1 |
#ifndef lint |
2 |
static const char RCSid[] = "$Id: rholo3.c,v 3.36 2003/05/13 17:58:33 greg Exp $"; |
3 |
#endif |
4 |
/* |
5 |
* Routines for tracking beam compuatations |
6 |
*/ |
7 |
|
8 |
#include "rholo.h" |
9 |
#include "view.h" |
10 |
|
11 |
#ifndef NFRAG2CHUNK |
12 |
#define NFRAG2CHUNK 4096 /* number of fragments to start chunking */ |
13 |
#endif |
14 |
|
15 |
#ifndef abs |
16 |
#define abs(x) ((x) > 0 ? (x) : -(x)) |
17 |
#endif |
18 |
#ifndef sgn |
19 |
#define sgn(x) ((x) > 0 ? 1 : (x) < 0 ? -1 : 0) |
20 |
#endif |
21 |
|
22 |
#define rchunk(n) (((n)+(RPACKSIZ/2))/RPACKSIZ) |
23 |
|
24 |
extern time_t time(); |
25 |
|
26 |
int chunkycmp = 0; /* clump beams together on disk */ |
27 |
|
28 |
static PACKHEAD *complist=NULL; /* list of beams to compute */ |
29 |
static int complen=0; /* length of complist */ |
30 |
static int listpos=0; /* current list position for next_packet */ |
31 |
static int lastin= -1; /* last ordered position in list */ |
32 |
|
33 |
|
34 |
int |
35 |
beamcmp(b0, b1) /* comparison for compute order */ |
36 |
register PACKHEAD *b0, *b1; |
37 |
{ |
38 |
BEAMI *bip0, *bip1; |
39 |
register long c; |
40 |
/* first check desired quantities */ |
41 |
if (chunkycmp) |
42 |
c = rchunk(b1->nr)*(rchunk(b0->nc)+1L) - |
43 |
rchunk(b0->nr)*(rchunk(b1->nc)+1L); |
44 |
else |
45 |
c = b1->nr*(b0->nc+1L) - b0->nr*(b1->nc+1L); |
46 |
if (c > 0) return(1); |
47 |
if (c < 0) return(-1); |
48 |
/* only one file, so skip the following: */ |
49 |
#if 0 |
50 |
/* next, check file descriptors */ |
51 |
c = hdlist[b0->hd]->fd - hdlist[b1->hd]->fd; |
52 |
if (c) return(c); |
53 |
#endif |
54 |
/* finally, check file positions */ |
55 |
bip0 = &hdlist[b0->hd]->bi[b0->bi]; |
56 |
bip1 = &hdlist[b1->hd]->bi[b1->bi]; |
57 |
/* put diskless beams last */ |
58 |
if (!bip0->nrd) |
59 |
return(bip1->nrd > 0); |
60 |
if (!bip1->nrd) |
61 |
return(-1); |
62 |
c = bip0->fo - bip1->fo; |
63 |
return(c < 0 ? -1 : c > 0); |
64 |
} |
65 |
|
66 |
|
67 |
int |
68 |
beamidcmp(b0, b1) /* comparison for beam searching */ |
69 |
register PACKHEAD *b0, *b1; |
70 |
{ |
71 |
register int c = b0->hd - b1->hd; |
72 |
|
73 |
if (c) return(c); |
74 |
return(b0->bi - b1->bi); |
75 |
} |
76 |
|
77 |
|
78 |
int |
79 |
dispbeam(b, hb) /* display a holodeck beam */ |
80 |
register BEAM *b; |
81 |
register HDBEAMI *hb; |
82 |
{ |
83 |
static int n = 0; |
84 |
static PACKHEAD *p = NULL; |
85 |
|
86 |
if (b == NULL) |
87 |
return; |
88 |
if (b->nrm > n) { /* (re)allocate packet holder */ |
89 |
n = b->nrm; |
90 |
if (p == NULL) p = (PACKHEAD *)malloc(packsiz(n)); |
91 |
else p = (PACKHEAD *)realloc((void *)p, packsiz(n)); |
92 |
CHECK(p==NULL, SYSTEM, "out of memory in dispbeam"); |
93 |
} |
94 |
/* assign packet fields */ |
95 |
bcopy((void *)hdbray(b), (void *)packra(p), b->nrm*sizeof(RAYVAL)); |
96 |
p->nr = p->nc = b->nrm; |
97 |
for (p->hd = 0; hdlist[p->hd] != hb->h; p->hd++) |
98 |
