ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/radiance/ray/src/rt/ambcomp.c
Revision: 2.29
Committed: Wed Apr 23 06:04:17 2014 UTC (10 years ago) by greg
Content type: text/plain
Branch: MAIN
Changes since 2.28: +43 -33 lines
Log Message: 
Fixed numerous errors in Hessian (-DNEWAMB) calculation, doubtless more to go

File Contents

# User Rev Content
1 greg 1.1 #ifndef lint
2 greg 2.29 static const char RCSid[] = "$Id: ambcomp.c,v 2.28 2014/04/19 19:20:47 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     float p[3]; /* intersection point */
32     } 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     FVECT r_i, r_i1, e_i;
39     double nf, I1, I2, J2;
40     } 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     hp->uy[0] = 0.1 - 0.2*frandom();
74     hp->uy[1] = 0.1 - 0.2*frandom();
75     hp->uy[2] = 0.1 - 0.2*frandom();
76 greg 2.26 for (i = 0; i < 3; i++)
77 greg 2.27 if (r->ron[i] < 0.6 && r->ron[i] > -0.6)
78 greg 2.26 break;
79     if (i >= 3)
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     if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0) {
107     setcolor(ap->v, 0., 0., 0.);
108 greg 2.27 VCOPY(ap->p, hp->rp->rop);
109 greg 2.28 return(NULL); /* no sample taken */
110 greg 2.26 }
111     if (ambacc > FTINY) {
112     multcolor(ar.rcoef, hp->acoef);
113     scalecolor(ar.rcoef, 1./AVGREFL);
114     }
115     /* generate hemispherical sample */
116 greg 2.27 SDsquare2disk(spt, (i+.1+.8*frandom())/hp->ns,
117 greg 2.28 (j+.1+.8*frandom())/hp->ns );
118 greg 2.26 zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]);
119     for (ii = 3; ii--; )
120     ar.rdir[ii] = spt[0]*hp->ux[ii] +
121     spt[1]*hp->uy[ii] +
122     zd*hp->rp->ron[ii];
123     checknorm(ar.rdir);
124     dimlist[ndims++] = i*hp->ns + j + 90171;
125     rayvalue(&ar); /* evaluate ray */
126     ndims--;
127     multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
128     copycolor(ap->v, ar.rcol);
129     if (ar.rt > 20.0*maxarad) /* limit vertex distance */
130 greg 2.29 VSUM(ap->p, ar.rorg, ar.rdir, 20.0*maxarad);
131     else
132     VCOPY(ap->p, ar.rop);
133 greg 2.28 return(ap);
134 greg 2.26 }
135    
136    
137 greg 2.27 /* Compute vectors and coefficients for Hessian/gradient calcs */
138     static void
139     comp_fftri(FFTRI *ftp, float ap0[3], float ap1[3], FVECT rop)
140     {
141     FVECT v1;
142     double dot_e, dot_er, dot_r, dot_r1;
143    
144     VSUB(ftp->r_i, ap0, rop);
145     VSUB(ftp->r_i1, ap1, rop);
146     VSUB(ftp->e_i, ap1, ap0);
147     VCROSS(v1, ftp->e_i, ftp->r_i);
148     ftp->nf = 1.0/DOT(v1,v1);
149     dot_e = DOT(ftp->e_i,ftp->e_i);
150     dot_er = DOT(ftp->e_i, ftp->r_i);
151     dot_r = DOT(ftp->r_i,ftp->r_i);
152     dot_r1 = DOT(ftp->r_i1,ftp->r_i1);
153 greg 2.29 ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) / sqrt(dot_r*dot_r1) ) *
154     sqrt( ftp->nf );
155 greg 2.27 ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)/dot_r1 - dot_er/dot_r +
156     dot_e*ftp->I1 )*0.5*ftp->nf;
157 greg 2.29 ftp->J2 = 0.5/dot_e*( 1.0/dot_r - 1.0/dot_r1 ) -
158 greg 2.27 dot_er/dot_e*ftp->I2;
159     }
160    
161    
162 greg 2.28 /* Compose 3x3 matrix from two vectors */
163 greg 2.27 static void
164     compose_matrix(FVECT mat[3], FVECT va, FVECT vb)
165     {
166     mat[0][0] = 2.0*va[0]*vb[0];
167     mat[1][1] = 2.0*va[1]*vb[1];
168     mat[2][2] = 2.0*va[2]*vb[2];
169     mat[0][1] = mat[1][0] = va[0]*vb[1] + va[1]*vb[0];
170     mat[0][2] = mat[2][0] = va[0]*vb[2] + va[2]*vb[0];
171     mat[1][2] = mat[2][1] = va[1]*vb[2] + va[2]*vb[1];
172     }
173    
174    
175     /* Compute partial 3x3 Hessian matrix for edge */
176     static void
177     comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm)
178     {
