ViewVC Help

View File | Revision Log | Show Annotations | Download File | Root Listing

root/radiance/ray/src/cv/pabopto2xml.c

Comparing ray/src/cv/pabopto2xml.c (file contents):
Revision 2.7 by greg, Sun Sep 2 15:33:15 2012 UTC vs.
Revision 2.17 by greg, Tue Oct 16 22:00:51 2012 UTC

#	Line 8 | Line 8 | static const char RCSid[] = "$Id$";
8		*      G.Ward
9		*/
10
11	+	#ifndef _WIN32
12	+	#include <unistd.h>
13	+	#include <sys/wait.h>
14	+	#include <sys/mman.h>
15	+	#endif
16		#define _USE_MATH_DEFINES
17		#include <stdio.h>
18		#include <stdlib.h>
#	Line 16 | Line 21 | static const char RCSid[] = "$Id$";
21		#include <math.h>
22		#include "bsdf.h"
23
24	+	#define DEBUG           1
25	+
26		#ifndef GRIDRES
27	<	#define GRIDRES         200             /* max. grid resolution per side */
27	>	#define GRIDRES         200             /* grid resolution per side */
28		#endif
29
30		#define RSCA            2.7             /* radius scaling factor (empirical) */
#	Line 51 | Line 58 | typedef struct s_rbfnode {
58		struct s_rbfnode        *next;          /* next in global RBF list */
59		MIGRATION               *ejl;           /* edge list for this vertex */
60		FVECT                   invec;          /* incident vector direction */
61	+	double                  vtotal;         /* volume for normalization */
62		int                     nrbf;           /* number of RBFs */
63		RBFVAL                  rbfa[1];        /* RBF array (extends struct) */
64	<	} RBFLIST;                      /* RBF representation of DSF @ 1 incidence */
64	>	} RBFNODE;                      /* RBF representation of DSF @ 1 incidence */
65
66		/* our loaded grid for this incident angle */
67	<	static double   theta_in_deg, phi_in_deg;
68	<	static GRIDVAL  dsf_grid[GRIDRES][GRIDRES];
67	>	static double           theta_in_deg, phi_in_deg;
68	>	static GRIDVAL          dsf_grid[GRIDRES][GRIDRES];
69
70	+	/* all incident angles in-plane so far? */
71	+	static int              single_plane_incident = -1;
72	+
73	+	/* input/output orientations */
74	+	static int              input_orient = 0;
75	+	static int              output_orient = 0;
76	+
77		/* processed incident DSF measurements */
78	<	static RBFLIST          *dsf_list = NULL;
78	>	static RBFNODE          *dsf_list = NULL;
79
80	<	/* edge (linking) matrices */
80	>	/* RBF-linking matrices (edges) */
81		static MIGRATION        *mig_list = NULL;
82
83	<	#define mtxval(m,i,j)   (m)->mtx[(i)*(m)->rbfv[1]->nrbf+(j)]
84	<	#define nextedge(rbf,m) (m)->enxt[(rbf)==(m)->rbfv[1]]
83	>	/* migration edges drawn in raster fashion */
84	>	static MIGRATION        *mig_grid[GRIDRES][GRIDRES];
85
86	+	#define mtx_nrows(m)    ((m)->rbfv[0]->nrbf)
87	+	#define mtx_ncols(m)    ((m)->rbfv[1]->nrbf)
88	+	#define mtx_ndx(m,i,j)  ((i)*mtx_ncols(m) + (j))
89	+	#define is_src(rbf,m)   ((rbf) == (m)->rbfv[0])
90	+	#define is_dest(rbf,m)  ((rbf) == (m)->rbfv[1])
91	+	#define nextedge(rbf,m) (m)->enxt[is_dest(rbf,m)]
92	+	#define opp_rbf(rbf,m)  (m)->rbfv[is_src(rbf,m)]
93	+
94	+	#define round(v)        (int)((v) + .5 - ((v) < -.5))
95	+
96	+	char                    *progname;
97	+
98	+	#ifdef DEBUG                    /* percentage to cull (<0 to turn off) */
99	+	int                     pctcull = -1;
100	+	#else
101	+	int                     pctcull = 90;
102	+	#endif
103	+	/* number of processes to run */
104	+	int                     nprocs = 1;
105	+	/* number of children (-1 in child) */
106	+	int                     nchild = 0;
107	+
108	+	/* sampling order (set by data density) */
109	+	int                     samp_order = 0;
110	+
111	+	/* Compute volume associated with Gaussian lobe */
112	+	static double
113	+	rbf_volume(const RBFVAL *rbfp)
114	+	{
115	+	double  rad = R2ANG(rbfp->crad);
116	+
117	+	return((2.*M_PI) * rbfp->peak * rad*rad);
118	+	}
119	+
120		/* Compute outgoing vector from grid position */
121		static void
122	<	vec_from_pos(FVECT vec, int xpos, int ypos)
122	>	ovec_from_pos(FVECT vec, int xpos, int ypos)
123		{
124		double  uv[2];
125		double  r2;
#	Line 81 | Line 130 | vec_from_pos(FVECT vec, int xpos, int ypos)
130		vec[0] = vec[1] = sqrt(2. - r2);
131		vec[0] *= uv[0];
132		vec[1] *= uv[1];
133	<	vec[2] = 1. - r2;
133	>	vec[2] = output_orient*(1. - r2);
134		}
135
136	<	/* Compute grid position from normalized outgoing vector */
136	>	/* Compute grid position from normalized input/output vector */
137		static void
138		pos_from_vec(int pos[2], const FVECT vec)
139		{
140		double  sq[2];          /* uniform hemispherical projection */
141	<	double  norm = 1./sqrt(1. + vec[2]);
141	>	double  norm = 1./sqrt(1. + fabs(vec[2]));
142
143		SDdisk2square(sq, vec[0]*norm, vec[1]*norm);
144
#	Line 99 | Line 148 | pos_from_vec(int pos[2], const FVECT vec)
148
149		/* Evaluate RBF for DSF at the given normalized outgoing direction */
150		static double
151	<	eval_rbfrep(const RBFLIST *rp, const FVECT outvec)
151	>	eval_rbfrep(const RBFNODE *rp, const FVECT outvec)
152		{
153		double          res = .0;
154		const RBFVAL    *rbfp;
#	Line 107 | Line 156 | eval_rbfrep(const RBFLIST *rp, const FVECT outvec)
156		double          sig2;
157		int             n;
158
159	+	if (rp == NULL)
160	+	return(.0);
161		rbfp = rp->rbfa;
162		for (n = rp->nrbf; n--; rbfp++) {
163	<	vec_from_pos(odir, rbfp->gx, rbfp->gy);
163	>	ovec_from_pos(odir, rbfp->gx, rbfp->gy);
164		sig2 = R2ANG(rbfp->crad);
165		sig2 = (DOT(odir,outvec) - 1.) / (sig2*sig2);
166		if (sig2 > -19.)
