pixman-mmx.c

嵌入式图形库

	    *(__m64*) (d + 24) = vfill;
	    *(__m64*) (d + 32) = vfill;
	    *(__m64*) (d + 40) = vfill;
	    *(__m64*) (d + 48) = vfill;
	    *(__m64*) (d + 56) = vfill;
#endif
	    w -= 64;
	    d += 64;
	}

	while (w >= 4)
	{
	    *(uint32_t *)d = xor;

	    w -= 4;
	    d += 4;
	}
	if (w >= 2)
	{
	    *(uint16_t *)d = xor;
	    w -= 2;
	    d += 2;
	}
    }

    _mm_empty();
    return TRUE;
}

void
fbCompositeSolidMaskSrc_nx8x8888mmx (pixman_op_t op,
				     pixman_image_t * pSrc,
				     pixman_image_t * pMask,
				     pixman_image_t * pDst,
				     int16_t      xSrc,
				     int16_t      ySrc,
				     int16_t      xMask,
				     int16_t      yMask,
				     int16_t      xDst,
				     int16_t      yDst,
				     uint16_t     width,
				     uint16_t     height)
{
    uint32_t	src, srca;
    uint32_t	*dstLine, *dst;
    uint8_t	*maskLine, *mask;
    int	dstStride, maskStride;
    uint16_t	w;
    __m64	vsrc, vsrca;
    ullong	srcsrc;

    CHECKPOINT();

    fbComposeGetSolid(pSrc, src, pDst->bits.format);

    srca = src >> 24;
    if (srca == 0)
    {
	pixman_fill_mmx (pDst->bits.bits, pDst->bits.rowstride, PIXMAN_FORMAT_BPP (pDst->bits.format),
			 xDst, yDst, width, height, 0);
	return;
    }

    srcsrc = (ullong)src << 32 | src;

    fbComposeGetStart (pDst, xDst, yDst, uint32_t, dstStride, dstLine, 1);
    fbComposeGetStart (pMask, xMask, yMask, uint8_t, maskStride, maskLine, 1);

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

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

	CHECKPOINT();

	while (w && (unsigned long)dst & 7)
	{
	    ullong m = *mask;

	    if (m)
	    {
		__m64 vdest = in(vsrc, expand_alpha_rev (M64(m)));
		*dst = store8888(vdest);
	    }
	    else
	    {
		*dst = 0;
	    }

	    w--;
	    mask++;
	    dst++;
	}

	CHECKPOINT();

	while (w >= 2)
	{
	    ullong m0, m1;
	    m0 = *mask;
	    m1 = *(mask + 1);

	    if (srca == 0xff && (m0 & m1) == 0xff)
	    {
		*(ullong *)dst = srcsrc;
	    }
	    else if (m0 | m1)
	    {
		__m64 vdest;
		__m64 dest0, dest1;

		vdest = *(__m64 *)dst;

		dest0 = in(vsrc, expand_alpha_rev (M64(m0)));
		dest1 = in(vsrc, expand_alpha_rev (M64(m1)));

		*(__m64 *)dst = pack8888(dest0, dest1);
	    }
	    else
	    {
		*(ullong *)dst = 0;
	    }

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

	CHECKPOINT();

	while (w)
	{
	    ullong m = *mask;

	    if (m)
	    {
		__m64 vdest = load8888(*dst);
		vdest = in(vsrc, expand_alpha_rev (M64(m)));
		*dst = store8888(vdest);
	    }
	    else
	    {
		*dst = 0;
	    }

	    w--;
	    mask++;
	    dst++;
	}
    }

    _mm_empty();
}

void
fbCompositeSolidMask_nx8x0565mmx (pixman_op_t op,
				  pixman_image_t * pSrc,
				  pixman_image_t * pMask,
				  pixman_image_t * pDst,
				  int16_t      xSrc,
				  int16_t      ySrc,
				  int16_t      xMask,
				  int16_t      yMask,
				  int16_t      xDst,
				  int16_t      yDst,
				  uint16_t     width,
				  uint16_t     height)
{
    uint32_t	src, srca;
    uint16_t	*dstLine, *dst;
    uint8_t	*maskLine, *mask;
    int	dstStride, maskStride;
    uint16_t	w;
    __m64	vsrc, vsrca, tmp;
    ullong srcsrcsrcsrc, src16;

    CHECKPOINT();

    fbComposeGetSolid(pSrc, src, pDst->bits.format);

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

    fbComposeGetStart (pDst, xDst, yDst, uint16_t, dstStride, dstLine, 1);
    fbComposeGetStart (pMask, xMask, yMask, uint8_t, maskStride, maskLine, 1);

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

    tmp = pack565(vsrc, _mm_setzero_si64(), 0);
    src16 = ULLONG(tmp);

    srcsrcsrcsrc = (ullong)src16 << 48 | (ullong)src16 << 32 |
	(ullong)src16 << 16 | (ullong)src16;

