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 |
|
|
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 */ |
36 |
|
FVECT ux, uy; /* tangent axis unit vectors */ |
37 |
|
int ns; /* number of samples per axis */ |
38 |
|
COLOR acoef; /* division contribution coefficient */ |
39 |
< |
struct s_ambsamp { |
30 |
< |
COLOR v; /* hemisphere sample value */ |
31 |
< |
float p[3]; /* intersection point */ |
32 |
< |
} sa[1]; /* sample array (extends struct) */ |
39 |
> |
AMBSAMP sa[1]; /* sample array (extends struct) */ |
40 |
|
} AMBHEMI; /* ambient sample hemisphere */ |
41 |
|
|
42 |
< |
#define ambsamp(h,i,j) (h)->sa[(i)*(h)->ns + (j)] |
42 |
> |
#define AI(h,i,j) ((i)*(h)->ns + (j)) |
43 |
> |
#define ambsam(h,i,j) (h)->sa[AI(h,i,j)] |
44 |
|
|
45 |
|
typedef struct { |
46 |
< |
FVECT r_i, r_i1, e_i; |
47 |
< |
double nf, I1, I2, J2; |
46 |
> |
FVECT r_i, r_i1, e_i, rcp, rI2_eJ2; |
47 |
> |
double I1, I2; |
48 |
|
} FFTRI; /* vectors and coefficients for Hessian calculation */ |
49 |
|
|
50 |
|
|
67 |
|
if (n < i) |
68 |
|
n = i; |
69 |
|
/* allocate sampling array */ |
70 |
< |
hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + |
63 |
< |
sizeof(struct s_ambsamp)*(n*n - 1)); |
70 |
> |
hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + sizeof(AMBSAMP)*(n*n - 1)); |
71 |
|
if (hp == NULL) |
72 |
|
return(NULL); |
73 |
|
hp->rp = r; |
76 |
|
copycolor(hp->acoef, ac); |
77 |
|
d = 1.0/(n*n); |
78 |
|
scalecolor(hp->acoef, d); |
79 |
< |
/* make tangent axes */ |
80 |
< |
hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0; |
81 |
< |
for (i = 0; i < 3; i++) |
82 |
< |
if (r->ron[i] < 0.6 && r->ron[i] > -0.6) |
79 |
> |
/* make tangent plane axes */ |
80 |
> |
hp->uy[0] = 0.5 - frandom(); |
81 |
> |
hp->uy[1] = 0.5 - frandom(); |
82 |
> |
hp->uy[2] = 0.5 - frandom(); |
83 |
> |
for (i = 3; i--; ) |
84 |
> |
if ((-0.6 < r->ron[i]) & (r->ron[i] < 0.6)) |
85 |
|
break; |
86 |
< |
if (i >= 3) |
87 |
< |
error(CONSISTENCY, "bad ray direction in inithemi()"); |
86 |
> |
if (i < 0) |
87 |
> |
error(CONSISTENCY, "bad ray direction in inithemi"); |
88 |
|
hp->uy[i] = 1.0; |
89 |
|
VCROSS(hp->ux, hp->uy, r->ron); |
90 |
|
normalize(hp->ux); |
94 |
|
} |
95 |
|
|
96 |
|
|
97 |
+ |
/* Sample ambient division and apply weighting coefficient */ |
98 |
|
static int |
99 |
< |
ambsample( /* sample an ambient direction */ |
90 |
< |
AMBHEMI *hp, |
91 |
< |
int i, |
92 |
< |
int j |
93 |
< |
) |
99 |
> |
getambsamp(RAY *arp, AMBHEMI *hp, int i, int j, int n) |
100 |
|
{ |
101 |
< |
struct s_ambsamp *ap = &ambsamp(hp,i,j); |
102 |
< |
RAY ar; |
97 |
< |
int hlist[3]; |
98 |
< |
double spt[2], zd; |
99 |
< |
int ii; |
101 |
> |
int hlist[3], ii; |
102 |
> |
double spt[2], zd; |
103 |
|
/* ambient coefficient for weight */ |
104 |
|
if (ambacc > FTINY) |
105 |
< |
setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL); |
105 |
> |
setcolor(arp->rcoef, AVGREFL, AVGREFL, AVGREFL); |
106 |
|
else |
107 |
< |
copycolor(ar.rcoef, hp->acoef); |
108 |
< |
if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0) { |
109 |
< |
setcolor(ap->v, 0., 0., 0.); |
107 |
< |
VCOPY(ap->p, hp->rp->rop); |
108 |
< |
return(0); /* no sample taken */ |
109 |
< |
} |
107 |
> |
copycolor(arp->rcoef, hp->acoef); |
108 |
> |
if (rayorigin(arp, AMBIENT, hp->rp, arp->rcoef) < 0) |
109 |
> |
return(0); |
110 |
|
if (ambacc > FTINY) { |
111 |
< |
multcolor(ar.rcoef, hp->acoef); |
112 |
< |
scalecolor(ar.rcoef, 1./AVGREFL); |
111 |
> |
multcolor(arp->rcoef, hp->acoef); |
112 |
> |
scalecolor(arp->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 |
> |
if (!n) { /* avoid border samples for n==0 */ |
119 |
> |
if ((spt[0] < 0.1) | (spt[0] >= 0.9)) |
120 |
> |
