ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/radiance/ray/src/rt/ambcomp.c
Revision: 2.39
Committed: Tue Apr 29 15:40:00 2014 UTC (10 years ago) by greg
Content type: text/plain
Branch: MAIN
Changes since 2.38: +2 -2 lines
Log Message: 
Fixed bug that was causing segmentation violations for all-black scenes

File Contents

# User Rev Content
1 greg 1.1 #ifndef lint
2 greg 2.39 static const char RCSid[] = "$Id: ambcomp.c,v 2.38 2014/04/26 15:54:17 greg Exp $";
3 greg 1.1 #endif
4     /*
5     * Routines to compute "ambient" values using Monte Carlo
6 greg 2.9 *
7 greg 2.27 * Hessian calculations based on "Practical Hessian-Based Error Control
8     * for Irradiance Caching" by Schwarzhaupt, Wann Jensen, & Jarosz
9     * from ACM SIGGRAPH Asia 2012 conference proceedings.
10     *
11 greg 2.9 * Declarations of external symbols in ambient.h
12     */
13    
14 greg 2.10 #include "copyright.h"
15 greg 1.1
16     #include "ray.h"
17 greg 2.25 #include "ambient.h"
18     #include "random.h"
19 greg 1.1
20 greg 2.25 #ifdef NEWAMB
21 greg 1.1
22 greg 2.26 extern void SDsquare2disk(double ds[2], double seedx, double seedy);
23    
24     typedef struct {
25     RAY *rp; /* originating ray sample */
26 greg 2.27 FVECT ux, uy; /* tangent axis unit vectors */
27 greg 2.26 int ns; /* number of samples per axis */
28     COLOR acoef; /* division contribution coefficient */
29     struct s_ambsamp {
30     COLOR v; /* hemisphere sample value */
31 greg 2.31 FVECT p; /* intersection point */
32 greg 2.26 } sa[1]; /* sample array (extends struct) */
33     } AMBHEMI; /* ambient sample hemisphere */
34    
35     #define ambsamp(h,i,j) (h)->sa[(i)*(h)->ns + (j)]
36    
37 greg 2.27 typedef struct {
38 greg 2.35 FVECT r_i, r_i1, e_i, rcp, rI2_eJ2;
39     double I1, I2;
40 greg 2.27 } FFTRI; /* vectors and coefficients for Hessian calculation */
41    
42 greg 2.26
43     static AMBHEMI *
44     inithemi( /* initialize sampling hemisphere */
45     COLOR ac,
46     RAY *r,
47     double wt
48     )
49     {
50     AMBHEMI *hp;
51     double d;
52     int n, i;
53     /* set number of divisions */
54     if (ambacc <= FTINY &&
55     wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight)))
56     wt = d; /* avoid ray termination */
57     n = sqrt(ambdiv * wt) + 0.5;
58 greg 2.27 i = 1 + 5*(ambacc > FTINY); /* minimum number of samples */
59 greg 2.26 if (n < i)
60     n = i;
61     /* allocate sampling array */
62     hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) +
63     sizeof(struct s_ambsamp)*(n*n - 1));
64     if (hp == NULL)
65     return(NULL);
66     hp->rp = r;
67     hp->ns = n;
68     /* assign coefficient */
69     copycolor(hp->acoef, ac);
70     d = 1.0/(n*n);
71     scalecolor(hp->acoef, d);
72 greg 2.28 /* make tangent plane axes */
73 greg 2.38 hp->uy[0] = 0.5 - frandom();
74     hp->uy[1] = 0.5 - frandom();
75     hp->uy[2] = 0.5 - frandom();
76 greg 2.36 for (i = 3; i--; )
77 greg 2.37 if ((-0.6 < r->ron[i]) & (r->ron[i] < 0.6))
78     break;
79     if (i < 0)
80     error(CONSISTENCY, "bad ray direction in inithemi");
81     hp->uy[i] = 1.0;
82 greg 2.27 VCROSS(hp->ux, hp->uy, r->ron);
83 greg 2.26 normalize(hp->ux);
84 greg 2.27 VCROSS(hp->uy, r->ron, hp->ux);
85 greg 2.26 /* we're ready to sample */
86     return(hp);
87     }
88    
89    
90 greg 2.28 static struct s_ambsamp *
91 greg 2.26 ambsample( /* sample an ambient direction */
92     AMBHEMI *hp,
93     int i,
94 greg 2.27 int j
95 greg 2.26 )
96     {
97     struct s_ambsamp *ap = &ambsamp(hp,i,j);
98     RAY ar;
99 greg 2.27 double spt[2], zd;
100 greg 2.26 int ii;
101     /* ambient coefficient for weight */
102     if (ambacc > FTINY)
103     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
104     else
105     copycolor(ar.rcoef, hp->acoef);
106 greg 2.31 if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0)
107     goto badsample;
108 greg 2.26 if (ambacc > FTINY) {
109     multcolor(ar.rcoef, hp->acoef);
110     scalecolor(ar.rcoef, 1./AVGREFL);
111     }
112     /* generate hemispherical sample */
113 greg 2.27 SDsquare2disk(spt, (i+.1+.8*frandom())/hp->ns,
114 greg 2.28 (j+.1+.8*frandom())/hp->ns );
115 greg 2.26 zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]);