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

	CHECKPOINT();

	while (w && (unsigned long)dst & 7)
	{
	    ullong m = *mask;

	    if (m)
	    {
		ullong d = *dst;
		__m64 vd = M64(d);
		__m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev (M64 (m)), expand565(vd, 0));
		vd = pack565(vdest, _mm_setzero_si64(), 0);
		*dst = ULLONG(vd);
	    }

	    w--;
	    mask++;
	    dst++;
	}

	CHECKPOINT();

	while (w >= 4)
	{
	    ullong m0, m1, m2, m3;
	    m0 = *mask;
	    m1 = *(mask + 1);
	    m2 = *(mask + 2);
	    m3 = *(mask + 3);

	    if (srca == 0xff && (m0 & m1 & m2 & m3) == 0xff)
	    {
		*(ullong *)dst = srcsrcsrcsrc;
	    }
	    else if (m0 | m1 | m2 | m3)
	    {
		__m64 vdest;
		__m64 vm0, vm1, vm2, vm3;

		vdest = *(__m64 *)dst;

		vm0 = M64(m0);
		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm0), expand565(vdest, 0)), vdest, 0);
		vm1 = M64(m1);
		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm1), expand565(vdest, 1)), vdest, 1);
		vm2 = M64(m2);
		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm2), expand565(vdest, 2)), vdest, 2);
		vm3 = M64(m3);
		vdest = pack565(in_over(vsrc, vsrca, expand_alpha_rev(vm3), expand565(vdest, 3)), vdest, 3);

		*(__m64 *)dst = vdest;
	    }

	    w -= 4;
	    mask += 4;
	    dst += 4;
	}

	CHECKPOINT();

	while (w)
	{
	    ullong m = *mask;

	    if (m)
	    {
		ullong d = *dst;
		__m64 vd = M64(d);
		__m64 vdest = in_over(vsrc, vsrca, expand_alpha_rev (M64(m)), expand565(vd, 0));
		vd = pack565(vdest, _mm_setzero_si64(), 0);
		*dst = ULLONG(vd);
	    }

	    w--;
	    mask++;
	    dst++;
	}
    }

    _mm_empty();
}

void
fbCompositeSrc_8888RevNPx0565mmx (pixman_op_t op,
				  pixman_image_t * pSrc,
				  pixman_image_t * pMask,
				  pixman_image_t * pDst,
				  int16_t      xSrc,
				  int16_t      ySrc,
				  int16_t      xMask,
				  int16_t      yMask,
				  int16_t      xDst,
				  int16_t      yDst,
				  uint16_t     width,
				  uint16_t     height)
{
    uint16_t	*dstLine, *dst;
    uint32_t	*srcLine, *src;
    int	dstStride, srcStride;
    uint16_t	w;

    CHECKPOINT();

    fbComposeGetStart (pDst, xDst, yDst, uint16_t, dstStride, dstLine, 1);
    fbComposeGetStart (pSrc, xSrc, ySrc, uint32_t, srcStride, srcLine, 1);

#if 0
    /* FIXME */
    assert (pSrc->pDrawable == pMask->pDrawable);
#endif

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

	CHECKPOINT();

	while (w && (unsigned long)dst & 7)
	{
	    __m64 vsrc = load8888 (*src);
	    ullong d = *dst;
	    __m64 vdest = expand565 (M64(d), 0);

	    vdest = pack565(over_rev_non_pre(vsrc, vdest), vdest, 0);

	    *dst = ULLONG(vdest);

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

	CHECKPOINT();

	while (w >= 4)
	{
	    uint32_t s0, s1, s2, s3;
	    unsigned char a0, a1, a2, a3;

	    s0 = *src;
	    s1 = *(src + 1);
	    s2 = *(src + 2);
	    s3 = *(src + 3);

	    a0 = (s0 >> 24);
	    a1 = (s1 >> 24);
	    a2 = (s2 >> 24);
	    a3 = (s3 >> 24);

	    if ((a0 & a1 & a2 & a3) == 0xFF)
	    {
		__m64 vdest;
		vdest = pack565(invert_colors(load8888(s0)), _mm_setzero_si64(), 0);
		vdest = pack565(invert_colors(load8888(s1)), vdest, 1);
		vdest = pack565(invert_colors(load8888(s2)), vdest, 2);
		vdest = pack565(invert_colors(load8888(s3)), vdest, 3);

		*(__m64 *)dst = vdest;
	    }
	    else if (a0 | a1 | a2 | a3)
	    {
		__m64 vdest = *(__m64 *)dst;

		vdest = pack565(over_rev_non_pre(load8888(s0), expand565(vdest, 0)), vdest, 0);
	        vdest = pack565(over_rev_non_pre(load8888(s1), expand565(vdest, 1)), vdest, 1);
		vdest = pack565(over_rev_non_pre(load8888(s2), expand565(vdest, 2)), vdest, 2);
		vdest = pack565(over_rev_non_pre(load8888(s3), expand565(vdest, 3)), vdest, 3);

