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 |
|
|
21 |
|
#include "ambient.h" |
22 |
|
#include "random.h" |
23 |
|
|
24 |
< |
#ifdef NEWAMB |
24 |
> |
#ifndef OLDAMB |
25 |
|
|
26 |
|
extern void SDsquare2disk(double ds[2], double seedx, double seedy); |
27 |
|
|
28 |
|
typedef struct { |
29 |
+ |
COLOR v; /* hemisphere sample value */ |
30 |
+ |
float d; /* reciprocal distance (1/rt) */ |
31 |
+ |
FVECT p; /* intersection point */ |
32 |
+ |
} AMBSAMP; /* sample value */ |
33 |
+ |
|
34 |
+ |
typedef struct { |
35 |
|
RAY *rp; /* originating ray sample */ |
26 |
– |
FVECT ux, uy; /* tangent axis unit vectors */ |
36 |
|
int ns; /* number of samples per axis */ |
37 |
+ |
int sampOK; /* acquired full sample set? */ |
38 |
|
COLOR acoef; /* division contribution coefficient */ |
39 |
< |
struct s_ambsamp { |
40 |
< |
COLOR v; /* hemisphere sample value */ |
41 |
< |
FVECT p; /* intersection point */ |
32 |
< |
} sa[1]; /* sample array (extends struct) */ |
39 |
> |
double acol[3]; /* accumulated color */ |
40 |
> |
FVECT ux, uy; /* tangent axis unit vectors */ |
41 |
> |
AMBSAMP sa[1]; /* sample array (extends struct) */ |
42 |
|
} AMBHEMI; /* ambient sample hemisphere */ |
43 |
|
|
44 |
< |
#define ambsamp(h,i,j) (h)->sa[(i)*(h)->ns + (j)] |
44 |
> |
#define AI(h,i,j) ((i)*(h)->ns + (j)) |
45 |
> |
#define ambsam(h,i,j) (h)->sa[AI(h,i,j)] |
46 |
|
|
47 |
|
typedef struct { |
48 |
|
FVECT r_i, r_i1, e_i, rcp, rI2_eJ2; |
50 |
|
} FFTRI; /* vectors and coefficients for Hessian calculation */ |
51 |
|
|
52 |
|
|
53 |
< |
static AMBHEMI * |
54 |
< |
inithemi( /* initialize sampling hemisphere */ |
55 |
< |
COLOR ac, |
56 |
< |
RAY *r, |
57 |
< |
double wt |
53 |
> |
static int |
54 |
> |
ambcollision( /* proposed direciton collides? */ |
55 |
> |
AMBHEMI *hp, |
56 |
> |
int i, |
57 |
> |
int j, |
58 |
> |
FVECT dv |
59 |
|
) |
60 |
|
{ |
61 |
< |
AMBHEMI *hp; |
62 |
< |
double d; |
63 |
< |
int n, i; |
64 |
< |
/* set number of divisions */ |
65 |
< |
if (ambacc <= FTINY && |
66 |
< |
wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight))) |
67 |
< |
wt = d; /* avoid ray termination */ |
68 |
< |
n = sqrt(ambdiv * wt) + 0.5; |
69 |
< |
i = 1 + 5*(ambacc > FTINY); /* minimum number of samples */ |
70 |
< |
if (n < i) |
71 |
< |
n = i; |
72 |
< |
/* allocate sampling array */ |
73 |
< |
hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + |
74 |
< |
sizeof(struct s_ambsamp)*(n*n - 1)); |
75 |
< |
if (hp == NULL) |
76 |
< |
return(NULL); |
77 |
< |
hp->rp = r; |
78 |
< |
hp->ns = n; |
79 |
< |
/* assign coefficient */ |
80 |
< |
copycolor(hp->acoef, ac); |
81 |
< |
d = 1.0/(n*n); |
82 |
< |
scalecolor(hp->acoef, d); |
83 |
< |
/* make tangent plane axes */ |
84 |
< |
hp->uy[0] = hp->uy[1] = hp->uy[2] = 0; |
85 |
< |
for (i = 3; i--; ) |
86 |
< |
if ((0.6 < r->ron[i]) & (r->ron[i] < 0.6)) |
76 |
< |
hp->uy[i] = 0.1+frandom(); |
77 |
< |
if (DOT(hp->uy,hp->uy) <= FTINY) |
78 |
< |
error(CONSISTENCY, "bad ray direction in inithemi()"); |
79 |
< |
VCROSS(hp->ux, hp->uy, r->ron); |
80 |
< |
normalize(hp->ux); |
81 |
< |
VCROSS(hp->uy, r->ron, hp->ux); |
82 |
< |
/* we're ready to sample */ |
83 |
< |
return(hp); |
61 |
> |
double cos_thresh; |
62 |
> |
int ii, jj; |
63 |
> |
/* min. spacing = 1/4th division */ |
64 |
> |
cos_thresh = (PI/4.)/(double)hp->ns; |
65 |
> |
cos_thresh = 1. - .5*cos_thresh*cos_thresh; |
66 |
> |
/* check existing neighbors */ |