179     FVECT v1, v2;
180     FVECT m1[3], m2[3], m3[3], m4[3];
181     double d1, d2, d3, d4;
182     double I3, J3, K3;
183     int i, j;
184     /* compute intermediate coefficients */
185     d1 = 1.0/DOT(ftp->r_i,ftp->r_i);
186     d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
187     d3 = 1.0/DOT(ftp->e_i,ftp->e_i);
188     d4 = DOT(ftp->e_i, ftp->r_i);
189     I3 = 0.25*ftp->nf*( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 +
190 greg 2.29 3.0/d3*ftp->I2 );
191 greg 2.27 J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3;
192     K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3);
193     /* intermediate matrices */
194     VCROSS(v1, nrm, ftp->e_i);
195     for (j = 3; j--; )
196 greg 2.28 v2[j] = ftp->I2*ftp->r_i[j] + ftp->J2*ftp->e_i[j];
197 greg 2.27 compose_matrix(m1, v1, v2);
198     compose_matrix(m2, ftp->r_i, ftp->r_i);
199     compose_matrix(m3, ftp->e_i, ftp->e_i);
200     compose_matrix(m4, ftp->r_i, ftp->e_i);
201     VCROSS(v1, ftp->r_i, ftp->e_i);
202     d1 = DOT(nrm, v1);
203     d2 = -d1*ftp->I2;
204     d1 *= 2.0;
205     for (i = 3; i--; ) /* final matrix sum */
206     for (j = 3; j--; ) {
207     hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] +
208     2.0*J3*m4[i][j] );
209     hess[i][j] += d2*(i==j);
210     hess[i][j] *= -1.0/PI;
211     }
212     }
213    
214    
215     /* Reverse hessian calculation result for edge in other direction */
216     static void
217     rev_hessian(FVECT hess[3])
218     {
219     int i;
220    
221     for (i = 3; i--; ) {
222     hess[i][0] = -hess[i][0];
223     hess[i][1] = -hess[i][1];
224     hess[i][2] = -hess[i][2];
225     }
226     }
227    
228    
229     /* Add to radiometric Hessian from the given triangle */
230     static void
231     add2hessian(FVECT hess[3], FVECT ehess1[3],
232     FVECT ehess2[3], FVECT ehess3[3], COLORV v)
233     {
234     int i, j;
235    
236     for (i = 3; i--; )
237     for (j = 3; j--; )
238     hess[i][j] += v*( ehess1[i][j] + ehess2[i][j] + ehess3[i][j] );
239     }
240    
241    
242     /* Compute partial displacement form factor gradient for edge */
243     static void
244     comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm)
245     {
246     FVECT vcp;
247     double f1;
248     int i;
249    
250     VCROSS(vcp, ftp->r_i, ftp->r_i1);
251     f1 = 2.0*DOT(nrm, vcp);
252     VCROSS(vcp, nrm, ftp->e_i);
253     for (i = 3; i--; )
254     grad[i] = (0.5/PI)*( ftp->I1*vcp[i] +
255     f1*(ftp->I2*ftp->r_i[i] + ftp->J2*ftp->e_i[i]) );
256     }
257    
258    
259     /* Reverse gradient calculation result for edge in other direction */
260     static void
261     rev_gradient(FVECT grad)
262     {
263     grad[0] = -grad[0];
264     grad[1] = -grad[1];
265     grad[2] = -grad[2];
266     }
267    
268    
269     /* Add to displacement gradient from the given triangle */
270     static void
271     add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v)
272     {
273     int i;
274    
275     for (i = 3; i--; )
276     grad[i] += v*( egrad1[i] + egrad2[i] + egrad3[i] );
277     }
278    
279    
280     /* Return brightness of furthest ambient sample */
281     static COLORV
282     back_ambval(struct s_ambsamp *ap1, struct s_ambsamp *ap2,
283     struct s_ambsamp *ap3, FVECT orig)
284     {
285     COLORV vback;
286     FVECT vec;
287     double d2, d2best;
288    
289     VSUB(vec, ap1->p, orig);
290     d2best = DOT(vec,vec);
291 greg 2.29 vback = colval(ap1->v,CIEY);
292 greg 2.27 VSUB(vec, ap2->p, orig);
293     d2 = DOT(vec,vec);
294     if (d2 > d2best) {
295     d2best = d2;
296 greg 2.29 vback = colval(ap2->v,CIEY);
297 greg 2.27 }
298     VSUB(vec, ap3->p, orig);
299     d2 = DOT(vec,vec);
300     if (d2 > d2best)
