--- ray/src/rt/ambient.c 1990/06/26 09:00:07 1.10 +++ ray/src/rt/ambient.c 2014/04/24 19:16:52 2.78 @@ -1,132 +1,297 @@ -/* Copyright (c) 1986 Regents of the University of California */ - #ifndef lint -static char SCCSid[] = "$SunId$ LBL"; +static const char RCSid[] = "$Id: ambient.c,v 2.78 2014/04/24 19:16:52 greg Exp $"; #endif - /* * ambient.c - routines dealing with ambient (inter-reflected) component. * - * The macro AMBFLUSH (if defined) is the number of ambient values - * to wait before flushing to the ambient file. - * - * 5/9/86 + * Declarations of external symbols in ambient.h */ -#include "ray.h" +#include "copyright.h" -#include "octree.h" +#include +#include "platform.h" +#include "ray.h" #include "otypes.h" - +#include "resolu.h" +#include "ambient.h" #include "random.h" -#define OCTSCALE 0.5 /* ceil((valid rad.)/(cube size)) */ +#ifndef OCTSCALE +#define OCTSCALE 1.0 /* ceil((valid rad.)/(cube size)) */ +#endif -extern CUBE thescene; /* contains space boundaries */ +extern char *shm_boundary; /* memory sharing boundary */ -extern COLOR ambval; /* global ambient component */ -extern double ambacc; /* ambient accuracy */ -extern int ambres; /* ambient resolution */ -extern int ambdiv; /* number of divisions for calculation */ -extern int ambssamp; /* number of super-samples */ -extern int ambounce; /* number of ambient bounces */ -extern char *amblist[]; /* ambient include/exclude list */ -extern int ambincl; /* include == 1, exclude == 0 */ +#ifndef MAXASET +#define MAXASET 4095 /* maximum number of elements in ambient set */ +#endif +OBJECT ambset[MAXASET+1]={0}; /* ambient include/exclude set */ -OBJECT ambset[256]={0}; /* ambient include/exclude set */ +double maxarad; /* maximum ambient radius */ +double minarad; /* minimum ambient radius */ -double maxarad; /* maximum ambient radius */ -double minarad; /* minimum ambient radius */ +static AMBTREE atrunk; /* our ambient trunk node */ -typedef struct ambval { - FVECT pos; /* position in space */ - FVECT dir; /* normal direction */ - int lvl; /* recursion level of parent ray */ - float weight; /* weight of parent ray */ - COLOR val; /* computed ambient value */ - float rad; /* validity radius */ - struct ambval *next; /* next in list */ -} AMBVAL; /* ambient value */ +static FILE *ambfp = NULL; /* ambient file pointer */ +static int nunflshed = 0; /* number of unflushed ambient values */ -typedef struct ambtree { - AMBVAL *alist; /* ambient value list */ - struct ambtree *kid; /* 8 child nodes */ -} AMBTREE; /* ambient octree */ +#ifndef SORT_THRESH +#ifdef SMLMEM +#define SORT_THRESH ((16L<<20)/sizeof(AMBVAL)) +#else +#define SORT_THRESH ((64L<<20)/sizeof(AMBVAL)) +#endif +#endif +#ifndef SORT_INTVL +#define SORT_INTVL (SORT_THRESH<<1) +#endif +#ifndef MAX_SORT_INTVL +#define MAX_SORT_INTVL (SORT_INTVL<<6) +#endif -typedef struct { - float k; /* error contribution per sample */ - COLOR v; /* ray sum */ - int n; /* number of samples */ - short t, p; /* theta, phi indices */ -} AMBSAMP; /* ambient sample */ -static AMBTREE atrunk; /* our ambient trunk node */ +static double qambacc = 0.; /* ambient accuracy to the 1/4 power */ +static double avsum = 0.; /* computed ambient value sum (log) */ +static unsigned int navsum = 0; /* number of values in avsum */ +static unsigned int nambvals = 0; /* total number of indirect values */ +static unsigned int nambshare = 0; /* number of values from file */ +static unsigned long ambclock = 0; /* ambient access clock */ +static unsigned long lastsort = 0; /* time of last value sort */ +static long sortintvl = SORT_INTVL; /* time until next sort */ +static FILE *ambinp = NULL; /* auxiliary file for input */ +static long lastpos = -1; /* last flush position */ -static FILE *ambfp = NULL; /* ambient file pointer */ +#define MAXACLOCK (1L<<30) /* clock turnover value */ + /* + * Track access times unless we are sharing ambient values + * through memory on a multiprocessor, when we want to avoid + * claiming our own memory (copy on write). Go ahead anyway + * if more than two thirds of our values are unshared. + * Compile with -Dtracktime=0 to turn this code off. + */ +#ifndef tracktime +#define tracktime (shm_boundary == NULL || nambvals > 3*nambshare) +#endif -#define newambval() (AMBVAL *)bmalloc(sizeof(AMBVAL)) +#define AMBFLUSH (BUFSIZ/AMBVALSIZ) -#define newambtree() (AMBTREE *)calloc(8, sizeof(AMBTREE)) +#define newambval() (AMBVAL *)malloc(sizeof(AMBVAL)) +#define freeav(av) free((void *)av); -double sumambient(), doambient(), makeambient(); +static void initambfile(int creat); +static void avsave(AMBVAL *av); +static AMBVAL *avstore(AMBVAL *aval); +static AMBTREE *newambtree(void); +static void freeambtree(AMBTREE *atp); +typedef void unloadtf_t(AMBVAL *); +static unloadtf_t avinsert; +static unloadtf_t av2list; +static unloadtf_t avfree; +static void unloadatree(AMBTREE *at, unloadtf_t *f); -setambient(afile) /* initialize calculation */ -char *afile; +static int aposcmp(const void *avp1, const void *avp2); +static int avlmemi(AMBVAL *avaddr); +static void sortambvals(int always); + +#ifdef F_SETLKW +static void aflock(int typ); +#endif + + +void +setambres( /* set ambient resolution */ + int ar +) { - long ftell(); - AMBVAL amb; + ambres = ar < 0 ? 