67 |
> |
for (ii = i-1; ii <= i+1; ii++) { |
68 |
> |
if (ii < 0) continue; |
69 |
> |
if (ii >= hp->ns) break; |
70 |
> |
for (jj = j-1; jj <= j+1; jj++) { |
71 |
> |
AMBSAMP *ap; |
72 |
> |
FVECT avec; |
73 |
> |
double dprod; |
74 |
> |
if (jj < 0) continue; |
75 |
> |
if (jj >= hp->ns) break; |
76 |
> |
if ((ii==i) & (jj==j)) continue; |
77 |
> |
ap = &ambsam(hp,ii,jj); |
78 |
> |
if (ap->d <= .5/FHUGE) |
79 |
> |
continue; /* no one home */ |
80 |
> |
VSUB(avec, ap->p, hp->rp->rop); |
81 |
> |
dprod = DOT(avec, dv); |
82 |
> |
if (dprod >= cos_thresh*VLEN(avec)) |
83 |
> |
return(1); /* collision */ |
84 |
> |
} |
85 |
> |
} |
86 |
> |
return(0); /* nothing to worry about */ |
87 |
|
} |
88 |
|
|
89 |
|
|
90 |
< |
static struct s_ambsamp * |
91 |
< |
ambsample( /* sample an ambient direction */ |
90 |
> |
static int |
91 |
> |
ambsample( /* initial ambient division sample */ |
92 |
|
AMBHEMI *hp, |
93 |
|
int i, |
94 |
< |
int j |
94 |
> |
int j, |
95 |
> |
int n |
96 |
|
) |
97 |
|
{ |
98 |
< |
struct s_ambsamp *ap = &ambsamp(hp,i,j); |
99 |
< |
RAY ar; |
100 |
< |
double spt[2], zd; |
101 |
< |
int ii; |
98 |
> |
AMBSAMP *ap = &ambsam(hp,i,j); |
99 |
> |
RAY ar; |
100 |
> |
int hlist[3], ii; |
101 |
> |
double spt[2], zd; |
102 |
> |
/* generate hemispherical sample */ |
103 |
|
/* ambient coefficient for weight */ |
104 |
|
if (ambacc > FTINY) |
105 |
|
setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL); |
106 |
|
else |
107 |
|
copycolor(ar.rcoef, hp->acoef); |
108 |
|
if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0) |
109 |
< |
goto badsample; |
109 |
> |
return(0); |
110 |
|
if (ambacc > FTINY) { |
111 |
|
multcolor(ar.rcoef, hp->acoef); |
112 |
|
scalecolor(ar.rcoef, 1./AVGREFL); |
113 |
|
} |
114 |
< |
/* generate hemispherical sample */ |
115 |
< |
SDsquare2disk(spt, (i+.1+.8*frandom())/hp->ns, |
116 |
< |
(j+.1+.8*frandom())/hp->ns ); |
114 |
> |
hlist[0] = hp->rp->rno; |
115 |
> |
hlist[1] = j; |
116 |
> |
hlist[2] = i; |
117 |
> |
multisamp(spt, 2, urand(ilhash(hlist,3)+n)); |
118 |
> |
resample: |
119 |
> |
SDsquare2disk(spt, (j+spt[1])/hp->ns, (i+spt[0])/hp->ns); |
120 |
|
zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]); |
121 |
|
for (ii = 3; ii--; ) |
122 |
|
ar.rdir[ii] = spt[0]*hp->ux[ii] + |
123 |
|
spt[1]*hp->uy[ii] + |
124 |
|
zd*hp->rp->ron[ii]; |
125 |
|
checknorm(ar.rdir); |
126 |
< |
dimlist[ndims++] = i*hp->ns + j + 90171; |
126 |
> |
/* avoid coincident samples */ |
127 |
> |
if (!n && ambcollision(hp, i, j, ar.rdir)) { |
128 |
> |
spt[0] = frandom(); spt[1] = frandom(); |
129 |
> |
goto resample; /* reject this sample */ |
130 |
> |
} |
131 |
> |
dimlist[ndims++] = AI(hp,i,j) + 90171; |
132 |
|
rayvalue(&ar); /* evaluate ray */ |
133 |
|
ndims--; |
134 |
< |
/* limit vertex distance */ |
135 |
< |
if (ar.rt > 10.0*thescene.cusize) |
123 |
< |
ar.rt = 10.0*thescene.cusize; |
124 |
< |
else if (ar.rt <= FTINY) /* should never happen! */ |
125 |
< |
goto badsample; |
126 |
< |
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
134 |
> |
if (ar.rt <= FTINY) |
135 |
> |
return(0); /* should never happen */ |
136 |
|
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
137 |
< |
copycolor(ap->v, ar.rcol); |
138 |
< |
return(ap); |
139 |
< |
badsample: |
140 |
< |
setcolor(ap->v, 0., 0., 0.); |
141 |
< |
VCOPY(ap->p, hp->rp->rop); |
142 |
< |
return(NULL); |
137 |
> |
if (ar.rt*ap->d < 1.0) /* new/closer distance? */ |