301 greg 2.29 return(colval(ap3->v,CIEY));
302 greg 2.27 return(vback);
303     }
304    
305    
306     /* Compute anisotropic radii and eigenvector directions */
307     static int
308     eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3])
309     {
310     double hess2[2][2];
311     FVECT a, b;
312     double evalue[2], slope1, xmag1;
313     int i;
314     /* project Hessian to sample plane */
315     for (i = 3; i--; ) {
316     a[i] = DOT(hessian[i], uv[0]);
317     b[i] = DOT(hessian[i], uv[1]);
318     }
319     hess2[0][0] = DOT(uv[0], a);
320     hess2[0][1] = DOT(uv[0], b);
321     hess2[1][0] = DOT(uv[1], a);
322     hess2[1][1] = DOT(uv[1], b);
323     /* compute eigenvalues */
324 greg 2.28 if ( quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1],
325 greg 2.27 hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]) != 2 ||
326 greg 2.28 (evalue[0] = fabs(evalue[0])) <= FTINY*FTINY ||
327     (evalue[1] = fabs(evalue[1])) <= FTINY*FTINY )
328 greg 2.27 error(INTERNAL, "bad eigenvalue calculation");
329    
330     if (evalue[0] > evalue[1]) {
331 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[0]));
332     ra[1] = sqrt(sqrt(4.0/evalue[1]));
333 greg 2.27 slope1 = evalue[1];
334     } else {
335 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[1]));
336     ra[1] = sqrt(sqrt(4.0/evalue[0]));
337 greg 2.27 slope1 = evalue[0];
338     }
339     /* compute unit eigenvectors */
340     if (fabs(hess2[0][1]) <= FTINY)
341     return; /* uv OK as is */
342     slope1 = (slope1 - hess2[0][0]) / hess2[0][1];
343     xmag1 = sqrt(1.0/(1.0 + slope1*slope1));
344     for (i = 3; i--; ) {
345     b[i] = xmag1*uv[0][i] + slope1*xmag1*uv[1][i];
346     a[i] = slope1*xmag1*uv[0][i] - xmag1*uv[1][i];
347     }
348     VCOPY(uv[0], a);
349     VCOPY(uv[1], b);
350     }
351    
352    
353 greg 2.26 static void
354     ambHessian( /* anisotropic radii & pos. gradient */
355     AMBHEMI *hp,
356     FVECT uv[2], /* returned */
357 greg 2.28 float ra[2], /* returned (optional) */
358     float pg[2] /* returned (optional) */
359 greg 2.26 )
360     {
361 greg 2.27 static char memerrmsg[] = "out of memory in ambHessian()";
362     FVECT (*hessrow)[3] = NULL;
363     FVECT *gradrow = NULL;
364     FVECT hessian[3];
365     FVECT gradient;
366     FFTRI fftr;
367     int i, j;
368     /* be sure to assign unit vectors */
369     VCOPY(uv[0], hp->ux);
370     VCOPY(uv[1], hp->uy);
371     /* clock-wise vertex traversal from sample POV */
372     if (ra != NULL) { /* initialize Hessian row buffer */
373 greg 2.28 hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1));
374 greg 2.27 if (hessrow == NULL)
375     error(SYSTEM, memerrmsg);
376     memset(hessian, 0, sizeof(hessian));
377     } else if (pg == NULL) /* bogus call? */
378     return;
379     if (pg != NULL) { /* initialize form factor row buffer */
380 greg 2.28 gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1));
381 greg 2.27 if (gradrow == NULL)
382     error(SYSTEM, memerrmsg);
383     memset(gradient, 0, sizeof(gradient));
384     }
385     /* compute first row of edges */
386     for (j = 0; j < hp->ns-1; j++) {
387     comp_fftri(&fftr, ambsamp(hp,0,j).p,
388     ambsamp(hp,0,j+1).p, hp->rp->rop);
389     if (hessrow != NULL)
390     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
391     if (gradrow != NULL)
392     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
393     }
394     /* sum each row of triangles */
395     for (i = 0; i < hp->ns-1; i++) {
396     FVECT hesscol[3]; /* compute first vertical edge */
397     FVECT gradcol;
398     comp_fftri(&fftr, ambsamp(hp,i,0).p,
399     ambsamp(hp,i+1,0).p, hp->rp->rop);
400     if (hessrow != NULL)
401     comp_hessian(hesscol, &fftr, hp->rp->ron);
