1 |
< |
/* Copyright (c) 1997 Silicon Graphics, Inc. */ |
1 |
> |
/* Copyright (c) 1998 Silicon Graphics, Inc. */ |
2 |
|
|
3 |
|
#ifndef lint |
4 |
|
static char SCCSid[] = "$SunId$ SGI"; |
9 |
|
*/ |
10 |
|
|
11 |
|
#include "rholo.h" |
12 |
+ |
#include <sys/types.h> |
13 |
|
|
14 |
+ |
#ifndef NFRAG2CHUNK |
15 |
+ |
#define NFRAG2CHUNK 4096 /* number of fragments to start chunking */ |
16 |
+ |
#endif |
17 |
+ |
|
18 |
+ |
#ifndef abs |
19 |
|
#define abs(x) ((x) > 0 ? (x) : -(x)) |
20 |
+ |
#endif |
21 |
+ |
#ifndef sgn |
22 |
|
#define sgn(x) ((x) > 0 ? 1 : (x) < 0 ? -1 : 0) |
23 |
+ |
#endif |
24 |
|
|
25 |
+ |
#define rchunk(n) (((n)+(RPACKSIZ/2))/RPACKSIZ) |
26 |
+ |
|
27 |
+ |
extern time_t time(); |
28 |
+ |
|
29 |
+ |
int chunkycmp = 0; /* clump beams together on disk */ |
30 |
+ |
|
31 |
|
static PACKHEAD *complist=NULL; /* list of beams to compute */ |
32 |
|
static int complen=0; /* length of complist */ |
33 |
|
static int listpos=0; /* current list position for next_packet */ |
35 |
|
|
36 |
|
|
37 |
|
int |
38 |
< |
beamcmp(b0, b1) /* comparison for descending compute order */ |
38 |
> |
beamcmp(b0, b1) /* comparison for compute order */ |
39 |
|
register PACKHEAD *b0, *b1; |
40 |
|
{ |
41 |
< |
return( b1->nr*(b0->nc+1) - b0->nr*(b1->nc+1) ); |
41 |
> |
BEAMI *bip0, *bip1; |
42 |
> |
register long c; |
43 |
> |
/* first check desired quantities */ |
44 |
> |
if (chunkycmp) |
45 |
> |
c = rchunk(b1->nr)*(rchunk(b0->nc)+1L) - |
46 |
> |
rchunk(b0->nr)*(rchunk(b1->nc)+1L); |
47 |
> |
else |
48 |
> |
c = b1->nr*(b0->nc+1L) - b0->nr*(b1->nc+1L); |
49 |
> |
if (c > 0) return(1); |
50 |
> |
if (c < 0) return(-1); |
51 |
> |
/* only one file, so skip the following: */ |
52 |
> |
#if 0 |
53 |
> |
/* next, check file descriptors */ |
54 |
> |
c = hdlist[b0->hd]->fd - hdlist[b1->hd]->fd; |
55 |
> |
if (c) return(c); |
56 |
> |
#endif |
57 |
> |
/* finally, check file positions */ |
58 |
> |
bip0 = &hdlist[b0->hd]->bi[b0->bi]; |
59 |
> |
bip1 = &hdlist[b1->hd]->bi[b1->bi]; |
60 |
> |
/* put diskless beams last */ |
61 |
> |
if (!bip0->nrd) |
62 |
> |
return(bip1->nrd > 0); |
63 |
> |
if (!bip1->nrd) |
64 |
> |
return(-1); |
65 |
> |
c = bip0->fo - bip1->fo; |
66 |
> |
return(c < 0 ? -1 : c > 0); |
67 |
|
} |
68 |
|
|
69 |
|
|
70 |
|
int |
71 |
< |
dispbeam(b, hp, bi) /* display a holodeck beam */ |
71 |
> |
beamidcmp(b0, b1) /* comparison for beam searching */ |
72 |
> |
register PACKHEAD *b0, *b1; |
73 |
> |
{ |
74 |
> |
register int c = b0->hd - b1->hd; |
75 |
> |
|
76 |
> |
if (c) return(c); |
77 |
> |
return(b0->bi - b1->bi); |
78 |
> |
} |
79 |
> |
|
80 |
> |
|
81 |
> |
int |
82 |
> |
dispbeam(b, hb) /* display a holodeck beam */ |
83 |
|
register BEAM *b; |
84 |
< |
HOLO *hp; |
