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 MINADIV |
25 |
> |
#define MINADIV 7 /* minimum # divisions in each dimension */ |
26 |
> |
#endif |
27 |
|
|
28 |
< |
extern void SDsquare2disk(double ds[2], double seedx, double seedy); |
28 |
> |
typedef struct { |
29 |
> |
COLOR v; /* hemisphere sample value */ |
30 |
> |
float d; /* reciprocal distance */ |
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 |
< |
float p[3]; /* 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; |
49 |
< |
double nf, I1, I2, J2; |
48 |
> |
FVECT r_i, r_i1, e_i, rcp, rI2_eJ2; |
49 |
> |
double I1, I2; |
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] = 0.1 - 0.2*frandom(); |
85 |
< |
hp->uy[1] = 0.1 - 0.2*frandom(); |
86 |
< |
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); |
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 |
< |
int hlist[3]; |
101 |
< |
double spt[2], zd; |
102 |
< |
int ii; |
98 |
> |
AMBSAMP *ap = &ambsam(hp,i,j); |
99 |
> |
RAY ar; |
100 |
> |
int hlist[3], ii; |
101 |
> |
RREAL spt[2]; |
102 |
> |
double zd; |
103 |
> |
/* generate hemispherical sample */ |
104 |
|
/* ambient coefficient for weight */ |
105 |
|
if (ambacc > FTINY) |
106 |
|
setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL); |
107 |
|
else |
108 |
|
copycolor(ar.rcoef, hp->acoef); |
109 |
< |
if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0) { |
110 |
< |
setcolor(ap->v, 0., 0., 0.); |
109 |
< |
VCOPY(ap->p, hp->rp->rop); |
110 |
< |
return(NULL); /* no sample taken */ |
111 |
< |
} |
109 |
> |
if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0) |
110 |
> |
return(0); |
111 |
|
if (ambacc > FTINY) { |
112 |
|
multcolor(ar.rcoef, hp->acoef); |
113 |
|
scalecolor(ar.rcoef, 1./AVGREFL); |
114 |
|
} |
115 |
< |
/* generate hemispherical sample */ |
116 |
< |
SDsquare2disk(spt, (i+.1+.8*frandom())/hp->ns, |
117 |
< |
(j+.1+.8*frandom())/hp->ns ); |
115 |
> |
hlist[0] = hp->rp->rno; |
116 |
> |
hlist[1] = j; |
117 |
> |
hlist[2] = i; |
118 |
> |
multisamp(spt, 2, urand(ilhash(hlist,3)+n)); |
119 |
> |
resample: |
120 |
> |
square2disk(spt, (j+spt[1])/hp->ns, (i+spt[0])/hp->ns); |
121 |
|
zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]); |
122 |
|
for (ii = 3; ii--; ) |
123 |
|
ar.rdir[ii] = spt[0]*hp->ux[ii] + |
124 |
|
spt[1]*hp->uy[ii] + |
125 |
|
zd*hp->rp->ron[ii]; |
126 |
|
checknorm(ar.rdir); |
127 |
< |
dimlist[ndims++] = i*hp->ns + j + 90171; |
127 |
> |
/* avoid coincident samples */ |
128 |
> |
if (!n && ambcollision(hp, i, j, ar.rdir)) { |
129 |
> |
spt[0] = frandom(); spt[1] = frandom(); |
130 |
> |
goto resample; /* reject this sample */ |
131 |
> |
} |
132 |
> |
dimlist[ndims++] = AI(hp,i,j) + 90171; |
133 |
|
rayvalue(&ar); /* evaluate ray */ |
134 |
|
ndims--; |
135 |
+ |
zd = raydistance(&ar); |
136 |
+ |
if (zd <= FTINY) |
137 |
+ |
return(0); /* should never happen */ |
138 |
|
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
139 |
< |
copycolor(ap->v, ar.rcol); |
140 |
< |
if (ar.rt > 20.0*maxarad) /* limit vertex distance */ |
141 |
< |
ar.rt = 20.0*maxarad; |
142 |
< |
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
143 |
< |
return(ap); |
139 |
> |
if (zd*ap->d < 1.0) /* new/closer distance? */ |
140 |
> |
ap->d = 1.0/zd; |
141 |
> |
if (!n) { /* record first vertex & value */ |
142 |
> |
if (zd > 10.0*thescene.cusize + 1000.) |
143 |
> |
zd = 10.0*thescene.cusize + 1000.; |
144 |
> |
VSUM(ap->p, ar.rorg, ar.rdir, zd); |
145 |
> |
copycolor(ap->v, ar.rcol); |
146 |
> |
} else { /* else update recorded value */ |
147 |
> |
hp->acol[RED] -= colval(ap->v,RED); |
148 |
> |
hp->acol[GRN] -= colval(ap->v,GRN); |
149 |
> |
hp->acol[BLU] -= colval(ap->v,BLU); |
150 |
> |
zd = 1.0/(double)(n+1); |
151 |
> |
scalecolor(ar.rcol, zd); |
152 |
> |
zd *= (double)n; |
153 |
> |
scalecolor(ap->v, zd); |
154 |
> |
addcolor(ap->v, ar.rcol); |
155 |
> |
} |
156 |
> |
addcolor(hp->acol, ap->v); /* add to our sum */ |
157 |
> |
return(1); |
158 |
|
} |
159 |
|
|
160 |
|
|
161 |
+ |
/* Estimate variance based on ambient division differences */ |
162 |
+ |
static float * |
163 |
+ |
getambdiffs(AMBHEMI *hp) |
164 |
+ |
{ |
165 |
+ |
const double normf = 1./bright(hp->acoef); |
166 |
+ |
float *earr = (float *)calloc(hp->ns*hp->ns, sizeof(float)); |
167 |
+ |
float *ep; |
168 |
+ |
AMBSAMP *ap; |
169 |
+ |
double b, b1, d2; |
170 |
+ |
int i, j; |
171 |
+ |
|
172 |
+ |
if (earr == NULL) /* out of memory? */ |
173 |
+ |
return(NULL); |
174 |
+ |
/* sum 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 |
+ |
b1 = bright(ap[-hp->ns].v); |
180 |
+ |
d2 = b - b1; |
181 |
+ |
d2 *= d2*normf/(b + b1); |
182 |
+ |
ep[0] += d2; |
183 |
+ |
ep[-hp->ns] += d2; |
184 |
+ |
} |
185 |
+ |
if (!j) continue; |
186 |
+ |
/* from behind */ |
187 |
+ |
b1 = bright(ap[-1].v); |
188 |
+ |
d2 = b - b1; |
189 |
+ |
d2 *= d2*normf/(b + b1); |
190 |
+ |
ep[0] += d2; |
191 |
+ |
ep[-1] += d2; |
192 |
+ |
if (!i) continue; |
193 |
+ |
/* diagonal */ |
194 |
+ |
b1 = bright(ap[-hp->ns-1].v); |
195 |
+ |
d2 = b - b1; |
196 |
+ |
d2 *= d2*normf/(b + b1); |
