windows-nt/Source/XPSP1/NT/drivers/video/ms/s3/disp/bitblt.c
2020-09-26 16:20:57 +08:00

1234 lines
42 KiB
C

/******************************Module*Header*******************************\
*
* *******************
* * GDI SAMPLE CODE *
* *******************
*
* Module Name: bitblt.c
*
* Contains the high-level DrvBitBlt and DrvCopyBits functions. The low-
* level stuff lives in the 'blt??.c' files.
*
* Note: The way we've implemented device-bitmaps has changed in NT5, with
* the advent of 'EngModifySurface' and 'DrvDeriveSurface'. Now,
* off-screen bitmaps will always have an iType of STYPE_BITMAP
* (meaning that GDI can draw directly on the bits if it needs to).
* Additionally, former off-screen bitmaps that have been converted
* by us to system-memory DIBs will still have an iType of STYPE_BITMAP.
*
* Copyright (c) 1992-1998 Microsoft Corporation
\**************************************************************************/
#include "precomp.h"
/******************************Public*Routine******************************\
* VOID vXferNativeSrccopy
*
* Does a SRCCOPY transfer of a bitmap to the screen using the frame
* buffer, because with USWC write-combining it's significantly faster
* than using the data transfer register.
*
\**************************************************************************/
VOID vXferNativeSrccopy( // Type FNXFER
PDEV* ppdev,
LONG c, // Count of rectangles, can't be zero
RECTL* prcl, // List of destination rectangles, in relative
// coordinates
ULONG rop4, // Not used
SURFOBJ* psoSrc, // Source surface
POINTL* pptlSrc, // Original unclipped source point
RECTL* prclDst, // Original unclipped destination rectangle
XLATEOBJ* pxlo) // Not used
{
LONG xOffset;
LONG yOffset;
LONG dx;
LONG dy;
RECTL rclDst;
POINTL ptlSrc;
ASSERTDD((pxlo == NULL) || (pxlo->flXlate & XO_TRIVIAL),
"Can handle trivial xlate only");
ASSERTDD(psoSrc->iBitmapFormat == ppdev->iBitmapFormat,
"Source must be same colour depth as screen");
ASSERTDD(c > 0, "Can't handle zero rectangles");
ASSERTDD(rop4 == 0xcccc, "Must be a SRCCOPY rop");
xOffset = ppdev->xOffset;
yOffset = ppdev->yOffset;
dx = pptlSrc->x - prclDst->left;
dy = pptlSrc->y - prclDst->top; // Add to destination to get source
while (TRUE)
{
ptlSrc.x = prcl->left + dx;
ptlSrc.y = prcl->top + dy;
// 'vPutBits' takes only absolute coordinates, so add in the
// off-screen bitmap offset here:
rclDst.left = prcl->left + xOffset;
rclDst.right = prcl->right + xOffset;
rclDst.top = prcl->top + yOffset;
rclDst.bottom = prcl->bottom + yOffset;
vPutBits(ppdev, psoSrc, &rclDst, &ptlSrc);
if (--c == 0)
return;
prcl++;
}
}
/******************************Public*Routine******************************\
* VOID vReadNativeSrccopy
*
* Does a SRCCOPY read from the screen to a system-memory bitmap. The only
* reason we do this here instead of punting to GDI is to ensure that we
* do dword reads that are aligned to the video-memory source and not the
* system-memory destination.
*
\**************************************************************************/
VOID vReadNativeSrccopy( // Type FNXFER
PDEV* ppdev,
LONG c, // Count of rectangles, can't be zero
RECTL* prcl, // List of destination rectangles, in relative
// coordinates
ULONG rop4, // Not used
SURFOBJ* psoDst, // Destination surface
POINTL* pptlSrc, // Original unclipped source point
RECTL* prclDst, // Original unclipped destination rectangle
XLATEOBJ* pxlo) // Not used
{
LONG xOffset;
LONG yOffset;
LONG dx;
LONG dy;
RECTL rclDst;
POINTL ptlSrc;
ASSERTDD((pxlo == NULL) || (pxlo->flXlate & XO_TRIVIAL),
"Can handle trivial xlate only");
ASSERTDD(psoDst->iBitmapFormat == ppdev->iBitmapFormat,
"Source must be same colour depth as screen");
ASSERTDD(c > 0, "Can't handle zero rectangles");
ASSERTDD(rop4 == 0xcccc, "Must be a SRCCOPY rop");
xOffset = ppdev->xOffset;
yOffset = ppdev->yOffset;
dx = pptlSrc->x - prclDst->left;
dy = pptlSrc->y - prclDst->top; // Add to destination to get source
while (TRUE)
{
// 'vGetBits' takes only absolute coordinates, so add in the
// off-screen bitmap offset here:
ptlSrc.x = prcl->left + dx + xOffset;
ptlSrc.y = prcl->top + dy + yOffset;
vGetBits(ppdev, psoDst, prcl, &ptlSrc);
if (--c == 0)
return;
prcl++;
}
}
/******************************Public*Routine******************************\
* BOOL bPuntBlt
*
* Has GDI do any drawing operations that we don't specifically handle
* in the driver.
