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 |
– |
/* vertex direction bit positions */ |
29 |
– |
#define VDB_xy 0 |
30 |
– |
#define VDB_y 01 |
31 |
– |
#define VDB_x 02 |
32 |
– |
#define VDB_Xy 03 |
33 |
– |
#define VDB_xY 04 |
34 |
– |
#define VDB_X 05 |
35 |
– |
#define VDB_Y 06 |
36 |
– |
#define VDB_XY 07 |
37 |
– |
/* get opposite vertex direction bit */ |
38 |
– |
#define VDB_OPP(f) (~(f) & 07) |
39 |
– |
/* adjacent triangle vertex flags */ |
40 |
– |
static const int adjacent_trifl[8] = { |
41 |
– |
0, /* forbidden diagonal */ |
42 |
– |
1<<VDB_x|1<<VDB_y|1<<VDB_Xy, |
43 |
– |
1<<VDB_y|1<<VDB_x|1<<VDB_xY, |
44 |
– |
1<<VDB_y|1<<VDB_Xy|1<<VDB_X, |
45 |
– |
1<<VDB_x|1<<VDB_xY|1<<VDB_Y, |
46 |
– |
1<<VDB_Xy|1<<VDB_X|1<<VDB_Y, |
47 |
– |
1<<VDB_xY|1<<VDB_Y|1<<VDB_X, |
48 |
– |
0, /* forbidden diagonal */ |
49 |
– |
}; |
50 |
– |
|
28 |
|
typedef struct { |
29 |
|
COLOR v; /* hemisphere sample value */ |
30 |
< |
float d; /* reciprocal distance (1/rt) */ |
30 |
> |
float d; /* reciprocal distance */ |
31 |
|
FVECT p; /* intersection point */ |
32 |
|
} AMBSAMP; /* sample value */ |
33 |
|
|
34 |
|
typedef struct { |
35 |
|
RAY *rp; /* originating ray sample */ |
59 |
– |
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 |
+ |
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 ambndx(h,i,j) ((i)*(h)->ns + (j)) |
45 |
< |
#define ambsam(h,i,j) (h)->sa[ambndx(h,i,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; |
49 |
|
double I1, I2; |
71 |
– |
int valid; |
50 |
|
} FFTRI; /* vectors and coefficients for Hessian calculation */ |
51 |
|
|
52 |
|
|
75 |
– |
/* Get index for adjacent vertex */ |
53 |
|
static int |
54 |
< |
adjacent_verti(AMBHEMI *hp, int i, int j, int dbit) |
54 |
> |
ambcollision( /* proposed direciton collides? */ |
55 |
> |
AMBHEMI *hp, |
56 |
> |
int i, |
57 |
> |
int j, |
58 |
> |
FVECT dv |
59 |
> |
) |
60 |
|
{ |
61 |
< |
int i0 = i*hp->ns + j; |
62 |
< |
|
63 |
< |
switch (dbit) { |
64 |
< |
case VDB_y: return(i0 - hp->ns); |
65 |
< |
case VDB_x: return(i0 - 1); |
66 |
< |
case VDB_Xy: return(i0 - hp->ns + 1); |
67 |
< |
case VDB_xY: return(i0 + hp->ns - 1); |
68 |
< |
case VDB_X: return(i0 + 1); |
69 |
< |
case VDB_Y: return(i0 + hp->ns); |
70 |
< |
/* the following should never occur */ |
71 |
< |
case VDB_xy: return(i0 - hp->ns - 1); |
72 |
< |
case VDB_XY: return(i0 + hp->ns + 1); |
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(-1); |
86 |
> |
return(0); /* nothing to worry about */ |
87 |
|
} |
88 |
|
|
89 |
|
|
96 |
– |
/* Get vertex direction bit for the opposite edge to complete triangle */ |
90 |
|
static int |
91 |
< |
vdb_edge(int db1, int db2) |
92 |
< |
{ |
93 |
< |
switch (db1) { |
94 |
< |
case VDB_x: return(db2==VDB_y ? VDB_Xy : VDB_Y); |
95 |
< |
case VDB_y: return(db2==VDB_x ? VDB_xY : VDB_X); |
103 |
< |
case VDB_X: return(db2==VDB_Xy ? VDB_y : VDB_xY); |
104 |
< |
case VDB_Y: return(db2==VDB_xY ? VDB_x : VDB_Xy); |
105 |
< |
case VDB_xY: return(db2==VDB_x ? VDB_y : VDB_X); |
106 |
< |
case VDB_Xy: return(db2==VDB_y ? VDB_x : VDB_Y); |
107 |
< |
} |
108 |
< |
error(INTERNAL, "forbidden diagonal in vdb_edge()"); |
109 |
< |
return(-1); |
110 |
< |
} |
111 |
< |
|
112 |
< |
|
113 |
< |
static AMBHEMI * |
114 |
< |
inithemi( /* initialize sampling hemisphere */ |
115 |
< |
COLOR ac, |
116 |
< |
RAY *r, |
117 |
< |
double wt |
91 |
> |
ambsample( /* initial ambient division sample */ |
92 |
> |
AMBHEMI *hp, |
93 |
> |
int i, |
94 |
> |
int j, |
95 |
> |
int n |
96 |
|
) |
97 |
|
{ |
98 |
< |
AMBHEMI *hp; |
99 |
< |
double d; |
122 |
< |
int n, i; |
123 |
< |
/* set number of divisions */ |
124 |
< |
if (ambacc <= FTINY && |
125 |
< |
wt > (d = 0.8*intens(ac)*r->rweight/(ambdiv*minweight))) |