197 |
+ |
ep[0] += d2; |
198 |
+ |
ep[-hp->ns-1] += d2; |
199 |
+ |
} |
200 |
+ |
/* correct for number of neighbors */ |
201 |
+ |
earr[0] *= 8./3.; |
202 |
+ |
earr[hp->ns-1] *= 8./3.; |
203 |
+ |
earr[(hp->ns-1)*hp->ns] *= 8./3.; |
204 |
+ |
earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 8./3.; |
205 |
+ |
for (i = 1; i < hp->ns-1; i++) { |
206 |
+ |
earr[i*hp->ns] *= 8./5.; |
207 |
+ |
earr[i*hp->ns + hp->ns-1] *= 8./5.; |
208 |
+ |
} |
209 |
+ |
for (j = 1; j < hp->ns-1; j++) { |
210 |
+ |
earr[j] *= 8./5.; |
211 |
+ |
earr[(hp->ns-1)*hp->ns + j] *= 8./5.; |
212 |
+ |
} |
213 |
+ |
return(earr); |
214 |
+ |
} |
215 |
+ |
|
216 |
+ |
|
217 |
+ |
/* Perform super-sampling on hemisphere (introduces bias) */ |
218 |
+ |
static void |
219 |
+ |
ambsupersamp(AMBHEMI *hp, int cnt) |
220 |
+ |
{ |
221 |
+ |
float *earr = getambdiffs(hp); |
222 |
+ |
double e2rem = 0; |
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 (i = 0; i < hp->ns; i++) |
233 |
+ |
for (j = 0; j < hp->ns; j++) { |
234 |
+ |
if (e2rem <= FTINY) |
235 |
+ |
goto done; /* nothing left to do */ |
236 |
+ |
nss = *ep/e2rem*cnt + frandom(); |
237 |
+ |
for (n = 1; n <= nss && ambsample(hp,i,j,n); n++) |
238 |
+ |
if (!--cnt) goto done; |
239 |
+ |
e2rem -= *ep++; /* update remainder */ |
240 |
+ |
} |
241 |
+ |
done: |
242 |
+ |
free(earr); |
243 |
+ |
} |
244 |
+ |
|
245 |
+ |
|
246 |
+ |
static AMBHEMI * |
247 |
+ |
samp_hemi( /* sample indirect hemisphere */ |
248 |
+ |
COLOR rcol, |
249 |
+ |
RAY *r, |
250 |
+ |
double wt |
251 |
+ |
) |
252 |
+ |
{ |
253 |
+ |
AMBHEMI *hp; |
254 |
+ |
double d; |
255 |
+ |
int n, i, j; |
256 |
+ |
/* insignificance check */ |
257 |
+ |
if (bright(rcol) <= FTINY) |
258 |
+ |
return(NULL); |
259 |
+ |
/* set number of divisions */ |
260 |
+ |
if (ambacc <= FTINY && |
261 |
+ |
wt > (d = 0.8*intens(rcol)*r->rweight/(ambdiv*minweight))) |
262 |
+ |
wt = d; /* avoid ray termination */ |
263 |
+ |
n = sqrt(ambdiv * wt) + 0.5; |
264 |
+ |
i = 1 + (MINADIV-1)*(ambacc > FTINY); |
265 |
+ |
if (n < i) /* use minimum number of samples? */ |
266 |
+ |
n = i; |
267 |
+ |
/* allocate sampling array */ |
268 |
+ |
hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + sizeof(AMBSAMP)*(n*n - 1)); |
269 |
+ |
if (hp == NULL) |
270 |
+ |
error(SYSTEM, "out of memory in samp_hemi"); |
271 |
+ |
hp->rp = r; |
272 |
+ |
hp->ns = n; |
273 |
+ |
hp->acol[RED] = hp->acol[GRN] = hp->acol[BLU] = 0.0; |
274 |
+ |
memset(hp->sa, 0, sizeof(AMBSAMP)*n*n); |
275 |
+ |
hp->sampOK = 0; |
276 |
+ |
/* assign coefficient */ |
277 |
+ |
copycolor(hp->acoef, rcol); |
278 |
+ |
d = 1.0/(n*n); |
279 |
+ |
scalecolor(hp->acoef, d); |
280 |
+ |
/* make tangent plane axes */ |
281 |
+ |
if (!getperpendicular(hp->ux, r->ron, 1)) |
282 |
+ |
error(CONSISTENCY, "bad ray direction in samp_hemi"); |
283 |
+ |
VCROSS(hp->uy, r->ron, hp->ux); |
284 |
+ |
/* sample divisions */ |
285 |
+ |
for (i = hp->ns; i--; ) |
286 |
+ |
for (j = hp->ns; j--; ) |
287 |
+ |
hp->sampOK += ambsample(hp, i, j, 0); |
288 |
+ |
copycolor(rcol, hp->acol); |
289 |
+ |
if (!hp->sampOK) { /* utter failure? */ |
290 |
+ |
free(hp); |
291 |
+ |
return(NULL); |
292 |
+ |
} |
293 |
+ |
if (hp->sampOK < hp->ns*hp->ns) { |
294 |
+ |
hp->sampOK *= -1; /* soft failure */ |
295 |
+ |
return(hp); |
296 |
+ |
} |
297 |
+ |
if (hp->sampOK <= MINADIV*MINADIV) |
298 |
+ |
return(hp); /* don't bother super-sampling */ |
299 |
+ |
n = ambssamp*wt + 0.5; |
300 |
+ |
if (n > 8) { /* perform super-sampling? */ |
301 |
+ |
ambsupersamp(hp, n); |
302 |
+ |
copycolor(rcol, hp->acol); |
303 |
+ |
} |
304 |
+ |
return(hp); /* all is well */ |
305 |
+ |
} |
306 |
+ |
|
307 |
+ |
|
308 |
+ |
/* Return brightness of farthest ambient sample */ |
309 |
+ |
static double |
310 |
+ |
back_ambval(AMBHEMI *hp, const int n1, const int n2, const int n3) |
311 |
+ |
{ |
312 |
+ |
if (hp->sa[n1].d <= hp->sa[n2].d) { |
313 |
+ |
if (hp->sa[n1].d <= hp->sa[n3].d) |
314 |
+ |
return(colval(hp->sa[n1].v,CIEY)); |
315 |
+ |
return(colval(hp->sa[n3].v,CIEY)); |
316 |
+ |
} |
317 |
+ |
if (hp->sa[n2].d <= hp->sa[n3].d) |
318 |
+ |
return(colval(hp->sa[n2].v,CIEY)); |
319 |
+ |
return(colval(hp->sa[n3].v,CIEY)); |
320 |
+ |
} |
321 |
+ |
|
322 |
+ |
|
323 |
|
/* Compute vectors and coefficients for Hessian/gradient calcs */ |
324 |
|
static void |
325 |
< |
comp_fftri(FFTRI *ftp, float ap0[3], float ap1[3], FVECT rop) |
325 |
> |
comp_fftri(FFTRI *ftp, AMBHEMI *hp, const int n0, const int n1) |
326 |
|
{ |
327 |
< |
FVECT v1; |
328 |
< |
double dot_e, dot_er, dot_r, dot_r1; |
327 |
> |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
328 |
> |
int ii; |
329 |
|
|
330 |
< |
VSUB(ftp->r_i, ap0, rop); |
331 |
< |
VSUB(ftp->r_i1, ap1, rop); |
332 |
< |
VSUB(ftp->e_i, ap1, ap0); |
333 |
< |
VCROSS(v1, ftp->e_i, ftp->r_i); |
334 |
< |
ftp->nf = 1.0/DOT(v1,v1); |
149 |
< |
VCROSS(v1, ftp->r_i, ftp->r_i1); |
150 |
< |
ftp->I1 = sqrt(DOT(v1,v1)*ftp->nf); |
330 |
> |
VSUB(ftp->r_i, hp->sa[n0].p, hp->rp->rop); |
331 |
> |
VSUB(ftp->r_i1, hp->sa[n1].p, hp->rp->rop); |
332 |
> |