0 : ar; /* may be done already */ + /* set min & max radii */ + if (ar <= 0) { + minarad = 0; + maxarad = thescene.cusize / 2.0; + } else { + minarad = thescene.cusize / ar; + maxarad = 64 * minarad; /* heuristic */ + if (maxarad > thescene.cusize / 2.0) + maxarad = thescene.cusize / 2.0; + } + if (minarad <= FTINY) + minarad = 10*FTINY; + if (maxarad <= minarad) + maxarad = 64 * minarad; +} - maxarad = thescene.cusize / 2.0; /* maximum radius */ - /* minimum radius */ - minarad = ambres > 0 ? thescene.cusize/ambres : 0.0; - /* open ambient file */ - if (afile != NULL) - if ((ambfp = fopen(afile, "r+")) != NULL) { - while (fread((char *)&amb,sizeof(AMBVAL),1,ambfp) == 1) - avinsert(&amb, &atrunk, thescene.cuorg, - thescene.cusize); - /* align */ - fseek(ambfp, -(ftell(ambfp)%sizeof(AMBVAL)), 1); - } else if ((ambfp = fopen(afile, "w")) == NULL) { - sprintf(errmsg, "cannot open ambient file \"%s\"", - afile); - error(SYSTEM, errmsg); +void +setambacc( /* set ambient accuracy */ + double newa +) +{ + double ambdiff; + + if (newa < 0.0) + newa = 0.0; + ambdiff = fabs(newa - ambacc); + if (ambdiff >= .01 && (ambacc = newa) > FTINY) { + qambacc = sqrt(sqrt(ambacc)); + if (nambvals > 0) + sortambvals(1); /* rebuild tree */ + } +} + + +void +setambient(void) /* initialize calculation */ +{ + int readonly = 0; + long flen; + AMBVAL amb; + /* make sure we're fresh */ + ambdone(); + /* init ambient limits */ + setambres(ambres); + qambacc = sqrt(sqrt(ambacc *= (ambacc > FTINY))); + if (ambfile == NULL || !ambfile[0]) + return; + if (ambacc <= FTINY) { + sprintf(errmsg, "zero ambient accuracy so \"%s\" not opened", + ambfile); + error(WARNING, errmsg); + return; + } + /* open ambient file */ + if ((ambfp = fopen(ambfile, "r+")) == NULL) + readonly = (ambfp = fopen(ambfile, "r")) != NULL; + if (ambfp != NULL) { + initambfile(0); /* file exists */ + lastpos = ftell(ambfp); + while (readambval(&amb, ambfp)) + avinsert(avstore(&amb)); + nambshare = nambvals; /* share loaded values */ + if (readonly) { + sprintf(errmsg, + "loaded %u values from read-only ambient file", + nambvals); + error(WARNING, errmsg); + fclose(ambfp); /* close file so no writes */ + ambfp = NULL; + return; /* avoid ambsync() */ } + /* align file pointer */ + lastpos += (long)nambvals*AMBVALSIZ; + flen = lseek(fileno(ambfp), (off_t)0, SEEK_END); + if (flen != lastpos) { + sprintf(errmsg, + "ignoring last %ld values in ambient file (corrupted)", + (flen - lastpos)/AMBVALSIZ); + error(WARNING, errmsg); + fseek(ambfp, lastpos, SEEK_SET); +#ifndef _WIN32 /* XXX we need a replacement for that one */ + ftruncate(fileno(ambfp), (off_t)lastpos); +#endif + } + } else if ((ambfp = fopen(ambfile, "w+")) != NULL) { + initambfile(1); /* else create new file */ + fflush(ambfp); + lastpos = ftell(ambfp); + } else { + sprintf(errmsg, "cannot open ambient file \"%s\"", ambfile); + error(SYSTEM, errmsg); + } +#ifdef getc_unlocked + flockfile(ambfp); /* application-level lock */ +#endif +#ifdef F_SETLKW + aflock(F_UNLCK); /* release file */ +#endif } -ambnotify(obj) /* record new modifier */ -OBJECT obj; +void +ambdone(void) /* close ambient file and free memory */ { - register OBJREC *o = objptr(obj); - register char **amblp; + if (ambfp != NULL) { /* close ambient file */ + ambsync(); + fclose(ambfp); + ambfp = NULL; + if (ambinp != NULL) { + fclose(ambinp); + ambinp = NULL; + } + lastpos = -1; + } + /* free ambient tree */ + unloadatree(&atrunk, &avfree); + /* reset state variables */ + avsum = 0.; + navsum = 0; + nambvals = 0; + nambshare = 0; + ambclock = 0; + lastsort = 0; + sortintvl = SORT_INTVL; +} - if (!ismodifier(o->otype)) + +void +ambnotify( /* record new modifier */ + OBJECT obj +) +{ + static int hitlimit = 0; + OBJREC *o; + char **amblp; + + if (obj == OVOID) { /* starting over */ + ambset[0] = 0; + hitlimit = 0; return; + } + o = objptr(obj); + if (hitlimit || !ismodifier(o->otype)) + return; for (amblp = amblist; *amblp != NULL; amblp++) if (!strcmp(o->oname, *amblp)) { + if (ambset[0] >= MAXASET) { + error(WARNING, "too many modifiers in ambient list"); + hitlimit++; + return; /* should this be fatal? */ + } insertelem(ambset, obj); return; } } +/************ THE FOLLOWING