pixman-mmx.c

嵌入式图形库

	    {
		__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_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;
    uint32_t	mask;
    __m64	vmask;
    int	dstStride, srcStride;
    uint16_t	w;
    __m64  srca;

    CHECKPOINT();

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

    fbComposeGetSolid (pMask, mask, pDst->bits.format);
    mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
    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 (in_over (s, expand_alpha (s), vmask, d));

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

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

	    *(__m64 *)dst = 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_x888xnx8888mmx (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;
    uint32_t	mask;
    __m64	vmask;
    int	dstStride, srcStride;
    uint16_t	w;
    __m64  srca;

    CHECKPOINT();

    fbComposeGetStart (pDst, xDst, yDst, uint32_t, dstStride, dstLine, 1);
    fbComposeGetStart (pSrc, xSrc, ySrc, uint32_t, srcStride, srcLine, 1);
    fbComposeGetSolid (pMask, mask, pDst->bits.format);

    mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
    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 | 0xff000000);
	    __m64 d = load8888 (*dst);

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

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

	while (w >= 16)
	{
	    __m64 vd0 = *(__m64 *)(dst + 0);
	    __m64 vd1 = *(__m64 *)(dst + 2);
	    __m64 vd2 = *(__m64 *)(dst + 4);
	    __m64 vd3 = *(__m64 *)(dst + 6);
	    __m64 vd4 = *(__m64 *)(dst + 8);
	    __m64 vd5 = *(__m64 *)(dst + 10);
	    __m64 vd6 = *(__m64 *)(dst + 12);
	    __m64 vd7 = *(__m64 *)(dst + 14);

	    __m64 vs0 = *(__m64 *)(src + 0);
	    __m64 vs1 = *(__m64 *)(src + 2);
	    __m64 vs2 = *(__m64 *)(src + 4);
	    __m64 vs3 = *(__m64 *)(src + 6);
	    __m64 vs4 = *(__m64 *)(src + 8);
	    __m64 vs5 = *(__m64 *)(src + 10);
	    __m64 vs6 = *(__m64 *)(src + 12);
	    __m64 vs7 = *(__m64 *)(src + 14);

	    vd0 = pack8888 (
		in_over (expand8888 (vs0, 0), srca, vmask, expand8888 (vd0, 0)),
		in_over (expand8888 (vs0, 1), srca, vmask, expand8888 (vd0, 1)));

	    vd1 = pack8888 (
		in_over (expand8888 (vs1, 0), srca, vmask, expand8888 (vd1, 0)),
		in_over (expand8888 (vs1, 1), srca, vmask, expand8888 (vd1, 1)));

	    vd2 = pack8888 (
		in_over (expand8888 (vs2, 0), srca, vmask, expand8888 (vd2, 0)),
		in_over (expand8888 (vs2, 1), srca, vmask, expand8888 (vd2, 1)));

	    vd3 = pack8888 (
		in_over (expand8888 (vs3, 0), srca, vmask, expand8888 (vd3, 0)),
		in_over (expand8888 (vs3, 1), srca, vmask, expand8888 (vd3, 1)));

	    vd4 = pack8888 (
		in_over (expand8888 (vs4, 0), srca, vmask, expand8888 (vd4, 0)),
		in_over (expand8888 (vs4, 1), srca, vmask, expand8888 (vd4, 1)));

	    vd5 = pack8888 (
		in_over (expand8888 (vs5, 0), srca, vmask, expand8888 (vd5, 0)),
		in_over (expand8888 (vs5, 1), srca, vmask, expand8888 (vd5, 1)));

            vd6 = pack8888 (
		in_over (expand8888 (vs6, 0), srca, vmask, expand8888 (vd6, 0)),
		in_over (expand8888 (vs6, 1), srca, vmask, expand8888 (vd6, 1)));

	    vd7 = pack8888 (
		in_over (expand8888 (vs7, 0), srca, vmask, expand8888 (vd7, 0)),
		in_over (expand8888 (vs7, 1), srca, vmask, expand8888 (vd7, 1)));

	    *(__m64 *)(dst + 0) = vd0;
	    *(__m64 *)(dst + 2) = vd1;
	    *(__m64 *)(dst + 4) = vd2;
	    *(__m64 *)(dst + 6) = vd3;
	    *(__m64 *)(dst + 8) = vd4;
	    *(__m64 *)(dst + 10) = vd5;
	    *(__m64 *)(dst + 12) = vd6;
	    *(__m64 *)(dst + 14) = vd7;

	    w -= 16;
	    dst += 16;
	    src += 16;
	}

	while (w)
	{
	    __m64 s = load8888 (*src | 0xff000000);
	    __m64 d = load8888 (*dst);

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

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

    _mm_empty();
}

void
fbCompositeSrc_8888x8888mmx (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;
    uint32_t    s;
    int	dstStride, srcStride;
    uint8_t     a;
    uint16_t	w;

    CHECKPOINT();

