ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/radiance/ray/src/rt/ambcomp.c
Revision: 2.73
Committed: Fri Oct 14 00:54:21 2016 UTC (7 years, 7 months ago) by greg
Content type: text/plain
Branch: MAIN
Changes since 2.72: +40 -9 lines
Log Message:
Fixed regression in genBSDF affecting Klems normalization

File Contents

# User Rev Content
1 greg 1.1 #ifndef lint
2 greg 2.73 static const char RCSid[] = "$Id: ambcomp.c,v 2.72 2016/03/16 15:43:04 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 2.46 * Added book-keeping optimization to avoid calculations that would
12     * cancel due to traversal both directions on edges that are adjacent
13     * to same-valued triangles. This cuts about half of Hessian math.
14     *
15 greg 2.9 * Declarations of external symbols in ambient.h
16     */
17    
18 greg 2.10 #include "copyright.h"
19 greg 1.1
20     #include "ray.h"
21 greg 2.25 #include "ambient.h"
22     #include "random.h"
23 greg 1.1
24 greg 2.63 #ifndef OLDAMB
25 greg 1.1
26 greg 2.26 extern void SDsquare2disk(double ds[2], double seedx, double seedy);
27    
28     typedef struct {
29 greg 2.44 COLOR v; /* hemisphere sample value */
30 greg 2.47 float d; /* reciprocal distance (1/rt) */
31 greg 2.44 FVECT p; /* intersection point */
32     } AMBSAMP; /* sample value */
33    
34     typedef struct {
35 greg 2.26 RAY *rp; /* originating ray sample */
36     int ns; /* number of samples per axis */
37 greg 2.61 int sampOK; /* acquired full sample set? */
38 greg 2.26 COLOR acoef; /* division contribution coefficient */
39 greg 2.61 double acol[3]; /* accumulated color */
40     FVECT ux, uy; /* tangent axis unit vectors */
41 greg 2.44 AMBSAMP sa[1]; /* sample array (extends struct) */
42 greg 2.26 } AMBHEMI; /* ambient sample hemisphere */
43    
44 greg 2.56 #define AI(h,i,j) ((i)*(h)->ns + (j))
45     #define ambsam(h,i,j) (h)->sa[AI(h,i,j)]
46 greg 2.26
47 greg 2.27 typedef struct {
48 greg 2.35 FVECT r_i, r_i1, e_i, rcp, rI2_eJ2;
49     double I1, I2;
50 greg 2.27 } FFTRI; /* vectors and coefficients for Hessian calculation */
51    
52 greg 2.26
53 greg 2.61 static int
54 greg 2.73 ambcollision( /* proposed direciton collides? */
55     AMBHEMI *hp,
56     int i,
57     int j,
58     FVECT dv
59     )
60     {
61     const double cos_thresh = 0.9999995; /* about 3.44 arcminutes */
62     int ii, jj;
63    
64     for (ii = i-1; ii <= i+1; ii++) {
65     if (ii < 0) continue;
66     if (ii >= hp->ns) break;
67     for (jj = j-1; jj <= j+1; jj++) {
68     AMBSAMP *ap;
69     FVECT avec;
70     double dprod;
71     if (jj < 0) continue;
72     if (jj >= hp->ns) break;
73     if ((ii==i) & (jj==j)) continue;
74     ap = &ambsam(hp,ii,jj);
75     if (ap->d <= .5/FHUGE) continue;
76     VSUB(avec, ap->p, hp->rp->rop);
77     dprod = DOT(avec, dv);
78     if (dprod >= cos_thresh*VLEN(avec))
79     return(1); /* collision */
80     }
81     }
82     return(0);
83     }
84    
85    
86     static int
87 greg 2.61 ambsample( /* initial ambient division sample */
88     AMBHEMI *hp,
89     int i,
90     int j,
91     int n
92 greg 2.26 )
93     {
94 greg 2.61 AMBSAMP *ap = &ambsam(hp,i,j);
95     RAY ar;
96 greg 2.41 int hlist[3], ii;
97     double spt[2], zd;
98 greg 2.61 /* generate hemispherical sample */
99 greg 2.26 /* ambient coefficient for weight */
100     if (ambacc > FTINY)
101 greg 2.61 setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
102 greg 2.26 else
103 greg 2.61 copycolor(ar.rcoef, hp->acoef);
104 greg 2.62 if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0)
105 greg 2.41 return(0);
106 greg 2.26 if (ambacc > FTINY) {
107 greg 2.61 multcolor(ar.rcoef, hp->acoef);
108     scalecolor(ar.rcoef, 1./AVGREFL);
109 greg 2.41 }
110     hlist[0] = hp->rp->rno;
111 greg 2.46 hlist[1] = j;
112     hlist[2] = i;
113 greg 2.41 multisamp(spt, 2, urand(ilhash(hlist,3)+n));
114 greg 2.73 resample:
115 greg 2.46 SDsquare2disk(spt, (j+spt[1])/hp->ns, (i+spt[0])/hp->ns);
116 greg 2.26 zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]);
117     for (ii = 3; ii--; )
118 greg 2.61 ar.rdir[ii] = spt[0]*hp->ux[ii] +
119 greg 2.26 spt[1]*hp->uy[ii] +
120     zd*hp->rp->ron[ii];
121 greg 2.61 checknorm(ar.rdir);
122 greg 2.73 /* avoid coincident samples */
123     if (!n && ambcollision(hp, i, j, ar.rdir)) {
124     spt[0] = frandom(); spt[1] = frandom();
125     goto resample;
126     }
127 greg 2.56 dimlist[ndims++] = AI(hp,i,j) + 90171;
128 greg 2.61 rayvalue(&ar); /* evaluate ray */
129     ndims--;
130     if (ar.rt <= FTINY)
131     return(0); /* should never happen */
132     multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
133 greg 2.62 if (ar.rt*ap->d < 1.0) /* new/closer distance? */
134 greg 2.61 ap->d = 1.0/ar.rt;
135     if (!n) { /* record first vertex & value */
136     if (ar.rt > 10.0*thescene.cusize)
137     ar.rt = 10.0*thescene.cusize;
138     VSUM(ap->p, ar.rorg, ar.rdir, ar.rt);
139     copycolor(ap->v, ar.rcol);
140     } else { /* else update recorded value */
141     hp->acol[RED] -= colval(ap->v,RED);
142     hp->acol[GRN] -= colval(ap->v,GRN);
