--- ray/src/rt/ambcomp.c	1991/06/07 13:43:05	1.2
+++ ray/src/rt/ambcomp.c	2005/04/19 01:15:06	2.14
@@ -1,83 +1,90 @@
-/* Copyright (c) 1991 Regents of the University of California */
-
 #ifndef lint
-static char SCCSid[] = "$SunId$ LBL";
+static const char	RCSid[] = "$Id: ambcomp.c,v 2.14 2005/04/19 01:15:06 greg Exp $";
 #endif
-
 /*
  * Routines to compute "ambient" values using Monte Carlo
+ *
+ *  Declarations of external symbols in ambient.h
  */
 
+#include "copyright.h"
+
 #include  "ray.h"
 
 #include  "ambient.h"
 
 #include  "random.h"
 
-typedef struct {
-	short  t, p;		/* theta, phi indices */
-	COLOR  v;		/* value sum */
-	float  r;		/* 1/distance sum */
-	float  k;		/* variance for this division */
-	int  n;			/* number of subsamples */
-}  AMBSAMP;		/* ambient sample division */
 
-typedef struct {
-	FVECT  ux, uy, uz;	/* x, y and z axis directions */
-	short  nt, np;		/* number of theta and phi directions */
-}  AMBHEMI;		/* ambient sample hemisphere */
-
-extern double  sin(), cos(), sqrt();
-
-
-static int
-ambcmp(d1, d2)				/* decreasing order */
-AMBSAMP  *d1, *d2;
+int
+inithemi(			/* initialize sampling hemisphere */
+	register AMBHEMI  *hp,
+	RAY  *r,
+	COLOR ac, 
+	double  wt
+)
 {
-	if (d1->k < d2->k)
-		return(1);
-	if (d1->k > d2->k)
-		return(-1);
-	return(0);
+	int	ns;
+	double	d;
+	register int  i;
+					/* set number of divisions */
+	hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
+	i = ambacc > FTINY ? 3 : 1;	/* minimum number of samples */
+	if (hp->nt < i)
+		hp->nt = i;
+	hp->np = PI * hp->nt + 0.5;
+					/* set number of super-samples */
+	ns = ambssamp * wt + 0.5;
+					/* assign coefficient */
+	d = 1.0/(hp->nt*hp->np + ns);	/* XXX weight not uniform if ns > 0 */
+	copycolor(hp->acoef, ac);
+	scalecolor(hp->acoef, d);
+					/* make axes */
+	VCOPY(hp->uz, r->ron);
+	hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
+	for (i = 0; i < 3; i++)
+		if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
+			break;
+	if (i >= 3)
+		error(CONSISTENCY, "bad ray direction in inithemi");
+	hp->uy[i] = 1.0;
+	fcross(hp->ux, hp->uy, hp->uz);
+	normalize(hp->ux);
+	fcross(hp->uy, hp->uz, hp->ux);
+	return(ns);
 }
 
 
-static int
-ambnorm(d1, d2)				/* standard order */
-AMBSAMP  *d1, *d2;
+int
+divsample(				/* sample a division */
+	register AMBSAMP  *dp,
+	AMBHEMI  *h,
+	RAY  *r
+)
 {
-	register int  c;
-
-	if (c = d1->t - d2->t)
-		return(c);
-	return(d1->p - d2->p);
-}
-
-
-double
-divsample(dp, h, r)			/* sample a division */
-register AMBSAMP  *dp;
-AMBHEMI  *h;
-RAY  *r;
-{
 	RAY  ar;
-	int  hlist[4];
+	int  hlist[3];
+	double  spt[2];
 	double  xd, yd, zd;
 	double  b2;
 	double  phi;
 	register int  i;
-
-	if (rayorigin(&ar, r, AMBIENT, 0.5) < 0)
-		return(0.0);
+					/* assign coefficient */
+	if (ambacc <= FTINY)		/* no storage, so report accurately */
+		copycolor(ar.rcoef, h->acoef);
+	else				/* else lie for sake of cache */
+		setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
+	if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
+		return(-1);
+	copycolor(ar.rcoef, h->acoef);	/* correct coefficient rtrace output */
 	hlist[0] = r->rno;
 	hlist[1] = dp->t;
 	hlist[2] = dp->p;
-	hlist[3] = 0;
-	zd = sqrt((dp->t+urand(ilhash(hlist,4)+dp->n))/h->nt);
-	hlist[3] = 1;
-	phi = 2.0*PI * (dp->p+urand(ilhash(hlist,4)+dp->n))/h->np;
-	xd = cos(phi) * zd;
-	yd = sin(phi) * zd;
+	multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n));
+	zd = sqrt((dp->t + spt[0])/h->nt);
+	phi = 2.0*PI * (dp->p + spt[1])/h->np;
+	xd = tcos(phi) * zd;
+	yd = tsin(phi) * zd;
 	zd = sqrt(1.0 - zd*zd);
 	for (i = 0; i < 3; i++)
 		ar.rdir[i] =	xd*h->ux[i] +
@@ -87,8 +94,9 @@ RAY  *r;
 	rayvalue(&ar);
 	ndims--;
 	addcolor(dp->v, ar.rcol);
-	if (ar.rot < FHUGE)
-		dp->r += 1.0/ar.rot;
+					/* use rt to improve gradient calc */
+	if (ar.rt > FTINY && ar.rt < FHUGE)
+		dp->r += 1.0/ar.rt;
 					/* (re)initialize error */
 	if (dp->n++) {
 		b2 = bright(dp->v)/dp->n - bright(ar.rcol);
@@ -96,15 +104,52 @@ RAY  *r;
 		dp->k = b2/(dp->n*dp->n);
 	} else
 		dp->k = 0.0;
-	return(ar.rot);
+	return(0);
 }
 
