8 |
|
* for Irradiance Caching" by Schwarzhaupt, Wann Jensen, & Jarosz |
9 |
|
* from ACM SIGGRAPH Asia 2012 conference proceedings. |
10 |
|
* |
11 |
+ |
* 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 |
|
* Declarations of external symbols in ambient.h |
16 |
|
*/ |
17 |
|
|
25 |
|
|
26 |
|
extern void SDsquare2disk(double ds[2], double seedx, double seedy); |
27 |
|
|
28 |
+ |
/* vertex direction bit positions */ |
29 |
+ |
#define VDB_xy 0 |
30 |
+ |
#define VDB_y 01 |
31 |
+ |
#define VDB_x 02 |
32 |
+ |
#define VDB_Xy 03 |
33 |
+ |
#define VDB_xY 04 |
34 |
+ |
#define VDB_X 05 |
35 |
+ |
#define VDB_Y 06 |
36 |
+ |
#define VDB_XY 07 |
37 |
+ |
/* get opposite vertex direction bit */ |
38 |
+ |
#define VDB_OPP(f) (~(f) & 07) |
39 |
+ |
/* adjacent triangle vertex flags */ |
40 |
+ |
static const int adjacent_trifl[8] = { |
41 |
+ |
0, /* forbidden diagonal */ |
42 |
+ |
1<<VDB_x|1<<VDB_y|1<<VDB_Xy, |
43 |
+ |
1<<VDB_y|1<<VDB_x|1<<VDB_xY, |
44 |
+ |
1<<VDB_y|1<<VDB_Xy|1<<VDB_X, |
45 |
+ |
1<<VDB_x|1<<VDB_xY|1<<VDB_Y, |
46 |
+ |
1<<VDB_Xy|1<<VDB_X|1<<VDB_Y, |
47 |
+ |
1<<VDB_xY|1<<VDB_Y|1<<VDB_X, |
48 |
+ |
0, /* forbidden diagonal */ |
49 |
+ |
}; |
50 |
+ |
|
51 |
|
typedef struct { |
52 |
|
COLOR v; /* hemisphere sample value */ |
53 |
+ |
float d; /* reciprocal distance (1/rt) */ |
54 |
|
FVECT p; /* intersection point */ |
55 |
|
} AMBSAMP; /* sample value */ |
56 |
|
|
62 |
|
AMBSAMP sa[1]; /* sample array (extends struct) */ |
63 |
|
} AMBHEMI; /* ambient sample hemisphere */ |
64 |
|
|
65 |
< |
#define ambsam(h,i,j) (h)->sa[(i)*(h)->ns + (j)] |
65 |
> |
#define ambndx(h,i,j) ((i)*(h)->ns + (j)) |
66 |
> |
#define ambsam(h,i,j) (h)->sa[ambndx(h,i,j)] |
67 |
|
|
68 |
|
typedef struct { |
69 |
|
FVECT r_i, r_i1, e_i, rcp, rI2_eJ2; |
70 |
|
double I1, I2; |
71 |
+ |
int valid; |
72 |
|
} FFTRI; /* vectors and coefficients for Hessian calculation */ |
73 |
|
|
74 |
|
|
75 |
+ |
/* Get index for adjacent vertex */ |
76 |
+ |
static int |
77 |
+ |
adjacent_verti(AMBHEMI *hp, int i, int j, int dbit) |
78 |
+ |
{ |
79 |
+ |
int i0 = i*hp->ns + j; |
80 |
+ |
|
81 |
+ |
switch (dbit) { |
82 |
+ |
case VDB_y: return(i0 - hp->ns); |
83 |
+ |
case VDB_x: return(i0 - 1); |
84 |
+ |
case VDB_Xy: return(i0 - hp->ns + 1); |
85 |
+ |
case VDB_xY: return(i0 + hp->ns - 1); |
86 |
+ |
case VDB_X: return(i0 + 1); |
87 |
+ |
case VDB_Y: return(i0 + hp->ns); |
88 |
+ |
/* the following should never occur */ |
89 |
+ |
case VDB_xy: return(i0 - hp->ns - 1); |
90 |
+ |
case VDB_XY: return(i0 + hp->ns + 1); |
91 |
+ |
} |
92 |
+ |
return(-1); |
93 |
+ |
} |
94 |
+ |
|
95 |
+ |
|
96 |
+ |
/* Get vertex direction bit for the opposite edge to complete triangle */ |
97 |
+ |
static int |
98 |
+ |
vdb_edge(int db1, int db2) |
99 |
+ |
{ |
100 |
+ |
switch (db1) { |
101 |
+ |
case VDB_x: return(db2==VDB_y ? VDB_Xy : VDB_Y); |
