Leaked source code of windows server 2003
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

1517 lines
49 KiB

/******************************Module*Header**********************************\
*
* *******************
* * GDI SAMPLE CODE *
* *******************
*
* Module Name: fillpath.c
*
* Copyright (c) 1994-1998 3Dlabs Inc. Ltd. All rights reserved.
* Copyright (c) 1995-1999 Microsoft Corporation. All rights reserved.
\*****************************************************************************/
// LATER identify convex polygons and special-case?
// LATER identify vertical edges and special-case?
// LATER move pointed-to variables into automatics in search loops
// LATER punt to the engine with segmented framebuffer callbacks
// LATER handle complex clipping
// LATER coalesce rectangles
#include "precomp.h"
#include "log.h"
#include "gdi.h"
#include "clip.h"
#define ALLOC_TAG ALLOC_TAG_IF2P
//-----------------------------Public*Routine----------------------------------
//
// DrvFillPath
//
// This function fills the specified path with the specified brush and ROP.
// This function detects single convex polygons, and will call to separate
// faster convex polygon code for those cases. This routine also detects
// polygons that are really rectangles, and handles those separately as well.
//
// Parameters
// pso---------Points to a SURFOBJ structure that defines the surface on which
// to draw.
// ppo---------Points to a PATHOBJ structure that defines the path to be filled
// The PATHOBJ_Xxx service routines are provided to enumerate the
// lines, Bezier curves, and other data that make up the path.
// pco---------Points to a CLIPOBJ structure. The CLIPOBJ_Xxx service routines
// are provided to enumerate the clip region as a set of rectangles.
// pbo---------Points to a BRUSHOBJ structure that defines the pattern and
// colors to fill with.
// pptlBrushOrg-Points to a POINTL structure that defines the brush origin,
// which is used to align the brush pattern on the device.
// mix---------Defines the foreground and background raster operations to use
// for the brush.
// flOptions---Specifies either FP_WINDINGMODE, indicating that a winding mode
// fill should be performed, or FP_ALTERNATEMODE, indicating that
// an alternating mode fill should be performed. All other flags
// should be ignored.
//
// Return Value
// The return value is TRUE if the driver is able to fill the path. If the
// path or clipping is too complex to be handled by the driver and should be
// handled by GDI, the return value is FALSE, and an error code is not logged.
// If the driver encounters an unexpected error, such as not being able to
// realize the brush, the return value is DDI_ERROR, and an error code is
// logged.
//
// Comments
// GDI can call DrvFillPath to fill a path on a device-managed surface. When
// deciding whether to call this function, GDI compares the fill requirements
// with the following flags in the flGraphicsCaps member of the DEVINFO
// structure: GCAPS_BEZIERS, GCAPS_ALTERNATEFILL, and GCAPS_WINDINGFILL.
//
// The mix mode defines how the incoming pattern should be mixed with the data
// already on the device surface. The MIX data type consists of two ROP2 values
// packed into a single ULONG. The low-order byte defines the foreground raster
// operation; the next byte defines the background raster operation.
//
// Multiple polygons in a path cannot be treated as being disjoint; The fill
// must consider all the points in the path. That is, if the path contains
// multiple polygons, you cannot simply draw one polygon after the other
// (unless they don't overlap).
//
// This function is an optional entry point for the driver. but is recommended
// for good performance. To get GDI to call this function, not only do you
// have to HOOK_FILLPATH, you have to set GCAPS_ALTERNATEFILL and/or
// GCAPS_WINDINGFILL.
//
//-----------------------------------------------------------------------------
BOOL
DrvFillPath(SURFOBJ* pso,
PATHOBJ* ppo,
CLIPOBJ* pco,
BRUSHOBJ* pbo,
POINTL* pptlBrush,
MIX mix,
FLONG flOptions)
{
GFNPB pb;
BYTE jClipping; // Clipping type
EDGE* pCurrentEdge;
EDGE AETHead; // Dummy head/tail node & sentinel for Active Edge
// Table
EDGE* pAETHead; // Pointer to AETHead
EDGE GETHead; // Dummy head/tail node & sentinel for Global Edge
// Table
EDGE* pGETHead; // Pointer to GETHead
EDGE* pFreeEdges = NULL; // Pointer to memory free for use to store edges
ULONG ulNumRects; // # of rectangles to draw currently in rectangle
// list
RECTL* prclRects; // Pointer to start of rectangle draw list
INT iCurrentY; // Scan line for which we're currently scanning out
// the fill
BOOL bMore;
PATHDATA pd;
RECTL ClipRect;
PDev* ppdev;
BOOL bRetVal=FALSE; // FALSE until proven TRUE
BOOL bMemAlloced=FALSE; // FALSE until proven TRUE
FLONG flFirstRecord;
POINTFIX* pPtFxTmp;
ULONG ulPtFxTmp;
POINTFIX aptfxBuf[NUM_BUFFER_POINTS];
ULONG ulRop4;
DBG_GDI((6, "DrvFillPath called"));
pb.psurfDst = (Surf*)pso->dhsurf;
pb.pco = pco;
ppdev = pb.psurfDst->ppdev;
pb.ppdev = ppdev;
pb.ulRop4 = gaMix[mix & 0xFF] | (gaMix[mix >> 8] << 8);
ulRop4 = pb.ulRop4;
//@@BEGIN_DDKSPLIT
#if MULTITHREADED
if(pb.ppdev->ulLockCount)
{
DBG_GDI((MT_LOG_LEVEL, "DrvBitBlt: re-entered! %d", pb.ppdev->ulLockCount));
}
EngAcquireSemaphore(pb.ppdev->hsemLock);
pb.ppdev->ulLockCount++;
#endif
//@@END_DDKSPLIT
vCheckGdiContext(ppdev);
//
// There's nothing to do if there are only one or two points
//
if ( ppo->cCurves <= 2 )
{
goto ReturnTrue;
}
//
// Pass the surface off to GDI if it's a device bitmap that we've uploaded
// to the system memory.
