| 8 |
|
* for Irradiance Caching" by Schwarzhaupt, Wann Jensen, & Jarosz |
| 9 |
|
* from ACM SIGGRAPH Asia 2012 conference proceedings. |
| 10 |
|
* |
| 11 |
+ |
* Added book-keeping optimization to avoid calculations that would |
| 12 |
+ |
* cancel due to traversal both directions on edges that are adjacent |
| 13 |
+ |
* to same-valued triangles. This cuts about half of Hessian math. |
| 14 |
+ |
* |
| 15 |
|
* Declarations of external symbols in ambient.h |
| 16 |
|
*/ |
| 17 |
|
|
| 25 |
|
|
| 26 |
|
extern void SDsquare2disk(double ds[2], double seedx, double seedy); |
| 27 |
|
|
| 28 |
+ |
/* vertex direction bit positions */ |
| 29 |
+ |
#define VDB_xy 0 |
| 30 |
+ |
#define VDB_y 01 |
| 31 |
+ |
#define VDB_x 02 |
| 32 |
+ |
#define VDB_Xy 03 |
| 33 |
+ |
#define VDB_xY 04 |
| 34 |
+ |
#define VDB_X 05 |
| 35 |
+ |
#define VDB_Y 06 |
| 36 |
+ |
#define VDB_XY 07 |
| 37 |
+ |
/* get opposite vertex direction bit */ |
| 38 |
+ |
#define VDB_OPP(f) (~(f) & 07) |
| 39 |
+ |
/* adjacent triangle vertex flags */ |
| 40 |
+ |
static const int adjacent_trifl[8] = { |
| 41 |
+ |
0, /* forbidden diagonal */ |
| 42 |
+ |
1<<VDB_x|1<<VDB_y|1<<VDB_Xy, |
| 43 |
+ |
1<<VDB_y|1<<VDB_x|1<<VDB_xY, |
| 44 |
+ |
1<<VDB_y|1<<VDB_Xy|1<<VDB_X, |
| 45 |
+ |
1<<VDB_x|1<<VDB_xY|1<<VDB_Y, |
| 46 |
+ |
1<<VDB_Xy|1<<VDB_X|1<<VDB_Y, |
| 47 |
+ |
1<<VDB_xY|1<<VDB_Y|1<<VDB_X, |
| 48 |
+ |
0, /* forbidden diagonal */ |
| 49 |
+ |
}; |
| 50 |
+ |
|
| 51 |
|
typedef struct { |
| 52 |
+ |
COLOR v; /* hemisphere sample value */ |
| 53 |
+ |
float d; /* reciprocal distance (1/rt) */ |
| 54 |
+ |
FVECT p; /* intersection point */ |
| 55 |
+ |
} AMBSAMP; /* sample value */ |
| 56 |
+ |
|
| 57 |
+ |
typedef struct { |
| 58 |
|
RAY *rp; /* originating ray sample */ |
| 59 |
|
FVECT ux, uy; /* tangent axis unit vectors */ |
| 60 |
|
int ns; /* number of samples per axis */ |
| 61 |
|
COLOR acoef; /* division contribution coefficient */ |
| 62 |
< |
struct s_ambsamp { |
| 30 |
< |
COLOR v; /* hemisphere sample value */ |
| 31 |
< |
FVECT p; /* intersection point */ |
| 32 |
< |
} sa[1]; /* sample array (extends struct) */ |
| 62 |
> |
AMBSAMP sa[1]; /* sample array (extends struct) */ |
| 63 |
|
} AMBHEMI; /* ambient sample hemisphere */ |
| 64 |
|
|
| 65 |
< |
typedef struct s_ambsamp AMBSAMP; |
| 65 |
> |
#define ambndx(h,i,j) ((i)*(h)->ns + (j)) |
| 66 |
> |
#define ambsam(h,i,j) (h)->sa[ambndx(h,i,j)] |
| 67 |
|
|
| 37 |
– |
#define ambsamp(h,i,j) (h)->sa[(i)*(h)->ns + (j)] |
| 38 |
– |
|
| 68 |
|
typedef struct { |
| 69 |
|
FVECT r_i, r_i1, e_i, rcp, rI2_eJ2; |
| 70 |
|
double I1, I2; |
| 71 |
+ |
int valid; |
| 72 |
|
} FFTRI; /* vectors and coefficients for Hessian calculation */ |
| 73 |
|
|
| 74 |
|
|
| 75 |
+ |
/* Get index for adjacent vertex */ |
| 76 |
+ |
static int |
| 77 |