*
\**************************************************************************/
BOOL bPuntBlt(
SURFOBJ* psoDst,
SURFOBJ* psoSrc,
SURFOBJ* psoMsk,
CLIPOBJ* pco,
XLATEOBJ* pxlo,
RECTL* prclDst,
POINTL* pptlSrc,
POINTL* pptlMsk,
BRUSHOBJ* pbo,
POINTL* pptlBrush,
ROP4 rop4)
{
PDEV* ppdev;
if (psoDst->dhsurf != NULL)
ppdev = (PDEV*) psoDst->dhpdev;
else
ppdev = (PDEV*) psoSrc->dhpdev;
#if DBG
{
//////////////////////////////////////////////////////////////////////
// Diagnostics
//
// Since calling the engine to do any drawing can be rather painful,
// particularly when the source is an off-screen DFB (since GDI will
// have to allocate a DIB and call us to make a temporary copy before
// it can even start drawing), we'll try to avoid it as much as
// possible.
//
// Here we simply spew out information describing the blt whenever
// this routine gets called (checked builds only, of course):
ULONG ulClip;
PDEV* ppdev;
if (psoDst->dhsurf != NULL)
ppdev = (PDEV*) psoDst->dhpdev;
else
ppdev = (PDEV*) psoSrc->dhpdev;
ulClip = (pco == NULL) ? DC_TRIVIAL : pco->iDComplexity;
DISPDBG((2, ">> Punt << Dst format: %li Dst type: %li Clip: %li Rop: %lx",
psoDst->iBitmapFormat, psoDst->iType, ulClip, rop4));
if (psoSrc != NULL)
{
DISPDBG((2, " << Src format: %li Src type: %li",
psoSrc->iBitmapFormat, psoSrc->iType));
if (psoSrc->iBitmapFormat == BMF_1BPP)
{
DISPDBG((2, " << Foreground: %lx Background: %lx",
pxlo->pulXlate[1], pxlo->pulXlate[0]));
}
}
if ((pxlo != NULL) && !(pxlo->flXlate & XO_TRIVIAL) && (psoSrc != NULL))
{
if (((psoSrc->dhsurf == NULL) &&
(psoSrc->iBitmapFormat != ppdev->iBitmapFormat)) ||
((psoDst->dhsurf == NULL) &&
(psoDst->iBitmapFormat != ppdev->iBitmapFormat)))
{
// Don't bother printing the 'xlate' message when the source
// is a different bitmap format from the destination -- in
// those cases we know there always has to be a translate.
}
else
{
DISPDBG((2, " << With xlate"));
}
}
// If the rop4 requires a pattern, and it's a non-solid brush...
if (((((rop4 >> 4) ^ (rop4)) & 0x0f0f) != 0) &&
(pbo->iSolidColor == -1))
{
if (pbo->pvRbrush == NULL)
DISPDBG((2, " << With brush -- Not created"));
else
DISPDBG((2, " << With brush -- Created Ok"));
}
}
#endif
if (DIRECT_ACCESS(ppdev))
{
//////////////////////////////////////////////////////////////////////
// Banked Framebuffer bPuntBlt
//
// This section of code handles a PuntBlt when GDI can directly draw
// on the framebuffer, but the drawing has to be done in banks:
BANK bnk;
BOOL b;
HSURF hsurfTmp;
SURFOBJ* psoTmp;
SIZEL sizl;
POINTL ptlSrc;
RECTL rclTmp;
RECTL rclDst;
DSURF* pdsurfDst;
DSURF* pdsurfSrc;
// We copy the original destination rectangle, and use that in every
// GDI call-back instead of the original because sometimes GDI is
// sneaky and points 'prclDst' to '&pco->rclBounds'. Because we
// modify 'rclBounds', that would affect 'prclDst', which we don't
// want to happen:
rclDst = *prclDst;
pdsurfDst = (DSURF*) psoDst->dhsurf;
pdsurfSrc = (psoSrc == NULL) ? NULL : (DSURF*) psoSrc->dhsurf;
if ((pdsurfSrc == NULL) || (pdsurfSrc->dt & DT_DIB))
{
// Do a memory-to-screen blt:
vBankStart(ppdev, &rclDst, pco, &bnk);
b = TRUE;
do {
b &= EngBitBlt(bnk.pso, psoSrc, psoMsk, bnk.pco, pxlo,
&rclDst, pptlSrc, pptlMsk, pbo, pptlBrush,
rop4);
} while (bBankEnum(&bnk));
}
else
{
b = FALSE; // Assume failure
// The screen is the source (it may be the destination too...)