//
if ( pb.psurfDst->flags == SF_SM )
{
DBG_GDI((1, "dest surface is in system memory. Punt it back"));
//@@BEGIN_DDKSPLIT
#if MULTITHREADED
pb.ppdev->ulLockCount--;
EngReleaseSemaphore(pb.ppdev->hsemLock);
#endif
//@@END_DDKSPLIT
return ( EngFillPath(pso, ppo, pco, pbo, pptlBrush, mix, flOptions));
}
//
// Set up the clipping
//
if ( pco == (CLIPOBJ*)NULL )
{
//
// No CLIPOBJ provided, so we don't have to worry about clipping
//
jClipping = DC_TRIVIAL;
}
else
{
//
// Use the CLIPOBJ-provided clipping
//
jClipping = pco->iDComplexity;
}
//
// Now we are sure the surface we are going to draw is in the video memory
//
// Set default fill as solid fill
//
pb.pgfn = vSolidFillWithRop;
pb.solidColor = 0; //Assume we don't need a pattern
pb.prbrush = NULL;
//
// It is too difficult to determine interaction between
// multiple paths, if there is more than one, skip this
//
PATHOBJ_vEnumStart(ppo);
bMore = PATHOBJ_bEnum(ppo, &pd);
//
// First we need to check if we need a pattern or not
//
if ( (((ulRop4 & 0xff00) >> 8) != (ulRop4 & 0x00ff))
|| ((((ulRop4 >> 4) ^ (ulRop4)) & 0xf0f) != 0) )
{
pb.solidColor = pbo->iSolidColor;
//
// Check to see if it is a non-solid brush (-1)
//
if ( pbo->iSolidColor == -1 )
{
//
// Get the driver's realized brush
//
pb.prbrush = (RBrush*)pbo->pvRbrush;
//
// If it hasn't been realized, do it
// Note: GDI will call DrvRealizeBrsuh to fullfill this task. So the
// driver should have this function ready
//
if ( pb.prbrush == NULL )
{
DBG_GDI((7, "Realizing brush"));
pb.prbrush = (RBrush*)BRUSHOBJ_pvGetRbrush(pbo);
if ( pb.prbrush == NULL )
{
//
// If we can't realize it, nothing we can do
//
//@@BEGIN_DDKSPLIT
#if MULTITHREADED
pb.ppdev->ulLockCount--;
EngReleaseSemaphore(pb.ppdev->hsemLock);
#endif
//@@END_DDKSPLIT
return(FALSE);
}
DBG_GDI((7, "Brsuh realizing done"));
}// Realize brush
pb.pptlBrush = pptlBrush;
//
// Check if brush pattern is 1 BPP or not
// Note: This is set in DrvRealizeBrush
//
if ( pb.prbrush->fl & RBRUSH_2COLOR )
{
//
// 1 BPP pattern. Do a Mono fill
//
pb.pgfn = vMonoPatFill;
}
else
{
//
// Pattern is more than 1 BPP. Do color pattern fill
//
pb.pgfn = vPatFill;
DBG_GDI((7, "Skip Fast Fill Color Pattern"));
//
// P2 can not handle fast filled patterns
//
goto SkipFastFill;
}
}// Handle non-solid brush
}// Blackness check
//
// For solid brush, we can use FastFill
//
if ( bMore )
{
//
// FastFill only knows how to take a single contiguous buffer
// of points. Unfortunately, GDI sometimes hands us paths
// that are split over multiple path data records. Convex
// figures such as Ellipses, Pies and RoundRects are almost
// always given in multiple records. Since probably 90% of
// multiple record paths could still be done by FastFill, for
// those cases we simply copy the points into a contiguous
// buffer...