138 |
> |
ap->d = 1.0/ar.rt; |
139 |
> |
if (!n) { /* record first vertex & value */ |
140 |
> |
if (ar.rt > 10.0*thescene.cusize + 1000.) |
141 |
> |
ar.rt = 10.0*thescene.cusize + 1000.; |
142 |
> |
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
143 |
> |
copycolor(ap->v, ar.rcol); |
144 |
> |
} else { /* else update recorded value */ |
145 |
> |
hp->acol[RED] -= colval(ap->v,RED); |
146 |
> |
hp->acol[GRN] -= colval(ap->v,GRN); |
147 |
> |
hp->acol[BLU] -= colval(ap->v,BLU); |
148 |
> |
zd = 1.0/(double)(n+1); |
149 |
> |
scalecolor(ar.rcol, zd); |
150 |
> |
zd *= (double)n; |
151 |
> |
scalecolor(ap->v, zd); |
152 |
> |
addcolor(ap->v, ar.rcol); |
153 |
> |
} |
154 |
> |
addcolor(hp->acol, ap->v); /* add to our sum */ |
155 |
> |
return(1); |
156 |
|
} |
157 |
|
|
158 |
|
|
159 |
+ |
/* Estimate errors based on ambient division differences */ |
160 |
+ |
static float * |
161 |
+ |
getambdiffs(AMBHEMI *hp) |
162 |
+ |
{ |
163 |
+ |
float *earr = (float *)calloc(hp->ns*hp->ns, sizeof(float)); |
164 |
+ |
float *ep; |
165 |
+ |
AMBSAMP *ap; |
166 |
+ |
double b, d2; |
167 |
+ |
int i, j; |
168 |
+ |
|
169 |
+ |
if (earr == NULL) /* out of memory? */ |
170 |
+ |
return(NULL); |
171 |
+ |
/* compute squared neighbor diffs */ |
172 |
+ |
for (ap = hp->sa, ep = earr, i = 0; i < hp->ns; i++) |
173 |
+ |
for (j = 0; j < hp->ns; j++, ap++, ep++) { |
174 |
+ |
b = bright(ap[0].v); |
175 |
+ |
if (i) { /* from above */ |
176 |
+ |
d2 = b - bright(ap[-hp->ns].v); |
177 |
+ |
d2 *= d2; |
178 |
+ |
ep[0] += d2; |
179 |
+ |
ep[-hp->ns] += d2; |
180 |
+ |
} |
181 |
+ |
if (!j) continue; |
182 |
+ |
/* from behind */ |
183 |
+ |
d2 = b - bright(ap[-1].v); |
184 |
+ |
d2 *= d2; |
185 |
+ |
ep[0] += d2; |
186 |
+ |
ep[-1] += d2; |
187 |
+ |
if (!i) continue; |
188 |
+ |
/* diagonal */ |
189 |
+ |
d2 = b - bright(ap[-hp->ns-1].v); |
190 |
+ |
d2 *= d2; |
191 |
+ |
ep[0] += d2; |
192 |
+ |
ep[-hp->ns-1] += d2; |
193 |
+ |
} |
194 |
+ |
/* correct for number of neighbors */ |
195 |
+ |
earr[0] *= 8./3.; |
196 |
+ |
earr[hp->ns-1] *= 8./3.; |
197 |
+ |
earr[(hp->ns-1)*hp->ns] *= 8./3.; |
198 |
+ |
earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 8./3.; |
199 |
+ |
for (i = 1; i < hp->ns-1; i++) { |
200 |
+ |
earr[i*hp->ns] *= 8./5.; |
201 |
+ |
earr[i*hp->ns + hp->ns-1] *= 8./5.; |
202 |
+ |
} |
203 |
+ |
for (j = 1; j < hp->ns-1; j++) { |
204 |
+ |
earr[j] *= 8./5.; |
205 |
+ |
earr[(hp->ns-1)*hp->ns + j] *= 8./5.; |
206 |
+ |
} |
207 |
+ |
return(earr); |
208 |
+ |
} |
209 |
+ |
|
210 |
+ |
|
211 |
+ |
/* Perform super-sampling on hemisphere (introduces bias) */ |
212 |
+ |
static void |
213 |
+ |
ambsupersamp(AMBHEMI *hp, int cnt) |
214 |
+ |
{ |
215 |
+ |
float *earr = getambdiffs(hp); |
216 |
+ |
double e2rem = 0; |
217 |
+ |
AMBSAMP *ap; |
218 |
+ |
float *ep; |
219 |
+ |
int i, j, n, nss; |
220 |
+ |
|
221 |
+ |
if (earr == NULL) /* just skip calc. if no memory */ |
222 |
+ |
return; |
223 |
+ |
/* accumulate estimated variances */ |
224 |
+ |
for (ep = earr + hp->ns*hp->ns; ep > earr; ) |
225 |
+ |
e2rem += *--ep; |
226 |
+ |
ep = earr; /* perform super-sampling */ |
227 |
+ |
for (ap = hp->sa, i = 0; i < hp->ns; i++) |
228 |
+ |
for (j = 0; j < hp->ns; j++, ap++) { |
229 |
+ |
if (e2rem <= FTINY) |
230 |
+ |
goto done; /* nothing left to do */ |
231 |
+ |
nss = *ep/e2rem*cnt + frandom(); |
232 |
+ |
for (n = 1; n <= nss && ambsample(hp,i,j,n); n++) |
