/******************************Module*Header*******************************\
* Module Name: Brush.c
*
* Handles all brush/pattern initialization and realization.
*
* Copyright (c) 1992-1996 Microsoft Corporation
*
\**************************************************************************/

#include "precomp.h"

/******************************Public*Routine******************************\
* VOID vRealizeDitherPattern
*
* Generates an 8x8 dither pattern, in our internal realization format, for
* the colour ulRGBToDither.  Note that the high byte of ulRGBToDither does
* not need to be set to zero, because vComputeSubspaces ignores it.
\**************************************************************************/

VOID vRealizeDitherPattern(
RBRUSH*     prb,
ULONG       ulRGBToDither)
{
    ULONG           ulNumVertices;
    VERTEX_DATA     vVertexData[4];
    VERTEX_DATA*    pvVertexData;
    LONG            i;

    // Calculate what colour subspaces are involved in the dither:

    pvVertexData = vComputeSubspaces(ulRGBToDither, vVertexData);

    // Now that we have found the bounding vertices and the number of
    // pixels to dither for each vertex, we can create the dither pattern

    ulNumVertices = pvVertexData - vVertexData;
                      // # of vertices with more than zero pixels in the dither

    // Do the actual dithering:

    vDitherColor(&prb->aulPattern[0], vVertexData, pvVertexData, ulNumVertices);

    // Initialize the fields we need:

    prb->ptlBrushOrg.x = LONG_MIN;
    prb->fl            = 0;

    for (i = 0; i < MAX_BOARDS; i++)
    {
        prb->apbe[i] = NULL;
    }
}

/******************************Public*Routine******************************\
* BOOL DrvRealizeBrush
*
* This function allows us to convert GDI brushes into an internal form
* we can use.  It may be called directly by GDI at SelectObject time, or
* it may be called by GDI as a result of us calling BRUSHOBJ_pvGetRbrush
* to create a realized brush in a function like DrvBitBlt.
*
* Note that we have no way of determining what the current Rop or brush
* alignment are at this point.
*
\**************************************************************************/

BOOL DrvRealizeBrush(
BRUSHOBJ*   pbo,
SURFOBJ*    psoDst,
SURFOBJ*    psoPattern,
SURFOBJ*    psoMask,
XLATEOBJ*   pxlo,
ULONG       iHatch)
{
    PDEV*       ppdev;
    ULONG       iPatternFormat;
    BYTE*       pjSrc;
    BYTE*       pjDst;
    LONG        lSrcDelta;
    LONG        cj;
    LONG        i;
    LONG        j;
    RBRUSH*     prb;
    ULONG*      pulXlate;
    SURFOBJ*    psoPunt;
    RECTL       rclDst;
    BOOL        b;

    ppdev = (PDEV*) psoDst->dhpdev;

    // We only handle brushes if we have an off-screen brush cache
    // available.  If there isn't one, we can simply fail the realization,
    // and eventually GDI will do the drawing for us (although a lot
    // slower than we could have done it):

    if (!(ppdev->flStatus & STAT_BRUSH_CACHE))
        goto ReturnFalse;

    // We have a fast path for dithers when we set GCAPS_DITHERONREALIZE:

    if (iHatch & RB_DITHERCOLOR)
    {
        // Implementing DITHERONREALIZE increased our score on a certain
        // unmentionable benchmark by 0.4 million 'megapixels'.  Too bad
        // this didn't work in the first version of NT.

        prb = BRUSHOBJ_pvAllocRbrush(pbo,
               sizeof(RBRUSH) + CONVERT_TO_BYTES(TOTAL_BRUSH_SIZE, ppdev));
        if (prb == NULL)
            goto ReturnFalse;

        vRealizeDitherPattern(prb, iHatch);
        goto ReturnTrue;
    }

    // We only accelerate 8x8 patterns.  Since Win3.1 and Chicago don't
    // support patterns of any other size, it's a safe bet that 99.9%
    // of the patterns we'll ever get will be 8x8:

    if ((psoPattern->sizlBitmap.cx != 8) ||
        (psoPattern->sizlBitmap.cy != 8))
        goto ReturnFalse;

    iPatternFormat = psoPattern->iBitmapFormat;

    prb = BRUSHOBJ_pvAllocRbrush(pbo,
          sizeof(RBRUSH) + CONVERT_TO_BYTES(TOTAL_BRUSH_SIZE, ppdev));
    if (prb == NULL)
        goto ReturnFalse;

    // Initialize the fields we need:

    prb->ptlBrushOrg.x = LONG_MIN;
    prb->fl            = 0;

    for (i = 0; i < MAX_BOARDS; i++)
    {
        prb->apbe[i] = NULL;
    }

    lSrcDelta = psoPattern->lDelta;
    pjSrc     = (BYTE*) psoPattern->pvScan0;
    pjDst     = (BYTE*) &prb->aulPattern[0];

    if ((ppdev->iBitmapFormat == iPatternFormat) &&
        ((pxlo == NULL) || (pxlo->flXlate & XO_TRIVIAL)))
    {
        DISPDBG((1, "Realizing un-translated brush"));