126 |
< |
wt = d; /* avoid ray termination */ |
127 |
< |
n = sqrt(ambdiv * wt) + 0.5; |
128 |
< |
i = 1 + 5*(ambacc > FTINY); /* minimum number of samples */ |
129 |
< |
if (n < i) |
130 |
< |
n = i; |
131 |
< |
/* allocate sampling array */ |
132 |
< |
hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + sizeof(AMBSAMP)*(n*n - 1)); |
133 |
< |
if (hp == NULL) |
134 |
< |
return(NULL); |
135 |
< |
hp->rp = r; |
136 |
< |
hp->ns = n; |
137 |
< |
/* assign coefficient */ |
138 |
< |
copycolor(hp->acoef, ac); |
139 |
< |
d = 1.0/(n*n); |
140 |
< |
scalecolor(hp->acoef, d); |
141 |
< |
/* make tangent plane axes */ |
142 |
< |
hp->uy[0] = 0.5 - frandom(); |
143 |
< |
hp->uy[1] = 0.5 - frandom(); |
144 |
< |
hp->uy[2] = 0.5 - frandom(); |
145 |
< |
for (i = 3; i--; ) |
146 |
< |
if ((-0.6 < r->ron[i]) & (r->ron[i] < 0.6)) |
147 |
< |
break; |
148 |
< |
if (i < 0) |
149 |
< |
error(CONSISTENCY, "bad ray direction in inithemi"); |
150 |
< |
hp->uy[i] = 1.0; |
151 |
< |
VCROSS(hp->ux, hp->uy, r->ron); |
152 |
< |
normalize(hp->ux); |
153 |
< |
VCROSS(hp->uy, r->ron, hp->ux); |
154 |
< |
/* we're ready to sample */ |
155 |
< |
return(hp); |
156 |
< |
} |
157 |
< |
|
158 |
< |
|
159 |
< |
/* Sample ambient division and apply weighting coefficient */ |
160 |
< |
static int |
161 |
< |
getambsamp(RAY *arp, AMBHEMI *hp, int i, int j, int n) |
162 |
< |
{ |
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(arp->rcoef, AVGREFL, AVGREFL, AVGREFL); |
105 |
> |
setcolor(ar.rcoef, AVGREFL, AVGREFL, AVGREFL); |
106 |
|
else |
107 |
< |
copycolor(arp->rcoef, hp->acoef); |
108 |
< |
if (rayorigin(arp, AMBIENT, hp->rp, arp->rcoef) < 0) |
107 |
> |
copycolor(ar.rcoef, hp->acoef); |
108 |
> |
if (rayorigin(&ar, AMBIENT, hp->rp, ar.rcoef) < 0) |
109 |
|
return(0); |
110 |
|
if (ambacc > FTINY) { |
111 |
< |
multcolor(arp->rcoef, hp->acoef); |
112 |
< |
scalecolor(arp->rcoef, 1./AVGREFL); |
111 |
> |
multcolor(ar.rcoef, hp->acoef); |
112 |
> |
scalecolor(ar.rcoef, 1./AVGREFL); |
113 |
|
} |
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 */ |
181 |
< |
if ((spt[0] < 0.1) | (spt[0] >= 0.9)) |
182 |
< |
spt[0] = 0.1 + 0.8*frandom(); |
183 |
< |
if ((spt[1] < 0.1) | (spt[1] >= 0.9)) |
184 |
< |
spt[1] = 0.1 + 0.8*frandom(); |
185 |
< |
} |
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 |
< |
arp->rdir[ii] = spt[0]*hp->ux[ii] + |
122 |
> |
ar.rdir[ii] = spt[0]*hp->ux[ii] + |
123 |
|
spt[1]*hp->uy[ii] + |
124 |
|
zd*hp->rp->ron[ii]; |
125 |
< |
checknorm(arp->rdir); |
126 |
< |
dimlist[ndims++] = ambndx(hp,i,j) + 90171; |
127 |
< |
rayvalue(arp); /* evaluate ray */ |
128 |
< |
ndims--; /* apply coefficient */ |
129 |
< |
multcolor(arp->rcol, arp->rcoef); |
125 |
> |
checknorm(ar.rdir); |
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 |
> |
zd = raydistance(&ar); |
135 |
> |
if (zd <= FTINY) |
136 |
> |
return(0); /* should never happen */ |
137 |
> |
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
138 |
> |
if (zd*ap->d < 1.0) /* new/closer distance? */ |
139 |
> |
ap->d = 1.0/zd; |
140 |
> |
if (!n) { /* record first vertex & value */ |
141 |
> |
if (zd > 10.0*thescene.cusize + 1000.) |
142 |
> |
zd = 10.0*thescene.cusize + 1000.; |
143 |
> |
VSUM(ap->p, ar.rorg, ar.rdir, zd); |
144 |
> |
copycolor(ap->v, ar.rcol); |
145 |
> |
} else { /* else update recorded value */ |
146 |
> |
hp->acol[RED] -= colval(ap->v,RED); |
147 |
> |
hp->acol[GRN] -= colval(ap->v,GRN); |
148 |
> |
hp->acol[BLU] -= colval(ap->v,BLU); |
149 |
> |
zd = 1.0/(double)(n+1); |
150 |
> |
scalecolor(ar.rcol, zd); |
151 |
> |
zd *= (double)n; |
152 |
> |
scalecolor(ap->v, zd); |
153 |
> |
addcolor(ap->v, ar.rcol); |
154 |
> |
} |
155 |
> |
addcolor(hp->acol, ap->v); /* add to our sum */ |