+ |
adjacent_verti(AMBHEMI *hp, int i, int j, int dbit) |
| 78 |
+ |
{ |
| 79 |
+ |
int i0 = i*hp->ns + j; |
| 80 |
+ |
|
| 81 |
+ |
switch (dbit) { |
| 82 |
+ |
case VDB_y: return(i0 - hp->ns); |
| 83 |
+ |
case VDB_x: return(i0 - 1); |
| 84 |
+ |
case VDB_Xy: return(i0 - hp->ns + 1); |
| 85 |
+ |
case VDB_xY: return(i0 + hp->ns - 1); |
| 86 |
+ |
case VDB_X: return(i0 + 1); |
| 87 |
+ |
case VDB_Y: return(i0 + hp->ns); |
| 88 |
+ |
/* the following should never occur */ |
| 89 |
+ |
case VDB_xy: return(i0 - hp->ns - 1); |
| 90 |
+ |
case VDB_XY: return(i0 + hp->ns + 1); |
| 91 |
+ |
} |
| 92 |
+ |
return(-1); |
| 93 |
+ |
} |
| 94 |
+ |
|
| 95 |
+ |
|
| 96 |
+ |
/* Get vertex direction bit for the opposite edge to complete triangle */ |
| 97 |
+ |
static int |
| 98 |
+ |
vdb_edge(int db1, int db2) |
| 99 |
+ |
{ |
| 100 |
+ |
switch (db1) { |
| 101 |
+ |
case VDB_x: return(db2==VDB_y ? VDB_Xy : VDB_Y); |
| 102 |
+ |
case VDB_y: return(db2==VDB_x ? VDB_xY : VDB_X); |
| 103 |
+ |
case VDB_X: return(db2==VDB_Xy ? VDB_y : VDB_xY); |
| 104 |
+ |
case VDB_Y: return(db2==VDB_xY ? VDB_x : VDB_Xy); |
| 105 |
+ |
case VDB_xY: return(db2==VDB_x ? VDB_y : VDB_X); |
| 106 |
+ |
case VDB_Xy: return(db2==VDB_y ? VDB_x : VDB_Y); |
| 107 |
+ |
} |
| 108 |
+ |
error(INTERNAL, "forbidden diagonal in vdb_edge()"); |
| 109 |
+ |
return(-1); |
| 110 |
+ |
} |
| 111 |
+ |
|
| 112 |
+ |
|
| 113 |
|
static AMBHEMI * |
| 114 |
|
inithemi( /* initialize sampling hemisphere */ |
| 115 |
|
COLOR ac, |
| 156 |
|
} |
| 157 |
|
|
| 158 |
|
|
| 159 |
< |
/* Prepare ambient division sample */ |
| 159 |
> |
/* Sample ambient division and apply weighting coefficient */ |
| 160 |
|
static int |
| 161 |
< |
prepambsamp(RAY *arp, AMBHEMI *hp, int i, int j, int n) |
| 161 |
> |
getambsamp(RAY *arp, AMBHEMI *hp, int i, int j, int n) |
| 162 |
|
{ |
| 163 |
|
int hlist[3], ii; |
| 164 |
|
double spt[2], zd; |
| 174 |
|
scalecolor(arp->rcoef, 1./AVGREFL); |
| 175 |
|
} |
| 176 |
|
hlist[0] = hp->rp->rno; |
| 177 |
< |
hlist[1] = i; |
| 178 |
< |
hlist[2] = j; |
| 177 |
> |
hlist[1] = j; |
| 178 |
> |
hlist[2] = i; |
| 179 |
|
multisamp(spt, 2, urand(ilhash(hlist,3)+n)); |
| 180 |
|
if (!n) { /* avoid border samples for n==0 */ |
| 181 |
< |
if ((spt[0] < 0.1) | (spt[0] > 0.9)) |
| 181 |
> |
if ((spt[0] < 0.1) | (spt[0] >= 0.9)) |
| 182 |
|
spt[0] = 0.1 + 0.8*frandom(); |
| 183 |
< |
if ((spt[1] < 0.1) | (spt[1] > 0.9)) |
| 183 |
> |
if ((spt[1] < 0.1) | (spt[1] >= 0.9)) |
| 184 |
|
spt[1] = 0.1 + 0.8*frandom(); |
| 185 |
|
} |
| 186 |
< |
SDsquare2disk(spt, (i+spt[0])/hp->ns, (j+spt[1])/hp->ns); |
| 186 |
> |
SDsquare2disk(spt, (j+spt[1])/hp->ns, (i+spt[0])/hp->ns); |
| 187 |
|
zd = sqrt(1. - spt[0]*spt[0] - spt[1]*spt[1]); |
| 188 |
|
for (ii = 3; ii--; ) |
| 189 |
|
arp->rdir[ii] = spt[0]*hp->ux[ii] + |
