Home | History | Annotate | Download | only in libtess
      1 /*
      2 ** License Applicability. Except to the extent portions of this file are
      3 ** made subject to an alternative license as permitted in the SGI Free
      4 ** Software License B, Version 1.1 (the "License"), the contents of this
      5 ** file are subject only to the provisions of the License. You may not use
      6 ** this file except in compliance with the License. You may obtain a copy
      7 ** of the License at Silicon Graphics, Inc., attn: Legal Services, 1600
      8 ** Amphitheatre Parkway, Mountain View, CA 94043-1351, or at:
      9 **
     10 ** http://oss.sgi.com/projects/FreeB
     11 **
     12 ** Note that, as provided in the License, the Software is distributed on an
     13 ** "AS IS" basis, with ALL EXPRESS AND IMPLIED WARRANTIES AND CONDITIONS
     14 ** DISCLAIMED, INCLUDING, WITHOUT LIMITATION, ANY IMPLIED WARRANTIES AND
     15 ** CONDITIONS OF MERCHANTABILITY, SATISFACTORY QUALITY, FITNESS FOR A
     16 ** PARTICULAR PURPOSE, AND NON-INFRINGEMENT.
     17 **
     18 ** Original Code. The Original Code is: OpenGL Sample Implementation,
     19 ** Version 1.2.1, released January 26, 2000, developed by Silicon Graphics,
     20 ** Inc. The Original Code is Copyright (c) 1991-2000 Silicon Graphics, Inc.
     21 ** Copyright in any portions created by third parties is as indicated
     22 ** elsewhere herein. All Rights Reserved.
     23 **
     24 ** Additional Notice Provisions: The application programming interfaces
     25 ** established by SGI in conjunction with the Original Code are The
     26 ** OpenGL(R) Graphics System: A Specification (Version 1.2.1), released
     27 ** April 1, 1999; The OpenGL(R) Graphics System Utility Library (Version
     28 ** 1.3), released November 4, 1998; and OpenGL(R) Graphics with the X
     29 ** Window System(R) (Version 1.3), released October 19, 1998. This software
     30 ** was created using the OpenGL(R) version 1.2.1 Sample Implementation
     31 ** published by SGI, but has not been independently verified as being
     32 ** compliant with the OpenGL(R) version 1.2.1 Specification.
     33 **
     34 */
     35 /*
     36 ** Author: Eric Veach, July 1994.
     37 **
     38 ** $Date$ $Revision$
     39 ** $Header: //depot/main/gfx/lib/glu/libtess/render.c#5 $
     40 */
     41 
     42 #include "gluos.h"
     43 #include <assert.h>
     44 #include <stddef.h>
     45 #include "mesh.h"
     46 #include "tess.h"
     47 #include "render.h"
     48 
     49 #define TRUE 1
     50 #define FALSE 0
     51 
     52 /* This structure remembers the information we need about a primitive
     53  * to be able to render it later, once we have determined which
     54  * primitive is able to use the most triangles.
     55  */
     56 struct FaceCount {
     57   long		size;		/* number of triangles used */
     58   GLUhalfEdge	*eStart;	/* edge where this primitive starts */
     59   void		(*render)(GLUtesselator *, GLUhalfEdge *, long);
     60                                 /* routine to render this primitive */
     61 };
     62 
     63 static struct FaceCount MaximumFan( GLUhalfEdge *eOrig );
     64 static struct FaceCount MaximumStrip( GLUhalfEdge *eOrig );
     65 
     66 static void RenderFan( GLUtesselator *tess, GLUhalfEdge *eStart, long size );
     67 static void RenderStrip( GLUtesselator *tess, GLUhalfEdge *eStart, long size );
     68 static void RenderTriangle( GLUtesselator *tess, GLUhalfEdge *eStart,
     69 			    long size );
     70 
     71 static void RenderMaximumFaceGroup( GLUtesselator *tess, GLUface *fOrig );
     72 static void RenderLonelyTriangles( GLUtesselator *tess, GLUface *head );
     73 
     74 
     75 
     76 /************************ Strips and Fans decomposition ******************/
     77 
     78 /* __gl_renderMesh( tess, mesh ) takes a mesh and breaks it into triangle
     79  * fans, strips, and separate triangles.  A substantial effort is made
     80  * to use as few rendering primitives as possible (ie. to make the fans
     81  * and strips as large as possible).
     82  *
     83  * The rendering output is provided as callbacks (see the api).