//
// First make sure that the entire path would fit in the
// temporary buffer, and make sure the path isn't comprised
// of more than one subpath:
//
if ( (ppo->cCurves >= NUM_BUFFER_POINTS)
||(pd.flags & PD_ENDSUBPATH) )
{
goto SkipFastFill;
}
pPtFxTmp = &aptfxBuf[0];
//
// Copy one vertex over to pPtFxTmp from pd(path data)
//
RtlCopyMemory(pPtFxTmp, pd.pptfx, sizeof(POINTFIX) * pd.count);
//
// Move the memory pointer over to next structure
//
pPtFxTmp += pd.count;
ulPtFxTmp = pd.count;
flFirstRecord = pd.flags; // Remember PD_BEGINSUBPATH flag
//
// Loop to get all the vertex info. After the loop, all the vertex info
// will be in array aptfxBuf[]
//
do
{
bMore = PATHOBJ_bEnum(ppo, &pd);
RtlCopyMemory(pPtFxTmp, pd.pptfx, sizeof(POINTFIX) * pd.count);
ulPtFxTmp += pd.count;
pPtFxTmp += pd.count;
} while ( !(pd.flags & PD_ENDSUBPATH) );
//
// Fake up the path data record
//
pd.pptfx = &aptfxBuf[0];
pd.count = ulPtFxTmp;
pd.flags |= flFirstRecord;
//
// If there's more than one subpath, we can't call FastFill
//
DBG_GDI((7, "More than one subpath!"));
if ( bMore )
{
goto SkipFastFill;
}
}// if ( bMore )
//
// Fast polygon fill
//
if ( bFillPolygon(ppdev, (Surf*)pso->dhsurf, pd.count,
pd.pptfx, pb.solidColor,
ulRop4,
pco, pb.prbrush, pptlBrush) )
{
DBG_GDI((7, "Fast Fill Succeeded"));
InputBufferFlush(ppdev);
//@@BEGIN_DDKSPLIT
#if MULTITHREADED
pb.ppdev->ulLockCount--;
EngReleaseSemaphore(pb.ppdev->hsemLock);
#endif
//@@END_DDKSPLIT
return (TRUE);
}
SkipFastFill:
DBG_GDI((7, "Fast Fill Skipped"));
if ( jClipping != DC_TRIVIAL )
{
if ( jClipping != DC_RECT )
{
DBG_GDI((7, "Complex Clipping"));
//
// There is complex clipping; let GDI fill the path
//
goto ReturnFalse;
}
//
// Clip to the clip rectangle
//
ClipRect = pco->rclBounds;
}
else
{
//
// So the y-clipping code doesn't do any clipping
// We don't blow the values out when we scale up to GIQ
//
ClipRect.top = (LONG_MIN + 1) / 16; // +1 to avoid compiler problem
ClipRect.bottom = LONG_MAX / 16;
}
//
// Set up working storage in the temporary buffer, storage for list of
// rectangles to draw
// Note: ppdev->pvTmpBuffer is allocated in DrvEnableSurface() in enable.c
// The purpose of using ppdev->pvTmpBuffer is to save us from having to
// allocate and free the temp space inside high frequency calls. It was
// allocated for TMP_BUFFER_SIZE bytes and will be freed in
// DrvDeleteSurface()
//
prclRects = (RECTL*)ppdev->pvTmpBuffer;
if ( !bMore )
{
RECTL* pTmpRect;
INT cPoints = pd.count;
//
// The count can't be less than three, because we got all the edges
// in this subpath, and above we checked that there were at least
// three edges
//
// If the count is four, check to see if the polygon is really a
// rectangle since we can really speed that up. We'll also check for
// five with the first and last points the same.
//
// ??? we have already done the memcpy for the pd data. shall we use it
//
if ( ( cPoints == 4 )
||( ( cPoints == 5 )
&&(pd.pptfx[0].x == pd.pptfx[4].x)
&&(pd.pptfx[0].y == pd.pptfx[4].y) ) )
{
//
// Get storage space for this temp rectangle
//
pTmpRect = prclRects;
//
// We have to start somewhere to assume that most
// applications specify the top left point first
// We want to check that the first two points are
// either vertically or horizontally aligned. If
// they are then we check that the last point [3]
// is either horizontally or vertically aligned,
// and finally that the 3rd point [2] is aligned
// with both the first point and the last point
//
pTmpRect->top = pd.pptfx[0].y - 1 & FIX_MASK;
pTmpRect->left = pd.pptfx[0].x - 1 & FIX_MASK;
pTmpRect->right = pd.pptfx[1].x - 1 & FIX_MASK;
//
// Check if the first two points are vertically alligned
//
if ( pTmpRect->left ^ pTmpRect->right )
{
//
// The first two points are not vertically alligned
// Let's see if these two points are horizontal alligned
//
if ( pTmpRect->top ^ (pd.pptfx[1].y - 1 & FIX_MASK) )
{
//
// The first two points are not horizontally alligned
// So it is not a rectangle
//
goto not_rectangle;
}
//
// Up to now, the first two points are horizontally alligned,
// but not vertically alligned. We need to check if the first
// point vertically alligned with the 4th point
//
if ( pTmpRect->left ^ (pd.pptfx[3].x - 1 & FIX_MASK) )
{
//
// The first point is not vertically alligned with the 4th
// point either. So this is not a rectangle
//
goto not_rectangle;
}
//
// Check if the 2nd point and the 3rd point are vertically aligned
//
if ( pTmpRect->right ^ (pd.pptfx[2].x - 1 & FIX_MASK) )
{
//
// The 2nd point and the 3rd point are not vertically aligned
// So this is not a rectangle
//
goto not_rectangle;
}
//
// Check to see if the 3rd and 4th points are horizontally
// alligned. If not, then it is not a rectangle
//
pTmpRect->bottom = pd.pptfx[2].y - 1 & FIX_MASK;
if ( pTmpRect->bottom ^ (pd.pptfx[3].y - 1 & FIX_MASK) )
{
goto not_rectangle;
}
}// Check if the first two points are vertically alligned
else
{
//