        // The pattern is the same colour depth as the screen, and
        // there's no translation to be done:

       cj = CONVERT_TO_BYTES(8, ppdev);  // Every pattern is 8 pels wide

        for (i = 8; i != 0; i--)
        {
            RtlCopyMemory(pjDst, pjSrc, cj);

            pjSrc += lSrcDelta;
            pjDst += cj;
        }
    }
    // Don't do monochrome expansion on 24 bpp due to s3 968 feature.
    else if ((iPatternFormat == BMF_1BPP) && (ppdev->iBitmapFormat != BMF_24BPP))
    {
        DISPDBG((1, "Realizing 1bpp brush"));

        // We word align the monochrome bitmap so that every row starts
        // on a new word (so that we can do word writes later to transfer
        // the bitmap):

        for (i = 8; i != 0; i--)
        {
            *pjDst = *pjSrc;
            pjDst += sizeof(WORD);
            pjSrc += lSrcDelta;
        }

        pulXlate         = pxlo->pulXlate;
        prb->fl         |= RBRUSH_2COLOR;
        prb->ulForeColor = pulXlate[1];
        prb->ulBackColor = pulXlate[0];
    }
    else if ((iPatternFormat == BMF_4BPP) && (ppdev->iBitmapFormat == BMF_8BPP))
    {
        DISPDBG((1, "Realizing 4bpp brush"));

        // The screen is 8bpp and the pattern is 4bpp:

        ASSERTDD((ppdev->iBitmapFormat == BMF_8BPP) &&
                 (iPatternFormat == BMF_4BPP),
                 "Messed up brush logic");

        pulXlate = pxlo->pulXlate;

        for (i = 8; i != 0; i--)
        {
            // Inner loop is repeated only 4 times because each loop
            // handles 2 pixels:

            for (j = 4; j != 0; j--)
            {
                *pjDst++ = (BYTE) pulXlate[*pjSrc >> 4];
                *pjDst++ = (BYTE) pulXlate[*pjSrc & 15];
                pjSrc++;
            }

            pjSrc += lSrcDelta - 4;
        }
    }
    else
    {
        // We've got a brush whose format we haven't special cased.  No
        // problem, we can have GDI convert it to our device's format.
        // We simply use a temporary surface object that was created with
        // the same format as the display, and point it to our brush
        // realization:

        DISPDBG((5, "Realizing funky brush"));

        psoPunt          = ppdev->psoBank;
        psoPunt->pvScan0 = pjDst;
        psoPunt->lDelta  = CONVERT_TO_BYTES(8, ppdev);

        rclDst.left      = 0;
        rclDst.top       = 0;
        rclDst.right     = 8;
        rclDst.bottom    = 8;

        b = EngCopyBits(psoPunt, psoPattern, NULL, pxlo,
                        &rclDst, (POINTL*) &rclDst);

        if (!b)
        {
            goto ReturnFalse;
        }
    }

ReturnTrue:

    if (!(ppdev->flCaps & CAPS_HW_PATTERNS))
    {
        // The last time I checked, GDI took some 500 odd instructions to
        // get from here back to whereever we called 'BRUSHOBJ_pvGetRbrush'.
        // We can at least use this time to get some overlap between the
        // CPU and the display hardware: we'll initialize the 72x72 off-
        // screen cache entry now, which will keep the accelerator busy for
        // a while.
        //
        // We don't do this if we have hardware patterns because:
        //
        //   a) S3 hardware patterns require that the off-screen cached
        //      brush be correctly aligned, and at this point we don't have
        //      access to the 'pptlBrush' brush origin (although we could
        //      have copied it into the PDEV before calling
        //      BRUSHOBJ_pvGetRbrush).
        //
        //   b) S3 hardware patterns require only an 8x8 copy of the
        //      pattern; it is not expanded to 72x72, so there isn't even
        //      any opportunity for CPU/accelerator processing overlap.

        vIoSlowPatRealize(ppdev, prb, FALSE);
    }

    return(TRUE);

ReturnFalse:

    if (psoPattern != NULL)
    {
        DISPDBG((1, "Failed realization -- Type: %li Format: %li cx: %li cy: %li",
                    psoPattern->iType, psoPattern->iBitmapFormat,
                    psoPattern->sizlBitmap.cx, psoPattern->sizlBitmap.cy));
    }

    return(FALSE);
}

/******************************Public*Routine******************************\
* BOOL bEnableBrushCache
*
* Allocates off-screen memory for storing the brush cache.
\**************************************************************************/