102 |
+ |
case VDB_y: return(db2==VDB_x ? VDB_xY : VDB_X); |
103 |
+ |
case VDB_X: return(db2==VDB_Xy ? VDB_y : VDB_xY); |
104 |
+ |
case VDB_Y: return(db2==VDB_xY ? VDB_x : VDB_Xy); |
105 |
+ |
case VDB_xY: return(db2==VDB_x ? VDB_y : VDB_X); |
106 |
+ |
case VDB_Xy: return(db2==VDB_y ? VDB_x : VDB_Y); |
107 |
+ |
} |
108 |
+ |
error(CONSISTENCY, "forbidden diagonal in vdb_edge()"); |
109 |
+ |
return(-1); |
110 |
+ |
} |
111 |
+ |
|
112 |
+ |
|
113 |
|
static AMBHEMI * |
114 |
|
inithemi( /* initialize sampling hemisphere */ |
115 |
|
COLOR ac, |
174 |
|
scalecolor(arp->rcoef, 1./AVGREFL); |
175 |
|
} |
176 |
|
hlist[0] = hp->rp->rno; |
177 |
< |
hlist[1] = i; |
178 |
< |
hlist[2] = j; |
177 |
> |
hlist[1] = j; |
178 |
> |
hlist[2] = i; |
179 |
|
multisamp(spt, 2, urand(ilhash(hlist,3)+n)); |
180 |
|
if (!n) { /* avoid border samples for n==0 */ |
181 |
< |
if ((spt[0] < 0.1) | (spt[0] > 0.9)) |
181 |
> |
if ((spt[0] < 0.1) | (spt[0] >= 0.9)) |
182 |
|
spt[0] = 0.1 + 0.8*frandom(); |
183 |
< |
if ((spt[1] < 0.1) | (spt[1] > 0.9)) |
183 |
> |
if ((spt[1] < 0.1) | (spt[1] >= 0.9)) |
184 |
|
spt[1] = 0.1 + 0.8*frandom(); |
185 |
|
} |
186 |
< |
SDsquare2disk(spt, (i+spt[0])/hp->ns, (j+spt[1])/hp->ns); |
186 |
> |
SDsquare2disk(spt, (j+spt[1])/hp->ns, (i+spt[0])/hp->ns); |
187 |
|
zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]); |
188 |
|
for (ii = 3; ii--; ) |
189 |
|
arp->rdir[ii] = spt[0]*hp->ux[ii] + |
190 |
|
spt[1]*hp->uy[ii] + |
191 |
|
zd*hp->rp->ron[ii]; |
192 |
|
checknorm(arp->rdir); |
193 |
< |
dimlist[ndims++] = i*hp->ns + j + 90171; |
193 |
> |
dimlist[ndims++] = ambndx(hp,i,j) + 90171; |
194 |
|
rayvalue(arp); /* evaluate ray */ |
195 |
|
ndims--; /* apply coefficient */ |
196 |
|
multcolor(arp->rcol, arp->rcoef); |
208 |
|
AMBSAMP *ap = &ambsam(hp,i,j); |
209 |
|
RAY ar; |
210 |
|
/* generate hemispherical sample */ |
211 |
< |
if (!getambsamp(&ar, hp, i, j, 0)) |
212 |
< |
goto badsample; |
213 |
< |
/* limit vertex distance */ |
211 |
> |
if (!getambsamp(&ar, hp, i, j, 0) || ar.rt <= FTINY) { |
212 |
> |
memset(ap, 0, sizeof(AMBSAMP)); |
213 |
> |
return(NULL); |
214 |
> |
} |
215 |
> |
ap->d = 1.0/ar.rt; /* limit vertex distance */ |
216 |
|
if (ar.rt > 10.0*thescene.cusize) |
217 |
|
ar.rt = 10.0*thescene.cusize; |
148 |
– |
else if (ar.rt <= FTINY) /* should never happen! */ |
149 |
– |
goto badsample; |
218 |
|
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
219 |
|
copycolor(ap->v, ar.rcol); |
220 |
|
return(ap); |
153 |
– |
badsample: |
154 |
– |
setcolor(ap->v, 0., 0., 0.); |
155 |
– |
VCOPY(ap->p, hp->rp->rop); |
156 |
– |
return(NULL); |
221 |
|
} |
222 |
|
|
223 |
|
|
225 |
|
static float * |
226 |
|
getambdiffs(AMBHEMI *hp) |
227 |
|
{ |
228 |
< |
float *earr = calloc(hp->ns*hp->ns, sizeof(float)); |
228 |
> |
