1 |
#ifndef lint |
2 |
static const char RCSid[] = "$Id: ambcomp.c,v 2.31 2014/04/24 06:03:15 greg Exp $"; |
3 |
#endif |
4 |
/* |
5 |
* Routines to compute "ambient" values using Monte Carlo |
6 |
* |
7 |
* 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 |
* Declarations of external symbols in ambient.h |
12 |
*/ |
13 |
|
14 |
#include "copyright.h" |
15 |
|
16 |
#include "ray.h" |
17 |
#include "ambient.h" |
18 |
#include "random.h" |
19 |
|
20 |
#ifdef NEWAMB |
21 |
|
22 |
extern void SDsquare2disk(double ds[2], double seedx, double seedy); |
23 |
|
24 |
typedef struct { |
25 |
RAY *rp; /* originating ray sample */ |
26 |
FVECT ux, uy; /* tangent axis unit vectors */ |
27 |
int ns; /* number of samples per axis */ |
28 |
COLOR acoef; /* division contribution coefficient */ |
29 |
struct s_ambsamp { |
30 |
COLOR v; /* hemisphere sample value */ |
31 |
FVECT p; /* 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 |
typedef struct { |
38 |
FVECT r_i, r_i1, e_i, rI2_eJ2; |
39 |
double nf, I1, I2; |
40 |
} FFTRI; /* vectors and coefficients for Hessian calculation */ |
41 |
|
42 |
|
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 |
i = 1 + 5*(ambacc > FTINY); /* minimum number of samples */ |
59 |
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 |
/* 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 |
for (i = 0; i < 3; i++) |
77 |
if (r->ron[i] < 0.6 && r->ron[i] > -0.6) |
78 |
break; |
79 |
if (i >= 3) |
80 |
error(CONSISTENCY, "bad ray direction in inithemi()"); |
81 |
hp->uy[i] = 1.0; |
82 |
VCROSS(hp->ux, hp->uy, r->ron); |
83 |
normalize(hp->ux); |
84 |
VCROSS(hp->uy, r->ron, hp->ux); |
85 |
/* we're ready to sample */ |
86 |
return(hp); |
87 |
} |
88 |
|
89 |
|
90 |
static struct s_ambsamp * |
91 |
ambsample( /* sample an ambient direction */ |
92 |
AMBHEMI *hp, |
93 |
int i, |
94 |
int j |
95 |
) |
96 |
{ |
97 |
struct s_ambsamp *ap = &ambsamp(hp,i,j); |
98 |
RAY ar; |
99 |
double spt[2], zd; |
100 |
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 |
goto badsample; |
108 |
if (ambacc > FTINY) { |
109 |
multcolor(ar.rcoef, hp->acoef); |
110 |
scalecolor(ar.rcoef, 1./AVGREFL); |
111 |
} |
112 |
/* generate hemispherical sample */ |
113 |
SDsquare2disk(spt, (i+.1+.8*frandom())/hp->ns, |
114 |
(j+.1+.8*frandom())/hp->ns ); |
115 |
zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]); |
116 |
for (ii = 3; ii--; ) |
117 |
ar.rdir[ii] = spt[0]*hp->ux[ii] + |
118 |
spt[1]*hp->uy[ii] + |
119 |
zd*hp->rp->ron[ii]; |
120 |
checknorm(ar.rdir); |
121 |
dimlist[ndims++] = i*hp->ns + j + 90171; |
122 |
rayvalue(&ar); /* evaluate ray */ |
123 |
ndims--; |
124 |
if (ar.rt > 20.0*maxarad) /* limit vertex distance */ |
125 |
ar.rt = 20.0*maxarad; |
126 |
else if (ar.rt <= FTINY) /* should never happen! */ |
127 |
goto badsample; |
128 |
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
129 |
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
130 |
copycolor(ap->v, ar.rcol); |
131 |
return(ap); |
132 |
badsample: |
133 |
setcolor(ap->v, 0., 0., 0.); |
134 |
VCOPY(ap->p, hp->rp->rop); |
135 |
return(NULL); |
136 |
} |
137 |
|
138 |
|
139 |
/* Compute vectors and coefficients for Hessian/gradient calcs */ |
140 |
static void |
141 |
comp_fftri(FFTRI *ftp, FVECT ap0, FVECT ap1, FVECT rop) |
142 |
{ |
143 |
FVECT vcp; |
144 |
double dot_e, dot_er, rdot_r, rdot_r1, J2; |
145 |
int i; |
146 |
|
147 |
VSUB(ftp->r_i, ap0, rop); |
148 |
VSUB(ftp->r_i1, ap1, rop); |
149 |
VSUB(ftp->e_i, ap1, ap0); |
150 |
VCROSS(vcp, ftp->e_i, ftp->r_i); |
151 |
ftp->nf = 1.0/DOT(vcp,vcp); |
152 |
dot_e = DOT(ftp->e_i,ftp->e_i); |
153 |
dot_er = DOT(ftp->e_i, ftp->r_i); |
154 |
rdot_r = 1.0/DOT(ftp->r_i,ftp->r_i); |
155 |