     84  */
     85 void __gl_renderMesh( GLUtesselator *tess, GLUmesh *mesh )
     86 {
     87   GLUface *f;
     88 
     89   /* Make a list of separate triangles so we can render them all at once */
     90   tess->lonelyTriList = NULL;
     91 
     92   for( f = mesh->fHead.next; f != &mesh->fHead; f = f->next ) {
     93     f->marked = FALSE;
     94   }
     95   for( f = mesh->fHead.next; f != &mesh->fHead; f = f->next ) {
     96 
     97     /* We examine all faces in an arbitrary order.  Whenever we find
     98      * an unprocessed face F, we output a group of faces including F
     99      * whose size is maximum.
    100      */
    101     if( f->inside && ! f->marked ) {
    102       RenderMaximumFaceGroup( tess, f );
    103       assert( f->marked );
    104     }
    105   }
    106   if( tess->lonelyTriList != NULL ) {
    107     RenderLonelyTriangles( tess, tess->lonelyTriList );
    108     tess->lonelyTriList = NULL;
    109   }
    110 }
    111 
    112 
    113 static void RenderMaximumFaceGroup( GLUtesselator *tess, GLUface *fOrig )
    114 {
    115   /* We want to find the largest triangle fan or strip of unmarked faces
    116    * which includes the given face fOrig.  There are 3 possible fans
    117    * passing through fOrig (one centered at each vertex), and 3 possible
    118    * strips (one for each CCW permutation of the vertices).  Our strategy
    119    * is to try all of these, and take the primitive which uses the most
    120    * triangles (a greedy approach).
    121    */
    122   GLUhalfEdge *e = fOrig->anEdge;
    123   struct FaceCount max, newFace;
    124 
    125   max.size = 1;
    126   max.eStart = e;
    127   max.render = &RenderTriangle;
    128 
    129   if( ! tess->flagBoundary ) {
    130     newFace = MaximumFan( e ); if( newFace.size > max.size ) { max = newFace; }
    131     newFace = MaximumFan( e->Lnext ); if( newFace.size > max.size ) { max = newFace; }
    132     newFace = MaximumFan( e->Lprev ); if( newFace.size > max.size ) { max = newFace; }
    133 
    134     newFace = MaximumStrip( e ); if( newFace.size > max.size ) { max = newFace; }
    135     newFace = MaximumStrip( e->Lnext ); if( newFace.size > max.size ) { max = newFace; }
    136     newFace = MaximumStrip( e->Lprev ); if( newFace.size > max.size ) { max = newFace; }
    137   }
    138   (*(max.render))( tess, max.eStart, max.size );
    139 }
    140 
    141 
    142 /* Macros which keep track of faces we have marked temporarily, and allow
    143  * us to backtrack when necessary.  With triangle fans, this is not
    144  * really necessary, since the only awkward case is a loop of triangles
    145  * around a single origin vertex.  However with strips the situation is
    146  * more complicated, and we need a general tracking method like the
    147  * one here.
    148  */
    149 #define Marked(f)	(! (f)->inside || (f)->marked)
    150 
    151 #define AddToTrail(f,t)	((f)->trail = (t), (t) = (f), (f)->marked = TRUE)
    152 
    153 #define FreeTrail(t)	do { \
    154 			  while( (t) != NULL ) { \
    155 			    (t)->marked = FALSE; t = (t)->trail; \
    156 			  } \
    157 			} while(0) /* absorb trailing semicolon */
    158 
    159 
    160 
    161 static struct FaceCount MaximumFan( GLUhalfEdge *eOrig )
    162 {
    163   /* eOrig->Lface is the face we want to render.  We want to find the size
    164    * of a maximal fan around eOrig->Org.  To do this we just walk around
    165    * the origin vertex as far as possible in both directions.
    166    */
    167   struct FaceCount newFace = { 0, NULL, &RenderFan };
    168   GLUface *trail = NULL;
    169   GLUhalfEdge *e;
    170 
    171   for( e = eOrig; ! Marked( e->Lface ); e = e->Onext ) {
    172     AddToTrail( e->Lface, trail );
    173     ++newFace.size;
    174   }
    175   for( e = eOrig; ! Marked( e->Rface ); e = e->Oprev ) {
    176     AddToTrail( e->Rface, trail );
    177     ++newFace.size;
    178   }
    179   newFace.eStart = e;
    180   /*LINTED*/
    181   FreeTrail( trail );
    182   return newFace;
    183 }
    184 
    185 
    186 #define IsEven(n)	(((n) & 1) == 0)
    187 
    188 static struct FaceCount MaximumStrip( GLUhalfEdge *eOrig )
    189 {
    190   /* Here we are looking for a maximal strip that contains the vertices
    191    * eOrig->Org, eOrig->Dst, eOrig->Lnext->Dst (in that order or the
    192    * reverse, such that all triangles are oriented CCW).