VSUB(ftp->e_i, hp->sa[n1].p, hp->sa[n0].p); |
333 |
> |
VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1); |
334 |
> |
rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp); |
335 |
|
dot_e = DOT(ftp->e_i,ftp->e_i); |
336 |
|
dot_er = DOT(ftp->e_i, ftp->r_i); |
337 |
< |
dot_r = DOT(ftp->r_i,ftp->r_i); |
338 |
< |
dot_r1 = DOT(ftp->r_i1,ftp->r_i1); |
339 |
< |
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)/dot_r1 - dot_er/dot_r + |
340 |
< |
dot_e*ftp->I1 )*0.5*ftp->nf; |
341 |
< |
ftp->J2 = 0.25*ftp->nf*( 1.0/dot_r - 1.0/dot_r1 ) - |
342 |
< |
dot_er/dot_e*ftp->I2; |
337 |
> |
rdot_r = 1.0/DOT(ftp->r_i,ftp->r_i); |
338 |
> |
rdot_r1 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
339 |
> |
ftp->I1 = acos( DOT(ftp->r_i, ftp->r_i1) * sqrt(rdot_r*rdot_r1) ) * |
340 |
> |
sqrt( rdot_cp ); |
341 |
> |
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r + |
342 |
> |
dot_e*ftp->I1 )*0.5*rdot_cp; |
343 |
> |
J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e; |
344 |
> |
for (ii = 3; ii--; ) |
345 |
> |
ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii]; |
346 |
|
} |
347 |
|
|
348 |
|
|
363 |
|
static void |
364 |
|
comp_hessian(FVECT hess[3], FFTRI *ftp, FVECT nrm) |
365 |
|
{ |
366 |
< |
FVECT v1, v2; |
366 |
> |
FVECT ncp; |
367 |
|
FVECT m1[3], m2[3], m3[3], m4[3]; |
368 |
|
double d1, d2, d3, d4; |
369 |
|
double I3, J3, K3; |
373 |
|
d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
374 |
|
d3 = 1.0/DOT(ftp->e_i,ftp->e_i); |
375 |
|
d4 = DOT(ftp->e_i, ftp->r_i); |
376 |
< |
I3 = 0.25*ftp->nf*( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + |
377 |
< |
3.0*d3*ftp->I2 ); |
376 |
> |
I3 = ( DOT(ftp->e_i, ftp->r_i1)*d2*d2 - d4*d1*d1 + 3.0/d3*ftp->I2 ) |
377 |
> |
/ ( 4.0*DOT(ftp->rcp,ftp->rcp) ); |
378 |
|
J3 = 0.25*d3*(d1*d1 - d2*d2) - d4*d3*I3; |
379 |
|
K3 = d3*(ftp->I2 - I3/d1 - 2.0*d4*J3); |
380 |
|
/* intermediate matrices */ |
381 |
< |
VCROSS(v1, nrm, ftp->e_i); |
382 |
< |
for (j = 3; j--; ) |
196 |
< |
v2[j] = ftp->I2*ftp->r_i[j] + ftp->J2*ftp->e_i[j]; |
197 |
< |
compose_matrix(m1, v1, v2); |
381 |
> |
VCROSS(ncp, nrm, ftp->e_i); |
382 |
> |
compose_matrix(m1, ncp, ftp->rI2_eJ2); |
383 |
|
compose_matrix(m2, ftp->r_i, ftp->r_i); |
384 |
|
compose_matrix(m3, ftp->e_i, ftp->e_i); |
385 |
|
compose_matrix(m4, ftp->r_i, ftp->e_i); |
386 |
< |
VCROSS(v1, ftp->r_i, ftp->e_i); |
202 |
< |
d1 = DOT(nrm, v1); |
386 |
> |
d1 = DOT(nrm, ftp->rcp); |
387 |
|
d2 = -d1*ftp->I2; |
388 |
|
d1 *= 2.0; |
389 |
|
for (i = 3; i--; ) /* final matrix sum */ |
413 |
|
/* Add to radiometric Hessian from the given triangle */ |
414 |
|
static void |
415 |
|
add2hessian(FVECT hess[3], FVECT ehess1[3], |
416 |
< |
FVECT ehess2[3], FVECT ehess3[3], COLORV v) |
416 |
> |
FVECT ehess2[3], FVECT ehess3[3], double v) |
417 |
|
{ |
418 |
|
int i, j; |
419 |
|
|
427 |
|
static void |
428 |
|
comp_gradient(FVECT grad, FFTRI *ftp, FVECT nrm) |
429 |
|
{ |
430 |
< |
FVECT vcp; |
430 |
> |
FVECT ncp; |
431 |
|
double f1; |
432 |
|
int i; |
433 |
|
|
434 |
< |
VCROSS(vcp, ftp->r_i, ftp->r_i1); |
435 |
< |
f1 = 2.0*DOT(nrm, vcp); |
252 |
< |
VCROSS(vcp, nrm, ftp->e_i); |
434 |
> |
f1 = 2.0*DOT(nrm, ftp->rcp); |
435 |
> |
VCROSS(ncp, nrm, ftp->e_i); |
436 |
|
for (i = 3; i--; ) |
437 |
< |
grad[i] = (0.5/PI)*( ftp->I1*vcp[i] + |
255 |
< |
f1*(ftp->I2*ftp->r_i[i] + ftp->J2*ftp->e_i[i]) ); |
437 |
> |
grad[i] = (0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
438 |
|
} |
439 |
|
|
440 |
|
|
450 |
|
|
451 |
|
/* Add to displacement gradient from the given triangle */ |
452 |
|
static void |
453 |
< |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v) |
453 |
> |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, double v) |
454 |
|
{ |
455 |
|
int i; |
456 |
|
|
459 |
|
} |
460 |
|
|
461 |
|
|
280 |
– |
/* Return brightness of furthest ambient sample */ |
281 |
– |
static COLORV |
282 |
– |
back_ambval(struct s_ambsamp *ap1, struct s_ambsamp *ap2, |
283 |
– |
struct s_ambsamp *ap3, FVECT orig) |
284 |
– |
{ |
285 |
– |
COLORV vback; |
286 |
– |
FVECT vec; |
287 |
– |
double d2, d2best; |
288 |
– |
|
289 |
– |
VSUB(vec, ap1->p, orig); |
290 |
– |
d2best = DOT(vec,vec); |
291 |
– |
vback = ap1->v[CIEY]; |
292 |
– |
VSUB(vec, ap2->p, orig); |
293 |
– |
d2 = DOT(vec,vec); |
294 |
– |
if (d2 > d2best) { |
295 |
– |
d2best = d2; |
296 |
– |
vback = ap2->v[CIEY]; |
297 |
– |
} |
298 |
– |
VSUB(vec, ap3->p, orig); |
299 |
– |
d2 = DOT(vec,vec); |
300 |
– |
if (d2 > d2best) |
301 |
– |
return(ap3->v[CIEY]); |
302 |
– |
return(vback); |
303 |
– |
} |
304 |
– |
|
305 |
– |
|
462 |
|
/* Compute anisotropic radii and eigenvector directions */ |
463 |
< |
static int |
463 |
> |
static void |
464 |
|
eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3]) |
465 |
|
{ |
466 |
|
double hess2[2][2]; |
476 |
|
hess2[0][1] = DOT(uv[0], b); |
477 |
|
hess2[1][0] = DOT(uv[1], a); |
478 |
|
hess2[1][1] = DOT(uv[1], b); |
479 |
< |
/* compute eigenvalues */ |
480 |
< |
if ( quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1], |
481 |
< |
hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]) != 2 || |
482 |
< |
(evalue[0] = fabs(evalue[0])) <= FTINY*FTINY || |
483 |
< |
(evalue[1] = fabs(evalue[1])) <= FTINY*FTINY ) |
484 |
< |
error(INTERNAL, "bad eigenvalue calculation"); |
485 |
< |
|
479 |
> |
/* compute eigenvalue(s) */ |
480 |
> |
i = quadratic(evalue, 1.0, -hess2[0][0]-hess2[1][1], |
481 |
> |
hess2[0][0]*hess2[1][1]-hess2[0][1]*hess2[1][0]); |
482 |
> |
if (i == 1) /* double-root (circle) */ |
483 |
> |
evalue[1] = evalue[0]; |
484 |
> |
if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) | |
485 |
> |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) { |
486 |
> |
ra[0] = ra[1] = maxarad; |
487 |
> |
return; |
488 |
> |
} |
489 |
|
if (evalue[0] > evalue[1]) { |
490 |
< |
ra[0] = 1.0/sqrt(sqrt(evalue[0])); |
491 |
< |
ra[1] = 1.0/sqrt(sqrt(evalue[1])); |
490 |
> |
ra[0] = sqrt(sqrt(4.0/evalue[0])); |
491 |
> |
ra[1] = sqrt(sqrt(4.0/evalue[1])); |
492 |
|
slope1 = evalue[1]; |
493 |
|
} else { |
494 |
< |
ra[0] = 1.0/sqrt(sqrt(evalue[1])); |
495 |
< |
ra[1] = 1.0/sqrt(sqrt(evalue[0])); |
494 |
> |
ra[0] = sqrt(sqrt(4.0/evalue[1])); |
495 |
> |
ra[1] = sqrt(sqrt(4.0/evalue[0])); |
496 |
|
slope1 = evalue[0]; |
497 |
|
} |
498 |
|
/* compute unit eigenvectors */ |
543 |
|
} |
544 |
|
/* compute first row of edges */ |
545 |
|
for (j = 0; j < hp->ns-1; j++) { |
546 |
< |
comp_fftri(&fftr, ambsamp(hp,0,j).p, |
388 |
< |
ambsamp(hp,0,j+1).p, hp->rp->rop); |
546 |
> |
comp_fftri(&fftr, hp, AI(hp,0,j), AI(hp,0,j+1)); |
547 |
|
if (hessrow != NULL) |
548 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
549 |
|
if (gradrow != NULL) |
553 |
|
for (i = 0; i < hp->ns-1; i++) { |
554 |
|
FVECT hesscol[3]; /* compute first vertical edge */ |
555 |
|
FVECT gradcol; |
556 |
< |
comp_fftri(&fftr, ambsamp(hp,i,0).p, |
399 |
< |
ambsamp(hp,i+1,0).p, hp->rp->rop); |
556 |
> |
comp_fftri(&fftr, hp, AI(hp,i,0), AI(hp,i+1,0)); |
557 |
|
if (hessrow != NULL) |
558 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
559 |
|
if (gradrow != NULL) |
561 |
|
for (j = 0; j < hp->ns-1; j++) { |
562 |
|
FVECT hessdia[3]; /* compute triangle contributions */ |
563 |
|
FVECT graddia; |
564 |
< |
COLORV backg; |
565 |
< |
backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1), |
566 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
564 |
> |
double backg; |
565 |
> |
backg = back_ambval(hp, AI(hp,i,j), |
566 |
> |
AI(hp,i,j+1), AI(hp,i+1,j)); |
567 |
|
/* diagonal (inner) edge */ |
568 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, |
412 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
568 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j)); |
569 |
|
if (hessrow != NULL) { |
570 |
|
comp_hessian(hessdia, &fftr, hp->rp->ron); |
571 |
|
rev_hessian(hesscol); |
572 |
|
add2hessian(hessian, hessrow[j], hessdia, hesscol, backg); |
573 |
|
} |
574 |
< |
if (gradient != NULL) { |
574 |
> |
if (gradrow != NULL) { |
575 |
|
comp_gradient(graddia, &fftr, hp->rp->ron); |
576 |
|
rev_gradient(gradcol); |
577 |
|
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
578 |
|
} |
579 |
|
/* initialize edge in next row */ |
580 |
< |
comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p, |
425 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
580 |
> |
comp_fftri(&fftr, hp, AI(hp,i+1,j+1), AI(hp,i+1,j)); |
581 |
|
if (hessrow != NULL) |
582 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
583 |
|
if (gradrow != NULL) |
584 |
|
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
585 |
|
/* new column edge & paired triangle */ |
586 |
< |
backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1), |
587 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
588 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p, |
434 |
< |
hp->rp->rop); |
586 |
> |
backg = back_ambval(hp, AI(hp,i+1,j+1), |
587 |
> |
AI(hp,i+1,j), AI(hp,i,j+1)); |
588 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j+1)); |
589 |
|
if (hessrow != NULL) { |
590 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
591 |
|
rev_hessian(hessdia); |
608 |
|
|
609 |
|
if (ra != NULL) /* extract eigenvectors & radii */ |
610 |
|
eigenvectors(uv, ra, hessian); |
611 |
< |
if (pg != NULL) { /* project position gradient */ |
611 |
> |
if (pg != NULL) { /* tangential position gradient */ |
612 |
|
pg[0] = DOT(gradient, uv[0]); |
613 |
|
pg[1] = DOT(gradient, uv[1]); |
614 |
|
} |
619 |
|
static void |
620 |
|
ambdirgrad(AMBHEMI *hp, FVECT uv[2], float dg[2]) |
621 |
|
{ |
622 |
< |
struct s_ambsamp *ap; |
623 |
< |
int n; |
624 |
< |
FVECT vd; |
625 |
< |
double gfact; |
622 |
> |
AMBSAMP *ap; |
623 |
> |
double dgsum[2]; |
624 |
> |
int n; |
625 |
> |
FVECT vd; |
626 |
> |
double gfact; |
627 |
|
|
628 |
< |
dg[0] = dg[1] = 0; |
628 |
> |
dgsum[0] = dgsum[1] = 0.0; /* sum values times -tan(theta) */ |
629 |
|
for (ap = hp->sa, n = hp->ns*hp->ns; n--; ap++) { |
630 |
|
/* use vector for azimuth + 90deg */ |
631 |
|
VSUB(vd, ap->p, hp->rp->rop); |
632 |
< |
/* brightness with tangent factor */ |
633 |
< |
gfact = ap->v[CIEY] / DOT(hp->rp->ron, vd); |
