ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/radiance/ray/src/rt/ambcomp.c
Revision: 2.36
Committed: Sat Apr 26 04:37:48 2014 UTC (10 years ago) by greg
Content type: text/plain
Branch: MAIN
Changes since 2.35: +6 -9 lines
Log Message: 
Randomize hemisphere orientation for better sampling

File Contents

# User Rev Content
1 greg 1.1 #ifndef lint
2 greg 2.36 static const char RCSid[] = "$Id: ambcomp.c,v 2.35 2014/04/25 18:39:22 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 2.36 hp->uy[0] = hp->uy[1] = hp->uy[2] = 0;
74     for (i = 3; i--; )
75     if ((0.6 < r->ron[i]) & (r->ron[i] < 0.6))
76     hp->uy[i] = 0.1+frandom();
77     if (DOT(hp->uy,hp->uy) <= FTINY)
78 greg 2.26 error(CONSISTENCY, "bad ray direction in inithemi()");
79 greg 2.27 VCROSS(hp->ux, hp->uy, r->ron);
80 greg 2.26 normalize(hp->ux);
81 greg 2.27 VCROSS(hp->uy, r->ron, hp->ux);
82 greg 2.26 /* we're ready to sample */
83     return(hp);
84     }
85    
86    
87 greg 2.28 static struct s_ambsamp *
88 greg 2.26 ambsample( /* sample an ambient direction */
89     AMBHEMI *hp,
90     int i,
91 greg 2.27 int j
92 greg 2.26 )
93     {
94     struct s_ambsamp *ap = &ambsamp(hp,i,j);
95     RAY ar;
96 greg 2.27 double spt[2], zd;
97 greg 2.26 int ii;
98     /* ambient coefficient for weight */
99     if (ambacc > FTINY)
100     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
101     else
102     copycolor(ar.rcoef, hp->acoef);
103 greg 2.31 if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0)
104     goto badsample;
105 greg 2.26 if (ambacc > FTINY) {
106     multcolor(ar.rcoef, hp->acoef);
107     scalecolor(ar.rcoef, 1./AVGREFL);
108     }
109     /* generate hemispherical sample */
110 greg 2.27 SDsquare2disk(spt, (i+.1+.8*frandom())/hp->ns,
111 greg 2.28 (j+.1+.8*frandom())/hp->ns );
112 greg 2.26 zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]);
113     for (ii = 3; ii--; )
114     ar.rdir[ii] = spt[0]*hp->ux[ii] +
115     spt[1]*hp->uy[ii] +
116     zd*hp->rp->ron[ii];
117     checknorm(ar.rdir);
118     dimlist[ndims++] = i*hp->ns + j + 90171;
119     rayvalue(&ar); /* evaluate ray */
120     ndims--;
121 greg 2.34 /* limit vertex distance */
122     if (ar.rt > 10.0*thescene.cusize)
123     ar.rt = 10.0*thescene.cusize;
124 greg 2.31 else if (ar.rt <= FTINY) /* should never happen! */
125     goto badsample;
126     VSUM(ap->p, ar.rorg, ar.rdir, ar.rt);
127 greg 2.26 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
128     copycolor(ap->v, ar.rcol);
129 greg 2.28 return(ap);
130 greg 2.31 badsample:
131     setcolor(ap->v, 0., 0., 0.);
132     VCOPY(ap->p, hp->rp->rop);
133     return(NULL);
134 greg 2.26 }
135    
136    
137 greg 2.27 /* Compute vectors and coefficients for Hessian/gradient calcs */
138     static void
139 greg 2.31 comp_fftri(FFTRI *ftp, FVECT ap0, FVECT ap1, FVECT rop)
140 greg 2.27 {
141 greg 2.35 double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2;
142 greg 2.30 int i;
143 greg 2.27
144     VSUB(ftp->r_i, ap0, rop);
145     VSUB(ftp->r_i1, ap1, rop);
146     VSUB(ftp->e_i, ap1, ap0);
147 greg 2.35 VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1);
148     rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp);
149 greg 2.27 dot_e = DOT(ftp->e_i,ftp->e_i);
150     dot_er = DOT(ftp->e_i, ftp->r_i);
151 greg 2.32 rdot_r = 1.0/DOT(ftp->r_i,ftp->r_i);
152     rdot_r1 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
153     ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) * sqrt(rdot_r*rdot_r1) ) *
154 greg 2.35 sqrt( rdot_cp );
155 greg 2.32 ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r +
156 greg 2.35 dot_e*ftp->I1 )*0.5*rdot_cp;
157 greg 2.32 J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e;
158 greg 2.30 for (i = 3; i--; )
159     ftp->rI2_eJ2[i] = ftp->I2*ftp->r_i[i] + J2*ftp->e_i[i];
160 greg 2.27 }
161    
162    
163 greg 2.28 /* Compose 3x3 matrix from two vectors */