		*(__m64 *)dst = vdest;
	    }

	    w -= 4;
	    dst += 4;
	    src += 4;
	}

	CHECKPOINT();

	while (w)
	{
	    __m64 vsrc = load8888 (*src);
	    ullong d = *dst;
	    __m64 vdest = expand565 (M64(d), 0);

	    vdest = pack565(over_rev_non_pre(vsrc, vdest), vdest, 0);

	    *dst = ULLONG(vdest);

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

    _mm_empty();
}

/* "8888RevNP" is GdkPixbuf's format: ABGR, non premultiplied */

void
fbCompositeSrc_8888RevNPx8888mmx (pixman_op_t op,
				  pixman_image_t * pSrc,
				  pixman_image_t * pMask,
				  pixman_image_t * pDst,
				  int16_t      xSrc,
				  int16_t      ySrc,
				  int16_t      xMask,
				  int16_t      yMask,
				  int16_t      xDst,
				  int16_t      yDst,
				  uint16_t     width,
				  uint16_t     height)
{
    uint32_t	*dstLine, *dst;
    uint32_t	*srcLine, *src;
    int	dstStride, srcStride;
    uint16_t	w;

    CHECKPOINT();

    fbComposeGetStart (pDst, xDst, yDst, uint32_t, dstStride, dstLine, 1);
    fbComposeGetStart (pSrc, xSrc, ySrc, uint32_t, srcStride, srcLine, 1);

#if 0
    /* FIXME */
    assert (pSrc->pDrawable == pMask->pDrawable);
#endif

    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_rev_non_pre (s, d));

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

	while (w >= 2)
	{
	    ullong s0, s1;
	    unsigned char a0, a1;
	    __m64 d0, d1;

	    s0 = *src;
	    s1 = *(src + 1);

	    a0 = (s0 >> 24);
	    a1 = (s1 >> 24);

	    if ((a0 & a1) == 0xFF)
	    {
		d0 = invert_colors(load8888(s0));
		d1 = invert_colors(load8888(s1));

		*(__m64 *)dst = pack8888 (d0, d1);
	    }
	    else if (a0 | a1)
	    {
		__m64 vdest = *(__m64 *)dst;

		d0 = over_rev_non_pre (load8888(s0), expand8888 (vdest, 0));
		d1 = over_rev_non_pre (load8888(s1), expand8888 (vdest, 1));

		*(__m64 *)dst = pack8888 (d0, d1);
	    }

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

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

	    *dst = store8888 (over_rev_non_pre (s, d));

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

    _mm_empty();
}

void
fbCompositeSolidMask_nx8888x0565Cmmx (pixman_op_t op,
				      pixman_image_t * pSrc,
				      pixman_image_t * pMask,
				      pixman_image_t * pDst,
				      int16_t      xSrc,
				      int16_t      ySrc,
				      int16_t      xMask,
				      int16_t      yMask,
				      int16_t      xDst,
				      int16_t      yDst,
				      uint16_t     width,
				      uint16_t     height)
{
    uint32_t	src, srca;
    uint16_t	*dstLine;
    uint32_t	*maskLine;
    int	dstStride, maskStride;
    __m64  vsrc, vsrca;

    CHECKPOINT();

    fbComposeGetSolid(pSrc, src, pDst->bits.format);

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

    fbComposeGetStart (pDst, xDst, yDst, uint16_t, dstStride, dstLine, 1);
    fbComposeGetStart (pMask, xMask, yMask, uint32_t, maskStride, maskLine, 1);

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

    while (height--)
    {
	int twidth = width;
	uint32_t *p = (uint32_t *)maskLine;
	uint16_t *q = (uint16_t *)dstLine;

	while (twidth && ((unsigned long)q & 7))
	{
	    uint32_t m = *(uint32_t *)p;

	    if (m)
	    {
		ullong d = *q;
		__m64 vdest = expand565 (M64(d), 0);
		vdest = pack565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
		*q = ULLONG(vdest);
	    }

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

	while (twidth >= 4)
	{
	    uint32_t m0, m1, m2, m3;

	    m0 = *p;
	    m1 = *(p + 1);
	    m2 = *(p + 2);
	    m3 = *(p + 3);

	    if ((m0 | m1 | m2 | m3))
	    {
		__m64 vdest = *(__m64 *)q;

		vdest = pack565(in_over(vsrc, vsrca, load8888(m0), expand565(vdest, 0)), vdest, 0);
		vdest = pack565(in_over(vsrc, vsrca, load8888(m1), expand565(vdest, 1)), vdest, 1);
		vdest = pack565(in_over(vsrc, vsrca, load8888(m2), expand565(vdest, 2)), vdest, 2);
		vdest = pack565(in_over(vsrc, vsrca, load8888(m3), expand565(vdest, 3)), vdest, 3);

		*(__m64 *)q = vdest;
	    }
	    twidth -= 4;
	    p += 4;
	    q += 4;
	}

	while (twidth)
	{
	    uint32_t m;

	    m = *(uint32_t *)p;
	    if (m)
	    {
		ullong d = *q;
		__m64 vdest = expand565(M64(d), 0);