156 |
|
return(1); |
157 |
|
} |
158 |
|
|
159 |
|
|
160 |
< |
static AMBSAMP * |
202 |
< |
ambsample( /* initial ambient division sample */ |
203 |
< |
AMBHEMI *hp, |
204 |
< |
int i, |
205 |
< |
int j |
206 |
< |
) |
207 |
< |
{ |
208 |
< |
AMBSAMP *ap = &ambsam(hp,i,j); |
209 |
< |
RAY ar; |
210 |
< |
/* generate hemispherical sample */ |
211 |
< |
if (!getambsamp(&ar, hp, i, j, 0) || ar.rt <= FTINY) { |
212 |
< |
memset(ap, 0, sizeof(AMBSAMP)); |
213 |
< |
return(NULL); |
214 |
< |
} |
215 |
< |
ap->d = 1.0/ar.rt; /* limit vertex distance */ |
216 |
< |
if (ar.rt > 10.0*thescene.cusize) |
217 |
< |
ar.rt = 10.0*thescene.cusize; |
218 |
< |
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
219 |
< |
copycolor(ap->v, ar.rcol); |
220 |
< |
return(ap); |
221 |
< |
} |
222 |
< |
|
223 |
< |
|
224 |
< |
/* Estimate errors based on ambient division differences */ |
160 |
> |
/* Estimate variance based on ambient division differences */ |
161 |
|
static float * |
162 |
|
getambdiffs(AMBHEMI *hp) |
163 |
|
{ |
164 |
+ |
const double normf = 1./bright(hp->acoef); |
165 |
|
float *earr = (float *)calloc(hp->ns*hp->ns, sizeof(float)); |
166 |
|
float *ep; |
167 |
|
AMBSAMP *ap; |
168 |
< |
double b, d2; |
168 |
> |
double b, b1, d2; |
169 |
|
int i, j; |
170 |
|
|
171 |
|
if (earr == NULL) /* out of memory? */ |
172 |
|
return(NULL); |
173 |
< |
/* compute squared neighbor diffs */ |
173 |
> |
/* sum squared neighbor diffs */ |
174 |
|
for (ap = hp->sa, ep = earr, i = 0; i < hp->ns; i++) |
175 |
|
for (j = 0; j < hp->ns; j++, ap++, ep++) { |
176 |
|
b = bright(ap[0].v); |
177 |
|
if (i) { /* from above */ |
178 |
< |
d2 = b - bright(ap[-hp->ns].v); |
179 |
< |
d2 *= d2; |
178 |
> |
b1 = bright(ap[-hp->ns].v); |
179 |
> |
d2 = b - b1; |
180 |
> |
d2 *= d2*normf/(b + b1); |
181 |
|
ep[0] += d2; |
182 |
|
ep[-hp->ns] += d2; |
183 |
|
} |
184 |
< |
if (j) { /* from behind */ |
185 |
< |
d2 = b - bright(ap[-1].v); |
186 |
< |
d2 *= d2; |
187 |
< |
ep[0] += d2; |
188 |
< |
ep[-1] += d2; |
189 |
< |
} |
184 |
> |
if (!j) continue; |
185 |
> |
/* from behind */ |
186 |
> |
b1 = bright(ap[-1].v); |
187 |
> |
d2 = b - b1; |
188 |
> |
d2 *= d2*normf/(b + b1); |
189 |
> |
ep[0] += d2; |
190 |
> |
ep[-1] += d2; |
191 |
> |
if (!i) continue; |
192 |
> |
/* diagonal */ |
193 |
> |
b1 = bright(ap[-hp->ns-1].v); |
194 |
> |
d2 = b - b1; |
195 |
> |
d2 *= d2*normf/(b + b1); |
196 |
> |
ep[0] += d2; |
197 |
> |
ep[-hp->ns-1] += d2; |
198 |
|
} |
199 |
|
/* correct for number of neighbors */ |
200 |
< |
earr[0] *= 2.f; |
201 |
< |
earr[hp->ns-1] *= 2.f; |
202 |
< |
earr[(hp->ns-1)*hp->ns] *= 2.f; |
203 |
< |
earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 2.f; |
200 |
> |
earr[0] *= 8./3.; |
201 |
> |
earr[hp->ns-1] *= 8./3.; |
202 |
> |
earr[(hp->ns-1)*hp->ns] *= 8./3.; |
203 |
> |
earr[(hp->ns-1)*hp->ns + hp->ns-1] *= 8./3.; |
204 |
|
for (i = 1; i < hp->ns-1; i++) { |
205 |
< |
earr[i*hp->ns] *= 4./3.; |
206 |
< |
earr[i*hp->ns + hp->ns-1] *= 4./3.; |
205 |
> |
earr[i*hp->ns] *= 8./5.; |
206 |
> |
earr[i*hp->ns + hp->ns-1] *= 8./5.; |
207 |
|
} |
208 |
|
for (j = 1; j < hp->ns-1; j++) { |
209 |
< |
earr[j] *= 4./3.; |
210 |
< |
earr[(hp->ns-1)*hp->ns + j] *= 4./3.; |
209 |
> |
earr[j] *= 8./5.; |
210 |
> |
earr[(hp->ns-1)*hp->ns + j] *= 8./5.; |
211 |
|
} |
212 |
|
return(earr); |
213 |
|
} |
215 |
|
|
216 |
|
/* Perform super-sampling on hemisphere (introduces bias) */ |
217 |
|
static void |
218 |
< |
ambsupersamp(double acol[3], AMBHEMI *hp, int cnt) |
218 |
> |
ambsupersamp(AMBHEMI *hp, int cnt) |
219 |
|
{ |
220 |
|
float *earr = getambdiffs(hp); |
221 |
< |
double e2sum = 0; |
276 |
< |
AMBSAMP *ap; |
277 |
< |
RAY ar; |
278 |
< |
COLOR asum; |