float *earr = (float *)malloc(sizeof(float)*hp->ns*hp->ns); |
229 |
|
float *ep; |
230 |
|
AMBSAMP *ap; |
231 |
|
double b, d2; |
236 |
|
/* compute squared neighbor diffs */ |
237 |
|
for (ap = hp->sa, ep = earr, i = 0; i < hp->ns; i++) |
238 |
|
for (j = 0; j < hp->ns; j++, ap++, ep++) { |
239 |
+ |
ep[0] = FTINY; |
240 |
|
b = bright(ap[0].v); |
241 |
|
if (i) { /* from above */ |
242 |
|
d2 = b - bright(ap[-hp->ns].v); |
273 |
|
ambsupersamp(double acol[3], AMBHEMI *hp, int cnt) |
274 |
|
{ |
275 |
|
float *earr = getambdiffs(hp); |
276 |
< |
double e2sum = 0; |
276 |
> |
double e2rem = 0; |
277 |
|
AMBSAMP *ap; |
278 |
|
RAY ar; |
279 |
< |
COLOR asum; |
279 |
> |
double asum[3]; |
280 |
|
float *ep; |
281 |
|
int i, j, n; |
282 |
|
|
283 |
|
if (earr == NULL) /* just skip calc. if no memory */ |
284 |
|
return; |
285 |
< |
/* add up estimated variances */ |
285 |
> |
/* accumulate estimated variances */ |
286 |
|
for (ep = earr + hp->ns*hp->ns; ep-- > earr; ) |
287 |
< |
e2sum += *ep; |
287 |
> |
e2rem += *ep; |
288 |
|
ep = earr; /* perform super-sampling */ |
289 |
|
for (ap = hp->sa, i = 0; i < hp->ns; i++) |
290 |
|
for (j = 0; j < hp->ns; j++, ap++) { |
291 |
< |
int nss = *ep/e2sum*cnt + frandom(); |
292 |
< |
setcolor(asum, 0., 0., 0.); |
291 |
> |
int nss = *ep/e2rem*cnt + frandom(); |
292 |
> |
asum[0] = asum[1] = asum[2] = 0.0; |
293 |
|
for (n = 1; n <= nss; n++) { |
294 |
|
if (!getambsamp(&ar, hp, i, j, n)) { |
295 |
|
nss = n-1; |
298 |
|
addcolor(asum, ar.rcol); |
299 |
|
} |
300 |
|
if (nss) { /* update returned ambient value */ |
301 |
< |
const double ssf = 1./(nss + 1); |
301 |
> |
const double ssf = 1./(nss + 1.); |
302 |
|
for (n = 3; n--; ) |
303 |
< |
acol[n] += ssf*colval(asum,n) + |
303 |
> |
acol[n] += ssf*asum[n] + |
304 |
|
(ssf - 1.)*colval(ap->v,n); |
305 |
|
} |
306 |
< |
e2sum -= *ep++; /* update remainders */ |
306 |
> |
e2rem -= *ep++; /* update remainders */ |
307 |
|
cnt -= nss; |
308 |
|
} |
309 |
|
free(earr); |
310 |
|
} |
311 |
|
|
312 |
|
|
313 |
+ |
/* Compute vertex flags, indicating farthest in each direction */ |
314 |
+ |
static uby8 * |
315 |
+ |
vertex_flags(AMBHEMI *hp) |
316 |
+ |
{ |
317 |
+ |
uby8 *vflags = (uby8 *)calloc(hp->ns*hp->ns, sizeof(uby8)); |
318 |
+ |
uby8 *vf; |
319 |
+ |
AMBSAMP *ap; |
320 |
+ |
int i, j; |
321 |
+ |
|
322 |
+ |
if (vflags == NULL) |
323 |
+ |
error(SYSTEM, "out of memory in vertex_flags()"); |
324 |
+ |
vf = vflags; |
325 |
+ |
ap = hp->sa; /* compute farthest along first row */ |
326 |
+ |
for (j = 0; j < hp->ns-1; j++, vf++, ap++) |
327 |
+ |
if (ap[0].d <= ap[1].d) |
328 |
+ |
vf[0] |= 1<<VDB_X; |
329 |
+ |
else |
330 |
+ |
vf[1] |= 1<<VDB_x; |
331 |
+ |
++vf; ++ap; |
332 |
+ |
/* flag subsequent rows */ |
333 |
+ |
for (i = 1; i < hp->ns; i++) { |
334 |
+ |
for (j = 0; j < hp->ns-1; j++, vf++, ap++) { |
335 |