ptlSrc.x = pptlSrc->x + ppdev->xOffset;
ptlSrc.y = pptlSrc->y + ppdev->yOffset;
if ((pco != NULL) && (pco->iDComplexity != DC_TRIVIAL))
{
// We have to intersect the destination rectangle with
// the clip bounds if there is one (consider the case
// where the app asked to blt a really, really big
// rectangle from the screen -- prclDst would be really,
// really big but pco->rclBounds would be the actual
// area of interest):
rclDst.left = max(rclDst.left, pco->rclBounds.left);
rclDst.top = max(rclDst.top, pco->rclBounds.top);
rclDst.right = min(rclDst.right, pco->rclBounds.right);
rclDst.bottom = min(rclDst.bottom, pco->rclBounds.bottom);
// Correspondingly, we have to offset the source point:
ptlSrc.x += (rclDst.left - prclDst->left);
ptlSrc.y += (rclDst.top - prclDst->top);
}
// We're now either going to do a screen-to-screen or screen-to-DIB
// blt. In either case, we're going to create a temporary copy of
// the source. (Why do we do this when GDI could do it for us?
// GDI would create a temporary copy of the DIB for every bank
// call-back!)
sizl.cx = rclDst.right - rclDst.left;
sizl.cy = rclDst.bottom - rclDst.top;
// Don't forget to convert from relative to absolute coordinates
// on the source! (vBankStart takes care of that for the
// destination.)
rclTmp.right = sizl.cx;
rclTmp.bottom = sizl.cy;
rclTmp.left = 0;
rclTmp.top = 0;
// GDI does guarantee us that the blt extents have already been
// clipped to the surface boundaries (we don't have to worry
// here about trying to read where there isn't video memory).
// Let's just assert to make sure:
ASSERTDD((ptlSrc.x >= 0) &&
(ptlSrc.y >= 0) &&
(ptlSrc.x + sizl.cx <= ppdev->cxMemory) &&
(ptlSrc.y + sizl.cy <= ppdev->cyMemory),
"Source rectangle out of bounds!");
hsurfTmp = (HSURF) EngCreateBitmap(sizl,
0, // Let GDI choose ulWidth
ppdev->iBitmapFormat,
0, // Don't need any options
NULL);// Let GDI allocate
if (hsurfTmp != 0)
{
psoTmp = EngLockSurface(hsurfTmp);
if (psoTmp != NULL)
{
vGetBits(ppdev, psoTmp, &rclTmp, &ptlSrc);
if ((pdsurfDst == NULL) || (pdsurfDst->dt & DT_DIB))
{
// It was a Screen-to-DIB blt; now it's a DIB-to-DIB
// blt. Note that the source point is (0, 0) in our
// temporary surface:
b = EngBitBlt(psoDst, psoTmp, psoMsk, pco, pxlo,
&rclDst, (POINTL*) &rclTmp, pptlMsk,
pbo, pptlBrush, rop4);
}
else
{
// It was a Screen-to-Screen blt; now it's a DIB-to-
// screen blt. Note that the source point is (0, 0)
// in our temporary surface:
vBankStart(ppdev, &rclDst, pco, &bnk);
b = TRUE;
do {
b &= EngBitBlt(bnk.pso, psoTmp, psoMsk, bnk.pco,
pxlo, &rclDst, (POINTL*) &rclTmp,
pptlMsk, pbo, pptlBrush, rop4);
} while (bBankEnum(&bnk));
}
EngUnlockSurface(psoTmp);
}
EngDeleteSurface(hsurfTmp);
}
}
return(b);
}
#if !defined(_X86_)
else
{
//////////////////////////////////////////////////////////////////////
// Really Slow bPuntBlt
//
// Here we handle a PuntBlt when GDI can't draw directly on the
// framebuffer (as on the Alpha, which can't do it because of its
// 32 bit bus). If you thought the banked version was slow, just
// look at this one. Guaranteed, there will be at least one bitmap
// allocation and extra copy involved; there could be two if it's a
// screen-to-screen operation.
