13 |
|
* These calls are designed similarly to the ones in raycalls.c, |
14 |
|
* but allow for multiple rendering processes on the same host |
15 |
|
* machine. There is no sense in specifying more child processes |
16 |
< |
* than you have processors, but one child may help by allowing |
16 |
> |
* than you have processor cores, but one child may help by allowing |
17 |
|
* asynchronous ray computation in an interactive program, and |
18 |
|
* will protect the caller from fatal rendering errors. |
19 |
|
* |
20 |
< |
* You should first read and undrstand the header in raycalls.c, |
20 |
> |
* You should first read and understand the header in raycalls.c, |
21 |
|
* as some things are explained there that are not repated here. |
22 |
|
* |
23 |
|
* The first step is opening one or more rendering processes |
24 |
|
* with a call to ray_pinit(oct, nproc). Before calling fork(), |
25 |
|
* ray_pinit() loads the octree and data structures into the |
26 |
< |
* caller's memory. This permits all sorts of queries that |
27 |
< |
* wouldn't be possible otherwise, without causing any real |
26 |
> |
* caller's memory, and ray_popen() synchronizes the ambient |
27 |
> |
* file, if any. Shared memory permits all sorts of queries |
28 |
> |
* that wouldn't be possible otherwise without causing any real |
29 |
|
* memory overhead, since all the static data are shared |
30 |
< |
* between processes. Rays are then traced using a simple |
30 |
> |
* between processes. Rays are traced using a simple |
31 |
|
* queuing mechanism, explained below. |
32 |
|
* |
33 |
< |
* The ray queue holds as many rays as there are rendering |
34 |
< |
* processes. Rays are queued and returned by a single |
33 |
> |
* The ray queue buffers RAYQLEN rays before sending to |
34 |
> |
* children, each of which may internally buffer RAYQLEN rays |
35 |
> |
* during evaluation. Rays are not returned in the order |
36 |
> |
* they are sent when multiple processes are open. |
37 |
> |
* |
38 |
> |
* Rays are queued and returned by a single |
39 |
|
* ray_pqueue() call. A ray_pqueue() return |
40 |
|
* value of 0 indicates that no rays are ready |
41 |
|
* and the queue is not yet full. A return value of 1 |
71 |
|
* |
72 |
|
* If the second argument is 1, the call won't block when |
73 |
|
* results aren't ready, but will immediately return 0. |
74 |
+ |
* (A special value of -1 returns 0 unless a ray is |
75 |
+ |
* ready in the queue and no system calls are needed.) |
76 |
|
* If the second argument is 0, the call will block |
77 |
|
* until a value is available, returning 0 only if the |
78 |
|
* queue is completely empty. A negative return value |
79 |
|
* indicates that a rendering process died. If this |
80 |
< |
* happens, ray_close(0) is automatically called to close |
80 |
> |
* happens, ray_pclose(0) is automatically called to close |
81 |
|
* all child processes, and ray_pnprocs is set to zero. |
82 |
|
* |
83 |
|
* If you just want to fill the ray queue without checking for |
97 |
|
* Any queued ray calculations will be awaited and discarded. |
98 |
|
* As with ray_done(), ray_pdone(0) hangs onto data files |
99 |
|
* and fonts that are likely to be used in subsequent renderings. |
100 |
< |
* Whether you want to bother cleaning up memory or not, you |
101 |
< |
* should at least call ray_pclose(0) to clean the child processes. |