rdot_r1 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
156 |
ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) * sqrt(rdot_r*rdot_r1) ) * |
157 |
sqrt( ftp->nf ); |
158 |
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r + |
159 |
dot_e*ftp->I1 )*0.5*ftp->nf; |
160 |
J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e; |
161 |
for (i = 3; i--; ) |
162 |
ftp->rI2_eJ2[i] = ftp->I2*ftp->r_i[i] + J2*ftp->e_i[i]; |
163 |
} |
164 |
|
165 |
|
166 |
/* Compose 3x3 matrix from two vectors */ |
167 |
static void |
168 |
compose_matrix(FVECT mat[3], FVECT va, FVECT vb) |
169 |
{ |
170 |
mat[0][0] = 2.0*va[0]*vb[0]; |
171 |
mat[1][1] = 2.0*va[1]*vb[1]; |
172 |
mat[2][2] = 2.0*va[2]*vb[2]; |
173 |
mat[0][1] = mat[1][0] = va[0]*vb[1] + va[1]*vb[0]; |
174 |
mat[0][2] = mat[2][0] = va[0]*vb[2] + va[2]*vb[0]; |
175 |
mat[1][2] = mat[2][1] = va[1]*vb[2] + va[2]*vb[1]; |
176 |
} |
177 |
|
178 |
|
179 |
/* Compute partial 3x3 Hessian matrix for edge */ |
180 |
static void |
181 |
comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm) |
182 |
{ |
183 |
FVECT vcp; |
184 |
FVECT m1[3], m2[3], m3[3], m4[3]; |
185 |
double d1, d2, d3, d4; |
186 |
double I3, J3, K3; |
187 |
int i, j; |
188 |
/* compute intermediate coefficients */ |
189 |
d1 = 1.0/DOT(ftp->r_i,ftp->r_i); |
190 |
d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
191 |
d3 = 1.0/DOT(ftp->e_i,ftp->e_i); |
192 |
d4 = DOT(ftp->e_i, ftp->r_i); |
193 |
I3 = 0.25*ftp->nf*( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + |
194 |
3.0/d3*ftp->I2 ); |
195 |
J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3; |
196 |
K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3); |
197 |
/* intermediate matrices */ |
198 |
VCROSS(vcp, nrm, ftp->e_i); |
199 |
compose_matrix(m1, vcp, ftp->rI2_eJ2); |
200 |
compose_matrix(m2, ftp->r_i, ftp->r_i); |
201 |
compose_matrix(m3, ftp->e_i, ftp->e_i); |
202 |
compose_matrix(m4, ftp->r_i, ftp->e_i); |
203 |
VCROSS(vcp, ftp->r_i, ftp->e_i); |
204 |
d1 = DOT(nrm, vcp); |
205 |
d2 = -d1*ftp->I2; |
206 |
d1 *= 2.0; |
207 |
for (i = 3; i--; ) /* final matrix sum */ |
208 |
for (j = 3; j--; ) { |
209 |
hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] + |
210 |
2.0*J3*m4[i][j] ); |
211 |
hess[i][j] += d2*(i==j); |
212 |
hess[i][j] *= 1.0/PI; |
213 |
} |
214 |
} |
215 |
|
216 |
|
217 |
/* Reverse hessian calculation result for edge in other direction */ |
218 |
static void |
219 |
rev_hessian(FVECT hess[3]) |
220 |
{ |
221 |
int i; |
222 |
|
223 |
for (i = 3; i--; ) { |
224 |
hess[i][0] = -hess[i][0]; |
225 |
hess[i][1] = -hess[i][1]; |
226 |
hess[i][2] = -hess[i][2]; |
227 |
} |
228 |
} |
229 |
|
230 |
|
231 |
/* Add to radiometric Hessian from the given triangle */ |
232 |
static void |
233 |
add2hessian(FVECT hess[3], FVECT ehess1[3], |
234 |
FVECT ehess2[3], FVECT ehess3[3], COLORV v) |
235 |
{ |
236 |
int i, j; |
237 |
|
238 |
for (i = 3; i--; ) |
239 |
for (j = 3; j--; ) |
240 |
hess[i][j] += v*( ehess1[i][j] + ehess2[i][j] + ehess3[i][j] ); |
241 |
} |
242 |
|
243 |
|
244 |
/* Compute partial displacement form factor gradient for edge */ |
245 |
static void |
246 |
comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm) |
247 |
{ |
248 |
FVECT vcp; |
249 |
double f1; |
250 |
int i; |
251 |
|
252 |
VCROSS(vcp, ftp->r_i, ftp->r_i1); |
253 |
f1 = 2.0*DOT(nrm, vcp); |
254 |
VCROSS(vcp, nrm, ftp->e_i); |
255 |
for (i = 3; i--; ) |
256 |
grad[i] = (-0.5/PI)*( ftp->I1*vcp[i] + f1*ftp->rI2_eJ2[i] ); |
257 |
} |
258 |
|
259 |
|
260 |
/* Reverse gradient calculation result for edge in other direction */ |
261 |
static void |
262 |
rev_gradient(FVECT grad) |
263 |
{ |
264 |
grad[0] = -grad[0]; |
265 |
grad[1] = -grad[1]; |
