| 1 |
/* Copyright (c) 1997 Silicon Graphics, Inc. */ |
| 2 |
|
| 3 |
#ifndef lint |
| 4 |
static char SCCSid[] = "$SunId$ SGI"; |
| 5 |
#endif |
| 6 |
|
| 7 |
/* |
| 8 |
* Routines for tracking beam compuatations |
| 9 |
*/ |
| 10 |
|
| 11 |
#include "rholo.h" |
| 12 |
|
| 13 |
#define abs(x) ((x) > 0 ? (x) : -(x)) |
| 14 |
#define sgn(x) ((x) > 0 ? 1 : (x) < 0 ? -1 : 0) |
| 15 |
|
| 16 |
static PACKHEAD *complist=NULL; /* list of beams to compute */ |
| 17 |
static int complen=0; /* length of complist */ |
| 18 |
static int listpos=0; /* current list position for next_packet */ |
| 19 |
static int lastin= -1; /* last ordered position in list */ |
| 20 |
|
| 21 |
|
| 22 |
int |
| 23 |
beamcmp(b0, b1) /* comparison for descending compute order */ |
| 24 |
register PACKHEAD *b0, *b1; |
| 25 |
{ |
| 26 |
return( b1->nr*(b0->nc+1) - b0->nr*(b1->nc+1) ); |
| 27 |
} |
| 28 |
|
| 29 |
|
| 30 |
bundle_set(op, clist, nents) /* bundle set operation */ |
| 31 |
int op; |
| 32 |
PACKHEAD *clist; |
| 33 |
int nents; |
| 34 |
{ |
| 35 |
BEAM *b; |
| 36 |
PACKHEAD *p; |
| 37 |
register int i, n; |
| 38 |
|
| 39 |
switch (op) { |
| 40 |
case BS_NEW: /* new computation set */ |
| 41 |
if (complen) |
| 42 |
free((char *)complist); |
| 43 |
if (nents <= 0) { |
| 44 |
complist = NULL; |
| 45 |
listpos = complen = 0; |
| 46 |
lastin = -1; |
| 47 |
return; |
| 48 |
} |
| 49 |
complist = (PACKHEAD *)malloc(nents*sizeof(PACKHEAD)); |
| 50 |
if (complist == NULL) |
| 51 |
goto memerr; |
| 52 |
bcopy((char *)clist, (char *)complist, nents*sizeof(PACKHEAD)); |
| 53 |
complen = nents; /* finish initialization below */ |
| 54 |
break; |
| 55 |
case BS_ADD: /* add to computation set */ |
| 56 |
case BS_ADJ: /* adjust set quantities */ |
| 57 |
if (nents <= 0) |
| 58 |
return; |
| 59 |
/* merge any common members */ |
| 60 |
for (n = 0; n < nents; n++) { |
| 61 |
for (i = 0; i < complen; i++) |
| 62 |
if (clist[n].bi == complist[i].bi && |
| 63 |
clist[n].hd == complist[i].hd) { |
| 64 |
int oldnr = complist[i].nr; |
| 65 |
if (op == BS_ADD) |
| 66 |
complist[i].nr += clist[n].nr; |
| 67 |
else /* op == BS_ADJ */ |
| 68 |
complist[i].nr = clist[n].nr; |
| 69 |
clist[n].nr = 0; |
| 70 |
clist[n].nc = complist[i].nc; |
| 71 |
if (complist[i].nr != oldnr) |
| 72 |
lastin = -1; /* flag sort */ |
| 73 |
break; |
| 74 |
} |
| 75 |
if (i >= complen) |
| 76 |
clist[n].nc = bnrays(hdlist[clist[n].hd], |
| 77 |
clist[n].bi); |
| 78 |
} |
| 79 |
/* sort updated list */ |
| 80 |
sortcomplist(); |
| 81 |
/* sort new entries */ |
| 82 |
qsort((char *)clist, nents, sizeof(PACKHEAD), beamcmp); |
| 83 |
/* what can't we satisfy? */ |
| 84 |
for (n = 0; n < nents && clist[n].nr > clist[n].nc; n++) |
| 85 |
; |
| 86 |
if (op == BS_ADJ) |
| 87 |
nents = n; |
| 88 |
if (n) { /* allocate space for merged list */ |
