ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/Development/ray/src/rt/ambcomp.c
Revision: 2.30
Committed: Wed Apr 23 17:30:10 2014 UTC (11 years, 6 months ago) by greg
Content type: text/plain
Branch: MAIN
Changes since 2.29: +17 -18 lines
Log Message: 
Bug fix and optimization

File Contents

# User Rev Content
1 greg 1.1 #ifndef lint
2 greg 2.30 static const char RCSid[] = "$Id: ambcomp.c,v 2.29 2014/04/23 06:04:17 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     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 greg 2.30 FVECT r_i, r_i1, e_i, rI2_eJ2;
39     double nf, 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     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 greg 2.30 FVECT vcp;
142     double dot_e, dot_er, dot_r, dot_r1, J2;
143     int i;
144 greg 2.27
145     VSUB(ftp->r_i, ap0, rop);
146     VSUB(ftp->r_i1, ap1, rop);
147     VSUB(ftp->e_i, ap1, ap0);
148 greg 2.30 VCROSS(vcp, ftp->e_i, ftp->r_i);
149     ftp->nf = 1.0/DOT(vcp,vcp);
150 greg 2.27 dot_e = DOT(ftp->e_i,ftp->e_i);
151     dot_er = DOT(ftp->e_i, ftp->r_i);
152     dot_r = DOT(ftp->r_i,ftp->r_i);
153     dot_r1 = DOT(ftp->r_i1,ftp->r_i1);
154 greg 2.29 ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) / sqrt(dot_r*dot_r1) ) *
155     sqrt( ftp->nf );
156 greg 2.27 ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)/dot_r1 - dot_er/dot_r +
157     dot_e*ftp->I1 )*0.5*ftp->nf;
158 greg 2.30 J2 = 0.5/dot_e*( 1.0/dot_r - 1.0/dot_r1 ) - dot_er/dot_e*ftp->I2;
159     for (i = 3; i--; )
160     ftp->rI2_eJ2[i] = ftp->I2*ftp->r_i[i] + J2*ftp->e_i[i];
161 greg 2.27 }
162    
163    
164 greg 2.28 /* Compose 3x3 matrix from two vectors */
165 greg 2.27 static void
166     compose_matrix(FVECT mat[3], FVECT va, FVECT vb)
167     {
168     mat[0][0] = 2.0*va[0]*vb[0];
169     mat[1][1] = 2.0*va[1]*vb[1];
170     mat[2][2] = 2.0*va[2]*vb[2];
171     mat[0][1] = mat[1][0] = va[0]*vb[1] + va[1]*vb[0];
172     mat[0][2] = mat[2][0] = va[0]*vb[2] + va[2]*vb[0];
173     mat[1][2] = mat[2][1] = va[1]*vb[2] + va[2]*vb[1];
174     }
175    
176    
177     /* Compute partial 3x3 Hessian matrix for edge */
178     static void
179     comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm)
180     {
181 greg 2.30 FVECT vcp;
182 greg 2.27 FVECT m1[3], m2[3], m3[3], m4[3];
183     double d1, d2, d3, d4;
184     double I3, J3, K3;
185     int i, j;
186     /* compute intermediate coefficients */
187     d1 = 1.0/DOT(ftp->r_i,ftp->r_i);
188     d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
189     d3 = 1.0/DOT(ftp->e_i,ftp->e_i);
190     d4 = DOT(ftp->e_i, ftp->r_i);
191     I3 = 0.25*ftp->nf*( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 +
192 greg 2.29 3.0/d3*ftp->I2 );
193 greg 2.27 J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3;
194     K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3);
195     /* intermediate matrices */
196 greg 2.30 VCROSS(vcp, nrm, ftp->e_i);
197     compose_matrix(m1, vcp, ftp->rI2_eJ2);
198 greg 2.27 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 greg 2.30 VCROSS(vcp, ftp->r_i, ftp->e_i);
202     d1 = DOT(nrm, vcp);