266 |
grad[2] = -grad[2]; |
267 |
} |
268 |
|
269 |
|
270 |
/* Add to displacement gradient from the given triangle */ |
271 |
static void |
272 |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v) |
273 |
{ |
274 |
int i; |
275 |
|
276 |
for (i = 3; i--; ) |
277 |
grad[i] += v*( egrad1[i] + egrad2[i] + egrad3[i] ); |
278 |
} |
279 |
|
280 |
|
281 |
/* Return brightness of furthest ambient sample */ |
282 |
static COLORV |
283 |
back_ambval(struct s_ambsamp *ap1, struct s_ambsamp *ap2, |
284 |
struct s_ambsamp *ap3, FVECT orig) |
285 |
{ |
286 |
COLORV vback; |
287 |
FVECT vec; |
288 |
double d2, d2best; |
289 |
|
290 |
VSUB(vec, ap1->p, orig); |
291 |
d2best = DOT(vec,vec); |
292 |
vback = colval(ap1->v,CIEY); |
293 |
VSUB(vec, ap2->p, orig); |
294 |
d2 = DOT(vec,vec); |
295 |
if (d2 > d2best) { |
296 |
d2best = d2; |
297 |
vback = colval(ap2->v,CIEY); |
298 |
} |
299 |
VSUB(vec, ap3->p, orig); |
300 |
d2 = DOT(vec,vec); |
301 |
if (d2 > d2best) |
302 |
return(colval(ap3->v,CIEY)); |
303 |
return(vback); |
304 |
} |
305 |
|
306 |
|
307 |
/* Compute anisotropic radii and eigenvector directions */ |
308 |
static int |
309 |
eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3]) |
310 |
{ |
311 |
double hess2[2][2]; |
312 |
FVECT a, b; |
313 |
double evalue[2], slope1, xmag1; |
314 |
int i; |
315 |
/* project Hessian to sample plane */ |
316 |
for (i = 3; i--; ) { |
317 |
a[i] = DOT(hessian[i], uv[0]); |
318 |
b[i] = DOT(hessian[i], uv[1]); |
319 |
} |
320 |
hess2[0][0] = DOT(uv[0], a); |
321 |
hess2[0][1] = DOT(uv[0], b); |
322 |
hess2[1][0] = DOT(uv[1], a); |
323 |
hess2[1][1] = DOT(uv[1], b); |
324 |
/* compute eigenvalues */ |
325 |
if ( quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1], |
326 |
hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]) != 2 || |
327 |
(evalue[0] = fabs(evalue[0])) <= FTINY*FTINY || |
328 |
(evalue[1] = fabs(evalue[1])) <= FTINY*FTINY ) |
329 |
error(INTERNAL, "bad eigenvalue calculation"); |
330 |
|
331 |
if (evalue[0] > evalue[1]) { |
332 |
ra[0] = sqrt(sqrt(4.0/evalue[0])); |
333 |
ra[1] = sqrt(sqrt(4.0/evalue[1])); |
334 |
slope1 = evalue[1]; |
335 |
} else { |
336 |
ra[0] = sqrt(sqrt(4.0/evalue[1])); |
337 |
ra[1] = sqrt(sqrt(4.0/evalue[0])); |
338 |
slope1 = evalue[0]; |
339 |
} |
340 |
/* compute unit eigenvectors */ |
341 |
if (fabs(hess2[0][1]) <= FTINY) |
342 |
return; /* uv OK as is */ |
343 |
slope1 = (slope1 - hess2[0][0]) / hess2[0][1]; |
344 |
xmag1 = sqrt(1.0/(1.0 + slope1*slope1)); |
345 |
for (i = 3; i--; ) { |
346 |
b[i] = xmag1*uv[0][i] + slope1*xmag1*uv[1][i]; |
347 |
a[i] = slope1*xmag1*uv[0][i] - xmag1*uv[1][i]; |
348 |
} |
349 |
VCOPY(uv[0], a); |
350 |
VCOPY(uv[1], b); |
351 |
} |
352 |
|
353 |
|
354 |
static void |
355 |
ambHessian( /* anisotropic radii & pos. gradient */ |
356 |
AMBHEMI *hp, |
357 |
FVECT uv[2], /* returned */ |
358 |
float ra[2], /* returned (optional) */ |
359 |
float pg[2] /* returned (optional) */ |
360 |
) |
361 |
{ |
362 |
static char memerrmsg[] = "out of memory in ambHessian()"; |
363 |
FVECT (*hessrow)[3] = NULL; |
364 |
FVECT *gradrow = NULL; |
365 |
FVECT hessian[3]; |
366 |
FVECT gradient; |
367 |
FFTRI fftr; |
368 |
int i, j; |
369 |
/* be sure to assign unit vectors */ |
370 |
VCOPY(uv[0], hp->ux); |
371 |
VCOPY(uv[1], hp->uy); |
372 |
/* clock-wise vertex traversal from sample POV */ |
373 |
if (ra != NULL) { /* initialize Hessian row buffer */ |
374 |
hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1)); |
375 |
if (hessrow == NULL) |
376 |
error(SYSTEM, memerrmsg); |
377 |
memset(hessian, 0, sizeof(hessian)); |
378 |
} else if (pg == NULL) /* bogus call? */ |
379 |
return; |
380 |