+ |
if (ap[0].d <= ap[-hp->ns].d) /* row before */ |
336 |
+ |
vf[0] |= 1<<VDB_y; |
337 |
+ |
else |
338 |
+ |
vf[-hp->ns] |= 1<<VDB_Y; |
339 |
+ |
if (ap[0].d <= ap[1-hp->ns].d) /* diagonal we care about */ |
340 |
+ |
vf[0] |= 1<<VDB_Xy; |
341 |
+ |
else |
342 |
+ |
vf[1-hp->ns] |= 1<<VDB_xY; |
343 |
+ |
if (ap[0].d <= ap[1].d) /* column after */ |
344 |
+ |
vf[0] |= 1<<VDB_X; |
345 |
+ |
else |
346 |
+ |
vf[1] |= 1<<VDB_x; |
347 |
+ |
} |
348 |
+ |
if (ap[0].d <= ap[-hp->ns].d) /* final column edge */ |
349 |
+ |
vf[0] |= 1<<VDB_y; |
350 |
+ |
else |
351 |
+ |
vf[-hp->ns] |= 1<<VDB_Y; |
352 |
+ |
++vf; ++ap; |
353 |
+ |
} |
354 |
+ |
return(vflags); |
355 |
+ |
} |
356 |
+ |
|
357 |
+ |
|
358 |
+ |
/* Return brightness of farthest ambient sample */ |
359 |
+ |
static double |
360 |
+ |
back_ambval(AMBHEMI *hp, int i, int j, int dbit1, int dbit2, const uby8 *vflags) |
361 |
+ |
{ |
362 |
+ |
const int v0 = ambndx(hp,i,j); |
363 |
+ |
const int tflags = (1<<dbit1 | 1<<dbit2); |
364 |
+ |
int v1, v2; |
365 |
+ |
|
366 |
+ |
if ((vflags[v0] & tflags) == tflags) /* is v0 the farthest? */ |
367 |
+ |
return(colval(hp->sa[v0].v,CIEY)); |
368 |
+ |
v1 = adjacent_verti(hp, i, j, dbit1); |
369 |
+ |
if (vflags[v0] & 1<<dbit2) /* v1 farthest if v0>v2 */ |
370 |
+ |
return(colval(hp->sa[v1].v,CIEY)); |
371 |
+ |
v2 = adjacent_verti(hp, i, j, dbit2); |
372 |
+ |
if (vflags[v0] & 1<<dbit1) /* v2 farthest if v0>v1 */ |
373 |
+ |
return(colval(hp->sa[v2].v,CIEY)); |
374 |
+ |
/* else check if v1>v2 */ |
375 |
+ |
if (vflags[v1] & 1<<vdb_edge(dbit1,dbit2)) |
376 |
+ |
return(colval(hp->sa[v1].v,CIEY)); |
377 |
+ |
return(colval(hp->sa[v2].v,CIEY)); |
378 |
+ |
} |
379 |
+ |
|
380 |
+ |
|
381 |
|
/* Compute vectors and coefficients for Hessian/gradient calcs */ |
382 |
|
static void |
383 |
< |
comp_fftri(FFTRI *ftp, FVECT ap0, FVECT ap1, FVECT rop) |
383 |
> |
comp_fftri(FFTRI *ftp, AMBHEMI *hp, int i, int j, int dbit, const uby8 *vflags) |
384 |
|
{ |
385 |
< |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
386 |
< |
int i; |
385 |
> |
const int i0 = ambndx(hp,i,j); |
386 |
> |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
387 |
> |
int i1, ii; |
388 |
|
|
389 |
< |
VSUB(ftp->r_i, ap0, rop); |
390 |
< |
VSUB(ftp->r_i1, ap1, rop); |
391 |
< |
VSUB(ftp->e_i, ap1, ap0); |
389 |
> |
ftp->valid = 0; /* check if we can skip this edge */ |
390 |
> |
ii = adjacent_trifl[dbit]; |
391 |
> |
if ((vflags[i0] & ii) == ii) /* cancels if vertex used as value */ |
392 |
> |
return; |
393 |
> |
i1 = adjacent_verti(hp, i, j, dbit); |
394 |
> |
ii = adjacent_trifl[VDB_OPP(dbit)]; |
395 |
> |
if ((vflags[i1] & ii) == ii) /* on either end (for both triangles) */ |
396 |
> |
return; |
397 |
> |
/* else go ahead with calculation */ |
398 |
> |
VSUB(ftp->r_i, hp->sa[i0].p, hp->rp->rop); |
399 |
> |