143     hp->acol[BLU] -= colval(ap->v,BLU);
144     zd = 1.0/(double)(n+1);
145     scalecolor(ar.rcol, zd);
146     zd *= (double)n;
147     scalecolor(ap->v, zd);
148     addcolor(ap->v, ar.rcol);
149     }
150     addcolor(hp->acol, ap->v); /* add to our sum */
151 greg 2.41 return(1);
152     }
153    
154    
155     /* Estimate errors based on ambient division differences */
156     static float *
157     getambdiffs(AMBHEMI *hp)
158     {
159 greg 2.55 float *earr = (float *)calloc(hp->ns*hp->ns, sizeof(float));
160 greg 2.41 float *ep;
161 greg 2.42 AMBSAMP *ap;
162 greg 2.41 double b, d2;
163     int i, j;
164    
165     if (earr == NULL) /* out of memory? */
166     return(NULL);
167     /* compute squared neighbor diffs */
168 greg 2.42 for (ap = hp->sa, ep = earr, i = 0; i < hp->ns; i++)
169     for (j = 0; j < hp->ns; j++, ap++, ep++) {
170     b = bright(ap[0].v);
171 greg 2.41 if (i) { /* from above */
172 greg 2.42 d2 = b - bright(ap[-hp->ns].v);
173 greg 2.41 d2 *= d2;
174     ep[0] += d2;
175     ep[-hp->ns] += d2;
176     }
177 greg 2.55 if (!j) continue;
178     /* from behind */
179     d2 = b - bright(ap[-1].v);
180     d2 *= d2;
181     ep[0] += d2;
182     ep[-1] += d2;
183     if (!i) continue;
184     /* diagonal */
185     d2 = b - bright(ap[-hp->ns-1].v);
186     d2 *= d2;
187     ep[0] += d2;
188     ep[-hp->ns-1] += d2;
189 greg 2.41 }
190     /* correct for number of neighbors */
191 greg 2.55 earr[0] *= 8./3.;
192     earr[hp->ns-1] *= 8./3.;
193     earr[(hp->ns-1)*hp->ns] *= 8./3.;
194     earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 8./3.;
195 greg 2.41 for (i = 1; i < hp->ns-1; i++) {
196 greg 2.55 earr[i*hp->ns] *= 8./5.;
197     earr[i*hp->ns + hp->ns-1] *= 8./5.;
198 greg 2.41 }
199     for (j = 1; j < hp->ns-1; j++) {
200 greg 2.55 earr[j] *= 8./5.;
201     earr[(hp->ns-1)*hp->ns + j] *= 8./5.;
202 greg 2.41 }
203     return(earr);
204     }
205    
206    
207 greg 2.43 /* Perform super-sampling on hemisphere (introduces bias) */
208 greg 2.41 static void
209 greg 2.61 ambsupersamp(AMBHEMI *hp, int cnt)
210 greg 2.41 {
211     float *earr = getambdiffs(hp);
212 greg 2.54 double e2rem = 0;
213 greg 2.41 AMBSAMP *ap;
214     float *ep;
215 greg 2.55 int i, j, n, nss;
216 greg 2.41
217     if (earr == NULL) /* just skip calc. if no memory */
218     return;
219 greg 2.54 /* accumulate estimated variances */
220 greg 2.55 for (ep = earr + hp->ns*hp->ns; ep > earr; )
221     e2rem += *--ep;
222 greg 2.41 ep = earr; /* perform super-sampling */
223     for (ap = hp->sa, i = 0; i < hp->ns; i++)
224     for (j = 0; j < hp->ns; j++, ap++) {
225 greg 2.55 if (e2rem <= FTINY)
226     goto done; /* nothing left to do */
227     nss = *ep/e2rem*cnt + frandom();
228 greg 2.62 for (n = 1; n <= nss && ambsample(hp,i,j,n); n++)
229     --cnt;
230 greg 2.61 e2rem -= *ep++; /* update remainder */
231 greg 2.41 }
232 greg 2.55 done:
233 greg 2.41 free(earr);
234     }
235    
236    
237 greg 2.61 static AMBHEMI *
238     samp_hemi( /* sample indirect hemisphere */
239     COLOR rcol,
240     RAY *r,
241     double wt
242     )
243     {
244     AMBHEMI *hp;
245     double d;
246     int n, i, j;
247     /* set number of divisions */
248     if (ambacc <= FTINY &&
249     wt > (d = 0.8*intens(rcol)*r->rweight/(ambdiv*minweight)))
250     wt = d; /* avoid ray termination */
251     n = sqrt(ambdiv * wt) + 0.5;
252 greg 2.68 i = 1 + 5*(ambacc > FTINY); /* minimum number of samples */
253 greg 2.61 if (n < i)
254     n = i;
255     /* allocate sampling array */
256     hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + sizeof(AMBSAMP)*(n*n - 1));
257     if (hp == NULL)
258     error(SYSTEM, "out of memory in samp_hemi");
259     hp->rp = r;
260     hp->ns = n;
261     hp->acol[RED] = hp->acol[GRN] = hp->acol[BLU] = 0.0;
262 greg 2.62 memset(hp->sa, 0, sizeof(AMBSAMP)*n*n);
263 greg 2.61 hp->sampOK = 0;
264     /* assign coefficient */
265     copycolor(hp->acoef, rcol);
266     d = 1.0/(n*n);
267     scalecolor(hp->acoef, d);
268     /* make tangent plane axes */
269 greg 2.70 if (!getperpendicular(hp->ux, r->ron, 1))
270 greg 2.61 error(CONSISTENCY, "bad ray direction in samp_hemi");
271     VCROSS(hp->uy, r->ron, hp->ux);
272     /* sample divisions */
273     for (i = hp->ns; i--; )
274     for (j = hp->ns; j--; )
275     hp->sampOK += ambsample(hp, i, j, 0);
276     copycolor(rcol, hp->acol);
277     if (!hp->sampOK) { /* utter failure? */
278     free(hp);
279     return(NULL);
280     }
281     if (hp->sampOK < hp->ns*hp->ns) {
282     hp->sampOK *= -1; /* soft failure */
283     return(hp);
284     }
285     n = ambssamp*wt + 0.5;
286     if (n > 8) { /* perform super-sampling? */
287     ambsupersamp(hp, n);
288     copycolor(rcol, hp->acol);
289     }
290     return(hp); /* all is well */
291     }
292    
293    
294 greg 2.46 /* Return brightness of farthest ambient sample */
295     static double
296 greg 2.56 back_ambval(AMBHEMI *hp, const int n1, const int n2, const int n3)
297 greg 2.46 {
298 greg 2.56 if (hp->sa[n1].d <= hp->sa[n2].d) {