if (pg != NULL) { /* initialize form factor row buffer */ |
381 |
gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1)); |
382 |
if (gradrow == NULL) |
383 |
error(SYSTEM, memerrmsg); |
384 |
memset(gradient, 0, sizeof(gradient)); |
385 |
} |
386 |
/* compute first row of edges */ |
387 |
for (j = 0; j < hp->ns-1; j++) { |
388 |
comp_fftri(&fftr, ambsamp(hp,0,j).p, |
389 |
ambsamp(hp,0,j+1).p, hp->rp->rop); |
390 |
if (hessrow != NULL) |
391 |
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
392 |
if (gradrow != NULL) |
393 |
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
394 |
} |
395 |
/* sum each row of triangles */ |
396 |
for (i = 0; i < hp->ns-1; i++) { |
397 |
FVECT hesscol[3]; /* compute first vertical edge */ |
398 |
FVECT gradcol; |
399 |
comp_fftri(&fftr, ambsamp(hp,i,0).p, |
400 |
ambsamp(hp,i+1,0).p, hp->rp->rop); |
401 |
if (hessrow != NULL) |
402 |
comp_hessian(hesscol, &fftr, hp->rp->ron); |
403 |
if (gradrow != NULL) |
404 |
comp_gradient(gradcol, &fftr, hp->rp->ron); |
405 |
for (j = 0; j < hp->ns-1; j++) { |
406 |
FVECT hessdia[3]; /* compute triangle contributions */ |
407 |
FVECT graddia; |
408 |
COLORV backg; |
409 |
backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1), |
410 |
&ambsamp(hp,i+1,j), hp->rp->rop); |
411 |
/* diagonal (inner) edge */ |
412 |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, |
413 |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
414 |
if (hessrow != NULL) { |
415 |
comp_hessian(hessdia, &fftr, hp->rp->ron); |
416 |
rev_hessian(hesscol); |
417 |
add2hessian(hessian, hessrow[j], hessdia, hesscol, backg); |
418 |
} |
419 |
if (gradient != NULL) { |
420 |
comp_gradient(graddia, &fftr, hp->rp->ron); |
421 |
rev_gradient(gradcol); |
422 |
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
423 |
} |
424 |
/* initialize edge in next row */ |
425 |
comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p, |
426 |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
427 |
if (hessrow != NULL) |
428 |
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
429 |
if (gradrow != NULL) |
430 |
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
431 |
/* new column edge & paired triangle */ |
432 |
backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1), |
433 |
&ambsamp(hp,i+1,j), hp->rp->rop); |
434 |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p, |
435 |
hp->rp->rop); |
436 |
if (hessrow != NULL) { |
437 |
comp_hessian(hesscol, &fftr, hp->rp->ron); |
438 |
rev_hessian(hessdia); |
439 |
add2hessian(hessian, hessrow[j], hessdia, hesscol, backg); |
440 |
if (i < hp->ns-2) |
441 |
rev_hessian(hessrow[j]); |
442 |
} |
443 |
if (gradrow != NULL) { |
444 |
comp_gradient(gradcol, &fftr, hp->rp->ron); |
445 |
rev_gradient(graddia); |
446 |
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
447 |
if (i < hp->ns-2) |
448 |
rev_gradient(gradrow[j]); |
449 |
} |
450 |
} |
451 |
} |
452 |
/* release row buffers */ |
453 |
if (hessrow != NULL) free(hessrow); |
454 |
if (gradrow != NULL) free(gradrow); |
455 |
|
456 |
if (ra != NULL) /* extract eigenvectors & radii */ |
457 |
eigenvectors(uv, ra, hessian); |
458 |
if (pg != NULL) { /* tangential position gradient */ |
459 |
pg[0] = DOT(gradient, uv[0]); |
460 |
pg[1] = DOT(gradient, uv[1]); |
461 |
} |
462 |
} |
463 |
|
464 |
|
465 |
/* Compute direction gradient from a hemispherical sampling */ |
466 |
static void |
467 |
ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2]) |
468 |
{ |
469 |
struct s_ambsamp *ap; |
470 |
double dgsum[2]; |
471 |
int n; |
472 |
FVECT vd; |
473 |
double gfact; |
474 |
|
475 |
dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */ |
476 |
for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) { |