spt[0] = 0.1 + 0.8*frandom(); |
121 |
> |
if ((spt[1] < 0.1) | (spt[1] >= 0.9)) |
122 |
> |
spt[1] = 0.1 + 0.8*frandom(); |
123 |
> |
} |
124 |
> |
SDsquare2disk(spt, (j+spt[1])/hp->ns, (i+spt[0])/hp->ns); |
125 |
|
zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]); |
126 |
|
for (ii = 3; ii--; ) |
127 |
< |
ar.rdir[ii] = spt[0]*hp->ux[ii] + |
127 |
> |
arp->rdir[ii] = spt[0]*hp->ux[ii] + |
128 |
|
spt[1]*hp->uy[ii] + |
129 |
|
zd*hp->rp->ron[ii]; |
130 |
< |
checknorm(ar.rdir); |
131 |
< |
dimlist[ndims++] = i*hp->ns + j + 90171; |
132 |
< |
rayvalue(&ar); /* evaluate ray */ |
133 |
< |
ndims--; |
134 |
< |
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
127 |
< |
copycolor(ap->v, ar.rcol); |
128 |
< |
if (ar.rt > 20.0*maxarad) /* limit vertex distance */ |
129 |
< |
ar.rt = 20.0*maxarad; |
130 |
< |
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
130 |
> |
checknorm(arp->rdir); |
131 |
> |
dimlist[ndims++] = AI(hp,i,j) + 90171; |
132 |
> |
rayvalue(arp); /* evaluate ray */ |
133 |
> |
ndims--; /* apply coefficient */ |
134 |
> |
multcolor(arp->rcol, arp->rcoef); |
135 |
|
return(1); |
136 |
|
} |
137 |
|
|
138 |
|
|
139 |
+ |
static AMBSAMP * |
140 |
+ |
ambsample( /* initial ambient division sample */ |
141 |
+ |
AMBHEMI *hp, |
142 |
+ |
int i, |
143 |
+ |
int j |
144 |
+ |
) |
145 |
+ |
{ |
146 |
+ |
AMBSAMP *ap = &ambsam(hp,i,j); |
147 |
+ |
RAY ar; |
148 |
+ |
/* generate hemispherical sample */ |
149 |
+ |
if (!getambsamp(&ar, hp, i, j, 0) || ar.rt <= FTINY) { |
150 |
+ |
memset(ap, 0, sizeof(AMBSAMP)); |
151 |
+ |
return(NULL); |
152 |
+ |
} |
153 |
+ |
ap->d = 1.0/ar.rt; /* limit vertex distance */ |
154 |
+ |
if (ar.rt > 10.0*thescene.cusize) |
155 |
+ |
ar.rt = 10.0*thescene.cusize; |
156 |
+ |
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
157 |
+ |
copycolor(ap->v, ar.rcol); |
158 |
+ |
return(ap); |
159 |
+ |
} |
160 |
+ |
|
161 |
+ |
|
162 |
+ |
/* Estimate errors based on ambient division differences */ |
163 |
+ |
static float * |
164 |
+ |
getambdiffs(AMBHEMI *hp) |
165 |
+ |
{ |
166 |
+ |
float *earr = (float *)calloc(hp->ns*hp->ns, sizeof(float)); |
167 |
+ |
float *ep; |
168 |
+ |
AMBSAMP *ap; |
169 |
+ |
double b, d2; |
170 |
+ |
int i, j; |
171 |
+ |
|
172 |
+ |
if (earr == NULL) /* out of memory? */ |
173 |
+ |
return(NULL); |
174 |
+ |
/* compute squared neighbor diffs */ |
175 |
+ |
for (ap = hp->sa, ep = earr, i = 0; i < hp->ns; i++) |
176 |
+ |
for (j = 0; j < hp->ns; j++, ap++, ep++) { |
177 |
+ |
b = bright(ap[0].v); |
178 |
+ |
if (i) { /* from above */ |
179 |
+ |
d2 = b - bright(ap[-hp->ns].v); |
180 |
+ |
d2 *= d2; |
181 |
+ |
ep[0] += d2; |
182 |
+ |
ep[-hp->ns] += d2; |
183 |
+ |
} |
184 |
+ |
if (!j) continue; |
185 |
+ |
/* from behind */ |
186 |
+ |
d2 = b - bright(ap[-1].v); |
187 |
+ |
d2 *= d2; |
188 |
+ |
ep[0] += d2; |
189 |
+ |
ep[-1] += d2; |
190 |
+ |
if (!i) continue; |
191 |
+ |
/* diagonal */ |
192 |
+ |
d2 = b - bright(ap[-hp->ns-1].v); |
193 |
+ |
d2 *= d2; |
194 |
+ |
ep[0] += d2; |
195 |
+ |
ep[-hp->ns-1] += d2; |
196 |
+ |
} |
197 |
+ |
/* correct for number of neighbors */ |
198 |
+ |
earr[0] *= 8./3.; |
199 |
+ |
earr[hp->ns-1] *= 8./3.; |
200 |
+ |
earr[(hp->ns-1)*hp->ns] *= 8./3.; |
201 |
+ |
earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 8./3.; |
202 |
+ |
for (i = 1; i < hp->ns-1; i++) { |
203 |
+ |
earr[i*hp->ns] *= 8./5.; |
204 |
+ |
earr[i*hp->ns + hp->ns-1] *= 8./5.; |
205 |
+ |
} |
206 |
+ |
for (j = 1; j < hp->ns-1; j++) { |
207 |
+ |
earr[j] *= 8./5.; |
208 |
+ |
earr[(hp->ns-1)*hp->ns + j] *= 8./5.; |
209 |
+ |
} |
210 |
+ |
return(earr); |
211 |
+ |
} |
212 |
+ |
|
213 |
+ |
|
214 |
+ |
/* Perform super-sampling on hemisphere (introduces bias) */ |
215 |