// The first two points are vertically alligned. Now we need to
// check if the 1st point and the 4th point are horizontally
// aligned. If not, then this is not a rectangle
//
if ( pTmpRect->top ^ (pd.pptfx[3].y - 1 & FIX_MASK) )
{
goto not_rectangle;
}
//
// Check if the 2nd point and the 3rd point are horizontally
// aligned. If not, then this is not a rectangle
//
pTmpRect->bottom = pd.pptfx[1].y - 1 & FIX_MASK;
if ( pTmpRect->bottom ^ (pd.pptfx[2].y - 1 & FIX_MASK) )
{
goto not_rectangle;
}
//
// Check if the 3rd point and the 4th point are vertically
// aligned. If not, then this is not a rectangle
//
pTmpRect->right = pd.pptfx[2].x - 1 & FIX_MASK;
if ( pTmpRect->right ^ (pd.pptfx[3].x - 1 & FIX_MASK) )
{
goto not_rectangle;
}
}
//
// We have a rectangle now. Do some adjustment here first
// If the left is greater than the right then
// swap them so the blt code won't have problem
//
if ( pTmpRect->left > pTmpRect->right )
{
FIX temp;
temp = pTmpRect->left;
pTmpRect->left = pTmpRect->right;
pTmpRect->right = temp;
}
else
{
//
// If left == right there's nothing to draw
//
if ( pTmpRect->left == pTmpRect->right )
{
DBG_GDI((7, "Nothing to draw"));
goto ReturnTrue;
}
}// Adjust right and left edge
//
// Shift the values to get pixel coordinates
//
pTmpRect->left = (pTmpRect->left >> FIX_SHIFT) + 1;
pTmpRect->right = (pTmpRect->right >> FIX_SHIFT) + 1;
//
// Adjust the top and bottom coordiantes if necessary
//
if ( pTmpRect->top > pTmpRect->bottom )
{
FIX temp;
temp = pTmpRect->top;
pTmpRect->top = pTmpRect->bottom;
pTmpRect->bottom = temp;
}
else
{
if ( pTmpRect->top == pTmpRect->bottom )
{
DBG_GDI((7, "Nothing to draw"));
goto ReturnTrue;
}
}
//
// Shift the values to get pixel coordinates
//
pTmpRect->top = (pTmpRect->top >> FIX_SHIFT) + 1;
pTmpRect->bottom = (pTmpRect->bottom >> FIX_SHIFT) + 1;
//
// Finally, check for clipping
//
if ( jClipping == DC_RECT )
{
//
// Clip to the clip rectangle
//
if ( !bIntersect(pTmpRect, &ClipRect, pTmpRect) )
{
//
// Totally clipped, nothing to do
//
DBG_GDI((7, "Nothing to draw"));
goto ReturnTrue;
}
}
//
// If we get here then the polygon is a rectangle,
// set count to 1 and goto bottom to draw it
//
ulNumRects = 1;
goto draw_remaining_rectangles;
}// Check to see if it is a rectangle
not_rectangle:
;
}// if ( !bMore )
//
// Do we have enough memory for all the edges?
// LATER does cCurves include closure????
//
if ( ppo->cCurves > MAX_EDGES )
{
//
// Try to allocate enough memory
//
pFreeEdges = (EDGE*)ENGALLOCMEM(0, (ppo->cCurves * sizeof(EDGE)),
ALLOC_TAG);
if ( pFreeEdges == NULL )
{
DBG_GDI((1, "Can't allocate memory for %d edges", ppo->cCurves));
//
// Too many edges; let GDI fill the path
//
goto ReturnFalse;
}
else
{
//
// Set a flag to indicate that we have allocate the memory so that
// we can free it later
//
bMemAlloced = TRUE;
}
}// if ( ppo->cCurves > MAX_EDGES )
else
{
//
// If the total number of edges doesn't exceed the MAX_EDGES, then just
// use our handy temporary buffer (it's big enough)
//
pFreeEdges = (EDGE*)((BYTE*)ppdev->pvTmpBuffer + RECT_BYTES);
}
//
// Initialize an empty list of rectangles to fill
//
ulNumRects = 0;
//
// Enumerate the path edges and build a Global Edge Table (GET) from them
// in YX-sorted order.
//
pGETHead = &GETHead;
if ( !bConstructGET(pGETHead, pFreeEdges, ppo, &pd, bMore, &ClipRect) )
{
DBG_GDI((7, "Outside Range"));
goto ReturnFalse; // outside GDI's 2**27 range
}
//
// Create an empty AET with the head node also a tail sentinel
//
pAETHead = &AETHead;
AETHead.pNext = pAETHead; // Mark that the AET is empty
AETHead.X = 0x7FFFFFFF; // This is greater than any valid X value, so
// searches will always terminate
//
// Top scan of polygon is the top of the first edge we come to
//
iCurrentY = ((EDGE*)GETHead.pNext)->Y;
//
// Loop through all the scans in the polygon, adding edges from the GET to
// the Active Edge Table (AET) as we come to their starts, and scanning out
// the AET at each scan into a rectangle list. Each time it fills up, the
// rectangle list is passed to the filling routine, and then once again at
// the end if any rectangles remain undrawn. We continue so long as there
// are edges to be scanned out
//
while ( 1 )
{
//
// Advance the edges in the AET one scan, discarding any that have
// reached the end (if there are any edges in the AET)
//
if ( AETHead.pNext != pAETHead )
{
vAdvanceAETEdges(pAETHead);
}
//
// If the AET is empty, done if the GET is empty, else jump ahead to
// the next edge in the GET; if the AET isn't empty, re-sort the AET
//
if ( AETHead.pNext == pAETHead )
{
if ( GETHead.pNext == pGETHead )
{
//
// Done if there are no edges in either the AET or the GET
//
break;
}
//
// There are no edges in the AET, so jump ahead to the next edge in
// the GET
//
iCurrentY = ((EDGE*)GETHead.pNext)->Y;
}
else
{
//
// Re-sort the edges in the AET by X coordinate, if there are at
// least two edges in the AET (there could be one edge if the
// balancing edge hasn't yet been added from the GET)
//
