ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/radiance/ray/src/rt/ambcomp.c
Revision: 2.65
Committed: Thu Aug 21 10:33:49 2014 UTC (9 years, 8 months ago) by greg
Content type: text/plain
Branch: MAIN
Changes since 2.64: +3 -3 lines
Log Message:
Added grazing angle extrapolation to BSDF interpolation

File Contents

# User Rev Content
1 greg 1.1 #ifndef lint
2 greg 2.65 static const char RCSid[] = "$Id: ambcomp.c,v 2.64 2014/08/19 15:04:40 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     ambsample( /* initial ambient division sample */
55     AMBHEMI *hp,
56     int i,
57     int j,
58     int n
59 greg 2.26 )
60     {
61 greg 2.61 AMBSAMP *ap = &ambsam(hp,i,j);
62     RAY ar;
63 greg 2.41 int hlist[3], ii;
64     double spt[2], zd;
65 greg 2.61 /* generate hemispherical sample */
66 greg 2.26 /* ambient coefficient for weight */
67     if (ambacc > FTINY)
68 greg 2.61 setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
69 greg 2.26 else
70 greg 2.61 copycolor(ar.rcoef, hp->acoef);
71 greg 2.62 if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0)
72 greg 2.41 return(0);
73 greg 2.26 if (ambacc > FTINY) {
74 greg 2.61 multcolor(ar.rcoef, hp->acoef);
75     scalecolor(ar.rcoef, 1./AVGREFL);
76 greg 2.41 }
77     hlist[0] = hp->rp->rno;
78 greg 2.46 hlist[1] = j;
79     hlist[2] = i;
80 greg 2.41 multisamp(spt, 2, urand(ilhash(hlist,3)+n));
81 greg 2.64 /* avoid coincident samples */
82     if (!n && (0 < i) & (i < hp->ns-1) &&
83     (0 < j) & (j < hp->ns-1)) {
84 greg 2.46 if ((spt[0] < 0.1) | (spt[0] >= 0.9))
85 greg 2.41 spt[0] = 0.1 + 0.8*frandom();
86 greg 2.46 if ((spt[1] < 0.1) | (spt[1] >= 0.9))
87 greg 2.41 spt[1] = 0.1 + 0.8*frandom();
88 greg 2.26 }
89 greg 2.46 SDsquare2disk(spt, (j+spt[1])/hp->ns, (i+spt[0])/hp->ns);
90 greg 2.26 zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]);
91     for (ii = 3; ii--; )
92 greg 2.61 ar.rdir[ii] = spt[0]*hp->ux[ii] +
93 greg 2.26 spt[1]*hp->uy[ii] +
94     zd*hp->rp->ron[ii];
95 greg 2.61 checknorm(ar.rdir);
96 greg 2.56 dimlist[ndims++] = AI(hp,i,j) + 90171;
97 greg 2.61 rayvalue(&ar); /* evaluate ray */
98     ndims--;
99     if (ar.rt <= FTINY)
100     return(0); /* should never happen */
101     multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
102 greg 2.62 if (ar.rt*ap->d < 1.0) /* new/closer distance? */
103 greg 2.61 ap->d = 1.0/ar.rt;
104     if (!n) { /* record first vertex & value */
105     if (ar.rt > 10.0*thescene.cusize)
106     ar.rt = 10.0*thescene.cusize;
107     VSUM(ap->p, ar.rorg, ar.rdir, ar.rt);
108     copycolor(ap->v, ar.rcol);
109     } else { /* else update recorded value */
110     hp->acol[RED] -= colval(ap->v,RED);
111     hp->acol[GRN] -= colval(ap->v,GRN);
112     hp->acol[BLU] -= colval(ap->v,BLU);
113     zd = 1.0/(double)(n+1);
114     scalecolor(ar.rcol, zd);
115     zd *= (double)n;
116     scalecolor(ap->v, zd);
117     addcolor(ap->v, ar.rcol);
118     }
119     addcolor(hp->acol, ap->v); /* add to our sum */
120 greg 2.41 return(1);
121     }
122    
123    
124     /* Estimate errors based on ambient division differences */
125     static float *
126     getambdiffs(AMBHEMI *hp)
127     {
128 greg 2.55 float *earr = (float *)calloc(hp->ns*hp->ns, sizeof(float));
129 greg 2.41 float *ep;
130 greg 2.42 AMBSAMP *ap;
131 greg 2.41 double b, d2;
132     int i, j;
133    
134     if (earr == NULL) /* out of memory? */
135     return(NULL);
136     /* compute squared neighbor diffs */
137 greg 2.42 for (ap = hp->sa, ep = earr, i = 0; i < hp->ns; i++)
138     for (j = 0; j < hp->ns; j++, ap++, ep++) {
139     b = bright(ap[0].v);
140 greg 2.41 if (i) { /* from above */
141 greg 2.42 d2 = b - bright(ap[-hp->ns].v);
142 greg 2.41 d2 *= d2;
143     ep[0] += d2;
144     ep[-hp->ns] += d2;
145     }
146 greg 2.55 if (!j) continue;
147     /* from behind */
148     d2 = b - bright(ap[-1].v);
149     d2 *= d2;
150     ep[0] += d2;
151     ep[-1] += d2;
152     if (!i) continue;
153     /* diagonal */
154     d2 = b - bright(ap[-hp->ns-1].v);
155     d2 *= d2;
156     ep[0] += d2;
157     ep[-hp->ns-1] += d2;
158 greg 2.41 }
159     /* correct for number of neighbors */
160 greg 2.55 earr[0] *= 8./3.;
161     earr[hp->ns-1] *= 8./3.;
162     earr[(hp->ns-1)*hp->ns] *= 8./3.;
163     earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 8./3.;
164 greg 2.41 for (i = 1; i < hp->ns-1; i++) {
165 greg 2.55 earr[i*hp->ns] *= 8./5.;
166     earr[i*hp->ns + hp->ns-1] *= 8./5.;
167 greg 2.41 }
168     for (j = 1; j < hp->ns-1; j++) {
169 greg 2.55 earr[j] *= 8./5.;
170     earr[(hp->ns-1)*hp->ns + j] *= 8./5.;
171 greg 2.41 }
172     return(earr);
173     }
174    
175    
176 greg 2.43 /* Perform super-sampling on hemisphere (introduces bias) */
177 greg 2.41 static void
178 greg 2.61 ambsupersamp(AMBHEMI *hp, int cnt)
179 greg 2.41 {