ROUTINES DIFFER BETWEEN NEW & OLD ***************/ -ambient(acol, r) /* compute ambient component for ray */ -COLOR acol; -register RAY *r; +#ifdef NEWAMB + +#define tfunc(lwr, x, upr) (((x)-(lwr))/((upr)-(lwr))) + +static double sumambient(COLOR acol, RAY *r, FVECT rn, int al, + AMBTREE *at, FVECT c0, double s); +static int makeambient(COLOR acol, RAY *r, FVECT rn, int al); +static void extambient(COLOR cr, AMBVAL *ap, FVECT pv, FVECT nv, + FVECT uvw[3]); + +void +multambient( /* compute ambient component & multiply by coef. */ + COLOR aval, + RAY *r, + FVECT nrm +) { static int rdepth = 0; /* ambient recursion */ - double wsum; + COLOR acol; + int ok; + double d, l; - rdepth++; /* increment level */ - if (ambdiv <= 0) /* no ambient calculation */ goto dumbamb; /* check number of bounces */ - if (rdepth > ambounce) + if (rdepth >= ambounce) goto dumbamb; /* check ambient list */ if (ambincl != -1 && r->ro != NULL && @@ -134,66 +299,397 @@ register RAY *r; goto dumbamb; if (ambacc <= FTINY) { /* no ambient storage */ - if (doambient(acol, r) == 0.0) + copycolor(acol, aval); + rdepth++; + ok = doambient(acol, r, r->rweight, NULL, NULL, NULL, NULL); + rdepth--; + if (!ok) goto dumbamb; - goto done; + copycolor(aval, acol); + return; } - /* get ambient value */ + + if (tracktime) /* sort to minimize thrashing */ + sortambvals(0); + /* interpolate ambient value */ setcolor(acol, 0.0, 0.0, 0.0); - wsum = sumambient(acol, r, &atrunk, thescene.cuorg, thescene.cusize); - if (wsum > FTINY) - scalecolor(acol, 1.0/wsum); - else if (makeambient(acol, r) == 0.0) + d = sumambient(acol, r, nrm, rdepth, + &atrunk, thescene.cuorg, thescene.cusize); + if (d > FTINY) { + d = 1.0/d; + scalecolor(acol, d); + multcolor(aval, acol); + return; + } + rdepth++; /* need to cache new value */ + ok = makeambient(acol, r, nrm, rdepth-1); + rdepth--; + if (ok) { + multcolor(aval, acol); /* computed new value */ + return; + } +dumbamb: /* return global value */ + if ((ambvwt <= 0) | (navsum == 0)) { + multcolor(aval, ambval); + return; + } + l = bright(ambval); /* average in computations */ + if (l > FTINY) { + d = (log(l)*(double)ambvwt + avsum) / + (double)(ambvwt + navsum); + d = exp(d) / l; + scalecolor(aval, d); + multcolor(aval, ambval); /* apply color of ambval */ + } else { + d = exp( avsum / (double)navsum ); + scalecolor(aval, d); /* neutral color */ + } +} + + +double +sumambient( /* get interpolated ambient value */ + COLOR acol, + RAY *r, + FVECT rn, + int al, + AMBTREE *at, + FVECT c0, + double s +) +{ /* initial limit is ambacc radians */ + const double maxangle = (ambacc-PI/2.)*pow(r->rweight,0.13) + PI/2.; + double wsum = 0.0; + FVECT ck0; + int i, j; + AMBVAL *av; + /* sum this node */ + for (av = at->alist; av != NULL; av = av->next) { + double d, delta_r2, delta_t2; + COLOR ct; + FVECT uvw[3]; + /* record access */ + if (tracktime) + av->latick = ambclock; + /* + * Ambient level test + */ + if (av->lvl > al) /* list sorted, so this works */ + break; + if (av->weight < 0.9*r->rweight) + continue; + /* + * Direction test using unperturbed normal + */ + decodedir(uvw[2], av->ndir); + d = DOT(uvw[2], r->ron); + if (d <= 0.0) /* >= 90 degrees */ + continue; + delta_r2 = 2.0 - 2.0*d; /* approx. radians^2 */ + if (delta_r2 >= maxangle*maxangle) + continue; + /* + * Elliptical radii test based on Hessian + */ + decodedir(uvw[0], av->udir); + VCROSS(uvw[1], uvw[2], uvw[0]); + VSUB(ck0, av->pos, r->rop); + d = DOT(ck0, uvw[0]) / av->rad[0]; + delta_t2 = d*d; + d = DOT(ck0, uvw[1]) / av->rad[1]; + delta_t2 += d*d; + if (delta_t2 >= qambacc*qambacc) + continue; + /* + * Intersection behind test + */ + d = 0.0; + for (j = 0; j < 3; j++) + d += (r->rop[j] - av->pos[j])*(uvw[2][j] + r->ron[j]); + if (d*0.5 < -minarad*qambacc-.001) + continue; + /* + * Extrapolate value and compute final weight (hat function) + */ + extambient(ct, av, r->rop, rn, uvw); + d = tfunc(maxangle, sqrt(delta_r2), 0.0) * + tfunc(qambacc, sqrt(delta_t2), 0.0); + scalecolor(ct, d); + addcolor(acol, ct); + wsum += d; + } + if (at->kid == NULL) + return(wsum); + /* sum children */ + s *= 0.5; + for (i = 0; i < 8; i++) { + for (j = 0; j < 3; j++) { + ck0[j] = c0[j]; + if (1<rop[j] < ck0[j] - OCTSCALE*s) + break; + if (r->rop[j] > ck0[j] + (1.0+OCTSCALE)*s) + break; + } + if (j == 3) + wsum += sumambient(acol, r, rn, al, + at->kid+i, ck0, s); + } + return(wsum); +} + + +int +makeambient( /* make a new ambient value for storage */ + COLOR acol, + RAY *r, + FVECT rn, + int al +) +{ + AMBVAL amb; + FVECT uvw[3]; + int i; + + amb.weight = 1.0; /* compute weight */ + for (i = al; i-- > 0; ) + amb.weight *= AVGREFL; + if (r->rweight < 0.1*amb.weight) /* heuristic override */ + amb.weight = 