164 greg 2.27 static void
165     compose_matrix(FVECT mat[3], FVECT va, FVECT vb)
166     {
167     mat[0][0] = 2.0*va[0]*vb[0];
168     mat[1][1] = 2.0*va[1]*vb[1];
169     mat[2][2] = 2.0*va[2]*vb[2];
170     mat[0][1] = mat[1][0] = va[0]*vb[1] + va[1]*vb[0];
171     mat[0][2] = mat[2][0] = va[0]*vb[2] + va[2]*vb[0];
172     mat[1][2] = mat[2][1] = va[1]*vb[2] + va[2]*vb[1];
173     }
174    
175    
176     /* Compute partial 3x3 Hessian matrix for edge */
177     static void
178     comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm)
179     {
180 greg 2.35 FVECT ncp;
181 greg 2.27 FVECT m1[3], m2[3], m3[3], m4[3];
182     double d1, d2, d3, d4;
183     double I3, J3, K3;
184     int i, j;
185     /* compute intermediate coefficients */
186     d1 = 1.0/DOT(ftp->r_i,ftp->r_i);
187     d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
188     d3 = 1.0/DOT(ftp->e_i,ftp->e_i);
189     d4 = DOT(ftp->e_i, ftp->r_i);
190 greg 2.35 I3 = ( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + 3.0/d3*ftp->I2 )
191     / ( 4.0*DOT(ftp->rcp,ftp->rcp) );
192 greg 2.27 J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3;
193     K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3);
194     /* intermediate matrices */
195 greg 2.35 VCROSS(ncp, nrm, ftp->e_i);
196     compose_matrix(m1, ncp, ftp->rI2_eJ2);
197 greg 2.27 compose_matrix(m2, ftp->r_i, ftp->r_i);
198     compose_matrix(m3, ftp->e_i, ftp->e_i);
199     compose_matrix(m4, ftp->r_i, ftp->e_i);
200 greg 2.35 d1 = DOT(nrm, ftp->rcp);
201 greg 2.27 d2 = -d1*ftp->I2;
202     d1 *= 2.0;
203     for (i = 3; i--; ) /* final matrix sum */
204     for (j = 3; j--; ) {
205     hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] +
206     2.0*J3*m4[i][j] );
207     hess[i][j] += d2*(i==j);
208 greg 2.32 hess[i][j] *= 1.0/PI;
209 greg 2.27 }
210     }
211    
212    
213     /* Reverse hessian calculation result for edge in other direction */
214     static void
215     rev_hessian(FVECT hess[3])
216     {
217     int i;
218    
219     for (i = 3; i--; ) {
220     hess[i][0] = -hess[i][0];
221     hess[i][1] = -hess[i][1];
222     hess[i][2] = -hess[i][2];
223     }
224     }
225    
226    
227     /* Add to radiometric Hessian from the given triangle */
228     static void
229     add2hessian(FVECT hess[3], FVECT ehess1[3],
230     FVECT ehess2[3], FVECT ehess3[3], COLORV v)
231     {
232     int i, j;
233    
234     for (i = 3; i--; )
235     for (j = 3; j--; )
236     hess[i][j] += v*( ehess1[i][j] + ehess2[i][j] + ehess3[i][j] );
237     }
238    
239    
240     /* Compute partial displacement form factor gradient for edge */
241     static void
242     comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm)
243     {
244 greg 2.35 FVECT ncp;
245 greg 2.27 double f1;
246     int i;
247    
248 greg 2.35 f1 = 2.0*DOT(nrm, ftp->rcp);
249     VCROSS(ncp, nrm, ftp->e_i);
250 greg 2.27 for (i = 3; i--; )
251 greg 2.35 grad[i] = (-0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] );
252 greg 2.27 }
253    
254    
255     /* Reverse gradient calculation result for edge in other direction */
256     static void
257     rev_gradient(FVECT grad)
258     {
259     grad[0] = -grad[0];
260     grad[1] = -grad[1];
261     grad[2] = -grad[2];
262     }
263    
264    
265     /* Add to displacement gradient from the given triangle */
266     static void
267     add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v)
268     {
269     int i;
270    
271     for (i = 3; i--; )
272     grad[i] += v*( egrad1[i] + egrad2[i] + egrad3[i] );
273     }
274    
275    
276     /* Return brightness of furthest ambient sample */
277     static COLORV
278     back_ambval(struct s_ambsamp *ap1, struct s_ambsamp *ap2,
279     struct s_ambsamp *ap3, FVECT orig)
280     {
281     COLORV vback;
282     FVECT vec;
283     double d2, d2best;
284    