233 |
+ |
--cnt; |
234 |
+ |
e2rem -= *ep++; /* update remainder */ |
235 |
+ |
} |
236 |
+ |
done: |
237 |
+ |
free(earr); |
238 |
+ |
} |
239 |
+ |
|
240 |
+ |
|
241 |
+ |
static AMBHEMI * |
242 |
+ |
samp_hemi( /* sample indirect hemisphere */ |
243 |
+ |
COLOR rcol, |
244 |
+ |
RAY *r, |
245 |
+ |
double wt |
246 |
+ |
) |
247 |
+ |
{ |
248 |
+ |
AMBHEMI *hp; |
249 |
+ |
double d; |
250 |
+ |
int n, i, j; |
251 |
+ |
/* set number of divisions */ |
252 |
+ |
if (ambacc <= FTINY && |
253 |
+ |
wt > (d = 0.8*intens(rcol)*r->rweight/(ambdiv*minweight))) |
254 |
+ |
wt = d; /* avoid ray termination */ |
255 |
+ |
n = sqrt(ambdiv * wt) + 0.5; |
256 |
+ |
i = 1 + 5*(ambacc > FTINY); /* minimum number of samples */ |
257 |
+ |
if (n < i) |
258 |
+ |
n = i; |
259 |
+ |
/* allocate sampling array */ |
260 |
+ |
hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + sizeof(AMBSAMP)*(n*n - 1)); |
261 |
+ |
if (hp == NULL) |
262 |
+ |
error(SYSTEM, "out of memory in samp_hemi"); |
263 |
+ |
hp->rp = r; |
264 |
+ |
hp->ns = n; |
265 |
+ |
hp->acol[RED] = hp->acol[GRN] = hp->acol[BLU] = 0.0; |
266 |
+ |
memset(hp->sa, 0, sizeof(AMBSAMP)*n*n); |
267 |
+ |
hp->sampOK = 0; |
268 |
+ |
/* assign coefficient */ |
269 |
+ |
copycolor(hp->acoef, rcol); |
270 |
+ |
d = 1.0/(n*n); |
271 |
+ |
scalecolor(hp->acoef, d); |
272 |
+ |
/* make tangent plane axes */ |
273 |
+ |
if (!getperpendicular(hp->ux, r->ron, 1)) |
274 |
+ |
error(CONSISTENCY, "bad ray direction in samp_hemi"); |
275 |
+ |
VCROSS(hp->uy, r->ron, hp->ux); |
276 |
+ |
/* sample divisions */ |
277 |
+ |
for (i = hp->ns; i--; ) |
278 |
+ |
for (j = hp->ns; j--; ) |
279 |
+ |
hp->sampOK += ambsample(hp, i, j, 0); |
280 |
+ |
copycolor(rcol, hp->acol); |
281 |
+ |
if (!hp->sampOK) { /* utter failure? */ |
282 |
+ |
free(hp); |
283 |
+ |
return(NULL); |
284 |
+ |
} |
285 |
+ |
if (hp->sampOK < hp->ns*hp->ns) { |
286 |
+ |
hp->sampOK *= -1; /* soft failure */ |
287 |
+ |
return(hp); |
288 |
+ |
} |
289 |
+ |
n = ambssamp*wt + 0.5; |
290 |
+ |
if (n > 8) { /* perform super-sampling? */ |
291 |
+ |
ambsupersamp(hp, n); |
292 |
+ |
copycolor(rcol, hp->acol); |
293 |
+ |
} |
294 |
+ |
return(hp); /* all is well */ |
295 |
+ |
} |
296 |
+ |
|
297 |
+ |
|
298 |
+ |
/* Return brightness of farthest ambient sample */ |
299 |
+ |
static double |
300 |
+ |
back_ambval(AMBHEMI *hp, const int n1, const int n2, const int n3) |
301 |
+ |
{ |
302 |
+ |
if (hp->sa[n1].d <= hp->sa[n2].d) { |
303 |
+ |
if (hp->sa[n1].d <= hp->sa[n3].d) |
304 |
+ |
return(colval(hp->sa[n1].v,CIEY)); |
305 |
+ |
return(colval(hp->sa[n3].v,CIEY)); |
306 |
+ |
} |
307 |
+ |
if (hp->sa[n2].d <= hp->sa[n3].d) |
308 |
+ |
return(colval(hp->sa[n2].v,CIEY)); |
309 |
+ |
return(colval(hp->sa[n3].v,CIEY)); |
310 |
+ |
} |
311 |
+ |
|
312 |
+ |
|
313 |
|
/* Compute vectors and coefficients for Hessian/gradient calcs */ |
314 |
|
static void |
315 |
< |
comp_fftri(FFTRI *ftp, FVECT ap0, FVECT ap1, FVECT rop) |
315 |
> |
comp_fftri(FFTRI *ftp, AMBHEMI *hp, const int n0, const int n1) |
316 |
|
{ |
317 |
|
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
318 |
< |
int i; |
318 |
> |
int ii; |
319 |
|
|
320 |
< |
VSUB(ftp->r_i, ap0, rop); |
321 |
< |
VSUB(ftp->r_i1, ap1, rop); |
322 |
< |
VSUB(ftp->e_i, ap1, ap0); |
320 |
> |
VSUB(ftp->r_i, hp->sa[n0].p, hp->rp->rop); |
321 |
> |
VSUB(ftp->r_i1, hp->sa[n1].p, hp->rp->rop); |
322 |
> |