if (hdlist[p->hd] == NULL) |
99 |
error(CONSISTENCY, "unregistered holodeck in dispbeam"); |
100 |
p->bi = hb->b; |
101 |
disp_packet(p); /* display it */ |
102 |
if (n >= 1024) { /* free ridiculous packets */ |
103 |
free((void *)p); |
104 |
p = NULL; n = 0; |
105 |
} |
106 |
} |
107 |
|
108 |
|
109 |
bundle_set(op, clist, nents) /* bundle set operation */ |
110 |
int op; |
111 |
PACKHEAD *clist; |
112 |
int nents; |
113 |
{ |
114 |
int oldnr, n; |
115 |
HDBEAMI *hbarr; |
116 |
register PACKHEAD *csm; |
117 |
register int i; |
118 |
/* search for common members */ |
119 |
for (csm = clist+nents; csm-- > clist; ) |
120 |
csm->nc = -1; |
121 |
qsort((void *)clist, nents, sizeof(PACKHEAD), beamidcmp); |
122 |
for (i = 0; i < complen; i++) { |
123 |
csm = (PACKHEAD *)bsearch((void *)(complist+i), (void *)clist, |
124 |
nents, sizeof(PACKHEAD), beamidcmp); |
125 |
if (csm == NULL) |
126 |
continue; |
127 |
oldnr = complist[i].nr; |
128 |
csm->nc = complist[i].nc; |
129 |
switch (op) { |
130 |
case BS_ADD: /* add to count */ |
131 |
complist[i].nr += csm->nr; |
132 |
csm->nr = 0; |
133 |
break; |
134 |
case BS_MAX: /* maximum of counts */ |
135 |
if (csm->nr > complist[i].nr) |
136 |
complist[i].nr = csm->nr; |
137 |
csm->nr = 0; |
138 |
break; |
139 |
case BS_ADJ: /* reset count */ |
140 |
complist[i].nr = csm->nr; |
141 |
csm->nr = 0; |
142 |
break; |
143 |
case BS_DEL: /* delete count */ |
144 |
if (csm->nr == 0 || csm->nr >= complist[i].nr) |
145 |
complist[i].nr = 0; |
146 |
else |
147 |
complist[i].nr -= csm->nr; |
148 |
break; |
149 |
} |
150 |
if (complist[i].nr != oldnr) |
151 |
lastin = -1; /* flag sort */ |
152 |
} |
153 |
/* record computed rays for uncommon beams */ |
154 |
for (csm = clist+nents; csm-- > clist; ) |
155 |
if (csm->nc < 0) |
156 |
csm->nc = bnrays(hdlist[csm->hd], csm->bi); |
157 |
/* complete list operations */ |
158 |
switch (op) { |
159 |
case BS_NEW: /* new computation set */ |
160 |
listpos = 0; lastin = -1; |
161 |
if (complen) /* free old list */ |
162 |
free((void *)complist); |
163 |
complist = NULL; |
164 |
if (!(complen = nents)) |
165 |
return; |
166 |
complist = (PACKHEAD *)malloc(nents*sizeof(PACKHEAD)); |
167 |
if (complist == NULL) |
168 |
goto memerr; |
169 |
bcopy((void *)clist, (void *)complist, nents*sizeof(PACKHEAD)); |
170 |
break; |
171 |
case BS_ADD: /* add to computation set */ |
172 |
case BS_MAX: /* maximum of quantities */ |
173 |
case BS_ADJ: /* adjust set quantities */ |
174 |
if (nents <= 0) |
175 |
return; |
176 |
sortcomplist(); /* sort updated list & new entries */ |
177 |
qsort((void *)clist, nents, sizeof(PACKHEAD), beamcmp); |
178 |
/* what can't we satisfy? */ |
179 |
for (i = nents, csm = clist; i-- && csm->nr > csm->nc; csm++) |
180 |
; |
181 |
n = csm - clist; |
182 |
if (op != BS_ADD) { /* don't regenerate adjusted beams */ |
183 |
for (++i; i-- && csm->nr > 0; csm++) |
184 |
; |
185 |
nents = csm - clist; |