 
+static int
+ambcmp(					/* decreasing order */
+	const void *p1,
+	const void *p2
+)
+{
+	const AMBSAMP	*d1 = (const AMBSAMP *)p1;
+	const AMBSAMP	*d2 = (const AMBSAMP *)p2;
+
+	if (d1->k < d2->k)
+		return(1);
+	if (d1->k > d2->k)
+		return(-1);
+	return(0);
+}
+
+
+static int
+ambnorm(				/* standard order */
+	const void *p1,
+	const void *p2
+)
+{
+	const AMBSAMP	*d1 = (const AMBSAMP *)p1;
+	const AMBSAMP	*d2 = (const AMBSAMP *)p2;
+	register int	c;
+
+	if ( (c = d1->t - d2->t) )
+		return(c);
+	return(d1->p - d2->p);
+}
+
+
 double
-doambient(acol, r, pg, dg)		/* compute ambient component */
-COLOR  acol;
-RAY  *r;
-FVECT  pg, dg;
+doambient(				/* compute ambient component */
+	COLOR  acol,
+	RAY  *r,
+	COLOR  ac,
+	double  wt,
+	FVECT  pg,
+	FVECT  dg
+)
 {
 	double  b, d;
 	AMBHEMI  hemi;
@@ -117,12 +162,11 @@ FVECT  pg, dg;
 					/* initialize color */
 	setcolor(acol, 0.0, 0.0, 0.0);
 					/* initialize hemisphere */
-	inithemi(&hemi, r);
+	ns = inithemi(&hemi, r, ac, wt);
 	ndivs = hemi.nt * hemi.np;
 	if (ndivs == 0)
 		return(0.0);
-					/* set number of super-samples */
-	ns = ambssamp * r->rweight + 0.5;
+					/* allocate super-samples */
 	if (ns > 0 || pg != NULL || dg != NULL) {
 		div = (AMBSAMP *)malloc(ndivs*sizeof(AMBSAMP));
 		if (div == NULL)
@@ -139,108 +183,107 @@ FVECT  pg, dg;
 			setcolor(dp->v, 0.0, 0.0, 0.0);
 			dp->r = 0.0;
 			dp->n = 0;
-			if ((d = divsample(dp, &hemi, r)) == 0.0)
+			if (divsample(dp, &hemi, r) < 0)
 				goto oopsy;
+			arad += dp->r;
 			if (div != NULL)
 				dp++;
-			else {
+			else
 				addcolor(acol, dp->v);
-				arad += dp->r;
-			}
 		}
-	if (ns > 0) {			/* perform super-sampling */
-		comperrs(div, hemi);			/* compute errors */
+	if (ns > 0 && arad > FTINY && ndivs/arad < minarad)
+		ns = 0;			/* close enough */
+	else if (ns > 0) {		/* else perform super-sampling */
+		comperrs(div, &hemi);			/* compute errors */
 		qsort(div, ndivs, sizeof(AMBSAMP), ambcmp);	/* sort divs */
-		dp = div + ndivs;			/* skim excess */
-		for (i = ndivs; i > ns; i--) {
-			dp--;
-			addcolor(acol, dp->v);
-			arad += dp->r;
-		}
 						/* super-sample */
 		for (i = ns; i > 0; i--) {
-			copystruct(&dnew, div);
-			if ((d = divsample(&dnew, &hemi)) == 0.0)
+			dnew = *div;
+			if (divsample(&dnew, &hemi, r) < 0)
 				goto oopsy;
-			if (d < FHUGE)
-				arad += 1.0 / d;
 							/* reinsert */
 			dp = div;
 			j = ndivs < i ? ndivs : i;
 			while (--j > 0 && dnew.k < dp[1].k) {
-				copystruct(dp, dp+1);
+				*dp = *(dp+1);
 				dp++;
 			}
-			copystruct(dp, &dnew);
-							/* extract darkest */
-			if (i <= ndivs) {
-				dp = div + i-1;
-				arad += dp->r;
-				if (dp->n > 1) {
-					b = 1.0/dp->n;
-					scalecolor(dp->v, b);
-					dp->r *= b;
-					dp->n = 1;
-				}
-				addcolor(acol, dp->v);
-			}
+			*dp = dnew;
 		}
 		if (pg != NULL || dg != NULL)	/* restore order */
 			qsort(div, ndivs, sizeof(AMBSAMP), ambnorm);
 	}
 					/* compute returned values */
-	if (pg != NULL)
-		posgradient(pg, div, &hemi);
-	if (dg != NULL)
-		dirgradient(dg, div, &hemi);
-	if (div != NULL)
-		free((char *)div);
+	if (div != NULL) {
+		arad = 0.0;
+		for (i = ndivs, dp = div; i-- > 0; dp++) {
+			arad += dp->r;
+			if (dp->n > 1) {
+				b = 1.0/dp->n;
+				scalecolor(dp->v, b);
+				dp->r *= b;
+				dp->n = 1;
+			}
+			addcolor(acol, dp->v);
+		}
+		b = bright(acol);
+		if (b > FTINY) {
+			b = ndivs/b;
+			if (pg != NULL) {
+				posgradient(pg, div, &hemi);
+				for (i = 0; i < 3; i++)
+					pg[i] *= b;
+			}
+			if (dg != NULL) {
+				dirgradient(dg, div, &hemi);
+				for (i = 0; i < 3; i++)
+					dg[i] *= b;
+			}
+		} else {
+			if (pg != NULL)
+				for (i = 0; i < 3; i++)
+					pg[i] = 0.0;
+			if (dg != NULL)
+				for (i = 0; i < 3; i++)
+					dg[i] = 0.0;
+		}
+		free((void *)div);
+	}
 	b = 1.0/ndivs;
 	scalecolor(acol, b);
 	if (arad <= FTINY)
-		arad = FHUGE;
+		arad = maxarad;
 	else
 		arad = (ndivs+ns)/arad;
-	if (arad > maxarad)
-		arad = maxarad;
-	else if (arad < minarad)
+	if (pg != NULL) {		/* reduce radius if gradient large */
+		d = DOT(pg,pg);
+		if (d*arad*arad > 1.0)
+			arad = 1.0/sqrt(d);
+	}
+	if (arad < minarad) {
 		arad = minarad;
-	arad /= sqrt(r->rweight);
+		if (pg != NULL && d*arad*arad > 1.0) {	/* cap gradient */
+			d = 1.0/arad/sqrt(d);
+			for (i = 0; i < 3; i++)
+				pg[i] *= d;
+		}
+	}
+	if ((arad /= sqrt(wt)) > maxarad)
+		arad = maxarad;
 	return(arad);
 oopsy:
 	if (div != NULL)
-		free((char *)div);
+		free((void *)div);
 	return(0.0);
 }
 