+ |
static void |
216 |
+ |
ambsupersamp(double acol[3], AMBHEMI *hp, int cnt) |
217 |
+ |
{ |
218 |
+ |
float *earr = getambdiffs(hp); |
219 |
+ |
double e2rem = 0; |
220 |
+ |
AMBSAMP *ap; |
221 |
+ |
RAY ar; |
222 |
+ |
double asum[3]; |
223 |
+ |
float *ep; |
224 |
+ |
int i, j, n, nss; |
225 |
+ |
|
226 |
+ |
if (earr == NULL) /* just skip calc. if no memory */ |
227 |
+ |
return; |
228 |
+ |
/* accumulate estimated variances */ |
229 |
+ |
for (ep = earr + hp->ns*hp->ns; ep > earr; ) |
230 |
+ |
e2rem += *--ep; |
231 |
+ |
ep = earr; /* perform super-sampling */ |
232 |
+ |
for (ap = hp->sa, i = 0; i < hp->ns; i++) |
233 |
+ |
for (j = 0; j < hp->ns; j++, ap++) { |
234 |
+ |
if (e2rem <= FTINY) |
235 |
+ |
goto done; /* nothing left to do */ |
236 |
+ |
nss = *ep/e2rem*cnt + frandom(); |
237 |
+ |
asum[0] = asum[1] = asum[2] = 0.0; |
238 |
+ |
for (n = 1; n <= nss; n++) { |
239 |
+ |
if (!getambsamp(&ar, hp, i, j, n)) { |
240 |
+ |
nss = n-1; |
241 |
+ |
break; |
242 |
+ |
} |
243 |
+ |
addcolor(asum, ar.rcol); |
244 |
+ |
} |
245 |
+ |
if (nss) { /* update returned ambient value */ |
246 |
+ |
const double ssf = 1./(nss + 1.); |
247 |
+ |
for (n = 3; n--; ) |
248 |
+ |
acol[n] += ssf*asum[n] + |
249 |
+ |
(ssf - 1.)*colval(ap->v,n); |
250 |
+ |
} |
251 |
+ |
e2rem -= *ep++; /* update remainders */ |
252 |
+ |
cnt -= nss; |
253 |
+ |
} |
254 |
+ |
done: |
255 |
+ |
free(earr); |
256 |
+ |
} |
257 |
+ |
|
258 |
+ |
|
259 |
+ |
/* Return brightness of farthest ambient sample */ |
260 |
+ |
static double |
261 |
+ |
back_ambval(AMBHEMI *hp, const int n1, const int n2, const int n3) |
262 |
+ |
{ |
263 |
+ |
if (hp->sa[n1].d <= hp->sa[n2].d) { |
264 |
+ |
if (hp->sa[n1].d <= hp->sa[n3].d) |
265 |
+ |
return(colval(hp->sa[n1].v,CIEY)); |
266 |
+ |
return(colval(hp->sa[n3].v,CIEY)); |
267 |
+ |
} |
268 |
+ |
if (hp->sa[n2].d <= hp->sa[n3].d) |
269 |
+ |
return(colval(hp->sa[n2].v,CIEY)); |
270 |
+ |
return(colval(hp->sa[n3].v,CIEY)); |
271 |
+ |
} |
272 |
+ |
|
273 |
+ |
|
274 |
|
/* Compute vectors and coefficients for Hessian/gradient calcs */ |
275 |
|
static void |
276 |
< |
comp_fftri(FFTRI *ftp, float ap0[3], float ap1[3], FVECT rop) |
276 |
> |
comp_fftri(FFTRI *ftp, AMBHEMI *hp, const int n0, const int n1) |
277 |
|
{ |
278 |
< |
FVECT v1; |
279 |
< |
double dot_e, dot_er, dot_r, dot_r1; |
278 |
> |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
279 |
> |
int ii; |
280 |
|
|
281 |
< |
VSUB(ftp->r_i, ap0, rop); |
282 |
< |
VSUB(ftp->r_i1, ap1, rop); |
283 |
< |
VSUB(ftp->e_i, ap1, ap0); |
284 |
< |
VCROSS(v1, ftp->e_i, ftp->r_i); |
285 |
< |
ftp->nf = 1.0/DOT(v1,v1); |
147 |
< |
VCROSS(v1, ftp->r_i, ftp->r_i1); |
148 |
< |
ftp->I1 = sqrt(DOT(v1,v1)*ftp->nf); |
281 |
> |
VSUB(ftp->r_i, hp->sa[n0].p, hp->rp->rop); |
282 |
> |
VSUB(ftp->r_i1, hp->sa[n1].p, hp->rp->rop); |
283 |
> |
VSUB(ftp->e_i, hp->sa[n1].p, hp->sa[n0].p); |
284 |
> |
VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1); |
285 |
> |
rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp); |
286 |
|
dot_e = DOT(ftp->e_i,ftp->e_i); |
287 |
|
dot_er = DOT(ftp->e_i, ftp->r_i); |
288 |
< |
dot_r = DOT(ftp->r_i,ftp->r_i); |
289 |
< |
dot_r1 = DOT(ftp->r_i1,ftp->r_i1); |
290 |
< |
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)/dot_r1 - dot_er/dot_r + |
291 |
< |
dot_e*ftp->I1 )*0.5*ftp->nf; |
292 |
< |
ftp->J2 = 0.25*ftp->nf*( 1.0/dot_r - 1.0/dot_r1 ) - |
293 |
< |
dot_er/dot_e*ftp->I2; |
288 |
> |
rdot_r = 1.0/DOT(ftp->r_i,ftp->r_i); |
289 |
> |
rdot_r1 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
290 |
> |
ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) * sqrt(rdot_r*rdot_r1) ) * |
291 |
> |
sqrt( rdot_cp ); |
292 |
> |