186 |
} |
187 |
if (n) { /* allocate space for merged list */ |
188 |
PACKHEAD *newlist; |
189 |
newlist = (PACKHEAD *)malloc( |
190 |
(complen+n)*sizeof(PACKHEAD) ); |
191 |
if (newlist == NULL) |
192 |
goto memerr; |
193 |
/* merge lists */ |
194 |
mergeclists(newlist, clist, n, complist, complen); |
195 |
if (complen) |
196 |
free((void *)complist); |
197 |
complist = newlist; |
198 |
complen += n; |
199 |
} |
200 |
listpos = 0; |
201 |
lastin = complen-1; /* list is now sorted */ |
202 |
break; |
203 |
case BS_DEL: /* delete from computation set */ |
204 |
return; /* already done */ |
205 |
default: |
206 |
error(CONSISTENCY, "bundle_set called with unknown operation"); |
207 |
} |
208 |
if (outdev == NULL || !nents) /* nothing to display? */ |
209 |
return; |
210 |
/* load and display beams we have */ |
211 |
hbarr = (HDBEAMI *)malloc(nents*sizeof(HDBEAMI)); |
212 |
for (i = nents; i--; ) { |
213 |
hbarr[i].h = hdlist[clist[i].hd]; |
214 |
hbarr[i].b = clist[i].bi; |
215 |
} |
216 |
hdloadbeams(hbarr, nents, dispbeam); |
217 |
free((void *)hbarr); |
218 |
if (hdfragflags&FF_READ) { |
219 |
listpos = 0; |
220 |
lastin = -1; /* need to re-sort list */ |
221 |
} |
222 |
return; |
223 |
memerr: |
224 |
error(SYSTEM, "out of memory in bundle_set"); |
225 |
} |
226 |
|
227 |
|
228 |
double |
229 |
beamvolume(hp, bi) /* compute approximate volume of a beam */ |
230 |
HOLO *hp; |
231 |
int bi; |
232 |
{ |
233 |
GCOORD gc[2]; |
234 |
FVECT cp[4], edgeA, edgeB, cent[2]; |
235 |
FVECT v, crossp[2], diffv; |
236 |
double vol[2]; |
237 |
register int i; |
238 |
/* get grid coordinates */ |
239 |
if (!hdbcoord(gc, hp, bi)) |
240 |
error(CONSISTENCY, "bad beam index in beamvolume"); |
241 |
for (i = 0; i < 2; i++) { /* compute cell area vectors */ |
242 |
hdcell(cp, hp, gc+i); |
243 |
VSUM(edgeA, cp[1], cp[0], -1.0); |
244 |
VSUM(edgeB, cp[3], cp[1], -1.0); |
245 |
fcross(crossp[i], edgeA, edgeB); |
246 |
/* compute center */ |
247 |
cent[i][0] = 0.5*(cp[0][0] + cp[2][0]); |
248 |
cent[i][1] = 0.5*(cp[0][1] + cp[2][1]); |
249 |
cent[i][2] = 0.5*(cp[0][2] + cp[2][2]); |
250 |
} |
251 |
/* compute difference vector */ |
252 |
VSUM(diffv, cent[1], cent[0], -1.0); |
253 |
for (i = 0; i < 2; i++) { /* compute volume contributions */ |
254 |
vol[i] = 0.5*DOT(crossp[i], diffv); |
255 |
if (vol[i] < 0.) vol[i] = -vol[i]; |
256 |
} |
257 |
return(vol[0] + vol[1]); /* return total volume */ |
258 |
} |
259 |
|
260 |
|
261 |
ambient_list() /* compute ambient beam list */ |
262 |
{ |
263 |
int32 wtotal, minrt; |
264 |
double frac; |
265 |
int i; |
266 |
register int j, k; |
267 |
|
268 |
complen = 0; |
269 |
for (j = 0; hdlist[j] != NULL; j++) |
270 |
complen += nbeams(hdlist[j]); |
271 |
complist = (PACKHEAD *)malloc(complen*sizeof(PACKHEAD)); |
272 |
CHECK(complist==NULL, SYSTEM, "out of memory in ambient_list"); |
273 |
/* compute beam weights */ |
274 |
k = 0; wtotal = 0; |
275 |