34 |
< |
int bi; |
84 |
> |
register HDBEAMI *hb; |
85 |
|
{ |
86 |
|
static int n = 0; |
87 |
|
static PACKHEAD *p = NULL; |
98 |
|
/* assign packet fields */ |
99 |
|
bcopy((char *)hdbray(b), (char *)packra(p), b->nrm*sizeof(RAYVAL)); |
100 |
|
p->nr = p->nc = b->nrm; |
101 |
< |
for (p->hd = 0; hdlist[p->hd] != hp; p->hd++) |
101 |
> |
for (p->hd = 0; hdlist[p->hd] != hb->h; p->hd++) |
102 |
|
if (hdlist[p->hd] == NULL) |
103 |
|
error(CONSISTENCY, "unregistered holodeck in dispbeam"); |
104 |
< |
p->bi = bi; |
104 |
> |
p->bi = hb->b; |
105 |
|
disp_packet(p); /* display it */ |
106 |
+ |
if (n >= 1024) { /* free ridiculous packets */ |
107 |
+ |
free((char *)p); |
108 |
+ |
p = NULL; n = 0; |
109 |
+ |
} |
110 |
|
} |
111 |
|
|
112 |
|
|
113 |
|
bundle_set(op, clist, nents) /* bundle set operation */ |
114 |
|
int op; |
115 |
< |
register PACKHEAD *clist; |
115 |
> |
PACKHEAD *clist; |
116 |
|
int nents; |
117 |
|
{ |
118 |
< |
register int i, n; |
119 |
< |
|
118 |
> |
int oldnr, n; |
119 |
> |
HDBEAMI *hbarr; |
120 |
> |
register PACKHEAD *csm; |
121 |
> |
register int i; |
122 |
> |
/* search for common members */ |
123 |
> |
for (csm = clist+nents; csm-- > clist; ) |
124 |
> |
csm->nc = -1; |
125 |
> |
qsort((char *)clist, nents, sizeof(PACKHEAD), beamidcmp); |
126 |
> |
for (i = 0; i < complen; i++) { |
127 |
> |
csm = (PACKHEAD *)bsearch((char *)(complist+i), (char *)clist, |
128 |
> |
nents, sizeof(PACKHEAD), beamidcmp); |
129 |
> |
if (csm == NULL) |
130 |
> |
continue; |
131 |
> |
oldnr = complist[i].nr; |
132 |
> |
csm->nc = complist[i].nc; |
133 |
> |
switch (op) { |
134 |
> |
case BS_ADD: /* add to count */ |
135 |
> |
complist[i].nr += csm->nr; |
136 |
> |
csm->nr = 0; |
137 |
> |
break; |
138 |
> |
case BS_ADJ: /* reset count */ |
139 |
> |
complist[i].nr = csm->nr; |
140 |
> |
csm->nr = 0; |
141 |
> |
break; |
142 |
> |
case BS_DEL: /* delete count */ |
143 |
> |
if (csm->nr == 0 || csm->nr >= complist[i].nr) |
144 |
> |
complist[i].nr = 0; |
145 |
> |
else |
146 |
> |
complist[i].nr -= csm->nr; |
147 |
> |
break; |
148 |
> |
} |
149 |
> |
if (complist[i].nr != oldnr) |
150 |
> |
lastin = -1; /* flag sort */ |
151 |
> |
} |
152 |
> |
/* record computed rays for uncommon beams */ |
153 |
> |
for (csm = clist+nents; csm-- > clist; ) |
154 |
> |
if (csm->nc < 0) |
155 |
> |
csm->nc = bnrays(hdlist[csm->hd], csm->bi); |
156 |
> |
/* complete list operations */ |
157 |
|
switch (op) { |
158 |
|
case BS_NEW: /* new computation set */ |
159 |
< |
if (complen) |
159 |
> |
listpos = 0; lastin = -1; |
160 |
> |
if (complen) /* free old list */ |
161 |
|
free((char *)complist); |
162 |
< |
if (nents <= 0) { |
163 |
< |
complist = NULL; |
72 |
< |
listpos = complen = 0; |
73 |
< |