VSUB(ftp->r_i1, hp->sa[i1].p, hp->rp->rop); |
400 |
> |
VSUB(ftp->e_i, hp->sa[i1].p, hp->sa[i0].p); |
401 |
|
VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1); |
402 |
|
rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp); |
403 |
|
dot_e = DOT(ftp->e_i,ftp->e_i); |
409 |
|
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r + |
410 |
|
dot_e*ftp->I1 )*0.5*rdot_cp; |
411 |
|
J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e; |
412 |
< |
for (i = 3; i--; ) |
413 |
< |
ftp->rI2_eJ2[i] = ftp->I2*ftp->r_i[i] + J2*ftp->e_i[i]; |
412 |
> |
for (ii = 3; ii--; ) |
413 |
> |
ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii]; |
414 |
> |
ftp->valid++; |
415 |
|
} |
416 |
|
|
417 |
|
|
437 |
|
double d1, d2, d3, d4; |
438 |
|
double I3, J3, K3; |
439 |
|
int i, j; |
440 |
+ |
|
441 |
+ |
if (!ftp->valid) { /* preemptive test */ |
442 |
+ |
memset(hess, 0, sizeof(FVECT)*3); |
443 |
+ |
return; |
444 |
+ |
} |
445 |
|
/* compute intermediate coefficients */ |
446 |
|
d1 = 1.0/DOT(ftp->r_i,ftp->r_i); |
447 |
|
d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
465 |
|
hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] + |
466 |
|
2.0*J3*m4[i][j] ); |
467 |
|
hess[i][j] += d2*(i==j); |
468 |
< |
hess[i][j] *= 1.0/PI; |
468 |
> |
hess[i][j] *= -1.0/PI; |
469 |
|
} |
470 |
|
} |
471 |
|
|
487 |
|
/* Add to radiometric Hessian from the given triangle */ |
488 |
|
static void |
489 |
|
add2hessian(FVECT hess[3], FVECT ehess1[3], |
490 |
< |
FVECT ehess2[3], FVECT ehess3[3], COLORV v) |
490 |
> |
FVECT ehess2[3], FVECT ehess3[3], double v) |
491 |
|
{ |
492 |
|
int i, j; |
493 |
|
|
505 |
|
double f1; |
506 |
|
int i; |
507 |
|
|
508 |
+ |
if (!ftp->valid) { /* preemptive test */ |
509 |
+ |
memset(grad, 0, sizeof(FVECT)); |
510 |
+ |
return; |
511 |
+ |
} |
512 |
|
f1 = 2.0*DOT(nrm, ftp->rcp); |
513 |
|
VCROSS(ncp, nrm, ftp->e_i); |
514 |
|
for (i = 3; i--; ) |
515 |
< |
grad[i] = (-0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
515 |
> |
grad[i] = (0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
516 |
|
} |
517 |
|
|
518 |
|
|
528 |
|
|
529 |
|
/* Add to displacement gradient from the given triangle */ |
530 |
|
static void |
531 |
< |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v) |
531 |
> |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, double v) |
532 |
|
{ |
533 |
|
int i; |
534 |
|
|
537 |
|
} |
538 |
|
|
539 |
|
|
387 |
– |
/* Return brightness of furthest ambient sample */ |
388 |
– |
static COLORV |
389 |
– |
back_ambval(AMBSAMP *ap1, AMBSAMP *ap2, AMBSAMP *ap3, FVECT orig) |
390 |
– |
{ |
391 |
– |
COLORV vback; |
392 |
– |
FVECT vec; |
393 |
– |
double d2, d2best; |
394 |
– |
|
395 |
– |
VSUB(vec, ap1->p, orig); |
396 |
– |
d2best = DOT(vec,vec); |
397 |
– |
vback = colval(ap1->v,CIEY); |
398 |
– |
VSUB(vec, ap2->p, orig); |
399 |
– |