VSUB(ftp->e_i, hp->sa[n1].p, hp->sa[n0].p); |
323 |
|
VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1); |
324 |
|
rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp); |
325 |
|
dot_e = DOT(ftp->e_i,ftp->e_i); |
331 |
|
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r + |
332 |
|
dot_e*ftp->I1 )*0.5*rdot_cp; |
333 |
|
J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e; |
334 |
< |
for (i = 3; i--; ) |
335 |
< |
ftp->rI2_eJ2[i] = ftp->I2*ftp->r_i[i] + J2*ftp->e_i[i]; |
334 |
> |
for (ii = 3; ii--; ) |
335 |
> |
ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii]; |
336 |
|
} |
337 |
|
|
338 |
|
|
381 |
|
hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] + |
382 |
|
2.0*J3*m4[i][j] ); |
383 |
|
hess[i][j] += d2*(i==j); |
384 |
< |
hess[i][j] *= 1.0/PI; |
384 |
> |
hess[i][j] *= -1.0/PI; |
385 |
|
} |
386 |
|
} |
387 |
|
|
403 |
|
/* Add to radiometric Hessian from the given triangle */ |
404 |
|
static void |
405 |
|
add2hessian(FVECT hess[3], FVECT ehess1[3], |
406 |
< |
FVECT ehess2[3], FVECT ehess3[3], COLORV v) |
406 |
> |
FVECT ehess2[3], FVECT ehess3[3], double v) |
407 |
|
{ |
408 |
|
int i, j; |
409 |
|
|
424 |
|
f1 = 2.0*DOT(nrm, ftp->rcp); |
425 |
|
VCROSS(ncp, nrm, ftp->e_i); |
426 |
|
for (i = 3; i--; ) |
427 |
< |
grad[i] = (-0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
427 |
> |
grad[i] = (0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
428 |
|
} |
429 |
|
|
430 |
|
|
440 |
|
|
441 |
|
/* Add to displacement gradient from the given triangle */ |
442 |
|
static void |
443 |
< |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v) |
443 |
> |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, double v) |
444 |
|
{ |
445 |
|
int i; |
446 |
|
|
449 |
|
} |
450 |
|
|
451 |
|
|
276 |
– |
/* Return brightness of furthest ambient sample */ |
277 |
– |
static COLORV |
278 |
– |
back_ambval(struct s_ambsamp *ap1, struct s_ambsamp *ap2, |
279 |
– |
struct s_ambsamp *ap3, FVECT orig) |
280 |
– |
{ |
281 |
– |
COLORV vback; |
282 |
– |
FVECT vec; |
283 |
– |
double d2, d2best; |
284 |
– |
|
285 |
– |
VSUB(vec, ap1->p, orig); |
286 |
– |
d2best = DOT(vec,vec); |
287 |
– |
vback = colval(ap1->v,CIEY); |
288 |
– |
VSUB(vec, ap2->p, orig); |
289 |
– |
d2 = DOT(vec,vec); |
290 |
– |
if (d2 > d2best) { |
291 |
– |
d2best = d2; |
292 |
– |
vback = colval(ap2->v,CIEY); |
293 |
– |
} |
294 |
– |
VSUB(vec, ap3->p, orig); |
295 |
– |
d2 = DOT(vec,vec); |
296 |
– |
if (d2 > d2best) |
297 |
– |
return(colval(ap3->v,CIEY)); |
298 |
– |
return(vback); |
299 |
– |
} |
300 |
– |
|
301 |
– |
|
452 |
|
/* Compute anisotropic radii and eigenvector directions */ |
453 |
< |
static int |
453 |
> |
static void |
454 |
|
eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3]) |
455 |
|
{ |
456 |
|
double hess2[2][2]; |
466 |
|
hess2[0][1] = DOT(uv[0], b); |
467 |
|
hess2[1][0] = DOT(uv[1], a); |
468 |
|
hess2[1][1] = DOT(uv[1], b); |
469 |
< |
/* compute eigenvalues */ |
470 |
< |
if ( quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1], |
471 |
< |
hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]) != 2 || |
472 |
< |
((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) | |
473 |
< |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) |
474 |
< |
error(INTERNAL, "bad eigenvalue calculation"); |
475 |
< |
|
469 |
> |
/* compute eigenvalue(s) */ |
470 |
> |
i = quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1], |