1.25*r->rweight; + setcolor(acol, AVGREFL, AVGREFL, AVGREFL); + /* compute ambient */ + i = doambient(acol, r, amb.weight, uvw, amb.rad, amb.gpos, amb.gdir); + scalecolor(acol, 1./AVGREFL); /* undo assumed reflectance */ + if (i <= 0 || amb.rad[0] <= FTINY) /* no Hessian or zero radius */ + return(i); + /* store value */ + VCOPY(amb.pos, r->rop); + amb.ndir = encodedir(r->ron); + amb.udir = encodedir(uvw[0]); + amb.lvl = al; + copycolor(amb.val, acol); + /* insert into tree */ + avsave(&amb); /* and save to file */ + if (rn != r->ron) { /* texture */ + VCOPY(uvw[2], r->ron); + extambient(acol, &amb, r->rop, rn, uvw); + } + return(1); +} + + +void +extambient( /* extrapolate value at pv, nv */ + COLOR cr, + AMBVAL *ap, + FVECT pv, + FVECT nv, + FVECT uvw[3] +) +{ + static FVECT my_uvw[3]; + FVECT v1; + int i; + double d = 1.0; /* zeroeth order */ + + if (uvw == NULL) { /* need local coordinates? */ + decodedir(my_uvw[2], ap->ndir); + decodedir(my_uvw[0], ap->udir); + VCROSS(my_uvw[1], my_uvw[2], my_uvw[0]); + uvw = my_uvw; + } + for (i = 3; i--; ) /* gradient due to translation */ + d += (pv[i] - ap->pos[i]) * + (ap->gpos[0]*uvw[0][i] + ap->gpos[1]*uvw[1][i]); + + VCROSS(v1, uvw[2], nv); /* gradient due to rotation */ + for (i = 3; i--; ) + d += v1[i] * (ap->gdir[0]*uvw[0][i] + ap->gdir[1]*uvw[1][i]); + + if (d <= 0.0) { + setcolor(cr, 0.0, 0.0, 0.0); + return; + } + copycolor(cr, ap->val); + scalecolor(cr, d); +} + + +static void +avinsert( /* insert ambient value in our tree */ + AMBVAL *av +) +{ + AMBTREE *at; + AMBVAL *ap; + AMBVAL avh; + FVECT ck0; + double s; + int branch; + int i; + + if (av->rad[0] <= FTINY) + error(CONSISTENCY, "zero ambient radius in avinsert"); + at = &atrunk; + VCOPY(ck0, thescene.cuorg); + s = thescene.cusize; + while (s*(OCTSCALE/2) > av->rad[1]*qambacc) { + if (at->kid == NULL) + if ((at->kid = newambtree()) == NULL) + error(SYSTEM, "out of memory in avinsert"); + s *= 0.5; + branch = 0; + for (i = 0; i < 3; i++) + if (av->pos[i] > ck0[i] + s) { + ck0[i] += s; + branch |= 1 << i; + } + at = at->kid + branch; + } + avh.next = at->alist; /* order by increasing level */ + for (ap = &avh; ap->next != NULL; ap = ap->next) + if (ap->next->lvl >= av->lvl) + break; + av->next = ap->next; + ap->next = (AMBVAL*)av; + at->alist = avh.next; +} + + +#else /* ! NEWAMB */ + +static double sumambient(COLOR acol, RAY *r, FVECT rn, int al, + AMBTREE *at, FVECT c0, double s); +static double makeambient(COLOR acol, RAY *r, FVECT rn, int al); +static void extambient(COLOR cr, AMBVAL *ap, FVECT pv, FVECT nv); + + +void +multambient( /* compute ambient component & multiply by coef. */ + COLOR aval, + RAY *r, + FVECT nrm +) +{ + static int rdepth = 0; /* ambient recursion */ + COLOR acol; + double d, l; + + if (ambdiv <= 0) /* no ambient calculation */ goto dumbamb; - goto done; + /* check number of bounces */ + if (rdepth >= ambounce) + goto dumbamb; + /* check ambient list */ + if (ambincl != -1 && r->ro != NULL && + ambincl != inset(ambset, r->ro->omod)) + goto dumbamb; -dumbamb: /* return global value */ - copycolor(acol, ambval); -done: /* must finish here! */ + if (ambacc <= FTINY) { /* no ambient storage */ + copycolor(acol, aval); + rdepth++; + d = doambient(acol, r, r->rweight, NULL, NULL); + rdepth--; + if (d <= FTINY) + goto dumbamb; + copycolor(aval, acol); + return; + } + + if (tracktime) /* sort to minimize thrashing */ + sortambvals(0); + /* interpolate ambient value */ + setcolor(acol, 0.0, 0.0, 0.0); + d = sumambient(acol, r, nrm, rdepth, + &atrunk, thescene.cuorg, thescene.cusize); + if (d > FTINY) { + d = 1.0/d; + scalecolor(acol, d); + multcolor(aval, acol); + return; + } + rdepth++; /* need to cache new value */ + d = makeambient(acol, r, nrm, rdepth-1); rdepth--; + if (d > FTINY) { + multcolor(aval, acol); /* got new value */ + return; + } +dumbamb: /* return global value */ + if ((ambvwt <= 0) | (navsum == 0)) { + multcolor(aval, ambval); + return; + } + l = bright(ambval); /* average in computations */ + if (l > FTINY) { + d = (log(l)*(double)ambvwt + avsum) / + (double)(ambvwt + navsum); + d = exp(d) / l; + scalecolor(aval, d); + multcolor(aval, ambval); /* apply color of ambval */ + } else { + d = exp( avsum / (double)navsum ); + scalecolor(aval, d); /* neutral color */ + } } -double -sumambient(acol, r, at, c0, s) /* get interpolated ambient value */ -COLOR acol; -register RAY *r; -AMBTREE *at; -FVECT c0; -double s; +static double +sumambient( /* get interpolated ambient value */ + COLOR acol, + RAY *r, + FVECT rn, + int al, + AMBTREE *at, + FVECT c0, + double s +) { - extern double sqrt(); - double d, e1, e2, wt, wsum; + double d, e1, e2, wt, wsum; COLOR ct; FVECT ck0; int i; - register int j; - register AMBVAL *av; - /* do