--- ray/src/rt/ambcomp.c	2014/05/02 21:58:50	2.46
+++ ray/src/rt/ambcomp.c	2014/05/09 16:05:09	2.55
@@ -1,5 +1,5 @@
 #ifndef lint
-static const char	RCSid[] = "$Id: ambcomp.c,v 2.46 2014/05/02 21:58:50 greg Exp $";
+static const char	RCSid[] = "$Id: ambcomp.c,v 2.55 2014/05/09 16:05:09 greg Exp $";
 #endif
 /*
  * Routines to compute "ambient" values using Monte Carlo
@@ -50,6 +50,7 @@ static const int  adjacent_trifl[8] = {
 
 typedef struct {
 	COLOR	v;		/* hemisphere sample value */
+	float	d;		/* reciprocal distance (1/rt) */
 	FVECT	p;		/* intersection point */
 } AMBSAMP;		/* sample value */
 
@@ -104,7 +105,7 @@ vdb_edge(int db1, int db2)
 	case VDB_xY:	return(db2==VDB_x ? VDB_y : VDB_X);
 	case VDB_Xy:	return(db2==VDB_y ? VDB_x : VDB_Y);
 	}
-	error(INTERNAL, "forbidden diagonal in vdb_edge()");
+	error(CONSISTENCY, "forbidden diagonal in vdb_edge()");
 	return(-1);
 }
 
@@ -207,20 +208,16 @@ ambsample(				/* initial ambient division sample */
 	AMBSAMP	*ap = &ambsam(hp,i,j);
 	RAY	ar;
 					/* generate hemispherical sample */
-	if (!getambsamp(&ar, hp, i, j, 0))
-		goto badsample;
-					/* limit vertex distance */
+	if (!getambsamp(&ar, hp, i, j, 0) || ar.rt <= FTINY) {
+		memset(ap, 0, sizeof(AMBSAMP));
+		return(NULL);
+	}
+	ap->d = 1.0/ar.rt;		/* limit vertex distance */
 	if (ar.rt > 10.0*thescene.cusize)
 		ar.rt = 10.0*thescene.cusize;
-	else if (ar.rt <= FTINY)	/* should never happen! */
-		goto badsample;
 	VSUM(ap->p, ar.rorg, ar.rdir, ar.rt);
 	copycolor(ap->v, ar.rcol);
 	return(ap);
-badsample:
-	setcolor(ap->v, 0., 0., 0.);
-	VCOPY(ap->p, hp->rp->rop);
-	return(NULL);
 }
 
 
@@ -246,25 +243,31 @@ getambdiffs(AMBHEMI *hp)
 			ep[0] += d2;
 			ep[-hp->ns] += d2;
 		}
-		if (j) {		/* from behind */
-			d2 = b - bright(ap[-1].v);
-			d2 *= d2;
-			ep[0] += d2;
-			ep[-1] += d2;
-		}
+		if (!j) continue;
+					/* from behind */
+		d2 = b - bright(ap[-1].v);
+		d2 *= d2;
+		ep[0] += d2;
+		ep[-1] += d2;
+		if (!i) continue;
+					/* diagonal */
+		d2 = b - bright(ap[-hp->ns-1].v);
+		d2 *= d2;
+		ep[0] += d2;
+		ep[-hp->ns-1] += d2;
 	    }
 					/* correct for number of neighbors */
-	earr[0] *= 2.f;
-	earr[hp->ns-1] *= 2.f;
-	earr[(hp->ns-1)*hp->ns] *= 2.f;
-	earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 2.f;
+	earr[0] *= 8./3.;
+	earr[hp->ns-1] *= 8./3.;
+	earr[(hp->ns-1)*hp->ns] *= 8./3.;
+	earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 8./3.;
 	for (i = 1; i < hp->ns-1; i++) {
-		earr[i*hp->ns] *= 4./3.;
-		earr[i*hp->ns + hp->ns-1] *= 4./3.;
+		earr[i*hp->ns] *= 8./5.;
+		earr[i*hp->ns + hp->ns-1] *= 8./5.;
 	}
 	for (j = 1; j < hp->ns-1; j++) {
-		earr[j] *= 4./3.;
-		earr[(hp->ns-1)*hp->ns + j] *= 4./3.;
+		earr[j] *= 8./5.;
+		earr[(hp->ns-1)*hp->ns + j] *= 8./5.;
 	}
 	return(earr);
 }
@@ -275,23 +278,25 @@ static void
 ambsupersamp(double acol[3], AMBHEMI *hp, int cnt)
 {
 	float	*earr = getambdiffs(hp);
-	double	e2sum = 0;
+	double	e2rem = 0;
 	AMBSAMP	*ap;
 	RAY	ar;
-	COLOR	asum;
+	double	asum[3];
 	float	*ep;
-	int	i, j, n;
+	int	i, j, n, nss;
 