285     VSUB(vec, ap1->p, orig);
286     d2best = DOT(vec,vec);
287 greg 2.29 vback = colval(ap1->v,CIEY);
288 greg 2.27 VSUB(vec, ap2->p, orig);
289     d2 = DOT(vec,vec);
290     if (d2 > d2best) {
291     d2best = d2;
292 greg 2.29 vback = colval(ap2->v,CIEY);
293 greg 2.27 }
294     VSUB(vec, ap3->p, orig);
295     d2 = DOT(vec,vec);
296     if (d2 > d2best)
297 greg 2.29 return(colval(ap3->v,CIEY));
298 greg 2.27 return(vback);
299     }
300    
301    
302     /* Compute anisotropic radii and eigenvector directions */
303     static int
304     eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3])
305     {
306     double hess2[2][2];
307     FVECT a, b;
308     double evalue[2], slope1, xmag1;
309     int i;
310     /* project Hessian to sample plane */
311     for (i = 3; i--; ) {
312     a[i] = DOT(hessian[i], uv[0]);
313     b[i] = DOT(hessian[i], uv[1]);
314     }
315     hess2[0][0] = DOT(uv[0], a);
316     hess2[0][1] = DOT(uv[0], b);
317     hess2[1][0] = DOT(uv[1], a);
318     hess2[1][1] = DOT(uv[1], b);
319     /* compute eigenvalues */
320 greg 2.28 if ( quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1],
321 greg 2.27 hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]) != 2 ||
322 greg 2.35 ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) |
323     ((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) )
324 greg 2.27 error(INTERNAL, "bad eigenvalue calculation");
325    
326     if (evalue[0] > evalue[1]) {
327 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[0]));
328     ra[1] = sqrt(sqrt(4.0/evalue[1]));
329 greg 2.27 slope1 = evalue[1];
330     } else {
331 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[1]));
332     ra[1] = sqrt(sqrt(4.0/evalue[0]));
333 greg 2.27 slope1 = evalue[0];
334     }
335     /* compute unit eigenvectors */
336     if (fabs(hess2[0][1]) <= FTINY)
337     return; /* uv OK as is */
338     slope1 = (slope1 - hess2[0][0]) / hess2[0][1];
339     xmag1 = sqrt(1.0/(1.0 + slope1*slope1));
340     for (i = 3; i--; ) {
341     b[i] = xmag1*uv[0][i] + slope1*xmag1*uv[1][i];
342     a[i] = slope1*xmag1*uv[0][i] - xmag1*uv[1][i];
343     }
344     VCOPY(uv[0], a);
345     VCOPY(uv[1], b);
346     }
347    
348    
349 greg 2.26 static void
350     ambHessian( /* anisotropic radii & pos. gradient */
351     AMBHEMI *hp,
352     FVECT uv[2], /* returned */
353 greg 2.28 float ra[2], /* returned (optional) */
354     float pg[2] /* returned (optional) */
355 greg 2.26 )
356     {
357 greg 2.27 static char memerrmsg[] = "out of memory in ambHessian()";
358     FVECT (*hessrow)[3] = NULL;
359     FVECT *gradrow = NULL;
360     FVECT hessian[3];
361     FVECT gradient;
362     FFTRI fftr;
363     int i, j;
364     /* be sure to assign unit vectors */
365     VCOPY(uv[0], hp->ux);
366     VCOPY(uv[1], hp->uy);
367     /* clock-wise vertex traversal from sample POV */
368     if (ra != NULL) { /* initialize Hessian row buffer */
369 greg 2.28 hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1));
370 greg 2.27 if (hessrow == NULL)
371     error(SYSTEM, memerrmsg);
372     memset(hessian, 0, sizeof(hessian));
373     } else if (pg == NULL) /* bogus call? */
374     return;
375     if (pg != NULL) { /* initialize form factor row buffer */
376 greg 2.28 gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1));
377 greg 2.27 if (gradrow == NULL)
378     error(SYSTEM, memerrmsg);
379     memset(gradient, 0, sizeof(gradient));
380     }
381     /* compute first row of edges */
382     for (j = 0; j < hp->ns-1; j++) {
383     comp_fftri(&fftr, ambsamp(hp,0,j).p,
384     ambsamp(hp,0,j+1).p, hp->rp->rop);
385     if (hessrow != NULL)
386     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
387     if (gradrow != NULL)
388     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
389     }
390     /* sum each row of triangles */
391     for (i = 0; i < hp->ns-1; i++) {