POINTL ptlSrc;
RECTL rclDst;
SIZEL sizl;
BOOL bSrcIsScreen;
HSURF hsurfSrc;
RECTL rclTmp;
BOOL b;
LONG lDelta;
BYTE* pjBits;
BYTE* pjScan0;
HSURF hsurfDst;
RECTL rclScreen;
b = FALSE; // For error cases, assume we'll fail
rclDst = *prclDst;
if (pptlSrc != NULL)
ptlSrc = *pptlSrc;
if ((pco != NULL) && (pco->iDComplexity != DC_TRIVIAL))
{
// We have to intersect the destination rectangle with
// the clip bounds if there is one (consider the case
// where the app asked to blt a really, really big
// rectangle from the screen -- prclDst would be really,
// really big but pco->rclBounds would be the actual
// area of interest):
rclDst.left = max(rclDst.left, pco->rclBounds.left);
rclDst.top = max(rclDst.top, pco->rclBounds.top);
rclDst.right = min(rclDst.right, pco->rclBounds.right);
rclDst.bottom = min(rclDst.bottom, pco->rclBounds.bottom);
ptlSrc.x += (rclDst.left - prclDst->left);
ptlSrc.y += (rclDst.top - prclDst->top);
}
sizl.cx = rclDst.right - rclDst.left;
sizl.cy = rclDst.bottom - rclDst.top;
// We only need to make a copy from the screen if the source is
// the screen, and the source is involved in the rop. Note that
// we have to check the rop before dereferencing 'psoSrc'
// (because 'psoSrc' may be NULL if the source isn't involved):
bSrcIsScreen = (((((rop4 >> 2) ^ (rop4)) & 0x3333) != 0) &&
(psoSrc->dhsurf != NULL));
if (bSrcIsScreen)
{
// We need to create a copy of the source rectangle:
hsurfSrc = (HSURF) EngCreateBitmap(sizl, 0, ppdev->iBitmapFormat,
0, NULL);
if (hsurfSrc == 0)
goto Error_0;
psoSrc = EngLockSurface(hsurfSrc);
if (psoSrc == NULL)
goto Error_1;
rclTmp.left = 0;
rclTmp.top = 0;
rclTmp.right = sizl.cx;
rclTmp.bottom = sizl.cy;
// vGetBits takes absolute coordinates for the source point:
ptlSrc.x += ppdev->xOffset;
ptlSrc.y += ppdev->yOffset;
vGetBits(ppdev, psoSrc, &rclTmp, &ptlSrc);
// The source will now come from (0, 0) of our temporary source
// surface:
ptlSrc.x = 0;
ptlSrc.y = 0;
}
if (psoDst->dhsurf == NULL)
{
b = EngBitBlt(psoDst, psoSrc, psoMsk, pco, pxlo, &rclDst, &ptlSrc,
pptlMsk, pbo, pptlBrush, rop4);
}
else
{
// We need to create a temporary work buffer. We have to do
// some fudging with the offsets so that the upper-left corner
// of the (relative coordinates) clip object bounds passed to
// GDI will be transformed to the upper-left corner of our
// temporary bitmap.
// The alignment doesn't have to be as tight as this at 16bpp
// and 32bpp, but it won't hurt:
lDelta = CONVERT_TO_BYTES((((rclDst.right + 3) & ~3L) -
(rclDst.left & ~3L)),
ppdev);
// We're actually only allocating a bitmap that is 'sizl.cx' x
// 'sizl.cy' in size:
pjBits = EngAllocMem(0, lDelta * sizl.cy, ALLOC_TAG);
if (pjBits == NULL)
goto Error_2;
// We now adjust the surface's 'pvScan0' so that when GDI thinks
// it's writing to pixel (rclDst.top, rclDst.left), it will
// actually be writing to the upper-left pixel of our temporary
// bitmap:
pjScan0 = pjBits - (rclDst.top * lDelta)
- CONVERT_TO_BYTES((rclDst.left & ~3L), ppdev);
ASSERTDD((((ULONG_PTR) pjScan0) & 3) == 0,
"pvScan0 must be dword aligned!");
// The checked build of GDI sometimes checks on blts that
// prclDst->right <= pso->sizl.cx, so we lie to it about
// the size of our bitmap:
sizl.cx = rclDst.right;
sizl.cy = rclDst.bottom;
hsurfDst = (HSURF) EngCreateBitmap(
sizl, // Bitmap covers rectangle
lDelta, // Use this delta
ppdev->iBitmapFormat, // Same colour depth
BMF_TOPDOWN, // Must have a positive delta
pjScan0); // Where (0, 0) would be
if ((hsurfDst == 0) ||
(!EngAssociateSurface(hsurfDst, ppdev->hdevEng, 0)))
goto Error_3;
psoDst = EngLockSurface(hsurfDst);
if (psoDst == NULL)
goto Error_4;
// Make sure that the rectangle we Get/Put from/to the screen
// is in absolute coordinates:
rclScreen.left = rclDst.left + ppdev->xOffset;
rclScreen.right = rclDst.right + ppdev->xOffset;
rclScreen.top = rclDst.top + ppdev->yOffset;
rclScreen.bottom = rclDst.bottom + ppdev->yOffset;
// It would be nice to get a copy of the destination rectangle
// only when the ROP involves the destination (or when the source
// is an RLE), but we can't do that. If the brush is truly NULL,
// GDI will immediately return TRUE from EngBitBlt, without
// modifying the temporary bitmap -- and we would proceed to
// copy the uninitialized temporary bitmap back to the screen.
vGetBits(ppdev, psoDst, &rclDst, (POINTL*) &rclScreen);
b = EngBitBlt(psoDst, psoSrc, psoMsk, pco, pxlo, &rclDst, &ptlSrc,
pptlMsk, pbo, pptlBrush, rop4);
vPutBits(ppdev, psoDst, &rclScreen, (POINTL*) &rclDst);
EngUnlockSurface(psoDst);
Error_4:
EngDeleteSurface(hsurfDst);
Error_3:
EngFreeMem(pjBits);
}
Error_2:
if (bSrcIsScreen)
{
EngUnlockSurface(psoSrc);
Error_1:
EngDeleteSurface(hsurfSrc);
}
Error_0:
return(b);
}
#endif
}
/******************************Public*Routine******************************\
* BOOL DrvBitBlt
*
* Implements the workhorse routine of a display driver.