| 190 |
|
spt[1]*hp->uy[ii] + |
| 191 |
|
zd*hp->rp->ron[ii]; |
| 192 |
|
checknorm(arp->rdir); |
| 193 |
+ |
dimlist[ndims++] = ambndx(hp,i,j) + 90171; |
| 194 |
+ |
rayvalue(arp); /* evaluate ray */ |
| 195 |
+ |
ndims--; /* apply coefficient */ |
| 196 |
+ |
multcolor(arp->rcol, arp->rcoef); |
| 197 |
|
return(1); |
| 198 |
|
} |
| 199 |
|
|
| 200 |
|
|
| 201 |
|
static AMBSAMP * |
| 202 |
< |
ambsample( /* sample an ambient direction */ |
| 202 |
> |
ambsample( /* initial ambient division sample */ |
| 203 |
|
AMBHEMI *hp, |
| 204 |
|
int i, |
| 205 |
|
int j |
| 206 |
|
) |
| 207 |
|
{ |
| 208 |
< |
AMBSAMP *ap = &ambsamp(hp,i,j); |
| 208 |
> |
AMBSAMP *ap = &ambsam(hp,i,j); |
| 209 |
|
RAY ar; |
| 210 |
|
/* generate hemispherical sample */ |
| 211 |
< |
if (!prepambsamp(&ar, hp, i, j, 0)) |
| 212 |
< |
goto badsample; |
| 213 |
< |
dimlist[ndims++] = i*hp->ns + j + 90171; |
| 214 |
< |
rayvalue(&ar); /* evaluate ray */ |
| 215 |
< |
ndims--; |
| 144 |
< |
/* limit vertex distance */ |
| 211 |
> |
if (!getambsamp(&ar, hp, i, j, 0) || ar.rt <= FTINY) { |
| 212 |
> |
memset(ap, 0, sizeof(AMBSAMP)); |
| 213 |
> |
return(NULL); |
| 214 |
> |
} |
| 215 |
> |
ap->d = 1.0/ar.rt; /* limit vertex distance */ |
| 216 |
|
if (ar.rt > 10.0*thescene.cusize) |
| 217 |
|
ar.rt = 10.0*thescene.cusize; |
| 147 |
– |
else if (ar.rt <= FTINY) /* should never happen! */ |
| 148 |
– |
goto badsample; |
| 218 |
|
VSUM(ap->p, ar.rorg, ar.rdir, ar.rt); |
| 150 |
– |
multcolor(ar.rcol, ar.rcoef); /* apply coefficient */ |
| 219 |
|
copycolor(ap->v, ar.rcol); |
| 220 |
|
return(ap); |
| 153 |
– |
badsample: |
| 154 |
– |
setcolor(ap->v, 0., 0., 0.); |
| 155 |
– |
VCOPY(ap->p, hp->rp->rop); |
| 156 |
– |
return(NULL); |
| 221 |
|
} |
| 222 |
|
|
| 223 |
|
|
| 225 |
|
static float * |
| 226 |
|
getambdiffs(AMBHEMI *hp) |
| 227 |
|
{ |
| 228 |
< |
float *earr = calloc(hp->ns*hp->ns, sizeof(float)); |
| 228 |
> |
float *earr = (float *)calloc(hp->ns*hp->ns, sizeof(float)); |
| 229 |
|
float *ep; |
| 230 |
+ |
AMBSAMP *ap; |
| 231 |
|
double b, d2; |
| 232 |
|
int i, j; |
| 233 |
|
|
| 234 |
|
if (earr == NULL) /* out of memory? */ |
| 235 |
|
return(NULL); |
| 236 |
|
/* compute squared neighbor diffs */ |
| 237 |
< |
for (ep = earr, i = 0; i < hp->ns; i++) |
| 238 |
< |
for (j = 0; j < hp->ns; j++, ep++) { |
| 239 |
< |
b = bright(ambsamp(hp,i,j).v); |
| 237 |
> |
for (ap = hp->sa, ep = earr, i = 0; i < hp->ns; i++) |
| 238 |
> |
for (j = 0; j < hp->ns; j++, ap++, ep++) { |
| 239 |
> |
b = bright(ap[0].v); |
| 240 |
|
if (i) { /* from above */ |
| 241 |
< |
d2 = b - bright(ambsamp(hp,i-1,j).v); |
| 241 |
> |
d2 = b - bright(ap[-hp->ns].v); |
| 242 |
|
d2 *= d2; |
| 243 |
|
ep[0] += d2; |
| 244 |
|
ep[-hp->ns] += d2; |
| 245 |
|
} |
| 246 |
|
if (j) { /* from behind */ |