this node */ + int j; + AMBVAL *av; + wsum = 0.0; + /* do this node */ for (av = at->alist; av != NULL; av = av->next) { + double rn_dot = -2.0; + if (tracktime) + av->latick = ambclock; /* - * Ray strength test. + * Ambient level test. */ - if (av->lvl > r->rlvl || av->weight < r->rweight-FTINY) + if (av->lvl > al) /* list sorted, so this works */ + break; + if (av->weight < 0.9*r->rweight) continue; /* * Ambient radius test. */ - e1 = 0.0; - for (j = 0; j < 3; j++) { - d = av->pos[j] - r->rop[j]; - e1 += d * d; - } - e1 /= av->rad * av->rad; + VSUB(ck0, av->pos, r->rop); + e1 = DOT(ck0, ck0) / (av->rad * av->rad); if (e1 > ambacc*ambacc*1.21) continue; /* - * Normal direction test. + * Direction test using closest normal. */ - e2 = (1.0 - DOT(av->dir, r->ron)) * r->rweight; - if (e2 < 0.0) e2 = 0.0; - if (e1 + e2 > ambacc*ambacc*1.21) + d = DOT(av->dir, r->ron); + if (rn != r->ron) { + rn_dot = DOT(av->dir, rn); + if (rn_dot > 1.0-FTINY) + rn_dot = 1.0-FTINY; + if (rn_dot >= d-FTINY) { + d = rn_dot; + rn_dot = -2.0; + } + } + e2 = (1.0 - d) * r->rweight; + if (e2 < 0.0) + e2 = 0.0; + else if (e1 + e2 > ambacc*ambacc*1.21) continue; /* * Ray behind test. @@ -202,21 +698,29 @@ double s; for (j = 0; j < 3; j++) d += (r->rop[j] - av->pos[j]) * (av->dir[j] + r->ron[j]); - if (d < -minarad) + if (d*0.5 < -minarad*ambacc-.001) continue; /* * Jittering final test reduces image artifacts. */ - wt = sqrt(e1) + sqrt(e2); - wt *= .9 + .2*frandom(); - if (wt > ambacc) + e1 = sqrt(e1); + e2 = sqrt(e2); + wt = e1 + e2; + if (wt > ambacc*(.9+.2*urand(9015+samplendx))) continue; + /* + * Recompute directional error using perturbed normal + */ + if (rn_dot > 0.0) { + e2 = sqrt((1.0 - rn_dot)*r->rweight); + wt = e1 + e2; + } if (wt <= 1e-3) wt = 1e3; else wt = 1.0 / wt; wsum += wt; - copycolor(ct, av->val); + extambient(ct, av, r->rop, rn); scalecolor(ct, wt); addcolor(acol, ct); } @@ -235,255 +739,513 @@ double s; break; } if (j == 3) - wsum += sumambient(acol, r, at->kid+i, ck0, s); + wsum += sumambient(acol, r, rn, al, + at->kid+i, ck0, s); } return(wsum); } -double -makeambient(acol, r) /* make a new ambient value */ -COLOR acol; -register RAY *r; +static double +makeambient( /* make a new ambient value for storage */ + COLOR acol, + RAY *r, + FVECT rn, + int al +) { - AMBVAL amb; + AMBVAL amb; + FVECT gp, gd; + int i; - amb.rad = doambient(acol, r); /* compute ambient */ - if (amb.rad == 0.0) + amb.weight = 1.0; /* compute weight */ + for (i = al; i-- > 0; ) + amb.weight *= AVGREFL; + if (r->rweight < 0.1*amb.weight) /* heuristic override */ + amb.weight = 1.25*r->rweight; + setcolor(acol, AVGREFL, AVGREFL, AVGREFL); + /* compute ambient */ + amb.rad = doambient(acol, r, amb.weight, gp, gd); + if (amb.rad <= FTINY) { + setcolor(acol, 0.0, 0.0, 0.0); return(0.0); - /* store it */ + } + scalecolor(acol, 1./AVGREFL); /* undo assumed reflectance */ + /* store value */ VCOPY(amb.pos, r->rop); VCOPY(amb.dir, r->ron); - amb.lvl = r->rlvl; - amb.weight = r->rweight; + amb.lvl = al; copycolor(amb.val, acol); + VCOPY(amb.gpos, gp); + VCOPY(amb.gdir, gd); /* insert into tree */ - avinsert(&amb, &atrunk, thescene.cuorg, thescene.cusize); - avsave(&amb); /* write to file */ + avsave(&amb); /* and save to file */ + if (rn != r->ron) + extambient(acol, &amb, r->rop, rn); /* texture */ return(amb.rad); } -double -doambient(acol, r) /* compute ambient component */ -COLOR acol; -register RAY *r; +static void +extambient( /* extrapolate value at pv, nv */ + COLOR cr, + AMBVAL *ap, + FVECT pv, + FVECT nv +) { - extern int ambcmp(); - extern double sin(), cos(), sqrt(); - double phi, xd, yd, zd; - double b, b2; - register AMBSAMP *div; - AMBSAMP dnew; - RAY ar; - FVECT ux, uy; - double arad; - int ndivs, nt, np, ns, ne, i, j; - register int k; + FVECT v1; + int i; + double d; - setcolor(acol, 0.0, 0.0, 0.0); - /* set number of divisions */ - nt = sqrt(ambdiv * r->rweight * 0.5) + 0.5; - np = 2 * nt; - ndivs = nt * np; - /* check first */ - if (ndivs == 0 || rayorigin(&ar, r, AMBIENT, 0.5) < 0) - return(0.0); - /* set number of super-samples */ - ns = ambssamp * r->rweight + 0.5; - if (ns > 0) { - div = (AMBSAMP *)malloc(ndivs*sizeof(AMBSAMP)); - if (div == NULL) - error(SYSTEM, "out of memory in doambient"); + d = 1.0; /* zeroeth order */ + /* gradient due to translation */ + for (i = 0; i < 3; i++) + d += ap->gpos[i]*(pv[i]-ap->pos[i]); + /* gradient due to rotation */ + VCROSS(v1, ap->dir, nv); + d += DOT(ap->gdir, v1); + if (d <= 0.0) { + setcolor(cr, 0.0, 0.0, 0.0); + return; } - /* make axes */ - uy[0] = uy[1] = uy[2] = 0.0; - for (k = 0; k < 3; k++) - if (r->ron[k] < 0.6 && r->ron[k] > -0.6) + copycolor(cr, ap->val); + scalecolor(cr, d); +} + + +static void +avinsert( /* insert