	    __m64 vs = *(__m64 *)dst;
	    __m64 vd = *(__m64 *)src;
	    __m64 vsrc0 = expand8888 (vs, 0);
	    __m64 vsrc1 = expand8888 (vs, 1);

	    *(__m64 *)dst = (__m64)pack8888 (
		in_over (vsrc0, expand_alpha (vsrc0), vmask, expand8888 (vd, 0)),
		in_over (vsrc1, expand_alpha (vsrc1), vmask, expand8888 (vd, 1)));

	    w -= 2;
	    dst += 2;
	    src += 2;
	}

	while (w)
	{
	    __m64 s = load8888 (*src);
	    __m64 d = load8888 (*dst);

	    *dst = store8888 (in_over (s, expand_alpha (s), vmask, d));

	    w--;
	    dst++;
	    src++;
	}
    }

    _mm_empty();
}

void
fbCompositeSrc_x888x8x8888mmx (pixman_operator_t	op,
			       PicturePtr pSrc,
			       PicturePtr pMask,
			       PicturePtr pDst,
			       INT16	xSrc,
			       INT16	ySrc,
			       INT16      xMask,
			       INT16      yMask,
			       INT16      xDst,
			       INT16      yDst,
			       CARD16     width,
			       CARD16     height)
{
    CARD32	*dstLine, *dst;
    CARD32	*srcLine, *src;
    CARD8	*maskLine;
    CARD32	mask;
    __m64	vmask;
    FbStride	dstStride, srcStride, maskStride;
    CARD16	w;
    __m64  srca;