 
-inithemi(hp, r)			/* initialize sampling hemisphere */
-register AMBHEMI  *hp;
-RAY  *r;
+void
+comperrs(			/* compute initial error estimates */
+	AMBSAMP  *da,	/* assumes standard ordering */
+	register AMBHEMI  *hp
+)
 {
-	register int  i;
-					/* set number of divisions */
-	hp->nt = sqrt(ambdiv * r->rweight * 0.5) + 0.5;
-	hp->np = 2 * hp->nt;
-					/* make axes */
-	VCOPY(hp->uz, r->ron);
-	hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
-	for (i = 0; i < 3; i++)
-		if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
-			break;
-	if (i >= 3)
-		error(CONSISTENCY, "bad ray direction in inithemi");
-	hp->uy[i] = 1.0;
-	fcross(hp->ux, hp->uz, hp->uy);
-	normalize(hp->ux);
-	fcross(hp->uy, hp->ux, hp->uz);
-}
-
-
-comperrs(da, hp)		/* compute initial error estimates */
-AMBSAMP  *da;		/* assumes standard ordering */
-register AMBHEMI  *hp;
-{
 	double  b, b2;
 	int  i, j;
 	register AMBSAMP  *dp;
@@ -248,6 +291,11 @@ register AMBHEMI  *hp;
 	dp = da;
 	for (i = 0; i < hp->nt; i++)
 		for (j = 0; j < hp->np; j++) {
+#ifdef  DEBUG
+			if (dp->t != i || dp->p != j)
+				error(CONSISTENCY,
+					"division order in comperrs");
+#endif
 			b = bright(dp[0].v);
 			if (i > 0) {		/* from above */
 				b2 = bright(dp[-hp->np].v) - b;
@@ -260,12 +308,11 @@ register AMBHEMI  *hp;
 				b2 *= b2 * 0.25;
 				dp[0].k += b2;
 				dp[-1].k += b2;
-			}
-			if (j == hp->np-1) {	/* around */
-				b2 = bright(dp[-(hp->np-1)].v) - b;
+			} else {		/* around */
+				b2 = bright(dp[hp->np-1].v) - b;
 				b2 *= b2 * 0.25;
 				dp[0].k += b2;
-				dp[-(hp->np-1)].k += b2;
+				dp[hp->np-1].k += b2;
 			}
 			dp++;
 		}
@@ -283,13 +330,15 @@ register AMBHEMI  *hp;
 }
 