402     if (gradrow != NULL)
403     comp_gradient(gradcol, &fftr, hp->rp->ron);
404     for (j = 0; j < hp->ns-1; j++) {
405     FVECT hessdia[3]; /* compute triangle contributions */
406     FVECT graddia;
407     COLORV backg;
408     backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1),
409     &ambsamp(hp,i+1,j), hp->rp->rop);
410     /* diagonal (inner) edge */
411     comp_fftri(&fftr, ambsamp(hp,i,j+1).p,
412     ambsamp(hp,i+1,j).p, hp->rp->rop);
413     if (hessrow != NULL) {
414     comp_hessian(hessdia, &fftr, hp->rp->ron);
415     rev_hessian(hesscol);
416     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
417     }
418     if (gradient != NULL) {
419     comp_gradient(graddia, &fftr, hp->rp->ron);
420     rev_gradient(gradcol);
421     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
422     }
423     /* initialize edge in next row */
424     comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p,
425     ambsamp(hp,i+1,j).p, hp->rp->rop);
426     if (hessrow != NULL)
427     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
428     if (gradrow != NULL)
429     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
430     /* new column edge & paired triangle */
431     backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1),
432     &ambsamp(hp,i+1,j), hp->rp->rop);
433     comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p,
434     hp->rp->rop);
435     if (hessrow != NULL) {
436     comp_hessian(hesscol, &fftr, hp->rp->ron);
437     rev_hessian(hessdia);
438     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
439     if (i < hp->ns-2)
440     rev_hessian(hessrow[j]);
441     }
442     if (gradrow != NULL) {
443     comp_gradient(gradcol, &fftr, hp->rp->ron);
444     rev_gradient(graddia);
445     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
446     if (i < hp->ns-2)
447     rev_gradient(gradrow[j]);
448     }
449     }
450     }
451     /* release row buffers */
452     if (hessrow != NULL) free(hessrow);
453     if (gradrow != NULL) free(gradrow);
454    
455     if (ra != NULL) /* extract eigenvectors & radii */
456     eigenvectors(uv, ra, hessian);
457 greg 2.29 if (pg != NULL) { /* tangential position gradient/PI */
458     pg[0] = DOT(gradient, uv[0]) / PI;
459     pg[1] = DOT(gradient, uv[1]) / PI;
460 greg 2.27 }
461     }
462    
463    
464     /* Compute direction gradient from a hemispherical sampling */
465     static void
466     ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2])
467     {
468     struct s_ambsamp *ap;
469 greg 2.29 double dgsum[2];
470 greg 2.27 int n;
471 greg 2.28 FVECT vd;
472     double gfact;
473 greg 2.27
474 greg 2.29 dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */
475 greg 2.27 for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) {
476     /* use vector for azimuth + 90deg */
477     VSUB(vd, ap->p, hp->rp->rop);
478 greg 2.29 /* brightness over cosine factor */
479     gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd);
480     /* -sine = -proj_radius/vd_length */
481     dgsum[0] += DOT(uv[1], vd) * gfact;
482     dgsum[1] -= DOT(uv[0], vd) * gfact;
483 greg 2.26 }
484 greg 2.29 dg[0] = dgsum[0] / (hp->ns*hp->ns);
485     dg[1] = dgsum[1] / (hp->ns*hp->ns);
486 greg 2.26 }
487    
488 greg 2.27
489 greg 2.26 int
490     doambient( /* compute ambient component */
491     COLOR rcol, /* input/output color */
492     RAY *r,
493     double wt,
494 greg 2.27 FVECT uv[2], /* returned (optional) */
495     float ra[2], /* returned (optional) */
496     float pg[2], /* returned (optional) */
497     float dg[2] /* returned (optional) */
498 greg 2.26 )
499     {
500 greg 2.28 AMBHEMI *hp = inithemi(rcol, r, wt);
501 greg 2.26 int cnt = 0;
502     FVECT my_uv[2];