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r + |
293 |
> |
dot_e*ftp->I1 )*0.5*rdot_cp; |
294 |
> |
J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e; |
295 |
> |
for (ii = 3; ii--; ) |
296 |
> |
ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii]; |
297 |
|
} |
298 |
|
|
299 |
|
|
300 |
< |
/* Compose matrix from two vectors */ |
300 |
> |
/* Compose 3x3 matrix from two vectors */ |
301 |
|
static void |
302 |
|
compose_matrix(FVECT mat[3], FVECT va, FVECT vb) |
303 |
|
{ |
314 |
|
static void |
315 |
|
comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm) |
316 |
|
{ |
317 |
< |
FVECT v1, v2; |
317 |
> |
FVECT ncp; |
318 |
|
FVECT m1[3], m2[3], m3[3], m4[3]; |
319 |
|
double d1, d2, d3, d4; |
320 |
|
double I3, J3, K3; |
324 |
|
d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
325 |
|
d3 = 1.0/DOT(ftp->e_i,ftp->e_i); |
326 |
|
d4 = DOT(ftp->e_i, ftp->r_i); |
327 |
< |
I3 = 0.25*ftp->nf*( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + |
328 |
< |
3.0*ftp->I2*d3 ); |
327 |
> |
I3 = ( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + 3.0/d3*ftp->I2 ) |
328 |
> |
/ ( 4.0*DOT(ftp->rcp,ftp->rcp) ); |
329 |
|
J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3; |
330 |
|
K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3); |
331 |
|
/* intermediate matrices */ |
332 |
< |
VCROSS(v1, nrm, ftp->e_i); |
333 |
< |
for (j = 3; j--; ) |
194 |
< |
v2[i] = ftp->I2*ftp->r_i[j] + ftp->J2*ftp->e_i[j]; |
195 |
< |
compose_matrix(m1, v1, v2); |
332 |
> |
VCROSS(ncp, nrm, ftp->e_i); |
333 |
> |
compose_matrix(m1, ncp, ftp->rI2_eJ2); |
334 |
|
compose_matrix(m2, ftp->r_i, ftp->r_i); |
335 |
|
compose_matrix(m3, ftp->e_i, ftp->e_i); |
336 |
|
compose_matrix(m4, ftp->r_i, ftp->e_i); |
337 |
< |
VCROSS(v1, ftp->r_i, ftp->e_i); |
200 |
< |
d1 = DOT(nrm, v1); |
337 |
> |
d1 = DOT(nrm, ftp->rcp); |
338 |
|
d2 = -d1*ftp->I2; |
339 |
|
d1 *= 2.0; |
340 |
|
for (i = 3; i--; ) /* final matrix sum */ |
364 |
|
/* Add to radiometric Hessian from the given triangle */ |
365 |
|
static void |
366 |
|
add2hessian(FVECT hess[3], FVECT ehess1[3], |
367 |
< |
FVECT ehess2[3], FVECT ehess3[3], COLORV v) |
367 |
> |
FVECT ehess2[3], FVECT ehess3[3], double v) |
368 |
|
{ |
369 |
|
int i, j; |
370 |
|
|
378 |
|
static void |
379 |
|
comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm) |
380 |
|
{ |
381 |
< |
FVECT vcp; |
381 |
> |
FVECT ncp; |
382 |
|
double f1; |
383 |
|
int i; |
384 |
|
|
385 |
< |
VCROSS(vcp, ftp->r_i, ftp->r_i1); |
386 |
< |
f1 = 2.0*DOT(nrm, vcp); |
250 |
< |
VCROSS(vcp, nrm, ftp->e_i); |
385 |
> |
f1 = 2.0*DOT(nrm, ftp->rcp); |
386 |
> |
VCROSS(ncp, nrm, ftp->e_i); |
387 |
|
for (i = 3; i--; ) |
388 |
< |
grad[i] = (0.5/PI)*( ftp->I1*vcp[i] + |
253 |
< |
f1*(ftp->I2*ftp->r_i[i] + ftp->J2*ftp->e_i[i]) ); |
388 |
> |
grad[i] = (0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
389 |
|
} |
390 |
|
|
391 |
|
|
401 |
|
|
402 |
|
/* Add to displacement gradient from the given triangle */ |
403 |
|
static void |
404 |
< |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v) |
404 |
> |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, double v) |
405 |
|
{ |
406 |
|
int i; |
407 |
|
|
410 |
|
} |
411 |
|
|
412 |
|
|
278 |
– |
/* Return brightness of furthest ambient sample */ |
279 |
– |
static COLORV |
280 |
– |
back_ambval(struct s_ambsamp *ap1, struct s_ambsamp *ap2, |
281 |
– |
struct s_ambsamp *ap3, FVECT orig) |
282 |
– |
{ |
283 |
– |
COLORV vback; |
284 |
– |
FVECT vec; |
285 |
– |
double d2, d2best; |
286 |
– |
|
287 |
– |
VSUB(vec, ap1->p, orig); |
288 |
– |
d2best = DOT(vec,vec); |
289 |
– |
vback = ap1->v[CIEY]; |
290 |
– |
VSUB(vec, ap2->p, orig); |
291 |
– |
d2 = DOT(vec,vec); |
292 |
– |
if (d2 > d2best) { |