116     for (ii = 3; ii--; )
117     ar.rdir[ii] = spt[0]*hp->ux[ii] +
118     spt[1]*hp->uy[ii] +
119     zd*hp->rp->ron[ii];
120     checknorm(ar.rdir);
121     dimlist[ndims++] = i*hp->ns + j + 90171;
122     rayvalue(&ar); /* evaluate ray */
123     ndims--;
124 greg 2.34 /* limit vertex distance */
125     if (ar.rt > 10.0*thescene.cusize)
126     ar.rt = 10.0*thescene.cusize;
127 greg 2.31 else if (ar.rt <= FTINY) /* should never happen! */
128     goto badsample;
129     VSUM(ap->p, ar.rorg, ar.rdir, ar.rt);
130 greg 2.26 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
131     copycolor(ap->v, ar.rcol);
132 greg 2.28 return(ap);
133 greg 2.31 badsample:
134     setcolor(ap->v, 0., 0., 0.);
135     VCOPY(ap->p, hp->rp->rop);
136     return(NULL);
137 greg 2.26 }
138    
139    
140 greg 2.27 /* Compute vectors and coefficients for Hessian/gradient calcs */
141     static void
142 greg 2.31 comp_fftri(FFTRI *ftp, FVECT ap0, FVECT ap1, FVECT rop)
143 greg 2.27 {
144 greg 2.35 double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2;
145 greg 2.30 int i;
146 greg 2.27
147     VSUB(ftp->r_i, ap0, rop);
148     VSUB(ftp->r_i1, ap1, rop);
149     VSUB(ftp->e_i, ap1, ap0);
150 greg 2.35 VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1);
151     rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp);
152 greg 2.27 dot_e = DOT(ftp->e_i,ftp->e_i);
153     dot_er = DOT(ftp->e_i, ftp->r_i);
154 greg 2.32 rdot_r = 1.0/DOT(ftp->r_i,ftp->r_i);
155     rdot_r1 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
156     ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) * sqrt(rdot_r*rdot_r1) ) *
157 greg 2.35 sqrt( rdot_cp );
158 greg 2.32 ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r +
159 greg 2.35 dot_e*ftp->I1 )*0.5*rdot_cp;
160 greg 2.32 J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e;
161 greg 2.30 for (i = 3; i--; )
162     ftp->rI2_eJ2[i] = ftp->I2*ftp->r_i[i] + J2*ftp->e_i[i];
163 greg 2.27 }
164    
165    
166 greg 2.28 /* Compose 3x3 matrix from two vectors */
167 greg 2.27 static void
168     compose_matrix(FVECT mat[3], FVECT va, FVECT vb)
169     {
170     mat[0][0] = 2.0*va[0]*vb[0];
171     mat[1][1] = 2.0*va[1]*vb[1];
172     mat[2][2] = 2.0*va[2]*vb[2];
173     mat[0][1] = mat[1][0] = va[0]*vb[1] + va[1]*vb[0];
174     mat[0][2] = mat[2][0] = va[0]*vb[2] + va[2]*vb[0];
175     mat[1][2] = mat[2][1] = va[1]*vb[2] + va[2]*vb[1];
176     }
177    
178    
179     /* Compute partial 3x3 Hessian matrix for edge */
180     static void
181     comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm)
182     {
183 greg 2.35 FVECT ncp;
184 greg 2.27 FVECT m1[3], m2[3], m3[3], m4[3];
185     double d1, d2, d3, d4;
186     double I3, J3, K3;
187     int i, j;
188     /* compute intermediate coefficients */
189     d1 = 1.0/DOT(ftp->r_i,ftp->r_i);
190     d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
191     d3 = 1.0/DOT(ftp->e_i,ftp->e_i);
192     d4 = DOT(ftp->e_i, ftp->r_i);
193 greg 2.35 I3 = ( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + 3.0/d3*ftp->I2 )
194     / ( 4.0*DOT(ftp->rcp,ftp->rcp) );
195 greg 2.27 J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3;
196     K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3);
197     /* intermediate matrices */
198 greg 2.35 VCROSS(ncp, nrm, ftp->e_i);
199     compose_matrix(m1, ncp, ftp->rI2_eJ2);
200 greg 2.27 compose_matrix(m2, ftp->r_i, ftp->r_i);
201     compose_matrix(m3, ftp->e_i, ftp->e_i);
202     compose_matrix(m4, ftp->r_i, ftp->e_i);
203 greg 2.35 d1 = DOT(nrm, ftp->rcp);
204 greg 2.27 d2 = -d1*ftp->I2;
205     d1 *= 2.0;
206     for (i = 3; i--; ) /* final matrix sum */
207     for (j = 3; j--; ) {
208     hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] +
209     2.0*J3*m4[i][j] );
210     hess[i][j] += d2*(i==j);
211 greg 2.32 hess[i][j] *= 1.0/PI;
212 greg 2.27 }
213     }
214    
215    
216     /* Reverse hessian calculation result for edge in other direction */
217     static void
218     rev_hessian(FVECT hess[3])
219     {
220     int i;
221    
222     for (i = 3; i--; ) {
223     hess[i][0] = -hess[i][0];
224     hess[i][1] = -hess[i][1];
225     hess[i][2] = -hess[i][2];