471 |
> |
hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]); |
472 |
> |
if (i == 1) /* double-root (circle) */ |
473 |
> |
evalue[1] = evalue[0]; |
474 |
> |
if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) | |
475 |
> |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) { |
476 |
> |
ra[0] = ra[1] = maxarad; |
477 |
> |
return; |
478 |
> |
} |
479 |
|
if (evalue[0] > evalue[1]) { |
480 |
|
ra[0] = sqrt(sqrt(4.0/evalue[0])); |
481 |
|
ra[1] = sqrt(sqrt(4.0/evalue[1])); |
533 |
|
} |
534 |
|
/* compute first row of edges */ |
535 |
|
for (j = 0; j < hp->ns-1; j++) { |
536 |
< |
comp_fftri(&fftr, ambsamp(hp,0,j).p, |
384 |
< |
ambsamp(hp,0,j+1).p, hp->rp->rop); |
536 |
> |
comp_fftri(&fftr, hp, AI(hp,0,j), AI(hp,0,j+1)); |
537 |
|
if (hessrow != NULL) |
538 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
539 |
|
if (gradrow != NULL) |
543 |
|
for (i = 0; i < hp->ns-1; i++) { |
544 |
|
FVECT hesscol[3]; /* compute first vertical edge */ |
545 |
|
FVECT gradcol; |
546 |
< |
comp_fftri(&fftr, ambsamp(hp,i,0).p, |
395 |
< |
ambsamp(hp,i+1,0).p, hp->rp->rop); |
546 |
> |
comp_fftri(&fftr, hp, AI(hp,i,0), AI(hp,i+1,0)); |
547 |
|
if (hessrow != NULL) |
548 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
549 |
|
if (gradrow != NULL) |
551 |
|
for (j = 0; j < hp->ns-1; j++) { |
552 |
|
FVECT hessdia[3]; /* compute triangle contributions */ |
553 |
|
FVECT graddia; |
554 |
< |
COLORV backg; |
555 |
< |
backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1), |
556 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
554 |
> |
double backg; |
555 |
> |
backg = back_ambval(hp, AI(hp,i,j), |
556 |
> |
AI(hp,i,j+1), AI(hp,i+1,j)); |
557 |
|
/* diagonal (inner) edge */ |
558 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, |
408 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
558 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j)); |
559 |
|
if (hessrow != NULL) { |
560 |
|
comp_hessian(hessdia, &fftr, hp->rp->ron); |
561 |
|
rev_hessian(hesscol); |
562 |
|
add2hessian(hessian, hessrow[j], hessdia, hesscol, backg); |
563 |
|
} |
564 |
< |
if (gradient != NULL) { |
564 |
> |
if (gradrow != NULL) { |
565 |
|
comp_gradient(graddia, &fftr, hp->rp->ron); |
566 |
|
rev_gradient(gradcol); |
567 |
|
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
568 |
|
} |
569 |
|
/* initialize edge in next row */ |
570 |
< |
comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p, |
421 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
570 |
> |
comp_fftri(&fftr, hp, AI(hp,i+1,j+1), AI(hp,i+1,j)); |
571 |
|
if (hessrow != NULL) |
572 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
573 |
|
if (gradrow != NULL) |
574 |
|
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
575 |
|
/* new column edge & paired triangle */ |
576 |
< |
backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1), |
577 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
578 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p, |
430 |
< |
hp->rp->rop); |
576 |
> |
backg = back_ambval(hp, AI(hp,i+1,j+1), |
577 |
> |
AI(hp,i+1,j), AI(hp,i,j+1)); |
578 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j+1)); |
579 |
|
if (hessrow != NULL) { |
580 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
581 |
|
rev_hessian(hessdia); |
609 |
|
static void |
610 |
|
ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2]) |
611 |
|
{ |
612 |
< |