293 |
– |
d2best = d2; |
294 |
– |
vback = ap2->v[CIEY]; |
295 |
– |
} |
296 |
– |
VSUB(vec, ap3->p, orig); |
297 |
– |
d2 = DOT(vec,vec); |
298 |
– |
if (d2 > d2best) |
299 |
– |
return(ap3->v[CIEY]); |
300 |
– |
return(vback); |
301 |
– |
} |
302 |
– |
|
303 |
– |
|
413 |
|
/* Compute anisotropic radii and eigenvector directions */ |
414 |
< |
static int |
414 |
> |
static void |
415 |
|
eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3]) |
416 |
|
{ |
417 |
|
double hess2[2][2]; |
427 |
|
hess2[0][1] = DOT(uv[0], b); |
428 |
|
hess2[1][0] = DOT(uv[1], a); |
429 |
|
hess2[1][1] = DOT(uv[1], b); |
430 |
< |
/* compute eigenvalues */ |
431 |
< |
if (quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1], |
432 |
< |
hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]) != 2 || |
433 |
< |
(evalue[0] = fabs(evalue[0])) <= FTINY*FTINY*FTINY || |
434 |
< |
(evalue[1] = fabs(evalue[1])) <= FTINY*FTINY*FTINY) |
435 |
< |
error(INTERNAL, "bad eigenvalue calculation"); |
436 |
< |
|
430 |
> |
/* compute eigenvalue(s) */ |
431 |
> |
i = quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1], |
432 |
> |
hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]); |
433 |
> |
if (i == 1) /* double-root (circle) */ |
434 |
> |
evalue[1] = evalue[0]; |
435 |
> |
if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) | |
436 |
> |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) { |
437 |
> |
ra[0] = ra[1] = maxarad; |
438 |
> |
return; |
439 |
> |
} |
440 |
|
if (evalue[0] > evalue[1]) { |
441 |
< |
ra[0] = 1.0/sqrt(sqrt(evalue[0])); |
442 |
< |
ra[1] = 1.0/sqrt(sqrt(evalue[1])); |
441 |
> |
ra[0] = sqrt(sqrt(4.0/evalue[0])); |
442 |
> |
ra[1] = sqrt(sqrt(4.0/evalue[1])); |
443 |
|
slope1 = evalue[1]; |
444 |
|
} else { |
445 |
< |
ra[0] = 1.0/sqrt(sqrt(evalue[1])); |
446 |
< |
ra[1] = 1.0/sqrt(sqrt(evalue[0])); |
445 |
> |
ra[0] = sqrt(sqrt(4.0/evalue[1])); |
446 |
> |
ra[1] = sqrt(sqrt(4.0/evalue[0])); |
447 |
|
slope1 = evalue[0]; |
448 |
|
} |
449 |
|
/* compute unit eigenvectors */ |
464 |
|
ambHessian( /* anisotropic radii & pos. gradient */ |
465 |
|
AMBHEMI *hp, |
466 |
|
FVECT uv[2], /* returned */ |
467 |
< |
float ra[2], /* returned */ |
468 |
< |
float pg[2] /* returned */ |
467 |
> |
float ra[2], /* returned (optional) */ |
468 |
> |
float pg[2] /* returned (optional) */ |
469 |
|
) |
470 |
|
{ |
471 |
|
static char memerrmsg[] = "out of memory in ambHessian()"; |
480 |
|
VCOPY(uv[1], hp->uy); |
481 |
|
/* clock-wise vertex traversal from sample POV */ |
482 |
|
if (ra != NULL) { /* initialize Hessian row buffer */ |
483 |
< |
hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*hp->ns); |
483 |
> |
hessrow = (FVECT (*)[3])malloc(sizeof(FVECT)*3*(hp->ns-1)); |
484 |
|
if (hessrow == NULL) |
485 |
|
error(SYSTEM, memerrmsg); |
486 |
|
memset(hessian, 0, sizeof(hessian)); |
487 |
|
} else if (pg == NULL) /* bogus call? */ |
488 |
|
return; |
489 |
|
if (pg != NULL) { /* initialize form factor row buffer */ |
490 |
< |
gradrow = (FVECT *)malloc(sizeof(FVECT)*hp->ns); |
490 |
> |
gradrow = (FVECT *)malloc(sizeof(FVECT)*(hp->ns-1)); |
491 |
|
if (gradrow == NULL) |
492 |
|
error(SYSTEM, memerrmsg); |
493 |
|
memset(gradient, 0, sizeof(gradient)); |
494 |
|
} |
495 |
|
/* compute first row of edges */ |
496 |
|
for (j = 0; j < hp->ns-1; j++) { |
497 |
< |
comp_fftri(&fftr, ambsamp(hp,0,j).p, |
386 |
< |
ambsamp(hp,0,j+1).p, hp->rp->rop); |
497 |
> |
comp_fftri(&fftr, hp, AI(hp,0,j), AI(hp,0,j+1)); |
498 |
|
if (hessrow != NULL) |
499 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
500 |
|
if (gradrow != NULL) |
504 |
|
for (i = 0; i < hp->ns-1; i++) { |
505 |
|
FVECT hesscol[3]; /* compute first vertical edge */ |
506 |
|