*
\**************************************************************************/
BOOL DrvBitBlt(
SURFOBJ* psoDst,
SURFOBJ* psoSrc,
SURFOBJ* psoMsk,
CLIPOBJ* pco,
XLATEOBJ* pxlo,
RECTL* prclDst,
POINTL* pptlSrc,
POINTL* pptlMsk,
BRUSHOBJ* pbo,
POINTL* pptlBrush,
ROP4 rop4)
{
PDEV* ppdev;
DSURF* pdsurfDst;
DSURF* pdsurfSrc;
POINTL ptlSrc;
BOOL bMore;
CLIPENUM ce;
LONG c;
RECTL rcl;
BYTE rop3;
FNFILL* pfnFill;
RBRUSH_COLOR rbc; // Realized brush or solid colour
FNXFER* pfnXfer;
ULONG iSrcBitmapFormat;
ULONG iDir;
BOOL bRet;
bRet = TRUE; // Assume success
pdsurfDst = (DSURF*) psoDst->dhsurf; // May be NULL
if (psoSrc == NULL)
{
pdsurfSrc = NULL;
if (!(pdsurfDst->dt & DT_DIB))
{
///////////////////////////////////////////////////////////////////
// Fills
///////////////////////////////////////////////////////////////////
// Fills are this function's "raison d'etre", so we handle them
// as quickly as possible:
ppdev = (PDEV*) psoDst->dhpdev;
ppdev->xOffset = pdsurfDst->x;
ppdev->yOffset = pdsurfDst->y;
// Make sure it doesn't involve a mask (i.e., it's really a
// Rop3):
rop3 = (BYTE) rop4;
if ((BYTE) (rop4 >> 8) == rop3)
{
// Since 'psoSrc' is NULL, the rop3 had better not indicate
// that we need a source.
ASSERTDD((((rop4 >> 2) ^ (rop4)) & 0x33) == 0,
"Need source but GDI gave us a NULL 'psoSrc'");
Fill_It:
pfnFill = ppdev->pfnFillSolid; // Default to solid fill
if ((((rop3 >> 4) ^ (rop3)) & 0xf) != 0)
{
// The rop says that a pattern is truly required
// (blackness, for instance, doesn't need one):
rbc.iSolidColor = pbo->iSolidColor;
if (rbc.iSolidColor == -1)
{
// Try and realize the pattern brush; by doing
// this call-back, GDI will eventually call us
// again through DrvRealizeBrush:
rbc.prb = pbo->pvRbrush;
if (rbc.prb == NULL)
{
rbc.prb = BRUSHOBJ_pvGetRbrush(pbo);
if (rbc.prb == NULL)
{
// If we couldn't realize the brush, punt
// the call (it may have been a non 8x8
// brush or something, which we can't be
// bothered to handle, so let GDI do the
// drawing):
goto Punt_It;
}
}
if ((ppdev->iBitmapFormat == BMF_24BPP) && ((BYTE) (rop4 >> 8) != rop3)) {
goto Punt_It;
}
pfnFill = ppdev->pfnFillPat;
}
}
// Note that these 2 'if's are more efficient than
// a switch statement:
if ((pco == NULL) || (pco->iDComplexity == DC_TRIVIAL))
{
pfnFill(ppdev, 1, prclDst, rop4, rbc, pptlBrush);
goto All_Done;
}
else if (pco->iDComplexity == DC_RECT)
{
if (bIntersect(prclDst, &pco->rclBounds, &rcl))
pfnFill(ppdev, 1, &rcl, rop4, rbc, pptlBrush);
goto All_Done;
}
else
{
CLIPOBJ_cEnumStart(pco, FALSE, CT_RECTANGLES, CD_ANY, 0);
do {
bMore = CLIPOBJ_bEnum(pco, sizeof(ce), (ULONG*) &ce);
c = cIntersect(prclDst, ce.arcl, ce.c);
if (c != 0)
pfnFill(ppdev, c, ce.arcl, rop4, rbc, pptlBrush);
} while (bMore);
goto All_Done;
}
}
}
else
{
// Thanks to EngModifySurface, the destination is really a
// plane old DIB, so we can forget about our DSURF structure
// (this will simplify checks later in this routine):
pdsurfDst = NULL;
}
}
else
{
pdsurfDst = (DSURF*) psoDst->dhsurf;
if ((pdsurfDst != NULL) && (pdsurfDst->dt & DT_DIB))
{
// The destination is really a plane old DIB.
pdsurfDst = NULL;
}
pdsurfSrc = (DSURF*) psoSrc->dhsurf;
if ((pdsurfSrc != NULL) && (pdsurfSrc->dt & DT_DIB))
{
// Here we consider putting a DIB DFB back into off-screen
// memory. If there's a translate, it's probably not worth
// moving since we won't be able to use the hardware to do
// the blt (a similar argument could be made for weird rops
// and stuff that we'll only end up having GDI simulate, but
// those should happen infrequently enough that I don't care).