180     float *earr = getambdiffs(hp);
181 greg 2.54 double e2rem = 0;
182 greg 2.41 AMBSAMP *ap;
183     RAY ar;
184     float *ep;
185 greg 2.55 int i, j, n, nss;
186 greg 2.41
187     if (earr == NULL) /* just skip calc. if no memory */
188     return;
189 greg 2.54 /* accumulate estimated variances */
190 greg 2.55 for (ep = earr + hp->ns*hp->ns; ep > earr; )
191     e2rem += *--ep;
192 greg 2.41 ep = earr; /* perform super-sampling */
193     for (ap = hp->sa, i = 0; i < hp->ns; i++)
194     for (j = 0; j < hp->ns; j++, ap++) {
195 greg 2.55 if (e2rem <= FTINY)
196     goto done; /* nothing left to do */
197     nss = *ep/e2rem*cnt + frandom();
198 greg 2.62 for (n = 1; n <= nss && ambsample(hp,i,j,n); n++)
199     --cnt;
200 greg 2.61 e2rem -= *ep++; /* update remainder */
201 greg 2.41 }
202 greg 2.55 done:
203 greg 2.41 free(earr);
204     }
205    
206    
207 greg 2.61 static AMBHEMI *
208     samp_hemi( /* sample indirect hemisphere */
209     COLOR rcol,
210     RAY *r,
211     double wt
212     )
213     {
214     AMBHEMI *hp;
215     double d;
216     int n, i, j;
217     /* set number of divisions */
218     if (ambacc <= FTINY &&
219     wt > (d = 0.8*intens(rcol)*r->rweight/(ambdiv*minweight)))
220     wt = d; /* avoid ray termination */
221     n = sqrt(ambdiv * wt) + 0.5;
222 greg 2.65 i = 1 + 8*(ambacc > FTINY); /* minimum number of samples */
223 greg 2.61 if (n < i)
224     n = i;
225     /* allocate sampling array */
226     hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + sizeof(AMBSAMP)*(n*n - 1));
227     if (hp == NULL)
228     error(SYSTEM, "out of memory in samp_hemi");
229     hp->rp = r;
230     hp->ns = n;
231     hp->acol[RED] = hp->acol[GRN] = hp->acol[BLU] = 0.0;
232 greg 2.62 memset(hp->sa, 0, sizeof(AMBSAMP)*n*n);
233 greg 2.61 hp->sampOK = 0;
234     /* assign coefficient */
235     copycolor(hp->acoef, rcol);
236     d = 1.0/(n*n);
237     scalecolor(hp->acoef, d);
238     /* make tangent plane axes */
239     hp->uy[0] = 0.5 - frandom();
240     hp->uy[1] = 0.5 - frandom();
241     hp->uy[2] = 0.5 - frandom();
242     for (i = 3; i--; )
243     if ((-0.6 < r->ron[i]) & (r->ron[i] < 0.6))
244     break;
245     if (i < 0)
246     error(CONSISTENCY, "bad ray direction in samp_hemi");
247     hp->uy[i] = 1.0;
248     VCROSS(hp->ux, hp->uy, r->ron);
249     normalize(hp->ux);
250     VCROSS(hp->uy, r->ron, hp->ux);
251     /* sample divisions */
252     for (i = hp->ns; i--; )
253     for (j = hp->ns; j--; )
254     hp->sampOK += ambsample(hp, i, j, 0);
255     copycolor(rcol, hp->acol);
256     if (!hp->sampOK) { /* utter failure? */
257     free(hp);
258     return(NULL);
259     }
260     if (hp->sampOK < hp->ns*hp->ns) {
261     hp->sampOK *= -1; /* soft failure */
262     return(hp);
263     }
264     n = ambssamp*wt + 0.5;
265     if (n > 8) { /* perform super-sampling? */
266     ambsupersamp(hp, n);
267     copycolor(rcol, hp->acol);
268     }
269     return(hp); /* all is well */
270     }
271    
272    
273 greg 2.46 /* Return brightness of farthest ambient sample */
274     static double
275 greg 2.56 back_ambval(AMBHEMI *hp, const int n1, const int n2, const int n3)
276 greg 2.46 {
277 greg 2.56 if (hp->sa[n1].d <= hp->sa[n2].d) {
278     if (hp->sa[n1].d <= hp->sa[n3].d)
279     return(colval(hp->sa[n1].v,CIEY));
280     return(colval(hp->sa[n3].v,CIEY));
281     }
282     if (hp->sa[n2].d <= hp->sa[n3].d)
283     return(colval(hp->sa[n2].v,CIEY));
284     return(colval(hp->sa[n3].v,CIEY));
285 greg 2.46 }
286    
287    
288 greg 2.27 /* Compute vectors and coefficients for Hessian/gradient calcs */
289     static void
290 greg 2.56 comp_fftri(FFTRI *ftp, AMBHEMI *hp, const int n0, const int n1)
291 greg 2.27 {
292 greg 2.56 double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2;
293     int ii;
294    
295     VSUB(ftp->r_i, hp->sa[n0].p, hp->rp->rop);
296     VSUB(ftp->r_i1, hp->sa[n1].p, hp->rp->rop);
297     VSUB(ftp->e_i, hp->sa[n1].p, hp->sa[n0].p);
298 greg 2.35 VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1);
299     rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp);
300 greg 2.27 dot_e = DOT(ftp->e_i,ftp->e_i);
301     dot_er = DOT(ftp->e_i, ftp->r_i);
302 greg 2.32 rdot_r = 1.0/DOT(ftp->r_i,ftp->r_i);
303     rdot_r1 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
304     ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) * sqrt(rdot_r*rdot_r1) ) *
305 greg 2.35 sqrt( rdot_cp );
306 greg 2.32 ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r +
307 greg 2.35 dot_e*ftp->I1 )*0.5*rdot_cp;
308 greg 2.32 J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e;
309 greg 2.46 for (ii = 3; ii--; )
310     ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii];
311 greg 2.27 }
312    
313    
314 greg 2.28 /* Compose 3x3 matrix from two vectors */