226     }
227     }
228    
229    
230     /* Add to radiometric Hessian from the given triangle */
231     static void
232     add2hessian(FVECT hess[3], FVECT ehess1[3],
233     FVECT ehess2[3], FVECT ehess3[3], COLORV v)
234     {
235     int i, j;
236    
237     for (i = 3; i--; )
238     for (j = 3; j--; )
239     hess[i][j] += v*( ehess1[i][j] + ehess2[i][j] + ehess3[i][j] );
240     }
241    
242    
243     /* Compute partial displacement form factor gradient for edge */
244     static void
245     comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm)
246     {
247 greg 2.35 FVECT ncp;
248 greg 2.27 double f1;
249     int i;
250    
251 greg 2.35 f1 = 2.0*DOT(nrm, ftp->rcp);
252     VCROSS(ncp, nrm, ftp->e_i);
253 greg 2.27 for (i = 3; i--; )
254 greg 2.35 grad[i] = (-0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] );
255 greg 2.27 }
256    
257    
258     /* Reverse gradient calculation result for edge in other direction */
259     static void
260     rev_gradient(FVECT grad)
261     {
262     grad[0] = -grad[0];
263     grad[1] = -grad[1];
264     grad[2] = -grad[2];
265     }
266    
267    
268     /* Add to displacement gradient from the given triangle */
269     static void
270     add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v)
271     {
272     int i;
273    
274     for (i = 3; i--; )
275     grad[i] += v*( egrad1[i] + egrad2[i] + egrad3[i] );
276     }
277    
278    
279     /* Return brightness of furthest ambient sample */
280     static COLORV
281     back_ambval(struct s_ambsamp *ap1, struct s_ambsamp *ap2,
282     struct s_ambsamp *ap3, FVECT orig)
283     {
284     COLORV vback;
285     FVECT vec;
286     double d2, d2best;
287    
288     VSUB(vec, ap1->p, orig);
289     d2best = DOT(vec,vec);
290 greg 2.29 vback = colval(ap1->v,CIEY);
291 greg 2.27 VSUB(vec, ap2->p, orig);
292     d2 = DOT(vec,vec);
293     if (d2 > d2best) {
294     d2best = d2;
295 greg 2.29 vback = colval(ap2->v,CIEY);
296 greg 2.27 }
297     VSUB(vec, ap3->p, orig);
298     d2 = DOT(vec,vec);
299     if (d2 > d2best)
300 greg 2.29 return(colval(ap3->v,CIEY));
301 greg 2.27 return(vback);
302     }
303    
304    
305     /* Compute anisotropic radii and eigenvector directions */
306     static int
307     eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3])
308     {
309     double hess2[2][2];
310     FVECT a, b;
311     double evalue[2], slope1, xmag1;
312     int i;
313     /* project Hessian to sample plane */
314     for (i = 3; i--; ) {
315     a[i] = DOT(hessian[i], uv[0]);
316     b[i] = DOT(hessian[i], uv[1]);
317     }
318     hess2[0][0] = DOT(uv[0], a);
319     hess2[0][1] = DOT(uv[0], b);
320     hess2[1][0] = DOT(uv[1], a);
321     hess2[1][1] = DOT(uv[1], b);
322 greg 2.38 /* compute eigenvalue(s) */
323     i = quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1],
324     hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]);
325     if (i == 1) /* double-root (circle) */
326     evalue[1] = evalue[0];
327     if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) |
328 greg 2.35 ((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) )
329 greg 2.27 error(INTERNAL, "bad eigenvalue calculation");
330    
331     if (evalue[0] > evalue[1]) {
332 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[0]));
333     ra[1] = sqrt(sqrt(4.0/evalue[1]));
334 greg 2.27 slope1 = evalue[1];
335     } else {
336 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[1]));
337     ra[1] = sqrt(sqrt(4.0/evalue[0]));
338 greg 2.27 slope1 = evalue[0];
339     }
340     /* compute unit eigenvectors */
341     if (fabs(hess2[0][1]) <= FTINY)
342     return; /* uv OK as is */
343     slope1 = (slope1 - hess2[0][0]) / hess2[0][1];
344     xmag1 = sqrt(1.0/(1.0 + slope1*slope1));
345     for (i = 3; i--; ) {
346     b[i] = xmag1*uv[0][i] + slope1*xmag1*uv[1][i];
347     a[i] = slope1*xmag1*uv[0][i] - xmag1*uv[1][i];
348     }
349     VCOPY(uv[0], a);
350     VCOPY(uv[1], b);
351     }
352    
353    
354 greg 2.26 static void
355     ambHessian( /* anisotropic radii & pos. gradient */
356     AMBHEMI *hp,
357     FVECT uv[2], /* returned */
358 greg 2.28 float ra[2], /* returned (optional) */
359     float pg[2] /* returned (optional) */