203 greg 2.27 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 greg 2.30 grad[i] = (0.5/PI)*( ftp->I1*vcp[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     /* compute eigenvalues */
323 greg 2.28 if ( quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1],
324 greg 2.27 hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]) != 2 ||
325 greg 2.28 (evalue[0] = fabs(evalue[0])) <= FTINY*FTINY ||
326     (evalue[1] = fabs(evalue[1])) <= FTINY*FTINY )
327 greg 2.27 error(INTERNAL, "bad eigenvalue calculation");
328    
329     if (evalue[0] > evalue[1]) {
330 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[0]));
331     ra[1] = sqrt(sqrt(4.0/evalue[1]));
332 greg 2.27 slope1 = evalue[1];
333     } else {
334 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[1]));
335     ra[1] = sqrt(sqrt(4.0/evalue[0]));
336 greg 2.27 slope1 = evalue[0];
337     }
338     /* compute unit eigenvectors */
339     if (fabs(hess2[0][1]) <= FTINY)
340     return; /* uv OK as is */
341     slope1 = (slope1 - hess2[0][0]) / hess2[0][1];
342     xmag1 = sqrt(1.0/(1.0 + slope1*slope1));
343     for (i = 3; i--; ) {
344     b[i] = xmag1*uv[0][i] + slope1*xmag1*uv[1][i];
345     a[i] = slope1*xmag1*uv[0][i] - xmag1*uv[1][i];
346     }
347     VCOPY(uv[0], a);
348     VCOPY(uv[1], b);
349     }
350    
351    
352 greg 2.26 static void
353     ambHessian( /* anisotropic radii & pos. gradient */
354     AMBHEMI *hp,
355     FVECT uv[2], /* returned */
356 greg 2.28 float ra[2], /* returned (optional) */
357     float pg[2] /* returned (optional) */
358 greg 2.26 )
359     {
360 greg 2.27 static char memerrmsg[] = "out of memory in ambHessian()";
361     FVECT (*hessrow)[3] = NULL;
362     FVECT *gradrow = NULL;
363     FVECT hessian[3];
364     FVECT gradient;
365     FFTRI fftr;
366     int i, j;
367     /* be sure to assign unit vectors */
368     VCOPY(uv[0], hp->ux);
369     VCOPY(uv[1], hp->uy);
370     /* clock-wise vertex traversal from sample POV */
371     if (ra != NULL) { /* initialize Hessian row buffer */
372 greg 2.28 hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1));
373 greg 2.27 if (hessrow == NULL)
374     error(SYSTEM, memerrmsg);
375     memset(hessian, 0, sizeof(hessian));
376     } else if (pg == NULL) /* bogus call? */
377     return;
378     if (pg != NULL) { /* initialize form factor row buffer */
379 greg 2.28 gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1));
380 greg 2.27 if (gradrow == NULL)
381     error(SYSTEM, memerrmsg);
382     memset(gradient, 0, sizeof(gradient));
383     }
384     /* compute first row of edges */
385     for (j = 0; j < hp->ns-1; j++) {
386     comp_fftri(&fftr, ambsamp(hp,0,j).p,
387     ambsamp(hp,0,j+1).p, hp->rp->rop);
388     if (hessrow != NULL)
389     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
390     if (gradrow != NULL)
391     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
392     }
393     /* sum each row of triangles */
394     for (i = 0; i < hp->ns-1; i++) {
395     FVECT hesscol[3]; /* compute first vertical edge */
396     FVECT gradcol;
397     comp_fftri(&fftr, ambsamp(hp,i,0).p,
398     ambsamp(hp,i+1,0).p, hp->rp->rop);