315 greg 2.27 static void
316     compose_matrix(FVECT mat[3], FVECT va, FVECT vb)
317     {
318     mat[0][0] = 2.0*va[0]*vb[0];
319     mat[1][1] = 2.0*va[1]*vb[1];
320     mat[2][2] = 2.0*va[2]*vb[2];
321     mat[0][1] = mat[1][0] = va[0]*vb[1] + va[1]*vb[0];
322     mat[0][2] = mat[2][0] = va[0]*vb[2] + va[2]*vb[0];
323     mat[1][2] = mat[2][1] = va[1]*vb[2] + va[2]*vb[1];
324     }
325    
326    
327     /* Compute partial 3x3 Hessian matrix for edge */
328     static void
329     comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm)
330     {
331 greg 2.35 FVECT ncp;
332 greg 2.27 FVECT m1[3], m2[3], m3[3], m4[3];
333     double d1, d2, d3, d4;
334     double I3, J3, K3;
335     int i, j;
336     /* compute intermediate coefficients */
337     d1 = 1.0/DOT(ftp->r_i,ftp->r_i);
338     d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1);
339     d3 = 1.0/DOT(ftp->e_i,ftp->e_i);
340     d4 = DOT(ftp->e_i, ftp->r_i);
341 greg 2.35 I3 = ( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + 3.0/d3*ftp->I2 )
342     / ( 4.0*DOT(ftp->rcp,ftp->rcp) );
343 greg 2.27 J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3;
344     K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3);
345     /* intermediate matrices */
346 greg 2.35 VCROSS(ncp, nrm, ftp->e_i);
347     compose_matrix(m1, ncp, ftp->rI2_eJ2);
348 greg 2.27 compose_matrix(m2, ftp->r_i, ftp->r_i);
349     compose_matrix(m3, ftp->e_i, ftp->e_i);
350     compose_matrix(m4, ftp->r_i, ftp->e_i);
351 greg 2.35 d1 = DOT(nrm, ftp->rcp);
352 greg 2.27 d2 = -d1*ftp->I2;
353     d1 *= 2.0;
354     for (i = 3; i--; ) /* final matrix sum */
355     for (j = 3; j--; ) {
356     hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] +
357     2.0*J3*m4[i][j] );
358     hess[i][j] += d2*(i==j);
359 greg 2.46 hess[i][j] *= -1.0/PI;
360 greg 2.27 }
361     }
362    
363    
364     /* Reverse hessian calculation result for edge in other direction */
365     static void
366     rev_hessian(FVECT hess[3])
367     {
368     int i;
369    
370     for (i = 3; i--; ) {
371     hess[i][0] = -hess[i][0];
372     hess[i][1] = -hess[i][1];
373     hess[i][2] = -hess[i][2];
374     }
375     }
376    
377    
378     /* Add to radiometric Hessian from the given triangle */
379     static void
380     add2hessian(FVECT hess[3], FVECT ehess1[3],
381 greg 2.46 FVECT ehess2[3], FVECT ehess3[3], double v)
382 greg 2.27 {
383     int i, j;
384    
385     for (i = 3; i--; )
386     for (j = 3; j--; )
387     hess[i][j] += v*( ehess1[i][j] + ehess2[i][j] + ehess3[i][j] );
388     }
389    
390    
391     /* Compute partial displacement form factor gradient for edge */
392     static void
393     comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm)
394     {
395 greg 2.35 FVECT ncp;
396 greg 2.27 double f1;
397     int i;
398    
399 greg 2.35 f1 = 2.0*DOT(nrm, ftp->rcp);
400     VCROSS(ncp, nrm, ftp->e_i);
401 greg 2.27 for (i = 3; i--; )
402 greg 2.46 grad[i] = (0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] );
403 greg 2.27 }
404    
405    
406     /* Reverse gradient calculation result for edge in other direction */
407     static void
408     rev_gradient(FVECT grad)
409     {
410     grad[0] = -grad[0];
411     grad[1] = -grad[1];
412     grad[2] = -grad[2];
413     }
414    
415    
416     /* Add to displacement gradient from the given triangle */
417     static void
418 greg 2.46 add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, double v)
419 greg 2.27 {
420     int i;
421    
422     for (i = 3; i--; )
423     grad[i] += v*( egrad1[i] + egrad2[i] + egrad3[i] );
424     }
425    
426    
427     /* Compute anisotropic radii and eigenvector directions */
428 greg 2.53 static void
429 greg 2.27 eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3])
430     {
431     double hess2[2][2];
432     FVECT a, b;
433     double evalue[2], slope1, xmag1;
434     int i;
435     /* project Hessian to sample plane */
436     for (i = 3; i--; ) {
437     a[i] = DOT(hessian[i], uv[0]);
438     b[i] = DOT(hessian[i], uv[1]);
439     }
440     hess2[0][0] = DOT(uv[0], a);
441     hess2[0][1] = DOT(uv[0], b);
442     hess2[1][0] = DOT(uv[1], a);
443     hess2[1][1] = DOT(uv[1], b);
444 greg 2.38 /* compute eigenvalue(s) */
445     i = quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1],
446     hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]);
447     if (i == 1) /* double-root (circle) */
448     evalue[1] = evalue[0];
449     if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) |
450 greg 2.53 ((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) {
451     ra[0] = ra[1] = maxarad;
452     return;
453     }
454 greg 2.27 if (evalue[0] > evalue[1]) {
455 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[0]));
456     ra[1] = sqrt(sqrt(4.0/evalue[1]));