632 |
> |
/* brightness over cosine factor */ |
633 |
> |
gfact = colval(ap->v,CIEY) / DOT(hp->rp->ron, vd); |
634 |
|
/* sine = proj_radius/vd_length */ |
635 |
< |
dg[0] -= DOT(uv[1], vd) * gfact; |
636 |
< |
dg[1] += DOT(uv[0], vd) * gfact; |
635 |
> |
dgsum[0] -= DOT(uv[1], vd) * gfact; |
636 |
> |
dgsum[1] += DOT(uv[0], vd) * gfact; |
637 |
|
} |
638 |
+ |
dg[0] = dgsum[0] / (hp->ns*hp->ns); |
639 |
+ |
dg[1] = dgsum[1] / (hp->ns*hp->ns); |
640 |
|
} |
641 |
|
|
642 |
|
|
643 |
+ |
/* Compute potential light leak direction flags for cache value */ |
644 |
+ |
static uint32 |
645 |
+ |
ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1) |
646 |
+ |
{ |
647 |
+ |
const double max_d = 1.0/(minarad*ambacc + 0.001); |
648 |
+ |
const double ang_res = 0.5*PI/hp->ns; |
649 |
+ |
const double ang_step = ang_res/((int)(16/PI*ang_res) + 1.01); |
650 |
+ |
double avg_d = 0; |
651 |
+ |
uint32 flgs = 0; |
652 |
+ |
FVECT vec; |
653 |
+ |
double u, v; |
654 |
+ |
double ang, a1; |
655 |
+ |
int i, j; |
656 |
+ |
/* don't bother for a few samples */ |
657 |
+ |
if (hp->ns < 8) |
658 |
+ |
return(0); |
659 |
+ |
/* check distances overhead */ |
660 |
+ |
for (i = hp->ns*3/4; i-- > hp->ns>>2; ) |
661 |
+ |
for (j = hp->ns*3/4; j-- > hp->ns>>2; ) |
662 |
+ |
avg_d += ambsam(hp,i,j).d; |
663 |
+ |
avg_d *= 4.0/(hp->ns*hp->ns); |
664 |
+ |
if (avg_d*r0 >= 1.0) /* ceiling too low for corral? */ |
665 |
+ |
return(0); |
666 |
+ |
if (avg_d >= max_d) /* insurance */ |
667 |
+ |
return(0); |
668 |
+ |
/* else circle around perimeter */ |
669 |
+ |
for (i = 0; i < hp->ns; i++) |
670 |
+ |
for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) { |
671 |
+ |
AMBSAMP *ap = &ambsam(hp,i,j); |
672 |
+ |
if ((ap->d <= FTINY) | (ap->d >= max_d)) |
673 |
+ |
continue; /* too far or too near */ |
674 |
+ |
VSUB(vec, ap->p, hp->rp->rop); |
675 |
+ |
u = DOT(vec, uv[0]); |
676 |
+ |
v = DOT(vec, uv[1]); |
677 |
+ |
if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= u*u + v*v) |
678 |
+ |
continue; /* occluder outside ellipse */ |
679 |
+ |
ang = atan2a(v, u); /* else set direction flags */ |
680 |
+ |
for (a1 = ang-ang_res; a1 <= ang+ang_res; a1 += ang_step) |
681 |
+ |
flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0))); |
682 |
+ |
} |
683 |
+ |
return(flgs); |
684 |
+ |
} |
685 |
+ |
|
686 |
+ |
|
687 |
|
int |
688 |
|
doambient( /* compute ambient component */ |
689 |
|
COLOR rcol, /* input/output color */ |
692 |
|
FVECT uv[2], /* returned (optional) */ |
693 |
|
float ra[2], /* returned (optional) */ |
694 |
|
float pg[2], /* returned (optional) */ |
695 |
< |
float dg[2] /* returned (optional) */ |
695 |
> |
float dg[2], /* returned (optional) */ |
696 |
> |
uint32 *crlp /* returned (optional) */ |
697 |
|
) |
698 |
|
{ |
699 |
< |
AMBHEMI *hp = inithemi(rcol, r, wt); |
700 |
< |
int cnt = 0; |
701 |
< |
FVECT my_uv[2]; |
702 |
< |
double d, acol[3]; |
703 |
< |
struct s_ambsamp *ap; |
704 |
< |
int i, j; |
503 |
< |
/* check/initialize */ |
504 |
< |
if (hp == NULL) |
505 |
< |
return(0); |
699 |
> |
AMBHEMI *hp = samp_hemi(rcol, r, wt); |
700 |
> |
FVECT my_uv[2]; |
701 |
> |
double d, K; |
702 |
> |
AMBSAMP *ap; |
703 |
> |
int i; |
704 |
> |
/* clear return values */ |
705 |
|
if (uv != NULL) |
706 |
|
memset(uv, 0, sizeof(FVECT)*2); |
707 |
|
if (ra != NULL) |
710 |
|
pg[0] = pg[1] = 0.0; |
711 |
|
if (dg != NULL) |
712 |
|
dg[0] = dg[1] = 0.0; |
713 |
< |
/* sample the hemisphere */ |
714 |
< |
acol[0] = acol[1] = acol[2] = 0.0; |
715 |
< |
for (i = hp->ns; i--; ) |
716 |
< |
for (j = hp->ns; j--; ) |
717 |
< |
if ((ap = ambsample(hp, i, j)) != NULL) { |
718 |
< |
addcolor(acol, ap->v); |
719 |
< |
++cnt; |
720 |
< |
} |
721 |
< |
if (!cnt) { |
523 |
< |
setcolor(rcol, 0.0, 0.0, 0.0); |
524 |
< |
free(hp); |
525 |
< |
return(0); /* no valid samples */ |
713 |
> |
if (crlp != NULL) |
714 |
> |
*crlp = 0; |
715 |
> |
if (hp == NULL) /* sampling falure? */ |
716 |
> |
return(0); |
717 |
> |
|
718 |
> |
if ((ra == NULL) & (pg == NULL) & (dg == NULL) || |
719 |
> |
(hp->sampOK < 0) | (hp->ns < MINADIV)) { |
720 |
> |
free(hp); /* Hessian not requested/possible */ |
721 |
> |
return(-1); /* value-only return value */ |
722 |
|
} |
723 |
< |
d = 1.0 / cnt; /* final indirect irradiance/PI */ |
724 |
< |
acol[0] *= d; acol[1] *= d; acol[2] *= d; |
725 |
< |
copycolor(rcol, acol); |
726 |
< |
if (cnt < hp->ns*hp->ns || /* incomplete sampling? */ |
727 |
< |
(ra == NULL) & (pg == NULL) & (dg == NULL)) { |
728 |
< |
free(hp); |
729 |
< |
return(-1); /* no radius or gradient calc. */ |
723 |
> |
if ((d = bright(rcol)) > FTINY) { /* normalize Y values */ |
724 |
> |
d = 0.99*(hp->ns*hp->ns)/d; |
725 |
> |
K = 0.01; |
726 |
> |
} else { /* or fall back on geometric Hessian */ |
727 |
> |
K = 1.0; |
728 |
> |
pg = NULL; |
729 |
> |
dg = NULL; |
730 |
> |
crlp = NULL; |
731 |
|
} |
732 |
< |
d = 0.01 * bright(rcol); /* add in 1% before Hessian comp. */ |
536 |
< |
if (d < FTINY) d = FTINY; |
537 |
< |
ap = hp->sa; /* using Y channel from here on... */ |
732 |
> |