| 247 |
< |
d2 = b - bright(ambsamp(hp,i,j-1).v); |
| 247 |
> |
d2 = b - bright(ap[-1].v); |
| 248 |
|
d2 *= d2; |
| 249 |
|
ep[0] += d2; |
| 250 |
|
ep[-1] += d2; |
| 267 |
|
} |
| 268 |
|
|
| 269 |
|
|
| 270 |
< |
/* Perform super-sampling on hemisphere */ |
| 270 |
> |
/* Perform super-sampling on hemisphere (introduces bias) */ |
| 271 |
|
static void |
| 272 |
|
ambsupersamp(double acol[3], AMBHEMI *hp, int cnt) |
| 273 |
|
{ |
| 290 |
|
int nss = *ep/e2sum*cnt + frandom(); |
| 291 |
|
setcolor(asum, 0., 0., 0.); |
| 292 |
|
for (n = 1; n <= nss; n++) { |
| 293 |
< |
if (!prepambsamp(&ar, hp, i, j, n)) { |
| 293 |
> |
if (!getambsamp(&ar, hp, i, j, n)) { |
| 294 |
|
nss = n-1; |
| 295 |
|
break; |
| 296 |
|
} |
| 232 |
– |
dimlist[ndims++] = i*hp->ns + j + 90171; |
| 233 |
– |
rayvalue(&ar); /* evaluate super-sample */ |
| 234 |
– |
ndims--; |
| 235 |
– |
multcolor(ar.rcol, ar.rcoef); |
| 297 |
|
addcolor(asum, ar.rcol); |
| 298 |
|
} |
| 299 |
|
if (nss) { /* update returned ambient value */ |
| 309 |
|
} |
| 310 |
|
|
| 311 |
|
|
| 312 |
+ |
/* Compute vertex flags, indicating farthest in each direction */ |
| 313 |
+ |
static uby8 * |
| 314 |
+ |
vertex_flags(AMBHEMI *hp) |
| 315 |
+ |
{ |
| 316 |
+ |
uby8 *vflags = (uby8 *)calloc(hp->ns*hp->ns, sizeof(uby8)); |
| 317 |
+ |
uby8 *vf; |
| 318 |
+ |
AMBSAMP *ap; |
| 319 |
+ |
int i, j; |
| 320 |
+ |
|
| 321 |
+ |
if (vflags == NULL) |
| 322 |
+ |
error(SYSTEM, "out of memory in vertex_flags()"); |
| 323 |
+ |
vf = vflags; |
| 324 |
+ |
ap = hp->sa; /* compute farthest along first row */ |
| 325 |
+ |
for (j = 0; j < hp->ns-1; j++, vf++, ap++) |
| 326 |
+ |
if (ap[0].d <= ap[1].d) |
| 327 |
+ |
vf[0] |= 1<<VDB_X; |
| 328 |
+ |
else |
| 329 |
+ |
vf[1] |= 1<<VDB_x; |
| 330 |
+ |
++vf; ++ap; |
| 331 |
+ |
/* flag subsequent rows */ |
| 332 |
+ |
for (i = 1; i < hp->ns; i++) { |
| 333 |
+ |
for (j = 0; j < hp->ns-1; j++, vf++, ap++) { |
| 334 |
+ |
if (ap[0].d <= ap[-hp->ns].d) /* row before */ |
| 335 |
+ |
vf[0] |= 1<<VDB_y; |
| 336 |
+ |
else |
| 337 |
+ |
vf[-hp->ns] |= 1<<VDB_Y; |
| 338 |
+ |
if (ap[0].d <= ap[1-hp->ns].d) /* diagonal we care about */ |
| 339 |
+ |
vf[0] |= 1<<VDB_Xy; |
| 340 |
+ |
else |
| 341 |
+ |
vf[1-hp->ns] |= 1<<VDB_xY; |
| 342 |
+ |
if (ap[0].d <= ap[1].d) /* column after */ |
| 343 |
+ |
vf[0] |= 1<<VDB_X; |
| 344 |
+ |
else |
| 345 |
+ |
vf[1] |= 1<<VDB_x; |
| 346 |
+ |
} |
| 347 |
+ |
if (ap[0].d <= ap[-hp->ns].d) /* final column edge */ |
| 348 |
+ |
vf[0] |= 1<<VDB_y; |
| 349 |
+ |
else |
| 350 |
+ |
vf[-hp->ns] |= 1<<VDB_Y; |
| 351 |
+ |
++vf; ++ap; |
| 352 |
+ |
} |
| 353 |
+ |
return(vflags); |
| 354 |
+ |
} |
| 355 |
+ |
|
| 356 |
+ |
|
| 357 |
+ |
/* Return brightness of farthest ambient sample */ |
| 358 |
+ |
static double |
| 359 |
+ |