FVECT gradcol; |
507 |
< |
comp_fftri(&fftr, ambsamp(hp,i,0).p, |
397 |
< |
ambsamp(hp,i+1,0).p, hp->rp->rop); |
507 |
> |
comp_fftri(&fftr, hp, AI(hp,i,0), AI(hp,i+1,0)); |
508 |
|
if (hessrow != NULL) |
509 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
510 |
|
if (gradrow != NULL) |
512 |
|
for (j = 0; j < hp->ns-1; j++) { |
513 |
|
FVECT hessdia[3]; /* compute triangle contributions */ |
514 |
|
FVECT graddia; |
515 |
< |
COLORV backg; |
516 |
< |
backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1), |
517 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
515 |
> |
double backg; |
516 |
> |
backg = back_ambval(hp, AI(hp,i,j), |
517 |
> |
AI(hp,i,j+1), AI(hp,i+1,j)); |
518 |
|
/* diagonal (inner) edge */ |
519 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, |
410 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
519 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j)); |
520 |
|
if (hessrow != NULL) { |
521 |
|
comp_hessian(hessdia, &fftr, hp->rp->ron); |
522 |
|
rev_hessian(hesscol); |
523 |
|
add2hessian(hessian, hessrow[j], hessdia, hesscol, backg); |
524 |
|
} |
525 |
< |
if (gradient != NULL) { |
525 |
> |
if (gradrow != NULL) { |
526 |
|
comp_gradient(graddia, &fftr, hp->rp->ron); |
527 |
|
rev_gradient(gradcol); |
528 |
|
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
529 |
|
} |
530 |
|
/* initialize edge in next row */ |
531 |
< |
comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p, |
423 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
531 |
> |
comp_fftri(&fftr, hp, AI(hp,i+1,j+1), AI(hp,i+1,j)); |
532 |
|
if (hessrow != NULL) |
533 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
534 |
|
if (gradrow != NULL) |
535 |
|
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
536 |
|
/* new column edge & paired triangle */ |
537 |
< |
backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1), |
538 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
539 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p, |
432 |
< |
hp->rp->rop); |
537 |
> |
backg = back_ambval(hp, AI(hp,i+1,j+1), |
538 |
> |
AI(hp,i+1,j), AI(hp,i,j+1)); |
539 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j+1)); |
540 |
|
if (hessrow != NULL) { |
541 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
542 |
|
rev_hessian(hessdia); |
559 |
|
|
560 |
|
if (ra != NULL) /* extract eigenvectors & radii */ |
561 |
|
eigenvectors(uv, ra, hessian); |
562 |
< |
if (pg != NULL) { /* project position gradient */ |
562 |
> |
if (pg != NULL) { /* tangential position gradient */ |
563 |
|
pg[0] = DOT(gradient, uv[0]); |
564 |
|
pg[1] = DOT(gradient, uv[1]); |
565 |
|
} |
570 |
|
static void |
571 |
|
ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2]) |
572 |
|
{ |
573 |
< |
struct s_ambsamp *ap; |
574 |
< |
int n; |
573 |
> |
AMBSAMP *ap; |
574 |
> |
double dgsum[2]; |
575 |
> |
int n; |
576 |
> |
FVECT vd; |
577 |
> |
double gfact; |
578 |
|
|
579 |
< |
dg[0] = dg[1] = 0; |
579 |
> |
dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */ |
580 |
|
for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) { |
471 |
– |
FVECT vd; |
472 |
– |
double gfact; |
581 |
|
/* use vector for azimuth + 90deg */ |
582 |
|
VSUB(vd, ap->p, hp->rp->rop); |
583 |
< |
/* brightness with tangent factor */ |
584 |
< |
gfact = ap->v[CIEY] / DOT(hp->rp->ron, vd); |
583 |
> |
/* brightness over cosine factor */ |
584 |
> |
gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd); |
585 |
|
/* sine = proj_radius/vd_length */ |
586 |
< |
dg[0] -= DOT(uv[1], vd) * gfact ; |
587 |
< |
dg[1] += DOT(uv[0], vd) * gfact; |
586 |
> |
dgsum[0] -= DOT(uv[1], vd) * gfact; |
587 |
> |
dgsum[1] += DOT(uv[0], vd) * gfact; |
588 |
|
} |
589 |