100 |
> |
* Whether you need to clean up memory or not, you should |
101 |
> |
* at least call ray_pclose(0) to await the child processes. |
102 |
> |
* The caller should define a quit() function that calls |
103 |
> |
* ray_pclose(0) if ray_pnprocs > 0. |
104 |
|
* |
105 |
|
* Warning: You cannot affect any of the rendering processes |
106 |
|
* by changing global parameter values onece ray_pinit() has |
114 |
|
* Note: These routines are written to coordinate with the |
115 |
|
* definitions in raycalls.c, and in fact depend on them. |
116 |
|
* If you want to trace a ray and get a result synchronously, |
117 |
< |
* use the ray_trace() call to compute it in the parent process |
117 |
> |
* use the ray_trace() call to compute it in the parent process. |
118 |
|
* This will not interfere with any subprocess calculations, |
119 |
|
* but beware that a fatal error may end with a call to quit(). |
120 |
|
* |
130 |
|
* returning a negative value from ray_pqueue() or |
131 |
|
* ray_presult(). If you get a negative value from either |
132 |
|
* of these calls, you can assume that the processes have |
133 |
< |
* been cleaned up with a call to ray_close(), though you |
133 |
> |
* been cleaned up with a call to ray_pclose(), though you |
134 |
|
* will have to call ray_pdone() yourself if you want to |
135 |
|
* free memory. Obviously, you cannot continue rendering |
136 |
|
* without risking further errors, but otherwise your |
137 |
|
* process should not be compromised. |
138 |
|
*/ |
139 |
|
|
131 |
– |
#include <stdio.h> |
132 |
– |
#include <sys/types.h> |
133 |
– |
#include <sys/wait.h> /* XXX platform */ |
134 |
– |
|
140 |
|
#include "rtprocess.h" |
141 |
|
#include "ray.h" |
142 |
|
#include "ambient.h" |
143 |
+ |
#include <sys/types.h> |
144 |
+ |
#include <sys/wait.h> |
145 |
|
#include "selcall.h" |
146 |
|
|
147 |
|
#ifndef RAYQLEN |
148 |
< |
#define RAYQLEN 16 /* # rays to send at once */ |
148 |
> |
#define RAYQLEN 12 /* # rays to send at once */ |
149 |
|
#endif |
150 |
|
|
151 |
|
#ifndef MAX_RPROCS |
166 |
|
int fd_send; /* write to child here */ |
167 |
|
int fd_recv; /* read from child here */ |
168 |
|
int npending; /* # rays in process */ |
169 |
< |
unsigned long rno[RAYQLEN]; /* working on these rays */ |
169 |
> |
RNUMBER rno[RAYQLEN]; /* working on these rays */ |
170 |
|
} r_proc[MAX_NPROCS]; /* our child processes */ |
171 |
|
|
172 |
|
static RAY r_queue[2*RAYQLEN]; /* ray i/o buffer */ |
173 |
< |
static int r_send_next; /* next send ray placement */ |
174 |
< |
static int r_recv_first; /* position of first unreported ray */ |
175 |
< |
static int r_recv_next; /* next receive ray placement */ |
173 |
> |
static int r_send_next = 0; /* next send ray placement */ |
174 |
> |
static int r_recv_first = RAYQLEN; /* position of first unreported ray */ |
175 |
> |
static int r_recv_next = RAYQLEN; /* next received ray placement */ |
176 |
|
|
177 |
|
#define sendq_full() (r_send_next >= RAYQLEN) |
178 |
|
|
179 |
|
static int ray_pflush(void); |
180 |
< |
static void ray_pchild(int fd_in, int fd_out); |
180 |
> |
static void ray_pchild(int fd_in, int fd_out); |
181 |
|
|
182 |
|
|
183 |
< |
extern void |
183 |
> |
void |
184 |
|
ray_pinit( /* initialize ray-tracing processes */ |
185 |
|
char *otnm, |
186 |
|
int nproc |
191 |
|
|
192 |
|
ray_init(otnm); /* load the shared scene */ |