if ( ((EDGE*)AETHead.pNext)->pNext != pAETHead )
{
vXSortAETEdges(pAETHead);
}
}
//
// Move any new edges that start on this scan from the GET to the AET;
// bother calling only if there's at least one edge to add
//
if ( ((EDGE*)GETHead.pNext)->Y == iCurrentY )
{
vMoveNewEdges(pGETHead, pAETHead, iCurrentY);
}
//
// Scan the AET into rectangles to fill (there's always at least one
// edge pair in the AET)
//
pCurrentEdge = (EDGE*)AETHead.pNext; // point to the first edge
do
{
INT iLeftEdge;
//
// The left edge of any given edge pair is easy to find; it's just
// wherever we happen to be currently
//
iLeftEdge = pCurrentEdge->X;
//
// Find the matching right edge according to the current fill rule
//
if ( (flOptions & FP_WINDINGMODE) != 0 )
{
INT iWindingCount;
//
// Do winding fill; scan across until we've found equal numbers
// of up and down edges
//
iWindingCount = pCurrentEdge->iWindingDirection;
do
{
pCurrentEdge = (EDGE*)pCurrentEdge->pNext;
iWindingCount += pCurrentEdge->iWindingDirection;
} while ( iWindingCount != 0 );
}
else
{
//
// Odd-even fill; the next edge is the matching right edge
//
pCurrentEdge = (EDGE*)pCurrentEdge->pNext;
}
//
// See if the resulting span encompasses at least one pixel, and
// add it to the list of rectangles to draw if so
//
if ( iLeftEdge < pCurrentEdge->X )
{
//
// We've got an edge pair to add to the list to be filled; see
// if there's room for one more rectangle
//
if ( ulNumRects >= MAX_PATH_RECTS )
{
//
// No more room; draw the rectangles in the list and reset
// it to empty
//
pb.lNumRects = ulNumRects;
pb.pRects = prclRects;
pb.pgfn(&pb);
//
// Reset the list to empty
//
ulNumRects = 0;
}
//
// Add the rectangle representing the current edge pair
//
if ( jClipping == DC_RECT )
{
//
// Clipped
// Clip to left
//
prclRects[ulNumRects].left = max(iLeftEdge, ClipRect.left);
//
// Clip to right
//
prclRects[ulNumRects].right =
min(pCurrentEdge->X, ClipRect.right);
//
// Draw only if not fully clipped
//
if ( prclRects[ulNumRects].left
< prclRects[ulNumRects].right )
{
prclRects[ulNumRects].top = iCurrentY;
prclRects[ulNumRects].bottom = iCurrentY + 1;
ulNumRects++;
}
}
else
{
//
// Unclipped
//
prclRects[ulNumRects].top = iCurrentY;
prclRects[ulNumRects].bottom = iCurrentY + 1;
prclRects[ulNumRects].left = iLeftEdge;
prclRects[ulNumRects].right = pCurrentEdge->X;
ulNumRects++;
}
}
} while ( (pCurrentEdge = (EDGE*)pCurrentEdge->pNext) != pAETHead );
iCurrentY++; // next scan
}// Loop through all the scans in the polygon
//
// Draw the remaining rectangles, if there are any
//
draw_remaining_rectangles:
if ( ulNumRects > 0 )
{
pb.lNumRects = ulNumRects;
pb.pRects = prclRects;
pb.pgfn(&pb);
}
ReturnTrue:
DBG_GDI((7, "Drawn"));
bRetVal = TRUE; // done successfully
ReturnFalse:
//
// bRetVal is originally false. If you jumped to ReturnFalse from somewhere,
// then it will remain false, and be returned.
//
if ( bMemAlloced )
{
//
// We did allocate memory, so release it
//
ENGFREEMEM(pFreeEdges);
}
DBG_GDI((6, "Returning %s", bRetVal ? "True" : "False"));
InputBufferFlush(ppdev);
//@@BEGIN_DDKSPLIT
#if MULTITHREADED
pb.ppdev->ulLockCount--;
EngReleaseSemaphore(pb.ppdev->hsemLock);
#endif
//@@END_DDKSPLIT
return (bRetVal);
}// DrvFillPath()
//-----------------------------------------------------------------------------
//
// void vAdvanceAETEdges(EDGE* pAETHead)
//
// Advance the edges in the AET to the next scan, dropping any for which we've
// done all scans. Assumes there is at least one edge in the AET.
//
//-----------------------------------------------------------------------------
VOID
vAdvanceAETEdges(EDGE* pAETHead)
{
EDGE* pLastEdge;
EDGE* pCurrentEdge;
pLastEdge = pAETHead;
pCurrentEdge = (EDGE*)pLastEdge->pNext;
do
{
//
// Count down this edge's remaining scans
//
if ( --pCurrentEdge->iScansLeft == 0 )
{
//
// We've done all scans for this edge; drop this edge from the AET
//
pLastEdge->pNext = pCurrentEdge->pNext;
}
else
{
//
// Advance the edge's X coordinate for a 1-scan Y advance
// Advance by the minimum amount
//
pCurrentEdge->X += pCurrentEdge->iXWhole;
//
// Advance the error term and see if we got one extra pixel this
// time
//
pCurrentEdge->iErrorTerm += pCurrentEdge->iErrorAdjustUp;
if ( pCurrentEdge->iErrorTerm >= 0 )
{
//
// The error term turned over, so adjust the error term and
// advance the extra pixel
//
pCurrentEdge->iErrorTerm -= pCurrentEdge->iErrorAdjustDown;
pCurrentEdge->X += pCurrentEdge->iXDirection;
}
pLastEdge = pCurrentEdge;
}
} while ((pCurrentEdge = (EDGE *)pLastEdge->pNext) != pAETHead);
}// vAdvanceAETEdges()
//-----------------------------------------------------------------------------
//
// VOID vXSortAETEdges(EDGE* pAETHead)
//
// X-sort the AET, because the edges may have moved around relative to
// one another when we advanced them. We'll use a multipass bubble
// sort, which is actually okay for this application because edges
// rarely move relative to one another, so we usually do just one pass.