221 |
> |
double e2rem = 0; |
222 |
|
float *ep; |
223 |
< |
int i, j, n; |
223 |
> |
int i, j, n, nss; |
224 |
|
|
225 |
|
if (earr == NULL) /* just skip calc. if no memory */ |
226 |
|
return; |
227 |
< |
/* add up estimated variances */ |
228 |
< |
for (ep = earr + hp->ns*hp->ns; ep-- > earr; ) |
229 |
< |
e2sum += *ep; |
227 |
> |
/* accumulate estimated variances */ |
228 |
> |
for (ep = earr + hp->ns*hp->ns; ep > earr; ) |
229 |
> |
e2rem += *--ep; |
230 |
|
ep = earr; /* perform super-sampling */ |
231 |
< |
for (ap = hp->sa, i = 0; i < hp->ns; i++) |
232 |
< |
for (j = 0; j < hp->ns; j++, ap++) { |
233 |
< |
int nss = *ep/e2sum*cnt + frandom(); |
234 |
< |
setcolor(asum, 0., 0., 0.); |
235 |
< |
for (n = 1; n <= nss; n++) { |
236 |
< |
if (!getambsamp(&ar, hp, i, j, n)) { |
237 |
< |
nss = n-1; |
238 |
< |
break; |
296 |
< |
} |
297 |
< |
addcolor(asum, ar.rcol); |
298 |
< |
} |
299 |
< |
if (nss) { /* update returned ambient value */ |
300 |
< |
const double ssf = 1./(nss + 1); |
301 |
< |
for (n = 3; n--; ) |
302 |
< |
acol[n] += ssf*colval(asum,n) + |
303 |
< |
(ssf - 1.)*colval(ap->v,n); |
304 |
< |
} |
305 |
< |
e2sum -= *ep++; /* update remainders */ |
306 |
< |
cnt -= nss; |
231 |
> |
for (i = 0; i < hp->ns; i++) |
232 |
> |
for (j = 0; j < hp->ns; j++) { |
233 |
> |
if (e2rem <= FTINY) |
234 |
> |
goto done; /* nothing left to do */ |
235 |
> |
nss = *ep/e2rem*cnt + frandom(); |
236 |
> |
for (n = 1; n <= nss && ambsample(hp,i,j,n); n++) |
237 |
> |
if (!--cnt) goto done; |
238 |
> |
e2rem -= *ep++; /* update remainder */ |
239 |
|
} |
240 |
+ |
done: |
241 |
|
free(earr); |
242 |
|
} |
243 |
|
|
244 |
|
|
245 |
< |
/* Compute vertex flags, indicating farthest in each direction */ |
246 |
< |
static uby8 * |
247 |
< |
vertex_flags(AMBHEMI *hp) |
245 |
> |
static AMBHEMI * |
246 |
> |
samp_hemi( /* sample indirect hemisphere */ |
247 |
> |
COLOR rcol, |
248 |
> |
RAY *r, |
249 |
> |
double wt |
250 |
> |
) |
251 |
|
{ |
252 |
< |
uby8 *vflags = (uby8 *)calloc(hp->ns*hp->ns, sizeof(uby8)); |
253 |
< |
uby8 *vf; |
254 |
< |
AMBSAMP *ap; |
255 |
< |
int i, j; |
256 |
< |
|
257 |
< |
if (vflags == NULL) |
258 |
< |
error(SYSTEM, "out of memory in vertex_flags()"); |
259 |
< |
vf = vflags; |
260 |
< |
ap = hp->sa; /* compute farthest along first row */ |
261 |
< |
for (j = 0; j < hp->ns-1; j++, vf++, ap++) |
262 |
< |
if (ap[0].d <= ap[1].d) |
263 |
< |
vf[0] |= 1<<VDB_X; |
264 |
< |
else |
265 |
< |
vf[1] |= 1<<VDB_x; |
266 |
< |
++vf; ++ap; |
267 |
< |
/* flag subsequent rows */ |
268 |
< |
for (i = 1; i < hp->ns; i++) { |
269 |
< |
for (j = 0; j < hp->ns-1; j++, vf++, ap++) { |
270 |
< |
if (ap[0].d <= ap[-hp->ns].d) /* row before */ |
271 |
< |
vf[0] |= 1<<VDB_y; |
272 |
< |
else |
273 |
< |
vf[-hp->ns] |= 1<<VDB_Y; |
274 |
< |
if (ap[0].d <= ap[1-hp->ns].d) /* diagonal we care about */ |
275 |
< |
vf[0] |= 1<<VDB_Xy; |
276 |
< |
else |
277 |
< |
vf[1-hp->ns] |= 1<<VDB_xY; |
278 |
< |
if (ap[0].d <= ap[1].d) /* column after */ |
279 |
< |
vf[0] |= 1<<VDB_X; |
280 |
< |
else |
281 |
< |
vf[1] |= 1<<VDB_x; |
282 |
< |
} |
283 |
< |
if (ap[0].d <= ap[-hp->ns].d) /* final column edge */ |
284 |
< |
vf[0] |= 1<<VDB_y; |
285 |
< |
else |
286 |
< |
vf[-hp->ns] |= 1<<VDB_Y; |
287 |
< |
++vf; ++ap; |
252 |
> |
AMBHEMI *hp; |
253 |
> |
double d; |
254 |
> |
int n, i, j; |
255 |
> |
/* insignificance check */ |
256 |
> |
if (bright(rcol) <= FTINY) |
257 |
> |
return(NULL); |
258 |
> |
/* set number of divisions */ |
259 |
> |
if (ambacc <= FTINY && |
260 |
> |
wt > (d = 0.8*intens(rcol)*r->rweight/(ambdiv*minweight))) |
261 |
> |
wt = d; /* avoid ray termination */ |
262 |
> |
n = sqrt(ambdiv * wt) + 0.5; |
263 |
> |