#	Line 118 | Line 169 | eval_rbfrep(const RBFLIST *rp, const FVECT outvec)
169		return(res);
170		}
171
172	+	/* Insert a new directional scattering function in our global list */
173	+	static void
174	+	insert_dsf(RBFNODE *newrbf)
175	+	{
176	+	RBFNODE         *rbf, *rbf_last;
177	+	/* check for redundant meas. */
178	+	for (rbf = dsf_list; rbf != NULL; rbf = rbf->next)
179	+	if (DOT(rbf->invec, newrbf->invec) >= 1.-FTINY) {
180	+	fputs("Duplicate incident measurement (ignored)\n", stderr);
181	+	free(newrbf);
182	+	return;
183	+	}
184	+	/* keep in ascending theta order */
185	+	for (rbf_last = NULL, rbf = dsf_list;
186	+	single_plane_incident & (rbf != NULL);
187	+	rbf_last = rbf, rbf = rbf->next)
188	+	if (input_orient*rbf->invec[2] < input_orient*newrbf->invec[2])
189	+	break;
190	+	if (rbf_last == NULL) {
191	+	newrbf->next = dsf_list;
192	+	dsf_list = newrbf;
193	+	return;
194	+	}
195	+	newrbf->next = rbf;
196	+	rbf_last->next = newrbf;
197	+	}
198	+
199		/* Count up filled nodes and build RBF representation from current grid */
200	<	static RBFLIST *
200	>	static RBFNODE *
201		make_rbfrep(void)
202		{
203		int     niter = 16;
204	+	int     minrad = ANG2R(pow(2., 1.-samp_order));
205		double  lastVar, thisVar = 100.;
206		int     nn;
207	<	RBFLIST *newnode;
207	>	RBFNODE *newnode;
208		int     i, j;
209
210		nn = 0;                 /* count selected bins */
#	Line 133 | Line 212 | make_rbfrep(void)
212		for (j = 0; j < GRIDRES; j++)
213		nn += dsf_grid[i][j].nval;
214		/* allocate RBF array */
215	<	newnode = (RBFLIST *)malloc(sizeof(RBFLIST) + sizeof(RBFVAL)*(nn-1));
215	>	newnode = (RBFNODE *)malloc(sizeof(RBFNODE) + sizeof(RBFVAL)*(nn-1));
216		if (newnode == NULL) {
217	<	fputs("Out of memory in make_rbfrep\n", stderr);
217	>	fputs("Out of memory in make_rbfrep()\n", stderr);
218		exit(1);
219		}
220		newnode->next = NULL;
#	Line 143 | Line 222 | make_rbfrep(void)
222		newnode->invec[2] = sin(M_PI/180.*theta_in_deg);
223		newnode->invec[0] = cos(M_PI/180.*phi_in_deg)*newnode->invec[2];
224		newnode->invec[1] = sin(M_PI/180.*phi_in_deg)*newnode->invec[2];
225	<	newnode->invec[2] = sqrt(1. - newnode->invec[2]*newnode->invec[2]);
225	>	newnode->invec[2] = input_orient*sqrt(1. - newnode->invec[2]*newnode->invec[2]);
226	>	newnode->vtotal = 0;
227		newnode->nrbf = nn;
228		nn = 0;                 /* fill RBF array */
229		for (i = 0; i < GRIDRES; i++)
#	Line 153 | Line 233 | make_rbfrep(void)
233		newnode->rbfa[nn].crad = RSCA*dsf_grid[i][j].crad + .5;
234		newnode->rbfa[nn].gx = i;
235		newnode->rbfa[nn].gy = j;
236	+	if (newnode->rbfa[nn].crad < minrad)
237	+	minrad = newnode->rbfa[nn].crad;
238		++nn;
239		}
240		/* iterate to improve interpolation accuracy */
241		do {
242	<	double  dsum = .0, dsum2 = .0;
242	>	double  dsum = 0, dsum2 = 0;
243		nn = 0;
244		for (i = 0; i < GRIDRES; i++)
245		for (j = 0; j < GRIDRES; j++)
246		if (dsf_grid[i][j].nval) {
247		FVECT   odir;
248		double  corr;
249	<	vec_from_pos(odir, i, j);
249	>	ovec_from_pos(odir, i, j);
250		newnode->rbfa[nn++].peak *= corr =
251		dsf_grid[i][j].vsum /
252		eval_rbfrep(newnode, odir);
#	Line 173 | Line 255 | make_rbfrep(void)
255		}
256		lastVar = thisVar;
257		thisVar = dsum2/(double)nn;
258	<	/*
258	>	#ifdef DEBUG
259		fprintf(stderr, "Avg., RMS error: %.1f%%  %.1f%%\n",
260		100.*dsum/(double)nn,
261		100.*sqrt(thisVar));
262	<	*/
262	>	#endif
263		} while (--niter > 0 && lastVar-thisVar > 0.02*lastVar);
264
265	<	newnode->next = dsf_list;
266	<	return(dsf_list = newnode);
265	>	nn = 0;                 /* compute sum for normalization */
266	>	while (nn < newnode->nrbf)
267	>	newnode->vtotal += rbf_volume(&newnode->rbfa[nn++]);
268	>
269	>	insert_dsf(newnode);
270	>	/* adjust sampling resolution */
271	>	samp_order = log(2./R2ANG(minrad))/M_LN2 + .5;
272	>
273	>	return(newnode);
274		}
275
276		/* Load a set of measurements corresponding to a particular incident angle */
277		static int
278	<	load_bsdf_meas(const char *fname)
278	>	load_pabopto_meas(const char *fname)
279		{
280		FILE    *fp = fopen(fname, "r");
281		int     inp_is_DSF = -1;
282	<	double  theta_out, phi_out, val;
282	>	double  new_phi, theta_out, phi_out, val;
283		char    buf[2048];
284		int     n, c;
285
#	Line 200 | Line 289 | load_bsdf_meas(const char *fname)
289		return(0);
290		}
291		memset(dsf_grid, 0, sizeof(dsf_grid));
292	+	#ifdef DEBUG
293	+	fprintf(stderr, "Loading measurement file '%s'...\n", fname);
294	+	#endif
295		/* read header information */
296		while ((c = getc(fp)) == '#' || c == EOF) {
297		if (fgets(buf, sizeof(buf), fp) == NULL) {
#	Line 218 | Line 310 | load_bsdf_meas(const char *fname)
310		}
311		if (sscanf(buf, "intheta %lf", &theta_in_deg) == 1)
312		continue;
313	<	if (sscanf(buf, "inphi %lf", &phi_in_deg) == 1)
313	>	if (sscanf(buf, "inphi %lf", &new_phi) == 1)
314		continue;
315		if (sscanf(buf, "incident_angle %lf %lf",
316	<	&theta_in_deg, &phi_in_deg) == 2)
316	>	&theta_in_deg, &new_phi) == 2)
317		continue;
318		}
319		if (inp_is_DSF < 0) {
#	Line 230 | Line 322 | load_bsdf_meas(const char *fname)
322		fclose(fp);
323		return(0);
324		}
325	<	ungetc(c, fp);          /* read actual data */
325	>	if (!input_orient)              /* check input orientation */
326	>	input_orient = 1 - 2*(theta_in_deg > 90.);
327	>	else if (input_orient > 0 ^ theta_in_deg < 90.) {
328	>	fputs("Cannot handle input angles on both sides of surface\n",
329	>	stderr);
330	>	exit(1);
331	>	}
332	>	if (single_plane_incident > 0)  /* check if still in plane */
333	>	single_plane_incident = (round(new_phi) == round(phi_in_deg));
334	>	else if (single_plane_incident < 0)
335	>	single_plane_incident = 1;
336	>	phi_in_deg = new_phi;
337	>	ungetc(c, fp);                  /* read actual data */
338		while (fscanf(fp, "%lf %lf %lf\n", &theta_out, &phi_out, &val) == 3) {
339		FVECT   ovec;
340		int     pos[2];
341
342	+	if (!output_orient)     /* check output orientation */
343	+	output_orient = 1 - 2*(theta_out > 90.);
344	+	else if (output_orient > 0 ^ theta_out < 90.) {
345	+	fputs("Cannot handle output angles on both sides of surface\n",
346	+	stderr);
347	+	exit(1);
348	+	}
349		ovec[2] = sin(M_PI/180.*theta_out);
350		ovec[0] = cos(M_PI/180.*phi_out) * ovec[2];
351		ovec[1] = sin(M_PI/180.*phi_out) * ovec[2];
#	Line 276 | Line 387 | compute_radii(void)
387		j = (i&1) ? jn : GRIDRES-1-jn;
388		if (dsf_grid[i][j].nval)        /* find empty grid pos. */
389		continue;
390	<	vec_from_pos(ovec0, i, j);
390	>	ovec_from_pos(ovec0, i, j);
391		inear = jnear = -1;             /* find nearest non-empty */
392		lastang2 = M_PI*M_PI;
393		for (ii = i-r; ii <= i+r; ii++) {
#	Line 287 | Line 398 | compute_radii(void)
398		if (jj >= GRIDRES) break;
399		if (!dsf_grid[ii][jj].nval)
400		continue;
401	<	vec_from_pos(ovec1, ii, jj);
401	>	ovec_from_pos(ovec1, ii, jj);
402		ang2 = 2. - 2.*DOT(ovec0,ovec1);
403		if (ang2 >= lastang2)