// Also, this makes it easy to keep just a singly-linked list. Assumes there
// are at least two edges in the AET.
//
//-----------------------------------------------------------------------------
VOID
vXSortAETEdges(EDGE *pAETHead)
{
BOOL bEdgesSwapped;
EDGE* pLastEdge;
EDGE* pCurrentEdge;
EDGE* pNextEdge;
do
{
bEdgesSwapped = FALSE;
pLastEdge = pAETHead;
pCurrentEdge = (EDGE *)pLastEdge->pNext;
pNextEdge = (EDGE *)pCurrentEdge->pNext;
do
{
if ( pNextEdge->X < pCurrentEdge->X )
{
//
// Next edge is to the left of the current edge; swap them
//
pLastEdge->pNext = pNextEdge;
pCurrentEdge->pNext = pNextEdge->pNext;
pNextEdge->pNext = pCurrentEdge;
bEdgesSwapped = TRUE;
//
// Continue sorting before the edge we just swapped; it might
// move farther yet
//
pCurrentEdge = pNextEdge;
}
pLastEdge = pCurrentEdge;
pCurrentEdge = (EDGE *)pLastEdge->pNext;
} while ( (pNextEdge = (EDGE*)pCurrentEdge->pNext) != pAETHead );
} while ( bEdgesSwapped );
}// vXSortAETEdges()
//-----------------------------------------------------------------------------
//
// VOID vMoveNewEdges(EDGE* pGETHead, EDGE* pAETHead, INT iCurrentY)
//
// Moves all edges that start on the current scan from the GET to the AET in
// X-sorted order. Parameters are pointer to head of GET and pointer to dummy
// edge at head of AET, plus current scan line. Assumes there's at least one
// edge to be moved.
//
//-----------------------------------------------------------------------------
VOID
vMoveNewEdges(EDGE* pGETHead,
EDGE* pAETHead,
INT iCurrentY)
{
EDGE* pCurrentEdge = pAETHead;
EDGE* pGETNext = (EDGE*)pGETHead->pNext;
do
{
//
// Scan through the AET until the X-sorted insertion point for this
// edge is found. We can continue from where the last search left
// off because the edges in the GET are in X sorted order, as is
// the AET. The search always terminates because the AET sentinel
// is greater than any valid X
//
while ( pGETNext->X > ((EDGE *)pCurrentEdge->pNext)->X )
{
pCurrentEdge = (EDGE*)pCurrentEdge->pNext;
}
//
// We've found the insertion point; add the GET edge to the AET, and
// remove it from the GET
//
pGETHead->pNext = pGETNext->pNext;
pGETNext->pNext = pCurrentEdge->pNext;
pCurrentEdge->pNext = pGETNext;
pCurrentEdge = pGETNext; // continue insertion search for the next
// GET edge after the edge we just added
pGETNext = (EDGE*)pGETHead->pNext;
} while (pGETNext->Y == iCurrentY);
}// vMoveNewEdges()
//-----------------------------------------------------------------------------
//
// BOOL (EDGE* pGETHead, EDGE* pAETHead, INT iCurrentY)
//
// Build the Global Edge Table from the path. There must be enough memory in
// the free edge area to hold all edges. The GET is constructed in Y-X order,
// and has a head/tail/sentinel node at pGETHead.
//
//-----------------------------------------------------------------------------
BOOL
bConstructGET(EDGE* pGETHead,
EDGE* pFreeEdges,
PATHOBJ* ppo,
PATHDATA* pd,
BOOL bMore,
RECTL* pClipRect)
{
POINTFIX pfxPathStart; // point that started the current subpath
POINTFIX pfxPathPrevious; // point before the current point in a subpath;
// starts the current edge
//
// Create an empty GET with the head node also a tail sentinel
//
pGETHead->pNext = pGETHead; // mark that the GET is empty
pGETHead->Y = 0x7FFFFFFF; // this is greater than any valid Y value, so
// Searches will always terminate
//
// Note: PATHOBJ_vEnumStart is implicitly performed by engine
// already and first path is enumerated by the caller
// so here we don't need to call it again.
//
next_subpath:
//
// Make sure the PATHDATA is not empty (is this necessary)???
//
if ( pd->count != 0 )
{
//
// If first point starts a subpath, remember it as such
// and go on to the next point, so we can get an edge
//
if ( pd->flags & PD_BEGINSUBPATH )
{
//
// The first point starts the subpath; Remember it
//
pfxPathStart = *pd->pptfx; // the subpath starts here
pfxPathPrevious = *pd->pptfx; // this point starts the next edge
pd->pptfx++; // advance to the next point
pd->count--; // count off this point
}
//
// Add edges in PATHDATA to GET, in Y-X sorted order
//
while ( pd->count-- )
{
if ( (pFreeEdges =
pAddEdgeToGET(pGETHead, pFreeEdges, &pfxPathPrevious,
pd->pptfx, pClipRect)) == NULL )
{
goto ReturnFalse;
}
pfxPathPrevious = *pd->pptfx; // current point becomes previous
pd->pptfx++; // advance to the next point
}// Loop through all the points
//
// If last point ends the subpath, insert the edge that
// connects to first point (is this built in already?)