299     if (hp->sa[n1].d <= hp->sa[n3].d)
300     return(colval(hp->sa[n1].v,CIEY));
301     return(colval(hp->sa[n3].v,CIEY));
302     }
303     if (hp->sa[n2].d <= hp->sa[n3].d)
304     return(colval(hp->sa[n2].v,CIEY));
305     return(colval(hp->sa[n3].v,CIEY));
306 greg 2.46 }
307    
308    
309 greg 2.27 /* Compute vectors and coefficients for Hessian/gradient calcs */
310     static void
311 greg 2.56 comp_fftri(FFTRI *ftp, AMBHEMI *hp, const int n0, const int n1)
312 greg 2.27 {
313 greg 2.56 double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2;
314     int ii;
315    
316     VSUB(ftp->r_i, hp->sa[n0].p, hp->rp->rop);
317     VSUB(ftp->r_i1, hp->sa[n1].p, hp->rp->rop);
318     VSUB(ftp->e_i, hp->sa[n1].p, hp->sa[n0].p);
319 greg 2.35 VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1);
320     rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp);
321 greg 2.27 dot_e = DOT(ftp->e_i,ftp->e_i);
322     dot_er = DOT(ftp->e_i, ftp->r_i);
323 greg 2.32 rdot_r = 1.0/DOT(ftp->r_i,ftp->r_i);
324     rdot_r1 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
325     ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) * sqrt(rdot_r*rdot_r1) ) *
326 greg 2.35 sqrt( rdot_cp );
327 greg 2.32 ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r +
328 greg 2.35 dot_e*ftp->I1 )*0.5*rdot_cp;
329 greg 2.32 J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e;
330 greg 2.46 for (ii = 3; ii--; )
331     ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii];
332 greg 2.27 }
333    
334    
335 greg 2.28 /* Compose 3x3 matrix from two vectors */
336 greg 2.27 static void
337     compose_matrix(FVECT mat[3], FVECT va, FVECT vb)
338     {
339     mat[0][0] = 2.0*va[0]*vb[0];
340     mat[1][1] = 2.0*va[1]*vb[1];
341     mat[2][2] = 2.0*va[2]*vb[2];
342     mat[0][1] = mat[1][0] = va[0]*vb[1] + va[1]*vb[0];
343     mat[0][2] = mat[2][0] = va[0]*vb[2] + va[2]*vb[0];
344     mat[1][2] = mat[2][1] = va[1]*vb[2] + va[2]*vb[1];
345     }
346    
347    
348     /* Compute partial 3x3 Hessian matrix for edge */
349     static void
350     comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm)
351     {
352 greg 2.35 FVECT ncp;
353 greg 2.27 FVECT m1[3], m2[3], m3[3], m4[3];
354     double d1, d2, d3, d4;
355     double I3, J3, K3;
356     int i, j;
357     /* compute intermediate coefficients */
358     d1 = 1.0/DOT(ftp->r_i,ftp->r_i);
359     d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
360     d3 = 1.0/DOT(ftp->e_i,ftp->e_i);
361     d4 = DOT(ftp->e_i, ftp->r_i);
362 greg 2.35 I3 = ( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + 3.0/d3*ftp->I2 )
363     / ( 4.0*DOT(ftp->rcp,ftp->rcp) );
364 greg 2.27 J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3;
365     K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3);
366     /* intermediate matrices */
367 greg 2.35 VCROSS(ncp, nrm, ftp->e_i);
368     compose_matrix(m1, ncp, ftp->rI2_eJ2);
369 greg 2.27 compose_matrix(m2, ftp->r_i, ftp->r_i);
370     compose_matrix(m3, ftp->e_i, ftp->e_i);
371     compose_matrix(m4, ftp->r_i, ftp->e_i);
372 greg 2.35 d1 = DOT(nrm, ftp->rcp);
373 greg 2.27 d2 = -d1*ftp->I2;
374     d1 *= 2.0;
375     for (i = 3; i--; ) /* final matrix sum */
376     for (j = 3; j--; ) {
377     hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] +
378     2.0*J3*m4[i][j] );
379     hess[i][j] += d2*(i==j);
380 greg 2.46 hess[i][j] *= -1.0/PI;
381 greg 2.27 }
382     }
383    
384    
385     /* Reverse hessian calculation result for edge in other direction */
386     static void
387     rev_hessian(FVECT hess[3])
388     {
389     int i;
390    
391     for (i = 3; i--; ) {
392     hess[i][0] = -hess[i][0];
393     hess[i][1] = -hess[i][1];
394     hess[i][2] = -hess[i][2];
395     }
396     }
397    
398    
399     /* Add to radiometric Hessian from the given triangle */
400     static void
401     add2hessian(FVECT hess[3], FVECT ehess1[3],
402 greg 2.46 FVECT ehess2[3], FVECT ehess3[3], double v)
403 greg 2.27 {
404     int i, j;
405    
406     for (i = 3; i--; )
407     for (j = 3; j--; )
408     hess[i][j] += v*( ehess1[i][j] + ehess2[i][j] + ehess3[i][j] );
409     }
410    
411    
412     /* Compute partial displacement form factor gradient for edge */
413     static void
414     comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm)
415     {
416 greg 2.35 FVECT ncp;
417 greg 2.27 double f1;
418     int i;
419    
420 greg 2.35 f1 = 2.0*DOT(nrm, ftp->rcp);
421     VCROSS(ncp, nrm, ftp->e_i);
422 greg 2.27 for (i = 3; i--; )
423 greg 2.46 grad[i] = (0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] );
424 greg 2.27 }
425    
426    
427     /* Reverse gradient calculation result for edge in other direction */
428     static void
429     rev_gradient(FVECT grad)
430     {
431     grad[0] = -grad[0];
432     grad[1] = -grad[1];
433     grad[2] = -grad[2];
434     }
435    
436    
437     /* Add to displacement gradient from the given triangle */
438     static void
439 greg 2.46 add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, double v)
440 greg 2.27 {
441     int i;
442    
443     for (i = 3; i--; )