399     if (hessrow != NULL)
400     comp_hessian(hesscol, &fftr, hp->rp->ron);
401     if (gradrow != NULL)
402     comp_gradient(gradcol, &fftr, hp->rp->ron);
403     for (j = 0; j < hp->ns-1; j++) {
404     FVECT hessdia[3]; /* compute triangle contributions */
405     FVECT graddia;
406     COLORV backg;
407     backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1),
408     &ambsamp(hp,i+1,j), hp->rp->rop);
409     /* diagonal (inner) edge */
410     comp_fftri(&fftr, ambsamp(hp,i,j+1).p,
411     ambsamp(hp,i+1,j).p, hp->rp->rop);
412     if (hessrow != NULL) {
413     comp_hessian(hessdia, &fftr, hp->rp->ron);
414     rev_hessian(hesscol);
415     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
416     }
417     if (gradient != NULL) {
418     comp_gradient(graddia, &fftr, hp->rp->ron);
419     rev_gradient(gradcol);
420     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
421     }
422     /* initialize edge in next row */
423     comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p,
424     ambsamp(hp,i+1,j).p, hp->rp->rop);
425     if (hessrow != NULL)
426     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
427     if (gradrow != NULL)
428     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
429     /* new column edge & paired triangle */
430     backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1),
431     &ambsamp(hp,i+1,j), hp->rp->rop);
432     comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p,
433     hp->rp->rop);
434     if (hessrow != NULL) {
435     comp_hessian(hesscol, &fftr, hp->rp->ron);
436     rev_hessian(hessdia);
437     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
438     if (i < hp->ns-2)
439     rev_hessian(hessrow[j]);
440     }
441     if (gradrow != NULL) {
442     comp_gradient(gradcol, &fftr, hp->rp->ron);
443     rev_gradient(graddia);
444     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
445     if (i < hp->ns-2)
446     rev_gradient(gradrow[j]);
447     }
448     }
449     }
450     /* release row buffers */
451     if (hessrow != NULL) free(hessrow);
452     if (gradrow != NULL) free(gradrow);
453    
454     if (ra != NULL) /* extract eigenvectors & radii */
455     eigenvectors(uv, ra, hessian);
456 greg 2.29 if (pg != NULL) { /* tangential position gradient/PI */
457     pg[0] = DOT(gradient, uv[0]) / PI;
458     pg[1] = DOT(gradient, uv[1]) / PI;
459 greg 2.27 }
460     }
461    
462    
463     /* Compute direction gradient from a hemispherical sampling */
464     static void
465     ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2])
466     {
467     struct s_ambsamp *ap;
468 greg 2.29 double dgsum[2];
469 greg 2.27 int n;
470 greg 2.28 FVECT vd;
471     double gfact;
472 greg 2.27
473 greg 2.29 dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */
474 greg 2.27 for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) {
475     /* use vector for azimuth + 90deg */
476     VSUB(vd, ap->p, hp->rp->rop);
477 greg 2.29 /* brightness over cosine factor */
478     gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd);
479     /* -sine = -proj_radius/vd_length */