477 |
/* use vector for azimuth + 90deg */ |
478 |
VSUB(vd, ap->p, hp->rp->rop); |
479 |
/* brightness over cosine factor */ |
480 |
gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd); |
481 |
/* -sine = -proj_radius/vd_length */ |
482 |
dgsum[0] += DOT(uv[1], vd) * gfact; |
483 |
dgsum[1] -= DOT(uv[0], vd) * gfact; |
484 |
} |
485 |
dg[0] = dgsum[0] / (hp->ns*hp->ns); |
486 |
dg[1] = dgsum[1] / (hp->ns*hp->ns); |
487 |
} |
488 |
|
489 |
|
490 |
int |
491 |
doambient( /* compute ambient component */ |
492 |
COLOR rcol, /* input/output color */ |
493 |
RAY *r, |
494 |
double wt, |
495 |
FVECT uv[2], /* returned (optional) */ |
496 |
float ra[2], /* returned (optional) */ |
497 |
float pg[2], /* returned (optional) */ |
498 |
float dg[2] /* returned (optional) */ |
499 |
) |
500 |
{ |
501 |
AMBHEMI *hp = inithemi(rcol, r, wt); |
502 |
int cnt = 0; |
503 |
FVECT my_uv[2]; |
504 |
double d, acol[3]; |
505 |
struct s_ambsamp *ap; |
506 |
int i, j; |
507 |
/* check/initialize */ |
508 |
if (hp == NULL) |
509 |
return(0); |
510 |
if (uv != NULL) |
511 |
memset(uv, 0, sizeof(FVECT)*2); |
512 |
if (ra != NULL) |
513 |
ra[0] = ra[1] = 0.0; |
514 |
if (pg != NULL) |
515 |
pg[0] = pg[1] = 0.0; |
516 |
if (dg != NULL) |
517 |
dg[0] = dg[1] = 0.0; |
518 |
/* sample the hemisphere */ |
519 |
acol[0] = acol[1] = acol[2] = 0.0; |
520 |
for (i = hp->ns; i--; ) |
521 |
for (j = hp->ns; j--; ) |
522 |
if ((ap = ambsample(hp, i, j)) != NULL) { |
523 |
addcolor(acol, ap->v); |
524 |
++cnt; |
525 |
} |
526 |
if (!cnt) { |
527 |
setcolor(rcol, 0.0, 0.0, 0.0); |
528 |
free(hp); |
529 |
return(0); /* no valid samples */ |
530 |
} |
531 |
copycolor(rcol, acol); /* final indirect irradiance/PI */ |
532 |
if (cnt < hp->ns*hp->ns || /* incomplete sampling? */ |
533 |
(ra == NULL) & (pg == NULL) & (dg == NULL)) { |
534 |
free(hp); |
535 |
return(-1); /* no radius or gradient calc. */ |
536 |
} |
537 |
if (bright(acol) > FTINY) /* normalize Y values */ |
538 |
d = cnt/bright(acol); |
539 |
else |
540 |
d = 0.0; |
541 |
ap = hp->sa; /* relative Y channel from here on... */ |
542 |
for (i = hp->ns*hp->ns; i--; ap++) |
543 |
colval(ap->v,CIEY) = bright(ap->v)*d + 0.01; |
544 |
|
545 |
if (uv == NULL) /* make sure we have axis pointers */ |
546 |
uv = my_uv; |
547 |
/* compute radii & pos. gradient */ |
548 |
ambHessian(hp, uv, ra, pg); |
549 |
|
550 |
if (dg != NULL) /* compute direction gradient */ |
551 |
ambdirgrad(hp, uv, dg); |
552 |
|
553 |
if (ra != NULL) { /* scale/clamp radii */ |
554 |
if (ra[0] < minarad) { |
555 |
ra[0] = minarad; |
556 |
if (ra[1] < minarad) |
557 |
ra[1] = minarad; |
558 |
} |
559 |
ra[0] *= d = 1.0/sqrt(sqrt(wt)); |
560 |
if ((ra[1] *= d) > 2.0*ra[0]) |
561 |
ra[1] = 2.0*ra[0]; |
562 |
if (ra[1] > maxarad) { |
563 |
ra[1] = maxarad; |
564 |
if (ra[0] > maxarad) |
565 |
ra[0] = maxarad; |
566 |
} |
567 |
} |
568 |
free(hp); /* clean up and return */ |
569 |
return(1); |
570 |
} |
571 |
|
572 |
|
573 |
#else /* ! NEWAMB */ |
574 |
|
575 |
|
576 |
void |
577 |
inithemi( /* initialize sampling hemisphere */ |
578 |
AMBHEMI *hp, |
579 |
COLOR ac, |
580 |
RAY *r, |
581 |
double wt |
582 |
) |
583 |
{ |
584 |
double d; |
585 |
int i; |
586 |
/* set number of divisions */ |
587 |
if (ambacc <= FTINY && |
588 |
wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight))) |
589 |
wt = d; /* avoid ray termination */ |
590 |
hp->nt = sqrt(ambdiv * wt / PI) + 0.5; |
591 |
i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */ |
592 |
if (hp->nt < i) |
593 |
hp->nt = i; |
594 |
hp->np = PI * hp->nt + 0.5; |
595 |
/* set number of super-samples */ |
596 |
hp->ns = ambssamp * wt + 0.5; |
597 |
/* assign coefficient */ |