ambient value in our tree */ + AMBVAL *av +) +{ + AMBTREE *at; + AMBVAL *ap; + AMBVAL avh; + FVECT ck0; + double s; + int branch; + int i; + + if (av->rad <= FTINY) + error(CONSISTENCY, "zero ambient radius in avinsert"); + at = &atrunk; + VCOPY(ck0, thescene.cuorg); + s = thescene.cusize; + while (s*(OCTSCALE/2) > av->rad*ambacc) { + if (at->kid == NULL) + if ((at->kid = newambtree()) == NULL) + error(SYSTEM, "out of memory in avinsert"); + s *= 0.5; + branch = 0; + for (i = 0; i < 3; i++) + if (av->pos[i] > ck0[i] + s) { + ck0[i] += s; + branch |= 1 << i; + } + at = at->kid + branch; + } + avh.next = at->alist; /* order by increasing level */ + for (ap = &avh; ap->next != NULL; ap = ap->next) + if (ap->next->lvl >= av->lvl) break; - uy[k] = 1.0; - fcross(ux, r->ron, uy); - normalize(ux); - fcross(uy, ux, r->ron); - /* sample divisions */ - arad = 0.0; - ne = 0; - for (i = 0; i < nt; i++) - for (j = 0; j < np; j++) { - rayorigin(&ar, r, AMBIENT, 0.5); /* pretested */ - zd = sqrt((i+frandom())/nt); - phi = 2.0*PI * (j+frandom())/np; - xd = cos(phi) * zd; - yd = sin(phi) * zd; - zd = sqrt(1.0 - zd*zd); - for (k = 0; k < 3; k++) - ar.rdir[k] = xd*ux[k]+yd*uy[k]+zd*r->ron[k]; - rayvalue(&ar); - if (ar.rot < FHUGE) - arad += 1.0 / ar.rot; - if (ns > 0) { /* save division */ - div[ne].k = 0.0; - copycolor(div[ne].v, ar.rcol); - div[ne].n = 0; - div[ne].t = i; div[ne].p = j; - /* sum errors */ - b = bright(ar.rcol); - if (i > 0) { /* from above */ - b2 = bright(div[ne-np].v) - b; - b2 *= b2 * 0.25; - div[ne].k += b2; - div[ne].n++; - div[ne-np].k += b2; - div[ne-np].n++; - } - if (j > 0) { /* from behind */ - b2 = bright(div[ne-1].v) - b; - b2 *= b2 * 0.25; - div[ne].k += b2; - div[ne].n++; - div[ne-1].k += b2; - div[ne-1].n++; - } - if (j == np-1) { /* around */ - b2 = bright(div[ne-(np-1)].v) - b; - b2 *= b2 * 0.25; - div[ne].k += b2; - div[ne].n++; - div[ne-(np-1)].k += b2; - div[ne-(np-1)].n++; - } - ne++; - } else - addcolor(acol, ar.rcol); - } - for (k = 0; k < ne; k++) { /* compute errors */ - if (div[k].n > 1) - div[k].k /= div[k].n; - div[k].n = 1; + av->next = ap->next; + ap->next = (AMBVAL*)av; + at->alist = avh.next; +} + +#endif /* ! NEWAMB */ + +/************* FOLLOWING ROUTINES SAME FOR NEW & OLD METHODS ***************/ + +static void +initambfile( /* initialize ambient file */ + int cre8 +) +{ + extern char *progname, *octname; + static char *mybuf = NULL; + +#ifdef F_SETLKW + aflock(cre8 ? F_WRLCK : F_RDLCK); +#endif + SET_FILE_BINARY(ambfp); + if (mybuf == NULL) + mybuf = (char *)bmalloc(BUFSIZ+8); + setbuf(ambfp, mybuf); + if (cre8) { /* new file */ + newheader("RADIANCE", ambfp); + fprintf(ambfp, "%s -av %g %g %g -aw %d -ab %d -aa %g ", + progname, colval(ambval,RED), + colval(ambval,GRN), colval(ambval,BLU), + ambvwt, ambounce, ambacc); + fprintf(ambfp, "-ad %d -as %d -ar %d ", + ambdiv, ambssamp, ambres); + if (octname != NULL) + fputs(octname, ambfp); + fputc('\n', ambfp); + fprintf(ambfp, "SOFTWARE= %s\n", VersionID); + fputnow(ambfp); + fputformat(AMBFMT, ambfp); + fputc('\n', ambfp); + putambmagic(ambfp); + } else if (checkheader(ambfp, AMBFMT, NULL) < 0 || !hasambmagic(ambfp)) + error(USER, "bad ambient file"); +} + + +static void +avsave( /* insert and save an ambient value */ + AMBVAL *av +) +{ + avinsert(avstore(av)); + if (ambfp == NULL) + return; + if (writambval(av, ambfp) < 0) + goto writerr; + if (++nunflshed >= AMBFLUSH) + if (ambsync() == EOF) + goto writerr; + return; +writerr: + error(SYSTEM, "error writing to ambient file"); +} + + +static AMBVAL * +avstore( /* allocate memory and store aval */ + AMBVAL *aval +) +{ + AMBVAL *av; + double d; + + if ((av = newambval()) == NULL) + error(SYSTEM, "out of memory in avstore"); + *av = *aval; + av->latick = ambclock; + av->next = NULL; + nambvals++; + d = bright(av->val); + if (d > FTINY) { /* add to log sum for averaging */ + avsum += log(d); + navsum++; } - /* sort the divisions */ - qsort(div, ne, sizeof(AMBSAMP), ambcmp); - /* skim excess */ - while (ne > ns) { - ne--; - addcolor(acol, div[ne].v); + return(av); +} + + +#define ATALLOCSZ 512 /* #/8 trees to allocate at once */ + +static AMBTREE *atfreelist = NULL; /* free ambient tree structures */ + + +static AMBTREE * +newambtree(void) /* allocate 8 ambient tree structs */ +{ + AMBTREE *atp, *upperlim; + + if (atfreelist == NULL) { /* get more nodes */ + atfreelist = (AMBTREE *)malloc(ATALLOCSZ*8*sizeof(AMBTREE)); + if (atfreelist == NULL) + return(NULL); + /* link new free list */ + upperlim = atfreelist + 8*(ATALLOCSZ-1); + for (atp = atfreelist; atp < upperlim; atp += 8) + atp->kid = atp + 8; + atp->kid = NULL; } - /* super-sample */ - for (i = ns; i > 0; i--) { - rayorigin(&ar, r, AMBIENT, 0.5); /* pretested */ - zd = sqrt((div[0].t+frandom())/nt); - phi = 2.0*PI * (div[0].p+frandom())/np; - xd = cos(phi) * zd; - yd = sin(phi) * zd; - zd = sqrt(1.0 - zd*zd); - for (k = 0; k < 3; k++) - ar.rdir[k] = xd*ux[k]+yd*uy[k]+zd*r->ron[k]; - rayvalue(&ar); - if (ar.rot < FHUGE) - arad += 1.0 / ar.rot; - /* recompute error */ - copycolor(dnew.v, div[0].v); - addcolor(dnew.v, ar.rcol); - dnew.n = div[0].n + 1; - dnew.t = div[0].t; dnew.p = div[0].p; - b2 = bright(dnew.v)/dnew.n - bright(ar.rcol); - b2 = b2*b2 + div[0].k*(div[0].n*div[0].n); - dnew.k = b2/(dnew.n*dnew.n); - /* reinsert */ - for (k = 0; k < ne-1 && dnew.k < div[k+1].k; k++) - copystruct(&div[k], &div[k+1]); - copystruct(&div[k], &dnew); + atp = atfreelist; + atfreelist = atp->kid; + memset((char *)atp, '\0', 8*sizeof(AMBTREE)); + return(atp); +} - if (ne >= i) { /* extract darkest division */ - ne--; - if (div[ne].n > 1) - scalecolor(div[ne].v, 1.0/div[ne].n); - addcolor(acol, div[ne].v); - } + +static void +freeambtree( /* free 8 ambient tree structs */ + AMBTREE *atp +) +{ + atp->kid = atfreelist; + atfreelist = atp; +} + + +static void +unloadatree( /* unload an ambient value tree */ + AMBTREE *at, + unloadtf_t *f +) +{ + AMBVAL *av; + int i; + /* transfer values at this node */ + for (av = at->alist; av != NULL; av = at->alist) { + at->alist = av->next; + (*f)(av); } - scalecolor(acol, 1.0/ndivs); - if (arad <= FTINY) - arad = FHUGE; - else - arad = (ndivs+ns) / arad / sqrt(r->rweight); - if (arad > maxarad) - arad = maxarad; - else if (arad < minarad) - arad = minarad; - if (ns > 0) - free((char *)div); - return(arad); + if (at->kid == NULL) + return; + for (i = 0; i < 8; i++) /* transfer and free children */ + unloadatree(at->kid+i, f); + freeambtree(at->kid); + at->kid = NULL; } +static struct avl { + AMBVAL *p; + unsigned long t; +} *avlist1; /* ambient value list with ticks */ +static AMBVAL **avlist2; /* memory positions for sorting */ +static int i_avlist; /* index for lists */ + +static int alatcmp(const void *av1, const void *av2); + +static void +avfree(AMBVAL *av) +{ + free(av); +} + +static void +av2list( + AMBVAL *av +) +{ +#ifdef DEBUG + if (i_avlist >= nambvals) + error(CONSISTENCY, "too many ambient values in av2list1"); +#endif + avlist1[i_avlist].p = avlist2[i_avlist] = (AMBVAL*)av; + avlist1[i_avlist++].t = av->latick; +} + + static int -ambcmp(d1, d2) /* decreasing order */ -AMBSAMP *d1, *d2; +alatcmp( /* compare ambient values for MRA */ + const void *av1, + const void *av2 +) { - if (d1->k < d2->k) - return(1); - if (d1->k > d2->k) + long lc = ((struct avl *)av2)->t - ((struct avl *)av1)->t; + return(lc<0 ? -1 : lc>0 ? 1 : 0); +} + + +/* GW NOTE 2002/10/3: + * I used to compare AMBVAL pointers, but found that this was the + * cause of a serious consistency error with gcc, since the optimizer + * uses some dangerous trick in pointer subtraction that + * assumes pointers differ by exact struct size increments. + */ +static int +aposcmp( /* compare ambient value positions */ + const void *avp1, + const void *avp2 +) +{ + long diff = *(char * const *)avp1 - *(char * const *)avp2; + if (diff < 0) return(-1); - return(0); + return(diff > 0); } -static -avsave(av) /* save an ambient value */ -AMBVAL *av; +static int +avlmemi( /* find list position from address */ + AMBVAL *avaddr +) { -#ifdef AMBFLUSH - static int nunflshed = 0; -#endif - if (ambfp == NULL) + AMBVAL **avlpp; + + avlpp = (AMBVAL **)bsearch((char *)&avaddr, (char *)avlist2, + nambvals, sizeof(AMBVAL *), &aposcmp); + if (avlpp == NULL) + error(CONSISTENCY, "address not found in avlmemi"); + return(avlpp - avlist2); +} + + +static void +sortambvals( /* resort ambient values */ + int always +) +{ + AMBTREE oldatrunk; + AMBVAL tav, *tap, *pnext; + int i, j; + /* see if it's time yet */ + if (!always && (ambclock++ < lastsort+sortintvl || + nambvals < SORT_THRESH)) return; - if (fwrite((char *)av, sizeof(AMBVAL), 1, ambfp) != 1) - goto writerr; -#ifdef AMBFLUSH - if (++nunflshed >= AMBFLUSH) { - if (fflush(ambfp) == EOF) - goto writerr; - nunflshed = 0; + /* + * The idea here is to minimize memory thrashing + * in VM systems by improving reference locality. + * We do this by periodically sorting our stored ambient + * values in memory in order of most recently to least + * recently accessed. This ordering was chosen so that new + * ambient values (which tend to be less important) go into + * higher memory with the infrequently accessed values. + * Since we expect our values to need sorting less + * frequently as the process continues, we double our + * waiting interval after each call. + * This routine is also called by setambacc() with + * the "always" parameter set to 1 so that the ambient + * tree will be