for (j = 0; hdlist[j] != NULL; j++) { |
276 |
frac = 512. * VLEN(hdlist[j]->wg[0]) * |
277 |
VLEN(hdlist[j]->wg[1]) * |
278 |
VLEN(hdlist[j]->wg[2]); |
279 |
for (i = nbeams(hdlist[j]); i > 0; i--) { |
280 |
complist[k].hd = j; |
281 |
complist[k].bi = i; |
282 |
complist[k].nr = frac*beamvolume(hdlist[j], i) + 0.5; |
283 |
complist[k].nc = bnrays(hdlist[j], i); |
284 |
wtotal += complist[k++].nr; |
285 |
} |
286 |
} |
287 |
/* adjust sample weights */ |
288 |
if (vdef(DISKSPACE)) |
289 |
frac = 1024.*1024.*vflt(DISKSPACE) / (wtotal*sizeof(RAYVAL)); |
290 |
else |
291 |
frac = 1024.*1024.*2048. / (wtotal*sizeof(RAYVAL)); |
292 |
minrt = .02*frac*wtotal/complen + .5; /* heuristic mimimum */ |
293 |
if (minrt > RPACKSIZ) |
294 |
minrt = RPACKSIZ; |
295 |
for (k = complen; k--; ) |
296 |
if ((complist[k].nr = frac*complist[k].nr + 0.5) < minrt) |
297 |
complist[k].nr = minrt; |
298 |
listpos = 0; lastin = -1; /* flag initial sort */ |
299 |
} |
300 |
|
301 |
|
302 |
view_list(fp) /* assign beam priority from view list */ |
303 |
FILE *fp; |
304 |
{ |
305 |
double pa = 1.; |
306 |
VIEW curview; |
307 |
int xr, yr; |
308 |
char *err; |
309 |
BEAMLIST blist; |
310 |
|
311 |
copystruct(&curview, &stdview); |
312 |
while (nextview(&curview, fp) != EOF) { |
313 |
if ((err = setview(&curview)) != NULL) { |
314 |
error(WARNING, err); |
315 |
continue; |
316 |
} |
317 |
xr = yr = 1024; |
318 |
normaspect(viewaspect(&curview), &pa, &xr, &yr); |
319 |
viewbeams(&curview, xr, yr, &blist); |
320 |
bundle_set(BS_MAX, blist.bl, blist.nb); |
321 |
free((void *)blist.bl); |
322 |
} |
323 |
} |
324 |
|
325 |
|
326 |
init_global() /* initialize global ray computation */ |
327 |
{ |
328 |
register int k; |
329 |
/* free old list and empty queue */ |
330 |
if (complen > 0) { |
331 |
free((void *)complist); |
332 |
done_packets(flush_queue()); |
333 |
} |
334 |
/* reseed random number generator */ |
335 |
srandom(time(NULL)); |
336 |
/* allocate beam list */ |
337 |
if (readinp) |
338 |
view_list(stdin); |
339 |
else |
340 |
ambient_list(); |
341 |
/* no view vicinity */ |
342 |
myeye.rng = 0; |
343 |
} |
344 |
|
345 |
|
346 |
mergeclists(cdest, cl1, n1, cl2, n2) /* merge two sorted lists */ |
347 |
register PACKHEAD *cdest; |
348 |
register PACKHEAD *cl1, *cl2; |
349 |
int n1, n2; |
350 |
{ |
351 |
register int cmp; |
352 |
|
353 |
while (n1 | n2) { |
354 |
if (!n1) cmp = 1; |
355 |
else if (!n2) cmp = -1; |
356 |
else cmp = beamcmp(cl1, cl2); |
357 |
if (cmp > 0) { |
358 |
copystruct(cdest, cl2); |
359 |
cl2++; n2--; |
360 |
} else { |
361 |
copystruct(cdest, cl1); |
362 |
cl1++; n1--; |
363 |
} |
364 |
cdest++; |
365 |
} |
366 |
} |
367 |
|
368 |
|
369 |
sortcomplist() /* fix our list order */ |
370 |
{ |
371 |
PACKHEAD *list2; |
372 |
int listlen; |
373 |
register int i; |
374 |
|
375 |
if (complen <= 0) /* check to see if there is even a list */ |
376 |
return; |
377 |