360 greg 2.26 )
361     {
362 greg 2.27 static char memerrmsg[] = "out of memory in ambHessian()";
363     FVECT (*hessrow)[3] = NULL;
364     FVECT *gradrow = NULL;
365     FVECT hessian[3];
366     FVECT gradient;
367     FFTRI fftr;
368     int i, j;
369     /* be sure to assign unit vectors */
370     VCOPY(uv[0], hp->ux);
371     VCOPY(uv[1], hp->uy);
372     /* clock-wise vertex traversal from sample POV */
373     if (ra != NULL) { /* initialize Hessian row buffer */
374 greg 2.28 hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1));
375 greg 2.27 if (hessrow == NULL)
376     error(SYSTEM, memerrmsg);
377     memset(hessian, 0, sizeof(hessian));
378     } else if (pg == NULL) /* bogus call? */
379     return;
380     if (pg != NULL) { /* initialize form factor row buffer */
381 greg 2.28 gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1));
382 greg 2.27 if (gradrow == NULL)
383     error(SYSTEM, memerrmsg);
384     memset(gradient, 0, sizeof(gradient));
385     }
386     /* compute first row of edges */
387     for (j = 0; j < hp->ns-1; j++) {
388     comp_fftri(&fftr, ambsamp(hp,0,j).p,
389     ambsamp(hp,0,j+1).p, hp->rp->rop);
390     if (hessrow != NULL)
391     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
392     if (gradrow != NULL)
393     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
394     }
395     /* sum each row of triangles */
396     for (i = 0; i < hp->ns-1; i++) {
397     FVECT hesscol[3]; /* compute first vertical edge */
398     FVECT gradcol;
399     comp_fftri(&fftr, ambsamp(hp,i,0).p,
400     ambsamp(hp,i+1,0).p, hp->rp->rop);
401     if (hessrow != NULL)
402     comp_hessian(hesscol, &fftr, hp->rp->ron);
403     if (gradrow != NULL)
404     comp_gradient(gradcol, &fftr, hp->rp->ron);
405     for (j = 0; j < hp->ns-1; j++) {
406     FVECT hessdia[3]; /* compute triangle contributions */
407     FVECT graddia;
408     COLORV backg;
409     backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1),
410     &ambsamp(hp,i+1,j), hp->rp->rop);
411     /* diagonal (inner) edge */
412     comp_fftri(&fftr, ambsamp(hp,i,j+1).p,
413     ambsamp(hp,i+1,j).p, hp->rp->rop);
414     if (hessrow != NULL) {
415     comp_hessian(hessdia, &fftr, hp->rp->ron);
416     rev_hessian(hesscol);
417     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
418     }
419 greg 2.39 if (gradrow != NULL) {
420 greg 2.27 comp_gradient(graddia, &fftr, hp->rp->ron);
421     rev_gradient(gradcol);
422     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
423     }
424     /* initialize edge in next row */
425     comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p,
426     ambsamp(hp,i+1,j).p, hp->rp->rop);
427     if (hessrow != NULL)
428     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
429     if (gradrow != NULL)
430     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
431     /* new column edge & paired triangle */
432     backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1),
433     &ambsamp(hp,i+1,j), hp->rp->rop);
434     comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p,
435     hp->rp->rop);
436     if (hessrow != NULL) {
437     comp_hessian(hesscol, &fftr, hp->rp->ron);
438     rev_hessian(hessdia);
439     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
440     if (i < hp->ns-2)
441     rev_hessian(hessrow[j]);
442     }
443     if (gradrow != NULL) {
444     comp_gradient(gradcol, &fftr, hp->rp->ron);
445     rev_gradient(graddia);
446     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
447     if (i < hp->ns-2)
448     rev_gradient(gradrow[j]);
449     }
450     }
451     }
452     /* release row buffers */
453     if (hessrow != NULL) free(hessrow);
454     if (gradrow != NULL) free(gradrow);
455    
456     if (ra != NULL) /* extract eigenvectors & radii */
457     eigenvectors(uv, ra, hessian);
458 greg 2.32 if (pg != NULL) { /* tangential position gradient */
459     pg[0] = DOT(gradient, uv[0]);
460     pg[1] = DOT(gradient, uv[1]);
461 greg 2.27 }
462     }
463    
464    
465     /* Compute direction gradient from a hemispherical sampling */
466     static void
467     ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2])
468     {