| 89 |
PACKHEAD *newlist; |
| 90 |
newlist = (PACKHEAD *)malloc( |
| 91 |
(complen+n)*sizeof(PACKHEAD) ); |
| 92 |
if (newlist == NULL) |
| 93 |
goto memerr; |
| 94 |
/* merge lists */ |
| 95 |
mergeclists(newlist, clist, n, complist, complen); |
| 96 |
if (complen) |
| 97 |
free((char *)complist); |
| 98 |
complist = newlist; |
| 99 |
complen += n; |
| 100 |
} |
| 101 |
listpos = 0; |
| 102 |
lastin = complen-1; /* list is now sorted */ |
| 103 |
break; |
| 104 |
case BS_DEL: /* delete from computation set */ |
| 105 |
if (nents <= 0) |
| 106 |
return; |
| 107 |
/* find each member */ |
| 108 |
for (i = 0; i < complen; i++) |
| 109 |
for (n = 0; n < nents; n++) |
| 110 |
if (clist[n].bi == complist[i].bi && |
| 111 |
clist[n].hd == complist[i].hd) { |
| 112 |
if (clist[n].nr == 0 || |
| 113 |
clist[n].nr >= complist[i].nr) |
| 114 |
complist[i].nr = 0; |
| 115 |
else |
| 116 |
complist[i].nr -= clist[n].nr; |
| 117 |
lastin = -1; /* flag full sort */ |
| 118 |
break; |
| 119 |
} |
| 120 |
return; /* no display */ |
| 121 |
default: |
| 122 |
error(CONSISTENCY, "bundle_set called with unknown operation"); |
| 123 |
} |
| 124 |
if (outdev == NULL) |
| 125 |
return; |
| 126 |
n = 32*RPACKSIZ; /* allocate packet holder */ |
| 127 |
p = (PACKHEAD *)malloc(packsiz(n)); |
| 128 |
if (p == NULL) |
| 129 |
goto memerr; |
| 130 |
/* display what we have */ |
| 131 |
for (i = 0; i < nents; i++) |
| 132 |
if ((b = hdgetbeam(hdlist[clist[i].hd], clist[i].bi)) != NULL) { |
| 133 |
if (b->nrm > n) { |
| 134 |
n = b->nrm; |
| 135 |
p = (PACKHEAD *)realloc((char *)p, packsiz(n)); |
| 136 |
if (p == NULL) |
| 137 |
goto memerr; |
| 138 |
} |
| 139 |
bcopy((char *)hdbray(b), (char *)packra(p), |
| 140 |
b->nrm*sizeof(RAYVAL)); |
| 141 |
p->hd = clist[i].hd; |
| 142 |
p->bi = clist[i].bi; |
| 143 |
p->nr = p->nc = b->nrm; |
| 144 |
disp_packet(p); |
| 145 |
} |
| 146 |
free((char *)p); /* clean up */ |
| 147 |
if (op == BS_NEW) { |
| 148 |
done_packets(flush_queue()); /* empty queue, so we can... */ |
| 149 |
for (i = 0; i < complen; i++) /* ...get number computed */ |
| 150 |
complist[i].nc = bnrays(hdlist[complist[i].hd], |
| 151 |
complist[i].bi); |
| 152 |
listpos = 0; |
| 153 |
lastin = -1; /* flag for initial sort */ |
| 154 |
} |
| 155 |
return; |
| 156 |
memerr: |
| 157 |
error(SYSTEM, "out of memory in bundle_set"); |
| 158 |
} |
| 159 |
|
| 160 |
|
| 161 |
double |
| 162 |
beamvolume(hp, bi) /* compute approximate volume of a beam */ |
| 163 |
HOLO *hp; |
| 164 |
int bi; |
| 165 |
{ |
| 166 |
GCOORD gc[2]; |
| 167 |
FVECT cp[4], edgeA, edgeB, cent[2]; |
| 168 |
FVECT v, crossp[2], diffv; |
| 169 |
double vol[2]; |
| 170 |
register int i; |
| 171 |
/* get grid coordinates */ |
| 172 |
if (!hdbcoord(gc, hp, bi)) |
| 173 |
error(CONSISTENCY, "bad beam index in beamvolume"); |