ap = hp->sa; /* relative Y channel from here on... */ |
733 |
|
for (i = hp->ns*hp->ns; i--; ap++) |
734 |
< |
colval(ap->v,CIEY) = bright(ap->v) + d; |
734 |
> |
colval(ap->v,CIEY) = bright(ap->v)*d + K; |
735 |
|
|
736 |
|
if (uv == NULL) /* make sure we have axis pointers */ |
737 |
|
uv = my_uv; |
738 |
|
/* compute radii & pos. gradient */ |
739 |
|
ambHessian(hp, uv, ra, pg); |
740 |
+ |
|
741 |
|
if (dg != NULL) /* compute direction gradient */ |
742 |
|
ambdirgrad(hp, uv, dg); |
743 |
+ |
|
744 |
|
if (ra != NULL) { /* scale/clamp radii */ |
745 |
< |
d = sqrt(sqrt((4.0/PI)*bright(rcol)/wt)); |
746 |
< |
ra[0] *= d; |
745 |
> |
if (pg != NULL) { |
746 |
> |
if (ra[0]*(d = fabs(pg[0])) > 1.0) |
747 |
> |
ra[0] = 1.0/d; |
748 |
> |
if (ra[1]*(d = fabs(pg[1])) > 1.0) |
749 |
> |
ra[1] = 1.0/d; |
750 |
> |
if (ra[0] > ra[1]) |
751 |
> |
ra[0] = ra[1]; |
752 |
> |
} |
753 |
> |
if (ra[0] < minarad) { |
754 |
> |
ra[0] = minarad; |
755 |
> |
if (ra[1] < minarad) |
756 |
> |
ra[1] = minarad; |
757 |
> |
} |
758 |
> |
ra[0] *= d = 1.0/sqrt(wt); |
759 |
|
if ((ra[1] *= d) > 2.0*ra[0]) |
760 |
|
ra[1] = 2.0*ra[0]; |
761 |
|
if (ra[1] > maxarad) { |
763 |
|
if (ra[0] > maxarad) |
764 |
|
ra[0] = maxarad; |
765 |
|
} |
766 |
+ |
/* flag encroached directions */ |
767 |
+ |
if (crlp != NULL) /* XXX doesn't update with changes to ambacc */ |
768 |
+ |
*crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc); |
769 |
+ |
if (pg != NULL) { /* cap gradient if necessary */ |
770 |
+ |
d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1]; |
771 |
+ |
if (d > 1.0) { |
772 |
+ |
d = 1.0/sqrt(d); |
773 |
+ |
pg[0] *= d; |
774 |
+ |
pg[1] *= d; |
775 |
+ |
} |
776 |
+ |
} |
777 |
|
} |
778 |
|
free(hp); /* clean up and return */ |
779 |
|
return(1); |
780 |
|
} |
561 |
– |
|
562 |
– |
|
563 |
– |
#else /* ! NEWAMB */ |
564 |
– |
|
565 |
– |
|
566 |
– |
void |
567 |
– |
inithemi( /* initialize sampling hemisphere */ |
568 |
– |
AMBHEMI *hp, |
569 |
– |
COLOR ac, |
570 |
– |
RAY *r, |
571 |
– |
double wt |
572 |
– |
) |
573 |
– |
{ |
574 |
– |
double d; |
575 |
– |
int i; |
576 |
– |
/* set number of divisions */ |
577 |
– |
if (ambacc <= FTINY && |
578 |
– |
wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight))) |
579 |
– |
wt = d; /* avoid ray termination */ |
580 |
– |
hp->nt = sqrt(ambdiv * wt / PI) + 0.5; |
581 |
– |
i = ambacc > FTINY ? 3 : 1; /* minimum number of samples */ |
582 |
– |
if (hp->nt < i) |
583 |
– |
hp->nt = i; |
584 |
– |
hp->np = PI * hp->nt + 0.5; |
585 |
– |
/* set number of super-samples */ |
586 |
– |
hp->ns = ambssamp * wt + 0.5; |
587 |
– |
/* assign coefficient */ |
588 |
– |
copycolor(hp->acoef, ac); |
589 |
– |
d = 1.0/(hp->nt*hp->np); |
590 |
– |
scalecolor(hp->acoef, d); |
591 |
– |
/* make axes */ |
592 |
– |
VCOPY(hp->uz, r->ron); |
593 |
– |
hp->uy[0] = hp->uy[1] = hp->uy[2] = 0.0; |
594 |
– |
for (i = 0; i < 3; i++) |
595 |
– |
if (hp->uz[i] < 0.6 && hp->uz[i] > -0.6) |
596 |
– |
break; |
597 |
– |
if (i >= 3) |
598 |
– |
error(CONSISTENCY, "bad ray direction in inithemi"); |
599 |
– |
hp->uy[i] = 1.0; |
600 |
– |
fcross(hp->ux, hp->uy, hp->uz); |
601 |
– |
normalize(hp->ux); |
602 |
– |
fcross(hp->uy, hp->uz, hp->ux); |
603 |
– |
} |
604 |
– |
|
605 |
– |
|
606 |
– |
int |
607 |
– |
divsample( /* sample a division */ |
608 |
– |
AMBSAMP *dp, |
609 |
– |
AMBHEMI *h, |
610 |
– |
RAY *r |
611 |
– |
) |
612 |
– |
{ |
613 |
– |
RAY ar; |
614 |
– |
int hlist[3]; |
615 |
– |
double spt[2]; |
616 |
– |
double xd, yd, zd; |
617 |
– |
double b2; |
618 |
– |
double phi; |
619 |
– |
int i; |
620 |
– |
/* ambient coefficient for weight */ |
621 |
– |
if (ambacc > FTINY) |
622 |
– |
setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL); |
623 |
– |
else |
624 |
– |
copycolor(ar.rcoef, h->acoef); |
625 |
– |
if (rayorigin(&ar, AMBIENT, r, ar.rcoef) < 0) |
626 |
– |
return(-1); |
627 |
– |
if (ambacc > FTINY) { |
628 |
– |
multcolor(ar.rcoef, h->acoef); |
629 |
– |
scalecolor(ar.rcoef, 1./AVGREFL); |
630 |
– |
} |
631 |
– |
hlist[0] = r->rno; |
632 |
– |
hlist[1] = dp->t; |
633 |
– |
hlist[2] = dp->p; |
634 |
– |
multisamp(spt, 2, urand(ilhash(hlist,3)+dp->n)); |
635 |
– |
zd = sqrt((dp->t + spt[0])/h->nt); |
636 |
– |
phi = 2.0*PI * (dp->p + spt[1])/h->np; |
637 |
– |
xd = tcos(phi) * zd; |
638 |
– |
yd = tsin(phi) * zd; |
639 |
– |
zd = sqrt(1.0 - zd*zd); |
640 |
– |
for (i = 0; i < 3; i++) |
641 |
– |
ar.rdir[i] = xd*h->ux[i] + |
642 |
– |
yd*h->uy[i] + |
643 |
– |
zd*h->uz[i]; |
644 |
– |
checknorm(ar.rdir); |
645 |
– |
dimlist[ndims++] = dp->t*h->np + dp->p + 90171; |
646 |
– |
rayvalue(&ar); |
647 |
– |
ndims--; |
648 |
– |
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
649 |
– |
addcolor(dp->v, ar.rcol); |
650 |
– |
/* use rt to improve gradient calc */ |
651 |
– |
if (ar.rt > FTINY && ar.rt < FHUGE) |
652 |
– |
dp->r += 1.0/ar.rt; |
653 |
– |
/* (re)initialize error */ |
654 |
– |
if (dp->n++) { |
655 |
– |
b2 = bright(dp->v)/dp->n - bright(ar.rcol); |
656 |
– |
b2 = b2*b2 + dp->k*((dp->n-1)*(dp->n-1)); |
657 |
– |
dp->k = b2/(dp->n*dp->n); |