469     struct s_ambsamp *ap;
470 greg 2.29 double dgsum[2];
471 greg 2.27 int n;
472 greg 2.28 FVECT vd;
473     double gfact;
474 greg 2.27
475 greg 2.29 dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */
476 greg 2.27 for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) {
477     /* use vector for azimuth + 90deg */
478     VSUB(vd, ap->p, hp->rp->rop);
479 greg 2.29 /* brightness over cosine factor */
480     gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd);
481     /* -sine = -proj_radius/vd_length */
482     dgsum[0] += DOT(uv[1], vd) * gfact;
483     dgsum[1] -= DOT(uv[0], vd) * gfact;
484 greg 2.26 }
485 greg 2.29 dg[0] = dgsum[0] / (hp->ns*hp->ns);
486     dg[1] = dgsum[1] / (hp->ns*hp->ns);
487 greg 2.26 }
488    
489 greg 2.27
490 greg 2.26 int
491     doambient( /* compute ambient component */
492     COLOR rcol, /* input/output color */
493     RAY *r,
494     double wt,
495 greg 2.27 FVECT uv[2], /* returned (optional) */
496     float ra[2], /* returned (optional) */
497     float pg[2], /* returned (optional) */
498     float dg[2] /* returned (optional) */
499 greg 2.26 )
500     {
501 greg 2.28 AMBHEMI *hp = inithemi(rcol, r, wt);
502 greg 2.26 int cnt = 0;
503     FVECT my_uv[2];
504 greg 2.38 double d, K, acol[3];
505 greg 2.26 struct s_ambsamp *ap;
506     int i, j;
507 greg 2.28 /* check/initialize */
508     if (hp == NULL)
509 greg 2.26 return(0);
510     if (uv != NULL)
511     memset(uv, 0, sizeof(FVECT)*2);
512     if (ra != NULL)
513     ra[0] = ra[1] = 0.0;
514     if (pg != NULL)
515     pg[0] = pg[1] = 0.0;
516     if (dg != NULL)
517     dg[0] = dg[1] = 0.0;
518     /* sample the hemisphere */
519     acol[0] = acol[1] = acol[2] = 0.0;
520 greg 2.27 for (i = hp->ns; i--; )
521     for (j = hp->ns; j--; )
522 greg 2.28 if ((ap = ambsample(hp, i, j)) != NULL) {
523 greg 2.26 addcolor(acol, ap->v);
524     ++cnt;
525     }
526     if (!cnt) {
527     setcolor(rcol, 0.0, 0.0, 0.0);
528     free(hp);
529     return(0); /* no valid samples */
530     }
531 greg 2.29 copycolor(rcol, acol); /* final indirect irradiance/PI */
532 greg 2.26 if (cnt < hp->ns*hp->ns || /* incomplete sampling? */
533     (ra == NULL) & (pg == NULL) & (dg == NULL)) {
534     free(hp);
535     return(-1); /* no radius or gradient calc. */
536     }
537 greg 2.38 if (bright(acol) > FTINY) { /* normalize Y values */
538     d = 0.99*cnt/bright(acol);
539     K = 0.01;
540     } else { /* geometric Hessian fall-back */
541 greg 2.29 d = 0.0;
542 greg 2.38 K = 1.0;
543     pg = NULL;
544     dg = NULL;
545     }
546 greg 2.29 ap = hp->sa; /* relative Y channel from here on... */
547 greg 2.26 for (i = hp->ns*hp->ns; i--; ap++)
548 greg 2.38 colval(ap->v,CIEY) = bright(ap->v)*d + K;
549 greg 2.26
550     if (uv == NULL) /* make sure we have axis pointers */
551     uv = my_uv;
552     /* compute radii & pos. gradient */
553     ambHessian(hp, uv, ra, pg);
554 greg 2.29
555 greg 2.26 if (dg != NULL) /* compute direction gradient */
556     ambdirgrad(hp, uv, dg);
557 greg 2.29
558 greg 2.28 if (ra != NULL) { /* scale/clamp radii */
559 greg 2.35 if (pg != NULL) {
560     if (ra[0]*(d = fabs(pg[0])) > 1.0)
561     ra[0] = 1.0/d;
562     if (ra[1]*(d = fabs(pg[1])) > 1.0)
563     ra[1] = 1.0/d;
564     if (ra[0] > ra[1])
565     ra[0] = ra[1];
566     }
567 greg 2.29 if (ra[0] < minarad) {
568     ra[0] = minarad;
569     if (ra[1] < minarad)
570     ra[1] = minarad;
571     }
572     ra[0] *= d = 1.0/sqrt(sqrt(wt));
573 greg 2.26 if ((ra[1] *= d) > 2.0*ra[0])
574     ra[1] = 2.0*ra[0];
575 greg 2.28 if (ra[1] > maxarad) {
576     ra[1] = maxarad;
577     if (ra[0] > maxarad)
578     ra[0] = maxarad;
579     }
580 greg 2.35 if (pg != NULL) { /* cap gradient if necessary */
581     d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1];
582     if (d > 1.0) {
583     d = 1.0/sqrt(d);
584     pg[0] *= d;
585     pg[1] *= d;
586     }
587     }
588 greg 2.26 }
589     free(hp); /* clean up and return */
590     return(1);
591     }
592    
593    
594 greg 2.25 #else /* ! NEWAMB */
595 greg 1.1
596    
597 greg 2.15 void
598 greg 2.14 inithemi( /* initialize sampling hemisphere */
599 greg 2.23 AMBHEMI *hp,
600 greg 2.16 COLOR ac,
