--- ray/src/rt/ambcomp.c	1991/11/04 11:14:46	1.16
+++ ray/src/rt/ambcomp.c	2005/08/22 16:07:26	2.19
@@ -1,64 +1,68 @@
-/* Copyright (c) 1991 Regents of the University of California */
-
 #ifndef lint
-static char SCCSid[] = "$SunId$ LBL";
+static const char	RCSid[] = "$Id: ambcomp.c,v 2.19 2005/08/22 16:07:26 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;
+void
+inithemi(			/* initialize sampling hemisphere */
+	register AMBHEMI  *hp,
+	COLOR ac,
+	RAY  *r,
+	double  wt
+)
 {
-	if (d1->k < d2->k)
-		return(1);
-	if (d1->k > d2->k)
-		return(-1);
-	return(0);
+	double	d;
+	register int  i;
+					/* set number of divisions */
+	if (ambacc <= FTINY &&
+			wt > (d = 0.8*bright(ac)*r->rweight/(ambdiv*minweight)))
+		wt = d;			/* avoid ray termination */
+	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 */
+	hp->ns = ambssamp * wt + 0.5;
+					/* assign coefficient */
+	copycolor(hp->acoef, ac);
+	d = 1.0/(hp->nt*hp->np);
+	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);
 }
 
 
-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);
-}
-
-
-divsample(dp, h, r)			/* sample a division */
-register AMBSAMP  *dp;
-AMBHEMI  *h;
-RAY  *r;
-{
 	RAY  ar;
 	int  hlist[3];
 	double  spt[2];
@@ -66,17 +70,25 @@ RAY  *r;
 	double  b2;
 	double  phi;
 	register int  i;
-
-	if (rayorigin(&ar, r, AMBIENT, AVGREFL) < 0)
+					/* ambient coefficient for weight */
+	if (ambacc > FTINY)
+		setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
+	else
+		copycolor(ar.rcoef, h->acoef);
+	if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
 		return(-1);
+	if (ambacc > FTINY) {
+		multcolor(ar.rcoef, h->acoef);
+		scalecolor(ar.rcoef, 1./AVGREFL);
+	}
 	hlist[0] = r->rno;
 	hlist[1] = dp->t;
 	hlist[2] = dp->p;
 	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 = cos(phi) * zd;
-	yd = sin(phi) * zd;
+	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] +
@@ -85,7 +97,9 @@ RAY  *r;
 	dimlist[ndims++] = dp->t*h->np + dp->p + 90171;
 	rayvalue(&ar);
 	ndims--;
+	multcolor(ar.rcol, ar.rcoef);	/* apply coefficient */
 	addcolor(dp->v, ar.rcol);
+					/* use rt to improve gradient calc */
 	if (ar.rt > FTINY && ar.rt < FHUGE)
 		dp->r += 1.0/ar.rt;
 					/* (re)initialize error */
@@ -99,12 +113,47 @@ RAY  *r;
 }
 