    193    *
    194    * Again we walk forward and backward as far as possible.  However for
    195    * strips there is a twist: to get CCW orientations, there must be
    196    * an *even* number of triangles in the strip on one side of eOrig.
    197    * We walk the strip starting on a side with an even number of triangles;
    198    * if both side have an odd number, we are forced to shorten one side.
    199    */
    200   struct FaceCount newFace = { 0, NULL, &RenderStrip };
    201   long headSize = 0, tailSize = 0;
    202   GLUface *trail = NULL;
    203   GLUhalfEdge *e, *eTail, *eHead;
    204 
    205   for( e = eOrig; ! Marked( e->Lface ); ++tailSize, e = e->Onext ) {
    206     AddToTrail( e->Lface, trail );
    207     ++tailSize;
    208     e = e->Dprev;
    209     if( Marked( e->Lface )) break;
    210     AddToTrail( e->Lface, trail );
    211   }
    212   eTail = e;
    213 
    214   for( e = eOrig; ! Marked( e->Rface ); ++headSize, e = e->Dnext ) {
    215     AddToTrail( e->Rface, trail );
    216     ++headSize;
    217     e = e->Oprev;
    218     if( Marked( e->Rface )) break;
    219     AddToTrail( e->Rface, trail );
    220   }
    221   eHead = e;
    222 
    223   newFace.size = tailSize + headSize;
    224   if( IsEven( tailSize )) {
    225     newFace.eStart = eTail->Sym;
    226   } else if( IsEven( headSize )) {
    227     newFace.eStart = eHead;
    228   } else {
    229     /* Both sides have odd length, we must shorten one of them.  In fact,
    230      * we must start from eHead to guarantee inclusion of eOrig->Lface.
    231      */
    232     --newFace.size;
    233     newFace.eStart = eHead->Onext;
    234   }
    235   /*LINTED*/
    236   FreeTrail( trail );
    237   return newFace;
    238 }
    239 
    240 
    241 static void RenderTriangle( GLUtesselator *tess, GLUhalfEdge *e, long size )
    242 {
    243   /* Just add the triangle to a triangle list, so we can render all
    244    * the separate triangles at once.
    245    */
    246   assert( size == 1 );
    247   AddToTrail( e->Lface, tess->lonelyTriList );
    248 }
    249 
    250 
    251 static void RenderLonelyTriangles( GLUtesselator *tess, GLUface *f )
    252 {
    253   /* Now we render all the separate triangles which could not be
    254    * grouped into a triangle fan or strip.
    255    */
    256   GLUhalfEdge *e;
    257   int newState;
    258   int edgeState = -1;	/* force edge state output for first vertex */
    259 
    260   CALL_BEGIN_OR_BEGIN_DATA( GL_TRIANGLES );
    261 
    262   for( ; f != NULL; f = f->trail ) {
    263     /* Loop once for each edge (there will always be 3 edges) */
    264 
    265     e = f->anEdge;
    266     do {
    267       if( tess->flagBoundary ) {
    268 	/* Set the "edge state" to TRUE just before we output the
    269 	 * first vertex of each edge on the polygon boundary.
    270 	 */
    271 	newState = ! e->Rface->inside;
    272 	if( edgeState != newState ) {
    273 	  edgeState = newState;
    274           CALL_EDGE_FLAG_OR_EDGE_FLAG_DATA( edgeState );
    275 	}
    276       }
    277       CALL_VERTEX_OR_VERTEX_DATA( e->Org->data );
    278 
    279       e = e->Lnext;
    280     } while( e != f->anEdge );
    281   }
    282   CALL_END_OR_END_DATA();
    283 }
    284 
    285 
    286 static void RenderFan( GLUtesselator *tess, GLUhalfEdge *e, long size )
    287 {
    288   /* Render as many CCW triangles as possible in a fan starting from
    289    * edge "e".  The fan *should* contain exactly "size" triangles
    290    * (otherwise we've goofed up somewhere).