392     FVECT hesscol[3]; /* compute first vertical edge */
393     FVECT gradcol;
394     comp_fftri(&fftr, ambsamp(hp,i,0).p,
395     ambsamp(hp,i+1,0).p, hp->rp->rop);
396     if (hessrow != NULL)
397     comp_hessian(hesscol, &fftr, hp->rp->ron);
398     if (gradrow != NULL)
399     comp_gradient(gradcol, &fftr, hp->rp->ron);
400     for (j = 0; j < hp->ns-1; j++) {
401     FVECT hessdia[3]; /* compute triangle contributions */
402     FVECT graddia;
403     COLORV backg;
404     backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1),
405     &ambsamp(hp,i+1,j), hp->rp->rop);
406     /* diagonal (inner) edge */
407     comp_fftri(&fftr, ambsamp(hp,i,j+1).p,
408     ambsamp(hp,i+1,j).p, hp->rp->rop);
409     if (hessrow != NULL) {
410     comp_hessian(hessdia, &fftr, hp->rp->ron);
411     rev_hessian(hesscol);
412     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
413     }
414     if (gradient != NULL) {
415     comp_gradient(graddia, &fftr, hp->rp->ron);
416     rev_gradient(gradcol);
417     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
418     }
419     /* initialize edge in next row */
420     comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p,
421     ambsamp(hp,i+1,j).p, hp->rp->rop);
422     if (hessrow != NULL)
423     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
424     if (gradrow != NULL)
425     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
426     /* new column edge & paired triangle */
427     backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1),
428     &ambsamp(hp,i+1,j), hp->rp->rop);
429     comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p,
430     hp->rp->rop);
431     if (hessrow != NULL) {
432     comp_hessian(hesscol, &fftr, hp->rp->ron);
433     rev_hessian(hessdia);
434     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
435     if (i < hp->ns-2)
436     rev_hessian(hessrow[j]);
437     }
438     if (gradrow != NULL) {
439     comp_gradient(gradcol, &fftr, hp->rp->ron);
440     rev_gradient(graddia);
441     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
442     if (i < hp->ns-2)
443     rev_gradient(gradrow[j]);
444     }
445     }
446     }
447     /* release row buffers */
448     if (hessrow != NULL) free(hessrow);
449     if (gradrow != NULL) free(gradrow);
450    
451     if (ra != NULL) /* extract eigenvectors & radii */
452     eigenvectors(uv, ra, hessian);
453 greg 2.32 if (pg != NULL) { /* tangential position gradient */
454     pg[0] = DOT(gradient, uv[0]);
455     pg[1] = DOT(gradient, uv[1]);
456 greg 2.27 }
457     }
458    
459    
460     /* Compute direction gradient from a hemispherical sampling */
461     static void
462     ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2])
463     {
464     struct s_ambsamp *ap;
465 greg 2.29 double dgsum[2];
466 greg 2.27 int n;
467 greg 2.28 FVECT vd;
468     double gfact;
469 greg 2.27
470 greg 2.29 dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */
471 greg 2.27 for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) {
472     /* use vector for azimuth + 90deg */
473     VSUB(vd, ap->p, hp->rp->rop);
474 greg 2.29 /* brightness over cosine factor */
475     gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd);
476     /* -sine = -proj_radius/vd_length */
477     dgsum[0] += DOT(uv[1], vd) * gfact;
478     dgsum[1] -= DOT(uv[0], vd) * gfact;
479 greg 2.26 }
480 greg 2.29 dg[0] = dgsum[0] / (hp->ns*hp->ns);
481     dg[1] = dgsum[1] / (hp->ns*hp->ns);
482 greg 2.26 }
483    
484 greg 2.27
485 greg 2.26 int
486     doambient( /* compute ambient component */
487     COLOR rcol, /* input/output color */
488     RAY *r,
489     double wt,
490 greg 2.27 FVECT uv[2], /* returned (optional) */
491     float ra[2], /* returned (optional) */
492     float pg[2], /* returned (optional) */
493     float dg[2] /* returned (optional) */
494 greg 2.26 )
495     {