457 greg 2.27 slope1 = evalue[1];
458     } else {
459 greg 2.29 ra[0] = sqrt(sqrt(4.0/evalue[1]));
460     ra[1] = sqrt(sqrt(4.0/evalue[0]));
461 greg 2.27 slope1 = evalue[0];
462     }
463     /* compute unit eigenvectors */
464     if (fabs(hess2[0][1]) <= FTINY)
465     return; /* uv OK as is */
466     slope1 = (slope1 - hess2[0][0]) / hess2[0][1];
467     xmag1 = sqrt(1.0/(1.0 + slope1*slope1));
468     for (i = 3; i--; ) {
469     b[i] = xmag1*uv[0][i] + slope1*xmag1*uv[1][i];
470     a[i] = slope1*xmag1*uv[0][i] - xmag1*uv[1][i];
471     }
472     VCOPY(uv[0], a);
473     VCOPY(uv[1], b);
474     }
475    
476    
477 greg 2.26 static void
478     ambHessian( /* anisotropic radii & pos. gradient */
479     AMBHEMI *hp,
480     FVECT uv[2], /* returned */
481 greg 2.28 float ra[2], /* returned (optional) */
482     float pg[2] /* returned (optional) */
483 greg 2.26 )
484     {
485 greg 2.27 static char memerrmsg[] = "out of memory in ambHessian()";
486     FVECT (*hessrow)[3] = NULL;
487     FVECT *gradrow = NULL;
488     FVECT hessian[3];
489     FVECT gradient;
490     FFTRI fftr;
491     int i, j;
492     /* be sure to assign unit vectors */
493     VCOPY(uv[0], hp->ux);
494     VCOPY(uv[1], hp->uy);
495     /* clock-wise vertex traversal from sample POV */
496     if (ra != NULL) { /* initialize Hessian row buffer */
497 greg 2.28 hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1));
498 greg 2.27 if (hessrow == NULL)
499     error(SYSTEM, memerrmsg);
500     memset(hessian, 0, sizeof(hessian));
501     } else if (pg == NULL) /* bogus call? */
502     return;
503     if (pg != NULL) { /* initialize form factor row buffer */
504 greg 2.28 gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1));
505 greg 2.27 if (gradrow == NULL)
506     error(SYSTEM, memerrmsg);
507     memset(gradient, 0, sizeof(gradient));
508     }
509     /* compute first row of edges */
510     for (j = 0; j < hp->ns-1; j++) {
511 greg 2.56 comp_fftri(&fftr, hp, AI(hp,0,j), AI(hp,0,j+1));
512 greg 2.27 if (hessrow != NULL)
513     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
514     if (gradrow != NULL)
515     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
516     }
517     /* sum each row of triangles */
518     for (i = 0; i < hp->ns-1; i++) {
519     FVECT hesscol[3]; /* compute first vertical edge */
520     FVECT gradcol;
521 greg 2.56 comp_fftri(&fftr, hp, AI(hp,i,0), AI(hp,i+1,0));
522 greg 2.27 if (hessrow != NULL)
523     comp_hessian(hesscol, &fftr, hp->rp->ron);
524     if (gradrow != NULL)
525     comp_gradient(gradcol, &fftr, hp->rp->ron);
526     for (j = 0; j < hp->ns-1; j++) {
527     FVECT hessdia[3]; /* compute triangle contributions */
528     FVECT graddia;
529 greg 2.46 double backg;
530 greg 2.56 backg = back_ambval(hp, AI(hp,i,j),
531     AI(hp,i,j+1), AI(hp,i+1,j));
532 greg 2.27 /* diagonal (inner) edge */
533 greg 2.56 comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j));
534 greg 2.27 if (hessrow != NULL) {
535     comp_hessian(hessdia, &fftr, hp->rp->ron);
536     rev_hessian(hesscol);
537     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
538     }
539 greg 2.39 if (gradrow != NULL) {
540 greg 2.27 comp_gradient(graddia, &fftr, hp->rp->ron);
541     rev_gradient(gradcol);
542     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
543     }
544     /* initialize edge in next row */
545 greg 2.56 comp_fftri(&fftr, hp, AI(hp,i+1,j+1), AI(hp,i+1,j));
546 greg 2.27 if (hessrow != NULL)
547     comp_hessian(hessrow[j], &fftr, hp->rp->ron);
548     if (gradrow != NULL)
549     comp_gradient(gradrow[j], &fftr, hp->rp->ron);
550     /* new column edge & paired triangle */
551 greg 2.56 backg = back_ambval(hp, AI(hp,i+1,j+1),
552     AI(hp,i+1,j), AI(hp,i,j+1));
553     comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j+1));
554 greg 2.27 if (hessrow != NULL) {
555     comp_hessian(hesscol, &fftr, hp->rp->ron);
556     rev_hessian(hessdia);
557     add2hessian(hessian, hessrow[j], hessdia, hesscol, backg);
558     if (i < hp->ns-2)
559     rev_hessian(hessrow[j]);
560     }
561     if (gradrow != NULL) {
562     comp_gradient(gradcol, &fftr, hp->rp->ron);
563     rev_gradient(graddia);
564     add2gradient(gradient, gradrow[j], graddia, gradcol, backg);
565     if (i < hp->ns-2)
566     rev_gradient(gradrow[j]);
567     }
568     }
569     }
570     /* release row buffers */
571     if (hessrow != NULL) free(hessrow);
572     if (gradrow != NULL) free(gradrow);
573    
574     if (ra != NULL) /* extract eigenvectors & radii */
575     eigenvectors(uv, ra, hessian);
576 greg 2.32 if (pg != NULL) { /* tangential position gradient */
577     pg[0] = DOT(gradient, uv[0]);
578     pg[1] = DOT(gradient, uv[1]);
579 greg 2.27 }
580     }
581    
582    
583     /* Compute direction gradient from a hemispherical sampling */