480     dgsum[0] += DOT(uv[1], vd) * gfact;
481     dgsum[1] -= DOT(uv[0], vd) * gfact;
482 greg 2.26 }
483 greg 2.29 dg[0] = dgsum[0] / (hp->ns*hp->ns);
484     dg[1] = dgsum[1] / (hp->ns*hp->ns);
485 greg 2.26 }
486    
487 greg 2.27
488 greg 2.26 int
489     doambient( /* compute ambient component */
490     COLOR rcol, /* input/output color */
491     RAY *r,
492     double wt,
493 greg 2.27 FVECT uv[2], /* returned (optional) */
494     float ra[2], /* returned (optional) */
495     float pg[2], /* returned (optional) */
496     float dg[2] /* returned (optional) */
497 greg 2.26 )
498     {
499 greg 2.28 AMBHEMI *hp = inithemi(rcol, r, wt);
500 greg 2.26 int cnt = 0;
501     FVECT my_uv[2];
502     double d, acol[3];
503     struct s_ambsamp *ap;
504     int i, j;
505 greg 2.28 /* check/initialize */
506     if (hp == NULL)
507 greg 2.26 return(0);
508     if (uv != NULL)
509     memset(uv, 0, sizeof(FVECT)*2);
510     if (ra != NULL)
511     ra[0] = ra[1] = 0.0;
512     if (pg != NULL)
513     pg[0] = pg[1] = 0.0;
514     if (dg != NULL)
515     dg[0] = dg[1] = 0.0;
516     /* sample the hemisphere */
517     acol[0] = acol[1] = acol[2] = 0.0;
518 greg 2.27 for (i = hp->ns; i--; )
519     for (j = hp->ns; j--; )
520 greg 2.28 if ((ap = ambsample(hp, i, j)) != NULL) {
521 greg 2.26 addcolor(acol, ap->v);
522     ++cnt;
523     }
524     if (!cnt) {
525     setcolor(rcol, 0.0, 0.0, 0.0);
526     free(hp);
527     return(0); /* no valid samples */
528     }
529 greg 2.29 copycolor(rcol, acol); /* final indirect irradiance/PI */
530 greg 2.26 if (cnt < hp->ns*hp->ns || /* incomplete sampling? */
531     (ra == NULL) & (pg == NULL) & (dg == NULL)) {
532     free(hp);
533     return(-1); /* no radius or gradient calc. */
534     }
535 greg 2.29 multcolor(acol, hp->acoef); /* normalize Y values */
536     if ((d = bright(acol)) > FTINY)
537     d = 1.0/d;
538     else
539     d = 0.0;
540     ap = hp->sa; /* relative Y channel from here on... */
541 greg 2.26 for (i = hp->ns*hp->ns; i--; ap++)
542 greg 2.29 colval(ap->v,CIEY) = bright(ap->v)*d + 0.0314;
543 greg 2.26
544     if (uv == NULL) /* make sure we have axis pointers */
545     uv = my_uv;
546     /* compute radii & pos. gradient */
547     ambHessian(hp, uv, ra, pg);
548 greg 2.29
549 greg 2.26 if (dg != NULL) /* compute direction gradient */
550     ambdirgrad(hp, uv, dg);
551 greg 2.29
552 greg 2.28 if (ra != NULL) { /* scale/clamp radii */
553 greg 2.29 if (ra[0] < minarad) {
554     ra[0] = minarad;
555     if (ra[1] < minarad)
556     ra[1] = minarad;
557     }
558     ra[0] *= d = 1.0/sqrt(sqrt(wt));
559 greg 2.26 if ((ra[1] *= d) > 2.0*ra[0])
560     ra[1] = 2.0*ra[0];
561 greg 2.28 if (ra[1] > maxarad) {
562     ra[1] = maxarad;
563     if (ra[0] > maxarad)
564     ra[0] = maxarad;
565     }
566 greg 2.26 }
567     free(hp); /* clean up and return */
568     return(1);
569     }
570    
571    
572 greg 2.25 #else /* ! NEWAMB */
573 greg 1.1
574    
575 greg 2.15 void
576 greg 2.14 inithemi( /* initialize sampling hemisphere */
577 greg 2.23 AMBHEMI *hp,
578 greg 2.16 COLOR ac,
579 greg 2.14 RAY *r,
580     double wt
581     )
582 greg 1.1 {