503     double d, acol[3];
504     struct s_ambsamp *ap;
505     int i, j;
506 greg 2.28 /* check/initialize */
507     if (hp == NULL)
508 greg 2.26 return(0);
509     if (uv != NULL)
510     memset(uv, 0, sizeof(FVECT)*2);
511     if (ra != NULL)
512     ra[0] = ra[1] = 0.0;
513     if (pg != NULL)
514     pg[0] = pg[1] = 0.0;
515     if (dg != NULL)
516     dg[0] = dg[1] = 0.0;
517     /* sample the hemisphere */
518     acol[0] = acol[1] = acol[2] = 0.0;
519 greg 2.27 for (i = hp->ns; i--; )
520     for (j = hp->ns; j--; )
521 greg 2.28 if ((ap = ambsample(hp, i, j)) != NULL) {
522 greg 2.26 addcolor(acol, ap->v);
523     ++cnt;
524     }
525     if (!cnt) {
526     setcolor(rcol, 0.0, 0.0, 0.0);
527     free(hp);
528     return(0); /* no valid samples */
529     }
530 greg 2.29 copycolor(rcol, acol); /* final indirect irradiance/PI */
531 greg 2.26 if (cnt < hp->ns*hp->ns || /* incomplete sampling? */
532     (ra == NULL) & (pg == NULL) & (dg == NULL)) {
533     free(hp);
534     return(-1); /* no radius or gradient calc. */
535     }
536 greg 2.29 multcolor(acol, hp->acoef); /* normalize Y values */
537     if ((d = bright(acol)) > FTINY)
538     d = 1.0/d;
539     else
540     d = 0.0;
541     ap = hp->sa; /* relative Y channel from here on... */
542 greg 2.26 for (i = hp->ns*hp->ns; i--; ap++)
543 greg 2.29 colval(ap->v,CIEY) = bright(ap->v)*d + 0.0314;
544 greg 2.26
545     if (uv == NULL) /* make sure we have axis pointers */
546     uv = my_uv;
547     /* compute radii & pos. gradient */
548     ambHessian(hp, uv, ra, pg);
549 greg 2.29
550 greg 2.26 if (dg != NULL) /* compute direction gradient */
551     ambdirgrad(hp, uv, dg);
552 greg 2.29
553 greg 2.28 if (ra != NULL) { /* scale/clamp radii */
554 greg 2.29 if (ra[0] < minarad) {
555     ra[0] = minarad;
556     if (ra[1] < minarad)
557     ra[1] = minarad;
558     }
559     ra[0] *= d = 1.0/sqrt(sqrt(wt));
560 greg 2.26 if ((ra[1] *= d) > 2.0*ra[0])
561     ra[1] = 2.0*ra[0];
562 greg 2.28 if (ra[1] > maxarad) {
563     ra[1] = maxarad;
564     if (ra[0] > maxarad)
565     ra[0] = maxarad;
566     }
567 greg 2.26 }
568     free(hp); /* clean up and return */
569     return(1);
570     }
571    
572    
573 greg 2.25 #else /* ! NEWAMB */
574 greg 1.1
575    
576 greg 2.15 void
577 greg 2.14 inithemi( /* initialize sampling hemisphere */
578 greg 2.23 AMBHEMI *hp,
579 greg 2.16 COLOR ac,
580 greg 2.14 RAY *r,
581     double wt
582     )
583 greg 1.1 {
584 greg 2.16 double d;
585 greg 2.23 int i;
586 greg 2.14 /* set number of divisions */
587 greg 2.16 if (ambacc <= FTINY &&
588 greg 2.20 wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight)))
589 greg 2.16 wt = d; /* avoid ray termination */
590     hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
591 greg 2.14 i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */
592     if (hp->nt < i)
593     hp->nt = i;
594     hp->np = PI * hp->nt + 0.5;
595     /* set number of super-samples */
596 greg 2.15 hp->ns = ambssamp * wt + 0.5;
597 greg 2.16 /* assign coefficient */
598 greg 2.14 copycolor(hp->acoef, ac);
599 greg 2.16 d = 1.0/(hp->nt*hp->np);
600     scalecolor(hp->acoef, d);
601 greg 2.14 /* make axes */
602     VCOPY(hp->uz, r->ron);
603     hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
604     for (i = 0; i < 3; i++)
605     if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
606     break;
607     if (i >= 3)
608     error(CONSISTENCY, "bad ray direction in inithemi");
609     hp->uy[i] = 1.0;
610     fcross(hp->ux, hp->uy, hp->uz);
611     normalize(hp->ux);
612     fcross(hp->uy, hp->uz, hp->ux);
613 greg 1.1 }
614    
615    
616 greg 2.9 int
617 greg 2.14 divsample( /* sample a division */
618 greg 2.23 AMBSAMP *dp,
619 greg 2.14 AMBHEMI *h,