598 |
copycolor(hp->acoef, ac); |
599 |
d = 1.0/(hp->nt*hp->np); |
600 |
scalecolor(hp->acoef, d); |
601 |
/* make axes */ |
602 |
VCOPY(hp->uz, r->ron); |
603 |
hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0; |
604 |
for (i = 0; i < 3; i++) |
605 |
if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6) |
606 |
break; |
607 |
if (i >= 3) |
608 |
error(CONSISTENCY, "bad ray direction in inithemi"); |
609 |
hp->uy[i] = 1.0; |
610 |
fcross(hp->ux, hp->uy, hp->uz); |
611 |
normalize(hp->ux); |
612 |
fcross(hp->uy, hp->uz, hp->ux); |
613 |
} |
614 |
|
615 |
|
616 |
int |
617 |
divsample( /* sample a division */ |
618 |
AMBSAMP *dp, |
619 |
AMBHEMI *h, |
620 |
RAY *r |
621 |
) |
622 |
{ |
623 |
RAY ar; |
624 |
int hlist[3]; |
625 |
double spt[2]; |
626 |
double xd, yd, zd; |
627 |
double b2; |
628 |
double phi; |
629 |
int i; |
630 |
/* ambient coefficient for weight */ |
631 |
if (ambacc > FTINY) |
632 |
setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL); |
633 |
else |
634 |
copycolor(ar.rcoef, h->acoef); |
635 |
if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0) |
636 |
return(-1); |
637 |
if (ambacc > FTINY) { |
638 |
multcolor(ar.rcoef, h->acoef); |
639 |
scalecolor(ar.rcoef, 1./AVGREFL); |
640 |
} |
641 |
hlist[0] = r->rno; |
642 |
hlist[1] = dp->t; |
643 |
hlist[2] = dp->p; |
644 |
multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n)); |
645 |
zd = sqrt((dp->t + spt[0])/h->nt); |
646 |
phi = 2.0*PI * (dp->p + spt[1])/h->np; |
647 |
xd = tcos(phi) * zd; |
648 |
yd = tsin(phi) * zd; |
649 |
zd = sqrt(1.0 - zd*zd); |
650 |
for (i = 0; i < 3; i++) |
651 |
ar.rdir[i] = xd*h->ux[i] + |
652 |
yd*h->uy[i] + |
653 |
zd*h->uz[i]; |
654 |
checknorm(ar.rdir); |
655 |
dimlist[ndims++] = dp->t*h->np + dp->p + 90171; |
656 |
rayvalue(&ar); |
657 |
ndims--; |
658 |
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
659 |
addcolor(dp->v, ar.rcol); |
660 |
/* use rt to improve gradient calc */ |
661 |
if (ar.rt > FTINY && ar.rt < FHUGE) |
662 |
dp->r += 1.0/ar.rt; |
663 |
/* (re)initialize error */ |
664 |
if (dp->n++) { |
665 |
b2 = bright(dp->v)/dp->n - bright(ar.rcol); |
666 |
b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1)); |
667 |
dp->k = b2/(dp->n*dp->n); |
668 |
} else |
669 |
dp->k = 0.0; |
670 |
return(0); |
671 |
} |
672 |
|
673 |
|
674 |
static int |
675 |
ambcmp( /* decreasing order */ |
676 |
const void *p1, |
677 |
const void *p2 |
678 |
) |
679 |
{ |
680 |
const AMBSAMP *d1 = (const AMBSAMP *)p1; |
681 |
const AMBSAMP *d2 = (const AMBSAMP *)p2; |
682 |
|
683 |
if (d1->k < d2->k) |
684 |
return(1); |
685 |
if (d1->k > d2->k) |
686 |
return(-1); |
687 |
return(0); |
688 |
} |
689 |
|
690 |
|
691 |
static int |
692 |
ambnorm( /* standard order */ |
693 |
const void *p1, |
694 |
const void *p2 |
695 |
) |
696 |
{ |
697 |
const AMBSAMP *d1 = (const AMBSAMP *)p1; |
698 |
const AMBSAMP *d2 = (const AMBSAMP *)p2; |
699 |
int c; |
700 |
|
701 |
if ( (c = d1->t - d2->t) ) |
702 |
return(c); |
703 |
return(d1->p - d2->p); |
704 |
} |
705 |
|
706 |
|
707 |
double |
708 |
doambient( /* compute ambient component */ |
709 |
COLOR rcol, |
710 |
RAY *r, |
711 |
double wt, |
712 |
FVECT pg, |
713 |
FVECT dg |
714 |
) |
715 |
{ |
716 |
double b, d=0; |
717 |
AMBHEMI hemi; |
718 |
AMBSAMP *div; |
719 |
AMBSAMP dnew; |
720 |
double acol[3]; |
721 |
AMBSAMP *dp; |
722 |
double arad; |
723 |
int divcnt; |
724 |
int i, j; |
725 |
/* initialize hemisphere */ |
726 |
inithemi(&hemi, rcol, r, wt); |
727 |
divcnt = hemi.nt * hemi.np; |
728 |
/* initialize */ |
729 |
if (pg != NULL) |
730 |
pg[0] = pg[1] = pg[2] = 0.0; |
731 |
if (dg != NULL) |
732 |
dg[0] = dg[1] = dg[2] = 0.0; |
733 |
setcolor(rcol, 0.0, 0.0, 0.0); |