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

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

	while (w--)
	{
	    s = *src++;
	    a = s >> 24;
	    if (a == 0xff)
		*dst = s;
	    else if (a) {
		__m64 ms, sa;
		ms = load8888(s);
		sa = expand_alpha(ms);
		*dst = store8888(over(ms, sa, load8888(*dst)));
	    }
	    dst++;
	}
    }
    _mm_empty();
}

void
fbCompositeSrc_8888x0565mmx (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(vsrc, expand_alpha(vsrc), vdest), vdest, 0);

	    *dst = ULLONG(vdest);

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

	CHECKPOINT();

	while (w >= 4)
	{
	    __m64 vsrc0, vsrc1, vsrc2, vsrc3;
	    __m64 vdest;

	    vsrc0 = load8888(*(src + 0));
	    vsrc1 = load8888(*(src + 1));
	    vsrc2 = load8888(*(src + 2));
	    vsrc3 = load8888(*(src + 3));

	    vdest = *(__m64 *)dst;

	    vdest = pack565(over(vsrc0, expand_alpha(vsrc0), expand565(vdest, 0)), vdest, 0);
	    vdest = pack565(over(vsrc1, expand_alpha(vsrc1), expand565(vdest, 1)), vdest, 1);
	    vdest = pack565(over(vsrc2, expand_alpha(vsrc2), expand565(vdest, 2)), vdest, 2);
	    vdest = pack565(over(vsrc3, expand_alpha(vsrc3), 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(vsrc, expand_alpha(vsrc), vdest), vdest, 0);

	    *dst = ULLONG(vdest);

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

    _mm_empty();
}

void
fbCompositeSolidMask_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)
	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_over(vsrc, vsrca, expand_alpha_rev (M64(m)), load8888(*dst));
		*dst = store8888(vdest);
	    }

	    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_over(vsrc, vsrca, expand_alpha_rev (M64(m0)), expand8888(vdest, 0));
		dest1 = in_over(vsrc, vsrca, expand_alpha_rev (M64(m1)), expand8888(vdest, 1));

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

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

	CHECKPOINT();

	while (w)
	{
	    ullong m = *mask;

	    if (m)
	    {
		__m64 vdest = load8888(*dst);
		vdest = in_over(vsrc, vsrca, expand_alpha_rev (M64(m)), vdest);
		*dst = store8888(vdest);
	    }

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

    _mm_empty();
}

pixman_bool_t
pixman_fill_mmx (uint32_t *bits,
		 int stride,
		 int bpp,
		 int x,
		 int y,
		 int width,
		 int height,
		 uint32_t xor)
{
    ullong	fill;
    __m64	vfill;
    uint32_t	byte_width;
    uint8_t	*byte_line;
#ifdef __GNUC__
    __m64	v1, v2, v3, v4, v5, v6, v7;
#endif

    if (bpp == 16 && (xor >> 16 != (xor & 0xffff)))
	return FALSE;

    if (bpp != 16 && bpp != 32)
	return FALSE;

    if (bpp == 16)
    {
	stride = stride * (int) sizeof (uint32_t) / 2;
	byte_line = (uint8_t *)(((uint16_t *)bits) + stride * y + x);
	byte_width = 2 * width;
	stride *= 2;
    }
    else
    {
	stride = stride * (int) sizeof (uint32_t) / 4;
	byte_line = (uint8_t *)(((uint32_t *)bits) + stride * y + x);
	byte_width = 4 * width;
	stride *= 4;
    }

    fill = ((ullong)xor << 32) | xor;
    vfill = M64(fill);

#ifdef __GNUC__
    __asm__ (
	"movq		%7,	%0\n"
	"movq		%7,	%1\n"
	"movq		%7,	%2\n"
	"movq		%7,	%3\n"
	"movq		%7,	%4\n"
	"movq		%7,	%5\n"
	"movq		%7,	%6\n"
	: "=y" (v1), "=y" (v2), "=y" (v3),
	  "=y" (v4), "=y" (v5), "=y" (v6), "=y" (v7)
	: "y" (vfill));
#endif

    while (height--)
    {
	int w;
	uint8_t *d = byte_line;
	byte_line += stride;
	w = byte_width;

	while (w >= 2 && ((unsigned long)d & 3))
	{
	    *(uint16_t *)d = xor;
	    w -= 2;
	    d += 2;
	}

	while (w >= 4 && ((unsigned long)d & 7))
	{
	    *(uint32_t *)d = xor;

	    w -= 4;
	    d += 4;
	}

	while (w >= 64)
	{
#ifdef __GNUC__
	    __asm__ (
		"movq	%1,	  (%0)\n"
		"movq	%2,	 8(%0)\n"
		"movq	%3,	16(%0)\n"
		"movq	%4,	24(%0)\n"
		"movq	%5,	32(%0)\n"
		"movq	%6,	40(%0)\n"
		"movq	%7,	48(%0)\n"
		"movq	%8,	56(%0)\n"
		:
		: "r" (d),
		  "y" (vfill), "y" (v1), "y" (v2), "y" (v3),
		  "y" (v4), "y" (v5), "y" (v6), "y" (v7)
		: "memory");
#else
	    *(__m64*) (d +  0) = vfill;
	    *(__m64*) (d +  8) = vfill;
	    *(__m64*) (d + 16) = vfill;