583 greg 2.16 double d;
584 greg 2.23 int i;
585 greg 2.14 /* set number of divisions */
586 greg 2.16 if (ambacc <= FTINY &&
587 greg 2.20 wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight)))
588 greg 2.16 wt = d; /* avoid ray termination */
589     hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
590 greg 2.14 i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */
591     if (hp->nt < i)
592     hp->nt = i;
593     hp->np = PI * hp->nt + 0.5;
594     /* set number of super-samples */
595 greg 2.15 hp->ns = ambssamp * wt + 0.5;
596 greg 2.16 /* assign coefficient */
597 greg 2.14 copycolor(hp->acoef, ac);
598 greg 2.16 d = 1.0/(hp->nt*hp->np);
599     scalecolor(hp->acoef, d);
600 greg 2.14 /* make axes */
601     VCOPY(hp->uz, r->ron);
602     hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
603     for (i = 0; i < 3; i++)
604     if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
605     break;
606     if (i >= 3)
607     error(CONSISTENCY, "bad ray direction in inithemi");
608     hp->uy[i] = 1.0;
609     fcross(hp->ux, hp->uy, hp->uz);
610     normalize(hp->ux);
611     fcross(hp->uy, hp->uz, hp->ux);
612 greg 1.1 }
613    
614    
615 greg 2.9 int
616 greg 2.14 divsample( /* sample a division */
617 greg 2.23 AMBSAMP *dp,
618 greg 2.14 AMBHEMI *h,
619     RAY *r
620     )
621 greg 1.1 {
622     RAY ar;
623 greg 1.11 int hlist[3];
624     double spt[2];
625 greg 1.1 double xd, yd, zd;
626     double b2;
627     double phi;
628 greg 2.23 int i;
629 greg 2.15 /* ambient coefficient for weight */
630 greg 2.16 if (ambacc > FTINY)
631     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
632     else
633     copycolor(ar.rcoef, h->acoef);
634 greg 2.14 if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
635 greg 1.4 return(-1);
636 greg 2.17 if (ambacc > FTINY) {
637     multcolor(ar.rcoef, h->acoef);
638     scalecolor(ar.rcoef, 1./AVGREFL);
639     }
640 greg 1.1 hlist[0] = r->rno;
641     hlist[1] = dp->t;
642     hlist[2] = dp->p;
643 greg 1.13 multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n));
644 greg 1.11 zd = sqrt((dp->t + spt[0])/h->nt);
645     phi = 2.0*PI * (dp->p + spt[1])/h->np;
646 gwlarson 2.8 xd = tcos(phi) * zd;
647     yd = tsin(phi) * zd;
648 greg 1.1 zd = sqrt(1.0 - zd*zd);
649 greg 1.2 for (i = 0; i < 3; i++)
650     ar.rdir[i] = xd*h->ux[i] +
651     yd*h->uy[i] +
652     zd*h->uz[i];
653 greg 2.22 checknorm(ar.rdir);
654 greg 1.2 dimlist[ndims++] = dp->t*h->np + dp->p + 90171;
655 greg 1.1 rayvalue(&ar);
656     ndims--;
657 greg 2.16 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
658 greg 1.1 addcolor(dp->v, ar.rcol);
659 greg 2.9 /* use rt to improve gradient calc */
660     if (ar.rt > FTINY && ar.rt < FHUGE)
661     dp->r += 1.0/ar.rt;
662 greg 1.1 /* (re)initialize error */
663     if (dp->n++) {
664     b2 = bright(dp->v)/dp->n - bright(ar.rcol);
665     b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1));
666     dp->k = b2/(dp->n*dp->n);
667     } else
668     dp->k = 0.0;
669 greg 1.4 return(0);
670 greg 1.1 }
671    
672    
673 greg 2.14 static int
674     ambcmp( /* decreasing order */