| 174 |
for (i = 0; i < 2; i++) { /* compute cell area vectors */ |
| 175 |
hdcell(cp, hp, gc+i); |
| 176 |
VSUM(edgeA, cp[1], cp[0], -1.0); |
| 177 |
VSUM(edgeB, cp[3], cp[1], -1.0); |
| 178 |
fcross(crossp[i], edgeA, edgeB); |
| 179 |
VSUM(edgeA, cp[2], cp[3], -1.0); |
| 180 |
VSUM(edgeB, cp[0], cp[2], -1.0); |
| 181 |
fcross(v, edgeA, edgeB); |
| 182 |
VSUM(crossp[i], crossp[i], v, 1.0); |
| 183 |
/* compute center */ |
| 184 |
cent[i][0] = 0.5*(cp[0][0] + cp[2][0]); |
| 185 |
cent[i][1] = 0.5*(cp[0][1] + cp[2][1]); |
| 186 |
cent[i][2] = 0.5*(cp[0][2] + cp[2][2]); |
| 187 |
} |
| 188 |
/* compute difference vector */ |
| 189 |
VSUM(diffv, cent[1], cent[0], -1.0); |
| 190 |
for (i = 0; i < 2; i++) { /* compute volume contributions */ |
| 191 |
vol[i] = 0.25*DOT(crossp[i], diffv); |
| 192 |
if (vol[i] < 0.) vol[i] = -vol[i]; |
| 193 |
} |
| 194 |
return(vol[0] + vol[1]); /* return total volume */ |
| 195 |
} |
| 196 |
|
| 197 |
|
| 198 |
init_global() /* initialize global ray computation */ |
| 199 |
{ |
| 200 |
long wtotal = 0; |
| 201 |
double frac; |
| 202 |
int i; |
| 203 |
register int j, k; |
| 204 |
/* free old list */ |
| 205 |
if (complen > 0) |
| 206 |
free((char *)complist); |
| 207 |
/* allocate beam list */ |
| 208 |
complen = 0; |
| 209 |
for (j = 0; hdlist[j] != NULL; j++) |
| 210 |
complen += nbeams(hdlist[j]); |
| 211 |
complist = (PACKHEAD *)malloc(complen*sizeof(PACKHEAD)); |
| 212 |
if (complist == NULL) |
| 213 |
error(SYSTEM, "out of memory in init_global"); |
| 214 |
/* compute beam weights */ |
| 215 |
k = 0; |
| 216 |
for (j = 0; hdlist[j] != NULL; j++) { |
| 217 |
frac = 512. * hdlist[j]->wg[0] * |
| 218 |
hdlist[j]->wg[1] * hdlist[j]->wg[2]; |
| 219 |
for (i = nbeams(hdlist[j]); i > 0; i--) { |
| 220 |
complist[k].hd = j; |
| 221 |
complist[k].bi = i; |
| 222 |
complist[k].nr = frac*beamvolume(hdlist[j], i) + 0.5; |
| 223 |
wtotal += complist[k++].nr; |
| 224 |
} |
| 225 |
} |
| 226 |
/* adjust weights */ |
| 227 |
if (vdef(DISKSPACE)) |
| 228 |
frac = 1024.*1024.*vflt(DISKSPACE) / (wtotal*sizeof(RAYVAL)); |
| 229 |
else |
| 230 |
frac = 1024.*1024.*16384. / (wtotal*sizeof(RAYVAL)); |
| 231 |
while (k--) |
| 232 |
complist[k].nr = frac * complist[k].nr; |
| 233 |
listpos = 0; lastin = -1; /* flag initial sort */ |
| 234 |
} |
| 235 |
|
| 236 |
|
| 237 |
mergeclists(cdest, cl1, n1, cl2, n2) /* merge two sorted lists */ |
| 238 |
PACKHEAD *cdest; |
| 239 |
PACKHEAD *cl1, *cl2; |
| 240 |
int n1, n2; |
| 241 |
{ |
| 242 |
int cmp; |
| 243 |
|
| 244 |
while (n1 | n2) { |
| 245 |
if (!n1) cmp = 1; |
| 246 |
else if (!n2) cmp = -1; |
| 247 |
else cmp = beamcmp(cl1, cl2); |
| 248 |
if (cmp > 0) { |
| 249 |
copystruct(cdest, cl2); |
| 250 |
cl2++; n2--; |
| 251 |
} else { |
| 252 |
copystruct(cdest, cl1); |
| 253 |