struct s_ambsamp *ap; |
613 |
< |
double dgsum[2]; |
614 |
< |
int n; |
615 |
< |
FVECT vd; |
616 |
< |
double gfact; |
612 |
> |
AMBSAMP *ap; |
613 |
> |
double dgsum[2]; |
614 |
> |
int n; |
615 |
> |
FVECT vd; |
616 |
> |
double gfact; |
617 |
|
|
618 |
|
dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */ |
619 |
|
for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) { |
621 |
|
VSUB(vd, ap->p, hp->rp->rop); |
622 |
|
/* brightness over cosine factor */ |
623 |
|
gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd); |
624 |
< |
/* -sine = -proj_radius/vd_length */ |
625 |
< |
dgsum[0] += DOT(uv[1], vd) * gfact; |
626 |
< |
dgsum[1] -= DOT(uv[0], vd) * gfact; |
624 |
> |
/* sine = proj_radius/vd_length */ |
625 |
> |
dgsum[0] -= DOT(uv[1], vd) * gfact; |
626 |
> |
dgsum[1] += DOT(uv[0], vd) * gfact; |
627 |
|
} |
628 |
|
dg[0] = dgsum[0] / (hp->ns*hp->ns); |
629 |
|
dg[1] = dgsum[1] / (hp->ns*hp->ns); |
630 |
|
} |
631 |
|
|
632 |
|
|
633 |
+ |
/* Compute potential light leak direction flags for cache value */ |
634 |
+ |
static uint32 |
635 |
+ |
ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1) |
636 |
+ |
{ |
637 |
+ |
const double max_d = 1.0/(minarad*ambacc + 0.001); |
638 |
+ |
const double ang_res = 0.5*PI/hp->ns; |
639 |
+ |
const double ang_step = ang_res/((int)(16/PI*ang_res) + 1.01); |
640 |
+ |
double avg_d = 0; |
641 |
+ |
uint32 flgs = 0; |
642 |
+ |
FVECT vec; |
643 |
+ |
double u, v; |
644 |
+ |
double ang, a1; |
645 |
+ |
int i, j; |
646 |
+ |
/* don't bother for a few samples */ |
647 |
+ |
if (hp->ns < 8) |
648 |
+ |
return(0); |
649 |
+ |
/* check distances overhead */ |
650 |
+ |
for (i = hp->ns*3/4; i-- > hp->ns>>2; ) |
651 |
+ |
for (j = hp->ns*3/4; j-- > hp->ns>>2; ) |
652 |
+ |
avg_d += ambsam(hp,i,j).d; |
653 |
+ |
avg_d *= 4.0/(hp->ns*hp->ns); |
654 |
+ |
if (avg_d*r0 >= 1.0) /* ceiling too low for corral? */ |
655 |
+ |
return(0); |
656 |
+ |
if (avg_d >= max_d) /* insurance */ |
657 |
+ |
return(0); |
658 |
+ |
/* else circle around perimeter */ |
659 |
+ |
for (i = 0; i < hp->ns; i++) |
660 |
+ |
for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) { |
661 |
+ |
AMBSAMP *ap = &ambsam(hp,i,j); |
662 |
+ |
if ((ap->d <= FTINY) | (ap->d >= max_d)) |
663 |
+ |
continue; /* too far or too near */ |
664 |
+ |
VSUB(vec, ap->p, hp->rp->rop); |
665 |
+ |
u = DOT(vec, uv[0]); |
666 |
+ |
v = DOT(vec, uv[1]); |
667 |
+ |
if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= u*u + v*v) |
668 |
+ |
continue; /* occluder outside ellipse */ |
669 |
+ |
ang = atan2a(v, u); /* else set direction flags */ |
670 |
+ |
for (a1 = ang-ang_res; a1 <= ang+ang_res; a1 += ang_step) |
671 |
+ |
flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0))); |
672 |
+ |
} |
673 |
+ |
/* add low-angle incident (< 20deg) */ |
674 |
+ |
if (fabs(hp->rp->rod) <= 0.342) { |
675 |
+ |
u = -DOT(hp->rp->rdir, uv[0]); |
676 |
+ |
v = -DOT(hp->rp->rdir, uv[1]); |
677 |
+ |
if ((r0*r0*u*u + r1*r1*v*v) > hp->rp->rot*hp->rp->rot) { |
678 |
+ |
ang = atan2a(v, u); |
679 |
+ |
ang += 2.*PI*(ang < 0); |
680 |
+ |
ang *= 16/PI; |
681 |
+ |
if ((ang < .5) | (ang >= 31.5)) |
682 |
+ |
flgs |= 0x80000001; |
683 |
+ |
else |
684 |
+ |
flgs |= 3L<<(int)(ang-.5); |
685 |
+ |
} |
686 |
+ |
} |
687 |
+ |
return(flgs); |
688 |
+ |
} |
689 |
+ |
|
690 |
+ |
|
691 |
|
int |
692 |
|
doambient( /* compute ambient component */ |
693 |
|
COLOR rcol, /* input/output color */ |