d2 = DOT(vec,vec); |
400 |
– |
if (d2 > d2best) { |
401 |
– |
d2best = d2; |
402 |
– |
vback = colval(ap2->v,CIEY); |
403 |
– |
} |
404 |
– |
VSUB(vec, ap3->p, orig); |
405 |
– |
d2 = DOT(vec,vec); |
406 |
– |
if (d2 > d2best) |
407 |
– |
return(colval(ap3->v,CIEY)); |
408 |
– |
return(vback); |
409 |
– |
} |
410 |
– |
|
411 |
– |
|
540 |
|
/* Compute anisotropic radii and eigenvector directions */ |
541 |
< |
static int |
541 |
> |
static void |
542 |
|
eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3]) |
543 |
|
{ |
544 |
|
double hess2[2][2]; |
560 |
|
if (i == 1) /* double-root (circle) */ |
561 |
|
evalue[1] = evalue[0]; |
562 |
|
if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) | |
563 |
< |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) |
564 |
< |
error(INTERNAL, "bad eigenvalue calculation"); |
565 |
< |
|
563 |
> |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) { |
564 |
> |
ra[0] = ra[1] = maxarad; |
565 |
> |
return; |
566 |
> |
} |
567 |
|
if (evalue[0] > evalue[1]) { |
568 |
|
ra[0] = sqrt(sqrt(4.0/evalue[0])); |
569 |
|
ra[1] = sqrt(sqrt(4.0/evalue[1])); |
598 |
|
static char memerrmsg[] = "out of memory in ambHessian()"; |
599 |
|
FVECT (*hessrow)[3] = NULL; |
600 |
|
FVECT *gradrow = NULL; |
601 |
+ |
uby8 *vflags; |
602 |
|
FVECT hessian[3]; |
603 |
|
FVECT gradient; |
604 |
|
FFTRI fftr; |
620 |
|
error(SYSTEM, memerrmsg); |
621 |
|
memset(gradient, 0, sizeof(gradient)); |
622 |
|
} |
623 |
+ |
/* get vertex position flags */ |
624 |
+ |
vflags = vertex_flags(hp); |
625 |
|
/* compute first row of edges */ |
626 |
|
for (j = 0; j < hp->ns-1; j++) { |
627 |
< |
comp_fftri(&fftr, ambsam(hp,0,j).p, |
496 |
< |
ambsam(hp,0,j+1).p, hp->rp->rop); |
627 |
> |
comp_fftri(&fftr, hp, 0, j, VDB_X, vflags); |
628 |
|
if (hessrow != NULL) |
629 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
630 |
|
if (gradrow != NULL) |
634 |
|
for (i = 0; i < hp->ns-1; i++) { |
635 |
|
FVECT hesscol[3]; /* compute first vertical edge */ |
636 |
|
FVECT gradcol; |
637 |
< |
comp_fftri(&fftr, ambsam(hp,i,0).p, |
507 |
< |
ambsam(hp,i+1,0).p, hp->rp->rop); |
637 |
> |
comp_fftri(&fftr, hp, i, 0, VDB_Y, vflags); |
638 |
|
if (hessrow != NULL) |
639 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
640 |
|
if (gradrow != NULL) |
642 |
|
for (j = 0; j < hp->ns-1; j++) { |
643 |
|
FVECT hessdia[3]; /* compute triangle contributions */ |
644 |
|
FVECT graddia; |
645 |
< |
COLORV backg; |
646 |
< |
backg = back_ambval(&ambsam(hp,i,j), &ambsam(hp,i,j+1), |
517 |
< |
&ambsam(hp,i+1,j), hp->rp->rop); |
645 |
> |
double backg; |
646 |
> |
backg = back_ambval(hp, i, j, VDB_X, VDB_Y, vflags); |
647 |
|
/* diagonal (inner) edge */ |
648 |
< |
comp_fftri(&fftr, ambsam(hp,i,j+1).p, |
520 |
< |
ambsam(hp,i+1,j).p, hp->rp->rop); |
648 |
> |
comp_fftri(&fftr, hp, i, j+1, VDB_xY, vflags); |