496 greg 2.28 AMBHEMI *hp = inithemi(rcol, r, wt);
497 greg 2.26 int cnt = 0;
498     FVECT my_uv[2];
499     double d, acol[3];
500     struct s_ambsamp *ap;
501     int i, j;
502 greg 2.28 /* check/initialize */
503     if (hp == NULL)
504 greg 2.26 return(0);
505     if (uv != NULL)
506     memset(uv, 0, sizeof(FVECT)*2);
507     if (ra != NULL)
508     ra[0] = ra[1] = 0.0;
509     if (pg != NULL)
510     pg[0] = pg[1] = 0.0;
511     if (dg != NULL)
512     dg[0] = dg[1] = 0.0;
513     /* sample the hemisphere */
514     acol[0] = acol[1] = acol[2] = 0.0;
515 greg 2.27 for (i = hp->ns; i--; )
516     for (j = hp->ns; j--; )
517 greg 2.28 if ((ap = ambsample(hp, i, j)) != NULL) {
518 greg 2.26 addcolor(acol, ap->v);
519     ++cnt;
520     }
521     if (!cnt) {
522     setcolor(rcol, 0.0, 0.0, 0.0);
523     free(hp);
524     return(0); /* no valid samples */
525     }
526 greg 2.29 copycolor(rcol, acol); /* final indirect irradiance/PI */
527 greg 2.26 if (cnt < hp->ns*hp->ns || /* incomplete sampling? */
528     (ra == NULL) & (pg == NULL) & (dg == NULL)) {
529     free(hp);
530     return(-1); /* no radius or gradient calc. */
531     }
532 greg 2.32 if (bright(acol) > FTINY) /* normalize Y values */
533     d = cnt/bright(acol);
534 greg 2.29 else
535     d = 0.0;
536     ap = hp->sa; /* relative Y channel from here on... */
537 greg 2.26 for (i = hp->ns*hp->ns; i--; ap++)
538 greg 2.32 colval(ap->v,CIEY) = bright(ap->v)*d + 0.01;
539 greg 2.26
540     if (uv == NULL) /* make sure we have axis pointers */
541     uv = my_uv;
542     /* compute radii & pos. gradient */
543     ambHessian(hp, uv, ra, pg);
544 greg 2.29
545 greg 2.26 if (dg != NULL) /* compute direction gradient */
546     ambdirgrad(hp, uv, dg);
547 greg 2.29
548 greg 2.28 if (ra != NULL) { /* scale/clamp radii */
549 greg 2.35 if (pg != NULL) {
550     if (ra[0]*(d = fabs(pg[0])) > 1.0)
551     ra[0] = 1.0/d;
552     if (ra[1]*(d = fabs(pg[1])) > 1.0)
553     ra[1] = 1.0/d;
554     if (ra[0] > ra[1])
555     ra[0] = ra[1];
556     }
557 greg 2.29 if (ra[0] < minarad) {
558     ra[0] = minarad;
559     if (ra[1] < minarad)
560     ra[1] = minarad;
561     }
562     ra[0] *= d = 1.0/sqrt(sqrt(wt));
563 greg 2.26 if ((ra[1] *= d) > 2.0*ra[0])
564     ra[1] = 2.0*ra[0];
565 greg 2.28 if (ra[1] > maxarad) {
566     ra[1] = maxarad;
567     if (ra[0] > maxarad)
568     ra[0] = maxarad;
569     }
570 greg 2.35 if (pg != NULL) { /* cap gradient if necessary */
571     d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1];
572     if (d > 1.0) {
573     d = 1.0/sqrt(d);
574     pg[0] *= d;
575     pg[1] *= d;
576     }
577     }
578 greg 2.26 }
579     free(hp); /* clean up and return */
580     return(1);
581     }
582    
583    
584 greg 2.25 #else /* ! NEWAMB */
585 greg 1.1
586    
587 greg 2.15 void
588 greg 2.14 inithemi( /* initialize sampling hemisphere */
589 greg 2.23 AMBHEMI *hp,
590 greg 2.16 COLOR ac,
591 greg 2.14 RAY *r,
592     double wt
593     )
594 greg 1.1 {
595 greg 2.16 double d;
596 greg 2.23 int i;
597 greg 2.14 /* set number of divisions */
598 greg 2.16 if (ambacc <= FTINY &&
599 greg 2.20 wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight)))
600 greg 2.16 wt = d; /* avoid ray termination */
601     hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
602 greg 2.14 i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */
603     if (hp->nt < i)
604     hp->nt = i;
605     hp->np = PI * hp->nt + 0.5;
606     /* set number of super-samples */
607 greg 2.15 hp->ns = ambssamp * wt + 0.5;
608 greg 2.16 /* assign coefficient */
609 greg 2.14 copycolor(hp->acoef, ac);
610 greg 2.16 d = 1.0/(hp->nt*hp->np);
611     scalecolor(hp->acoef, d);
612 greg 2.14 /* make axes */
613     VCOPY(hp->uz, r->ron);
614     hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
615     for (i = 0; i < 3; i++)
616     if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