601 greg 2.14 RAY *r,
602     double wt
603     )
604 greg 1.1 {
605 greg 2.16 double d;
606 greg 2.23 int i;
607 greg 2.14 /* set number of divisions */
608 greg 2.16 if (ambacc <= FTINY &&
609 greg 2.20 wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight)))
610 greg 2.16 wt = d; /* avoid ray termination */
611     hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
612 greg 2.14 i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */
613     if (hp->nt < i)
614     hp->nt = i;
615     hp->np = PI * hp->nt + 0.5;
616     /* set number of super-samples */
617 greg 2.15 hp->ns = ambssamp * wt + 0.5;
618 greg 2.16 /* assign coefficient */
619 greg 2.14 copycolor(hp->acoef, ac);
620 greg 2.16 d = 1.0/(hp->nt*hp->np);
621     scalecolor(hp->acoef, d);
622 greg 2.14 /* make axes */
623     VCOPY(hp->uz, r->ron);
624     hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
625     for (i = 0; i < 3; i++)
626     if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
627     break;
628     if (i >= 3)
629     error(CONSISTENCY, "bad ray direction in inithemi");
630     hp->uy[i] = 1.0;
631     fcross(hp->ux, hp->uy, hp->uz);
632     normalize(hp->ux);
633     fcross(hp->uy, hp->uz, hp->ux);
634 greg 1.1 }
635    
636    
637 greg 2.9 int
638 greg 2.14 divsample( /* sample a division */
639 greg 2.23 AMBSAMP *dp,
640 greg 2.14 AMBHEMI *h,
641     RAY *r
642     )
643 greg 1.1 {
644     RAY ar;
645 greg 1.11 int hlist[3];
646     double spt[2];
647 greg 1.1 double xd, yd, zd;
648     double b2;
649     double phi;
650 greg 2.23 int i;
651 greg 2.15 /* ambient coefficient for weight */
652 greg 2.16 if (ambacc > FTINY)
653     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
654     else
655     copycolor(ar.rcoef, h->acoef);
656 greg 2.14 if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
657 greg 1.4 return(-1);
658 greg 2.17 if (ambacc > FTINY) {
659     multcolor(ar.rcoef, h->acoef);
660     scalecolor(ar.rcoef, 1./AVGREFL);
661     }
662 greg 1.1 hlist[0] = r->rno;
663     hlist[1] = dp->t;
664     hlist[2] = dp->p;
665 greg 1.13 multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n));
666 greg 1.11 zd = sqrt((dp->t + spt[0])/h->nt);
667     phi = 2.0*PI * (dp->p + spt[1])/h->np;
668 gwlarson 2.8 xd = tcos(phi) * zd;
669     yd = tsin(phi) * zd;
670 greg 1.1 zd = sqrt(1.0 - zd*zd);
671 greg 1.2 for (i = 0; i < 3; i++)
672     ar.rdir[i] = xd*h->ux[i] +
673     yd*h->uy[i] +
674     zd*h->uz[i];
675 greg 2.22 checknorm(ar.rdir);
676 greg 1.2 dimlist[ndims++] = dp->t*h->np + dp->p + 90171;
677 greg 1.1 rayvalue(&ar);
678     ndims--;
679 greg 2.16 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
680 greg 1.1 addcolor(dp->v, ar.rcol);
681 greg 2.9 /* use rt to improve gradient calc */
682     if (ar.rt > FTINY && ar.rt < FHUGE)
683     dp->r += 1.0/ar.rt;
684 greg 1.1 /* (re)initialize error */
685     if (dp->n++) {
686     b2 = bright(dp->v)/dp->n - bright(ar.rcol);
687     b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1));
688     dp->k = b2/(dp->n*dp->n);
689     } else
690     dp->k = 0.0;
691 greg 1.4 return(0);
692 greg 1.1 }
693    
694    
695 greg 2.14 static int
696     ambcmp( /* decreasing order */
697     const void *p1,
698     const void *p2
699     )
700     {
701     const AMBSAMP *d1 = (const AMBSAMP *)p1;
702     const AMBSAMP *d2 = (const AMBSAMP *)p2;
703    
704     if (d1->k < d2->k)
705     return(1);
706     if (d1->k > d2->k)
707     return(-1);
708     return(0);
709     }
710    
711    
712     static int
713     ambnorm( /* standard order */
714     const void *p1,
715     const void *p2
716     )
717     {
718     const AMBSAMP *d1 = (const AMBSAMP *)p1;
719     const AMBSAMP *d2 = (const AMBSAMP *)p2;
720 greg 2.23 int c;
721 greg 2.14
722     if ( (c = d1->t - d2->t) )
723     return(c);
724     return(d1->p - d2->p);
725     }
726    
727    
728 greg 1.1 double
729 greg 2.14 doambient( /* compute ambient component */
730 greg 2.23 COLOR rcol,
731 greg 2.14 RAY *r,
732     double wt,
733     FVECT pg,
734     FVECT dg
735     )
736 greg 1.1 {
737 greg 2.24 double b, d=0;
738 greg 1.1 AMBHEMI hemi;
739     AMBSAMP *div;
740     AMBSAMP dnew;
741 greg 2.23 double acol[3];
742     AMBSAMP *dp;