404		continue;
#	Line 304 | Line 415 | compute_radii(void)
415		if (r > dsf_grid[inear][jnear].crad)
416		dsf_grid[inear][jnear].crad = r;
417		/* next search radius */
418	<	r = ang2*(2.*GRIDRES/M_PI) + 1;
418	>	r = ang2*(2.*GRIDRES/M_PI) + 3;
419		}
420		/* blur radii over hemisphere */
421		memset(fill_grid, 0, sizeof(fill_grid));
#	Line 348 | Line 459 | cull_values(void)
459		continue;
460		if (!dsf_grid[i][j].crad)
461		continue;               /* shouldn't happen */
462	<	vec_from_pos(ovec0, i, j);
462	>	ovec_from_pos(ovec0, i, j);
463		maxang = 2.*R2ANG(dsf_grid[i][j].crad);
464		if (maxang > ovec0[2])          /* clamp near horizon */
465		maxang = ovec0[2];
#	Line 364 | Line 475 | cull_values(void)
475		continue;
476		if ((ii == i) & (jj == j))
477		continue;       /* don't get self-absorbed */
478	<	vec_from_pos(ovec1, ii, jj);
478	>	ovec_from_pos(ovec1, ii, jj);
479		if (2. - 2.*DOT(ovec0,ovec1) >= maxang2)
480		continue;
481		/* absorb sum */
#	Line 385 | Line 496 | cull_values(void)
496		}
497		}
498
499	+	/* Compute (and allocate) migration price matrix for optimization */
500	+	static float *
501	+	price_routes(const RBFNODE *from_rbf, const RBFNODE *to_rbf)
502	+	{
503	+	float   *pmtx = (float *)malloc(sizeof(float) *
504	+	from_rbf->nrbf * to_rbf->nrbf);
505	+	FVECT   *vto = (FVECT *)malloc(sizeof(FVECT) * to_rbf->nrbf);
506	+	int     i, j;
507
508	+	if ((pmtx == NULL) | (vto == NULL)) {
509	+	fputs("Out of memory in migration_costs()\n", stderr);
510	+	exit(1);
511	+	}
512	+	for (j = to_rbf->nrbf; j--; )           /* save repetitive ops. */
513	+	ovec_from_pos(vto[j], to_rbf->rbfa[j].gx, to_rbf->rbfa[j].gy);
514	+
515	+	for (i = from_rbf->nrbf; i--; ) {
516	+	const double        from_ang = R2ANG(from_rbf->rbfa[i].crad);
517	+	FVECT               vfrom;
518	+	ovec_from_pos(vfrom, from_rbf->rbfa[i].gx, from_rbf->rbfa[i].gy);
519	+	for (j = to_rbf->nrbf; j--; )
520	+	pmtx[i*to_rbf->nrbf + j] = acos(DOT(vfrom, vto[j])) +
521	+	fabs(R2ANG(to_rbf->rbfa[j].crad) - from_ang);
522	+	}
523	+	free(vto);
524	+	return(pmtx);
525	+	}
526	+
527	+	/* Comparison routine needed for sorting price row */
528	+	static const float      *price_arr;
529	+	static int
530	+	msrt_cmp(const void *p1, const void *p2)
531	+	{
532	+	float   c1 = price_arr[*(const int *)p1];
533	+	float   c2 = price_arr[*(const int *)p2];
534	+
535	+	if (c1 > c2) return(1);
536	+	if (c1 < c2) return(-1);
537	+	return(0);
538	+	}
539	+
540	+	/* Compute minimum (optimistic) cost for moving the given source material */
541	+	static double
542	+	min_cost(double amt2move, const double *avail, const float *price, int n)
543	+	{
544	+	static int      *price_sort = NULL;
545	+	static int      n_alloc = 0;
546	+	double          total_cost = 0;
547	+	int             i;
548	+
549	+	if (amt2move <= FTINY)                  /* pre-emptive check */
550	+	return(0.);
551	+	if (n > n_alloc) {                      /* (re)allocate sort array */
552	+	if (n_alloc) free(price_sort);
553	+	price_sort = (int *)malloc(sizeof(int)*n);
554	+	if (price_sort == NULL) {
555	+	fputs("Out of memory in min_cost()\n", stderr);
556	+	exit(1);
557	+	}
558	+	n_alloc = n;
559	+	}
560	+	for (i = n; i--; )
561	+	price_sort[i] = i;
562	+	price_arr = price;
563	+	qsort(price_sort, n, sizeof(int), &msrt_cmp);
564	+	/* move cheapest first */
565	+	for (i = 0; i < n && amt2move > FTINY; i++) {
566	+	int     d = price_sort[i];
567	+	double  amt = (amt2move < avail[d]) ? amt2move : avail[d];
568	+
569	+	total_cost += amt * price[d];
570	+	amt2move -= amt;
571	+	}
572	+	return(total_cost);
573	+	}
574	+
575	+	/* Take a step in migration by choosing optimal bucket to transfer */
576	+	static double
577	+	migration_step(MIGRATION *mig, double *src_rem, double *dst_rem, const float *pmtx)
578	+	{
579	+	static double   *src_cost = NULL;
580	+	static int      n_alloc = 0;
581	+	const double    maxamt = .1; /* 2./(mtx_nrows(mig)*mtx_ncols(mig)); */
582	+	double          amt = 0;
583	+	struct {
584	+	int     s, d;   /* source and destination */
585	+	double  price;  /* price estimate per amount moved */
586	+	double  amt;    /* amount we can move */
587	+	} cur, best;
588	+	int             i;
589	+
590	+	if (mtx_nrows(mig) > n_alloc) {         /* allocate cost array */
591	+	if (n_alloc)
592	+	free(src_cost);
593	+	src_cost = (double *)malloc(sizeof(double)*mtx_nrows(mig));
594	+	if (src_cost == NULL) {
595	+	fputs("Out of memory in migration_step()\n", stderr);
596	+	exit(1);
597	+	}
598	+	n_alloc = mtx_nrows(mig);
599	+	}
600	+	for (i = mtx_nrows(mig); i--; )         /* starting costs for diff. */
601	+	src_cost[i] = min_cost(src_rem[i], dst_rem,
602	+	pmtx+i*mtx_ncols(mig), mtx_ncols(mig));
603	+
604	+	/* find best source & dest. */
605	+	best.s = best.d = -1; best.price = FHUGE; best.amt = 0;
606	+	for (cur.s = mtx_nrows(mig); cur.s--; ) {
607	+	const float *price = pmtx + cur.s*mtx_ncols(mig);
608	+	double      cost_others = 0;
609	+	if (src_rem[cur.s] <= FTINY)
610	+	continue;
611	+	cur.d = -1;                         /* examine cheapest dest. */
612	+	for (i = mtx_ncols(mig); i--; )
613	+	if (dst_rem[i] > FTINY &&
614	+	(cur.d < 0 || price[i] < price[cur.d]))
615	+	cur.d = i;
616	+	if (cur.d < 0)
617	+	return(.0);
618	+	if ((cur.price = price[cur.d]) >= best.price)
619	+	continue;                   /* no point checking further */
620	+	cur.amt = (src_rem[cur.s] < dst_rem[cur.d]) ?
621	+	src_rem[cur.s] : dst_rem[cur.d];
622	+	if (cur.amt > maxamt) cur.amt = maxamt;
623	+	dst_rem[cur.d] -= cur.amt;          /* add up differential costs */
624	+	for (i = mtx_nrows(mig); i--; ) {
625	+	if (i == cur.s) continue;
626	+	cost_others += min_cost(src_rem[i], dst_rem, price, mtx_ncols(mig))
627	+	- src_cost[i];
628	+	}
629	+	dst_rem[cur.d] += cur.amt;          /* undo trial move */
630	+	cur.price += cost_others/cur.amt;   /* adjust effective price */
631	+	if (cur.price < best.price)         /* are we better than best? */
632	+	best = cur;
633	+	}
634	+	if ((best.s < 0) | (best.d < 0))
635	+	return(.0);
636	+	/* make the actual move */
637	+	mig->mtx[mtx_ndx(mig,best.s,best.d)] += best.amt;
638	+	src_rem[best.s] -= best.amt;
639	+	dst_rem[best.d] -= best.amt;
640	+	return(best.amt);
641	+	}
642	+
643	+	#ifdef DEBUG
644	+	static char *
645	+	thetaphi(const FVECT v)
646	+	{
647	+	static char     buf[128];
648	+	double          theta, phi;
649	+
650	+	theta = 180./M_PI*acos(v[2]);
651	+	phi = 180./M_PI*atan2(v[1],v[0]);
652	+	sprintf(buf, "(%.0f,%.0f)", theta, phi);
653	+
654	+	return(buf);
655	+	}
656	+	#endif
657	+
658	+	/* Create a new migration holder (sharing memory for multiprocessing) */
659	+	static MIGRATION *
660	+	new_migration(RBFNODE *from_rbf, RBFNODE *to_rbf)
661	+	{
662	+	size_t          memlen = sizeof(MIGRATION) +
663	+	sizeof(float)*(from_rbf->nrbf*to_rbf->nrbf - 1);
664	+	MIGRATION       *newmig;
665	+	#ifdef _WIN32