617     break;
618     if (i >= 3)
619     error(CONSISTENCY, "bad ray direction in inithemi");
620     hp->uy[i] = 1.0;
621     fcross(hp->ux, hp->uy, hp->uz);
622     normalize(hp->ux);
623     fcross(hp->uy, hp->uz, hp->ux);
624 greg 1.1 }
625    
626    
627 greg 2.9 int
628 greg 2.14 divsample( /* sample a division */
629 greg 2.23 AMBSAMP *dp,
630 greg 2.14 AMBHEMI *h,
631     RAY *r
632     )
633 greg 1.1 {
634     RAY ar;
635 greg 1.11 int hlist[3];
636     double spt[2];
637 greg 1.1 double xd, yd, zd;
638     double b2;
639     double phi;
640 greg 2.23 int i;
641 greg 2.15 /* ambient coefficient for weight */
642 greg 2.16 if (ambacc > FTINY)
643     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
644     else
645     copycolor(ar.rcoef, h->acoef);
646 greg 2.14 if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
647 greg 1.4 return(-1);
648 greg 2.17 if (ambacc > FTINY) {
649     multcolor(ar.rcoef, h->acoef);
650     scalecolor(ar.rcoef, 1./AVGREFL);
651     }
652 greg 1.1 hlist[0] = r->rno;
653     hlist[1] = dp->t;
654     hlist[2] = dp->p;
655 greg 1.13 multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n));
656 greg 1.11 zd = sqrt((dp->t + spt[0])/h->nt);
657     phi = 2.0*PI * (dp->p + spt[1])/h->np;
658 gwlarson 2.8 xd = tcos(phi) * zd;
659     yd = tsin(phi) * zd;
660 greg 1.1 zd = sqrt(1.0 - zd*zd);
661 greg 1.2 for (i = 0; i < 3; i++)
662     ar.rdir[i] = xd*h->ux[i] +
663     yd*h->uy[i] +
664     zd*h->uz[i];
665 greg 2.22 checknorm(ar.rdir);
666 greg 1.2 dimlist[ndims++] = dp->t*h->np + dp->p + 90171;
667 greg 1.1 rayvalue(&ar);
668     ndims--;
669 greg 2.16 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
670 greg 1.1 addcolor(dp->v, ar.rcol);
671 greg 2.9 /* use rt to improve gradient calc */
672     if (ar.rt > FTINY && ar.rt < FHUGE)
673     dp->r += 1.0/ar.rt;
674 greg 1.1 /* (re)initialize error */
675     if (dp->n++) {
676     b2 = bright(dp->v)/dp->n - bright(ar.rcol);
677     b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1));
678     dp->k = b2/(dp->n*dp->n);
679     } else
680     dp->k = 0.0;
681 greg 1.4 return(0);
682 greg 1.1 }
683    
684    
685 greg 2.14 static int
686     ambcmp( /* decreasing order */
687     const void *p1,
688     const void *p2
689     )
690     {
691     const AMBSAMP *d1 = (const AMBSAMP *)p1;
692     const AMBSAMP *d2 = (const AMBSAMP *)p2;
693    
694     if (d1->k < d2->k)
695     return(1);
696     if (d1->k > d2->k)
697     return(-1);
698     return(0);
699     }
700    
701    
702     static int
703     ambnorm( /* standard order */
704     const void *p1,
705     const void *p2
706     )
707     {
708     const AMBSAMP *d1 = (const AMBSAMP *)p1;
709     const AMBSAMP *d2 = (const AMBSAMP *)p2;
710 greg 2.23 int c;
711 greg 2.14
712     if ( (c = d1->t - d2->t) )
713     return(c);
714     return(d1->p - d2->p);
715     }
716    
717    
718 greg 1.1 double
719 greg 2.14 doambient( /* compute ambient component */
720 greg 2.23 COLOR rcol,
721 greg 2.14 RAY *r,
722     double wt,
723     FVECT pg,
724     FVECT dg
725     )
726 greg 1.1 {
727 greg 2.24 double b, d=0;
728 greg 1.1 AMBHEMI hemi;
729     AMBSAMP *div;
730     AMBSAMP dnew;
731 greg 2.23 double acol[3];
732     AMBSAMP *dp;
733 greg 1.1 double arad;
734 greg 2.19 int divcnt;
735 greg 2.23 int i, j;
736 greg 1.1 /* initialize hemisphere */
737 greg 2.23 inithemi(&hemi, rcol, r, wt);
738 greg 2.19 divcnt = hemi.nt * hemi.np;
739 greg 2.17 /* initialize */
740     if (pg != NULL)
741     pg[0] = pg[1] = pg[2] = 0.0;
742     if (dg != NULL)
743     dg[0] = dg[1] = dg[2] = 0.0;
744 greg 2.23 setcolor(rcol, 0.0, 0.0, 0.0);
745 greg 2.19 if (divcnt == 0)
746 greg 1.1 return(0.0);
747 greg 2.14 /* allocate super-samples */
748 greg 2.15 if (hemi.ns > 0 || pg != NULL || dg != NULL) {
749 greg 2.19 div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP));