cl1++; n1--; |
| 254 |
} |
| 255 |
cdest++; |
| 256 |
} |
| 257 |
} |
| 258 |
|
| 259 |
|
| 260 |
sortcomplist() /* fix our list order */ |
| 261 |
{ |
| 262 |
PACKHEAD *list2; |
| 263 |
register int i; |
| 264 |
|
| 265 |
if (complen <= 0) /* check to see if there is even a list */ |
| 266 |
return; |
| 267 |
if (lastin < 0 || listpos*4 >= complen*3) |
| 268 |
qsort((char *)complist, complen, sizeof(PACKHEAD), beamcmp); |
| 269 |
else if (listpos) { /* else sort and merge sublist */ |
| 270 |
list2 = (PACKHEAD *)malloc(listpos*sizeof(PACKHEAD)); |
| 271 |
if (list2 == NULL) |
| 272 |
error(SYSTEM, "out of memory in sortcomplist"); |
| 273 |
bcopy((char *)complist,(char *)list2,listpos*sizeof(PACKHEAD)); |
| 274 |
qsort((char *)list2, listpos, sizeof(PACKHEAD), beamcmp); |
| 275 |
mergeclists(complist, list2, listpos, |
| 276 |
complist+listpos, complen-listpos); |
| 277 |
free((char *)list2); |
| 278 |
} |
| 279 |
/* drop satisfied requests */ |
| 280 |
for (i = complen; i-- && complist[i].nr <= complist[i].nc; ) |
| 281 |
; |
| 282 |
if (i < 0) { |
| 283 |
free((char *)complist); |
| 284 |
complist = NULL; |
| 285 |
complen = 0; |
| 286 |
} else if (i < complen-1) { |
| 287 |
list2 = (PACKHEAD *)realloc((char *)complist, |
| 288 |
(i+1)*sizeof(PACKHEAD)); |
| 289 |
if (list2 != NULL) { |
| 290 |
complist = list2; |
| 291 |
complen = i+1; |
| 292 |
} |
| 293 |
} |
| 294 |
listpos = 0; lastin = i; |
| 295 |
} |
| 296 |
|
| 297 |
|
| 298 |
/* |
| 299 |
* The following routine works on the assumption that the bundle weights are |
| 300 |
* more or less evenly distributed, such that computing a packet causes |
| 301 |
* a given bundle to move way down in the computation order. We keep |
| 302 |
* track of where the computed bundle with the highest priority would end |
| 303 |
* up, and if we get further in our compute list than this, we resort the |
| 304 |
* list and start again from the beginning. Since |
| 305 |
* a merge sort is used, the sorting costs are minimal. |
| 306 |
*/ |
| 307 |
next_packet(p) /* prepare packet for computation */ |
| 308 |
register PACKET *p; |
| 309 |
{ |
| 310 |
register int i; |
| 311 |
|
| 312 |
if (listpos > lastin) /* time to sort the list */ |
| 313 |
sortcomplist(); |
| 314 |
if (complen <= 0) |
| 315 |
return(0); |
| 316 |
p->hd = complist[listpos].hd; |
| 317 |
p->bi = complist[listpos].bi; |
| 318 |
p->nc = complist[listpos].nc; |
| 319 |
p->nr = complist[listpos].nr - p->nc; |
| 320 |
if (p->nr <= 0) |
| 321 |
return(0); |
| 322 |
if (p->nr > RPACKSIZ) |
| 323 |
p->nr = RPACKSIZ; |
| 324 |
complist[listpos].nc += p->nr; /* find where this one would go */ |
| 325 |
while (lastin > listpos && |
| 326 |
beamcmp(complist+lastin, complist+listpos) > 0) |
| 327 |
lastin--; |
| 328 |
listpos++; |
| 329 |
return(1); |
| 330 |
} |