666	+	newmig = (MIGRATION *)malloc(memlen);
667	+	#else
668	+	if (nprocs <= 1) {                      /* single process? */
669	+	newmig = (MIGRATION *)malloc(memlen);
670	+	} else {                                /* else need to share memory */
671	+	newmig = (MIGRATION *)mmap(NULL, memlen, PROT_READ|PROT_WRITE,
672	+	MAP_ANON|MAP_SHARED, -1, 0);
673	+	if ((void *)newmig == MAP_FAILED)
674	+	newmig = NULL;
675	+	}
676	+	#endif
677	+	if (newmig == NULL) {
678	+	fprintf(stderr, "%s: cannot allocate new migration\n", progname);
679	+	exit(1);
680	+	}
681	+	newmig->rbfv[0] = from_rbf;
682	+	newmig->rbfv[1] = to_rbf;
683	+	/* insert in edge lists */
684	+	newmig->enxt[0] = from_rbf->ejl;
685	+	from_rbf->ejl = newmig;
686	+	newmig->enxt[1] = to_rbf->ejl;
687	+	to_rbf->ejl = newmig;
688	+	newmig->next = mig_list;                /* push onto global list */
689	+	return(mig_list = newmig);
690	+	}
691	+
692	+	#ifdef _WIN32
693	+	#define await_children(n)       (void)(n)
694	+	#define run_subprocess()        0
695	+	#define end_subprocess()        (void)0
696	+	#else
697	+
698	+	/* Wait for the specified number of child processes to complete */
699	+	static void
700	+	await_children(int n)
701	+	{
702	+	int     exit_status = 0;
703	+
704	+	if (n > nchild)
705	+	n = nchild;
706	+	while (n-- > 0) {
707	+	int     status;
708	+	if (wait(&status) < 0) {
709	+	fprintf(stderr, "%s: missing child(ren)!\n", progname);
710	+	nchild = 0;
711	+	break;
712	+	}
713	+	--nchild;
714	+	if (status) {                   /* something wrong */
715	+	if ((status = WEXITSTATUS(status)))
716	+	exit_status = status;
717	+	else
718	+	exit_status += !exit_status;
719	+	fprintf(stderr, "%s: subprocess died\n", progname);
720	+	n = nchild;             /* wait for the rest */
721	+	}
722	+	}
723	+	if (exit_status)
724	+	exit(exit_status);
725	+	}
726	+
727	+	/* Start child process if multiprocessing selected */
728	+	static pid_t
729	+	run_subprocess(void)
730	+	{
731	+	int     status;
732	+	pid_t   pid;
733	+
734	+	if (nprocs <= 1)                        /* any children requested? */
735	+	return(0);
736	+	await_children(nchild + 1 - nprocs);    /* free up child process */
737	+	if ((pid = fork())) {
738	+	if (pid < 0) {
739	+	fprintf(stderr, "%s: cannot fork subprocess\n",
740	+	progname);
741	+	exit(1);
742	+	}
743	+	++nchild;                       /* subprocess started */
744	+	return(pid);
745	+	}
746	+	nchild = -1;
747	+	return(0);                              /* put child to work */
748	+	}
749	+
750	+	/* If we are in subprocess, call exit */
751	+	#define end_subprocess()        if (nchild < 0) _exit(0); else
752	+
753	+	#endif  /* ! _WIN32 */
754	+
755	+	/* Compute and insert migration along directed edge (may fork child) */
756	+	static MIGRATION *
757	+	create_migration(RBFNODE *from_rbf, RBFNODE *to_rbf)
758	+	{
759	+	const double    end_thresh = 0.1/(from_rbf->nrbf*to_rbf->nrbf);
760	+	const double    rel_thresh = 0.0001;
761	+	float           *pmtx;
762	+	MIGRATION       *newmig;
763	+	double          *src_rem, *dst_rem;
764	+	double          total_rem = 1., move_amt;
765	+	int             i;
766	+	/* check if exists already */
767	+	for (newmig = from_rbf->ejl; newmig != NULL;
768	+	newmig = nextedge(from_rbf,newmig))
769	+	if (newmig->rbfv[1] == to_rbf)
770	+	return(NULL);
771	+	/* else allocate */
772	+	newmig = new_migration(from_rbf, to_rbf);
773	+	if (run_subprocess())
774	+	return(newmig);                 /* child continues */
775	+	pmtx = price_routes(from_rbf, to_rbf);
776	+	src_rem = (double *)malloc(sizeof(double)*from_rbf->nrbf);
777	+	dst_rem = (double *)malloc(sizeof(double)*to_rbf->nrbf);
778	+	if ((src_rem == NULL) | (dst_rem == NULL)) {
779	+	fputs("Out of memory in create_migration()\n", stderr);
780	+	exit(1);
781	+	}
782	+	#ifdef DEBUG
783	+	fprintf(stderr, "Building path from (theta,phi) %s ",
784	+	thetaphi(from_rbf->invec));
785	+	fprintf(stderr, "to %s", thetaphi(to_rbf->invec));
786	+	/* if (nchild) */ fputc('\n', stderr);
787	+	#endif
788	+	/* starting quantities */
789	+	memset(newmig->mtx, 0, sizeof(float)*from_rbf->nrbf*to_rbf->nrbf);
790	+	for (i = from_rbf->nrbf; i--; )
791	+	src_rem[i] = rbf_volume(&from_rbf->rbfa[i]) / from_rbf->vtotal;
792	+	for (i = to_rbf->nrbf; i--; )
793	+	dst_rem[i] = rbf_volume(&to_rbf->rbfa[i]) / to_rbf->vtotal;
794	+	do {                                    /* move a bit at a time */
795	+	move_amt = migration_step(newmig, src_rem, dst_rem, pmtx);
796	+	total_rem -= move_amt;
797	+	#ifdef DEBUG
798	+	if (!nchild)
799	+	/* fputc('.', stderr); */
800	+	fprintf(stderr, "%.9f remaining...\r", total_rem);
801	+	#endif
802	+	} while ((total_rem > end_thresh) & (move_amt > rel_thresh*total_rem));
803	+	#ifdef DEBUG
804	+	if (!nchild) fputs("\ndone.\n", stderr);
805	+	#endif
806	+
807	+	free(pmtx);                             /* free working arrays */
808	+	free(src_rem);
809	+	free(dst_rem);
810	+	for (i = from_rbf->nrbf; i--; ) {       /* normalize final matrix */
811	+	float       nf = rbf_volume(&from_rbf->rbfa[i]);
812	+	int         j;
813	+	if (nf <= FTINY) continue;
814	+	nf = from_rbf->vtotal / nf;
815	+	for (j = to_rbf->nrbf; j--; )
816	+	newmig->mtx[mtx_ndx(newmig,i,j)] *= nf;
817	+	}
818	+	end_subprocess();                       /* exit here if subprocess */
819	+	return(newmig);
820	+	}
821	+
822	+	/* Get triangle surface orientation (unnormalized) */
823	+	static void
824	+	tri_orient(FVECT vres, const FVECT v1, const FVECT v2, const FVECT v3)
825	+	{
826	+	FVECT   v2minus1, v3minus2;
827	+
828	+	VSUB(v2minus1, v2, v1);
829	+	VSUB(v3minus2, v3, v2);
830	+	VCROSS(vres, v2minus1, v3minus2);
831	+	}
832	+
833	+	/* Determine if vertex order is reversed (inward normal) */
834	+	static int
835	+	is_rev_tri(const FVECT v1, const FVECT v2, const FVECT v3)
836	+	{
837	+	FVECT   tor;
838	+
839	+	tri_orient(tor, v1, v2, v3);
840	+
841	+	return(DOT(tor, v2) < 0.);
842	+	}
843	+
844	+	/* Find vertices completing triangles on either side of the given edge */
845	+	static int
846	+	get_triangles(RBFNODE *rbfv[2], const MIGRATION *mig)
847	+	{
848	+	const MIGRATION *ej, *ej2;
849	+	RBFNODE         *tv;
850	+
851	+	rbfv[0] = rbfv[1] = NULL;
852	+	if (mig == NULL)
853	+	return(0);
854	+	for (ej = mig->rbfv[0]->ejl; ej != NULL;
855	+	ej = nextedge(mig->rbfv[0],ej)) {
856	+	if (ej == mig)
857	+	continue;
858	+	tv = opp_rbf(mig->rbfv[0],ej);
859	+	for (ej2 = tv->ejl; ej2 != NULL; ej2 = nextedge(tv,ej2))
860	+	if (opp_rbf(tv,ej2) == mig->rbfv[1]) {
861	+	rbfv[is_rev_tri(mig->rbfv[0]->invec,
862	+	mig->rbfv[1]->invec,
863	+	tv->invec)] = tv;
864	+	break;
865	+	}
866	+	}
867	+	return((rbfv[0] != NULL) + (rbfv[1] != NULL));
868	+	}
869	+
870	+	/* Check if prospective vertex would create overlapping triangle */
871	+	static int
872	+	overlaps_tri(const RBFNODE *bv0, const RBFNODE *bv1, const RBFNODE *pv)
873	+	{
874	+	const MIGRATION *ej;
875	+	RBFNODE         *vother[2];