//
if ( pd->flags & PD_ENDSUBPATH )
{
if ( (pFreeEdges = pAddEdgeToGET(pGETHead, pFreeEdges, &pfxPathPrevious,
&pfxPathStart, pClipRect)) == NULL )
{
goto ReturnFalse;
}
}
}// if ( pd->count != 0 )
//
// The initial loop conditions preclude a do, while or for
//
if ( bMore )
{
bMore = PATHOBJ_bEnum(ppo, pd);
goto next_subpath;
}
return(TRUE); // done successfully
ReturnFalse:
return(FALSE); // failed
}// bConstructGET()
//-----------------------------------------------------------------------------
//
// EDGE* pAddEdgeToGET(EDGE* pGETHead, EDGE* pFreeEdge, POINTFIX* ppfxEdgeStart,
// POINTFIX* ppfxEdgeEnd, RECTL* pClipRect)
//
// Adds the edge described by the two passed-in points to the Global Edge
// Table (GET), if the edge spans at least one pixel vertically.
//
//-----------------------------------------------------------------------------
EDGE*
pAddEdgeToGET(EDGE* pGETHead,
EDGE* pFreeEdge,
POINTFIX* ppfxEdgeStart,
POINTFIX* ppfxEdgeEnd,
RECTL* pClipRect)
{
int iYStart;
int iYEnd;
int iXStart;
int iXEnd;
int iYHeight;
int iXWidth;
int yJump;
int yTop;
//
// Set the winding-rule direction of the edge, and put the endpoints in
// top-to-bottom order
//
iYHeight = ppfxEdgeEnd->y - ppfxEdgeStart->y;
if ( iYHeight == 0 )
{
//
// Zero height; ignore this edge
//
return(pFreeEdge);
}
else if ( iYHeight > 0 )
{
//
// Top-to-bottom
//
iXStart = ppfxEdgeStart->x;
iYStart = ppfxEdgeStart->y;
iXEnd = ppfxEdgeEnd->x;
iYEnd = ppfxEdgeEnd->y;
pFreeEdge->iWindingDirection = 1;
}
else
{
iYHeight = -iYHeight;
iXEnd = ppfxEdgeStart->x;
iYEnd = ppfxEdgeStart->y;
iXStart = ppfxEdgeEnd->x;
iYStart = ppfxEdgeEnd->y;
pFreeEdge->iWindingDirection = -1;
}
if ( iYHeight & 0x80000000 )
{
//
// Too large; outside 2**27 GDI range
//
return(NULL);
}
//
// Set the error term and adjustment factors, all in GIQ coordinates for
// now
//
iXWidth = iXEnd - iXStart;
if ( iXWidth >= 0 )
{
//
// Left to right, so we change X as soon as we move at all
//
pFreeEdge->iXDirection = 1;
pFreeEdge->iErrorTerm = -1;
}
else
{
//
// Right to left, so we don't change X until we've moved a full GIQ
// coordinate
//
iXWidth = -iXWidth;
pFreeEdge->iXDirection = -1;
pFreeEdge->iErrorTerm = -iYHeight;
}
if ( iXWidth & 0x80000000 )
{
//
// Too large; outside 2**27 GDI range
//
return(NULL);
}
if ( iXWidth >= iYHeight )
{
//
// Calculate base run length (minimum distance advanced in X for a 1-
// scan advance in Y)
//
pFreeEdge->iXWhole = iXWidth / iYHeight;
//
// Add sign back into base run length if going right to left
//
if ( pFreeEdge->iXDirection == -1 )
{
pFreeEdge->iXWhole = -pFreeEdge->iXWhole;
}
pFreeEdge->iErrorAdjustUp = iXWidth % iYHeight;
}
else
{
//
// Base run length is 0, because line is closer to vertical than
// horizontal
//
pFreeEdge->iXWhole = 0;
pFreeEdge->iErrorAdjustUp = iXWidth;
}
pFreeEdge->iErrorAdjustDown = iYHeight;
//
// Calculate the number of pixels spanned by this edge, accounting for
// clipping
//
// Top true pixel scan in GIQ coordinates
// Shifting to divide and multiply by 16 is okay because the clip rect
// always contains positive numbers
//
yTop = max(pClipRect->top << 4, (iYStart + 15) & ~0x0F);
//
// Initial scan line on which to fill edge
//
pFreeEdge->Y = yTop >> 4;
//
// Calculate # of scans to actually fill, accounting for clipping
//
if ( (pFreeEdge->iScansLeft = min(pClipRect->bottom, ((iYEnd + 15) >> 4))
- pFreeEdge->Y) <= 0 )
{
//
// No pixels at all are spanned, so we can ignore this edge
//
return(pFreeEdge);
}
//
// If the edge doesn't start on a pixel scan (that is, it starts at a
// fractional GIQ coordinate), advance it to the first pixel scan it
// intersects. Ditto if there's top clipping. Also clip to the bottom if
// needed
//
if ( iYStart != yTop )
{
//
// Jump ahead by the Y distance in GIQ coordinates to the first pixel
// to draw
//
yJump = yTop - iYStart;
//
// Advance x the minimum amount for the number of scans traversed
//
iXStart += pFreeEdge->iXWhole * yJump;
vAdjustErrorTerm(&pFreeEdge->iErrorTerm, pFreeEdge->iErrorAdjustUp,
pFreeEdge->iErrorAdjustDown, yJump, &iXStart,
pFreeEdge->iXDirection);
}
//
// Turn the calculations into pixel rather than GIQ calculations
//
// Move the X coordinate to the nearest pixel, and adjust the error term
// accordingly
// Dividing by 16 with a shift is okay because X is always positive
pFreeEdge->X = (iXStart + 15) >> 4; // convert from GIQ to pixel coordinates
//
// LATER adjust only if needed (if prestepped above)?