lastin = -1; |
162 |
> |
complist = NULL; |
163 |
> |
if (!(complen = nents)) |
164 |
|
return; |
75 |
– |
} |
165 |
|
complist = (PACKHEAD *)malloc(nents*sizeof(PACKHEAD)); |
166 |
|
if (complist == NULL) |
167 |
|
goto memerr; |
168 |
|
bcopy((char *)clist, (char *)complist, nents*sizeof(PACKHEAD)); |
80 |
– |
complen = nents; /* finish initialization below */ |
169 |
|
break; |
170 |
|
case BS_ADD: /* add to computation set */ |
171 |
|
case BS_ADJ: /* adjust set quantities */ |
172 |
|
if (nents <= 0) |
173 |
|
return; |
174 |
< |
/* merge any common members */ |
87 |
< |
for (n = 0; n < nents; n++) { |
88 |
< |
for (i = 0; i < complen; i++) |
89 |
< |
if (clist[n].bi == complist[i].bi && |
90 |
< |
clist[n].hd == complist[i].hd) { |
91 |
< |
int oldnr = complist[i].nr; |
92 |
< |
if (op == BS_ADD) |
93 |
< |
complist[i].nr += clist[n].nr; |
94 |
< |
else /* op == BS_ADJ */ |
95 |
< |
complist[i].nr = clist[n].nr; |
96 |
< |
clist[n].nr = 0; |
97 |
< |
clist[n].nc = complist[i].nc; |
98 |
< |
if (complist[i].nr != oldnr) |
99 |
< |
lastin = -1; /* flag sort */ |
100 |
< |
break; |
101 |
< |
} |
102 |
< |
if (i >= complen) |
103 |
< |
clist[n].nc = bnrays(hdlist[clist[n].hd], |
104 |
< |
clist[n].bi); |
105 |
< |
} |
106 |
< |
/* sort updated list */ |
107 |
< |
sortcomplist(); |
108 |
< |
/* sort new entries */ |
174 |
> |
sortcomplist(); /* sort updated list & new entries */ |
175 |
|
qsort((char *)clist, nents, sizeof(PACKHEAD), beamcmp); |
176 |
|
/* what can't we satisfy? */ |
177 |
< |
for (n = 0; n < nents && clist[n].nr > clist[n].nc; n++) |
177 |
> |
for (i = nents, csm = clist; i-- && csm->nr > csm->nc; csm++) |
178 |
|
; |
179 |
< |
if (op == BS_ADJ) |
180 |
< |
nents = n; |
179 |
> |
n = csm - clist; |
180 |
> |
if (op == BS_ADJ) { /* don't regenerate adjusted beams */ |
181 |
> |
for (++i; i-- && csm->nr > 0; csm++) |
182 |
> |
; |
183 |
> |
nents = csm - clist; |
184 |
> |
} |
185 |
|
if (n) { /* allocate space for merged list */ |
186 |
|
PACKHEAD *newlist; |
187 |
|
newlist = (PACKHEAD *)malloc( |
199 |
|
lastin = complen-1; /* list is now sorted */ |
200 |
|
break; |
201 |
|
case BS_DEL: /* delete from computation set */ |
202 |
< |
if (nents <= 0) |
133 |
< |
return; |
134 |
< |
/* find each member */ |
135 |
< |
for (i = 0; i < complen; i++) |
136 |
< |
for (n = 0; n < nents; n++) |
137 |
< |
if (clist[n].bi == complist[i].bi && |
138 |
< |
clist[n].hd == complist[i].hd) { |
139 |
< |
if (clist[n].nr == 0 || |
140 |
< |
clist[n].nr >= complist[i].nr) |
141 |
< |
complist[i].nr = 0; |
142 |
< |
else |
143 |
< |
complist[i].nr -= clist[n].nr; |
144 |
< |
lastin = -1; /* flag full sort */ |
145 |
< |
break; |
146 |
< |
} |
147 |
< |
return; /* no display */ |
202 |
> |