 
-posgradient(gv, da, hp)				/* compute position gradient */
-FVECT  gv;
-AMBSAMP  *da;			/* assumes standard ordering */
-AMBHEMI  *hp;
+void
+posgradient(					/* compute position gradient */
+	FVECT  gv,
+	AMBSAMP  *da,			/* assumes standard ordering */
+	register AMBHEMI  *hp
+)
 {
 	register int  i, j;
-	double  b, d;
+	double  nextsine, lastsine, b, d;
 	double  mag0, mag1;
 	double  phi, cosp, sinp, xd, yd;
 	register AMBSAMP  *dp;
@@ -298,6 +347,7 @@ AMBHEMI  *hp;
 	for (j = 0; j < hp->np; j++) {
 		dp = da + j;
 		mag0 = mag1 = 0.0;
+		lastsine = 0.0;
 		for (i = 0; i < hp->nt; i++) {
 #ifdef  DEBUG
 			if (dp->t != i || dp->p != j)
@@ -308,38 +358,42 @@ AMBHEMI  *hp;
 			if (i > 0) {
 				d = dp[-hp->np].r;
 				if (dp[0].r > d) d = dp[0].r;
-				d *= 1.0 - sqrt((double)i/hp->nt);
+							/* sin(t)*cos(t)^2 */
+				d *= lastsine * (1.0 - (double)i/hp->nt);
 				mag0 += d*(b - bright(dp[-hp->np].v));
 			}
+			nextsine = sqrt((double)(i+1)/hp->nt);
 			if (j > 0) {
 				d = dp[-1].r;
 				if (dp[0].r > d) d = dp[0].r;
-				mag1 += d*(b - bright(dp[-1].v));
+				mag1 += d * (nextsine - lastsine) *
+						(b - bright(dp[-1].v));
 			} else {
 				d = dp[hp->np-1].r;
 				if (dp[0].r > d) d = dp[0].r;
-				mag1 += d*(b - bright(dp[hp->np-1].v));
+				mag1 += d * (nextsine - lastsine) *
+						(b - bright(dp[hp->np-1].v));
 			}
 			dp += hp->np;
+			lastsine = nextsine;
 		}
-		if (hp->nt > 1) {
-			mag0 /= (double)(hp->nt-1);
-			mag1 /= (double)hp->nt;
-		}
+		mag0 *= 2.0*PI / hp->np;
 		phi = 2.0*PI * (double)j/hp->np;
-		cosp = cos(phi); sinp = sin(phi);
+		cosp = tcos(phi); sinp = tsin(phi);
 		xd += mag0*cosp - mag1*sinp;
 		yd += mag0*sinp + mag1*cosp;
 	}
 	for (i = 0; i < 3; i++)
-		gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])/hp->np;
+		gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])/PI;
 }
 
 
-dirgradient(gv, da, hp)				/* compute direction gradient */
-FVECT  gv;
-AMBSAMP  *da;			/* assumes standard ordering */
-AMBHEMI  *hp;
+void
+dirgradient(					/* compute direction gradient */
+	FVECT  gv,
+	AMBSAMP  *da,			/* assumes standard ordering */
+	register AMBHEMI  *hp
+)
 {
 	register int  i, j;
 	double  mag;
@@ -356,12 +410,13 @@ AMBHEMI  *hp;
 				error(CONSISTENCY,
 					"division order in dirgradient");
 #endif
-			mag += sqrt((i+.5)/hp->nt)*bright(dp->v);
+							/* tan(t) */
+			mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0);
 			dp += hp->np;
 		}
 		phi = 2.0*PI * (j+.5)/hp->np + PI/2.0;
-		xd += mag * cos(phi);
-		yd += mag * sin(phi);
+		xd += mag * tcos(phi);
+		yd += mag * tsin(phi);
 	}
 	for (i = 0; i < 3; i++)
 		gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])/(hp->nt*hp->np);