584     static void
585     ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2])
586     {
587 greg 2.41 AMBSAMP *ap;
588     double dgsum[2];
589     int n;
590     FVECT vd;
591     double gfact;
592 greg 2.27
593 greg 2.29 dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */
594 greg 2.27 for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) {
595     /* use vector for azimuth + 90deg */
596     VSUB(vd, ap->p, hp->rp->rop);
597 greg 2.29 /* brightness over cosine factor */
598     gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd);
599 greg 2.40 /* sine = proj_radius/vd_length */
600     dgsum[0] -= DOT(uv[1], vd) * gfact;
601     dgsum[1] += DOT(uv[0], vd) * gfact;
602 greg 2.26 }
603 greg 2.29 dg[0] = dgsum[0] / (hp->ns*hp->ns);
604     dg[1] = dgsum[1] / (hp->ns*hp->ns);
605 greg 2.26 }
606    
607 greg 2.27
608 greg 2.49 /* Compute potential light leak direction flags for cache value */
609     static uint32
610     ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1)
611 greg 2.47 {
612 greg 2.50 const double max_d = 1.0/(minarad*ambacc + 0.001);
613     const double ang_res = 0.5*PI/(hp->ns-1);
614     const double ang_step = ang_res/((int)(16/PI*ang_res) + (1+FTINY));
615 greg 2.51 double avg_d = 0;
616 greg 2.50 uint32 flgs = 0;
617 greg 2.58 FVECT vec;
618 greg 2.62 double u, v;
619 greg 2.58 double ang, a1;
620 greg 2.50 int i, j;
621 greg 2.52 /* don't bother for a few samples */
622     if (hp->ns < 12)
623     return(0);
624     /* check distances overhead */
625 greg 2.51 for (i = hp->ns*3/4; i-- > hp->ns>>2; )
626     for (j = hp->ns*3/4; j-- > hp->ns>>2; )
627     avg_d += ambsam(hp,i,j).d;
628     avg_d *= 4.0/(hp->ns*hp->ns);
629 greg 2.52 if (avg_d*r0 >= 1.0) /* ceiling too low for corral? */
630     return(0);
631     if (avg_d >= max_d) /* insurance */
632 greg 2.51 return(0);
633     /* else circle around perimeter */
634 greg 2.47 for (i = 0; i < hp->ns; i++)
635     for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) {
636     AMBSAMP *ap = &ambsam(hp,i,j);
637 greg 2.50 if ((ap->d <= FTINY) | (ap->d >= max_d))
638     continue; /* too far or too near */
639 greg 2.47 VSUB(vec, ap->p, hp->rp->rop);
640 greg 2.62 u = DOT(vec, uv[0]);
641     v = DOT(vec, uv[1]);
642     if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= u*u + v*v)
643 greg 2.49 continue; /* occluder outside ellipse */
644     ang = atan2a(v, u); /* else set direction flags */
645 greg 2.50 for (a1 = ang-.5*ang_res; a1 <= ang+.5*ang_res; a1 += ang_step)
646     flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0)));
647 greg 2.47 }
648 greg 2.58 /* add low-angle incident (< 20deg) */
649     if (fabs(hp->rp->rod) <= 0.342) {
650     u = -DOT(hp->rp->rdir, uv[0]);
651     v = -DOT(hp->rp->rdir, uv[1]);
652     if ((r0*r0*u*u + r1*r1*v*v) > hp->rp->rot*hp->rp->rot) {
653     ang = atan2a(v, u);
654     ang += 2.*PI*(ang < 0);
655     ang *= 16/PI;
656     if ((ang < .5) | (ang >= 31.5))
657     flgs |= 0x80000001;
658     else
659     flgs |= 3L<<(int)(ang-.5);
660     }
661     }
662 greg 2.49 return(flgs);
663 greg 2.47 }
664    
665    
666 greg 2.26 int
667     doambient( /* compute ambient component */
668     COLOR rcol, /* input/output color */
669     RAY *r,
670     double wt,
671 greg 2.27 FVECT uv[2], /* returned (optional) */
672     float ra[2], /* returned (optional) */
673     float pg[2], /* returned (optional) */
674 greg 2.49 float dg[2], /* returned (optional) */
675     uint32 *crlp /* returned (optional) */
676 greg 2.26 )
677     {
678 greg 2.61 AMBHEMI *hp = samp_hemi(rcol, r, wt);
679 greg 2.41 FVECT my_uv[2];
680 greg 2.61 double d, K;
681 greg 2.41 AMBSAMP *ap;
682 greg 2.61 int i;
683     /* clear return values */
684 greg 2.26 if (uv != NULL)
685     memset(uv, 0, sizeof(FVECT)*2);
686     if (ra != NULL)
687     ra[0] = ra[1] = 0.0;
688     if (pg != NULL)
689     pg[0] = pg[1] = 0.0;
690     if (dg != NULL)
691     dg[0] = dg[1] = 0.0;
692 greg 2.49 if (crlp != NULL)
693     *crlp = 0;
694 greg 2.61 if (hp == NULL) /* sampling falure? */
695     return(0);
696    
697     if ((ra == NULL) & (pg == NULL) & (dg == NULL) ||
698 greg 2.65 (hp->sampOK < 0) | (hp->ns < 9)) {
699 greg 2.61 free(hp); /* Hessian not requested/possible */
700     return(-1); /* value-only return value */
701 greg 2.26 }
702 greg 2.61 if ((d = bright(rcol)) > FTINY) { /* normalize Y values */
703 greg 2.45 d = 0.99*(hp->ns*hp->ns)/d;
704 greg 2.38 K = 0.01;
705 greg 2.45 } else { /* or fall back on geometric Hessian */
706 greg 2.38 K = 1.0;
707     pg = NULL;
708     dg = NULL;
709 greg 2.53 crlp = NULL;
710 greg 2.38 }
711 greg 2.29 ap = hp->sa; /* relative Y channel from here on... */
712 greg 2.26 for (i = hp->ns*hp->ns; i--; ap++)
713 greg 2.38 colval(ap->v,CIEY) = bright(ap->v)*d + K;