658 |
– |
} else |
659 |
– |
dp->k = 0.0; |
660 |
– |
return(0); |
661 |
– |
} |
662 |
– |
|
663 |
– |
|
664 |
– |
static int |
665 |
– |
ambcmp( /* decreasing order */ |
666 |
– |
const void *p1, |
667 |
– |
const void *p2 |
668 |
– |
) |
669 |
– |
{ |
670 |
– |
const AMBSAMP *d1 = (const AMBSAMP *)p1; |
671 |
– |
const AMBSAMP *d2 = (const AMBSAMP *)p2; |
672 |
– |
|
673 |
– |
if (d1->k < d2->k) |
674 |
– |
return(1); |
675 |
– |
if (d1->k > d2->k) |
676 |
– |
return(-1); |
677 |
– |
return(0); |
678 |
– |
} |
679 |
– |
|
680 |
– |
|
681 |
– |
static int |
682 |
– |
ambnorm( /* standard order */ |
683 |
– |
const void *p1, |
684 |
– |
const void *p2 |
685 |
– |
) |
686 |
– |
{ |
687 |
– |
const AMBSAMP *d1 = (const AMBSAMP *)p1; |
688 |
– |
const AMBSAMP *d2 = (const AMBSAMP *)p2; |
689 |
– |
int c; |
690 |
– |
|
691 |
– |
if ( (c = d1->t - d2->t) ) |
692 |
– |
return(c); |
693 |
– |
return(d1->p - d2->p); |
694 |
– |
} |
695 |
– |
|
696 |
– |
|
697 |
– |
double |
698 |
– |
doambient( /* compute ambient component */ |
699 |
– |
COLOR rcol, |
700 |
– |
RAY *r, |
701 |
– |
double wt, |
702 |
– |
FVECT pg, |
703 |
– |
FVECT dg |
704 |
– |
) |
705 |
– |
{ |
706 |
– |
double b, d=0; |
707 |
– |
AMBHEMI hemi; |
708 |
– |
AMBSAMP *div; |
709 |
– |
AMBSAMP dnew; |
710 |
– |
double acol[3]; |
711 |
– |
AMBSAMP *dp; |
712 |
– |
double arad; |
713 |
– |
int divcnt; |
714 |
– |
int i, j; |
715 |
– |
/* initialize hemisphere */ |
716 |
– |
inithemi(&hemi, rcol, r, wt); |
717 |
– |
divcnt = hemi.nt * hemi.np; |
718 |
– |
/* initialize */ |
719 |
– |
if (pg != NULL) |
720 |
– |
pg[0] = pg[1] = pg[2] = 0.0; |
721 |
– |
if (dg != NULL) |
722 |
– |
dg[0] = dg[1] = dg[2] = 0.0; |
723 |
– |
setcolor(rcol, 0.0, 0.0, 0.0); |
724 |
– |
if (divcnt == 0) |
725 |
– |
return(0.0); |
726 |
– |
/* allocate super-samples */ |
727 |
– |
if (hemi.ns > 0 || pg != NULL || dg != NULL) { |
728 |
– |
div = (AMBSAMP *)malloc(divcnt*sizeof(AMBSAMP)); |
729 |
– |
if (div == NULL) |
730 |
– |
error(SYSTEM, "out of memory in doambient"); |
731 |
– |
} else |
732 |
– |
div = NULL; |
733 |
– |
/* sample the divisions */ |
734 |
– |
arad = 0.0; |
735 |
– |
acol[0] = acol[1] = acol[2] = 0.0; |
736 |
– |
if ((dp = div) == NULL) |
737 |
– |
dp = &dnew; |
738 |
– |
divcnt = 0; |
739 |
– |
for (i = 0; i < hemi.nt; i++) |
740 |
– |
for (j = 0; j < hemi.np; j++) { |
741 |
– |
dp->t = i; dp->p = j; |
742 |
– |
setcolor(dp->v, 0.0, 0.0, 0.0); |
743 |
– |
dp->r = 0.0; |
744 |
– |
dp->n = 0; |
745 |
– |
if (divsample(dp, &hemi, r) < 0) { |
746 |
– |
if (div != NULL) |
747 |
– |
dp++; |
748 |
– |
continue; |
749 |
– |
} |
750 |
– |
arad += dp->r; |
751 |
– |
divcnt++; |
752 |
– |
if (div != NULL) |
753 |
– |
dp++; |
754 |
– |
else |
755 |
– |
addcolor(acol, dp->v); |
756 |
– |
} |
757 |
– |
if (!divcnt) { |
758 |
– |
if (div != NULL) |
759 |
– |
free((void *)div); |
760 |
– |
return(0.0); /* no samples taken */ |
761 |
– |
} |
762 |
– |
if (divcnt < hemi.nt*hemi.np) { |
763 |
– |
pg = dg = NULL; /* incomplete sampling */ |
764 |
– |
hemi.ns = 0; |
765 |
– |
} else if (arad > FTINY && divcnt/arad < minarad) { |
766 |
– |
hemi.ns = 0; /* close enough */ |
767 |
– |
} else if (hemi.ns > 0) { /* else perform super-sampling? */ |
768 |
– |
comperrs(div, &hemi); /* compute errors */ |
769 |
– |
qsort(div, divcnt, sizeof(AMBSAMP), ambcmp); /* sort divs */ |
770 |
– |
/* super-sample */ |
771 |
– |
for (i = hemi.ns; i > 0; i--) { |
772 |
– |
dnew = *div; |
773 |
– |
if (divsample(&dnew, &hemi, r) < 0) { |
774 |
– |
dp++; |
775 |
– |
continue; |
776 |
– |
} |
777 |
– |
dp = div; /* reinsert */ |
778 |
– |
j = divcnt < i ? divcnt : i; |
779 |
– |
while (--j > 0 && dnew.k < dp[1].k) { |
780 |
– |
*dp = *(dp+1); |
781 |
– |
dp++; |
782 |
– |
} |
783 |
– |
*dp = dnew; |
784 |
– |
} |
785 |
– |
if (pg != NULL || dg != NULL) /* restore order */ |
786 |
– |
qsort(div, divcnt, sizeof(AMBSAMP), ambnorm); |
787 |
– |
} |
788 |
– |
/* compute returned values */ |
789 |
– |
if (div != NULL) { |
790 |
– |
arad = 0.0; /* note: divcnt may be < nt*np */ |
791 |
– |
for (i = hemi.nt*hemi.np, dp = div; i-- > 0; dp++) { |
792 |
– |
arad += dp->r; |
793 |
– |
if (dp->n > 1) { |
794 |
– |
b = 1.0/dp->n; |
795 |
– |
scalecolor(dp->v, b); |
796 |
– |
dp->r *= b; |
797 |
– |
dp->n = 1; |
798 |
– |
} |
799 |
– |
addcolor(acol, dp->v); |
800 |
– |
} |
801 |
– |
b = bright(acol); |
802 |
– |
if (b > FTINY) { |
803 |
– |
b = 1.0/b; /* compute & normalize gradient(s) */ |
804 |
– |
if (pg != NULL) { |
805 |
– |
posgradient(pg, div, &hemi); |
806 |
– |
for (i = 0; i < 3; i++) |
807 |
– |
pg[i] *= b; |
808 |
– |
} |
809 |
– |
if (dg != NULL) { |
810 |
– |
dirgradient(dg, div, &hemi); |
811 |
– |
for (i = 0; i < 3; i++) |
812 |
– |
dg[i] *= b; |
813 |
– |
} |
814 |
– |
} |
815 |
– |
free((void *)div); |
816 |
– |
} |
817 |
– |
copycolor(rcol, acol); |
818 |
– |
if (arad <= FTINY) |
819 |
– |
arad = maxarad; |
820 |
– |
else |
821 |
– |
arad = (divcnt+hemi.ns)/arad; |
822 |
– |
if (pg != NULL) { /* reduce radius if gradient large */ |
823 |
– |
d = DOT(pg,pg); |