 
+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, wt, pg, dg)		/* compute ambient component */
-COLOR  acol;
-RAY  *r;
-double  wt;
-FVECT  pg, dg;
+doambient(				/* compute ambient component */
+	COLOR  acol,
+	RAY  *r,
+	double  wt,
+	FVECT  pg,
+	FVECT  dg
+)
 {
 	double  b, d;
 	AMBHEMI  hemi;
@@ -112,19 +161,22 @@ FVECT  pg, dg;
 	AMBSAMP  dnew;
 	register AMBSAMP  *dp;
 	double  arad;
-	int  ndivs, ns;
+	int  divcnt;
 	register int  i, j;
-					/* initialize color */
-	setcolor(acol, 0.0, 0.0, 0.0);
 					/* initialize hemisphere */
-	inithemi(&hemi, r, wt);
-	ndivs = hemi.nt * hemi.np;
-	if (ndivs == 0)
+	inithemi(&hemi, acol, r, wt);
+	divcnt = hemi.nt * hemi.np;
+					/* initialize */
+	if (pg != NULL)
+		pg[0] = pg[1] = pg[2] = 0.0;
+	if (dg != NULL)
+		dg[0] = dg[1] = dg[2] = 0.0;
+	setcolor(acol, 0.0, 0.0, 0.0);
+	if (divcnt == 0)
 		return(0.0);
-					/* set number of super-samples */
-	ns = ambssamp * wt + 0.5;
-	if (ns > 0 || pg != NULL || dg != NULL) {
-		div = (AMBSAMP *)malloc(ndivs*sizeof(AMBSAMP));
+					/* allocate super-samples */
+	if (hemi.ns > 0 || pg != NULL || dg != NULL) {
+		div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP));
 		if (div == NULL)
 			error(SYSTEM, "out of memory in doambient");
 	} else
@@ -133,44 +185,57 @@ FVECT  pg, dg;
 	arad = 0.0;
 	if ((dp = div) == NULL)
 		dp = &dnew;
+	divcnt = 0;
 	for (i = 0; i < hemi.nt; i++)
 		for (j = 0; j < hemi.np; j++) {
 			dp->t = i; dp->p = j;
 			setcolor(dp->v, 0.0, 0.0, 0.0);
 			dp->r = 0.0;
 			dp->n = 0;
-			if (divsample(dp, &hemi, r) < 0)
-				goto oopsy;
+			if (divsample(dp, &hemi, r) < 0) {
+				if (div != NULL)
+					dp++;
+				continue;
+			}
+			arad += dp->r;
+			divcnt++;
 			if (div != NULL)
 				dp++;
-			else {
+			else
 				addcolor(acol, dp->v);
-				arad += dp->r;
-			}
 		}
-	if (ns > 0) {			/* perform super-sampling */
+	if (!divcnt)
+		return(0.0);		/* no samples taken */
+	if (divcnt < hemi.nt*hemi.np) {
+		pg = dg = NULL;		/* incomplete sampling */
+		hemi.ns = 0;
+	} else if (arad > FTINY && divcnt/arad < minarad) {
+		hemi.ns = 0;		/* close enough */
+	} else if (hemi.ns > 0) {	/* else perform super-sampling? */
 		comperrs(div, &hemi);			/* compute errors */
-		qsort(div, ndivs, sizeof(AMBSAMP), ambcmp);	/* sort divs */
+		qsort(div, divcnt, sizeof(AMBSAMP), ambcmp);	/* sort divs */
 						/* super-sample */
-		for (i = ns; i > 0; i--) {
-			copystruct(&dnew, div);
-			if (divsample(&dnew, &hemi, r) < 0)
-				goto oopsy;
-							/* reinsert */
-			dp = div;
-			j = ndivs < i ? ndivs : i;
+		for (i = hemi.ns; i > 0; i--) {
+			dnew = *div;
+			if (divsample(&dnew, &hemi, r) < 0) {
+				dp++;
+				continue;
+			}
+			dp = div;		/* reinsert */
+			j = divcnt < i ? divcnt : i;
 			while (--j > 0 && dnew.k < dp[1].k) {
-				copystruct(dp, dp+1);
+				*dp = *(dp+1);
 				dp++;
 			}
-			copystruct(dp, &dnew);
+			*dp = dnew;
 		}
 		if (pg != NULL || dg != NULL)	/* restore order */
-			qsort(div, ndivs, sizeof(AMBSAMP), ambnorm);
+			qsort(div, divcnt, sizeof(AMBSAMP), ambnorm);
 	}
 					/* compute returned values */
 	if (div != NULL) {
-		for (i = ndivs, dp = div; i-- > 0; dp++) {
+		arad = 0.0;		/* note: divcnt may be < nt*np */
+		for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) {
 			arad += dp->r;
 			if (dp->n > 1) {
 				b = 1.0/dp->n;
@@ -182,7 +247,7 @@ FVECT  pg, dg;
 		}
 		b = bright(acol);
 		if (b > FTINY) {
-			b = ndivs/b;
+			b = 1.0/b;	/* compute & normalize gradient(s) */
 			if (pg != NULL) {
 				posgradient(pg, div, &hemi);
 				for (i = 0; i < 3; i++)
@@ -193,25 +258,13 @@ FVECT  pg, dg;
 				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((char *)div);
+		free((void *)div);
 	}
-	b = 1.0/ndivs;
-	scalecolor(acol, b);
 	if (arad <= FTINY)
 		arad = maxarad;
-	else {
-		arad = (ndivs+ns)/arad;
-		if (arad > maxarad)
-			arad = maxarad;
-	}
+	else
+		arad = (divcnt+hemi.ns)/arad;
 	if (pg != NULL) {		/* reduce radius if gradient large */
 		d = DOT(pg,pg);
 		if (d*arad*arad > 1.0)
@@ -225,46 +278,18 @@ FVECT  pg, dg;
 				pg[i] *= d;
 		}
 	}
-	return(arad/sqrt(wt));
-oopsy:
-	if (div != NULL)
-		free((char *)div);
-	return(0.0);
+	if ((arad /= sqrt(wt)) > maxarad)
+		arad = maxarad;
+	return(arad);
 }
 
 
-inithemi(hp, r, wt)		/* initialize sampling hemisphere */
-register AMBHEMI  *hp;
-RAY  *r;
-double  wt;
+void
+comperrs(			/* compute initial error estimates */
+	AMBSAMP  *da,	/* assumes standard ordering */
+	register AMBHEMI  *hp
+)
 {
-	register int  i;
-					/* set number of divisions */
-	if (wt < (.25*PI)/ambdiv+FTINY) {
-		hp->nt = hp->np = 0;
-		return;			/* zero samples */
-	}
-	hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
-	hp->np = PI * hp->nt + 0.5;
-					/* 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);
-}
-
-
-comperrs(da, hp)		/* compute initial error estimates */
-AMBSAMP  *da;		/* assumes standard ordering */
-register AMBHEMI  *hp;
-{
 	double  b, b2;
 	int  i, j;
 	register AMBSAMP  *dp;
@@ -311,13 +336,15 @@ 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;
@@ -326,6 +353,7 @@ 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)
@@ -336,38 +364,42 @@ AMBHEMI  *hp;
 			if (i > 0) {
 				d = dp[-hp->np].r;
 				if (dp[0].r > d) d = dp[0].r;
-				d *= 1.0 - (double)i/hp->nt;	/* cos(t)^2 */
+							/* 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->np;
-			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])/PI;
+		gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/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;
@@ -384,13 +416,14 @@ AMBHEMI  *hp;
 				error(CONSISTENCY,
 					"division order in dirgradient");
 #endif
-			mag += sqrt((i+.5)/hp->nt)*bright(dp->v);  /* sin(t) */
+							/* 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])*PI/(hp->nt*hp->np);
+		gv[i] = xd*hp->ux[i] + yd*hp->uy[i];
 }