back_ambval(AMBHEMI *hp, int i, int j, int dbit1, int dbit2, const uby8 *vflags) |
| 360 |
+ |
{ |
| 361 |
+ |
const int v0 = ambndx(hp,i,j); |
| 362 |
+ |
const int tflags = (1<<dbit1 | 1<<dbit2); |
| 363 |
+ |
int v1, v2; |
| 364 |
+ |
|
| 365 |
+ |
if ((vflags[v0] & tflags) == tflags) /* is v0 the farthest? */ |
| 366 |
+ |
return(colval(hp->sa[v0].v,CIEY)); |
| 367 |
+ |
v1 = adjacent_verti(hp, i, j, dbit1); |
| 368 |
+ |
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)); |
| 377 |
+ |
} |
| 378 |
+ |
|
| 379 |
+ |
|
| 380 |
|
/* Compute vectors and coefficients for Hessian/gradient calcs */ |
| 381 |
|
static void |
| 382 |
< |
comp_fftri(FFTRI *ftp, FVECT ap0, FVECT ap1, FVECT rop) |
| 382 |
> |
comp_fftri(FFTRI *ftp, AMBHEMI *hp, int i, int j, int dbit, const uby8 *vflags) |
| 383 |
|
{ |
| 384 |
< |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
| 385 |
< |
int i; |
| 384 |
> |
const int i0 = ambndx(hp,i,j); |
| 385 |
> |
double rdot_cp, dot_e, dot_er, rdot_r, rdot_r1, J2; |
| 386 |
> |
int i1, ii; |
| 387 |
|
|
| 388 |
< |
VSUB(ftp->r_i, ap0, rop); |
| 389 |
< |
VSUB(ftp->r_i1, ap1, rop); |
| 390 |
< |
VSUB(ftp->e_i, ap1, ap0); |
| 388 |
> |
ftp->valid = 0; /* check if we can skip this edge */ |
| 389 |
> |
ii = adjacent_trifl[dbit]; |
| 390 |
> |
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); |
| 400 |
|
VCROSS(ftp->rcp, ftp->r_i, ftp->r_i1); |
| 401 |
|
rdot_cp = 1.0/DOT(ftp->rcp,ftp->rcp); |
| 402 |
|
dot_e = DOT(ftp->e_i,ftp->e_i); |
| 408 |
|
ftp->I2 = ( DOT(ftp->e_i, ftp->r_i1)*rdot_r1 - dot_er*rdot_r + |
| 409 |
|
dot_e*ftp->I1 )*0.5*rdot_cp; |
| 410 |
|
J2 = ( 0.5*(rdot_r - rdot_r1) - dot_er*ftp->I2 ) / dot_e; |
| 411 |
< |
for (i = 3; i--; ) |
| 412 |
< |
ftp->rI2_eJ2[i] = ftp->I2*ftp->r_i[i] + J2*ftp->e_i[i]; |
| 411 |
> |
for (ii = 3; ii--; ) |
| 412 |
> |
ftp->rI2_eJ2[ii] = ftp->I2*ftp->r_i[ii] + J2*ftp->e_i[ii]; |
| 413 |
> |
ftp->valid++; |
| 414 |
|
} |
| 415 |
|
|
| 416 |
|
|
| 436 |
|
double d1, d2, d3, d4; |
| 437 |
|
double I3, J3, K3; |
| 438 |
|
int i, j; |
| 439 |
+ |
|
| 440 |
+ |
if (!ftp->valid) { /* preemptive test */ |
| 441 |
+ |
memset(hess, 0, sizeof(FVECT)*3); |
| 442 |
+ |
return; |
| 443 |
+ |
} |
| 444 |
|
/* compute intermediate coefficients */ |
| 445 |
|
d1 = 1.0/DOT(ftp->r_i,ftp->r_i); |
| 446 |
|
d2 = 1.0/DOT(ftp->r_i1,ftp->r_i1); |
| 464 |
|
hess[i][j] = m1[i][j] + d1*( I3*m2[i][j] + K3*m3[i][j] + |
| 465 |
|
2.0*J3*m4[i][j] ); |
| 466 |
|
hess[i][j] += d2*(i==j); |
| 467 |
< |
hess[i][j] *= 1.0/PI; |
| 467 |
> |
hess[i][j] *= -1.0/PI; |
| 468 |
|
} |
| 469 |
|
} |
| 470 |
|
|
| 486 |
|
/* Add to radiometric Hessian from the given triangle */ |
| 487 |
|
static void |
| 488 |
|
add2hessian(FVECT hess[3], FVECT ehess1[3], |
| 489 |
< |
FVECT ehess2[3], FVECT ehess3[3], COLORV v) |
| 489 |
> |