649 |
|
if (hessrow != NULL) { |
650 |
|
comp_hessian(hessdia, &fftr, hp->rp->ron); |
651 |
|
rev_hessian(hesscol); |
657 |
|
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
658 |
|
} |
659 |
|
/* initialize edge in next row */ |
660 |
< |
comp_fftri(&fftr, ambsam(hp,i+1,j+1).p, |
533 |
< |
ambsam(hp,i+1,j).p, hp->rp->rop); |
660 |
> |
comp_fftri(&fftr, hp, i+1, j+1, VDB_x, vflags); |
661 |
|
if (hessrow != NULL) |
662 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
663 |
|
if (gradrow != NULL) |
664 |
|
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
665 |
|
/* new column edge & paired triangle */ |
666 |
< |
backg = back_ambval(&ambsam(hp,i,j+1), &ambsam(hp,i+1,j+1), |
667 |
< |
&ambsam(hp,i+1,j), hp->rp->rop); |
541 |
< |
comp_fftri(&fftr, ambsam(hp,i,j+1).p, ambsam(hp,i+1,j+1).p, |
542 |
< |
hp->rp->rop); |
666 |
> |
backg = back_ambval(hp, i+1, j+1, VDB_x, VDB_y, vflags); |
667 |
> |
comp_fftri(&fftr, hp, i, j+1, VDB_Y, vflags); |
668 |
|
if (hessrow != NULL) { |
669 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
670 |
|
rev_hessian(hessdia); |
684 |
|
/* release row buffers */ |
685 |
|
if (hessrow != NULL) free(hessrow); |
686 |
|
if (gradrow != NULL) free(gradrow); |
687 |
+ |
free(vflags); |
688 |
|
|
689 |
|
if (ra != NULL) /* extract eigenvectors & radii */ |
690 |
|
eigenvectors(uv, ra, hessian); |
720 |
|
} |
721 |
|
|
722 |
|
|
723 |
+ |
/* Compute potential light leak direction flags for cache value */ |
724 |
+ |
static uint32 |
725 |
+ |
ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1) |
726 |
+ |
{ |
727 |
+ |
const double max_d = 1.0/(minarad*ambacc + 0.001); |
728 |
+ |
const double ang_res = 0.5*PI/(hp->ns-1); |
729 |
+ |
const double ang_step = ang_res/((int)(16/PI*ang_res) + (1+FTINY)); |
730 |
+ |
double avg_d = 0; |
731 |
+ |
uint32 flgs = 0; |
732 |
+ |
int i, j; |
733 |
+ |
/* don't bother for a few samples */ |
734 |
+ |
if (hp->ns < 12) |
735 |
+ |
return(0); |
736 |
+ |
/* check distances overhead */ |
737 |
+ |
for (i = hp->ns*3/4; i-- > hp->ns>>2; ) |
738 |
+ |
for (j = hp->ns*3/4; j-- > hp->ns>>2; ) |
739 |
+ |
avg_d += ambsam(hp,i,j).d; |
740 |
+ |
avg_d *= 4.0/(hp->ns*hp->ns); |
741 |
+ |
if (avg_d*r0 >= 1.0) /* ceiling too low for corral? */ |
742 |
+ |
return(0); |
743 |
+ |
if (avg_d >= max_d) /* insurance */ |
744 |
+ |
return(0); |
745 |
+ |
/* else circle around perimeter */ |
746 |
+ |
for (i = 0; i < hp->ns; i++) |
747 |
+ |
for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) { |
748 |
+ |
AMBSAMP *ap = &ambsam(hp,i,j); |
749 |
+ |
FVECT vec; |
750 |
+ |
double u, v; |
751 |
+ |
double ang, a1; |
752 |
+ |
int abp; |
753 |
+ |
if ((ap->d <= FTINY) | (ap->d >= max_d)) |
754 |
+ |
continue; /* too far or too near */ |
755 |
+ |
VSUB(vec, ap->p, hp->rp->rop); |
756 |
+ |
u = DOT(vec, uv[0]) * ap->d; |