i = 1 + 5*(ambacc > FTINY); /* minimum number of samples */ |
264 |
> |
if (n < i) |
265 |
> |
n = i; |
266 |
> |
/* allocate sampling array */ |
267 |
> |
hp = (AMBHEMI *)malloc(sizeof(AMBHEMI) + sizeof(AMBSAMP)*(n*n - 1)); |
268 |
> |
if (hp == NULL) |
269 |
> |
error(SYSTEM, "out of memory in samp_hemi"); |
270 |
> |
hp->rp = r; |
271 |
> |
hp->ns = n; |
272 |
> |
hp->acol[RED] = hp->acol[GRN] = hp->acol[BLU] = 0.0; |
273 |
> |
memset(hp->sa, 0, sizeof(AMBSAMP)*n*n); |
274 |
> |
hp->sampOK = 0; |
275 |
> |
/* assign coefficient */ |
276 |
> |
copycolor(hp->acoef, rcol); |
277 |
> |
d = 1.0/(n*n); |
278 |
> |
scalecolor(hp->acoef, d); |
279 |
> |
/* make tangent plane axes */ |
280 |
> |
if (!getperpendicular(hp->ux, r->ron, 1)) |
281 |
> |
error(CONSISTENCY, "bad ray direction in samp_hemi"); |
282 |
> |
VCROSS(hp->uy, r->ron, hp->ux); |
283 |
> |
/* sample divisions */ |
284 |
> |
for (i = hp->ns; i--; ) |
285 |
> |
for (j = hp->ns; j--; ) |
286 |
> |
hp->sampOK += ambsample(hp, i, j, 0); |
287 |
> |
copycolor(rcol, hp->acol); |
288 |
> |
if (!hp->sampOK) { /* utter failure? */ |
289 |
> |
free(hp); |
290 |
> |
return(NULL); |
291 |
|
} |
292 |
< |
return(vflags); |
292 |
> |
if (hp->sampOK < hp->ns*hp->ns) { |
293 |
> |
hp->sampOK *= -1; /* soft failure */ |
294 |
> |
return(hp); |
295 |
> |
} |
296 |
> |
if (hp->sampOK < 64) |
297 |
> |
return(hp); /* insufficient for super-sampling */ |
298 |
> |
n = ambssamp*wt + 0.5; |
299 |
> |
if (n > 8) { /* perform super-sampling? */ |
300 |
> |
ambsupersamp(hp, n); |
301 |
> |
copycolor(rcol, hp->acol); |
302 |
> |
} |
303 |
> |
return(hp); /* all is well */ |
304 |
|
} |
305 |
|
|
306 |
|
|
307 |
|
/* Return brightness of farthest ambient sample */ |
308 |
|
static double |
309 |
< |
back_ambval(AMBHEMI *hp, int i, int j, int dbit1, int dbit2, const uby8 *vflags) |
309 |
> |
back_ambval(AMBHEMI *hp, const int n1, const int n2, const int n3) |
310 |
|
{ |
311 |
< |
const int v0 = ambndx(hp,i,j); |
312 |
< |
const int tflags = (1<<dbit1 | 1<<dbit2); |
313 |
< |
int v1, v2; |
314 |
< |
|
315 |
< |
if ((vflags[v0] & tflags) == tflags) /* is v0 the farthest? */ |
316 |
< |
return(colval(hp->sa[v0].v,CIEY)); |
317 |
< |
v1 = adjacent_verti(hp, i, j, dbit1); |
318 |
< |
if (vflags[v0] & 1<<dbit2) /* v1 farthest if v0>v2 */ |
369 |
< |
return(colval(hp->sa[v1].v,CIEY)); |
370 |
< |
v2 = adjacent_verti(hp, i, j, dbit2); |
371 |
< |
if (vflags[v0] & 1<<dbit1) /* v2 farthest if v0>v1 */ |
372 |
< |
return(colval(hp->sa[v2].v,CIEY)); |
373 |
< |
/* else check if v1>v2 */ |
374 |
< |
if (vflags[v1] & 1<<vdb_edge(dbit1,dbit2)) |
375 |
< |
return(colval(hp->sa[v1].v,CIEY)); |
376 |
< |
return(colval(hp->sa[v2].v,CIEY)); |
311 |
> |
if (hp->sa[n1].d <= hp->sa[n2].d) { |
312 |
> |
if (hp->sa[n1].d <= hp->sa[n3].d) |
313 |
> |
return(colval(hp->sa[n1].v,CIEY)); |
314 |
> |
return(colval(hp->sa[n3].v,CIEY)); |
315 |
> |
} |
316 |
> |
if (hp->sa[n2].d <= hp->sa[n3].d) |
317 |
> |
return(colval(hp->sa[n2].v,CIEY)); |
318 |
> |
return(colval(hp->sa[n3].v,CIEY)); |
319 |
|
} |
320 |
|
|
321 |
|
|
322 |
|
/* Compute vectors and coefficients for Hessian/gradient calcs */ |
323 |
|
static void |
324 |
< |
comp_fftri(FFTRI *ftp, AMBHEMI *hp, int i, int j, int dbit, const uby8 *vflags) |
324 |
> |
comp_fftri(FFTRI *ftp, AMBHEMI *hp, const int n0, const int n1) |
325 |
|
{ |
326 |
< |
const int i0 = ambndx(hp,i,j); |
327 |
< |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
386 |
< |
int i1, ii; |
326 |
> |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
327 |
> |
int ii; |
328 |
|
|
329 |
< |
ftp->valid = 0; /* check if we can skip this edge */ |
330 |
< |
ii = adjacent_trifl[dbit]; |
331 |
< |
if ((vflags[i0] & ii) == ii) /* cancels if vertex used as value */ |