675     const void *p1,
676     const void *p2
677     )
678     {
679     const AMBSAMP *d1 = (const AMBSAMP *)p1;
680     const AMBSAMP *d2 = (const AMBSAMP *)p2;
681    
682     if (d1->k < d2->k)
683     return(1);
684     if (d1->k > d2->k)
685     return(-1);
686     return(0);
687     }
688    
689    
690     static int
691     ambnorm( /* standard order */
692     const void *p1,
693     const void *p2
694     )
695     {
696     const AMBSAMP *d1 = (const AMBSAMP *)p1;
697     const AMBSAMP *d2 = (const AMBSAMP *)p2;
698 greg 2.23 int c;
699 greg 2.14
700     if ( (c = d1->t - d2->t) )
701     return(c);
702     return(d1->p - d2->p);
703     }
704    
705    
706 greg 1.1 double
707 greg 2.14 doambient( /* compute ambient component */
708 greg 2.23 COLOR rcol,
709 greg 2.14 RAY *r,
710     double wt,
711     FVECT pg,
712     FVECT dg
713     )
714 greg 1.1 {
715 greg 2.24 double b, d=0;
716 greg 1.1 AMBHEMI hemi;
717     AMBSAMP *div;
718     AMBSAMP dnew;
719 greg 2.23 double acol[3];
720     AMBSAMP *dp;
721 greg 1.1 double arad;
722 greg 2.19 int divcnt;
723 greg 2.23 int i, j;
724 greg 1.1 /* initialize hemisphere */
725 greg 2.23 inithemi(&hemi, rcol, r, wt);
726 greg 2.19 divcnt = hemi.nt * hemi.np;
727 greg 2.17 /* initialize */
728     if (pg != NULL)
729     pg[0] = pg[1] = pg[2] = 0.0;
730     if (dg != NULL)
731     dg[0] = dg[1] = dg[2] = 0.0;
732 greg 2.23 setcolor(rcol, 0.0, 0.0, 0.0);
733 greg 2.19 if (divcnt == 0)
734 greg 1.1 return(0.0);
735 greg 2.14 /* allocate super-samples */
736 greg 2.15 if (hemi.ns > 0 || pg != NULL || dg != NULL) {
737 greg 2.19 div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP));
738 greg 1.1 if (div == NULL)
739     error(SYSTEM, "out of memory in doambient");
740     } else
741     div = NULL;
742     /* sample the divisions */
743     arad = 0.0;
744 greg 2.23 acol[0] = acol[1] = acol[2] = 0.0;
745 greg 1.1 if ((dp = div) == NULL)
746     dp = &dnew;
747 greg 2.19 divcnt = 0;
748 greg 1.1 for (i = 0; i < hemi.nt; i++)
749     for (j = 0; j < hemi.np; j++) {
750     dp->t = i; dp->p = j;
751     setcolor(dp->v, 0.0, 0.0, 0.0);
752 greg 1.2 dp->r = 0.0;
753 greg 1.1 dp->n = 0;
754 greg 2.16 if (divsample(dp, &hemi, r) < 0) {
755 greg 2.19 if (div != NULL)
756     dp++;
757 greg 2.16 continue;
758     }
759 greg 2.6 arad += dp->r;
760 greg 2.19 divcnt++;
761 greg 1.1 if (div != NULL)
762     dp++;
763 greg 2.6 else
764 greg 1.1 addcolor(acol, dp->v);
765     }
766 greg 2.21 if (!divcnt) {
767     if (div != NULL)
768     free((void *)div);
769 greg 2.19 return(0.0); /* no samples taken */
770 greg 2.21 }
771 greg 2.19 if (divcnt < hemi.nt*hemi.np) {
772     pg = dg = NULL; /* incomplete sampling */
773     hemi.ns = 0;
774     } else if (arad > FTINY && divcnt/arad < minarad) {
775 greg 2.15 hemi.ns = 0; /* close enough */
776 greg 2.19 } else if (hemi.ns > 0) { /* else perform super-sampling? */
777 greg 1.4 comperrs(div, &hemi); /* compute errors */
778 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */
779 greg 1.1 /* super-sample */
780 greg 2.15 for (i = hemi.ns; i > 0; i--) {
781 schorsch 2.11 dnew = *div;