620     RAY *r
621     )
622 greg 1.1 {
623     RAY ar;
624 greg 1.11 int hlist[3];
625     double spt[2];
626 greg 1.1 double xd, yd, zd;
627     double b2;
628     double phi;
629 greg 2.23 int i;
630 greg 2.15 /* ambient coefficient for weight */
631 greg 2.16 if (ambacc > FTINY)
632     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
633     else
634     copycolor(ar.rcoef, h->acoef);
635 greg 2.14 if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
636 greg 1.4 return(-1);
637 greg 2.17 if (ambacc > FTINY) {
638     multcolor(ar.rcoef, h->acoef);
639     scalecolor(ar.rcoef, 1./AVGREFL);
640     }
641 greg 1.1 hlist[0] = r->rno;
642     hlist[1] = dp->t;
643     hlist[2] = dp->p;
644 greg 1.13 multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n));
645 greg 1.11 zd = sqrt((dp->t + spt[0])/h->nt);
646     phi = 2.0*PI * (dp->p + spt[1])/h->np;
647 gwlarson 2.8 xd = tcos(phi) * zd;
648     yd = tsin(phi) * zd;
649 greg 1.1 zd = sqrt(1.0 - zd*zd);
650 greg 1.2 for (i = 0; i < 3; i++)
651     ar.rdir[i] = xd*h->ux[i] +
652     yd*h->uy[i] +
653     zd*h->uz[i];
654 greg 2.22 checknorm(ar.rdir);
655 greg 1.2 dimlist[ndims++] = dp->t*h->np + dp->p + 90171;
656 greg 1.1 rayvalue(&ar);
657     ndims--;
658 greg 2.16 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
659 greg 1.1 addcolor(dp->v, ar.rcol);
660 greg 2.9 /* use rt to improve gradient calc */
661     if (ar.rt > FTINY && ar.rt < FHUGE)
662     dp->r += 1.0/ar.rt;
663 greg 1.1 /* (re)initialize error */
664     if (dp->n++) {
665     b2 = bright(dp->v)/dp->n - bright(ar.rcol);
666     b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1));
667     dp->k = b2/(dp->n*dp->n);
668     } else
669     dp->k = 0.0;
670 greg 1.4 return(0);
671 greg 1.1 }
672    
673    
674 greg 2.14 static int
675     ambcmp( /* decreasing order */
676     const void *p1,
677     const void *p2
678     )
679     {
680     const AMBSAMP *d1 = (const AMBSAMP *)p1;
681     const AMBSAMP *d2 = (const AMBSAMP *)p2;
682    
683     if (d1->k < d2->k)
684     return(1);
685     if (d1->k > d2->k)
686     return(-1);
687     return(0);
688     }
689    
690    
691     static int
692     ambnorm( /* standard order */
693     const void *p1,
694     const void *p2
695     )
696     {
697     const AMBSAMP *d1 = (const AMBSAMP *)p1;
698     const AMBSAMP *d2 = (const AMBSAMP *)p2;
699 greg 2.23 int c;
700 greg 2.14
701     if ( (c = d1->t - d2->t) )
702     return(c);
703     return(d1->p - d2->p);
704     }
705    
706    
707 greg 1.1 double
708 greg 2.14 doambient( /* compute ambient component */
709 greg 2.23 COLOR rcol,
710 greg 2.14 RAY *r,
711     double wt,
712     FVECT pg,
713     FVECT dg
714     )
715 greg 1.1 {
716 greg 2.24 double b, d=0;
717 greg 1.1 AMBHEMI hemi;
718     AMBSAMP *div;
719     AMBSAMP dnew;
720 greg 2.23 double acol[3];
721     AMBSAMP *dp;
722 greg 1.1 double arad;
723 greg 2.19 int divcnt;
724 greg 2.23 int i, j;
725 greg 1.1 /* initialize hemisphere */
726 greg 2.23 inithemi(&hemi, rcol, r, wt);
727 greg 2.19 divcnt = hemi.nt * hemi.np;
728 greg 2.17 /* initialize */
729     if (pg != NULL)
730     pg[0] = pg[1] = pg[2] = 0.0;
731     if (dg != NULL)
732     dg[0] = dg[1] = dg[2] = 0.0;
733 greg 2.23 setcolor(rcol, 0.0, 0.0, 0.0);
734 greg 2.19 if (divcnt == 0)
735 greg 1.1 return(0.0);
736 greg 2.14 /* allocate super-samples */
737 greg 2.15 if (hemi.ns > 0 || pg != NULL || dg != NULL) {
738 greg 2.19 div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP));
739 greg 1.1 if (div == NULL)
740     error(SYSTEM, "out of memory in doambient");
741     } else
742     div = NULL;
743     /* sample the divisions */
744     arad = 0.0;
745 greg 2.23 acol[0] = acol[1] = acol[2] = 0.0;
746 greg 1.1 if ((dp = div) == NULL)