return; /* already done */ |
203 |
|
default: |
204 |
|
error(CONSISTENCY, "bundle_set called with unknown operation"); |
205 |
|
} |
206 |
< |
if (outdev != NULL) { /* load and display beams we have */ |
207 |
< |
register HDBEAMI *hb; |
208 |
< |
|
209 |
< |
hb = (HDBEAMI *)malloc(nents*sizeof(HDBEAMI)); |
210 |
< |
for (i = 0; i < nents; i++) { |
211 |
< |
hb[i].h = hdlist[clist[i].hd]; |
212 |
< |
hb[i].b = clist[i].bi; |
158 |
< |
} |
159 |
< |
hdloadbeams(hb, nents, dispbeam); |
160 |
< |
free((char *)hb); |
206 |
> |
if (outdev == NULL || !nents) /* nothing to display? */ |
207 |
> |
return; |
208 |
> |
/* load and display beams we have */ |
209 |
> |
hbarr = (HDBEAMI *)malloc(nents*sizeof(HDBEAMI)); |
210 |
> |
for (i = nents; i--; ) { |
211 |
> |
hbarr[i].h = hdlist[clist[i].hd]; |
212 |
> |
hbarr[i].b = clist[i].bi; |
213 |
|
} |
214 |
< |
if (op == BS_NEW) { |
215 |
< |
done_packets(flush_queue()); /* empty queue, so we can... */ |
216 |
< |
for (i = 0; i < complen; i++) /* ...get number computed */ |
165 |
< |
complist[i].nc = bnrays(hdlist[complist[i].hd], |
166 |
< |
complist[i].bi); |
214 |
> |
hdloadbeams(hbarr, nents, dispbeam); |
215 |
> |
free((char *)hbarr); |
216 |
> |
if (hdfragflags&FF_READ) { |
217 |
|
listpos = 0; |
218 |
< |
lastin = -1; /* flag for initial sort */ |
218 |
> |
lastin = -1; /* need to re-sort list */ |
219 |
|
} |
220 |
|
return; |
221 |
|
memerr: |
262 |
|
double frac; |
263 |
|
int i; |
264 |
|
register int j, k; |
265 |
< |
/* free old list */ |
266 |
< |
if (complen > 0) |
265 |
> |
/* free old list and empty queue */ |
266 |
> |
if (complen > 0) { |
267 |
|
free((char *)complist); |
268 |
+ |
done_packets(flush_queue()); |
269 |
+ |
} |
270 |
+ |
/* reseed random number generator */ |
271 |
+ |
srandom(time(NULL)); |
272 |
|
/* allocate beam list */ |
273 |
|
complen = 0; |
274 |
|
for (j = 0; hdlist[j] != NULL; j++) |
279 |
|
/* compute beam weights */ |
280 |
|
k = 0; |
281 |
|
for (j = 0; hdlist[j] != NULL; j++) { |
282 |
< |
frac = 512. * hdlist[j]->wg[0] * |
283 |
< |
hdlist[j]->wg[1] * hdlist[j]->wg[2]; |
284 |
< |
if (frac < 0.) frac = -frac; |
282 |
> |
frac = 512. * VLEN(hdlist[j]->wg[0]) * |
283 |
> |
VLEN(hdlist[j]->wg[1]) * |
284 |
> |
VLEN(hdlist[j]->wg[2]); |
285 |
|
for (i = nbeams(hdlist[j]); i > 0; i--) { |
286 |
|
complist[k].hd = j; |
287 |
|
complist[k].bi = i; |
288 |
|
complist[k].nr = frac*beamvolume(hdlist[j], i) + 0.5; |
289 |
+ |
complist[k].nc = bnrays(hdlist[j], i); |
290 |
|
wtotal += complist[k++].nr; |
291 |
|
} |
292 |
|
} |
296 |
|
else |
297 |
|
frac = 1024.*1024.*16384. / (wtotal*sizeof(RAYVAL)); |
298 |
|
while (k--) |
299 |
< |
complist[k].nr = frac * complist[k].nr; |
300 |
< |
listpos = 0; lastin = -1; /* flag initial sort */ |