444     grad[i] += v*( egrad1[i] + egrad2[i] + egrad3[i] );
445     }
446    
447    
448     /* Compute anisotropic radii and eigenvector directions */
449 greg 2.53 static void
450 greg 2.27 eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3])
451     {
452     double hess2[2][2];
453     FVECT a, b;
454     double evalue[2], slope1, xmag1;
455     int i;
456     /* project Hessian to sample plane */
457     for (i = 3; i--; ) {
458     a[i] = DOT(hessian[i], uv[0]);
459     b[i] = DOT(hessian[i], uv[1]);
460     }
461     hess2[0][0] = DOT(uv[0], a);
462     hess2[0][1] = DOT(uv[0], b);
463     hess2[1][0] = DOT(uv[1], a);
464     hess2[1][1] = DOT(uv[1], b);
465 greg 2.38 /* compute eigenvalue(s) */
466     i = quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1],
467     hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]);
468     if (i == 1) /* double-root (circle) */
469     evalue[1] = evalue[0];
470     if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) |
471 greg 2.53 ((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) {
472     ra[0] = ra[1] = maxarad;
473     return;
474     }
475 greg 2.27 if (evalue[0] > evalue[1]) {
476 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[0]));
477     ra[1] = sqrt(sqrt(4.0/evalue[1]));
478 greg 2.27 slope1 = evalue[1];
479     } else {
480 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[1]));
481     ra[1] = sqrt(sqrt(4.0/evalue[0]));
482 greg 2.27 slope1 = evalue[0];
483     }
484     /* compute unit eigenvectors */
485     if (fabs(hess2[0][1]) <= FTINY)
486     return; /* uv OK as is */
487     slope1 = (slope1 - hess2[0][0]) / hess2[0][1];
488     xmag1 = sqrt(1.0/(1.0 + slope1*slope1));
489     for (i = 3; i--; ) {
490     b[i] = xmag1*uv[0][i] + slope1*xmag1*uv[1][i];
491     a[i] = slope1*xmag1*uv[0][i] - xmag1*uv[1][i];
492     }
493     VCOPY(uv[0], a);
494     VCOPY(uv[1], b);
495     }
496    
497    
498 greg 2.26 static void
499     ambHessian( /* anisotropic radii & pos. gradient */
500     AMBHEMI *hp,
501     FVECT uv[2], /* returned */
502 greg 2.28 float ra[2], /* returned (optional) */
503     float pg[2] /* returned (optional) */
504 greg 2.26 )
505     {
506 greg 2.27 static char memerrmsg[] = "out of memory in ambHessian()";
507     FVECT (*hessrow)[3] = NULL;
508     FVECT *gradrow = NULL;
509     FVECT hessian[3];
510     FVECT gradient;
511     FFTRI fftr;
512     int i, j;
513     /* be sure to assign unit vectors */
514     VCOPY(uv[0], hp->ux);
515     VCOPY(uv[1], hp->uy);
516     /* clock-wise vertex traversal from sample POV */
517     if (ra != NULL) { /* initialize Hessian row buffer */
518 greg 2.28 hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1));
519 greg 2.27 if (hessrow == NULL)
520     error(SYSTEM, memerrmsg);
521     memset(hessian, 0, sizeof(hessian));
522     } else if (pg == NULL) /* bogus call? */
523     return;
524     if (pg != NULL) { /* initialize form factor row buffer */
525 greg 2.28 gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1));
526 greg 2.27 if (gradrow == NULL)
527     error(SYSTEM, memerrmsg);
528     memset(gradient, 0, sizeof(gradient));
529     }
530     /* compute first row of edges */
531     for (j = 0; j < hp->ns-1; j++) {
532 greg 2.56 comp_fftri(&fftr, hp, AI(hp,0,j), AI(hp,0,j+1));
533 greg 2.27 if (hessrow != NULL)
534     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
535     if (gradrow != NULL)
536     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
537     }
538     /* sum each row of triangles */
539     for (i = 0; i < hp->ns-1; i++) {
540     FVECT hesscol[3]; /* compute first vertical edge */
541     FVECT gradcol;
542 greg 2.56 comp_fftri(&fftr, hp, AI(hp,i,0), AI(hp,i+1,0));
543 greg 2.27 if (hessrow != NULL)
544     comp_hessian(hesscol, &fftr, hp->rp->ron);
545     if (gradrow != NULL)
546     comp_gradient(gradcol, &fftr, hp->rp->ron);
547     for (j = 0; j < hp->ns-1; j++) {
548     FVECT hessdia[3]; /* compute triangle contributions */
549     FVECT graddia;
550 greg 2.46 double backg;
551 greg 2.56 backg = back_ambval(hp, AI(hp,i,j),
552     AI(hp,i,j+1), AI(hp,i+1,j));
553 greg 2.27 /* diagonal (inner) edge */
554 greg 2.56 comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j));
555 greg 2.27 if (hessrow != NULL) {
556     comp_hessian(hessdia, &fftr, hp->rp->ron);
557     rev_hessian(hesscol);
558     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
559     }
560 greg 2.39 if (gradrow != NULL) {
561 greg 2.27 comp_gradient(graddia, &fftr, hp->rp->ron);
562     rev_gradient(gradcol);
563     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
564     }
565     /* initialize edge in next row */
566 greg 2.56 comp_fftri(&fftr, hp, AI(hp,i+1,j+1), AI(hp,i+1,j));
567 greg 2.27 if (hessrow != NULL)
568     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
569     if (gradrow != NULL)
570     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
571     /* new column edge & paired triangle */
572 greg 2.56 backg = back_ambval(hp, AI(hp,i+1,j+1),
573     AI(hp,i+1,j), AI(hp,i,j+1));