750 greg 1.1 if (div == NULL)
751     error(SYSTEM, "out of memory in doambient");
752     } else
753     div = NULL;
754     /* sample the divisions */
755     arad = 0.0;
756 greg 2.23 acol[0] = acol[1] = acol[2] = 0.0;
757 greg 1.1 if ((dp = div) == NULL)
758     dp = &dnew;
759 greg 2.19 divcnt = 0;
760 greg 1.1 for (i = 0; i < hemi.nt; i++)
761     for (j = 0; j < hemi.np; j++) {
762     dp->t = i; dp->p = j;
763     setcolor(dp->v, 0.0, 0.0, 0.0);
764 greg 1.2 dp->r = 0.0;
765 greg 1.1 dp->n = 0;
766 greg 2.16 if (divsample(dp, &hemi, r) < 0) {
767 greg 2.19 if (div != NULL)
768     dp++;
769 greg 2.16 continue;
770     }
771 greg 2.6 arad += dp->r;
772 greg 2.19 divcnt++;
773 greg 1.1 if (div != NULL)
774     dp++;
775 greg 2.6 else
776 greg 1.1 addcolor(acol, dp->v);
777     }
778 greg 2.21 if (!divcnt) {
779     if (div != NULL)
780     free((void *)div);
781 greg 2.19 return(0.0); /* no samples taken */
782 greg 2.21 }
783 greg 2.19 if (divcnt < hemi.nt*hemi.np) {
784     pg = dg = NULL; /* incomplete sampling */
785     hemi.ns = 0;
786     } else if (arad > FTINY && divcnt/arad < minarad) {
787 greg 2.15 hemi.ns = 0; /* close enough */
788 greg 2.19 } else if (hemi.ns > 0) { /* else perform super-sampling? */
789 greg 1.4 comperrs(div, &hemi); /* compute errors */
790 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */
791 greg 1.1 /* super-sample */
792 greg 2.15 for (i = hemi.ns; i > 0; i--) {
793 schorsch 2.11 dnew = *div;
794 greg 2.16 if (divsample(&dnew, &hemi, r) < 0) {
795     dp++;
796     continue;
797     }
798     dp = div; /* reinsert */
799 greg 2.19 j = divcnt < i ? divcnt : i;
800 greg 1.1 while (--j > 0 && dnew.k < dp[1].k) {
801 schorsch 2.11 *dp = *(dp+1);
802 greg 1.1 dp++;
803     }
804 schorsch 2.11 *dp = dnew;
805 greg 1.1 }
806 greg 1.2 if (pg != NULL || dg != NULL) /* restore order */
807 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambnorm);
808 greg 1.1 }
809     /* compute returned values */
810 greg 1.3 if (div != NULL) {
811 greg 2.19 arad = 0.0; /* note: divcnt may be < nt*np */
812     for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) {
813 greg 1.3 arad += dp->r;
814     if (dp->n > 1) {
815     b = 1.0/dp->n;
816     scalecolor(dp->v, b);
817     dp->r *= b;
818     dp->n = 1;
819     }
820     addcolor(acol, dp->v);
821     }
822 greg 1.5 b = bright(acol);
823 greg 1.6 if (b > FTINY) {
824 greg 2.17 b = 1.0/b; /* compute & normalize gradient(s) */
825 greg 1.6 if (pg != NULL) {
826     posgradient(pg, div, &hemi);
827     for (i = 0; i < 3; i++)
828     pg[i] *= b;
829     }
830     if (dg != NULL) {
831     dirgradient(dg, div, &hemi);
832     for (i = 0; i < 3; i++)
833     dg[i] *= b;
834     }
835 greg 1.5 }
836 greg 2.9 free((void *)div);
837 greg 1.3 }
838 greg 2.23 copycolor(rcol, acol);
839 greg 1.1 if (arad <= FTINY)
840 greg 1.16 arad = maxarad;
841 greg 2.3 else
842 greg 2.19 arad = (divcnt+hemi.ns)/arad;
843 greg 1.15 if (pg != NULL) { /* reduce radius if gradient large */
844     d = DOT(pg,pg);
845     if (d*arad*arad > 1.0)
846     arad = 1.0/sqrt(d);
847     }
848 greg 1.16 if (arad < minarad) {
849 greg 1.1 arad = minarad;
850 greg 1.16 if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */
851     d = 1.0/arad/sqrt(d);
852     for (i = 0; i < 3; i++)
853     pg[i] *= d;
854     }
855     }
856 greg 2.3 if ((arad /= sqrt(wt)) > maxarad)
857     arad = maxarad;
858     return(arad);
859 greg 1.1 }
860    
861    
862 greg 2.9 void
863 greg 2.14 comperrs( /* compute initial error estimates */
864     AMBSAMP *da, /* assumes standard ordering */
865 greg 2.23 AMBHEMI *hp
866 greg 2.14 )
867 greg 1.1 {
868     double b, b2;
869     int i, j;
870 greg 2.23 AMBSAMP *dp;
871 greg 1.1 /* sum differences from neighbors */
872     dp = da;
873     for (i = 0; i < hp->nt; i++)
874     for (j = 0; j < hp->np; j++) {
875 greg 1.6 #ifdef DEBUG