 	if (earr == NULL)		/* just skip calc. if no memory */
 		return;
-					/* add up estimated variances */
-	for (ep = earr + hp->ns*hp->ns; ep-- > earr; )
-		e2sum += *ep;
+					/* accumulate estimated variances */
+	for (ep = earr + hp->ns*hp->ns; ep > earr; )
+		e2rem += *--ep;
 	ep = earr;			/* perform super-sampling */
 	for (ap = hp->sa, i = 0; i < hp->ns; i++)
 	    for (j = 0; j < hp->ns; j++, ap++) {
-		int	nss = *ep/e2sum*cnt + frandom();
-		setcolor(asum, 0., 0., 0.);
+		if (e2rem <= FTINY)
+			goto done;	/* nothing left to do */
+		nss = *ep/e2rem*cnt + frandom();
+		asum[0] = asum[1] = asum[2] = 0.0;
 		for (n = 1; n <= nss; n++) {
 			if (!getambsamp(&ar, hp, i, j, n)) {
 				nss = n-1;
@@ -300,14 +305,15 @@ ambsupersamp(double acol[3], AMBHEMI *hp, int cnt)
 			addcolor(asum, ar.rcol);
 		}
 		if (nss) {		/* update returned ambient value */
-			const double	ssf = 1./(nss + 1);
+			const double	ssf = 1./(nss + 1.);
 			for (n = 3; n--; )
-				acol[n] += ssf*colval(asum,n) +
+				acol[n] += ssf*asum[n] +
 						(ssf - 1.)*colval(ap->v,n);
 		}
-		e2sum -= *ep++;		/* update remainders */
+		e2rem -= *ep++;		/* update remainders */
 		cnt -= nss;
 	}
+done:
 	free(earr);
 }
 
@@ -317,51 +323,42 @@ static uby8 *
 vertex_flags(AMBHEMI *hp)
 {
 	uby8	*vflags = (uby8 *)calloc(hp->ns*hp->ns, sizeof(uby8));
-	double	*dist2a = (double *)malloc(sizeof(double)*hp->ns);
 	uby8	*vf;
+	AMBSAMP	*ap;
 	int	i, j;
 
-	if ((vflags == NULL) | (dist2a == NULL))
+	if (vflags == NULL)
 		error(SYSTEM, "out of memory in vertex_flags()");
-	vf = vflags;		/* compute distances along first row */
-	for (j = 0; j < hp->ns; j++) {
-		dist2a[j] = dist2(ambsam(hp,0,j).p, hp->rp->rop);
-		++vf;
-		if (!j) continue;
-		if (dist2a[j] >= dist2a[j-1])
-			vf[0] |= 1<<VDB_x;
+	vf = vflags;
+	ap = hp->sa;		/* compute farthest along first row */
+	for (j = 0; j < hp->ns-1; j++, vf++, ap++)
+		if (ap[0].d <= ap[1].d)
+			vf[0] |= 1<<VDB_X;
 		else
-			vf[-1] |= 1<<VDB_X;
-	}
+			vf[1] |= 1<<VDB_x;
+	++vf; ++ap;
 				/* flag subsequent rows */
 	for (i = 1; i < hp->ns; i++) {
-	    double	d2n = dist2(ambsam(hp,i,0).p, hp->rp->rop);
-	    for (j = 0; j < hp->ns-1; j++) {
-		double	d2 = d2n;
-		if (d2 >= dist2a[j])	/* row before */
+	    for (j = 0; j < hp->ns-1; j++, vf++, ap++) {
+		if (ap[0].d <= ap[-hp->ns].d)	/* row before */
 			vf[0] |= 1<<VDB_y;
 		else
 			vf[-hp->ns] |= 1<<VDB_Y;
-		dist2a[j] = d2n;
-		if (d2 >= dist2a[j+1])	/* diagonal we care about */
+		if (ap[0].d <= ap[1-hp->ns].d)	/* diagonal we care about */
 			vf[0] |= 1<<VDB_Xy;
 		else
 			vf[1-hp->ns] |= 1<<VDB_xY;
-		d2n = dist2(ambsam(hp,i,j+1).p, hp->rp->rop);
-		if (d2 >= d2n)		/* column after */
+		if (ap[0].d <= ap[1].d)		/* column after */
 			vf[0] |= 1<<VDB_X;
 		else
 			vf[1] |= 1<<VDB_x;
-		++vf;
 	    }
-	    if (d2n >= dist2a[j])	/* final column edge */
+	    if (ap[0].d <= ap[-hp->ns].d)	/* final column edge */
 		vf[0] |= 1<<VDB_y;
 	    else
 		vf[-hp->ns] |= 1<<VDB_Y;
-	    dist2a[j] = d2n;
-	    ++vf;
+	    ++vf; ++ap;
 	}
-	free(dist2a);
 	return(vflags);
 }
 