714 greg 2.26
715     if (uv == NULL) /* make sure we have axis pointers */
716     uv = my_uv;
717     /* compute radii & pos. gradient */
718     ambHessian(hp, uv, ra, pg);
719 greg 2.29
720 greg 2.26 if (dg != NULL) /* compute direction gradient */
721     ambdirgrad(hp, uv, dg);
722 greg 2.29
723 greg 2.28 if (ra != NULL) { /* scale/clamp radii */
724 greg 2.35 if (pg != NULL) {
725     if (ra[0]*(d = fabs(pg[0])) > 1.0)
726     ra[0] = 1.0/d;
727     if (ra[1]*(d = fabs(pg[1])) > 1.0)
728     ra[1] = 1.0/d;
729 greg 2.48 if (ra[0] > ra[1])
730     ra[0] = ra[1];
731 greg 2.35 }
732 greg 2.29 if (ra[0] < minarad) {
733     ra[0] = minarad;
734     if (ra[1] < minarad)
735     ra[1] = minarad;
736     }
737 greg 2.60 ra[0] *= d = 1.0/sqrt(wt);
738 greg 2.26 if ((ra[1] *= d) > 2.0*ra[0])
739     ra[1] = 2.0*ra[0];
740 greg 2.28 if (ra[1] > maxarad) {
741     ra[1] = maxarad;
742     if (ra[0] > maxarad)
743     ra[0] = maxarad;
744     }
745 greg 2.53 /* flag encroached directions */
746 greg 2.57 if ((wt >= 0.89*AVGREFL) & (crlp != NULL))
747 greg 2.49 *crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc);
748 greg 2.35 if (pg != NULL) { /* cap gradient if necessary */
749     d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1];
750     if (d > 1.0) {
751     d = 1.0/sqrt(d);
752     pg[0] *= d;
753     pg[1] *= d;
754     }
755     }
756 greg 2.26 }
757     free(hp); /* clean up and return */
758     return(1);
759     }
760    
761    
762 greg 2.25 #else /* ! NEWAMB */
763 greg 1.1
764    
765 greg 2.15 void
766 greg 2.14 inithemi( /* initialize sampling hemisphere */
767 greg 2.23 AMBHEMI *hp,
768 greg 2.16 COLOR ac,
769 greg 2.14 RAY *r,
770     double wt
771     )
772 greg 1.1 {
773 greg 2.16 double d;
774 greg 2.23 int i;
775 greg 2.14 /* set number of divisions */
776 greg 2.16 if (ambacc <= FTINY &&
777 greg 2.20 wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight)))
778 greg 2.16 wt = d; /* avoid ray termination */
779     hp->nt = sqrt(ambdiv * wt / PI) + 0.5;
780 greg 2.14 i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */
781     if (hp->nt < i)
782     hp->nt = i;
783     hp->np = PI * hp->nt + 0.5;
784     /* set number of super-samples */
785 greg 2.15 hp->ns = ambssamp * wt + 0.5;
786 greg 2.16 /* assign coefficient */
787 greg 2.14 copycolor(hp->acoef, ac);
788 greg 2.16 d = 1.0/(hp->nt*hp->np);
789     scalecolor(hp->acoef, d);
790 greg 2.14 /* make axes */
791     VCOPY(hp->uz, r->ron);
792     hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0;
793     for (i = 0; i < 3; i++)
794     if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6)
795     break;
796     if (i >= 3)
797     error(CONSISTENCY, "bad ray direction in inithemi");
798     hp->uy[i] = 1.0;
799     fcross(hp->ux, hp->uy, hp->uz);
800     normalize(hp->ux);
801     fcross(hp->uy, hp->uz, hp->ux);
802 greg 1.1 }
803    
804    
805 greg 2.9 int
806 greg 2.14 divsample( /* sample a division */
807 greg 2.23 AMBSAMP *dp,
808 greg 2.14 AMBHEMI *h,
809     RAY *r
810     )
811 greg 1.1 {
812     RAY ar;
813 greg 1.11 int hlist[3];
814     double spt[2];
815 greg 1.1 double xd, yd, zd;
816     double b2;
817     double phi;
818 greg 2.23 int i;
819 greg 2.15 /* ambient coefficient for weight */
820 greg 2.16 if (ambacc > FTINY)
821     setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL);
822     else
823     copycolor(ar.rcoef, h->acoef);
824 greg 2.14 if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0)
825 greg 1.4 return(-1);
826 greg 2.17 if (ambacc > FTINY) {
827     multcolor(ar.rcoef, h->acoef);
828     scalecolor(ar.rcoef, 1./AVGREFL);
829     }
830 greg 1.1 hlist[0] = r->rno;
831     hlist[1] = dp->t;
832     hlist[2] = dp->p;
833 greg 1.13 multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n));
834 greg 1.11 zd = sqrt((dp->t + spt[0])/h->nt);
835     phi = 2.0*PI * (dp->p + spt[1])/h->np;
836 gwlarson 2.8 xd = tcos(phi) * zd;
837     yd = tsin(phi) * zd;
838 greg 1.1 zd = sqrt(1.0 - zd*zd);
839 greg 1.2 for (i = 0; i < 3; i++)
840     ar.rdir[i] = xd*h->ux[i] +
841     yd*h->uy[i] +
842     zd*h->uz[i];
843 greg 2.22 checknorm(ar.rdir);
844 greg 1.2 dimlist[ndims++] = dp->t*h->np + dp->p + 90171;
845 greg 1.1 rayvalue(&ar);
846     ndims--;
847 greg 2.16 multcolor(ar.rcol, ar.rcoef); /* apply coefficient */
848 greg 1.1 addcolor(dp->v, ar.rcol);
849 greg 2.9 /* use rt to improve gradient calc */
850     if (ar.rt > FTINY && ar.rt < FHUGE)
851     dp->r += 1.0/ar.rt;
852 greg 1.1 /* (re)initialize error */
853     if (dp->n++) {
854     b2 = bright(dp->v)/dp->n - bright(ar.rcol);
855     b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1));
856     dp->k = b2/(dp->n*dp->n);
857     } else
858     dp->k = 0.0;
859 greg 1.4 return(0);
860 greg 1.1 }
861    
862    
863 greg 2.14 static int