//
// This is only worth doing if the destination is in off-
// screen memory, though!
if ((pdsurfDst != NULL) &&
((pxlo == NULL) || (pxlo->flXlate & XO_TRIVIAL)))
{
ppdev = pdsurfSrc->ppdev;
// See 'DrvCopyBits' for some more comments on how this
// moving-it-back-into-off-screen-memory thing works:
if (pdsurfSrc->iUniq == ppdev->iHeapUniq)
{
if (--pdsurfSrc->cBlt == 0)
{
if (bMoveDibToOffscreenDfbIfRoom(ppdev, pdsurfSrc))
goto Continue_It;
}
}
else
{
// Some space was freed up in off-screen memory,
// so reset the counter for this DFB:
pdsurfSrc->iUniq = ppdev->iHeapUniq;
pdsurfSrc->cBlt = HEAP_COUNT_DOWN;
}
}
// The source is really a plane old DIB.
pdsurfSrc = NULL;
}
}
Continue_It:
ASSERTDD((pdsurfSrc == NULL) || !(pdsurfSrc->dt & DT_DIB),
"pdsurfSrc should be non-NULL only if in off-screen memory");
ASSERTDD((pdsurfDst == NULL) || !(pdsurfDst->dt & DT_DIB),
"pdsurfDst should be non-NULL only if in off-screen memory");
if (pdsurfDst != NULL)
{
// The destination is in video memory.
if (pdsurfSrc != NULL)
{
// The source is also in video memory. This is effectively
// a screen-to-screen blt, so adjust the source point:
ptlSrc.x = pptlSrc->x - (pdsurfDst->x - pdsurfSrc->x);
ptlSrc.y = pptlSrc->y - (pdsurfDst->y - pdsurfSrc->y);
pptlSrc = &ptlSrc;
}
ppdev = pdsurfDst->ppdev;
ppdev->xOffset = pdsurfDst->x;
ppdev->yOffset = pdsurfDst->y;
}
else
{
// The destination is a DIB.
if (pdsurfSrc == NULL)
{
// The source is a DIB, too. Let GDI handle it.
goto EngBitBlt_It;
}
ppdev = pdsurfSrc->ppdev;
ppdev->xOffset = pdsurfSrc->x;
ppdev->yOffset = pdsurfSrc->y;
}
if (((rop4 >> 8) & 0xff) == (rop4 & 0xff))
{
// Since we've already handled the cases where the ROP4 is really
// a ROP3 and no source is required, we can assert...
ASSERTDD((psoSrc != NULL) && (pptlSrc != NULL),
"Expected no-source case to already have been handled");
///////////////////////////////////////////////////////////////////
// Bitmap transfers
///////////////////////////////////////////////////////////////////
// Since the foreground and background ROPs are the same, we
// don't have to worry about no stinking masks (it's a simple
// Rop3).
rop3 = (BYTE) rop4; // Make it into a Rop3 (we keep the rop4
// around in case we decide to punt)
if (pdsurfDst != NULL)
{
// The destination is the screen. See if the ROP3 requires a
// pattern:
if ((rop3 >> 4) == (rop3 & 0xf))
{
// Nope, the ROP3 doesn't require a pattern.
if (pdsurfSrc == NULL)
{
//////////////////////////////////////////////////
// DIB-to-screen blt
iSrcBitmapFormat = psoSrc->iBitmapFormat;
if (iSrcBitmapFormat == BMF_1BPP)
{
pfnXfer = ppdev->pfnXfer1bpp;
goto Xfer_It;
}
else if ((iSrcBitmapFormat == ppdev->iBitmapFormat) &&
((pxlo == NULL) || (pxlo->flXlate & XO_TRIVIAL)))
{
if ((rop3 & 0xf) != 0xc)
{
pfnXfer = ppdev->pfnXferNative;
}
else
{
// Thanks to USWC write-combining, for SRCCOPY
// blts it will be much stupendously faster to copy
// directly to the frame buffer than to use the
// transfer register. Note that this is true for
// almost any video adapter (including yours).
pfnXfer = vXferNativeSrccopy;
}
goto Xfer_It;
}
// Expansions from 4bpp are pretty frequent with a ROP, and
// should really be done for all color depths, not just 4bpp.