574     comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j+1));
575 greg 2.27 if (hessrow != NULL) {
576     comp_hessian(hesscol, &fftr, hp->rp->ron);
577     rev_hessian(hessdia);
578     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
579     if (i < hp->ns-2)
580     rev_hessian(hessrow[j]);
581     }
582     if (gradrow != NULL) {
583     comp_gradient(gradcol, &fftr, hp->rp->ron);
584     rev_gradient(graddia);
585     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
586     if (i < hp->ns-2)
587     rev_gradient(gradrow[j]);
588     }
589     }
590     }
591     /* release row buffers */
592     if (hessrow != NULL) free(hessrow);
593     if (gradrow != NULL) free(gradrow);
594    
595     if (ra != NULL) /* extract eigenvectors & radii */
596     eigenvectors(uv, ra, hessian);
597 greg 2.32 if (pg != NULL) { /* tangential position gradient */
598     pg[0] = DOT(gradient, uv[0]);
599     pg[1] = DOT(gradient, uv[1]);
600 greg 2.27 }
601     }
602    
603    
604     /* Compute direction gradient from a hemispherical sampling */
605     static void
606     ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2])
607     {
608 greg 2.41 AMBSAMP *ap;
609     double dgsum[2];
610     int n;
611     FVECT vd;
612     double gfact;
613 greg 2.27
614 greg 2.29 dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */
615 greg 2.27 for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) {
616     /* use vector for azimuth + 90deg */
617     VSUB(vd, ap->p, hp->rp->rop);
618 greg 2.29 /* brightness over cosine factor */
619     gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd);
620 greg 2.40 /* sine = proj_radius/vd_length */
621     dgsum[0] -= DOT(uv[1], vd) * gfact;
622     dgsum[1] += DOT(uv[0], vd) * gfact;
623 greg 2.26 }
624 greg 2.29 dg[0] = dgsum[0] / (hp->ns*hp->ns);
625     dg[1] = dgsum[1] / (hp->ns*hp->ns);
626 greg 2.26 }
627    
628 greg 2.27
629 greg 2.49 /* Compute potential light leak direction flags for cache value */
630     static uint32
631     ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1)
632 greg 2.47 {
633 greg 2.50 const double max_d = 1.0/(minarad*ambacc + 0.001);
634 greg 2.66 const double ang_res = 0.5*PI/hp->ns;
635     const double ang_step = ang_res/((int)(16/PI*ang_res) + 1.01);
636 greg 2.51 double avg_d = 0;
637 greg 2.50 uint32 flgs = 0;
638 greg 2.58 FVECT vec;
639 greg 2.62 double u, v;
640 greg 2.58 double ang, a1;
641 greg 2.50 int i, j;
642 greg 2.52 /* don't bother for a few samples */
643 greg 2.72 if (hp->ns < 8)
644 greg 2.52 return(0);
645     /* check distances overhead */
646 greg 2.51 for (i = hp->ns*3/4; i-- > hp->ns>>2; )
647     for (j = hp->ns*3/4; j-- > hp->ns>>2; )
648     avg_d += ambsam(hp,i,j).d;
649     avg_d *= 4.0/(hp->ns*hp->ns);
650 greg 2.52 if (avg_d*r0 >= 1.0) /* ceiling too low for corral? */
651     return(0);
652     if (avg_d >= max_d) /* insurance */
653 greg 2.51 return(0);
654     /* else circle around perimeter */
655 greg 2.47 for (i = 0; i < hp->ns; i++)
656     for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) {
657     AMBSAMP *ap = &ambsam(hp,i,j);
658 greg 2.50 if ((ap->d <= FTINY) | (ap->d >= max_d))
659     continue; /* too far or too near */
660 greg 2.47 VSUB(vec, ap->p, hp->rp->rop);
661 greg 2.62 u = DOT(vec, uv[0]);
662     v = DOT(vec, uv[1]);
663     if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= u*u + v*v)
664 greg 2.49 continue; /* occluder outside ellipse */
665     ang = atan2a(v, u); /* else set direction flags */
666 greg 2.66 for (a1 = ang-ang_res; a1 <= ang+ang_res; a1 += ang_step)
667 greg 2.50 flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0)));
668 greg 2.47 }
669 greg 2.58 /* add low-angle incident (< 20deg) */
670     if (fabs(hp->rp->rod) <= 0.342) {
671     u = -DOT(hp->rp->rdir, uv[0]);
672     v = -DOT(hp->rp->rdir, uv[1]);
673     if ((r0*r0*u*u + r1*r1*v*v) > hp->rp->rot*hp->rp->rot) {
674     ang = atan2a(v, u);
675     ang += 2.*PI*(ang < 0);
676     ang *= 16/PI;
677     if ((ang < .5) | (ang >= 31.5))
678     flgs |= 0x80000001;
679     else
680     flgs |= 3L<<(int)(ang-.5);
681     }
682     }
683 greg 2.49 return(flgs);
684 greg 2.47 }
685    
686    
687 greg 2.26 int
688     doambient( /* compute ambient component */
689     COLOR rcol, /* input/output color */
690     RAY *r,
691     double wt,
692 greg 2.27 FVECT uv[2], /* returned (optional) */
693     float ra[2], /* returned (optional) */
694     float pg[2], /* returned (optional) */
695 greg 2.49 float dg[2], /* returned (optional) */
696     uint32 *crlp /* returned (optional) */
697 greg 2.26 )
698     {
699 greg 2.61 AMBHEMI *hp = samp_hemi(rcol, r, wt);
700 greg 2.41 FVECT my_uv[2];
701 greg 2.61 double d, K;
702 greg 2.41 AMBSAMP *ap;
703 greg 2.61 int i;
704     /* clear return values */
705 greg 2.26 if (uv != NULL)
706     memset(uv, 0, sizeof(FVECT)*2);
707     if (ra != NULL)
708     ra[0] = ra[1] = 0.0;
709     if (pg != NULL)
710     pg[0] = pg[1] = 0.0;
711     if (dg != NULL)
712     dg[0] = dg[1] = 0.0;
713 greg 2.49 if (crlp != NULL)
714     *crlp = 0;