747     dp = &dnew;
748 greg 2.19 divcnt = 0;
749 greg 1.1 for (i = 0; i < hemi.nt; i++)
750     for (j = 0; j < hemi.np; j++) {
751     dp->t = i; dp->p = j;
752     setcolor(dp->v, 0.0, 0.0, 0.0);
753 greg 1.2 dp->r = 0.0;
754 greg 1.1 dp->n = 0;
755 greg 2.16 if (divsample(dp, &hemi, r) < 0) {
756 greg 2.19 if (div != NULL)
757     dp++;
758 greg 2.16 continue;
759     }
760 greg 2.6 arad += dp->r;
761 greg 2.19 divcnt++;
762 greg 1.1 if (div != NULL)
763     dp++;
764 greg 2.6 else
765 greg 1.1 addcolor(acol, dp->v);
766     }
767 greg 2.21 if (!divcnt) {
768     if (div != NULL)
769     free((void *)div);
770 greg 2.19 return(0.0); /* no samples taken */
771 greg 2.21 }
772 greg 2.19 if (divcnt < hemi.nt*hemi.np) {
773     pg = dg = NULL; /* incomplete sampling */
774     hemi.ns = 0;
775     } else if (arad > FTINY && divcnt/arad < minarad) {
776 greg 2.15 hemi.ns = 0; /* close enough */
777 greg 2.19 } else if (hemi.ns > 0) { /* else perform super-sampling? */
778 greg 1.4 comperrs(div, &hemi); /* compute errors */
779 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */
780 greg 1.1 /* super-sample */
781 greg 2.15 for (i = hemi.ns; i > 0; i--) {
782 schorsch 2.11 dnew = *div;
783 greg 2.16 if (divsample(&dnew, &hemi, r) < 0) {
784     dp++;
785     continue;
786     }
787     dp = div; /* reinsert */
788 greg 2.19 j = divcnt < i ? divcnt : i;
789 greg 1.1 while (--j > 0 && dnew.k < dp[1].k) {
790 schorsch 2.11 *dp = *(dp+1);
791 greg 1.1 dp++;
792     }
793 schorsch 2.11 *dp = dnew;
794 greg 1.1 }
795 greg 1.2 if (pg != NULL || dg != NULL) /* restore order */
796 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambnorm);
797 greg 1.1 }
798     /* compute returned values */
799 greg 1.3 if (div != NULL) {
800 greg 2.19 arad = 0.0; /* note: divcnt may be < nt*np */
801     for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) {
802 greg 1.3 arad += dp->r;
803     if (dp->n > 1) {
804     b = 1.0/dp->n;
805     scalecolor(dp->v, b);
806     dp->r *= b;
807     dp->n = 1;
808     }
809     addcolor(acol, dp->v);
810     }
811 greg 1.5 b = bright(acol);
812 greg 1.6 if (b > FTINY) {
813 greg 2.17 b = 1.0/b; /* compute & normalize gradient(s) */
814 greg 1.6 if (pg != NULL) {
815     posgradient(pg, div, &hemi);
816     for (i = 0; i < 3; i++)
817     pg[i] *= b;
818     }
819     if (dg != NULL) {
820     dirgradient(dg, div, &hemi);
821     for (i = 0; i < 3; i++)
822     dg[i] *= b;
823     }
824 greg 1.5 }
825 greg 2.9 free((void *)div);
826 greg 1.3 }
827 greg 2.23 copycolor(rcol, acol);
828 greg 1.1 if (arad <= FTINY)
829 greg 1.16 arad = maxarad;
830 greg 2.3 else
831 greg 2.19 arad = (divcnt+hemi.ns)/arad;
832 greg 1.15 if (pg != NULL) { /* reduce radius if gradient large */
833     d = DOT(pg,pg);
834     if (d*arad*arad > 1.0)
835     arad = 1.0/sqrt(d);
836     }
837 greg 1.16 if (arad < minarad) {
838 greg 1.1 arad = minarad;
839 greg 1.16 if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */
840     d = 1.0/arad/sqrt(d);
841     for (i = 0; i < 3; i++)
842     pg[i] *= d;
843     }
844     }
845 greg 2.3 if ((arad /= sqrt(wt)) > maxarad)
846     arad = maxarad;
847     return(arad);
848 greg 1.1 }
849    
850    
851 greg 2.9 void
852 greg 2.14 comperrs( /* compute initial error estimates */
853     AMBSAMP *da, /* assumes standard ordering */
854 greg 2.23 AMBHEMI *hp
855 greg 2.14 )
856 greg 1.1 {
857     double b, b2;
858     int i, j;
859 greg 2.23 AMBSAMP *dp;
860 greg 1.1 /* sum differences from neighbors */
861     dp = da;
862     for (i = 0; i < hp->nt; i++)
863     for (j = 0; j < hp->np; j++) {
864 greg 1.6 #ifdef DEBUG
865     if (dp->t != i || dp->p != j)
866     error(CONSISTENCY,
867     "division order in comperrs");
868     #endif
869 greg 1.1 b = bright(dp[0].v);