//
// Note, though, that USWC means it's faster to punt to GDI
// for all SRCCOPY cases.
else if ((iSrcBitmapFormat == BMF_4BPP) &&
(ppdev->iBitmapFormat == BMF_8BPP) &&
(rop4 != 0xcccc))
{
pfnXfer = ppdev->pfnXfer4bpp;
goto Xfer_It;
}
}
else // pdsurfSrc != NULL
{
if ((pxlo == NULL) || (pxlo->flXlate & XO_TRIVIAL))
{
//////////////////////////////////////////////////
// Screen-to-screen blt with no translate
if ((pco == NULL) || (pco->iDComplexity == DC_TRIVIAL))
{
(ppdev->pfnCopyBlt)(ppdev, 1, prclDst, rop4,
pptlSrc, prclDst);
goto All_Done;
}
else if (pco->iDComplexity == DC_RECT)
{
if (bIntersect(prclDst, &pco->rclBounds, &rcl))
{
(ppdev->pfnCopyBlt)(ppdev, 1, &rcl, rop4,
pptlSrc, prclDst);
}
goto All_Done;
}
else
{
// Don't forget that we'll have to draw the
// rectangles in the correct direction:
if (pptlSrc->y >= prclDst->top)
{
if (pptlSrc->x >= prclDst->left)
iDir = CD_RIGHTDOWN;
else
iDir = CD_LEFTDOWN;
}
else
{
if (pptlSrc->x >= prclDst->left)
iDir = CD_RIGHTUP;
else
iDir = CD_LEFTUP;
}
CLIPOBJ_cEnumStart(pco, FALSE, CT_RECTANGLES,
iDir, 0);
do {
bMore = CLIPOBJ_bEnum(pco, sizeof(ce),
(ULONG*) &ce);
c = cIntersect(prclDst, ce.arcl, ce.c);
if (c != 0)
{
(ppdev->pfnCopyBlt)(ppdev, c, ce.arcl,
rop4, pptlSrc, prclDst);
}
} while (bMore);
goto All_Done;
}
}
}
}
}
else if (((pxlo == NULL) || (pxlo->flXlate & XO_TRIVIAL)) &&
(rop4 == 0xcccc) &&
(psoDst->iBitmapFormat == ppdev->iBitmapFormat))
{
//////////////////////////////////////////////////
// Screen-to-DIB SRCCOPY blt with no translate
//
// The only way to read bits from video memory on the S3 is to
// have the CPU read directly from the frame-buffer. Unfortunately,
// reads from video memory are pathetically slow.
//
// Have you ever benchmarked reads? On a Pentium II with USWC
// enabled, consecutive writes to the frame buffer via PCI on a
// typical video card is typically on the order of 80 to 100 MB/s.
// Dword reads max out at 6 MB/s! Byte reads (or worse, unaligned
// Dword reads) max out at a very small 1.5 MB/s!
//
// The problem is that if we just punt to GDI, GDI doesn't realize
// that the source is video-memory and the destination is system-
// memory. It will proceed to align its copy to the destination,
// which means that it may do misaligned dword reads from video
// memory. So we just dropped our throughput by a factor of 4!
//
// So the net-result is that we special-case reads here simply so
// that we can do aligned dword reads from video memory.
pfnXfer = vReadNativeSrccopy;
// The Xfer_It routine expects the system-memory surface to come
// in as 'psoSrc'.
psoSrc = psoDst;
// It might also be a thought to convert an off-screen DFB to a
// DIB at this point.
goto Xfer_It;
}
}
else if ((psoMsk == NULL) && (rop4 == 0xaaf0))
{
// The only time GDI will ask us to do a true rop4 using the brush
// mask is when the brush is 1bpp, and the background rop is AA
// (meaning it's a NOP):
rop3 = (BYTE) rop4;
goto Fill_It;
}
// Just fall through to Punt_It...
Punt_It:
bRet = bPuntBlt(psoDst,
psoSrc,
psoMsk,
pco,
pxlo,
prclDst,
pptlSrc,
pptlMsk,
pbo,
pptlBrush,
rop4);
goto All_Done;
//////////////////////////////////////////////////////////////////////
// Common bitmap transfer
Xfer_It:
if ((pco == NULL) || (pco->iDComplexity == DC_TRIVIAL))
{
pfnXfer(ppdev, 1, prclDst, rop4, psoSrc, pptlSrc, prclDst, pxlo);
goto All_Done;
}
else if (pco->iDComplexity == DC_RECT)
{
if (bIntersect(prclDst, &pco->rclBounds, &rcl))
pfnXfer(ppdev, 1, &rcl, rop4, psoSrc, pptlSrc, prclDst, pxlo);
goto All_Done;
}
else
{
CLIPOBJ_cEnumStart(pco, FALSE, CT_RECTANGLES,
CD_ANY, 0);
do {
bMore = CLIPOBJ_bEnum(pco, sizeof(ce),
(ULONG*) &ce);
c = cIntersect(prclDst, ce.arcl, ce.c);
if (c != 0)
{
pfnXfer(ppdev, c, ce.arcl, rop4, psoSrc,
pptlSrc, prclDst, pxlo);
}
} while (bMore);
goto All_Done;
}
////////////////////////////////////////////////////////////////////////
// Common DIB blt
EngBitBlt_It:
// Our driver doesn't handle any blt's between two DIBs. Normally
// a driver doesn't have to worry about this, but we do because
// we have DFBs that may get moved from off-screen memory to a DIB,
// where we have GDI do all the drawing. GDI does DIB drawing at
// a reasonable speed (unless one of the surfaces is a device-
// managed surface...)