757 |
+ |
v = DOT(vec, uv[1]) * ap->d; |
758 |
+ |
if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= 1.0) |
759 |
+ |
continue; /* occluder outside ellipse */ |
760 |
+ |
ang = atan2a(v, u); /* else set direction flags */ |
761 |
+ |
for (a1 = ang-.5*ang_res; a1 <= ang+.5*ang_res; a1 += ang_step) |
762 |
+ |
flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0))); |
763 |
+ |
} |
764 |
+ |
return(flgs); |
765 |
+ |
} |
766 |
+ |
|
767 |
+ |
|
768 |
|
int |
769 |
|
doambient( /* compute ambient component */ |
770 |
|
COLOR rcol, /* input/output color */ |
773 |
|
FVECT uv[2], /* returned (optional) */ |
774 |
|
float ra[2], /* returned (optional) */ |
775 |
|
float pg[2], /* returned (optional) */ |
776 |
< |
float dg[2] /* returned (optional) */ |
776 |
> |
float dg[2], /* returned (optional) */ |
777 |
> |
uint32 *crlp /* returned (optional) */ |
778 |
|
) |
779 |
|
{ |
780 |
|
AMBHEMI *hp = inithemi(rcol, r, wt); |
781 |
< |
int cnt = 0; |
781 |
> |
int cnt; |
782 |
|
FVECT my_uv[2]; |
783 |
|
double d, K, acol[3]; |
784 |
|
AMBSAMP *ap; |
794 |
|
pg[0] = pg[1] = 0.0; |
795 |
|
if (dg != NULL) |
796 |
|
dg[0] = dg[1] = 0.0; |
797 |
+ |
if (crlp != NULL) |
798 |
+ |
*crlp = 0; |
799 |
|
/* sample the hemisphere */ |
800 |
|
acol[0] = acol[1] = acol[2] = 0.0; |
801 |
+ |
cnt = 0; |
802 |
|
for (i = hp->ns; i--; ) |
803 |
|
for (j = hp->ns; j--; ) |
804 |
|
if ((ap = ambsample(hp, i, j)) != NULL) { |
816 |
|
return(-1); /* return value w/o Hessian */ |
817 |
|
} |
818 |
|
cnt = ambssamp*wt + 0.5; /* perform super-sampling? */ |
819 |
< |
if (cnt > 0) |
819 |
> |
if (cnt > 8) |
820 |
|
ambsupersamp(acol, hp, cnt); |
821 |
|
copycolor(rcol, acol); /* final indirect irradiance/PI */ |
822 |
|
if ((ra == NULL) & (pg == NULL) & (dg == NULL)) { |
823 |
|
free(hp); |
824 |
|
return(-1); /* no radius or gradient calc. */ |
825 |
|
} |
826 |
< |
if (bright(acol) > FTINY) { /* normalize Y values */ |
827 |
< |
d = 0.99*cnt/bright(acol); |
826 |
> |
if ((d = bright(acol)) > FTINY) { /* normalize Y values */ |
827 |
> |
d = 0.99*(hp->ns*hp->ns)/d; |
828 |
|
K = 0.01; |
829 |
< |
} else { /* geometric Hessian fall-back */ |
655 |
< |
d = 0.0; |
829 |
> |
} else { /* or fall back on geometric Hessian */ |
830 |
|
K = 1.0; |
831 |
|
pg = NULL; |
832 |
|
dg = NULL; |
833 |
+ |
crlp = NULL; |
834 |
|
} |
835 |
|
ap = hp->sa; /* relative Y channel from here on... */ |
836 |
|
for (i = hp->ns*hp->ns; i--; ap++) |
866 |
|
if (ra[0] > maxarad) |
867 |
|
ra[0] = maxarad; |
868 |
|
} |
869 |
+ |
/* flag encroached directions */ |
870 |
+ |
if ((wt >= 0.5-FTINY) & (crlp != NULL)) |
871 |
+ |
*crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc); |
872 |
|
if (pg != NULL) { /* cap gradient if necessary */ |
873 |
|
d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1]; |
874 |
|
if (d > 1.0) { |