743 greg 1.1 double arad;
744 greg 2.19 int divcnt;
745 greg 2.23 int i, j;
746 greg 1.1 /* initialize hemisphere */
747 greg 2.23 inithemi(&hemi, rcol, r, wt);
748 greg 2.19 divcnt = hemi.nt * hemi.np;
749 greg 2.17 /* initialize */
750     if (pg != NULL)
751     pg[0] = pg[1] = pg[2] = 0.0;
752     if (dg != NULL)
753     dg[0] = dg[1] = dg[2] = 0.0;
754 greg 2.23 setcolor(rcol, 0.0, 0.0, 0.0);
755 greg 2.19 if (divcnt == 0)
756 greg 1.1 return(0.0);
757 greg 2.14 /* allocate super-samples */
758 greg 2.15 if (hemi.ns > 0 || pg != NULL || dg != NULL) {
759 greg 2.19 div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP));
760 greg 1.1 if (div == NULL)
761     error(SYSTEM, "out of memory in doambient");
762     } else
763     div = NULL;
764     /* sample the divisions */
765     arad = 0.0;
766 greg 2.23 acol[0] = acol[1] = acol[2] = 0.0;
767 greg 1.1 if ((dp = div) == NULL)
768     dp = &dnew;
769 greg 2.19 divcnt = 0;
770 greg 1.1 for (i = 0; i < hemi.nt; i++)
771     for (j = 0; j < hemi.np; j++) {
772     dp->t = i; dp->p = j;
773     setcolor(dp->v, 0.0, 0.0, 0.0);
774 greg 1.2 dp->r = 0.0;
775 greg 1.1 dp->n = 0;
776 greg 2.16 if (divsample(dp, &hemi, r) < 0) {
777 greg 2.19 if (div != NULL)
778     dp++;
779 greg 2.16 continue;
780     }
781 greg 2.6 arad += dp->r;
782 greg 2.19 divcnt++;
783 greg 1.1 if (div != NULL)
784     dp++;
785 greg 2.6 else
786 greg 1.1 addcolor(acol, dp->v);
787     }
788 greg 2.21 if (!divcnt) {
789     if (div != NULL)
790     free((void *)div);
791 greg 2.19 return(0.0); /* no samples taken */
792 greg 2.21 }
793 greg 2.19 if (divcnt < hemi.nt*hemi.np) {
794     pg = dg = NULL; /* incomplete sampling */
795     hemi.ns = 0;
796     } else if (arad > FTINY && divcnt/arad < minarad) {
797 greg 2.15 hemi.ns = 0; /* close enough */
798 greg 2.19 } else if (hemi.ns > 0) { /* else perform super-sampling? */
799 greg 1.4 comperrs(div, &hemi); /* compute errors */
800 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */
801 greg 1.1 /* super-sample */
802 greg 2.15 for (i = hemi.ns; i > 0; i--) {
803 schorsch 2.11 dnew = *div;
804 greg 2.16 if (divsample(&dnew, &hemi, r) < 0) {
805     dp++;
806     continue;
807     }
808     dp = div; /* reinsert */
809 greg 2.19 j = divcnt < i ? divcnt : i;
810 greg 1.1 while (--j > 0 && dnew.k < dp[1].k) {
811 schorsch 2.11 *dp = *(dp+1);
812 greg 1.1 dp++;
813     }
814 schorsch 2.11 *dp = dnew;
815 greg 1.1 }
816 greg 1.2 if (pg != NULL || dg != NULL) /* restore order */
817 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambnorm);
818 greg 1.1 }
819     /* compute returned values */
820 greg 1.3 if (div != NULL) {
821 greg 2.19 arad = 0.0; /* note: divcnt may be < nt*np */
822     for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) {
823 greg 1.3 arad += dp->r;
824     if (dp->n > 1) {
825     b = 1.0/dp->n;
826     scalecolor(dp->v, b);
827     dp->r *= b;
828     dp->n = 1;
829     }
830     addcolor(acol, dp->v);
831     }
832 greg 1.5 b = bright(acol);
833 greg 1.6 if (b > FTINY) {
834 greg 2.17 b = 1.0/b; /* compute & normalize gradient(s) */
835 greg 1.6 if (pg != NULL) {
836     posgradient(pg, div, &hemi);
837     for (i = 0; i < 3; i++)
838     pg[i] *= b;
839     }
840     if (dg != NULL) {
841     dirgradient(dg, div, &hemi);
842     for (i = 0; i < 3; i++)
843     dg[i] *= b;
844     }
845 greg 1.5 }
846 greg 2.9 free((void *)div);
847 greg 1.3 }
848 greg 2.23 copycolor(rcol, acol);
849 greg 1.1 if (arad <= FTINY)
850 greg 1.16 arad = maxarad;
851 greg 2.3 else
852 greg 2.19 arad = (divcnt+hemi.ns)/arad;
853 greg 1.15 if (pg != NULL) { /* reduce radius if gradient large */
854     d = DOT(pg,pg);
855     if (d*arad*arad > 1.0)
856     arad = 1.0/sqrt(d);
857     }
858 greg 1.16 if (arad < minarad) {
859 greg 1.1 arad = minarad;
860 greg 1.16 if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */
861     d = 1.0/arad/sqrt(d);
862     for (i = 0; i < 3; i++)
863     pg[i] *= d;
864     }
865     }
866 greg 2.3 if ((arad /= sqrt(wt)) > maxarad)
867     arad = maxarad;
868     return(arad);
869 greg 1.1 }
870    
871    
872 greg 2.9 void
873 greg 2.14 comperrs( /* compute initial error estimates */
874     AMBSAMP *da, /* assumes standard ordering */