rebuilt with the new accuracy parameter. + */ + if (tracktime) { /* allocate pointer arrays to sort */ + avlist2 = (AMBVAL **)malloc(nambvals*sizeof(AMBVAL *)); + avlist1 = (struct avl *)malloc(nambvals*sizeof(struct avl)); + } else { + avlist2 = NULL; + avlist1 = NULL; } + if (avlist1 == NULL) { /* no time tracking -- rebuild tree? */ + if (avlist2 != NULL) + free((void *)avlist2); + if (always) { /* rebuild without sorting */ + oldatrunk = atrunk; + atrunk.alist = NULL; + atrunk.kid = NULL; + unloadatree(&oldatrunk, &avinsert); + } + } else { /* sort memory by last access time */ + /* + * Sorting memory is tricky because it isn't contiguous. + * We have to sort an array of pointers by MRA and also + * by memory position. We then copy values in "loops" + * to minimize memory hits. Nevertheless, we will visit + * everyone at least twice, and this is an expensive process + * when we're thrashing, which is when we need to do it. + */ +#ifdef DEBUG + sprintf(errmsg, "sorting %u ambient values at ambclock=%lu...", + nambvals, ambclock); + eputs(errmsg); #endif - return; -writerr: - error(SYSTEM, "error writing ambient file"); + i_avlist = 0; + unloadatree(&atrunk, &av2list); /* empty current tree */ +#ifdef DEBUG + if (i_avlist < nambvals) + error(CONSISTENCY, "missing ambient values in sortambvals"); +#endif + qsort((char *)avlist1, nambvals, sizeof(struct avl), &alatcmp); + qsort((char *)avlist2, nambvals, sizeof(AMBVAL *), &aposcmp); + for (i = 0; i < nambvals; i++) { + if (avlist1[i].p == NULL) + continue; + tap = avlist2[i]; + tav = *tap; + for (j = i; (pnext = avlist1[j].p) != tap; + j = avlmemi(pnext)) { + *(avlist2[j]) = *pnext; + avinsert(avlist2[j]); + avlist1[j].p = NULL; + } + *(avlist2[j]) = tav; + avinsert(avlist2[j]); + avlist1[j].p = NULL; + } + free((void *)avlist1); + free((void *)avlist2); + /* compute new sort interval */ + sortintvl = ambclock - lastsort; + if (sortintvl >= MAX_SORT_INTVL/2) + sortintvl = MAX_SORT_INTVL; + else + sortintvl <<= 1; /* wait twice as long next */ +#ifdef DEBUG + eputs("done\n"); +#endif + } + if (ambclock >= MAXACLOCK) + ambclock = MAXACLOCK/2; + lastsort = ambclock; } -static -avinsert(aval, at, c0, s) /* insert ambient value in a tree */ -AMBVAL *aval; -register AMBTREE *at; -FVECT c0; -double s; +#ifdef F_SETLKW + +static void +aflock( /* lock/unlock ambient file */ + int typ +) { - FVECT ck0; - int branch; - register AMBVAL *av; - register int i; + static struct flock fls; /* static so initialized to zeroes */ - if ((av = newambval()) == NULL) - goto memerr; - copystruct(av, aval); - VCOPY(ck0, c0); - while (s*(OCTSCALE/2) > av->rad*ambacc) { - if (at->kid == NULL) - if ((at->kid = newambtree()) == NULL) - goto memerr; - s *= 0.5; - branch = 0; - for (i = 0; i < 3; i++) - if (av->pos[i] > ck0[i] + s) { - ck0[i] += s; - branch |= 1 << i; + if (typ == fls.l_type) /* already called? */ + return; + fls.l_type = typ; + if (fcntl(fileno(ambfp), F_SETLKW, &fls) < 0) + error(SYSTEM, "cannot (un)lock ambient file"); +} + + +int +ambsync(void) /* synchronize ambient file */ +{ + long flen; + AMBVAL avs; + int n; + + if (ambfp == NULL) /* no ambient file? */ + return(0); + /* gain appropriate access */ + aflock(nunflshed ? F_WRLCK : F_RDLCK); + /* see if file has grown */ + if ((flen = lseek(fileno(ambfp), (off_t)0, SEEK_END)) < 0) + goto seekerr; + if ((n = flen - lastpos) > 0) { /* file has grown */ + if (ambinp == NULL) { /* use duplicate filedes */ + ambinp = fdopen(dup(fileno(ambfp)), "r"); + if (ambinp == NULL) + error(SYSTEM, "fdopen failed in ambsync"); + } + if (fseek(ambinp, lastpos, SEEK_SET) < 0) + goto seekerr; + while (n >= AMBVALSIZ) { /* load contributed values */ + if (!readambval(&avs, ambinp)) { + sprintf(errmsg, + "ambient file \"%s\" corrupted near character %ld", + ambfile, flen - n); + error(WARNING, errmsg); + break; } - at = at->kid + branch; + avinsert(avstore(&avs)); + n -= AMBVALSIZ; + } + lastpos = flen - n; + /*** seek always as safety measure + if (n) ***/ /* alignment */ + if (lseek(fileno(ambfp), (off_t)lastpos, SEEK_SET) < 0) + goto seekerr; } - av->next = at->alist; - at->alist = av; - return; -memerr: - error(SYSTEM, "out of memory in avinsert"); + n = fflush(ambfp); /* calls write() at last */ + if (n != EOF) + lastpos += (long)nunflshed*AMBVALSIZ; + else if ((lastpos = lseek(fileno(ambfp), (off_t)0, SEEK_CUR)) < 0) + goto seekerr; + + aflock(F_UNLCK); /* release file */ + nunflshed = 0; + return(n); +seekerr: + error(SYSTEM, "seek failed in ambsync"); + return -1; /* pro forma return */ } + +#else /* ! F_SETLKW */ + +int +ambsync(void) /* flush ambient file */ +{ + if (ambfp == NULL) + return(0); + nunflshed = 0; + return(fflush(ambfp)); +} + +#endif /* ! F_SETLKW */