FVECT ehess2[3], FVECT ehess3[3], double v) |
| 490 |
|
{ |
| 491 |
|
int i, j; |
| 492 |
|
|
| 504 |
|
double f1; |
| 505 |
|
int i; |
| 506 |
|
|
| 507 |
+ |
if (!ftp->valid) { /* preemptive test */ |
| 508 |
+ |
memset(grad, 0, sizeof(FVECT)); |
| 509 |
+ |
return; |
| 510 |
+ |
} |
| 511 |
|
f1 = 2.0*DOT(nrm, ftp->rcp); |
| 512 |
|
VCROSS(ncp, nrm, ftp->e_i); |
| 513 |
|
for (i = 3; i--; ) |
| 514 |
< |
grad[i] = (-0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
| 514 |
> |
grad[i] = (0.5/PI)*( ftp->I1*ncp[i] + f1*ftp->rI2_eJ2[i] ); |
| 515 |
|
} |
| 516 |
|
|
| 517 |
|
|
| 527 |
|
|
| 528 |
|
/* Add to displacement gradient from the given triangle */ |
| 529 |
|
static void |
| 530 |
< |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, COLORV v) |
| 530 |
> |
add2gradient(FVECT grad, FVECT egrad1, FVECT egrad2, FVECT egrad3, double v) |
| 531 |
|
{ |
| 532 |
|
int i; |
| 533 |
|
|
| 536 |
|
} |
| 537 |
|
|
| 538 |
|
|
| 390 |
– |
/* Return brightness of furthest ambient sample */ |
| 391 |
– |
static COLORV |
| 392 |
– |
back_ambval(AMBSAMP *ap1, AMBSAMP *ap2, AMBSAMP *ap3, FVECT orig) |
| 393 |
– |
{ |
| 394 |
– |
COLORV vback; |
| 395 |
– |
FVECT vec; |
| 396 |
– |
double d2, d2best; |
| 397 |
– |
|
| 398 |
– |
VSUB(vec, ap1->p, orig); |
| 399 |
– |
d2best = DOT(vec,vec); |
| 400 |
– |
vback = colval(ap1->v,CIEY); |
| 401 |
– |
VSUB(vec, ap2->p, orig); |
| 402 |
– |
d2 = DOT(vec,vec); |
| 403 |
– |
if (d2 > d2best) { |
| 404 |
– |
d2best = d2; |
| 405 |
– |
vback = colval(ap2->v,CIEY); |
| 406 |
– |
} |
| 407 |
– |
VSUB(vec, ap3->p, orig); |
| 408 |
– |
d2 = DOT(vec,vec); |
| 409 |
– |
if (d2 > d2best) |
| 410 |
– |
return(colval(ap3->v,CIEY)); |
| 411 |
– |
return(vback); |
| 412 |
– |
} |
| 413 |
– |
|
| 414 |
– |
|
| 539 |
|
/* Compute anisotropic radii and eigenvector directions */ |
| 540 |
|
static int |
| 541 |
|
eigenvectors(FVECT uv[2], float ra[2], FVECT hessian[3]) |
| 596 |
|
static char memerrmsg[] = "out of memory in ambHessian()"; |
| 597 |
|
FVECT (*hessrow)[3] = NULL; |
| 598 |
|
FVECT *gradrow = NULL; |
| 599 |
+ |
uby8 *vflags; |
| 600 |
|
FVECT hessian[3]; |
| 601 |
|
FVECT gradient; |
| 602 |
|
FFTRI fftr; |
| 618 |
|
error(SYSTEM, memerrmsg); |
| 619 |
|
memset(gradient, 0, sizeof(gradient)); |
| 620 |
|
} |
| 621 |
+ |
/* get vertex position flags */ |
| 622 |
+ |
vflags = vertex_flags(hp); |
| 623 |
|
/* compute first row of edges */ |
| 624 |
|
for (j = 0; j < hp->ns-1; j++) { |
| 625 |
< |
comp_fftri(&fftr, ambsamp(hp,0,j).p, |
| 499 |
< |
ambsamp(hp,0,j+1).p, hp->rp->rop); |
| 625 |
> |
comp_fftri(&fftr, hp, 0, j, VDB_X, vflags); |
| 626 |
|
if (hessrow != NULL) |
| 627 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
| 628 |
|
if (gradrow != NULL) |
| 632 |
|
for (i = 0; i < hp->ns-1; i++) { |
| 633 |
|