//
if ( pFreeEdge->iXDirection == 1 )
{
//
// Left to right
//
pFreeEdge->iErrorTerm -= pFreeEdge->iErrorAdjustDown
* (((iXStart + 15) & ~0x0F) - iXStart);
}
else
{
//
// Right to left
//
pFreeEdge->iErrorTerm -= pFreeEdge->iErrorAdjustDown
* ((iXStart - 1) & 0x0F);
}
//
// Scale the error term down 16 times to switch from GIQ to pixels.
// Shifts work to do the multiplying because these values are always
// non-negative
//
pFreeEdge->iErrorTerm >>= 4;
//
// Insert the edge into the GET in YX-sorted order. The search always ends
// because the GET has a sentinel with a greater-than-possible Y value
//
while ( (pFreeEdge->Y > ((EDGE*)pGETHead->pNext)->Y)
||( (pFreeEdge->Y == ((EDGE*)pGETHead->pNext)->Y)
&&(pFreeEdge->X > ((EDGE*)pGETHead->pNext)->X) ) )
{
pGETHead = (EDGE*)pGETHead->pNext;
}
pFreeEdge->pNext = pGETHead->pNext; // link the edge into the GET
pGETHead->pNext = pFreeEdge;
//
// Point to the next edge storage location for next time
//
return(++pFreeEdge);
}// pAddEdgeToGET()
//-----------------------------------------------------------------------------
//
// void vAdjustErrorTerm(int *pErrorTerm, int iErrorAdjustUp,
// int iErrorAdjustDown, int yJump, int *pXStart,
// int iXDirection)
// Adjust the error term for a skip ahead in y. This is in ASM because there's
// a multiply/divide that may involve a larger than 32-bit value.
//
//-----------------------------------------------------------------------------
void
vAdjustErrorTerm(int* pErrorTerm,
int iErrorAdjustUp,
int iErrorAdjustDown,
int yJump,
int* pXStart,
int iXDirection)
{
#if defined(_X86_) || defined(i386)
//
// Adjust the error term up by the number of y coordinates we'll skip
// *pErrorTerm += iErrorAdjustUp * yJump;
//
_asm mov ebx,pErrorTerm
_asm mov eax,iErrorAdjustUp
_asm mul yJump
_asm add eax,[ebx]
_asm adc edx,-1 // the error term starts out negative
//
// See if the error term turned over even once while skipping
//
_asm js short NoErrorTurnover
//
// # of times we'll turn over the error term and step an extra x
// coordinate while skipping
// NumAdjustDowns = (*pErrorTerm / iErrorAdjustDown) + 1;
//
_asm div iErrorAdjustDown
_asm inc eax
//
// Note that EDX is the remainder; (EDX - iErrorAdjustDown) is where
// the error term ends up ultimately
//
// Advance x appropriately for the # of times the error term
// turned over
// if (iXDirection == 1)
// {
// *pXStart += NumAdjustDowns;
// }
// else
// {
// *pXStart -= NumAdjustDowns;
// }
//
_asm mov ecx,pXStart
_asm cmp iXDirection,1
_asm jz short GoingRight
_asm neg eax
GoingRight:
_asm add [ecx],eax
// Adjust the error term down to its proper post-skip value
// *pErrorTerm -= iErrorAdjustDown * NumAdjustDowns;
_asm sub edx,iErrorAdjustDown
_asm mov eax,edx // put into EAX for storing to pErrorTerm next
NoErrorTurnover:
_asm mov [ebx],eax
#else
//
// LONGLONGS are 64 bit integers (We hope!) as the multiply could
// overflow 32 bit integers. If 64 bit ints are unsupported, the
// LONGLONG will end up as a double. Hopefully there will be no
// noticable difference in accuracy.
LONGLONG NumAdjustDowns;
LONGLONG tmpError = *pErrorTerm;
//
// Adjust the error term up by the number of y coordinates we'll skip
//
tmpError += (LONGLONG)iErrorAdjustUp * (LONGLONG)yJump;
//
// See if the error term turned over even once while skipping
//
if ( tmpError >= 0 )
{
//
// # of times we'll turn over the error term and step an extra x
// coordinate while skipping
//
NumAdjustDowns = (tmpError / (LONGLONG)iErrorAdjustDown) + 1;
//
// Advance x appropriately for the # of times the error term
// turned over
//
if ( iXDirection == 1 )
{
*pXStart += (LONG)NumAdjustDowns;
}
else
{
*pXStart -= (LONG) NumAdjustDowns;
}
//
// Adjust the error term down to its proper post-skip value
//
tmpError -= (LONGLONG)iErrorAdjustDown * NumAdjustDowns;
}
*pErrorTerm = (LONG)tmpError;
#endif // X86
}// vAdjustErrorTerm()