//
// If either the source or destination surface in an EngBitBlt
// call-back is a device-managed surface (meaning it's not a DIB
// that GDI can draw with), GDI will automatically allocate memory
// and call the driver's DrvCopyBits routine to create a DIB copy
// that it can use. So this means that this could handle all 'punts',
// and we could conceivably get rid of bPuntBlt. But this would have
// a bad performance impact because of the extra memory allocations
// and bitmap copies -- you really don't want to do this unless you
// have to (or your surface was created such that GDI can draw
// directly onto it) -- I've been burned by this because it's not
// obvious that the performance impact is so bad.
//
// That being said, we only call EngBitBlt when all the surfaces
// are DIBs:
bRet = EngBitBlt(psoDst, psoSrc, psoMsk, pco, pxlo, prclDst,
pptlSrc, pptlMsk, pbo, pptlBrush, rop4);
All_Done:
return(bRet);
}
/******************************Public*Routine******************************\
* BOOL DrvCopyBits
*
* Do fast bitmap copies.
*
* DrvCopyBits is just a special-case of DrvBitBlt. Since DrvBitBlt is
* plenty fast, we let DrvBitBlt handle all the cases.
*
* (I used to have a bunch of extra code here to optimize the SRCCOPY
* cases, but the performance win was immeasurable. There's no point in
* the adding code complexity or the working set hit.)
*
\**************************************************************************/
BOOL DrvCopyBits(
SURFOBJ* psoDst,
SURFOBJ* psoSrc,
CLIPOBJ* pco,
XLATEOBJ* pxlo,
RECTL* prclDst,
POINTL* pptlSrc)
{
return(DrvBitBlt(psoDst, psoSrc, NULL, pco, pxlo, prclDst, pptlSrc, NULL,
NULL, NULL, 0x0000CCCC));
}
/******************************Public*Routine******************************\
* BOOL DrvTransparentBlt
*
* Do blt using a source color-key.
*
\**************************************************************************/
BOOL DrvTransparentBlt(
SURFOBJ* psoDst,
SURFOBJ* psoSrc,
CLIPOBJ* pco,
XLATEOBJ* pxlo,
RECTL* prclDst,
RECTL* prclSrc,
ULONG iTransparentColor,
ULONG ulReserved)
{
DSURF* pdsurfSrc;
DSURF* pdsurfDst;
PDEV* ppdev;
ULONG c;
BOOL bMore;
POINTL ptlSrc;
RECTL rcl;
CLIPENUM ce;
pdsurfSrc = (DSURF*) psoSrc->dhsurf;
pdsurfDst = (DSURF*) psoDst->dhsurf;
// We only handle the case when both surfaces are in video memory
// and when no stretching is involved. (GDI using USWC write-
// combining is perfectly fast for the case where the source is
// a DIB and the destination is video memory.)
if (((pdsurfSrc == NULL) || (pdsurfSrc->dt & DT_DIB)) ||
((pdsurfDst == NULL) || (pdsurfDst->dt & DT_DIB)) ||
((pxlo != NULL) && !(pxlo->flXlate & XO_TRIVIAL)) ||
((prclSrc->right - prclSrc->left) != (prclDst->right - prclDst->left)) ||
((prclSrc->bottom - prclSrc->top) != (prclDst->bottom - prclDst->top)))
{
return(EngTransparentBlt(psoDst, psoSrc, pco, pxlo, prclDst, prclSrc,
iTransparentColor, ulReserved));
}
ppdev = (PDEV*) psoDst->dhpdev;
ppdev->xOffset = pdsurfDst->x;
ppdev->yOffset = pdsurfDst->y;
ptlSrc.x = prclSrc->left - (pdsurfDst->x - pdsurfSrc->x);
ptlSrc.y = prclSrc->top - (pdsurfDst->y - pdsurfSrc->y);
if ((pco == NULL) || (pco->iDComplexity == DC_TRIVIAL))
{
ppdev->pfnCopyTransparent(ppdev, 1, prclDst, &ptlSrc,
prclDst, iTransparentColor);
}
else if (pco->iDComplexity == DC_RECT)
{
if (bIntersect(prclDst, &pco->rclBounds, &rcl))
ppdev->pfnCopyTransparent(ppdev, 1, &rcl, &ptlSrc,
prclDst, iTransparentColor);
}
else
{
CLIPOBJ_cEnumStart(pco, FALSE, CT_RECTANGLES, CD_ANY, 0);
do {
bMore = CLIPOBJ_bEnum(pco, sizeof(ce), (ULONG*) &ce);
c = cIntersect(prclDst, ce.arcl, ce.c);
if (c != 0)
{
ppdev->pfnCopyTransparent(ppdev, c, ce.arcl, &ptlSrc,
prclDst, iTransparentColor);
}
} while (bMore);
}
return(TRUE);
}