391 |
< |
return; |
392 |
< |
i1 = adjacent_verti(hp, i, j, dbit); |
393 |
< |
ii = adjacent_trifl[VDB_OPP(dbit)]; |
394 |
< |
if ((vflags[i1] & ii) == ii) /* on either end (for both triangles) */ |
395 |
< |
return; |
396 |
< |
/* else go ahead with calculation */ |
397 |
< |
VSUB(ftp->r_i, hp->sa[i0].p, hp->rp->rop); |
398 |
< |
VSUB(ftp->r_i1, hp->sa[i1].p, hp->rp->rop); |
399 |
< |
VSUB(ftp->e_i, hp->sa[i1].p, hp->sa[i0].p); |
329 |
> |
VSUB(ftp->r_i, hp->sa[n0].p, hp->rp->rop); |
330 |
> |
VSUB(ftp->r_i1, hp->sa[n1].p, hp->rp->rop); |
331 |
> |
VSUB(ftp->e_i, hp->sa[n1].p, hp->sa[n0].p); |
332 |
|
VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1); |
333 |
|
rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp); |
334 |
|
dot_e = DOT(ftp->e_i,ftp->e_i); |
342 |
|
J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e; |
343 |
|
for (ii = 3; ii--; ) |
344 |
|
ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii]; |
413 |
– |
ftp->valid++; |
345 |
|
} |
346 |
|
|
347 |
|
|
367 |
|
double d1, d2, d3, d4; |
368 |
|
double I3, J3, K3; |
369 |
|
int i, j; |
439 |
– |
|
440 |
– |
if (!ftp->valid) { /* preemptive test */ |
441 |
– |
memset(hess, 0, sizeof(FVECT)*3); |
442 |
– |
return; |
443 |
– |
} |
370 |
|
/* compute intermediate coefficients */ |
371 |
|
d1 = 1.0/DOT(ftp->r_i,ftp->r_i); |
372 |
|
d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
430 |
|
double f1; |
431 |
|
int i; |
432 |
|
|
507 |
– |
if (!ftp->valid) { /* preemptive test */ |
508 |
– |
memset(grad, 0, sizeof(FVECT)); |
509 |
– |
return; |
510 |
– |
} |
433 |
|
f1 = 2.0*DOT(nrm, ftp->rcp); |
434 |
|
VCROSS(ncp, nrm, ftp->e_i); |
435 |
|
for (i = 3; i--; ) |
459 |
|
|
460 |
|
|
461 |
|
/* Compute anisotropic radii and eigenvector directions */ |
462 |
< |
static int |
462 |
> |
static void |
463 |
|
eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3]) |
464 |
|
{ |
465 |
|
double hess2[2][2]; |
481 |
|
if (i == 1) /* double-root (circle) */ |
482 |
|
evalue[1] = evalue[0]; |
483 |
|
if (!i || ((evalue[0] = fabs(evalue[0])) <= FTINY*FTINY) | |
484 |
< |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) |
485 |
< |
error(INTERNAL, "bad eigenvalue calculation"); |
486 |
< |
|
484 |
> |
((evalue[1] = fabs(evalue[1])) <= FTINY*FTINY) ) { |
485 |
> |
ra[0] = ra[1] = maxarad; |
486 |
> |
return; |
487 |
> |
} |
488 |
|
if (evalue[0] > evalue[1]) { |
489 |
|
ra[0] = sqrt(sqrt(4.0/evalue[0])); |
490 |
|
ra[1] = sqrt(sqrt(4.0/evalue[1])); |
519 |
|
static char memerrmsg[] = "out of memory in ambHessian()"; |
520 |
|
FVECT (*hessrow)[3] = NULL; |
521 |
|
FVECT *gradrow = NULL; |
599 |
– |
uby8 *vflags; |
522 |
|
FVECT hessian[3]; |
523 |
|
FVECT gradient; |
524 |
|
FFTRI fftr; |
540 |
|
error(SYSTEM, memerrmsg); |
541 |
|
memset(gradient, 0, sizeof(gradient)); |
542 |
|
} |
621 |
– |
/* get vertex position flags */ |
622 |
– |
vflags = vertex_flags(hp); |
543 |
|
/* compute first row of edges */ |
544 |
|
for (j = 0; j < hp->ns-1; j++) { |
545 |
< |
comp_fftri(&fftr, hp, 0, j, VDB_X, vflags); |
545 |
> |
comp_fftri(&fftr, hp, AI(hp,0,j), AI(hp,0,j+1)); |
546 |
|
if (hessrow != NULL) |
547 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
548 |
|
if (gradrow != NULL) |
552 |
|
for (i = 0; i < hp->ns-1; i++) { |
553 |
|
FVECT hesscol[3]; /* compute first vertical edge */ |
554 |
|
FVECT gradcol; |
555 |
< |
comp_fftri(&fftr, hp, i, 0, VDB_Y, vflags); |
555 |
> |
comp_fftri(&fftr, hp, AI(hp,i,0), AI(hp,i+1,0)); |
556 |
|
if (hessrow != NULL) |
557 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
558 |
|
if (gradrow != NULL) |
561 |
|
FVECT hessdia[3]; /* compute triangle contributions */ |
562 |
|
FVECT graddia; |
563 |
|
double backg; |
564 |
< |