    291    */
    292   CALL_BEGIN_OR_BEGIN_DATA( GL_TRIANGLE_FAN );
    293   CALL_VERTEX_OR_VERTEX_DATA( e->Org->data );
    294   CALL_VERTEX_OR_VERTEX_DATA( e->Dst->data );
    295 
    296   while( ! Marked( e->Lface )) {
    297     e->Lface->marked = TRUE;
    298     --size;
    299     e = e->Onext;
    300     CALL_VERTEX_OR_VERTEX_DATA( e->Dst->data );
    301   }
    302 
    303   assert( size == 0 );
    304   CALL_END_OR_END_DATA();
    305 }
    306 
    307 
    308 static void RenderStrip( GLUtesselator *tess, GLUhalfEdge *e, long size )
    309 {
    310   /* Render as many CCW triangles as possible in a strip starting from
    311    * edge "e".  The strip *should* contain exactly "size" triangles
    312    * (otherwise we've goofed up somewhere).
    313    */
    314   CALL_BEGIN_OR_BEGIN_DATA( GL_TRIANGLE_STRIP );
    315   CALL_VERTEX_OR_VERTEX_DATA( e->Org->data );
    316   CALL_VERTEX_OR_VERTEX_DATA( e->Dst->data );
    317 
    318   while( ! Marked( e->Lface )) {
    319     e->Lface->marked = TRUE;
    320     --size;
    321     e = e->Dprev;
    322     CALL_VERTEX_OR_VERTEX_DATA( e->Org->data );
    323     if( Marked( e->Lface )) break;
    324 
    325     e->Lface->marked = TRUE;
    326     --size;
    327     e = e->Onext;
    328     CALL_VERTEX_OR_VERTEX_DATA( e->Dst->data );
    329   }
    330 
    331   assert( size == 0 );
    332   CALL_END_OR_END_DATA();
    333 }
    334 
    335 
    336 /************************ Boundary contour decomposition ******************/
    337 
    338 /* __gl_renderBoundary( tess, mesh ) takes a mesh, and outputs one
    339  * contour for each face marked "inside".  The rendering output is
    340  * provided as callbacks (see the api).
    341  */
    342 void __gl_renderBoundary( GLUtesselator *tess, GLUmesh *mesh )
    343 {
    344   GLUface *f;
    345   GLUhalfEdge *e;
    346 
    347   for( f = mesh->fHead.next; f != &mesh->fHead; f = f->next ) {
    348     if( f->inside ) {
    349       CALL_BEGIN_OR_BEGIN_DATA( GL_LINE_LOOP );
    350       e = f->anEdge;
    351       do {
    352         CALL_VERTEX_OR_VERTEX_DATA( e->Org->data );
    353 	e = e->Lnext;
    354       } while( e != f->anEdge );
    355       CALL_END_OR_END_DATA();
    356     }
    357   }
    358 }
    359 
    360 
    361 /************************ Quick-and-dirty decomposition ******************/
    362 
    363 #define SIGN_INCONSISTENT 2
    364 
    365 static int ComputeNormal( GLUtesselator *tess, GLdouble norm[3], int check )
    366 /*
    367  * If check==FALSE, we compute the polygon normal and place it in norm[].
    368  * If check==TRUE, we check that each triangle in the fan from v0 has a
    369  * consistent orientation with respect to norm[].  If triangles are
    370  * consistently oriented CCW, return 1; if CW, return -1; if all triangles
    371  * are degenerate return 0; otherwise (no consistent orientation) return
    372  * SIGN_INCONSISTENT.
    373  */
    374 {
    375   CachedVertex *v0 = tess->cache;
    376   CachedVertex *vn = v0 + tess->cacheCount;
    377   CachedVertex *vc;
    378   GLdouble dot, xc, yc, zc, xp, yp, zp, n[3];
    379   int sign = 0;
    380 
    381   /* Find the polygon normal.  It is important to get a reasonable
    382    * normal even when the polygon is self-intersecting (eg. a bowtie).
    383    * Otherwise, the computed normal could be very tiny, but perpendicular
    384    * to the true plane of the polygon due to numerical noise.  Then all
    385    * the triangles would appear to be degenerate and we would incorrectly
    386    * decompose the polygon as a fan (or simply not render it at all).
    387    *
    388    * We use a sum-of-triangles normal algorithm rather than the more
    389    * efficient sum-of-trapezoids method (used in CheckOrientation()
    390    * in normal.c).  This lets us explicitly reverse the signed area
    391    * of some triangles to get a reasonable normal in the self-intersecting
    392    * case.