876	+	int             im_rev;
877	+	/* find shared edge in mesh */
878	+	for (ej = pv->ejl; ej != NULL; ej = nextedge(pv,ej)) {
879	+	const RBFNODE   *tv = opp_rbf(pv,ej);
880	+	if (tv == bv0) {
881	+	im_rev = is_rev_tri(ej->rbfv[0]->invec,
882	+	ej->rbfv[1]->invec, bv1->invec);
883	+	break;
884	+	}
885	+	if (tv == bv1) {
886	+	im_rev = is_rev_tri(ej->rbfv[0]->invec,
887	+	ej->rbfv[1]->invec, bv0->invec);
888	+	break;
889	+	}
890	+	}
891	+	if (!get_triangles(vother, ej))         /* triangle on same side? */
892	+	return(0);
893	+	return(vother[im_rev] != NULL);
894	+	}
895	+
896	+	/* Find context hull vertex to complete triangle (oriented call) */
897	+	static RBFNODE *
898	+	find_chull_vert(const RBFNODE *rbf0, const RBFNODE *rbf1)
899	+	{
900	+	FVECT   vmid, vejn, vp;
901	+	RBFNODE *rbf, *rbfbest = NULL;
902	+	double  dprod, area2, bestarea2 = FHUGE, bestdprod = -.5;
903	+
904	+	VSUB(vejn, rbf1->invec, rbf0->invec);
905	+	VADD(vmid, rbf0->invec, rbf1->invec);
906	+	if (normalize(vejn) == 0 || normalize(vmid) == 0)
907	+	return(NULL);
908	+	/* XXX exhaustive search */
909	+	/* Find triangle with minimum rotation from perpendicular */
910	+	for (rbf = dsf_list; rbf != NULL; rbf = rbf->next) {
911	+	if ((rbf == rbf0) | (rbf == rbf1))
912	+	continue;
913	+	tri_orient(vp, rbf0->invec, rbf1->invec, rbf->invec);
914	+	if (DOT(vp, vmid) <= FTINY)
915	+	continue;               /* wrong orientation */
916	+	area2 = .25*DOT(vp,vp);
917	+	VSUB(vp, rbf->invec, rbf0->invec);
918	+	dprod = -DOT(vp, vejn);
919	+	VSUM(vp, vp, vejn, dprod);      /* above guarantees non-zero */
920	+	dprod = DOT(vp, vmid) / VLEN(vp);
921	+	if (dprod <= bestdprod + FTINY*(1 - 2*(area2 < bestarea2)))
922	+	continue;               /* found better already */
923	+	if (overlaps_tri(rbf0, rbf1, rbf))
924	+	continue;               /* overlaps another triangle */
925	+	rbfbest = rbf;
926	+	bestdprod = dprod;              /* new one to beat */
927	+	bestarea2 = area2;
928	+	}
929	+	return(rbfbest);
930	+	}
931	+
932	+	/* Create new migration edge and grow mesh recursively around it */
933	+	static void
934	+	mesh_from_edge(MIGRATION *edge)
935	+	{
936	+	MIGRATION       *ej0, *ej1;
937	+	RBFNODE         *tvert[2];
938	+
939	+	if (edge == NULL)
940	+	return;
941	+	/* triangle on either side? */
942	+	get_triangles(tvert, edge);
943	+	if (tvert[0] == NULL) {                 /* grow mesh on right */
944	+	tvert[0] = find_chull_vert(edge->rbfv[0], edge->rbfv[1]);
945	+	if (tvert[0] != NULL) {
946	+	if (tvert[0] > edge->rbfv[0])
947	+	ej0 = create_migration(edge->rbfv[0], tvert[0]);
948	+	else
949	+	ej0 = create_migration(tvert[0], edge->rbfv[0]);
950	+	if (tvert[0] > edge->rbfv[1])
951	+	ej1 = create_migration(edge->rbfv[1], tvert[0]);
952	+	else
953	+	ej1 = create_migration(tvert[0], edge->rbfv[1]);
954	+	mesh_from_edge(ej0);
955	+	mesh_from_edge(ej1);
956	+	}
957	+	} else if (tvert[1] == NULL) {          /* grow mesh on left */
958	+	tvert[1] = find_chull_vert(edge->rbfv[1], edge->rbfv[0]);
959	+	if (tvert[1] != NULL) {
960	+	if (tvert[1] > edge->rbfv[0])
961	+	ej0 = create_migration(edge->rbfv[0], tvert[1]);
962	+	else
963	+	ej0 = create_migration(tvert[1], edge->rbfv[0]);
964	+	if (tvert[1] > edge->rbfv[1])
965	+	ej1 = create_migration(edge->rbfv[1], tvert[1]);
966	+	else
967	+	ej1 = create_migration(tvert[1], edge->rbfv[1]);
968	+	mesh_from_edge(ej0);
969	+	mesh_from_edge(ej1);
970	+	}
971	+	}
972	+	}
973	+
974	+	#ifdef DEBUG
975	+	#include "random.h"
976	+	#include "bmpfile.h"
977	+	/* Hash pointer to byte value (must return 0 for NULL) */
978	+	static int
979	+	byte_hash(const void *p)
980	+	{
981	+	size_t  h = (size_t)p;
982	+	h ^= (size_t)p >> 8;
983	+	h ^= (size_t)p >> 16;
984	+	h ^= (size_t)p >> 24;
985	+	return(h & 0xff);
986	+	}
987	+	/* Write out BMP image showing edges */
988	+	static void
989	+	write_edge_image(const char *fname)
990	+	{
991	+	BMPHeader       *hdr = BMPmappedHeader(GRIDRES, GRIDRES, 0, 256);
992	+	BMPWriter       *wtr;
993	+	int             i, j;
994	+
995	+	fprintf(stderr, "Writing incident mesh drawing to '%s'\n", fname);
996	+	hdr->compr = BI_RLE8;
997	+	for (i = 256; --i; ) {                  /* assign random color map */
998	+	hdr->palette[i].r = random() & 0xff;
999	+	hdr->palette[i].g = random() & 0xff;
1000	+	hdr->palette[i].b = random() & 0xff;
1001	+	/* reject dark colors */
1002	+	i += (hdr->palette[i].r + hdr->palette[i].g +
1003	+	hdr->palette[i].b < 128);
1004	+	}
1005	+	hdr->palette[0].r = hdr->palette[0].g = hdr->palette[0].b = 0;
1006	+	/* open output */
1007	+	wtr = BMPopenOutputFile(fname, hdr);
1008	+	if (wtr == NULL) {
1009	+	free(hdr);
1010	+	return;
1011	+	}
1012	+	for (i = 0; i < GRIDRES; i++) {         /* write scanlines */
1013	+	for (j = 0; j < GRIDRES; j++)
1014	+	wtr->scanline[j] = byte_hash(mig_grid[i][j]);
1015	+	if (BMPwriteScanline(wtr) != BIR_OK)
1016	+	break;
1017	+	}
1018	+	BMPcloseOutput(wtr);                    /* close & clean up */
1019	+	}
1020	+	#endif
1021	+
1022	+	/* Draw edge list into mig_grid array */
1023	+	static void
1024	+	draw_edges()
1025	+	{
1026	+	int             nnull = 0, ntot = 0;
1027	+	MIGRATION       *ej;
1028	+	int             p0[2], p1[2];
1029	+
1030	+	/* memset(mig_grid, 0, sizeof(mig_grid)); */
1031	+	for (ej = mig_list; ej != NULL; ej = ej->next) {
1032	+	++ntot;
1033	+	pos_from_vec(p0, ej->rbfv[0]->invec);
1034	+	pos_from_vec(p1, ej->rbfv[1]->invec);
1035	+	if ((p0[0] == p1[0]) & (p0[1] == p1[1])) {
1036	+	++nnull;
1037	+	mig_grid[p0[0]][p0[1]] = ej;
1038	+	continue;
1039	+	}
1040	+	if (abs(p1[0]-p0[0]) > abs(p1[1]-p0[1])) {
1041	+	const int       xstep = 2*(p1[0] > p0[0]) - 1;
1042	+	const double    ystep = (double)((p1[1]-p0[1])*xstep) /
1043	+	(double)(p1[0]-p0[0]);
1044	+	int             x;
1045	+	double          y;
1046	+	for (x = p0[0], y = p0[1]+.5; x != p1[0];
1047	+	x += xstep, y += ystep)
1048	+	mig_grid[x][(int)y] = ej;
1049	+	mig_grid[x][(int)y] = ej;
1050	+	} else {
1051	+	const int       ystep = 2*(p1[1] > p0[1]) - 1;
1052	+	const double    xstep = (double)((p1[0]-p0[0])*ystep) /
1053	+	(double)(p1[1]-p0[1]);
1054	+	int             y;
1055	+	double          x;
1056	+	for (y = p0[1], x = p0[0]+.5; y != p1[1];
1057	+	y += ystep, x += xstep)
1058	+	mig_grid[(int)x][y] = ej;
1059	+	mig_grid[(int)x][y] = ej;
1060	+	}
1061	+	}
1062	+	if (nnull)
1063	+	fprintf(stderr, "Warning: %d of %d edges are null\n",
1064	+	nnull, ntot);
1065	+	#ifdef DEBUG
1066	+	write_edge_image("bsdf_edges.bmp");
1067	+	#endif
1068	+	}
1069	+
1070	+	/* Build our triangle mesh from recorded RBFs */
1071	+	static void
1072	+	build_mesh()
1073	+	{
1074	+	double          best2 = M_PI*M_PI;
1075	+	RBFNODE         *shrt_edj[2];
1076	+	RBFNODE         *rbf0, *rbf1;
1077	+	/* check if isotropic */
1078	+	if (single_plane_incident) {
1079	+	for (rbf0 = dsf_list; rbf0 != NULL; rbf0 = rbf0->next)
1080	+	if (rbf0->next != NULL)