193 |
|
|
187 |
– |
preload_objs(); /* preload auxiliary data */ |
188 |
– |
|
189 |
– |
/* set shared memory boundary */ |
190 |
– |
shm_boundary = (char *)malloc(16); |
191 |
– |
strcpy(shm_boundary, "SHM_BOUNDARY"); |
192 |
– |
|
193 |
– |
r_send_next = 0; /* set up queue */ |
194 |
– |
r_recv_first = r_recv_next = RAYQLEN; |
195 |
– |
|
194 |
|
ray_popen(nproc); /* fork children */ |
195 |
|
} |
196 |
|
|
230 |
|
} |
231 |
|
|
232 |
|
|
233 |
< |
extern void |
233 |
> |
void |
234 |
|
ray_psend( /* add a ray to our send queue */ |
235 |
|
RAY *r |
236 |
|
) |
241 |
|
if (sendq_full() && ray_pflush() <= 0) |
242 |
|
error(INTERNAL, "ray_pflush failed in ray_psend"); |
243 |
|
|
244 |
< |
r_queue[r_send_next] = *r; |
247 |
< |
r_send_next++; |
244 |
> |
r_queue[r_send_next++] = *r; |
245 |
|
} |
246 |
|
|
247 |
|
|
248 |
< |
extern int |
248 |
> |
int |
249 |
|
ray_pqueue( /* queue a ray for computation */ |
250 |
|
RAY *r |
251 |
|
) |
254 |
|
return(0); |
255 |
|
/* check for full send queue */ |
256 |
|
if (sendq_full()) { |
257 |
< |
RAY mySend; |
261 |
< |
int rval; |
262 |
< |
mySend = *r; |
257 |
> |
RAY mySend = *r; |
258 |
|
/* wait for a result */ |
259 |
< |
rval = ray_presult(r, 0); |
259 |
> |
if (ray_presult(r, 0) <= 0) |
260 |
> |
return(-1); |
261 |
|
/* put new ray in queue */ |
262 |
< |
r_queue[r_send_next] = mySend; |
263 |
< |
r_send_next++; |
264 |
< |
return(rval); /* done */ |
262 |
> |
r_queue[r_send_next++] = mySend; |
263 |
> |
/* XXX r_send_next may now be > RAYQLEN */ |
264 |
> |
return(1); |
265 |
|
} |
266 |
< |
/* add ray to send queue */ |
267 |
< |
r_queue[r_send_next] = *r; |
272 |
< |
r_send_next++; |
266 |
> |
/* else add ray to send queue */ |
267 |
> |
r_queue[r_send_next++] = *r; |
268 |
|
/* check for returned ray... */ |
269 |
|
if (r_recv_first >= r_recv_next) |
270 |
|
return(0); |
271 |
|
/* ...one is sitting in queue */ |
272 |
< |
*r = r_queue[r_recv_first]; |
278 |
< |
r_recv_first++; |
272 |
> |
*r = r_queue[r_recv_first++]; |
273 |
|
return(1); |
274 |
|
} |
275 |
|
|
276 |
|
|
277 |
< |
extern int |
277 |
> |
int |
278 |
|
ray_presult( /* check for a completed ray */ |
279 |
|
RAY *r, |
280 |
|
int poll |
289 |
|
return(0); |
290 |
|
/* check queued results first */ |
291 |
|
if (r_recv_first < r_recv_next) { |
292 |
< |
*r = r_queue[r_recv_first]; |
299 |
< |
r_recv_first++; |
292 |
> |
*r = r_queue[r_recv_first++]; |
293 |
|
return(1); |
294 |
|
} |
295 |
+ |
if (poll < 0) /* immediate polling mode? */ |
296 |
+ |
return(0); |
297 |
+ |
|
298 |
|
n = ray_pnprocs - ray_pnidle; /* pending before flush? */ |
299 |
|
|
300 |
|
if (ray_pflush() < 0) /* send new rays to process */ |
306 |
|
n = ray_pnprocs - ray_pnidle; |
307 |
|
if (n <= 0) /* return if nothing to await */ |
308 |
|
return(0); |
309 |
+ |
if (!poll && ray_pnprocs == 1) /* one process -> skip select() */ |
310 |
+ |
FD_SET(r_proc[0].fd_recv, &readset); |
311 |
+ |
|
312 |
|
getready: /* any children waiting for us? */ |
313 |
|
for (pn = ray_pnprocs; pn--; ) |
314 |
|
if (FD_ISSET(r_proc[pn].fd_recv, &readset) || |
315 |
|
FD_ISSET(r_proc[pn].fd_recv, &errset)) |
316 |
|
break; |
317 |
< |
/* call select if we must */ |
317 |
> |
/* call select() if we must */ |
318 |
|
if (pn < 0) { |
319 |
|
FD_ZERO(&readset); FD_ZERO(&errset); n = 0; |
320 |
|
for (pn = ray_pnprocs; pn--; ) { |