870     if (i > 0) { /* from above */
871     b2 = bright(dp[-hp->np].v) - b;
872     b2 *= b2 * 0.25;
873     dp[0].k += b2;
874     dp[-hp->np].k += b2;
875     }
876     if (j > 0) { /* from behind */
877     b2 = bright(dp[-1].v) - b;
878     b2 *= b2 * 0.25;
879     dp[0].k += b2;
880     dp[-1].k += b2;
881 greg 1.4 } else { /* around */
882     b2 = bright(dp[hp->np-1].v) - b;
883 greg 1.1 b2 *= b2 * 0.25;
884     dp[0].k += b2;
885 greg 1.4 dp[hp->np-1].k += b2;
886 greg 1.1 }
887     dp++;
888     }
889     /* divide by number of neighbors */
890     dp = da;
891     for (j = 0; j < hp->np; j++) /* top row */
892     (dp++)->k *= 1.0/3.0;
893     if (hp->nt < 2)
894     return;
895     for (i = 1; i < hp->nt-1; i++) /* central region */
896     for (j = 0; j < hp->np; j++)
897     (dp++)->k *= 0.25;
898     for (j = 0; j < hp->np; j++) /* bottom row */
899     (dp++)->k *= 1.0/3.0;
900     }
901    
902    
903 greg 2.9 void
904 greg 2.14 posgradient( /* compute position gradient */
905     FVECT gv,
906     AMBSAMP *da, /* assumes standard ordering */
907 greg 2.23 AMBHEMI *hp
908 greg 2.14 )
909 greg 1.1 {
910 greg 2.23 int i, j;
911 greg 2.2 double nextsine, lastsine, b, d;
912 greg 1.2 double mag0, mag1;
913     double phi, cosp, sinp, xd, yd;
914 greg 2.23 AMBSAMP *dp;
915 greg 1.2
916     xd = yd = 0.0;
917     for (j = 0; j < hp->np; j++) {
918     dp = da + j;
919     mag0 = mag1 = 0.0;
920 greg 2.2 lastsine = 0.0;
921 greg 1.2 for (i = 0; i < hp->nt; i++) {
922     #ifdef DEBUG
923     if (dp->t != i || dp->p != j)
924     error(CONSISTENCY,
925     "division order in posgradient");
926     #endif
927     b = bright(dp->v);
928     if (i > 0) {
929     d = dp[-hp->np].r;
930     if (dp[0].r > d) d = dp[0].r;
931 greg 2.2 /* sin(t)*cos(t)^2 */
932     d *= lastsine * (1.0 - (double)i/hp->nt);
933 greg 1.2 mag0 += d*(b - bright(dp[-hp->np].v));
934     }
935 greg 2.2 nextsine = sqrt((double)(i+1)/hp->nt);
936 greg 1.2 if (j > 0) {
937     d = dp[-1].r;
938     if (dp[0].r > d) d = dp[0].r;
939 greg 2.2 mag1 += d * (nextsine - lastsine) *
940     (b - bright(dp[-1].v));
941 greg 1.2 } else {
942     d = dp[hp->np-1].r;
943     if (dp[0].r > d) d = dp[0].r;
944 greg 2.2 mag1 += d * (nextsine - lastsine) *
945     (b - bright(dp[hp->np-1].v));
946 greg 1.2 }
947     dp += hp->np;
948 greg 2.2 lastsine = nextsine;
949 greg 1.2 }
950 greg 2.2 mag0 *= 2.0*PI / hp->np;
951 greg 1.2 phi = 2.0*PI * (double)j/hp->np;
952 gwlarson 2.8 cosp = tcos(phi); sinp = tsin(phi);
953 greg 1.2 xd += mag0*cosp - mag1*sinp;
954     yd += mag0*sinp + mag1*cosp;
955     }
956     for (i = 0; i < 3; i++)
957 greg 2.16 gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI;
958 greg 1.1 }
959    
960    
961 greg 2.9 void
962 greg 2.14 dirgradient( /* compute direction gradient */
963     FVECT gv,
964     AMBSAMP *da, /* assumes standard ordering */
965 greg 2.23 AMBHEMI *hp
966 greg 2.14 )
967 greg 1.1 {
968 greg 2.23 int i, j;
969 greg 1.2 double mag;
970     double phi, xd, yd;
971 greg 2.23 AMBSAMP *dp;
972 greg 1.2
973     xd = yd = 0.0;
974     for (j = 0; j < hp->np; j++) {
975     dp = da + j;
976     mag = 0.0;
977     for (i = 0; i < hp->nt; i++) {
978     #ifdef DEBUG
979     if (dp->t != i || dp->p != j)
980     error(CONSISTENCY,
981     "division order in dirgradient");
982     #endif
983 greg 2.2 /* tan(t) */
984     mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0);
985 greg 1.2 dp += hp->np;
986     }
987     phi = 2.0*PI * (j+.5)/hp->np + PI/2.0;
988 gwlarson 2.8 xd += mag * tcos(phi);
989     yd += mag * tsin(phi);
990 greg 1.2 }
991     for (i = 0; i < 3; i++)
992 greg 2.16 gv[i] = xd*hp->ux[i] + yd*hp->uy[i];
993 greg 1.1 }
994 greg 2.25
995     #endif /* ! NEWAMB */