864     ambcmp( /* decreasing order */
865     const void *p1,
866     const void *p2
867     )
868     {
869     const AMBSAMP *d1 = (const AMBSAMP *)p1;
870     const AMBSAMP *d2 = (const AMBSAMP *)p2;
871    
872     if (d1->k < d2->k)
873     return(1);
874     if (d1->k > d2->k)
875     return(-1);
876     return(0);
877     }
878    
879    
880     static int
881     ambnorm( /* standard order */
882     const void *p1,
883     const void *p2
884     )
885     {
886     const AMBSAMP *d1 = (const AMBSAMP *)p1;
887     const AMBSAMP *d2 = (const AMBSAMP *)p2;
888 greg 2.23 int c;
889 greg 2.14
890     if ( (c = d1->t - d2->t) )
891     return(c);
892     return(d1->p - d2->p);
893     }
894    
895    
896 greg 1.1 double
897 greg 2.14 doambient( /* compute ambient component */
898 greg 2.23 COLOR rcol,
899 greg 2.14 RAY *r,
900     double wt,
901     FVECT pg,
902     FVECT dg
903     )
904 greg 1.1 {
905 greg 2.24 double b, d=0;
906 greg 1.1 AMBHEMI hemi;
907     AMBSAMP *div;
908     AMBSAMP dnew;
909 greg 2.23 double acol[3];
910     AMBSAMP *dp;
911 greg 1.1 double arad;
912 greg 2.19 int divcnt;
913 greg 2.23 int i, j;
914 greg 1.1 /* initialize hemisphere */
915 greg 2.23 inithemi(&hemi, rcol, r, wt);
916 greg 2.19 divcnt = hemi.nt * hemi.np;
917 greg 2.17 /* initialize */
918     if (pg != NULL)
919     pg[0] = pg[1] = pg[2] = 0.0;
920     if (dg != NULL)
921     dg[0] = dg[1] = dg[2] = 0.0;
922 greg 2.23 setcolor(rcol, 0.0, 0.0, 0.0);
923 greg 2.19 if (divcnt == 0)
924 greg 1.1 return(0.0);
925 greg 2.14 /* allocate super-samples */
926 greg 2.15 if (hemi.ns > 0 || pg != NULL || dg != NULL) {
927 greg 2.19 div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP));
928 greg 1.1 if (div == NULL)
929     error(SYSTEM, "out of memory in doambient");
930     } else
931     div = NULL;
932     /* sample the divisions */
933     arad = 0.0;
934 greg 2.23 acol[0] = acol[1] = acol[2] = 0.0;
935 greg 1.1 if ((dp = div) == NULL)
936     dp = &dnew;
937 greg 2.19 divcnt = 0;
938 greg 1.1 for (i = 0; i < hemi.nt; i++)
939     for (j = 0; j < hemi.np; j++) {
940     dp->t = i; dp->p = j;
941     setcolor(dp->v, 0.0, 0.0, 0.0);
942 greg 1.2 dp->r = 0.0;
943 greg 1.1 dp->n = 0;
944 greg 2.16 if (divsample(dp, &hemi, r) < 0) {
945 greg 2.19 if (div != NULL)
946     dp++;
947 greg 2.16 continue;
948     }
949 greg 2.6 arad += dp->r;
950 greg 2.19 divcnt++;
951 greg 1.1 if (div != NULL)
952     dp++;
953 greg 2.6 else
954 greg 1.1 addcolor(acol, dp->v);
955     }
956 greg 2.21 if (!divcnt) {
957     if (div != NULL)
958     free((void *)div);
959 greg 2.19 return(0.0); /* no samples taken */
960 greg 2.21 }
961 greg 2.19 if (divcnt < hemi.nt*hemi.np) {
962     pg = dg = NULL; /* incomplete sampling */
963     hemi.ns = 0;
964     } else if (arad > FTINY && divcnt/arad < minarad) {
965 greg 2.15 hemi.ns = 0; /* close enough */
966 greg 2.19 } else if (hemi.ns > 0) { /* else perform super-sampling? */
967 greg 1.4 comperrs(div, &hemi); /* compute errors */
968 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */
969 greg 1.1 /* super-sample */
970 greg 2.15 for (i = hemi.ns; i > 0; i--) {
971 schorsch 2.11 dnew = *div;
972 greg 2.16 if (divsample(&dnew, &hemi, r) < 0) {
973     dp++;
974     continue;
975     }
976     dp = div; /* reinsert */
977 greg 2.19 j = divcnt < i ? divcnt : i;
978 greg 1.1 while (--j > 0 && dnew.k < dp[1].k) {
979 schorsch 2.11 *dp = *(dp+1);
980 greg 1.1 dp++;
981     }
982 schorsch 2.11 *dp = dnew;
983 greg 1.1 }
984 greg 1.2 if (pg != NULL || dg != NULL) /* restore order */
985 greg 2.19 qsort(div, divcnt, sizeof(AMBSAMP), ambnorm);
986 greg 1.1 }
987     /* compute returned values */
988 greg 1.3 if (div != NULL) {
989 greg 2.19 arad = 0.0; /* note: divcnt may be < nt*np */
990     for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) {
991 greg 1.3 arad += dp->r;
992     if (dp->n > 1) {
993     b = 1.0/dp->n;
994     scalecolor(dp->v, b);
995     dp->r *= b;
996     dp->n = 1;
997     }
998     addcolor(acol, dp->v);
999     }
1000 greg 1.5 b = bright(acol);
1001 greg 1.6 if (b > FTINY) {
1002 greg 2.17 b = 1.0/b; /* compute & normalize gradient(s) */
1003 greg 1.6 if (pg != NULL) {
1004     posgradient(pg, div, &hemi);
1005     for (i = 0; i < 3; i++)
1006     pg[i] *= b;
1007     }
1008     if (dg != NULL) {
1009     dirgradient(dg, div, &hemi);
1010     for (i = 0; i < 3; i++)
1011     dg[i] *= b;
1012     }
1013 greg 1.5 }
1014 greg 2.9 free((void *)div);
1015 greg 1.3 }
1016 greg 2.23 copycolor(rcol, acol);
1017 greg 1.1 if (arad <= FTINY)
1018 greg 1.16 arad = maxarad;
1019 greg 2.3 else
1020 greg 2.19 arad = (divcnt+hemi.ns)/arad;
1021 greg 1.15 if (pg != NULL) { /* reduce radius if gradient large */
1022     d = DOT(pg,pg);
1023     if (d*arad*arad > 1.0)
1024     arad = 1.0/sqrt(d);
1025     }
1026 greg 1.16 if (arad < minarad) {