backg = back_ambval(hp, i, j, VDB_X, VDB_Y, vflags); |
564 |
> |
backg = back_ambval(hp, AI(hp,i,j), |
565 |
> |
AI(hp,i,j+1), AI(hp,i+1,j)); |
566 |
|
/* diagonal (inner) edge */ |
567 |
< |
comp_fftri(&fftr, hp, i, j+1, VDB_xY, vflags); |
567 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j)); |
568 |
|
if (hessrow != NULL) { |
569 |
|
comp_hessian(hessdia, &fftr, hp->rp->ron); |
570 |
|
rev_hessian(hesscol); |
576 |
|
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
577 |
|
} |
578 |
|
/* initialize edge in next row */ |
579 |
< |
comp_fftri(&fftr, hp, i+1, j+1, VDB_x, vflags); |
579 |
> |
comp_fftri(&fftr, hp, AI(hp,i+1,j+1), AI(hp,i+1,j)); |
580 |
|
if (hessrow != NULL) |
581 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
582 |
|
if (gradrow != NULL) |
583 |
|
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
584 |
|
/* new column edge & paired triangle */ |
585 |
< |
backg = back_ambval(hp, i+1, j+1, VDB_x, VDB_y, vflags); |
586 |
< |
comp_fftri(&fftr, hp, i, j+1, VDB_Y, vflags); |
585 |
> |
backg = back_ambval(hp, AI(hp,i+1,j+1), |
586 |
> |
AI(hp,i+1,j), AI(hp,i,j+1)); |
587 |
> |
comp_fftri(&fftr, hp, AI(hp,i,j+1), AI(hp,i+1,j+1)); |
588 |
|
if (hessrow != NULL) { |
589 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
590 |
|
rev_hessian(hessdia); |
604 |
|
/* release row buffers */ |
605 |
|
if (hessrow != NULL) free(hessrow); |
606 |
|
if (gradrow != NULL) free(gradrow); |
685 |
– |
free(vflags); |
607 |
|
|
608 |
|
if (ra != NULL) /* extract eigenvectors & radii */ |
609 |
|
eigenvectors(uv, ra, hessian); |
643 |
|
static uint32 |
644 |
|
ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1) |
645 |
|
{ |
646 |
< |
uint32 flgs = 0; |
647 |
< |
int i, j; |
648 |
< |
/* circle around perimeter */ |
646 |
> |
const double max_d = 1.0/(minarad*ambacc + 0.001); |
647 |
> |
const double ang_res = 0.5*PI/hp->ns; |
648 |
> |
const double ang_step = ang_res/((int)(16/PI*ang_res) + 1.01); |
649 |
> |
double avg_d = 0; |
650 |
> |
uint32 flgs = 0; |
651 |
> |
FVECT vec; |
652 |
> |
double u, v; |
653 |
> |
double ang, a1; |
654 |
> |
int i, j; |
655 |
> |
/* don't bother for a few samples */ |
656 |
> |
if (hp->ns < 8) |
657 |
> |
return(0); |
658 |
> |
/* check distances overhead */ |
659 |
> |
for (i = hp->ns*3/4; i-- > hp->ns>>2; ) |
660 |
> |
for (j = hp->ns*3/4; j-- > hp->ns>>2; ) |
661 |
> |
avg_d += ambsam(hp,i,j).d; |
662 |
> |
avg_d *= 4.0/(hp->ns*hp->ns); |
663 |
> |
if (avg_d*r0 >= 1.0) /* ceiling too low for corral? */ |
664 |
> |
return(0); |
665 |
> |
if (avg_d >= max_d) /* insurance */ |
666 |
> |
return(0); |
667 |
> |
/* else circle around perimeter */ |
668 |
|
for (i = 0; i < hp->ns; i++) |
669 |
|
for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) { |
670 |
|
AMBSAMP *ap = &ambsam(hp,i,j); |
671 |
< |
FVECT vec; |
672 |
< |
double u, v; |
733 |
< |
double ang; |
734 |
< |
int abp; |
735 |
< |
if (ap->d <= FTINY) |
736 |
< |
continue; |
671 |
> |
if ((ap->d <= FTINY) | (ap->d >= max_d)) |
672 |
> |
continue; /* too far or too near */ |
673 |
|
VSUB(vec, ap->p, hp->rp->rop); |
674 |
< |
u = DOT(vec, uv[0]) * ap->d; |
675 |
< |
v = DOT(vec, uv[1]) * ap->d; |
676 |
< |
if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= 1.0) |
674 |
> |
u = DOT(vec, uv[0]); |
675 |
> |
v = DOT(vec, uv[1]); |
676 |
> |
if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= u*u + v*v) |
677 |
|
continue; /* occluder outside ellipse */ |
678 |
|
ang = atan2a(v, u); /* else set direction flags */ |
679 |
< |
ang += 2.0*PI*(ang < 0); |
680 |
< |
ang *= 16./PI; |
745 |
< |
if ((ang < .5) | (ang >= 31.5)) |
746 |
< |
flgs |= 0x80000001; |
747 |
< |
else |
748 |
< |
flgs |= 3L<<(int)(ang-.5); |
679 |
> |