@@ -549,7 +546,7 @@ add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, F
 
 
 /* Compute anisotropic radii and eigenvector directions */
-static int
+static void
 eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3])
 {
 	double	hess2[2][2];
@@ -571,9 +568,10 @@ eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3
 	if (i == 1)			/* double-root (circle) */
 		evalue[1] = evalue[0];
 	if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) |
-			((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) )
-		error(INTERNAL, "bad eigenvalue calculation");
-
+			((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) {
+		ra[0] = ra[1] = maxarad;
+		return;
+	}
 	if (evalue[0] > evalue[1]) {
 		ra[0] = sqrt(sqrt(4.0/evalue[0]));
 		ra[1] = sqrt(sqrt(4.0/evalue[1]));
@@ -730,6 +728,51 @@ ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2])
 }
 
 
+/* Compute potential light leak direction flags for cache value */
+static uint32
+ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1)
+{
+	const double	max_d = 1.0/(minarad*ambacc + 0.001);
+	const double	ang_res = 0.5*PI/(hp->ns-1);
+	const double	ang_step = ang_res/((int)(16/PI*ang_res) + (1+FTINY));
+	double		avg_d = 0;
+	uint32		flgs = 0;
+	int		i, j;
+					/* don't bother for a few samples */
+	if (hp->ns < 12)
+		return(0);
+					/* check distances overhead */
+	for (i = hp->ns*3/4; i-- > hp->ns>>2; )
+	    for (j = hp->ns*3/4; j-- > hp->ns>>2; )
+		avg_d += ambsam(hp,i,j).d;
+	avg_d *= 4.0/(hp->ns*hp->ns);
+	if (avg_d*r0 >= 1.0)		/* ceiling too low for corral? */
+		return(0);
+	if (avg_d >= max_d)		/* insurance */
+		return(0);
+					/* else circle around perimeter */
+	for (i = 0; i < hp->ns; i++)
+	    for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) {
+		AMBSAMP	*ap = &ambsam(hp,i,j);
+		FVECT	vec;
+		double	u, v;
+		double	ang, a1;
+		int	abp;
+		if ((ap->d <= FTINY) | (ap->d >= max_d))
+			continue;	/* too far or too near */
+		VSUB(vec, ap->p, hp->rp->rop);
+		u = DOT(vec, uv[0]) * ap->d;
+		v = DOT(vec, uv[1]) * ap->d;
+		if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= 1.0)
+			continue;	/* occluder outside ellipse */
+		ang = atan2a(v, u);	/* else set direction flags */
+		for (a1 = ang-.5*ang_res; a1 <= ang+.5*ang_res; a1 += ang_step)
+			flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0)));
+	    }
+	return(flgs);
+}
+
+
 int
 doambient(				/* compute ambient component */
 	COLOR	rcol,			/* input/output color */
@@ -738,7 +781,8 @@ doambient(				/* compute ambient component */
 	FVECT	uv[2],			/* returned (optional) */
 	float	ra[2],			/* returned (optional) */
 	float	pg[2],			/* returned (optional) */
-	float	dg[2]			/* returned (optional) */
+	float	dg[2],			/* returned (optional) */
+	uint32	*crlp			/* returned (optional) */
 )
 {
 	AMBHEMI	*hp = inithemi(rcol, r, wt);
@@ -758,6 +802,8 @@ doambient(				/* compute ambient component */
 		pg[0] = pg[1] = 0.0;
 	if (dg != NULL)
 		dg[0] = dg[1] = 0.0;
+	if (crlp != NULL)
+		*crlp = 0;
 					/* sample the hemisphere */
 	acol[0] = acol[1] = acol[2] = 0.0;
 	cnt = 0;
@@ -778,7 +824,7 @@ doambient(				/* compute ambient component */
 		return(-1);		/* return value w/o Hessian */
 	}
 	cnt = ambssamp*wt + 0.5;	/* perform super-sampling? */
-	if (cnt > 0)
+	if (cnt > 8)
 		ambsupersamp(acol, hp, cnt);
 	copycolor(rcol, acol);		/* final indirect irradiance/PI */
 	if ((ra == NULL) & (pg == NULL) & (dg == NULL)) {
@@ -792,6 +838,7 @@ doambient(				/* compute ambient component */
 		K = 1.0;
 		pg = NULL;
 		dg = NULL;
+		crlp = NULL;
 	}
 	ap = hp->sa;			/* relative Y channel from here on... */
 	for (i = hp->ns*hp->ns; i--; ap++)
@@ -827,6 +874,9 @@ doambient(				/* compute ambient component */
 			if (ra[0] > maxarad)
 				ra[0] = maxarad;
 		}
+					/* flag encroached directions */
+		if ((wt >= 0.5-FTINY) & (crlp != NULL))
+			*crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc);
 		if (pg != NULL) {	/* cap gradient if necessary */
 			d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1];
 			if (d > 1.0) {