875 greg 2.23 AMBHEMI *hp
876 greg 2.14 )
877 greg 1.1 {
878     double b, b2;
879     int i, j;
880 greg 2.23 AMBSAMP *dp;
881 greg 1.1 /* sum differences from neighbors */
882     dp = da;
883     for (i = 0; i < hp->nt; i++)
884     for (j = 0; j < hp->np; j++) {
885 greg 1.6 #ifdef DEBUG
886     if (dp->t != i || dp->p != j)
887     error(CONSISTENCY,
888     "division order in comperrs");
889     #endif
890 greg 1.1 b = bright(dp[0].v);
891     if (i > 0) { /* from above */
892     b2 = bright(dp[-hp->np].v) - b;
893     b2 *= b2 * 0.25;
894     dp[0].k += b2;
895     dp[-hp->np].k += b2;
896     }
897     if (j > 0) { /* from behind */
898     b2 = bright(dp[-1].v) - b;
899     b2 *= b2 * 0.25;
900     dp[0].k += b2;
901     dp[-1].k += b2;
902 greg 1.4 } else { /* around */
903     b2 = bright(dp[hp->np-1].v) - b;
904 greg 1.1 b2 *= b2 * 0.25;
905     dp[0].k += b2;
906 greg 1.4 dp[hp->np-1].k += b2;
907 greg 1.1 }
908     dp++;
909     }
910     /* divide by number of neighbors */
911     dp = da;
912     for (j = 0; j < hp->np; j++) /* top row */
913     (dp++)->k *= 1.0/3.0;
914     if (hp->nt < 2)
915     return;
916     for (i = 1; i < hp->nt-1; i++) /* central region */
917     for (j = 0; j < hp->np; j++)
918     (dp++)->k *= 0.25;
919     for (j = 0; j < hp->np; j++) /* bottom row */
920     (dp++)->k *= 1.0/3.0;
921     }
922    
923    
924 greg 2.9 void
925 greg 2.14 posgradient( /* compute position gradient */
926     FVECT gv,
927     AMBSAMP *da, /* assumes standard ordering */
928 greg 2.23 AMBHEMI *hp
929 greg 2.14 )
930 greg 1.1 {
931 greg 2.23 int i, j;
932 greg 2.2 double nextsine, lastsine, b, d;
933 greg 1.2 double mag0, mag1;
934     double phi, cosp, sinp, xd, yd;
935 greg 2.23 AMBSAMP *dp;
936 greg 1.2
937     xd = yd = 0.0;
938     for (j = 0; j < hp->np; j++) {
939     dp = da + j;
940     mag0 = mag1 = 0.0;
941 greg 2.2 lastsine = 0.0;
942 greg 1.2 for (i = 0; i < hp->nt; i++) {
943     #ifdef DEBUG
944     if (dp->t != i || dp->p != j)
945     error(CONSISTENCY,
946     "division order in posgradient");
947     #endif
948     b = bright(dp->v);
949     if (i > 0) {
950     d = dp[-hp->np].r;
951     if (dp[0].r > d) d = dp[0].r;
952 greg 2.2 /* sin(t)*cos(t)^2 */
953     d *= lastsine * (1.0 - (double)i/hp->nt);
954 greg 1.2 mag0 += d*(b - bright(dp[-hp->np].v));
955     }
956 greg 2.2 nextsine = sqrt((double)(i+1)/hp->nt);
957 greg 1.2 if (j > 0) {
958     d = dp[-1].r;
959     if (dp[0].r > d) d = dp[0].r;
960 greg 2.2 mag1 += d * (nextsine - lastsine) *
961     (b - bright(dp[-1].v));
962 greg 1.2 } else {
963     d = dp[hp->np-1].r;
964     if (dp[0].r > d) d = dp[0].r;
965 greg 2.2 mag1 += d * (nextsine - lastsine) *
966     (b - bright(dp[hp->np-1].v));
967 greg 1.2 }
968     dp += hp->np;
969 greg 2.2 lastsine = nextsine;
970 greg 1.2 }
971 greg 2.2 mag0 *= 2.0*PI / hp->np;
972 greg 1.2 phi = 2.0*PI * (double)j/hp->np;
973 gwlarson 2.8 cosp = tcos(phi); sinp = tsin(phi);
974 greg 1.2 xd += mag0*cosp - mag1*sinp;
975     yd += mag0*sinp + mag1*cosp;
976     }
977     for (i = 0; i < 3; i++)
978 greg 2.16 gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI;
979 greg 1.1 }
980    
981    
982 greg 2.9 void
983 greg 2.14 dirgradient( /* compute direction gradient */
984     FVECT gv,
985     AMBSAMP *da, /* assumes standard ordering */
986 greg 2.23 AMBHEMI *hp
987 greg 2.14 )
988 greg 1.1 {
989 greg 2.23 int i, j;
990 greg 1.2 double mag;
991     double phi, xd, yd;
992 greg 2.23 AMBSAMP *dp;
993 greg 1.2
994     xd = yd = 0.0;
995     for (j = 0; j < hp->np; j++) {
996     dp = da + j;
997     mag = 0.0;
998     for (i = 0; i < hp->nt; i++) {
999     #ifdef DEBUG
1000     if (dp->t != i || dp->p != j)
1001     error(CONSISTENCY,
1002     "division order in dirgradient");
1003     #endif
1004 greg 2.2 /* tan(t) */
1005     mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0);
1006 greg 1.2 dp += hp->np;
1007     }
1008     phi = 2.0*PI * (j+.5)/hp->np + PI/2.0;
1009 gwlarson 2.8 xd += mag * tcos(phi);
1010     yd += mag * tsin(phi);
1011 greg 1.2 }
1012     for (i = 0; i < 3; i++)
1013 greg 2.16 gv[i] = xd*hp->ux[i] + yd*hp->uy[i];
1014 greg 1.1 }
1015 greg 2.25
1016     #endif /* ! NEWAMB */