    393    */
    394   if( ! check ) {
    395     norm[0] = norm[1] = norm[2] = 0.0;
    396   }
    397 
    398   vc = v0 + 1;
    399   xc = vc->coords[0] - v0->coords[0];
    400   yc = vc->coords[1] - v0->coords[1];
    401   zc = vc->coords[2] - v0->coords[2];
    402   while( ++vc < vn ) {
    403     xp = xc; yp = yc; zp = zc;
    404     xc = vc->coords[0] - v0->coords[0];
    405     yc = vc->coords[1] - v0->coords[1];
    406     zc = vc->coords[2] - v0->coords[2];
    407 
    408     /* Compute (vp - v0) cross (vc - v0) */
    409     n[0] = yp*zc - zp*yc;
    410     n[1] = zp*xc - xp*zc;
    411     n[2] = xp*yc - yp*xc;
    412 
    413     dot = n[0]*norm[0] + n[1]*norm[1] + n[2]*norm[2];
    414     if( ! check ) {
    415       /* Reverse the contribution of back-facing triangles to get
    416        * a reasonable normal for self-intersecting polygons (see above)
    417        */
    418       if( dot >= 0 ) {
    419 	norm[0] += n[0]; norm[1] += n[1]; norm[2] += n[2];
    420       } else {
    421 	norm[0] -= n[0]; norm[1] -= n[1]; norm[2] -= n[2];
    422       }
    423     } else if( dot != 0 ) {
    424       /* Check the new orientation for consistency with previous triangles */
    425       if( dot > 0 ) {
    426 	if( sign < 0 ) return SIGN_INCONSISTENT;
    427 	sign = 1;
    428       } else {
    429 	if( sign > 0 ) return SIGN_INCONSISTENT;
    430 	sign = -1;
    431       }
    432     }
    433   }
    434   return sign;
    435 }
    436 
    437 /* __gl_renderCache( tess ) takes a single contour and tries to render it
    438  * as a triangle fan.  This handles convex polygons, as well as some
    439  * non-convex polygons if we get lucky.
    440  *
    441  * Returns TRUE if the polygon was successfully rendered.  The rendering
    442  * output is provided as callbacks (see the api).
    443  */
    444 GLboolean __gl_renderCache( GLUtesselator *tess )
    445 {
    446   CachedVertex *v0 = tess->cache;
    447   CachedVertex *vn = v0 + tess->cacheCount;
    448   CachedVertex *vc;
    449   GLdouble norm[3];
    450   int sign;
    451 
    452   if( tess->cacheCount < 3 ) {
    453     /* Degenerate contour -- no output */
    454     return TRUE;
    455   }
    456 
    457   norm[0] = tess->normal[0];
    458   norm[1] = tess->normal[1];
    459   norm[2] = tess->normal[2];
    460   if( norm[0] == 0 && norm[1] == 0 && norm[2] == 0 ) {
    461     ComputeNormal( tess, norm, FALSE );
    462   }
    463 
    464   sign = ComputeNormal( tess, norm, TRUE );
    465   if( sign == SIGN_INCONSISTENT ) {
    466     /* Fan triangles did not have a consistent orientation */
    467     return FALSE;
    468   }
    469   if( sign == 0 ) {
    470     /* All triangles were degenerate */
    471     return TRUE;
    472   }
    473 
    474   /* Make sure we do the right thing for each winding rule */
    475   switch( tess->windingRule ) {
    476   case GLU_TESS_WINDING_ODD:
    477   case GLU_TESS_WINDING_NONZERO:
    478     break;
    479   case GLU_TESS_WINDING_POSITIVE:
    480     if( sign < 0 ) return TRUE;
    481     break;
    482   case GLU_TESS_WINDING_NEGATIVE:
    483     if( sign > 0 ) return TRUE;
    484     break;
    485   case GLU_TESS_WINDING_ABS_GEQ_TWO:
    486     return TRUE;
    487   }
    488 
    489   CALL_BEGIN_OR_BEGIN_DATA( tess->boundaryOnly ? GL_LINE_LOOP
    490 			  : (tess->cacheCount > 3) ? GL_TRIANGLE_FAN
    491 			  : GL_TRIANGLES );
    492 
    493   CALL_VERTEX_OR_VERTEX_DATA( v0->data );
    494   if( sign > 0 ) {
    495     for( vc = v0+1; vc < vn; ++vc ) {
    496       CALL_VERTEX_OR_VERTEX_DATA( vc->data );
    497     }
    498   } else {
    499     for( vc = vn-1; vc > v0; --vc ) {
    500       CALL_VERTEX_OR_VERTEX_DATA( vc->data );
    501     }
    502   }
    503   CALL_END_OR_END_DATA();
    504   return TRUE;
    505 }
    506