782 greg 2.16 if (divsample(&dnew, &hemi, r) < 0) {
783     dp++;
784     continue;
785     }
786     dp = div; /* reinsert */
787 greg 2.19 j = divcnt < i ? divcnt : i;
788 greg 1.1 while (--j > 0 && dnew.k < dp[1].k) {
789 schorsch 2.11 *dp = *(dp+1);
790 greg 1.1 dp++;
791     }
792 schorsch 2.11 *dp = dnew;
793 greg 1.1 }
794 greg 1.2 if (pg != NULL || dg != NULL) /* restore order */
795 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambnorm);
796 greg 1.1 }
797     /* compute returned values */
798 greg 1.3 if (div != NULL) {
799 greg 2.19 arad = 0.0; /* note: divcnt may be < nt*np */
800     for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) {
801 greg 1.3 arad += dp->r;
802     if (dp->n > 1) {
803     b = 1.0/dp->n;
804     scalecolor(dp->v, b);
805     dp->r *= b;
806     dp->n = 1;
807     }
808     addcolor(acol, dp->v);
809     }
810 greg 1.5 b = bright(acol);
811 greg 1.6 if (b > FTINY) {
812 greg 2.17 b = 1.0/b; /* compute & normalize gradient(s) */
813 greg 1.6 if (pg != NULL) {
814     posgradient(pg, div, &hemi);
815     for (i = 0; i < 3; i++)
816     pg[i] *= b;
817     }
818     if (dg != NULL) {
819     dirgradient(dg, div, &hemi);
820     for (i = 0; i < 3; i++)
821     dg[i] *= b;
822     }
823 greg 1.5 }
824 greg 2.9 free((void *)div);
825 greg 1.3 }
826 greg 2.23 copycolor(rcol, acol);
827 greg 1.1 if (arad <= FTINY)
828 greg 1.16 arad = maxarad;
829 greg 2.3 else
830 greg 2.19 arad = (divcnt+hemi.ns)/arad;
831 greg 1.15 if (pg != NULL) { /* reduce radius if gradient large */
832     d = DOT(pg,pg);
833     if (d*arad*arad > 1.0)
834     arad = 1.0/sqrt(d);
835     }
836 greg 1.16 if (arad < minarad) {
837 greg 1.1 arad = minarad;
838 greg 1.16 if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */
839     d = 1.0/arad/sqrt(d);
840     for (i = 0; i < 3; i++)
841     pg[i] *= d;
842     }
843     }
844 greg 2.3 if ((arad /= sqrt(wt)) > maxarad)
845     arad = maxarad;
846     return(arad);
847 greg 1.1 }
848    
849    
850 greg 2.9 void
851 greg 2.14 comperrs( /* compute initial error estimates */
852     AMBSAMP *da, /* assumes standard ordering */
853 greg 2.23 AMBHEMI *hp
854 greg 2.14 )
855 greg 1.1 {
856     double b, b2;
857     int i, j;
858 greg 2.23 AMBSAMP *dp;
859 greg 1.1 /* sum differences from neighbors */
860     dp = da;
861     for (i = 0; i < hp->nt; i++)
862     for (j = 0; j < hp->np; j++) {
863 greg 1.6 #ifdef DEBUG
864     if (dp->t != i || dp->p != j)
865     error(CONSISTENCY,
866     "division order in comperrs");
867     #endif
868 greg 1.1 b = bright(dp[0].v);
869     if (i > 0) { /* from above */
870     b2 = bright(dp[-hp->np].v) - b;
871     b2 *= b2 * 0.25;
872     dp[0].k += b2;
873     dp[-hp->np].k += b2;
874     }
875     if (j > 0) { /* from behind */
876     b2 = bright(dp[-1].v) - b;
877     b2 *= b2 * 0.25;
878     dp[0].k += b2;
879     dp[-1].k += b2;
880 greg 1.4 } else { /* around */
881     b2 = bright(dp[hp->np-1].v) - b;
882 greg 1.1 b2 *= b2 * 0.25;
883     dp[0].k += b2;
884 greg 1.4 dp[hp->np-1].k += b2;
885 greg 1.1 }
886     dp++;
887     }
888     /* divide by number of neighbors */
889     dp = da;
890     for (j = 0; j < hp->np; j++) /* top row */