715 greg 2.61 if (hp == NULL) /* sampling falure? */
716     return(0);
717    
718     if ((ra == NULL) & (pg == NULL) & (dg == NULL) ||
719 greg 2.68 (hp->sampOK < 0) | (hp->ns < 6)) {
720 greg 2.61 free(hp); /* Hessian not requested/possible */
721     return(-1); /* value-only return value */
722 greg 2.26 }
723 greg 2.61 if ((d = bright(rcol)) > FTINY) { /* normalize Y values */
724 greg 2.45 d = 0.99*(hp->ns*hp->ns)/d;
725 greg 2.38 K = 0.01;
726 greg 2.45 } else { /* or fall back on geometric Hessian */
727 greg 2.38 K = 1.0;
728     pg = NULL;
729     dg = NULL;
730 greg 2.53 crlp = NULL;
731 greg 2.38 }
732 greg 2.29 ap = hp->sa; /* relative Y channel from here on... */
733 greg 2.26 for (i = hp->ns*hp->ns; i--; ap++)
734 greg 2.38 colval(ap->v,CIEY) = bright(ap->v)*d + K;
735 greg 2.26
736     if (uv == NULL) /* make sure we have axis pointers */
737     uv = my_uv;
738     /* compute radii & pos. gradient */
739     ambHessian(hp, uv, ra, pg);
740 greg 2.29
741 greg 2.26 if (dg != NULL) /* compute direction gradient */
742     ambdirgrad(hp, uv, dg);
743 greg 2.29
744 greg 2.28 if (ra != NULL) { /* scale/clamp radii */
745 greg 2.35 if (pg != NULL) {
746     if (ra[0]*(d = fabs(pg[0])) > 1.0)
747     ra[0] = 1.0/d;
748     if (ra[1]*(d = fabs(pg[1])) > 1.0)
749     ra[1] = 1.0/d;
750 greg 2.48 if (ra[0] > ra[1])
751     ra[0] = ra[1];
752 greg 2.35 }
753 greg 2.29 if (ra[0] < minarad) {
754     ra[0] = minarad;
755     if (ra[1] < minarad)
756     ra[1] = minarad;
757     }
758 greg 2.60 ra[0] *= d = 1.0/sqrt(wt);
759 greg 2.26 if ((ra[1] *= d) > 2.0*ra[0])
760     ra[1] = 2.0*ra[0];
761 greg 2.28 if (ra[1] > maxarad) {
762     ra[1] = maxarad;
763     if (ra[0] > maxarad)
764     ra[0] = maxarad;
765     }
766 greg 2.53 /* flag encroached directions */
767 greg 2.71 if (crlp != NULL)
768 greg 2.49 *crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc);
769 greg 2.35 if (pg != NULL) { /* cap gradient if necessary */
770     d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1];
771     if (d > 1.0) {
772     d = 1.0/sqrt(d);
773     pg[0] *= d;
774     pg[1] *= d;
775     }
776     }
777 greg 2.26 }
778     free(hp); /* clean up and return */
779     return(1);
780     }
781    
782    
783 greg 2.25 #else /* ! NEWAMB */
784 greg 1.1
785    
786 greg 2.15 void
787 greg 2.14 inithemi( /* initialize sampling hemisphere */
788 greg 2.23 AMBHEMI *hp,
789 greg 2.16 COLOR ac,
790 greg 2.14 RAY *r,
791     double wt
792     )
793 greg 1.1 {
794 greg 2.16 double d;
795 greg 2.23 int i;
796 greg 2.14 /* set number of divisions */
797 greg 2.16 if (ambacc <= FTINY &&
798 greg 2.20 wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight)))
799 greg 2.16 wt = d; /* avoid ray termination */
800     hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
801 greg 2.14 i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */
802     if (hp->nt < i)
803     hp->nt = i;
804     hp->np = PI * hp->nt + 0.5;
805     /* set number of super-samples */
806 greg 2.15 hp->ns = ambssamp * wt + 0.5;
807 greg 2.16 /* assign coefficient */
808 greg 2.14 copycolor(hp->acoef, ac);
809 greg 2.16 d = 1.0/(hp->nt*hp->np);
810     scalecolor(hp->acoef, d);
811 greg 2.14 /* make axes */
812     VCOPY(hp->uz, r->ron);
813     hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
814     for (i = 0; i < 3; i++)
815     if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
816     break;
817     if (i >= 3)
818     error(CONSISTENCY, "bad ray direction in inithemi");
819     hp->uy[i] = 1.0;
820     fcross(hp->ux, hp->uy, hp->uz);
821     normalize(hp->ux);
822     fcross(hp->uy, hp->uz, hp->ux);
823 greg 1.1 }
824    
825    
826 greg 2.9 int
827 greg 2.14 divsample( /* sample a division */
828 greg 2.23 AMBSAMP *dp,
829 greg 2.14 AMBHEMI *h,
830     RAY *r
831     )
832 greg 1.1 {
833     RAY ar;
834 greg 1.11 int hlist[3];
835     double spt[2];
836 greg 1.1 double xd, yd, zd;
837     double b2;
838     double phi;
839 greg 2.23 int i;
840 greg 2.15 /* ambient coefficient for weight */
841 greg 2.16 if (ambacc > FTINY)
842     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
843     else
844     copycolor(ar.rcoef, h->acoef);
845 greg 2.14 if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
846 greg 1.4 return(-1);
847 greg 2.17 if (ambacc > FTINY) {
848     multcolor(ar.rcoef, h->acoef);
849     scalecolor(ar.rcoef, 1./AVGREFL);
850     }
851 greg 1.1 hlist[0] = r->rno;
852     hlist[1] = dp->t;
853     hlist[2] = dp->p;
854 greg 1.13 multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n));
855 greg 1.11 zd = sqrt((dp->t + spt[0])/h->nt);
856     phi = 2.0*PI * (dp->p + spt[1])/h->np;
857 gwlarson 2.8 xd = tcos(phi) * zd;
858     yd = tsin(phi) * zd;
859 greg 1.1 zd = sqrt(1.0 - zd*zd);
860 greg 1.2 for (i = 0; i < 3; i++)
861     ar.rdir[i] = xd*h->ux[i] +
862     yd*h->uy[i] +
863     zd*h->uz[i];
864 greg 2.22 checknorm(ar.rdir);
865 greg 1.2 dimlist[ndims++] = dp->t*h->np + dp->p + 90171;
866 greg 1.1 rayvalue(&ar);
867     ndims--;
868 greg 2.16 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