if (!chunkycmp) /* check to see if fragment list is full */ |
378 |
if (!hdfragOK(hdlist[0]->fd, &listlen, NULL) |
379 |
#if NFRAG2CHUNK |
380 |
|| listlen >= NFRAG2CHUNK |
381 |
#endif |
382 |
) { |
383 |
chunkycmp++; /* use "chunky" comparison */ |
384 |
lastin = -1; /* need to re-sort list */ |
385 |
#ifdef DEBUG |
386 |
error(WARNING, "using chunky comparison mode"); |
387 |
#endif |
388 |
} |
389 |
if (lastin < 0 || listpos*4 >= complen*3) |
390 |
qsort((void *)complist, complen, sizeof(PACKHEAD), beamcmp); |
391 |
else if (listpos) { /* else sort and merge sublist */ |
392 |
list2 = (PACKHEAD *)malloc(listpos*sizeof(PACKHEAD)); |
393 |
CHECK(list2==NULL, SYSTEM, "out of memory in sortcomplist"); |
394 |
bcopy((void *)complist,(void *)list2,listpos*sizeof(PACKHEAD)); |
395 |
qsort((void *)list2, listpos, sizeof(PACKHEAD), beamcmp); |
396 |
mergeclists(complist, list2, listpos, |
397 |
complist+listpos, complen-listpos); |
398 |
free((void *)list2); |
399 |
} |
400 |
/* drop satisfied requests */ |
401 |
for (i = complen; i-- && complist[i].nr <= complist[i].nc; ) |
402 |
; |
403 |
if (i < 0) { |
404 |
free((void *)complist); |
405 |
complist = NULL; |
406 |
complen = 0; |
407 |
} else if (i < complen-1) { |
408 |
list2 = (PACKHEAD *)realloc((void *)complist, |
409 |
(i+1)*sizeof(PACKHEAD)); |
410 |
if (list2 != NULL) |
411 |
complist = list2; |
412 |
complen = i+1; |
413 |
} |
414 |
listpos = 0; lastin = i; |
415 |
} |
416 |
|
417 |
|
418 |
/* |
419 |
* The following routine works on the assumption that the bundle weights are |
420 |
* more or less evenly distributed, such that computing a packet causes |
421 |
* a given bundle to move way down in the computation order. We keep |
422 |
* track of where the computed bundle with the highest priority would end |
423 |
* up, and if we get further in our compute list than this, we re-sort the |
424 |
* list and start again from the beginning. Since |
425 |
* a merge sort is used, the sorting costs are minimal. |
426 |
*/ |
427 |
next_packet(p, n) /* prepare packet for computation */ |
428 |
register PACKET *p; |
429 |
int n; |
430 |
{ |
431 |
register int i; |
432 |
|
433 |
if (listpos > lastin) /* time to sort the list */ |
434 |
sortcomplist(); |
435 |
if (complen <= 0) |
436 |
return(0); |
437 |
p->hd = complist[listpos].hd; |
438 |
p->bi = complist[listpos].bi; |
439 |
p->nc = complist[listpos].nc; |
440 |
p->nr = complist[listpos].nr - p->nc; |
441 |
if (p->nr <= 0) |
442 |
return(0); |
443 |
DCHECK(n < 1 | n > RPACKSIZ, |
444 |
CONSISTENCY, "next_packet called with bad n value"); |
445 |
if (p->nr > n) |
446 |
p->nr = n; |
447 |
complist[listpos].nc += p->nr; /* find where this one would go */ |
448 |
if (hdgetbeam(hdlist[p->hd], p->bi) != NULL) |
449 |
hdfreefrag(hdlist[p->hd], p->bi); |
450 |
while (lastin > listpos && |
451 |
beamcmp(complist+lastin, complist+listpos) > 0) |
452 |
lastin--; |
453 |
listpos++; |
454 |
return(1); |
455 |
} |