FVECT hesscol[3]; /* compute first vertical edge */ |
| 634 |
|
FVECT gradcol; |
| 635 |
< |
comp_fftri(&fftr, ambsamp(hp,i,0).p, |
| 510 |
< |
ambsamp(hp,i+1,0).p, hp->rp->rop); |
| 635 |
> |
comp_fftri(&fftr, hp, i, 0, VDB_Y, vflags); |
| 636 |
|
if (hessrow != NULL) |
| 637 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
| 638 |
|
if (gradrow != NULL) |
| 640 |
|
for (j = 0; j < hp->ns-1; j++) { |
| 641 |
|
FVECT hessdia[3]; /* compute triangle contributions */ |
| 642 |
|
FVECT graddia; |
| 643 |
< |
COLORV backg; |
| 644 |
< |
backg = back_ambval(&ambsamp(hp,i,j), &ambsamp(hp,i,j+1), |
| 520 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
| 643 |
> |
double backg; |
| 644 |
> |
backg = back_ambval(hp, i, j, VDB_X, VDB_Y, vflags); |
| 645 |
|
/* diagonal (inner) edge */ |
| 646 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, |
| 523 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
| 646 |
> |
comp_fftri(&fftr, hp, i, j+1, VDB_xY, vflags); |
| 647 |
|
if (hessrow != NULL) { |
| 648 |
|
comp_hessian(hessdia, &fftr, hp->rp->ron); |
| 649 |
|
rev_hessian(hesscol); |
| 655 |
|
add2gradient(gradient, gradrow[j], graddia, gradcol, backg); |
| 656 |
|
} |
| 657 |
|
/* initialize edge in next row */ |
| 658 |
< |
comp_fftri(&fftr, ambsamp(hp,i+1,j+1).p, |
| 536 |
< |
ambsamp(hp,i+1,j).p, hp->rp->rop); |
| 658 |
> |
comp_fftri(&fftr, hp, i+1, j+1, VDB_x, vflags); |
| 659 |
|
if (hessrow != NULL) |
| 660 |
|
comp_hessian(hessrow[j], &fftr, hp->rp->ron); |
| 661 |
|
if (gradrow != NULL) |
| 662 |
|
comp_gradient(gradrow[j], &fftr, hp->rp->ron); |
| 663 |
|
/* new column edge & paired triangle */ |
| 664 |
< |
backg = back_ambval(&ambsamp(hp,i,j+1), &ambsamp(hp,i+1,j+1), |
| 665 |
< |
&ambsamp(hp,i+1,j), hp->rp->rop); |
| 544 |
< |
comp_fftri(&fftr, ambsamp(hp,i,j+1).p, ambsamp(hp,i+1,j+1).p, |
| 545 |
< |
hp->rp->rop); |
| 664 |
> |
backg = back_ambval(hp, i+1, j+1, VDB_x, VDB_y, vflags); |
| 665 |
> |
comp_fftri(&fftr, hp, i, j+1, VDB_Y, vflags); |
| 666 |
|
if (hessrow != NULL) { |
| 667 |
|
comp_hessian(hesscol, &fftr, hp->rp->ron); |
| 668 |
|
rev_hessian(hessdia); |
| 682 |
|
/* release row buffers */ |
| 683 |
|
if (hessrow != NULL) free(hessrow); |
| 684 |
|
if (gradrow != NULL) free(gradrow); |
| 685 |
+ |
free(vflags); |
| 686 |
|
|
| 687 |
|
if (ra != NULL) /* extract eigenvectors & radii */ |
| 688 |
|
eigenvectors(uv, ra, hessian); |
| 718 |
|
} |
| 719 |
|
|
| 720 |
|
|
| 721 |
+ |
/* Compute potential light leak direction flags for cache value */ |
| 722 |
+ |
static uint32 |
| 723 |
+ |
ambcorral(AMBHEMI *hp, FVECT uv[2], const double r0, const double r1) |
| 724 |
+ |
{ |
| 725 |
+ |
uint32 flgs = 0; |
| 726 |
+ |
int i, j; |
| 727 |
+ |
/* circle around perimeter */ |
| 728 |
+ |
for (i = 0; i < hp->ns; i++) |
| 729 |
+ |