869 greg 1.1 addcolor(dp->v, ar.rcol);
870 greg 2.9 /* use rt to improve gradient calc */
871     if (ar.rt > FTINY && ar.rt < FHUGE)
872     dp->r += 1.0/ar.rt;
873 greg 1.1 /* (re)initialize error */
874     if (dp->n++) {
875     b2 = bright(dp->v)/dp->n - bright(ar.rcol);
876     b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1));
877     dp->k = b2/(dp->n*dp->n);
878     } else
879     dp->k = 0.0;
880 greg 1.4 return(0);
881 greg 1.1 }
882    
883    
884 greg 2.14 static int
885     ambcmp( /* decreasing order */
886     const void *p1,
887     const void *p2
888     )
889     {
890     const AMBSAMP *d1 = (const AMBSAMP *)p1;
891     const AMBSAMP *d2 = (const AMBSAMP *)p2;
892    
893     if (d1->k < d2->k)
894     return(1);
895     if (d1->k > d2->k)
896     return(-1);
897     return(0);
898     }
899    
900    
901     static int
902     ambnorm( /* standard order */
903     const void *p1,
904     const void *p2
905     )
906     {
907     const AMBSAMP *d1 = (const AMBSAMP *)p1;
908     const AMBSAMP *d2 = (const AMBSAMP *)p2;
909 greg 2.23 int c;
910 greg 2.14
911     if ( (c = d1->t - d2->t) )
912     return(c);
913     return(d1->p - d2->p);
914     }
915    
916    
917 greg 1.1 double
918 greg 2.14 doambient( /* compute ambient component */
919 greg 2.23 COLOR rcol,
920 greg 2.14 RAY *r,
921     double wt,
922     FVECT pg,
923     FVECT dg
924     )
925 greg 1.1 {
926 greg 2.24 double b, d=0;
927 greg 1.1 AMBHEMI hemi;
928     AMBSAMP *div;
929     AMBSAMP dnew;
930 greg 2.23 double acol[3];
931     AMBSAMP *dp;
932 greg 1.1 double arad;
933 greg 2.19 int divcnt;
934 greg 2.23 int i, j;
935 greg 1.1 /* initialize hemisphere */
936 greg 2.23 inithemi(&hemi, rcol, r, wt);
937 greg 2.19 divcnt = hemi.nt * hemi.np;
938 greg 2.17 /* initialize */
939     if (pg != NULL)
940     pg[0] = pg[1] = pg[2] = 0.0;
941     if (dg != NULL)
942     dg[0] = dg[1] = dg[2] = 0.0;
943 greg 2.23 setcolor(rcol, 0.0, 0.0, 0.0);
944 greg 2.19 if (divcnt == 0)
945 greg 1.1 return(0.0);
946 greg 2.14 /* allocate super-samples */
947 greg 2.15 if (hemi.ns > 0 || pg != NULL || dg != NULL) {
948 greg 2.19 div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP));
949 greg 1.1 if (div == NULL)
950     error(SYSTEM, "out of memory in doambient");
951     } else
952     div = NULL;
953     /* sample the divisions */
954     arad = 0.0;
955 greg 2.23 acol[0] = acol[1] = acol[2] = 0.0;
956 greg 1.1 if ((dp = div) == NULL)
957     dp = &dnew;
958 greg 2.19 divcnt = 0;
959 greg 1.1 for (i = 0; i < hemi.nt; i++)
960     for (j = 0; j < hemi.np; j++) {
961     dp->t = i; dp->p = j;
962     setcolor(dp->v, 0.0, 0.0, 0.0);
963 greg 1.2 dp->r = 0.0;
964 greg 1.1 dp->n = 0;
965 greg 2.16 if (divsample(dp, &hemi, r) < 0) {
966 greg 2.19 if (div != NULL)
967     dp++;
968 greg 2.16 continue;
969     }
970 greg 2.6 arad += dp->r;
971 greg 2.19 divcnt++;
972 greg 1.1 if (div != NULL)
973     dp++;
974 greg 2.6 else
975 greg 1.1 addcolor(acol, dp->v);
976     }
977 greg 2.21 if (!divcnt) {
978     if (div != NULL)
979     free((void *)div);
980 greg 2.19 return(0.0); /* no samples taken */
981 greg 2.21 }
982 greg 2.19 if (divcnt < hemi.nt*hemi.np) {
983     pg = dg = NULL; /* incomplete sampling */
984     hemi.ns = 0;
985     } else if (arad > FTINY && divcnt/arad < minarad) {
986 greg 2.15 hemi.ns = 0; /* close enough */
987 greg 2.19 } else if (hemi.ns > 0) { /* else perform super-sampling? */
988 greg 1.4 comperrs(div, &hemi); /* compute errors */
989 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */
990 greg 1.1 /* super-sample */
991 greg 2.15 for (i = hemi.ns; i > 0; i--) {
992 schorsch 2.11 dnew = *div;
993 greg 2.16 if (divsample(&dnew, &hemi, r) < 0) {
994     dp++;
995     continue;
996     }
997     dp = div; /* reinsert */
998 greg 2.19 j = divcnt < i ? divcnt : i;
999 greg 1.1 while (--j > 0 && dnew.k < dp[1].k) {
1000 schorsch 2.11 *dp = *(dp+1);
1001 greg 1.1 dp++;
1002     }
1003 schorsch 2.11 *dp = dnew;
1004 greg 1.1 }
1005 greg 1.2 if (pg != NULL || dg != NULL) /* restore order */
1006 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambnorm);
1007 greg 1.1 }
1008     /* compute returned values */
1009 greg 1.3 if (div != NULL) {
1010 greg 2.19 arad = 0.0; /* note: divcnt may be < nt*np */
1011     for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) {
1012 greg 1.3 arad += dp->r;
1013     if (dp->n > 1) {
1014     b = 1.0/dp->n;
1015     scalecolor(dp->v, b);
1016     dp->r *= b;
1017     dp->n = 1;
1018     }
1019     addcolor(acol, dp->v);
1020     }
1021 greg 1.5 b = bright(acol);
1022 greg 1.6 if (b > FTINY) {
1023 greg 2.17 b = 1.0/b; /* compute & normalize gradient(s) */
1024 greg 1.6 if (pg != NULL) {
1025     posgradient(pg, div, &hemi);
1026     for (i = 0; i < 3; i++)
1027     pg[i] *= b;
1028     }
1029     if (dg != NULL) {
1030     dirgradient(dg, div, &hemi);
1031     for (i = 0; i < 3; i++)
1032     dg[i] *= b;
1033     }
1034 greg 1.5 }
1035 greg 2.9 free((void *)div);
1036 greg 1.3 }
1037 greg 2.23 copycolor(rcol, acol);
1038 greg 1.1 if (arad <= FTINY)
1039 greg 1.16 arad = maxarad;