824 |
– |
if (d*arad*arad > 1.0) |
825 |
– |
arad = 1.0/sqrt(d); |
826 |
– |
} |
827 |
– |
if (arad < minarad) { |
828 |
– |
arad = minarad; |
829 |
– |
if (pg != NULL && d*arad*arad > 1.0) { /* cap gradient */ |
830 |
– |
d = 1.0/arad/sqrt(d); |
831 |
– |
for (i = 0; i < 3; i++) |
832 |
– |
pg[i] *= d; |
833 |
– |
} |
834 |
– |
} |
835 |
– |
if ((arad /= sqrt(wt)) > maxarad) |
836 |
– |
arad = maxarad; |
837 |
– |
return(arad); |
838 |
– |
} |
839 |
– |
|
840 |
– |
|
841 |
– |
void |
842 |
– |
comperrs( /* compute initial error estimates */ |
843 |
– |
AMBSAMP *da, /* assumes standard ordering */ |
844 |
– |
AMBHEMI *hp |
845 |
– |
) |
846 |
– |
{ |
847 |
– |
double b, b2; |
848 |
– |
int i, j; |
849 |
– |
AMBSAMP *dp; |
850 |
– |
/* sum differences from neighbors */ |
851 |
– |
dp = da; |
852 |
– |
for (i = 0; i < hp->nt; i++) |
853 |
– |
for (j = 0; j < hp->np; j++) { |
854 |
– |
#ifdef DEBUG |
855 |
– |
if (dp->t != i || dp->p != j) |
856 |
– |
error(CONSISTENCY, |
857 |
– |
"division order in comperrs"); |
858 |
– |
#endif |
859 |
– |
b = bright(dp[0].v); |
860 |
– |
if (i > 0) { /* from above */ |
861 |
– |
b2 = bright(dp[-hp->np].v) - b; |
862 |
– |
b2 *= b2 * 0.25; |
863 |
– |
dp[0].k += b2; |
864 |
– |
dp[-hp->np].k += b2; |
865 |
– |
} |
866 |
– |
if (j > 0) { /* from behind */ |
867 |
– |
b2 = bright(dp[-1].v) - b; |
868 |
– |
b2 *= b2 * 0.25; |
869 |
– |
dp[0].k += b2; |
870 |
– |
dp[-1].k += b2; |
871 |
– |
} else { /* around */ |
872 |
– |
b2 = bright(dp[hp->np-1].v) - b; |
873 |
– |
b2 *= b2 * 0.25; |
874 |
– |
dp[0].k += b2; |
875 |
– |
dp[hp->np-1].k += b2; |
876 |
– |
} |
877 |
– |
dp++; |
878 |
– |
} |
879 |
– |
/* divide by number of neighbors */ |
880 |
– |
dp = da; |
881 |
– |
for (j = 0; j < hp->np; j++) /* top row */ |
882 |
– |
(dp++)->k *= 1.0/3.0; |
883 |
– |
if (hp->nt < 2) |
884 |
– |
return; |
885 |
– |
for (i = 1; i < hp->nt-1; i++) /* central region */ |
886 |
– |
for (j = 0; j < hp->np; j++) |
887 |
– |
(dp++)->k *= 0.25; |
888 |
– |
for (j = 0; j < hp->np; j++) /* bottom row */ |
889 |
– |
(dp++)->k *= 1.0/3.0; |
890 |
– |
} |
891 |
– |
|
892 |
– |
|
893 |
– |
void |
894 |
– |
posgradient( /* compute position gradient */ |
895 |
– |
FVECT gv, |
896 |
– |
AMBSAMP *da, /* assumes standard ordering */ |
897 |
– |
AMBHEMI *hp |
898 |
– |
) |
899 |
– |
{ |
900 |
– |
int i, j; |
901 |
– |
double nextsine, lastsine, b, d; |
902 |
– |
double mag0, mag1; |
903 |
– |
double phi, cosp, sinp, xd, yd; |
904 |
– |
AMBSAMP *dp; |
905 |
– |
|
906 |
– |
xd = yd = 0.0; |
907 |
– |
for (j = 0; j < hp->np; j++) { |
908 |
– |
dp = da + j; |
909 |
– |
mag0 = mag1 = 0.0; |
910 |
– |
lastsine = 0.0; |
911 |
– |
for (i = 0; i < hp->nt; i++) { |
912 |
– |
#ifdef DEBUG |
913 |
– |
if (dp->t != i || dp->p != j) |
914 |
– |
error(CONSISTENCY, |
915 |
– |
"division order in posgradient"); |
916 |
– |
#endif |
917 |
– |
b = bright(dp->v); |
918 |
– |
if (i > 0) { |
919 |
– |
d = dp[-hp->np].r; |
920 |
– |
if (dp[0].r > d) d = dp[0].r; |
921 |
– |
/* sin(t)*cos(t)^2 */ |
922 |
– |
d *= lastsine * (1.0 - (double)i/hp->nt); |
923 |
– |
mag0 += d*(b - bright(dp[-hp->np].v)); |
924 |
– |
} |
925 |
– |
nextsine = sqrt((double)(i+1)/hp->nt); |
926 |
– |
if (j > 0) { |
927 |
– |
d = dp[-1].r; |
928 |
– |
if (dp[0].r > d) d = dp[0].r; |
929 |
– |
mag1 += d * (nextsine - lastsine) * |
930 |
– |
(b - bright(dp[-1].v)); |
931 |
– |
} else { |
932 |
– |
d = dp[hp->np-1].r; |
933 |
– |
if (dp[0].r > d) d = dp[0].r; |
934 |
– |
mag1 += d * (nextsine - lastsine) * |
935 |
– |
(b - bright(dp[hp->np-1].v)); |
936 |
– |
} |
937 |
– |
dp += hp->np; |
938 |
– |
lastsine = nextsine; |
939 |
– |
} |
940 |
– |
mag0 *= 2.0*PI / hp->np; |
941 |
– |
phi = 2.0*PI * (double)j/hp->np; |
942 |
– |
cosp = tcos(phi); sinp = tsin(phi); |
943 |
– |
xd += mag0*cosp - mag1*sinp; |
944 |
– |
yd += mag0*sinp + mag1*cosp; |
945 |
– |
} |
946 |
– |
for (i = 0; i < 3; i++) |
947 |
– |
gv[i] = (xd*hp->ux[i] + yd*hp->uy[i])*(hp->nt*hp->np)/PI; |
948 |
– |
} |
949 |
– |
|
950 |
– |
|
951 |
– |
void |
952 |
– |
dirgradient( /* compute direction gradient */ |
953 |
– |
FVECT gv, |
954 |
– |
AMBSAMP *da, /* assumes standard ordering */ |
955 |
– |
AMBHEMI *hp |
956 |
– |
) |
957 |
– |
{ |
958 |
– |
int i, j; |
959 |
– |
double mag; |
960 |
– |
double phi, xd, yd; |
961 |
– |
AMBSAMP *dp; |
962 |
– |
|
963 |
– |
xd = yd = 0.0; |
964 |
– |
for (j = 0; j < hp->np; j++) { |
965 |
– |
dp = da + j; |
966 |
– |
mag = 0.0; |
967 |
– |
for (i = 0; i < hp->nt; i++) { |
968 |
– |
#ifdef DEBUG |
969 |
– |
if (dp->t != i || dp->p != j) |
970 |
– |
error(CONSISTENCY, |
971 |
– |
"division order in dirgradient"); |
972 |
– |
#endif |
973 |
– |
/* tan(t) */ |
974 |
– |
mag += bright(dp->v)/sqrt(hp->nt/(i+.5) - 1.0); |
975 |
– |
dp += hp->np; |
976 |
– |
} |
977 |
– |
phi = 2.0*PI * (j+.5)/hp->np + PI/2.0; |
978 |
– |
xd += mag * tcos(phi); |
979 |
– |
yd += mag * tsin(phi); |
980 |
– |
} |
981 |
– |
for (i = 0; i < 3; i++) |
982 |
– |
gv[i] = xd*hp->ux[i] + yd*hp->uy[i]; |
983 |
– |
} |
984 |
– |
|
985 |
– |
#endif /* ! NEWAMB */ |