734 |
if (divcnt == 0) |
735 |
return(0.0); |
736 |
/* allocate super-samples */ |
737 |
if (hemi.ns > 0 || pg != NULL || dg != NULL) { |
738 |
div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP)); |
739 |
if (div == NULL) |
740 |
error(SYSTEM, "out of memory in doambient"); |
741 |
} else |
742 |
div = NULL; |
743 |
/* sample the divisions */ |
744 |
arad = 0.0; |
745 |
acol[0] = acol[1] = acol[2] = 0.0; |
746 |
if ((dp = div) == NULL) |
747 |
dp = &dnew; |
748 |
divcnt = 0; |
749 |
for (i = 0; i < hemi.nt; i++) |
750 |
for (j = 0; j < hemi.np; j++) { |
751 |
dp->t = i; dp->p = j; |
752 |
setcolor(dp->v, 0.0, 0.0, 0.0); |
753 |
dp->r = 0.0; |
754 |
dp->n = 0; |
755 |
if (divsample(dp, &hemi, r) < 0) { |
756 |
if (div != NULL) |
757 |
dp++; |
758 |
continue; |
759 |
} |
760 |
arad += dp->r; |
761 |
divcnt++; |
762 |
if (div != NULL) |
763 |
dp++; |
764 |
else |
765 |
addcolor(acol, dp->v); |
766 |
} |
767 |
if (!divcnt) { |
768 |
if (div != NULL) |
769 |
free((void *)div); |
770 |
return(0.0); /* no samples taken */ |
771 |
} |
772 |
if (divcnt < hemi.nt*hemi.np) { |
773 |
pg = dg = NULL; /* incomplete sampling */ |
774 |
hemi.ns = 0; |
775 |
} else if (arad > FTINY && divcnt/arad < minarad) { |
776 |
hemi.ns = 0; /* close enough */ |
777 |
} else if (hemi.ns > 0) { /* else perform super-sampling? */ |
778 |
comperrs(div, &hemi); /* compute errors */ |
779 |
qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */ |
780 |
/* super-sample */ |
781 |
for (i = hemi.ns; i > 0; i--) { |
782 |
dnew = *div; |
783 |
if (divsample(&dnew, &hemi, r) < 0) { |
784 |
dp++; |
785 |
continue; |
786 |
} |
787 |
dp = div; /* reinsert */ |
788 |
j = divcnt < i ? divcnt : i; |
789 |
while (--j > 0 && dnew.k < dp[1].k) { |
790 |
*dp = *(dp+1); |
791 |
dp++; |
792 |
} |
793 |
*dp = dnew; |
794 |
} |
795 |
if (pg != NULL || dg != NULL) /* restore order */ |
796 |
qsort(div, divcnt, sizeof(AMBSAMP), ambnorm); |
797 |
} |
798 |
/* compute returned values */ |
799 |
if (div != NULL) { |
800 |
arad = 0.0; /* note: divcnt may be < nt*np */ |
801 |
for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) { |
802 |
arad += dp->r; |
803 |
if (dp->n > 1) { |
804 |
b = 1.0/dp->n; |
805 |
scalecolor(dp->v, b); |
806 |
dp->r *= b; |
807 |
dp->n = 1; |
808 |
} |
809 |
addcolor(acol, dp->v); |
810 |
} |
811 |
b = bright(acol); |
812 |
if (b > FTINY) { |
813 |
b = 1.0/b; /* compute & normalize gradient(s) */ |
814 |
if (pg != NULL) { |
815 |
posgradient(pg, div, &hemi); |
816 |
for (i = 0; i < 3; i++) |
817 |
pg[i] *= b; |
818 |
} |
819 |
if (dg != NULL) { |
820 |
dirgradient(dg, div, &hemi); |
821 |
for (i = 0; i < 3; i++) |
822 |
dg[i] *= b; |
823 |
} |
824 |
} |
825 |
free((void *)div); |
826 |
} |
827 |
copycolor(rcol, acol); |
828 |
if (arad <= FTINY) |
829 |
arad = maxarad; |
830 |
else |
831 |
arad = (divcnt+hemi.ns)/arad; |
832 |
if (pg != NULL) { /* reduce radius if gradient large */ |
833 |
d = DOT(pg,pg); |
834 |
if (d*arad*arad > 1.0) |
835 |
arad = 1.0/sqrt(d); |
836 |
} |
837 |
if (arad < minarad) { |
838 |
arad = minarad; |
839 |
if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */ |
840 |
d = 1.0/arad/sqrt(d); |
841 |
for (i = 0; i < 3; i++) |
842 |
pg[i] *= d; |
843 |
} |
844 |
} |
845 |
if ((arad /= sqrt(wt)) > maxarad) |
846 |
arad = maxarad; |
847 |
return(arad); |
848 |
} |
849 |
|
850 |
|
851 |
void |
852 |
comperrs( /* compute initial error estimates */ |
853 |
AMBSAMP *da, /* assumes standard ordering */ |
854 |
AMBHEMI *hp |
855 |
) |
856 |
{ |
857 |
double b, b2; |
858 |
int i, j; |
859 |
AMBSAMP *dp; |
860 |
/* sum differences from neighbors */ |
861 |
dp = da; |
862 |
for (i = 0; i < hp->nt; i++) |
863 |