1081	+	create_migration(rbf0, rbf0->next);
1082	+	await_children(nchild);
1083	+	return;
1084	+	}
1085	+	/* start w/ shortest edge */
1086	+	shrt_edj[0] = shrt_edj[1] = NULL;
1087	+	for (rbf0 = dsf_list; rbf0 != NULL; rbf0 = rbf0->next)
1088	+	for (rbf1 = rbf0->next; rbf1 != NULL; rbf1 = rbf1->next) {
1089	+	double  dist2 = 2. - 2.*DOT(rbf0->invec,rbf1->invec);
1090	+	if (dist2 < best2) {
1091	+	shrt_edj[0] = rbf0;
1092	+	shrt_edj[1] = rbf1;
1093	+	best2 = dist2;
1094	+	}
1095	+	}
1096	+	if (shrt_edj[0] == NULL) {
1097	+	fputs("Cannot find shortest edge\n", stderr);
1098	+	exit(1);
1099	+	}
1100	+	/* build mesh from this edge */
1101	+	if (shrt_edj[0] < shrt_edj[1])
1102	+	mesh_from_edge(create_migration(shrt_edj[0], shrt_edj[1]));
1103	+	else
1104	+	mesh_from_edge(create_migration(shrt_edj[1], shrt_edj[0]));
1105	+	/* complete migrations */
1106	+	await_children(nchild);
1107	+	/* draw edge list into grid */
1108	+	draw_edges();
1109	+	}
1110	+
1111	+	/* Identify enclosing triangle for this position (flood fill raster check) */
1112	+	static int
1113	+	identify_tri(MIGRATION *miga[3], unsigned char vmap[GRIDRES][(GRIDRES+7)/8],
1114	+	int px, int py)
1115	+	{
1116	+	const int       btest = 1<<(py&07);
1117	+
1118	+	if (vmap[px][py>>3] & btest)            /* already visited here? */
1119	+	return(1);
1120	+	/* else mark it */
1121	+	vmap[px][py>>3] |= btest;
1122	+
1123	+	if (mig_grid[px][py] != NULL) {         /* are we on an edge? */
1124	+	int     i;
1125	+	for (i = 0; i < 3; i++) {
1126	+	if (miga[i] == mig_grid[px][py])
1127	+	return(1);
1128	+	if (miga[i] != NULL)
1129	+	continue;
1130	+	miga[i] = mig_grid[px][py];
1131	+	return(1);
1132	+	}
1133	+	return(0);                      /* outside triangle! */
1134	+	}
1135	+	/* check neighbors (flood) */
1136	+	if (px > 0 && !identify_tri(miga, vmap, px-1, py))
1137	+	return(0);
1138	+	if (px < GRIDRES-1 && !identify_tri(miga, vmap, px+1, py))
1139	+	return(0);
1140	+	if (py > 0 && !identify_tri(miga, vmap, px, py-1))
1141	+	return(0);
1142	+	if (py < GRIDRES-1 && !identify_tri(miga, vmap, px, py+1))
1143	+	return(0);
1144	+	return(1);                              /* this neighborhood done */
1145	+	}
1146	+
1147	+	/* Insert vertex in ordered list */
1148	+	static void
1149	+	insert_vert(RBFNODE **vlist, RBFNODE *v)
1150	+	{
1151	+	int     i, j;
1152	+
1153	+	for (i = 0; vlist[i] != NULL; i++) {
1154	+	if (v == vlist[i])
1155	+	return;
1156	+	if (v < vlist[i])
1157	+	break;
1158	+	}
1159	+	for (j = i; vlist[j] != NULL; j++)
1160	+	;
1161	+	while (j > i) {
1162	+	vlist[j] = vlist[j-1];
1163	+	--j;
1164	+	}
1165	+	vlist[i] = v;
1166	+	}
1167	+
1168	+	/* Sort triangle edges in standard order */
1169	+	static int
1170	+	order_triangle(MIGRATION *miga[3])
1171	+	{
1172	+	RBFNODE         *vert[7];
1173	+	MIGRATION       *ord[3];
1174	+	int             i;
1175	+	/* order vertices, first */
1176	+	memset(vert, 0, sizeof(vert));
1177	+	for (i = 3; i--; ) {
1178	+	if (miga[i] == NULL)
1179	+	return(0);
1180	+	insert_vert(vert, miga[i]->rbfv[0]);
1181	+	insert_vert(vert, miga[i]->rbfv[1]);
1182	+	}
1183	+	/* should be just 3 vertices */
1184	+	if ((vert[3] == NULL) | (vert[4] != NULL))
1185	+	return(0);
1186	+	/* identify edge 0 */
1187	+	for (i = 3; i--; )
1188	+	if (miga[i]->rbfv[0] == vert[0] &&
1189	+	miga[i]->rbfv[1] == vert[1]) {
1190	+	ord[0] = miga[i];
1191	+	break;
1192	+	}
1193	+	if (i < 0)
1194	+	return(0);
1195	+	/* identify edge 1 */
1196	+	for (i = 3; i--; )
1197	+	if (miga[i]->rbfv[0] == vert[1] &&
1198	+	miga[i]->rbfv[1] == vert[2]) {
1199	+	ord[1] = miga[i];
1200	+	break;
1201	+	}
1202	+	if (i < 0)
1203	+	return(0);
1204	+	/* identify edge 2 */
1205	+	for (i = 3; i--; )
1206	+	if (miga[i]->rbfv[0] == vert[0] &&
1207	+	miga[i]->rbfv[1] == vert[2]) {
1208	+	ord[2] = miga[i];
1209	+	break;
1210	+	}
1211	+	if (i < 0)
1212	+	return(0);
1213	+	/* reassign order */
1214	+	miga[0] = ord[0]; miga[1] = ord[1]; miga[2] = ord[2];
1215	+	return(1);
1216	+	}
1217	+
1218	+	/* Find edge(s) for interpolating the given incident vector */
1219	+	static int
1220	+	get_interp(MIGRATION *miga[3], const FVECT invec)
1221	+	{
1222	+	miga[0] = miga[1] = miga[2] = NULL;
1223	+	if (single_plane_incident) {            /* isotropic BSDF? */
1224	+	RBFNODE *rbf;                   /* find edge we're on */
1225	+	for (rbf = dsf_list; rbf != NULL; rbf = rbf->next) {
1226	+	if (input_orient*rbf->invec[2] < input_orient*invec[2])
1227	+	break;
1228	+	if (rbf->next != NULL &&
1229	+	input_orient*rbf->next->invec[2] <
1230	+	input_orient*invec[2]) {
1231	+	for (miga[0] = rbf->ejl; miga[0] != NULL;
1232	+	miga[0] = nextedge(rbf,miga[0]))
1233	+	if (opp_rbf(rbf,miga[0]) == rbf->next)
1234	+	return(1);
1235	+	break;
1236	+	}
1237	+	}
1238	+	return(0);                      /* outside range! */
1239	+	}
1240	+	{                                       /* else use triangle mesh */
1241	+	unsigned char   floodmap[GRIDRES][(GRIDRES+7)/8];
1242	+	int             pstart[2];
1243	+	RBFNODE         *vother;
1244	+	MIGRATION       *ej;
1245	+	int             i;
1246	+
1247	+	pos_from_vec(pstart, invec);
1248	+	memset(floodmap, 0, sizeof(floodmap));
1249	+	/* call flooding function */
1250	+	if (!identify_tri(miga, floodmap, pstart[0], pstart[1]))
1251	+	return(0);              /* outside mesh */
1252	+	if ((miga[0] == NULL) | (miga[2] == NULL))
1253	+	return(0);              /* should never happen */
1254	+	if (miga[1] == NULL)
1255	+	return(1);              /* on edge */
1256	+	/* verify triangle */
1257	+	if (!order_triangle(miga)) {
1258	+	#ifdef DEBUG
1259	+	fputs("Munged triangle in get_interp()\n", stderr);
1260	+	#endif
1261	+	vother = NULL;          /* find triangle from edge */
1262	+	for (i = 3; i--; ) {
1263	+	RBFNODE     *tpair[2];
1264	+	if (get_triangles(tpair, miga[i]) &&
1265	+	(vother = tpair[ is_rev_tri(
1266	+	miga[i]->rbfv[0]->invec,
1267	+	miga[i]->rbfv[1]->invec,
1268	+	invec) ]) != NULL)
1269	+	break;
1270	+	}
1271	+	if (vother == NULL) {   /* couldn't find 3rd vertex */
1272	+	#ifdef DEBUG
1273	+	fputs("No triangle in get_interp()\n", stderr);
1274	+	#endif
1275	+	return(0);
1276	+	}
1277	+	/* reassign other two edges */
1278	+	for (ej = vother->ejl; ej != NULL;
1279	+	ej = nextedge(vother,ej)) {
1280	+	RBFNODE *vorig = opp_rbf(vother,ej);
1281	+	if (vorig == miga[i]->rbfv[0])
1282	+	miga[(i+1)%3] = ej;
1283	+	else if (vorig == miga[i]->rbfv[1])
1284	+	miga[(i+2)%3] = ej;
1285	+	}
1286	+	if (!order_triangle(miga)) {
1287	+	#ifdef DEBUG
1288	+	fputs("Bad triangle in get_interp()\n", stderr);
1289	+	#endif
1290	+	return(0);
1291	+	}
1292	+	}
1293	+	return(3);                      /* return in standard order */
1294	+	}
1295	+	}
1296	+
1297	+	/* Advect and allocate new RBF along edge */
1298	+	static RBFNODE *
1299	+	e_advect_rbf(const MIGRATION *mig, const FVECT invec)
1300	+	{
1301	+	RBFNODE         *rbf;
1302	+	int             n, i, j;