1040 greg 2.3 else
1041 greg 2.19 arad = (divcnt+hemi.ns)/arad;
1042 greg 1.15 if (pg != NULL) { /* reduce radius if gradient large */
1043     d = DOT(pg,pg);
1044     if (d*arad*arad > 1.0)
1045     arad = 1.0/sqrt(d);
1046     }
1047 greg 1.16 if (arad < minarad) {
1048 greg 1.1 arad = minarad;
1049 greg 1.16 if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */
1050     d = 1.0/arad/sqrt(d);
1051     for (i = 0; i < 3; i++)
1052     pg[i] *= d;
1053     }
1054     }
1055 greg 2.3 if ((arad /= sqrt(wt)) > maxarad)
1056     arad = maxarad;
1057     return(arad);
1058 greg 1.1 }
1059    
1060    
1061 greg 2.9 void
1062 greg 2.14 comperrs( /* compute initial error estimates */
1063     AMBSAMP *da, /* assumes standard ordering */
1064 greg 2.23 AMBHEMI *hp
1065 greg 2.14 )
1066 greg 1.1 {
1067     double b, b2;
1068     int i, j;
1069 greg 2.23 AMBSAMP *dp;
1070 greg 1.1 /* sum differences from neighbors */
1071     dp = da;
1072     for (i = 0; i < hp->nt; i++)
1073     for (j = 0; j < hp->np; j++) {
1074 greg 1.6 #ifdef DEBUG
1075     if (dp->t != i || dp->p != j)
1076     error(CONSISTENCY,
1077     "division order in comperrs");
1078     #endif
1079 greg 1.1 b = bright(dp[0].v);
1080     if (i > 0) { /* from above */
1081     b2 = bright(dp[-hp->np].v) - b;
1082     b2 *= b2 * 0.25;
1083     dp[0].k += b2;
1084     dp[-hp->np].k += b2;
1085     }
1086     if (j > 0) { /* from behind */
1087     b2 = bright(dp[-1].v) - b;
1088     b2 *= b2 * 0.25;
1089     dp[0].k += b2;
1090     dp[-1].k += b2;
1091 greg 1.4 } else { /* around */
1092     b2 = bright(dp[hp->np-1].v) - b;
1093 greg 1.1 b2 *= b2 * 0.25;
1094     dp[0].k += b2;
1095 greg 1.4 dp[hp->np-1].k += b2;
1096 greg 1.1 }
1097     dp++;
1098     }
1099     /* divide by number of neighbors */
1100     dp = da;
1101     for (j = 0; j < hp->np; j++) /* top row */
1102     (dp++)->k *= 1.0/3.0;
1103     if (hp->nt < 2)
1104     return;
1105     for (i = 1; i < hp->nt-1; i++) /* central region */
1106     for (j = 0; j < hp->np; j++)
1107     (dp++)->k *= 0.25;
1108     for (j = 0; j < hp->np; j++) /* bottom row */
1109     (dp++)->k *= 1.0/3.0;
1110     }
1111    
1112    
1113 greg 2.9 void
1114 greg 2.14 posgradient( /* compute position gradient */
1115     FVECT gv,
1116     AMBSAMP *da, /* assumes standard ordering */
1117 greg 2.23 AMBHEMI *hp
1118 greg 2.14 )
1119 greg 1.1 {
1120 greg 2.23 int i, j;
1121 greg 2.2 double nextsine, lastsine, b, d;
1122 greg 1.2 double mag0, mag1;
1123     double phi, cosp, sinp, xd, yd;
1124 greg 2.23 AMBSAMP *dp;
1125 greg 1.2
1126     xd = yd = 0.0;
1127     for (j = 0; j < hp->np; j++) {
1128     dp = da + j;
1129     mag0 = mag1 = 0.0;
1130 greg 2.2 lastsine = 0.0;
1131 greg 1.2 for (i = 0; i < hp->nt; i++) {
1132     #ifdef DEBUG
1133     if (dp->t != i || dp->p != j)
1134     error(CONSISTENCY,
1135     "division order in posgradient");
1136     #endif
1137     b = bright(dp->v);
1138     if (i > 0) {
1139     d = dp[-hp->np].r;
1140     if (dp[0].r > d) d = dp[0].r;
1141 greg 2.2 /* sin(t)*cos(t)^2 */
1142     d *= lastsine * (1.0 - (double)i/hp->nt);
1143 greg 1.2 mag0 += d*(b - bright(dp[-hp->np].v));
1144     }
1145 greg 2.2 nextsine = sqrt((double)(i+1)/hp->nt);
1146 greg 1.2 if (j > 0) {
1147     d = dp[-1].r;
1148     if (dp[0].r > d) d = dp[0].r;
1149 greg 2.2 mag1 += d * (nextsine - lastsine) *
1150     (b - bright(dp[-1].v));
1151 greg 1.2 } else {
1152     d = dp[hp->np-1].r;
1153     if (dp[0].r > d) d = dp[0].r;
1154 greg 2.2 mag1 += d * (nextsine - lastsine) *
1155     (b - bright(dp[hp->np-1].v));
1156 greg 1.2 }
1157     dp += hp->np;
1158 greg 2.2 lastsine = nextsine;
1159 greg 1.2 }
1160 greg 2.2 mag0 *= 2.0*PI / hp->np;
1161 greg 1.2 phi = 2.0*PI * (double)j/hp->np;
1162 gwlarson 2.8 cosp = tcos(phi); sinp = tsin(phi);
1163 greg 1.2 xd += mag0*cosp - mag1*sinp;
1164     yd += mag0*sinp + mag1*cosp;
1165     }
1166     for (i = 0; i < 3; i++)
1167 greg 2.16 gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI;
1168 greg 1.1 }
1169    
1170    
1171 greg 2.9 void
1172 greg 2.14 dirgradient( /* compute direction gradient */
1173     FVECT gv,
1174     AMBSAMP *da, /* assumes standard ordering */
1175 greg 2.23 AMBHEMI *hp
1176 greg 2.14 )
1177 greg 1.1 {
1178 greg 2.23 int i, j;
1179 greg 1.2 double mag;
1180     double phi, xd, yd;
1181 greg 2.23 AMBSAMP *dp;
1182 greg 1.2
1183     xd = yd = 0.0;
1184     for (j = 0; j < hp->np; j++) {
1185     dp = da + j;
1186     mag = 0.0;
1187     for (i = 0; i < hp->nt; i++) {
1188     #ifdef DEBUG
1189     if (dp->t != i || dp->p != j)
1190     error(CONSISTENCY,
1191     "division order in dirgradient");
1192     #endif
1193 greg 2.2 /* tan(t) */
1194     mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0);
1195 greg 1.2 dp += hp->np;
1196     }
1197     phi = 2.0*PI * (j+.5)/hp->np + PI/2.0;
1198 gwlarson 2.8 xd += mag * tcos(phi);
1199     yd += mag * tsin(phi);
1200 greg 1.2 }
1201     for (i = 0; i < 3; i++)
1202 greg 2.16 gv[i] = xd*hp->ux[i] + yd*hp->uy[i];
1203 greg 1.1 }
1204 greg 2.25
1205     #endif /* ! NEWAMB */