BOOL bEnableBrushCache(
PDEV*   ppdev)
{
    OH*         poh;            // Points to off-screen chunk of memory
    BRUSHENTRY* pbe;            // Pointer to the brush-cache entry
    LONG        i;

    pbe = &ppdev->abe[0];       // Points to where we'll put the first brush
                                //   cache entry

    if (ppdev->flCaps & CAPS_HW_PATTERNS)
    {
        LONG x;
        LONG y;

        // Allocate the width of the screen so that we bias the off-screen
        // heap manager to horizontal allocations:

        ASSERTDD(ppdev->cxMemory >= (FAST_BRUSH_COUNT + 1) * FAST_BRUSH_ALLOCATION + 1,
            "cxMemory allocation won't be big enough");

        poh = pohAllocate(ppdev,
                          NULL,
                          ppdev->cxMemory,
                          FAST_BRUSH_ALLOCATION,
                          FLOH_MAKE_PERMANENT);
        if (poh == NULL)
            goto ReturnTrue;    // See note about why we can return TRUE...

        ppdev->cBrushCache = FAST_BRUSH_COUNT;

        // Hardware brushes require that the x-coordinate start on an 8
        // pixel boundary.  The heap manager doesn't guarantee us any such
        // alignment, so we allocate a bit of extra room so that we can
        // do the alignment ourselves:

        x = (poh->x + 7) & ~7L;
        y = poh->y;

        for (i = FAST_BRUSH_COUNT; i != 0; i--)
        {
            // If we hadn't allocated 'ppdev' so that it was zero initialized,
            // we would have to initialize pbe->prbVerify too...

            pbe->x = x;
            pbe->y = y;

            x += FAST_BRUSH_ALLOCATION;
            pbe++;
        }

        // Remember the location of our 1x8 work area, which will be at
        // the right end of our brush array:

        ppdev->ptlReRealize.x = x;
        ppdev->ptlReRealize.y = y;
    }
    else
    {
        LONG j;

        ppdev->pfnFillPat = vIoFillPatSlow;           // Override FillPatFast

        // Typically, we'll be running at 1024x768x256 on a 1meg board,
        // giving us off-screen memory of the dimension 1024x253 (accounting
        // for the space taken by the hardware pointer).  If we allocate
        // the brush cache as one long one-high row of brushes, the heap
        // manager would shave that amount off the largest chunk of memory
        // we could allocate (meaning the largest bitmap potentially stored
        // in off-screen memory couldn't be larger than 253 - 64 = 189 pels
        // high, but it could be 1024 wide).
        //
        // To make this more square, I want to shave off a left-side chunk
        // for the brush cache, and I want at least 8 brushes cached.
        // Since floor(253/64) = 3, we'll allocate a 3 x 3 cache:

        poh = pohAllocate(ppdev,
                          NULL,
                          SLOW_BRUSH_CACHE_DIM * SLOW_BRUSH_ALLOCATION,
                          SLOW_BRUSH_CACHE_DIM * SLOW_BRUSH_ALLOCATION,
                          FLOH_MAKE_PERMANENT);

        if (poh == NULL)
            goto ReturnTrue;    // See note about why we can return TRUE...

        ppdev->cBrushCache = SLOW_BRUSH_COUNT;

        for (i = 0; i < SLOW_BRUSH_CACHE_DIM; i++)
        {
            for (j = 0; j < SLOW_BRUSH_CACHE_DIM; j++)
            {
                pbe->x = poh->x + (i * SLOW_BRUSH_ALLOCATION);
                pbe->y = poh->y + (j * SLOW_BRUSH_ALLOCATION);
                pbe++;
            }
        }
    }

    // Note that we don't have to remember 'poh' for when we have
    // to disable brushes -- the off-screen heap frees any
    // off-screen heap allocations automatically.

    // We successfully allocated the brush cache, so let's turn
    // on the switch showing that we can use it:

    ppdev->flStatus |= STAT_BRUSH_CACHE;

ReturnTrue:

    // If we couldn't allocate a brush cache, it's not a catastrophic
    // failure; patterns will still work, although they'll be a bit
    // slower since they'll go through GDI.  As a result we don't
    // actually have to fail this call:

    DISPDBG((5, "Passed bEnableBrushCache"));

    return(TRUE);
}

/******************************Public*Routine******************************\
* VOID vDisableBrushCache
*
* Cleans up anything done in bEnableBrushCache.
\**************************************************************************/

VOID vDisableBrushCache(PDEV* ppdev)
{
    // We ain't gotta do nothin'
}

/******************************Public*Routine******************************\
* VOID vAssertModeBrushCache
*
* Resets the brush cache when we exit out of full-screen.
\**************************************************************************/

VOID vAssertModeBrushCache(
PDEV*   ppdev,
BOOL    bEnable)
{
    BRUSHENTRY* pbe;
    LONG        i;

    if (bEnable)
    {
        // Invalidate the brush cache:

        pbe = &ppdev->abe[0];

        for (i = ppdev->cBrushCache; i != 0; i--)
        {
            pbe->prbVerify = NULL;
            pbe++;
        }
    }
}