891     (dp++)->k *= 1.0/3.0;
892     if (hp->nt < 2)
893     return;
894     for (i = 1; i < hp->nt-1; i++) /* central region */
895     for (j = 0; j < hp->np; j++)
896     (dp++)->k *= 0.25;
897     for (j = 0; j < hp->np; j++) /* bottom row */
898     (dp++)->k *= 1.0/3.0;
899     }
900    
901    
902 greg 2.9 void
903 greg 2.14 posgradient( /* compute position gradient */
904     FVECT gv,
905     AMBSAMP *da, /* assumes standard ordering */
906 greg 2.23 AMBHEMI *hp
907 greg 2.14 )
908 greg 1.1 {
909 greg 2.23 int i, j;
910 greg 2.2 double nextsine, lastsine, b, d;
911 greg 1.2 double mag0, mag1;
912     double phi, cosp, sinp, xd, yd;
913 greg 2.23 AMBSAMP *dp;
914 greg 1.2
915     xd = yd = 0.0;
916     for (j = 0; j < hp->np; j++) {
917     dp = da + j;
918     mag0 = mag1 = 0.0;
919 greg 2.2 lastsine = 0.0;
920 greg 1.2 for (i = 0; i < hp->nt; i++) {
921     #ifdef DEBUG
922     if (dp->t != i || dp->p != j)
923     error(CONSISTENCY,
924     "division order in posgradient");
925     #endif
926     b = bright(dp->v);
927     if (i > 0) {
928     d = dp[-hp->np].r;
929     if (dp[0].r > d) d = dp[0].r;
930 greg 2.2 /* sin(t)*cos(t)^2 */
931     d *= lastsine * (1.0 - (double)i/hp->nt);
932 greg 1.2 mag0 += d*(b - bright(dp[-hp->np].v));
933     }
934 greg 2.2 nextsine = sqrt((double)(i+1)/hp->nt);
935 greg 1.2 if (j > 0) {
936     d = dp[-1].r;
937     if (dp[0].r > d) d = dp[0].r;
938 greg 2.2 mag1 += d * (nextsine - lastsine) *
939     (b - bright(dp[-1].v));
940 greg 1.2 } else {
941     d = dp[hp->np-1].r;
942     if (dp[0].r > d) d = dp[0].r;
943 greg 2.2 mag1 += d * (nextsine - lastsine) *
944     (b - bright(dp[hp->np-1].v));
945 greg 1.2 }
946     dp += hp->np;
947 greg 2.2 lastsine = nextsine;
948 greg 1.2 }
949 greg 2.2 mag0 *= 2.0*PI / hp->np;
950 greg 1.2 phi = 2.0*PI * (double)j/hp->np;
951 gwlarson 2.8 cosp = tcos(phi); sinp = tsin(phi);
952 greg 1.2 xd += mag0*cosp - mag1*sinp;
953     yd += mag0*sinp + mag1*cosp;
954     }
955     for (i = 0; i < 3; i++)
956 greg 2.16 gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI;
957 greg 1.1 }
958    
959    
960 greg 2.9 void
961 greg 2.14 dirgradient( /* compute direction gradient */
962     FVECT gv,
963     AMBSAMP *da, /* assumes standard ordering */
964 greg 2.23 AMBHEMI *hp
965 greg 2.14 )
966 greg 1.1 {
967 greg 2.23 int i, j;
968 greg 1.2 double mag;
969     double phi, xd, yd;
970 greg 2.23 AMBSAMP *dp;
971 greg 1.2
972     xd = yd = 0.0;
973     for (j = 0; j < hp->np; j++) {
974     dp = da + j;
975     mag = 0.0;
976     for (i = 0; i < hp->nt; i++) {
977     #ifdef DEBUG
978     if (dp->t != i || dp->p != j)
979     error(CONSISTENCY,
980     "division order in dirgradient");
981     #endif
982 greg 2.2 /* tan(t) */
983     mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0);
984 greg 1.2 dp += hp->np;
985     }
986     phi = 2.0*PI * (j+.5)/hp->np + PI/2.0;
987 gwlarson 2.8 xd += mag * tcos(phi);
988     yd += mag * tsin(phi);
989 greg 1.2 }
990     for (i = 0; i < 3; i++)
991 greg 2.16 gv[i] = xd*hp->ux[i] + yd*hp->uy[i];
992 greg 1.1 }
993 greg 2.25
994     #endif /* ! NEWAMB */