299 |
> |
complist[k].nr = frac*complist[k].nr + 0.5; |
300 |
> |
listpos = 0; lastin = -1; /* perform initial sort */ |
301 |
> |
sortcomplist(); |
302 |
> |
/* no view vicinity */ |
303 |
> |
myeye.rng = 0; |
304 |
|
} |
305 |
|
|
306 |
|
|
330 |
|
sortcomplist() /* fix our list order */ |
331 |
|
{ |
332 |
|
PACKHEAD *list2; |
333 |
+ |
int listlen; |
334 |
|
register int i; |
335 |
|
|
336 |
|
if (complen <= 0) /* check to see if there is even a list */ |
337 |
|
return; |
338 |
+ |
if (!chunkycmp) /* check to see if fragment list is full */ |
339 |
+ |
if (!hdfragOK(hdlist[0]->fd, &listlen, NULL) |
340 |
+ |
#if NFRAG2CHUNK |
341 |
+ |
|| listlen >= NFRAG2CHUNK |
342 |
+ |
#endif |
343 |
+ |
) { |
344 |
+ |
chunkycmp++; /* use "chunky" comparison */ |
345 |
+ |
lastin = -1; /* need to re-sort list */ |
346 |
+ |
#ifdef DEBUG |
347 |
+ |
error(WARNING, "using chunky comparison mode"); |
348 |
+ |
#endif |
349 |
+ |
} |
350 |
|
if (lastin < 0 || listpos*4 >= complen*3) |
351 |
|
qsort((char *)complist, complen, sizeof(PACKHEAD), beamcmp); |
352 |
|
else if (listpos) { /* else sort and merge sublist */ |
369 |
|
} else if (i < complen-1) { |
370 |
|
list2 = (PACKHEAD *)realloc((char *)complist, |
371 |
|
(i+1)*sizeof(PACKHEAD)); |
372 |
< |
if (list2 != NULL) { |
372 |
> |
if (list2 != NULL) |
373 |
|
complist = list2; |
374 |
< |
complen = i+1; |
304 |
< |
} |
374 |
> |
complen = i+1; |
375 |
|
} |
376 |
|
listpos = 0; lastin = i; |
377 |
|
} |
382 |
|
* more or less evenly distributed, such that computing a packet causes |
383 |
|
* a given bundle to move way down in the computation order. We keep |
384 |
|
* track of where the computed bundle with the highest priority would end |
385 |
< |
* up, and if we get further in our compute list than this, we resort the |
385 |
> |
* up, and if we get further in our compute list than this, we re-sort the |
386 |
|
* list and start again from the beginning. Since |
387 |
|
* a merge sort is used, the sorting costs are minimal. |
388 |
|
*/ |
402 |
|
p->nr = complist[listpos].nr - p->nc; |
403 |
|
if (p->nr <= 0) |
404 |
|
return(0); |
405 |
< |
#ifdef DEBUG |
406 |
< |
if (n < 1 | n > RPACKSIZ) |
337 |
< |
error(CONSISTENCY, "next_packet called with bad n value"); |
338 |
< |
#endif |
405 |
> |
DCHECK(n < 1 | n > RPACKSIZ, |
406 |
> |
CONSISTENCY, "next_packet called with bad n value"); |
407 |
|
if (p->nr > n) |
408 |
|
p->nr = n; |
409 |
|
complist[listpos].nc += p->nr; /* find where this one would go */ |
410 |
+ |
if (hdgetbeam(hdlist[p->hd], p->bi) != NULL) |
411 |
+ |
hdfreefrag(hdlist[p->hd], p->bi); |
412 |
|
while (lastin > listpos && |
413 |
|
beamcmp(complist+lastin, complist+listpos) > 0) |
414 |
|
lastin--; |