+ |
dg[0] = dgsum[0] / (hp->ns*hp->ns); |
590 |
+ |
dg[1] = dgsum[1] / (hp->ns*hp->ns); |
591 |
|
} |
592 |
|
|
593 |
|
|
594 |
+ |
/* Compute potential light leak direction flags for cache value */ |
595 |
+ |
static uint32 |
596 |
+ |
ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1) |
597 |
+ |
{ |
598 |
+ |
const double max_d = 1.0/(minarad*ambacc + 0.001); |
599 |
+ |
const double ang_res = 0.5*PI/(hp->ns-1); |
600 |
+ |
const double ang_step = ang_res/((int)(16/PI*ang_res) + (1+FTINY)); |
601 |
+ |
double avg_d = 0; |
602 |
+ |
uint32 flgs = 0; |
603 |
+ |
FVECT vec; |
604 |
+ |
double u, v; |
605 |
+ |
double ang, a1; |
606 |
+ |
int i, j; |
607 |
+ |
/* don't bother for a few samples */ |
608 |
+ |
if (hp->ns < 12) |
609 |
+ |
return(0); |
610 |
+ |
/* check distances overhead */ |
611 |
+ |
for (i = hp->ns*3/4; i-- > hp->ns>>2; ) |
612 |
+ |
for (j = hp->ns*3/4; j-- > hp->ns>>2; ) |
613 |
+ |
avg_d += ambsam(hp,i,j).d; |
614 |
+ |
avg_d *= 4.0/(hp->ns*hp->ns); |
615 |
+ |
if (avg_d*r0 >= 1.0) /* ceiling too low for corral? */ |
616 |
+ |
return(0); |
617 |
+ |
if (avg_d >= max_d) /* insurance */ |
618 |
+ |
return(0); |
619 |
+ |
/* else circle around perimeter */ |
620 |
+ |
for (i = 0; i < hp->ns; i++) |
621 |
+ |
for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) { |
622 |
+ |
AMBSAMP *ap = &ambsam(hp,i,j); |
623 |
+ |
if ((ap->d <= FTINY) | (ap->d >= max_d)) |
624 |
+ |
continue; /* too far or too near */ |
625 |
+ |
VSUB(vec, ap->p, hp->rp->rop); |
626 |
+ |
u = DOT(vec, uv[0]) * ap->d; |
627 |
+ |
v = DOT(vec, uv[1]) * ap->d; |
628 |
+ |
if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= 1.0) |
629 |
+ |
continue; /* occluder outside ellipse */ |
630 |
+ |
ang = atan2a(v, u); /* else set direction flags */ |
631 |
+ |
for (a1 = ang-.5*ang_res; a1 <= ang+.5*ang_res; a1 += ang_step) |
632 |
+ |
flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0))); |
633 |
+ |
} |
634 |
+ |
/* add low-angle incident (< 20deg) */ |
635 |
+ |
if (fabs(hp->rp->rod) <= 0.342) { |
636 |
+ |
u = -DOT(hp->rp->rdir, uv[0]); |
637 |
+ |
v = -DOT(hp->rp->rdir, uv[1]); |
638 |
+ |
if ((r0*r0*u*u + r1*r1*v*v) > hp->rp->rot*hp->rp->rot) { |
639 |
+ |
ang = atan2a(v, u); |
640 |
+ |
ang += 2.*PI*(ang < 0); |
641 |
+ |
ang *= 16/PI; |
642 |
+ |
if ((ang < .5) | (ang >= 31.5)) |
643 |
+ |
flgs |= 0x80000001; |
644 |
+ |
else |
645 |
+ |
flgs |= 3L<<(int)(ang-.5); |
646 |
+ |
} |
647 |
+ |
} |
648 |
+ |
return(flgs); |
649 |
+ |
} |
650 |
+ |
|
651 |
+ |
|
652 |
|
int |
653 |
|
doambient( /* compute ambient component */ |
654 |
|
COLOR rcol, /* input/output color */ |
657 |
|
FVECT uv[2], /* returned (optional) */ |
658 |
|
float ra[2], /* returned (optional) */ |
659 |
|
float pg[2], /* returned (optional) */ |
660 |
< |
float dg[2] /* returned (optional) */ |
660 |
> |
float dg[2], /* returned (optional) */ |
661 |
> |
uint32 *crlp /* returned (optional) */ |
662 |
|
) |
663 |
|
{ |
664 |
< |
int cnt = 0; |
665 |
< |
FVECT my_uv[2]; |
666 |
< |
AMBHEMI *hp; |
667 |
< |
double d, acol[3]; |
668 |
< |
struct s_ambsamp *ap; |
669 |
< |
int i, j; |
670 |
< |
/* initialize */ |
671 |
< |
if ((hp = inithemi(rcol, r, wt)) == NULL) |
664 |
> |
AMBHEMI *hp = inithemi(rcol, r, wt); |
665 |
> |
int cnt; |
666 |
> |
FVECT my_uv[2]; |
667 |
> |
double d, K, acol[3]; |
668 |
> |
AMBSAMP *ap; |
669 |
> |
int i, j; |
670 |
> |
/* check/initialize */ |
671 |
> |
if (hp == NULL) |
672 |
|
return(0); |
673 |
|
if (uv != NULL) |
674 |
|
memset(uv, 0, sizeof(FVECT)*2); |
678 |
|
pg[0] = pg[1] = 0.0; |
679 |
|
if (dg != NULL) |
680 |
|
dg[0] = dg[1] = 0.0; |
681 |
+ |
if (crlp != NULL) |
682 |
+ |
*crlp = 0; |
683 |
|
/* sample the hemisphere */ |
684 |
|