696 |
|
FVECT uv[2], /* returned (optional) */ |
697 |
|
float ra[2], /* returned (optional) */ |
698 |
|
float pg[2], /* returned (optional) */ |
699 |
< |
float dg[2] /* returned (optional) */ |
699 |
> |
float dg[2], /* returned (optional) */ |
700 |
> |
uint32 *crlp /* returned (optional) */ |
701 |
|
) |
702 |
|
{ |
703 |
< |
AMBHEMI *hp = inithemi(rcol, r, wt); |
704 |
< |
int cnt = 0; |
705 |
< |
FVECT my_uv[2]; |
706 |
< |
double d, acol[3]; |
707 |
< |
struct s_ambsamp *ap; |
708 |
< |
int i, j; |
502 |
< |
/* check/initialize */ |
503 |
< |
if (hp == NULL) |
504 |
< |
return(0); |
703 |
> |
AMBHEMI *hp = samp_hemi(rcol, r, wt); |
704 |
> |
FVECT my_uv[2]; |
705 |
> |
double d, K; |
706 |
> |
AMBSAMP *ap; |
707 |
> |
int i; |
708 |
> |
/* clear return values */ |
709 |
|
if (uv != NULL) |
710 |
|
memset(uv, 0, sizeof(FVECT)*2); |
711 |
|
if (ra != NULL) |
714 |
|
pg[0] = pg[1] = 0.0; |
715 |
|
if (dg != NULL) |
716 |
|
dg[0] = dg[1] = 0.0; |
717 |
< |
/* sample the hemisphere */ |
718 |
< |
acol[0] = acol[1] = acol[2] = 0.0; |
719 |
< |
for (i = hp->ns; i--; ) |
720 |
< |
for (j = hp->ns; j--; ) |
721 |
< |
if ((ap = ambsample(hp, i, j)) != NULL) { |
722 |
< |
addcolor(acol, ap->v); |
723 |
< |
++cnt; |
724 |
< |
} |
725 |
< |
if (!cnt) { |
522 |
< |
setcolor(rcol, 0.0, 0.0, 0.0); |
523 |
< |
free(hp); |
524 |
< |
return(0); /* no valid samples */ |
717 |
> |
if (crlp != NULL) |
718 |
> |
*crlp = 0; |
719 |
> |
if (hp == NULL) /* sampling falure? */ |
720 |
> |
return(0); |
721 |
> |
|
722 |
> |
if ((ra == NULL) & (pg == NULL) & (dg == NULL) || |
723 |
> |
(hp->sampOK < 0) | (hp->ns < 6)) { |
724 |
> |
free(hp); /* Hessian not requested/possible */ |
725 |
> |
return(-1); /* value-only return value */ |
726 |
|
} |
727 |
< |
copycolor(rcol, acol); /* final indirect irradiance/PI */ |
728 |
< |
if (cnt < hp->ns*hp->ns || /* incomplete sampling? */ |
729 |
< |
(ra == NULL) & (pg == NULL) & (dg == NULL)) { |
730 |
< |
free(hp); |
731 |
< |
return(-1); /* no radius or gradient calc. */ |
727 |
> |
if ((d = bright(rcol)) > FTINY) { /* normalize Y values */ |
728 |
> |
d = 0.99*(hp->ns*hp->ns)/d; |
729 |
> |
K = 0.01; |
730 |
> |
} else { /* or fall back on geometric Hessian */ |
731 |
> |
K = 1.0; |
732 |
> |
pg = NULL; |
733 |
> |
dg = NULL; |
734 |
> |
crlp = NULL; |
735 |
|
} |
532 |
– |
if (bright(acol) > FTINY) /* normalize Y values */ |
533 |
– |
d = cnt/bright(acol); |
534 |
– |
else |
535 |
– |
d = 0.0; |
736 |
|
ap = hp->sa; /* relative Y channel from here on... */ |
737 |
|
for (i = hp->ns*hp->ns; i--; ap++) |
738 |
< |
colval(ap->v,CIEY) = bright(ap->v)*d + 0.01; |
738 |
> |
colval(ap->v,CIEY) = bright(ap->v)*d + K; |
739 |
|
|
740 |
|
if (uv == NULL) /* make sure we have axis pointers */ |
741 |
|
uv = my_uv; |
759 |
|
if (ra[1] < minarad) |
760 |
|
ra[1] = minarad; |
761 |
|
} |
762 |
< |
ra[0] *= d = 1.0/sqrt(sqrt(wt)); |
762 |
> |
ra[0] *= d = 1.0/sqrt(wt); |
763 |
|
if ((ra[1] *= d) > 2.0*ra[0]) |
764 |
|
ra[1] = 2.0*ra[0]; |
765 |
|
if (ra[1] > maxarad) { |
767 |
|
if (ra[0] > maxarad) |
768 |
|
ra[0] = maxarad; |
769 |
|
} |
770 |
+ |
/* flag encroached directions */ |
771 |
+ |
if (crlp != NULL) |
772 |
+ |
*crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc); |
773 |
|
if (pg != NULL) { /* cap gradient if necessary */ |
774 |
|
d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1]; |
775 |
|
if (d > 1.0) { |