1027 greg 1.1 arad = minarad;
1028 greg 1.16 if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */
1029     d = 1.0/arad/sqrt(d);
1030     for (i = 0; i < 3; i++)
1031     pg[i] *= d;
1032     }
1033     }
1034 greg 2.3 if ((arad /= sqrt(wt)) > maxarad)
1035     arad = maxarad;
1036     return(arad);
1037 greg 1.1 }
1038    
1039    
1040 greg 2.9 void
1041 greg 2.14 comperrs( /* compute initial error estimates */
1042     AMBSAMP *da, /* assumes standard ordering */
1043 greg 2.23 AMBHEMI *hp
1044 greg 2.14 )
1045 greg 1.1 {
1046     double b, b2;
1047     int i, j;
1048 greg 2.23 AMBSAMP *dp;
1049 greg 1.1 /* sum differences from neighbors */
1050     dp = da;
1051     for (i = 0; i < hp->nt; i++)
1052     for (j = 0; j < hp->np; j++) {
1053 greg 1.6 #ifdef DEBUG
1054     if (dp->t != i || dp->p != j)
1055     error(CONSISTENCY,
1056     "division order in comperrs");
1057     #endif
1058 greg 1.1 b = bright(dp[0].v);
1059     if (i > 0) { /* from above */
1060     b2 = bright(dp[-hp->np].v) - b;
1061     b2 *= b2 * 0.25;
1062     dp[0].k += b2;
1063     dp[-hp->np].k += b2;
1064     }
1065     if (j > 0) { /* from behind */
1066     b2 = bright(dp[-1].v) - b;
1067     b2 *= b2 * 0.25;
1068     dp[0].k += b2;
1069     dp[-1].k += b2;
1070 greg 1.4 } else { /* around */
1071     b2 = bright(dp[hp->np-1].v) - b;
1072 greg 1.1 b2 *= b2 * 0.25;
1073     dp[0].k += b2;
1074 greg 1.4 dp[hp->np-1].k += b2;
1075 greg 1.1 }
1076     dp++;
1077     }
1078     /* divide by number of neighbors */
1079     dp = da;
1080     for (j = 0; j < hp->np; j++) /* top row */
1081     (dp++)->k *= 1.0/3.0;
1082     if (hp->nt < 2)
1083     return;
1084     for (i = 1; i < hp->nt-1; i++) /* central region */
1085     for (j = 0; j < hp->np; j++)
1086     (dp++)->k *= 0.25;
1087     for (j = 0; j < hp->np; j++) /* bottom row */
1088     (dp++)->k *= 1.0/3.0;
1089     }
1090    
1091    
1092 greg 2.9 void
1093 greg 2.14 posgradient( /* compute position gradient */
1094     FVECT gv,
1095     AMBSAMP *da, /* assumes standard ordering */
1096 greg 2.23 AMBHEMI *hp
1097 greg 2.14 )
1098 greg 1.1 {
1099 greg 2.23 int i, j;
1100 greg 2.2 double nextsine, lastsine, b, d;
1101 greg 1.2 double mag0, mag1;
1102     double phi, cosp, sinp, xd, yd;
1103 greg 2.23 AMBSAMP *dp;
1104 greg 1.2
1105     xd = yd = 0.0;
1106     for (j = 0; j < hp->np; j++) {
1107     dp = da + j;
1108     mag0 = mag1 = 0.0;
1109 greg 2.2 lastsine = 0.0;
1110 greg 1.2 for (i = 0; i < hp->nt; i++) {
1111     #ifdef DEBUG
1112     if (dp->t != i || dp->p != j)
1113     error(CONSISTENCY,
1114     "division order in posgradient");
1115     #endif
1116     b = bright(dp->v);
1117     if (i > 0) {
1118     d = dp[-hp->np].r;
1119     if (dp[0].r > d) d = dp[0].r;
1120 greg 2.2 /* sin(t)*cos(t)^2 */
1121     d *= lastsine * (1.0 - (double)i/hp->nt);
1122 greg 1.2 mag0 += d*(b - bright(dp[-hp->np].v));
1123     }
1124 greg 2.2 nextsine = sqrt((double)(i+1)/hp->nt);
1125 greg 1.2 if (j > 0) {
1126     d = dp[-1].r;
1127     if (dp[0].r > d) d = dp[0].r;
1128 greg 2.2 mag1 += d * (nextsine - lastsine) *
1129     (b - bright(dp[-1].v));
1130 greg 1.2 } else {
1131     d = dp[hp->np-1].r;
1132     if (dp[0].r > d) d = dp[0].r;
1133 greg 2.2 mag1 += d * (nextsine - lastsine) *
1134     (b - bright(dp[hp->np-1].v));
1135 greg 1.2 }
1136     dp += hp->np;
1137 greg 2.2 lastsine = nextsine;
1138 greg 1.2 }
1139 greg 2.2 mag0 *= 2.0*PI / hp->np;
1140 greg 1.2 phi = 2.0*PI * (double)j/hp->np;
1141 gwlarson 2.8 cosp = tcos(phi); sinp = tsin(phi);
1142 greg 1.2 xd += mag0*cosp - mag1*sinp;
1143     yd += mag0*sinp + mag1*cosp;
1144     }
1145     for (i = 0; i < 3; i++)
1146 greg 2.16 gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI;
1147 greg 1.1 }
1148    
1149    
1150 greg 2.9 void
1151 greg 2.14 dirgradient( /* compute direction gradient */
1152     FVECT gv,
1153     AMBSAMP *da, /* assumes standard ordering */
1154 greg 2.23 AMBHEMI *hp
1155 greg 2.14 )
1156 greg 1.1 {
1157 greg 2.23 int i, j;
1158 greg 1.2 double mag;
1159     double phi, xd, yd;
1160 greg 2.23 AMBSAMP *dp;
1161 greg 1.2
1162     xd = yd = 0.0;
1163     for (j = 0; j < hp->np; j++) {
1164     dp = da + j;
1165     mag = 0.0;
1166     for (i = 0; i < hp->nt; i++) {
1167     #ifdef DEBUG
1168     if (dp->t != i || dp->p != j)
1169     error(CONSISTENCY,
1170     "division order in dirgradient");
1171     #endif
1172 greg 2.2 /* tan(t) */
1173     mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0);
1174 greg 1.2 dp += hp->np;
1175     }
1176     phi = 2.0*PI * (j+.5)/hp->np + PI/2.0;
1177 gwlarson 2.8 xd += mag * tcos(phi);
1178     yd += mag * tsin(phi);
1179 greg 1.2 }
1180     for (i = 0; i < 3; i++)
1181 greg 2.16 gv[i] = xd*hp->ux[i] + yd*hp->uy[i];
1182 greg 1.1 }
1183 greg 2.25
1184     #endif /* ! NEWAMB */