acol[0] = acol[1] = acol[2] = 0.0; |
685 |
+ |
cnt = 0; |
686 |
|
for (i = hp->ns; i--; ) |
687 |
|
for (j = hp->ns; j--; ) |
688 |
< |
if (ambsample(hp, i, j)) { |
517 |
< |
ap = &ambsamp(hp,i,j); |
688 |
> |
if ((ap = ambsample(hp, i, j)) != NULL) { |
689 |
|
addcolor(acol, ap->v); |
690 |
|
++cnt; |
691 |
|
} |
692 |
< |
if (!cnt) { |
693 |
< |
setcolor(rcol, 0.0, 0.0, 0.0); |
694 |
< |
free(hp); |
695 |
< |
return(0); /* no valid samples */ |
692 |
> |
if ((hp->ns < 4) | (cnt < hp->ns*hp->ns)) { |
693 |
> |
free(hp); /* inadequate sampling */ |
694 |
> |
copycolor(rcol, acol); |
695 |
> |
return(-cnt); /* value-only result */ |
696 |
|
} |
697 |
< |
d = 1.0 / cnt; /* final indirect irradiance/PI */ |
698 |
< |
acol[0] *= d; acol[1] *= d; acol[2] *= d; |
699 |
< |
copycolor(rcol, acol); |
700 |
< |
if (cnt < hp->ns*hp->ns || /* incomplete sampling? */ |
701 |
< |
(ra == NULL) & (pg == NULL) & (dg == NULL)) { |
697 |
> |
cnt = ambssamp*wt + 0.5; /* perform super-sampling? */ |
698 |
> |
if (cnt > 8) |
699 |
> |
ambsupersamp(acol, hp, cnt); |
700 |
> |
copycolor(rcol, acol); /* final indirect irradiance/PI */ |
701 |
> |
if ((ra == NULL) & (pg == NULL) & (dg == NULL)) { |
702 |
|
free(hp); |
703 |
< |
return(-1); /* no radius or gradient calc. */ |
703 |
> |
return(-1); /* no Hessian or gradients requested */ |
704 |
|
} |
705 |
< |
d = 0.01 * bright(rcol); /* add in 1% before Hessian comp. */ |
706 |
< |
if (d < FTINY) d = FTINY; |
707 |
< |
ap = hp->sa; /* using Y channel from here on... */ |
705 |
> |
if ((d = bright(acol)) > FTINY) { /* normalize Y values */ |
706 |
> |
d = 0.99*(hp->ns*hp->ns)/d; |
707 |
> |
K = 0.01; |
708 |
> |
} else { /* or fall back on geometric Hessian */ |
709 |
> |
K = 1.0; |
710 |
> |
pg = NULL; |
711 |
> |
dg = NULL; |
712 |
> |
crlp = NULL; |
713 |
> |
} |
714 |
> |
ap = hp->sa; /* relative Y channel from here on... */ |
715 |
|
for (i = hp->ns*hp->ns; i--; ap++) |
716 |
< |
colval(ap->v,CIEY) = bright(ap->v) + d; |
716 |
> |
colval(ap->v,CIEY) = bright(ap->v)*d + K; |
717 |
|
|
718 |
|
if (uv == NULL) /* make sure we have axis pointers */ |
719 |
|
uv = my_uv; |
720 |
|
/* compute radii & pos. gradient */ |
721 |
|
ambHessian(hp, uv, ra, pg); |
722 |
+ |
|
723 |
|
if (dg != NULL) /* compute direction gradient */ |
724 |
|
ambdirgrad(hp, uv, dg); |
725 |
< |
if (ra != NULL) { /* adjust/clamp radii */ |
726 |
< |
d = sqrt(sqrt((4.0/PI)*bright(rcol)/wt)); |
727 |
< |
if ((ra[0] *= d) > maxarad) |
728 |
< |
ra[0] = maxarad; |
725 |
> |
|
726 |
> |
if (ra != NULL) { /* scale/clamp radii */ |
727 |
> |
if (pg != NULL) { |
728 |
> |
if (ra[0]*(d = fabs(pg[0])) > 1.0) |
729 |
> |
ra[0] = 1.0/d; |
730 |
> |
if (ra[1]*(d = fabs(pg[1])) > 1.0) |
731 |
> |
ra[1] = 1.0/d; |
732 |
> |
if (ra[0] > ra[1]) |
733 |
> |
ra[0] = ra[1]; |
734 |
> |
} |
735 |
> |
if (ra[0] < minarad) { |
736 |
> |
ra[0] = minarad; |
737 |
> |
if (ra[1] < minarad) |
738 |
> |
ra[1] = minarad; |
739 |
> |
} |
740 |
> |
ra[0] *= d = 1.0/sqrt(wt); |
741 |
|
if ((ra[1] *= d) > 2.0*ra[0]) |
742 |
|
ra[1] = 2.0*ra[0]; |
743 |
+ |
if (ra[1] > maxarad) { |
744 |
+ |
ra[1] = maxarad; |
745 |
+ |
if (ra[0] > maxarad) |
746 |
+ |
ra[0] = maxarad; |
747 |
+ |
} |
748 |
+ |
/* flag encroached directions */ |
749 |
+ |
if ((wt >= 0.89*AVGREFL) & (crlp != NULL)) |
750 |
+ |
*crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc); |
751 |
+ |
if (pg != NULL) { /* cap gradient if necessary */ |
752 |
+ |
d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1]; |
753 |
+ |
if (d > 1.0) { |
754 |
+ |
d = 1.0/sqrt(d); |
755 |
+ |
pg[0] *= d; |
756 |
+ |
pg[1] *= d; |
757 |
+ |
} |
758 |
+ |
} |
759 |
|
} |
760 |
|
free(hp); /* clean up and return */ |
761 |
|
return(1); |