876     if (dp->t != i || dp->p != j)
877     error(CONSISTENCY,
878     "division order in comperrs");
879     #endif
880 greg 1.1 b = bright(dp[0].v);
881     if (i > 0) { /* from above */
882     b2 = bright(dp[-hp->np].v) - b;
883     b2 *= b2 * 0.25;
884     dp[0].k += b2;
885     dp[-hp->np].k += b2;
886     }
887     if (j > 0) { /* from behind */
888     b2 = bright(dp[-1].v) - b;
889     b2 *= b2 * 0.25;
890     dp[0].k += b2;
891     dp[-1].k += b2;
892 greg 1.4 } else { /* around */
893     b2 = bright(dp[hp->np-1].v) - b;
894 greg 1.1 b2 *= b2 * 0.25;
895     dp[0].k += b2;
896 greg 1.4 dp[hp->np-1].k += b2;
897 greg 1.1 }
898     dp++;
899     }
900     /* divide by number of neighbors */
901     dp = da;
902     for (j = 0; j < hp->np; j++) /* top row */
903     (dp++)->k *= 1.0/3.0;
904     if (hp->nt < 2)
905     return;
906     for (i = 1; i < hp->nt-1; i++) /* central region */
907     for (j = 0; j < hp->np; j++)
908     (dp++)->k *= 0.25;
909     for (j = 0; j < hp->np; j++) /* bottom row */
910     (dp++)->k *= 1.0/3.0;
911     }
912    
913    
914 greg 2.9 void
915 greg 2.14 posgradient( /* compute position gradient */
916     FVECT gv,
917     AMBSAMP *da, /* assumes standard ordering */
918 greg 2.23 AMBHEMI *hp
919 greg 2.14 )
920 greg 1.1 {
921 greg 2.23 int i, j;
922 greg 2.2 double nextsine, lastsine, b, d;
923 greg 1.2 double mag0, mag1;
924     double phi, cosp, sinp, xd, yd;
925 greg 2.23 AMBSAMP *dp;
926 greg 1.2
927     xd = yd = 0.0;
928     for (j = 0; j < hp->np; j++) {
929     dp = da + j;
930     mag0 = mag1 = 0.0;
931 greg 2.2 lastsine = 0.0;
932 greg 1.2 for (i = 0; i < hp->nt; i++) {
933     #ifdef DEBUG
934     if (dp->t != i || dp->p != j)
935     error(CONSISTENCY,
936     "division order in posgradient");
937     #endif
938     b = bright(dp->v);
939     if (i > 0) {
940     d = dp[-hp->np].r;
941     if (dp[0].r > d) d = dp[0].r;
942 greg 2.2 /* sin(t)*cos(t)^2 */
943     d *= lastsine * (1.0 - (double)i/hp->nt);
944 greg 1.2 mag0 += d*(b - bright(dp[-hp->np].v));
945     }
946 greg 2.2 nextsine = sqrt((double)(i+1)/hp->nt);
947 greg 1.2 if (j > 0) {
948     d = dp[-1].r;
949     if (dp[0].r > d) d = dp[0].r;
950 greg 2.2 mag1 += d * (nextsine - lastsine) *
951     (b - bright(dp[-1].v));
952 greg 1.2 } else {
953     d = dp[hp->np-1].r;
954     if (dp[0].r > d) d = dp[0].r;
955 greg 2.2 mag1 += d * (nextsine - lastsine) *
956     (b - bright(dp[hp->np-1].v));
957 greg 1.2 }
958     dp += hp->np;
959 greg 2.2 lastsine = nextsine;
960 greg 1.2 }
961 greg 2.2 mag0 *= 2.0*PI / hp->np;
962 greg 1.2 phi = 2.0*PI * (double)j/hp->np;
963 gwlarson 2.8 cosp = tcos(phi); sinp = tsin(phi);
964 greg 1.2 xd += mag0*cosp - mag1*sinp;
965     yd += mag0*sinp + mag1*cosp;
966     }
967     for (i = 0; i < 3; i++)
968 greg 2.16 gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI;
969 greg 1.1 }
970    
971    
972 greg 2.9 void
973 greg 2.14 dirgradient( /* compute direction gradient */
974     FVECT gv,
975     AMBSAMP *da, /* assumes standard ordering */
976 greg 2.23 AMBHEMI *hp
977 greg 2.14 )
978 greg 1.1 {
979 greg 2.23 int i, j;
980 greg 1.2 double mag;
981     double phi, xd, yd;
982 greg 2.23 AMBSAMP *dp;
983 greg 1.2
984     xd = yd = 0.0;
985     for (j = 0; j < hp->np; j++) {
986     dp = da + j;
987     mag = 0.0;
988     for (i = 0; i < hp->nt; i++) {
989     #ifdef DEBUG
990     if (dp->t != i || dp->p != j)
991     error(CONSISTENCY,
992     "division order in dirgradient");
993     #endif
994 greg 2.2 /* tan(t) */
995     mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0);
996 greg 1.2 dp += hp->np;
997     }
998     phi = 2.0*PI * (j+.5)/hp->np + PI/2.0;
999 gwlarson 2.8 xd += mag * tcos(phi);
1000     yd += mag * tsin(phi);
1001 greg 1.2 }
1002     for (i = 0; i < 3; i++)
1003 greg 2.16 gv[i] = xd*hp->ux[i] + yd*hp->uy[i];
1004 greg 1.1 }
1005 greg 2.25
1006     #endif /* ! NEWAMB */