369 |
|
rp->slights = NULL; |
370 |
|
} |
371 |
|
/* return first ray received */ |
372 |
< |
*r = r_queue[r_recv_first]; |
374 |
< |
r_recv_first++; |
372 |
> |
*r = r_queue[r_recv_first++]; |
373 |
|
return(1); |
374 |
|
} |
375 |
|
|
376 |
|
|
377 |
< |
extern void |
377 |
> |
void |
378 |
|
ray_pdone( /* reap children and free data */ |
379 |
|
int freall |
380 |
|
) |
385 |
|
free((void *)shm_boundary); |
386 |
|
shm_boundary = NULL; |
387 |
|
} |
388 |
+ |
|
389 |
|
ray_done(freall); /* free rendering data */ |
390 |
|
} |
391 |
|
|
398 |
|
{ |
399 |
|
int n; |
400 |
|
register int i; |
401 |
+ |
/* flag child process for quit() */ |
402 |
+ |
ray_pnprocs = -1; |
403 |
|
/* read each ray request set */ |
404 |
|
while ((n = read(fd_in, (char *)r_queue, sizeof(r_queue))) > 0) { |
405 |
|
int n2; |
406 |
< |
if (n % sizeof(RAY)) |
406 |
> |
if (n < sizeof(RAY)) |
407 |
|
break; |
407 |
– |
n /= sizeof(RAY); |
408 |
|
/* get smuggled set length */ |
409 |
< |
n2 = r_queue[0].crtype - n; |
409 |
> |
n2 = sizeof(RAY)*r_queue[0].crtype - n; |
410 |
|
if (n2 < 0) |
411 |
|
error(INTERNAL, "buffer over-read in ray_pchild"); |
412 |
|
if (n2 > 0) { /* read the rest of the set */ |
413 |
< |
i = readbuf(fd_in, (char *)(r_queue+n), |
414 |
< |
sizeof(RAY)*n2); |
415 |
< |
if (i != sizeof(RAY)*n2) |
413 |
> |
i = readbuf(fd_in, (char *)r_queue + n, n2); |
414 |
> |
if (i != n2) |
415 |
|
break; |
416 |
|
n += n2; |
417 |
|
} |
418 |
+ |
n /= sizeof(RAY); |
419 |
|
/* evaluate rays */ |
420 |
|
for (i = 0; i < n; i++) { |
421 |
|
r_queue[i].crtype = r_queue[i].rtype; |
422 |
|
r_queue[i].parent = NULL; |
423 |
|
r_queue[i].clipset = NULL; |
424 |
|
r_queue[i].slights = NULL; |
425 |
+ |
r_queue[i].rlvl = 0; |
426 |
|
samplendx++; |
427 |
|
rayclear(&r_queue[i]); |
428 |
|
rayvalue(&r_queue[i]); |
439 |
|
} |
440 |
|
|
441 |
|
|
442 |
< |
extern void |
442 |
> |
void |
443 |
|
ray_popen( /* open the specified # processes */ |
444 |
|
int nadd |
445 |
|
) |
449 |
|
nadd = MAX_NPROCS - ray_pnprocs; |
450 |
|
if (nadd <= 0) |
451 |
|
return; |
452 |
< |
fflush(stderr); /* clear pending output */ |
453 |
< |
fflush(stdout); |
452 |
> |
ambsync(); /* load any new ambient values */ |
453 |
> |
if (shm_boundary == NULL) { /* first child process? */ |
454 |
> |
preload_objs(); /* preload auxiliary data */ |
455 |
> |
/* set shared memory boundary */ |
456 |
> |
shm_boundary = (char *)malloc(16); |
457 |
> |
strcpy(shm_boundary, "SHM_BOUNDARY"); |
458 |
> |
} |
459 |
> |
fflush(NULL); /* clear pending output */ |
460 |
|
while (nadd--) { /* fork each new process */ |
461 |
|
int p0[2], p1[2]; |
462 |
|
if (pipe(p0) < 0 || pipe(p1) < 0) |
489 |
|
} |
490 |
|
|
491 |
|
|
492 |
< |
extern void |
492 |
> |
void |
493 |
|
ray_pclose( /* close one or more child processes */ |
494 |
|
int nsub |
495 |
|
) |
506 |
|
/* clear our ray queue */ |
507 |
|
while (ray_presult(&res,0) > 0) |
508 |
|
; |
509 |
+ |
r_send_next = 0; /* hard reset in case of error */ |
510 |
+ |
r_recv_first = r_recv_next = RAYQLEN; |
511 |
|
/* clean up children */ |
512 |
|
while (nsub--) { |
513 |
|
int status; |
525 |
|
ray_pnidle--; |
526 |
|
} |
527 |
|
inclose--; |
519 |
– |
} |
520 |
– |
|
521 |
– |
|
522 |
– |
void |
523 |
– |
quit(ec) /* make sure exit is called */ |
524 |
– |
int ec; |
525 |
– |
{ |
526 |
– |
exit(ec); |
528 |
|
} |