for (j = 0; j < hp->np; j++) { |
864 |
#ifdef DEBUG |
865 |
if (dp->t != i || dp->p != j) |
866 |
error(CONSISTENCY, |
867 |
"division order in comperrs"); |
868 |
#endif |
869 |
b = bright(dp[0].v); |
870 |
if (i > 0) { /* from above */ |
871 |
b2 = bright(dp[-hp->np].v) - b; |
872 |
b2 *= b2 * 0.25; |
873 |
dp[0].k += b2; |
874 |
dp[-hp->np].k += b2; |
875 |
} |
876 |
if (j > 0) { /* from behind */ |
877 |
b2 = bright(dp[-1].v) - b; |
878 |
b2 *= b2 * 0.25; |
879 |
dp[0].k += b2; |
880 |
dp[-1].k += b2; |
881 |
} else { /* around */ |
882 |
b2 = bright(dp[hp->np-1].v) - b; |
883 |
b2 *= b2 * 0.25; |
884 |
dp[0].k += b2; |
885 |
dp[hp->np-1].k += b2; |
886 |
} |
887 |
dp++; |
888 |
} |
889 |
/* divide by number of neighbors */ |
890 |
dp = da; |
891 |
for (j = 0; j < hp->np; j++) /* top row */ |
892 |
(dp++)->k *= 1.0/3.0; |
893 |
if (hp->nt < 2) |
894 |
return; |
895 |
for (i = 1; i < hp->nt-1; i++) /* central region */ |
896 |
for (j = 0; j < hp->np; j++) |
897 |
(dp++)->k *= 0.25; |
898 |
for (j = 0; j < hp->np; j++) /* bottom row */ |
899 |
(dp++)->k *= 1.0/3.0; |
900 |
} |
901 |
|
902 |
|
903 |
void |
904 |
posgradient( /* compute position gradient */ |
905 |
FVECT gv, |
906 |
AMBSAMP *da, /* assumes standard ordering */ |
907 |
AMBHEMI *hp |
908 |
) |
909 |
{ |
910 |
int i, j; |
911 |
double nextsine, lastsine, b, d; |
912 |
double mag0, mag1; |
913 |
double phi, cosp, sinp, xd, yd; |
914 |
AMBSAMP *dp; |
915 |
|
916 |
xd = yd = 0.0; |
917 |
for (j = 0; j < hp->np; j++) { |
918 |
dp = da + j; |
919 |
mag0 = mag1 = 0.0; |
920 |
lastsine = 0.0; |
921 |
for (i = 0; i < hp->nt; i++) { |
922 |
#ifdef DEBUG |
923 |
if (dp->t != i || dp->p != j) |
924 |
error(CONSISTENCY, |
925 |
"division order in posgradient"); |
926 |
#endif |
927 |
b = bright(dp->v); |
928 |
if (i > 0) { |
929 |
d = dp[-hp->np].r; |
930 |
if (dp[0].r > d) d = dp[0].r; |
931 |
/* sin(t)*cos(t)^2 */ |
932 |
d *= lastsine * (1.0 - (double)i/hp->nt); |
933 |
mag0 += d*(b - bright(dp[-hp->np].v)); |
934 |
} |
935 |
nextsine = sqrt((double)(i+1)/hp->nt); |
936 |
if (j > 0) { |
937 |
d = dp[-1].r; |
938 |
if (dp[0].r > d) d = dp[0].r; |
939 |
mag1 += d * (nextsine - lastsine) * |
940 |
(b - bright(dp[-1].v)); |
941 |
} else { |
942 |
d = dp[hp->np-1].r; |
943 |
if (dp[0].r > d) d = dp[0].r; |
944 |
mag1 += d * (nextsine - lastsine) * |
945 |
(b - bright(dp[hp->np-1].v)); |
946 |
} |
947 |
dp += hp->np; |
948 |
lastsine = nextsine; |
949 |
} |
950 |
mag0 *= 2.0*PI / hp->np; |
951 |
phi = 2.0*PI * (double)j/hp->np; |
952 |
cosp = tcos(phi); sinp = tsin(phi); |
953 |
xd += mag0*cosp - mag1*sinp; |
954 |
yd += mag0*sinp + mag1*cosp; |
955 |
} |
956 |
for (i = 0; i < 3; i++) |
957 |
gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI; |
958 |
} |
959 |
|
960 |
|
961 |
void |
962 |
dirgradient( /* compute direction gradient */ |
963 |
FVECT gv, |
964 |
AMBSAMP *da, /* assumes standard ordering */ |
965 |
AMBHEMI *hp |
966 |
) |
967 |
{ |
968 |
int i, j; |
969 |
double mag; |
970 |
double phi, xd, yd; |
971 |
AMBSAMP *dp; |
972 |
|
973 |
xd = yd = 0.0; |
974 |
for (j = 0; j < hp->np; j++) { |
975 |
dp = da + j; |
976 |
mag = 0.0; |
977 |
for (i = 0; i < hp->nt; i++) { |
978 |
#ifdef DEBUG |
979 |
if (dp->t != i || dp->p != j) |
980 |
error(CONSISTENCY, |
981 |
"division order in dirgradient"); |
982 |
#endif |
983 |
/* tan(t) */ |
984 |
mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0); |
985 |
dp += hp->np; |
986 |
} |
987 |
phi = 2.0*PI * (j+.5)/hp->np + PI/2.0; |
988 |
xd += mag * tcos(phi); |
989 |
yd += mag * tsin(phi); |
990 |
} |
991 |
for (i = 0; i < 3; i++) |
992 |
gv[i] = xd*hp->ux[i] + yd*hp->uy[i]; |
993 |
} |
994 |
|
995 |
#endif /* ! NEWAMB */ |