for (j = 0; j < hp->ns; j += !i|(i==hp->ns-1) ? 1 : hp->ns-1) { |
| 730 |
+ |
AMBSAMP *ap = &ambsam(hp,i,j); |
| 731 |
+ |
FVECT vec; |
| 732 |
+ |
double u, v; |
| 733 |
+ |
double ang; |
| 734 |
+ |
int abp; |
| 735 |
+ |
if (ap->d <= FTINY) |
| 736 |
+ |
continue; |
| 737 |
+ |
VSUB(vec, ap->p, hp->rp->rop); |
| 738 |
+ |
u = DOT(vec, uv[0]) * ap->d; |
| 739 |
+ |
v = DOT(vec, uv[1]) * ap->d; |
| 740 |
+ |
if ((r0*r0*u*u + r1*r1*v*v) * ap->d*ap->d <= 1.0) |
| 741 |
+ |
continue; /* occluder outside ellipse */ |
| 742 |
+ |
ang = atan2a(v, u); /* else set direction flags */ |
| 743 |
+ |
ang += 2.0*PI*(ang < 0); |
| 744 |
+ |
ang *= 16./PI; |
| 745 |
+ |
if ((ang < .5) | (ang >= 31.5)) |
| 746 |
+ |
flgs |= 0x80000001; |
| 747 |
+ |
else |
| 748 |
+ |
flgs |= 3L<<(int)(ang-.5); |
| 749 |
+ |
} |
| 750 |
+ |
return(flgs); |
| 751 |
+ |
} |
| 752 |
+ |
|
| 753 |
+ |
|
| 754 |
|
int |
| 755 |
|
doambient( /* compute ambient component */ |
| 756 |
|
COLOR rcol, /* input/output color */ |
| 759 |
|
FVECT uv[2], /* returned (optional) */ |
| 760 |
|
float ra[2], /* returned (optional) */ |
| 761 |
|
float pg[2], /* returned (optional) */ |
| 762 |
< |
float dg[2] /* returned (optional) */ |
| 762 |
> |
float dg[2], /* returned (optional) */ |
| 763 |
> |
uint32 *crlp /* returned (optional) */ |
| 764 |
|
) |
| 765 |
|
{ |
| 766 |
|
AMBHEMI *hp = inithemi(rcol, r, wt); |
| 767 |
< |
int cnt = 0; |
| 767 |
> |
int cnt; |
| 768 |
|
FVECT my_uv[2]; |
| 769 |
|
double d, K, acol[3]; |
| 770 |
|
AMBSAMP *ap; |
| 780 |
|
pg[0] = pg[1] = 0.0; |
| 781 |
|
if (dg != NULL) |
| 782 |
|
dg[0] = dg[1] = 0.0; |
| 783 |
+ |
if (crlp != NULL) |
| 784 |
+ |
*crlp = 0; |
| 785 |
|
/* sample the hemisphere */ |
| 786 |
|
acol[0] = acol[1] = acol[2] = 0.0; |
| 787 |
+ |
cnt = 0; |
| 788 |
|
for (i = hp->ns; i--; ) |
| 789 |
|
for (j = hp->ns; j--; ) |
| 790 |
|
if ((ap = ambsample(hp, i, j)) != NULL) { |
| 809 |
|
free(hp); |
| 810 |
|
return(-1); /* no radius or gradient calc. */ |
| 811 |
|
} |
| 812 |
< |
if (bright(acol) > FTINY) { /* normalize Y values */ |
| 813 |
< |
d = 0.99*cnt/bright(acol); |
| 812 |
> |
if ((d = bright(acol)) > FTINY) { /* normalize Y values */ |
| 813 |
> |
d = 0.99*(hp->ns*hp->ns)/d; |
| 814 |
|
K = 0.01; |
| 815 |
< |
} else { /* geometric Hessian fall-back */ |
| 658 |
< |
d = 0.0; |
| 815 |
> |
} else { /* or fall back on geometric Hessian */ |
| 816 |
|
K = 1.0; |
| 817 |
|
pg = NULL; |
| 818 |
|
dg = NULL; |
| 851 |
|
if (ra[0] > maxarad) |
| 852 |
|
ra[0] = maxarad; |
| 853 |
|
} |
| 854 |
+ |
if (crlp != NULL) /* flag encroached directions */ |
| 855 |
+ |
*crlp = ambcorral(hp, uv, ra[0]*ambacc, ra[1]*ambacc); |
| 856 |
|
if (pg != NULL) { /* cap gradient if necessary */ |
| 857 |
|
d = pg[0]*pg[0]*ra[0]*ra[0] + pg[1]*pg[1]*ra[1]*ra[1]; |
| 858 |
|
if (d > 1.0) { |