for (a1 = ang-ang_res; a1 <= ang+ang_res; a1 += ang_step) |
680 |
> |
flgs |= 1L<<(int)(16/PI*(a1 + 2.*PI*(a1 < 0))); |
681 |
|
} |
682 |
|
return(flgs); |
683 |
|
} |
695 |
|
uint32 *crlp /* returned (optional) */ |
696 |
|
) |
697 |
|
{ |
698 |
< |
AMBHEMI *hp = inithemi(rcol, r, wt); |
767 |
< |
int cnt; |
698 |
> |
AMBHEMI *hp = samp_hemi(rcol, r, wt); |
699 |
|
FVECT my_uv[2]; |
700 |
< |
double d, K, acol[3]; |
700 |
> |
double d, K; |
701 |
|
AMBSAMP *ap; |
702 |
< |
int i, j; |
703 |
< |
/* check/initialize */ |
773 |
< |
if (hp == NULL) |
774 |
< |
return(0); |
702 |
> |
int i; |
703 |
> |
/* clear return values */ |
704 |
|
if (uv != NULL) |
705 |
|
memset(uv, 0, sizeof(FVECT)*2); |
706 |
|
if (ra != NULL) |
711 |
|
dg[0] = dg[1] = 0.0; |
712 |
|
if (crlp != NULL) |
713 |
|
*crlp = 0; |
714 |
< |
/* sample the hemisphere */ |
715 |
< |
acol[0] = acol[1] = acol[2] = 0.0; |
716 |
< |
cnt = 0; |
717 |
< |
for (i = hp->ns; i--; ) |
718 |
< |
for (j = hp->ns; j--; ) |
719 |
< |
if ((ap = ambsample(hp, i, j)) != NULL) { |
720 |
< |
addcolor(acol, ap->v); |
792 |
< |
++cnt; |
793 |
< |
} |
794 |
< |
if (!cnt) { |
795 |
< |
setcolor(rcol, 0.0, 0.0, 0.0); |
796 |
< |
free(hp); |
797 |
< |
return(0); /* no valid samples */ |
714 |
> |
if (hp == NULL) /* sampling falure? */ |
715 |
> |
return(0); |
716 |
> |
|
717 |
> |
if ((ra == NULL) & (pg == NULL) & (dg == NULL) || |
718 |
> |
(hp->sampOK < 0) | (hp->ns < 6)) { |
719 |
> |
free(hp); /* Hessian not requested/possible */ |
720 |
> |
return(-1); /* value-only return value */ |
721 |
|
} |
722 |
< |
if (cnt < hp->ns*hp->ns) { /* incomplete sampling? */ |
800 |
< |
copycolor(rcol, acol); |
801 |
< |
free(hp); |
802 |
< |
return(-1); /* return value w/o Hessian */ |
803 |
< |
} |
804 |
< |
cnt = ambssamp*wt + 0.5; /* perform super-sampling? */ |
805 |
< |
if (cnt > 0) |
806 |
< |
ambsupersamp(acol, hp, cnt); |
807 |
< |
copycolor(rcol, acol); /* final indirect irradiance/PI */ |
808 |
< |
if ((ra == NULL) & (pg == NULL) & (dg == NULL)) { |
809 |
< |
free(hp); |
810 |
< |
return(-1); /* no radius or gradient calc. */ |
811 |
< |
} |
812 |
< |
if ((d = bright(acol)) > FTINY) { /* normalize Y values */ |
722 |
> |
if ((d = bright(rcol)) > FTINY) { /* normalize Y values */ |
723 |
|
d = 0.99*(hp->ns*hp->ns)/d; |
724 |
|
K = 0.01; |
725 |
|
} else { /* or fall back on geometric Hessian */ |
726 |
|
K = 1.0; |
727 |
|
pg = NULL; |
728 |
|
dg = NULL; |
729 |
+ |
crlp = NULL; |
730 |
|
} |
731 |
|
ap = hp->sa; /* relative Y channel from here on... */ |
732 |
|
for (i = hp->ns*hp->ns; i--; ap++) |
754 |
|
if (ra[1] < minarad) |
755 |
|
ra[1] = minarad; |
756 |
|
} |
757 |
< |
ra[0] *= d = 1.0/sqrt(sqrt(wt)); |
757 |
> |
ra[0] *= d = 1.0/sqrt(wt); |
758 |
|
if ((ra[1] *= d) > 2.0*ra[0]) |
759 |
|
ra[1] = 2.0*ra[0]; |
760 |
|
if (ra[1] > maxarad) { |
762 |
|
if (ra[0] > maxarad) |
763 |
|
ra[0] = maxarad; |
764 |
|
} |
765 |
< |
if (crlp != NULL) /* flag encroached directions */ |
765 |
> |
/* flag encroached directions */ |
766 |
> |
if (crlp != NULL) |
767 |
|
*crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc); |
768 |
|
if (pg != NULL) { /* cap gradient if necessary */ |
769 |
|
d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1]; |
866 |
|
ndims--; |
867 |
|
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
868 |
|
addcolor(dp->v, ar.rcol); |
869 |
< |
/* use rt to improve gradient calc */ |
870 |
< |
if (ar.rt > FTINY && ar.rt < FHUGE) |
871 |
< |
dp->r += 1.0/ar.rt; |
869 |
> |
/* use rxt to improve gradient calc */ |
870 |
> |
if (ar.rxt > FTINY && ar.rxt < FHUGE) |
871 |
> |
dp->r += 1.0/ar.rxt; |
872 |
|
/* (re)initialize error */ |
873 |
|
if (dp->n++) { |
874 |
|
b2 = bright(dp->v)/dp->n - bright(ar.rcol); |