1303	+	double          t, full_dist;
1304	+	/* get relative position */
1305	+	t = acos(DOT(invec, mig->rbfv[0]->invec));
1306	+	if (t < M_PI/GRIDRES) {                 /* near first DSF */
1307	+	n = sizeof(RBFNODE) + sizeof(RBFVAL)*(mig->rbfv[0]->nrbf-1);
1308	+	rbf = (RBFNODE *)malloc(n);
1309	+	if (rbf == NULL)
1310	+	goto memerr;
1311	+	memcpy(rbf, mig->rbfv[0], n);   /* just duplicate */
1312	+	return(rbf);
1313	+	}
1314	+	full_dist = acos(DOT(mig->rbfv[0]->invec, mig->rbfv[1]->invec));
1315	+	if (t > full_dist-M_PI/GRIDRES) {       /* near second DSF */
1316	+	n = sizeof(RBFNODE) + sizeof(RBFVAL)*(mig->rbfv[1]->nrbf-1);
1317	+	rbf = (RBFNODE *)malloc(n);
1318	+	if (rbf == NULL)
1319	+	goto memerr;
1320	+	memcpy(rbf, mig->rbfv[1], n);   /* just duplicate */
1321	+	return(rbf);
1322	+	}
1323	+	t /= full_dist;
1324	+	n = 0;                                  /* count migrating particles */
1325	+	for (i = 0; i < mtx_nrows(mig); i++)
1326	+	for (j = 0; j < mtx_ncols(mig); j++)
1327	+	n += (mig->mtx[mtx_ndx(mig,i,j)] > FTINY);
1328	+	#ifdef DEBUG
1329	+	fprintf(stderr, "Input RBFs have %d, %d nodes -> output has %d\n",
1330	+	mig->rbfv[0]->nrbf, mig->rbfv[1]->nrbf, n);
1331	+	#endif
1332	+	rbf = (RBFNODE *)malloc(sizeof(RBFNODE) + sizeof(RBFVAL)*(n-1));
1333	+	if (rbf == NULL)
1334	+	goto memerr;
1335	+	rbf->next = NULL; rbf->ejl = NULL;
1336	+	VCOPY(rbf->invec, invec);
1337	+	rbf->nrbf = n;
1338	+	rbf->vtotal = 1.-t + t*mig->rbfv[1]->vtotal/mig->rbfv[0]->vtotal;
1339	+	n = 0;                                  /* advect RBF lobes */
1340	+	for (i = 0; i < mtx_nrows(mig); i++) {
1341	+	const RBFVAL        *rbf0i = &mig->rbfv[0]->rbfa[i];
1342	+	const float         peak0 = rbf0i->peak;
1343	+	const double        rad0 = R2ANG(rbf0i->crad);
1344	+	FVECT               v0;
1345	+	float               mv;
1346	+	ovec_from_pos(v0, rbf0i->gx, rbf0i->gy);
1347	+	for (j = 0; j < mtx_ncols(mig); j++)
1348	+	if ((mv = mig->mtx[mtx_ndx(mig,i,j)]) > FTINY) {
1349	+	const RBFVAL    *rbf1j = &mig->rbfv[1]->rbfa[j];
1350	+	double          rad1 = R2ANG(rbf1j->crad);
1351	+	FVECT           v;
1352	+	int             pos[2];
1353	+	rbf->rbfa[n].peak = peak0 * mv * rbf->vtotal;
1354	+	rbf->rbfa[n].crad = ANG2R(sqrt(rad0*rad0*(1.-t) +
1355	+	rad1*rad1*t));
1356	+	ovec_from_pos(v, rbf1j->gx, rbf1j->gy);
1357	+	geodesic(v, v0, v, t, GEOD_REL);
1358	+	pos_from_vec(pos, v);
1359	+	rbf->rbfa[n].gx = pos[0];
1360	+	rbf->rbfa[n].gy = pos[1];
1361	+	++n;
1362	+	}
1363	+	}
1364	+	rbf->vtotal *= mig->rbfv[0]->vtotal;    /* turn ratio into actual */
1365	+	return(rbf);
1366	+	memerr:
1367	+	fputs("Out of memory in e_advect_rbf()\n", stderr);
1368	+	exit(1);
1369	+	return(NULL);   /* pro forma return */
1370	+	}
1371	+
1372	+	/* Partially advect between recorded incident angles and allocate new RBF */
1373	+	static RBFNODE *
1374	+	advect_rbf(const FVECT invec)
1375	+	{
1376	+	MIGRATION       *miga[3];
1377	+	RBFNODE         *rbf;
1378	+	float           mbfact, mcfact;
1379	+	int             n, i, j, k;
1380	+	FVECT           v0, v1, v2;
1381	+	double          s, t;
1382	+
1383	+	if (!get_interp(miga, invec))           /* can't interpolate? */
1384	+	return(NULL);
1385	+	if (miga[1] == NULL)                    /* advect along edge? */
1386	+	return(e_advect_rbf(miga[0], invec));
1387	+	#ifdef DEBUG
1388	+	if (miga[0]->rbfv[0] != miga[2]->rbfv[0] |
1389	+	miga[0]->rbfv[1] != miga[1]->rbfv[0] |
1390	+	miga[1]->rbfv[1] != miga[2]->rbfv[1]) {
1391	+	fputs("Triangle vertex screw-up!\n", stderr);
1392	+	exit(1);
1393	+	}
1394	+	#endif
1395	+	/* figure out position */
1396	+	fcross(v0, miga[2]->rbfv[0]->invec, miga[2]->rbfv[1]->invec);
1397	+	normalize(v0);
1398	+	fcross(v2, miga[1]->rbfv[0]->invec, miga[1]->rbfv[1]->invec);
1399	+	normalize(v2);
1400	+	fcross(v1, invec, miga[1]->rbfv[1]->invec);
1401	+	normalize(v1);
1402	+	s = acos(DOT(v0,v1)) / acos(DOT(v0,v2));
1403	+	geodesic(v1, miga[0]->rbfv[0]->invec, miga[0]->rbfv[1]->invec,
1404	+	s, GEOD_REL);
1405	+	t = acos(DOT(v1,invec)) / acos(DOT(v1,miga[1]->rbfv[1]->invec));
1406	+	n = 0;                                  /* count migrating particles */
1407	+	for (i = 0; i < mtx_nrows(miga[0]); i++)
1408	+	for (j = 0; j < mtx_ncols(miga[0]); j++)
1409	+	for (k = (miga[0]->mtx[mtx_ndx(miga[0],i,j)] > FTINY) *
1410	+	mtx_ncols(miga[2]); k--; )
1411	+	n += (miga[2]->mtx[mtx_ndx(miga[2],i,k)] > FTINY &&
1412	+	miga[1]->mtx[mtx_ndx(miga[1],j,k)] > FTINY);
1413	+	#ifdef DEBUG
1414	+	fprintf(stderr, "Input RBFs have %d, %d, %d nodes -> output has %d\n",
1415	+	miga[0]->rbfv[0]->nrbf, miga[0]->rbfv[1]->nrbf,
1416	+	miga[2]->rbfv[1]->nrbf, n);
1417	+	#endif
1418	+	rbf = (RBFNODE *)malloc(sizeof(RBFNODE) + sizeof(RBFVAL)*(n-1));
1419	+	if (rbf == NULL) {
1420	+	fputs("Out of memory in advect_rbf()\n", stderr);
1421	+	exit(1);
1422	+	}
1423	+	rbf->next = NULL; rbf->ejl = NULL;
1424	+	VCOPY(rbf->invec, invec);
1425	+	rbf->nrbf = n;
1426	+	n = 0;                                  /* compute RBF lobes */
1427	+	mbfact = s * miga[0]->rbfv[1]->vtotal/miga[0]->rbfv[0]->vtotal *
1428	+	(1.-t + t*miga[1]->rbfv[1]->vtotal/miga[1]->rbfv[0]->vtotal);
1429	+	mcfact = (1.-s) *
1430	+	(1.-t + t*miga[2]->rbfv[1]->vtotal/miga[2]->rbfv[0]->vtotal);
1431	+	for (i = 0; i < mtx_nrows(miga[0]); i++) {
1432	+	const RBFVAL        *rbf0i = &miga[0]->rbfv[0]->rbfa[i];
1433	+	const float         w0i = rbf0i->peak;
1434	+	const double        rad0i = R2ANG(rbf0i->crad);
1435	+	ovec_from_pos(v0, rbf0i->gx, rbf0i->gy);
1436	+	for (j = 0; j < mtx_ncols(miga[0]); j++) {
1437	+	const float     ma = miga[0]->mtx[mtx_ndx(miga[0],i,j)];
1438	+	const RBFVAL    *rbf1j;
1439	+	double          rad1j, srad2;
1440	+	if (ma <= FTINY)
1441	+	continue;
1442	+	rbf1j = &miga[0]->rbfv[1]->rbfa[j];
1443	+	rad1j = R2ANG(rbf1j->crad);
1444	+	srad2 = (1.-s)*(1.-t)*rad0i*rad0i + s*(1.-t)*rad1j*rad1j;
1445	+	ovec_from_pos(v1, rbf1j->gx, rbf1j->gy);
1446	+	geodesic(v1, v0, v1, s, GEOD_REL);
1447	+	for (k = 0; k < mtx_ncols(miga[2]); k++) {
1448	+	float               mb = miga[1]->mtx[mtx_ndx(miga[1],j,k)];
1449	+	float               mc = miga[2]->mtx[mtx_ndx(miga[2],i,k)];
1450	+	const RBFVAL        *rbf2k;
1451	+	double              rad2k;
1452	+	FVECT               vout;
1453	+	int                 pos[2];
1454	+	if ((mb <= FTINY) | (mc <= FTINY))
1455	+	continue;
1456	+	rbf2k = &miga[2]->rbfv[1]->rbfa[k];
1457	+	rbf->rbfa[n].peak = w0i * ma * (mb*mbfact + mc*mcfact);
1458	+	rad2k = R2ANG(rbf2k->crad);
1459	+	rbf->rbfa[n].crad = ANG2R(sqrt(srad2 + t*rad2k*rad2k));
1460	+	ovec_from_pos(v2, rbf2k->gx, rbf2k->gy);
1461	+	geodesic(vout, v1, v2, t, GEOD_REL);
1462	+	pos_from_vec(pos, vout);
1463	+	rbf->rbfa[n].gx = pos[0];
1464	+	rbf->rbfa[n].gy = pos[1];
1465	+	++n;
1466	+	}
1467	+	}
1468	+	}
1469	+	rbf->vtotal = miga[0]->rbfv[0]->vtotal * (mbfact + mcfact);
1470	+	return(rbf);
1471	+	}
1472	+
1473	+	/* Interpolate and output isotropic BSDF data */
1474	+	static void
1475	+	interp_isotropic()
1476	+	{
1477	+	const int       sqres = 1<<samp_order;
1478	+	FILE            *ofp = NULL;
1479	+	char            cmd[128];
1480	+	int             ix, ox, oy;
1481	+	FVECT           ivec, ovec;
1482	+	double          bsdf;
1483	+	#if DEBUG
1484	+	fprintf(stderr, "Writing isotropic order %d ", samp_order);
1485	+	if (pctcull >= 0) fprintf(stderr, "data with %d%% culling\n", pctcull);
1486	+	else fputs("raw data\n", stderr);
1487	+	#endif
1488	+	if (pctcull >= 0) {                     /* begin output */
1489	+	sprintf(cmd, "rttree_reduce -h -a -fd -r 3 -t %d -g %d",
1490	+	pctcull, samp_order);
1491	+	fflush(stdout);
1492	+	ofp = popen(cmd, "w");
1493	+	if (ofp == NULL) {
1494	+	fprintf(stderr, "%s: cannot create pipe to rttree_reduce\n",
1495	+	progname);
1496	+	exit(1);
1497	+	}
1498	+	} else
1499	+	fputs("{\n", stdout);
1500	+	/* run through directions */
1501	+	for (ix = 0; ix < sqres/2; ix++) {
1502	+	RBFNODE *rbf;
1503	+	SDsquare2disk(ivec, (ix+.5)/sqres, .5);
1504	+	ivec[2] = input_orient *
1505	+	sqrt(1. - ivec[0]*ivec[0] - ivec[1]*ivec[1]);
1506	+	rbf = advect_rbf(ivec);
1507	+	for (ox = 0; ox < sqres; ox++)
1508	+	for (oy = 0; oy < sqres; oy++) {
1509	+	SDsquare2disk(ovec, (ox+.5)/sqres, (oy+.5)/sqres);
1510	+	ovec[2] = output_orient *
1511	+	sqrt(1. - ovec[0]*ovec[0] - ovec[1]*ovec[1]);
1512	+	bsdf = eval_rbfrep(rbf, ovec) / fabs(ovec[2]);
1513	+	if (pctcull >= 0)
1514	+	fwrite(&bsdf, sizeof(bsdf), 1, ofp);
1515	+	else
1516	+	printf("\t%.3e\n", bsdf);
1517	+	}
1518	+	free(rbf);
1519	+	}
1520	+	if (pctcull >= 0) {                     /* finish output */
1521	+	if (pclose(ofp)) {
1522	+	fprintf(stderr, "%s: error running '%s'\n",
1523	+	progname, cmd);
1524	+	exit(1);
1525	+	}
1526	+	} else {
1527	+	for (ix = sqres*sqres*sqres/2; ix--; )
1528	+	fputs("\t0\n", stdout);
1529	+	fputs("}\n", stdout);
1530	+	}
1531	+	}
1532	+
1533	+	/* Interpolate and output anisotropic BSDF data */
1534	+	static void
1535	+	interp_anisotropic()
1536	+	{
1537	+	const int       sqres = 1<<samp_order;
1538	+	FILE            *ofp = NULL;
1539	+	char            cmd[128];
1540	+	int             ix, iy, ox, oy;
1541	+	FVECT           ivec, ovec;
1542	+	double          bsdf;
1543	+	#if DEBUG
1544	+	fprintf(stderr, "Writing anisotropic order %d ", samp_order);
1545	+	if (pctcull >= 0) fprintf(stderr, "data with %d%% culling\n", pctcull);
1546	+	else fputs("raw data\n", stderr);
1547	+	#endif
1548	+	if (pctcull >= 0) {                     /* begin output */
1549	+	sprintf(cmd, "rttree_reduce -h -a -fd -r 4 -t %d -g %d",
1550	+	pctcull, samp_order);
1551	+	fflush(stdout);
1552	+	ofp = popen(cmd, "w");
1553	+	if (ofp == NULL) {
1554	+	fprintf(stderr, "%s: cannot create pipe to rttree_reduce\n",
1555	+	progname);
1556	+	exit(1);
1557	+	}
1558	+	} else
1559	+	fputs("{\n", stdout);
1560	+	/* run through directions */
1561	+	for (ix = 0; ix < sqres; ix++)
1562	+	for (iy = 0; iy < sqres; iy++) {
1563	+	RBFNODE *rbf;
1564	+	SDsquare2disk(ivec, (ix+.5)/sqres, (iy+.5)/sqres);
1565	+	ivec[2] = input_orient *
1566	+	sqrt(1. - ivec[0]*ivec[0] - ivec[1]*ivec[1]);
1567	+	rbf = advect_rbf(ivec);
1568	+	for (ox = 0; ox < sqres; ox++)
1569	+	for (oy = 0; oy < sqres; oy++) {
1570	+	SDsquare2disk(ovec, (ox+.5)/sqres, (oy+.5)/sqres);
1571	+	ovec[2] = output_orient *
1572	+	sqrt(1. - ovec[0]*ovec[0] - ovec[1]*ovec[1]);
1573	+	bsdf = eval_rbfrep(rbf, ovec) / fabs(ovec[2]);
1574	+	if (pctcull >= 0)
1575	+	fwrite(&bsdf, sizeof(bsdf), 1, ofp);
1576	+	else
1577	+	printf("\t%.3e\n", bsdf);
1578	+	}
1579	+	free(rbf);
1580	+	}
1581	+	if (pctcull >= 0) {                     /* finish output */
1582	+	if (pclose(ofp)) {
1583	+	fprintf(stderr, "%s: error running '%s'\n",
1584	+	progname, cmd);
1585	+	exit(1);
1586	+	}
1587	+	} else
1588	+	fputs("}\n", stdout);
1589	+	}
1590	+
1591		#if 1
1592	+	/* Read in BSDF files and interpolate as tensor tree representation */
1593	+	int
1594	+	main(int argc, char *argv[])
1595	+	{
1596	+	RBFNODE         *rbf;
1597	+	double          bsdf;
1598	+	int             i;
1599	+
1600	+	progname = argv[0];                     /* get options */
1601	+	while (argc > 2 && argv[1][0] == '-') {
1602	+	switch (argv[1][1]) {
1603	+	case 'n':
1604	+	nprocs = atoi(argv[2]);
1605	+	break;
1606	+	case 't':
1607	+	pctcull = atoi(argv[2]);
1608	+	break;
1609	+	default:
1610	+	goto userr;
1611	+	}
1612	+	argv += 2; argc -= 2;
1613	+	}
1614	+	if (argc < 3)
1615	+	goto userr;
1616	+	#ifdef _WIN32
1617	+	if (nprocs > 1) {
1618	+	fprintf(stderr, "%s: multiprocessing not supported\n",
1619	+	progname);
1620	+	return(1);
1621	+	}
1622	+	#endif
1623	+	for (i = 1; i < argc; i++) {            /* compile measurements */
1624	+	if (!load_pabopto_meas(argv[i]))
1625	+	return(1);
1626	+	compute_radii();
1627	+	cull_values();
1628	+	make_rbfrep();
1629	+	}
1630	+	build_mesh();                           /* create interpolation */
1631	+	/* xml_prologue();                              /* start XML output */
1632	+	if (single_plane_incident)              /* resample dist. */
1633	+	interp_isotropic();
1634	+	else
1635	+	interp_anisotropic();
1636	+	/* xml_epilogue();                              /* finish XML output */
1637	+	return(0);
1638	+	userr:
1639	+	fprintf(stderr,
1640	+	"Usage: %s [-n nprocs][-t pctcull] meas1.dat meas2.dat .. > bsdf.xml\n",
1641	+	progname);
1642	+	return(1);
1643	+	}
1644	+	#else
1645		/* Test main produces a Radiance model from the given input file */
1646		int
1647		main(int argc, char *argv[])
#	Line 401 | Line 1656 | main(int argc, char *argv[])
1656		fprintf(stderr, "Usage: %s input.dat > output.rad\n", argv[0]);
1657		return(1);
1658		}
1659	<	if (!load_bsdf_meas(argv[1]))
1659	>	if (!load_pabopto_meas(argv[1]))
1660		return(1);
1661
1662		compute_radii();
#	Line 414 | Line 1669 | main(int argc, char *argv[])
1669		for (i = 0; i < GRIDRES; i++)
1670		for (j = 0; j < GRIDRES; j++)
1671		if (dsf_grid[i][j].vsum > .0f) {
1672	<	vec_from_pos(dir, i, j);
1672	>	ovec_from_pos(dir, i, j);
1673		bsdf = dsf_grid[i][j].vsum / dir[2];
1674		if (dsf_grid[i][j].nval) {
1675		printf("pink cone c%04d\n0\n0\n8\n", ++n);
#	Line 425 | Line 1680 | main(int argc, char *argv[])
1680		dir[2]*(bsdf+.005));
1681		puts("\t.003\t0\n");
1682		} else {
1683	<	vec_from_pos(dir, i, j);
1683	>	ovec_from_pos(dir, i, j);
1684		printf("yellow sphere s%04d\n0\n0\n", ++n);
1685		printf("4 %.6g %.6g %.6g .0015\n\n",
1686		dir[0]*bsdf, dir[1]*bsdf, dir[2]*bsdf);
#	Line 443 | Line 1698 | main(int argc, char *argv[])
1698		}
1699		for (i = 0; i < GRIDRES; i++)
1700		for (j = 0; j < GRIDRES; j++) {
1701	<	vec_from_pos(dir, i, j);
1701	>	ovec_from_pos(dir, i, j);
1702		bsdf = eval_rbfrep(dsf_list, dir) / dir[2];
1703		fprintf(pfp, "%.8e %.8e %.8e\n",
1704		dir[0]*bsdf, dir[1]*bsdf, dir[2]*bsdf);

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines