ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/radiance/src/rt/raypcalls.c
(Generate patch)

Comparing src/rt/raypcalls.c (file contents):
Revision 2.5 by schorsch, Sun Jul 27 22:12:03 2003 UTC vs.
Revision 2.34 by greg, Tue Jun 16 17:58:11 2020 UTC

# Line 13 | Line 13 | static const char      RCSid[] = "$Id$";
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
# Line 43 | Line 48 | static const char      RCSid[] = "$Id$";
48   *      myRay.rorg = ( ray origin point )
49   *      myRay.rdir = ( normalized ray direction )
50   *      myRay.rmax = ( maximum length, or zero for no limit )
51 < *      rayorigin(&myRay, NULL, PRIMARY, 1.0);
51 > *      rayorigin(&myRay, PRIMARY, NULL, NULL);
52   *      myRay.rno = ( my personal ray identifier )
53   *      if (ray_pqueue(&myRay) == 1)
54   *              { do something with results }
# Line 51 | Line 56 | static const char      RCSid[] = "$Id$";
56   *  Note the differences between this and the simpler ray_trace()
57   *  call.  In particular, the call may or may not return a value
58   *  in the passed ray structure.  Also, you need to call rayorigin()
59 < *  yourself, which is normally for you by ray_trace().  The
60 < *  great thing is that ray_pqueue() will trace rays faster in
59 > *  yourself, which is normally called for you by ray_trace().  The
60 > *  benefit is that ray_pqueue() will trace rays faster in
61   *  proportion to the number of CPUs you have available on your
62   *  system.  If the ray queue is full before the call, ray_pqueue()
63   *  will block until a result is ready so it can queue this one.
# Line 68 | Line 73 | static const char      RCSid[] = "$Id$";
73   *  results aren't ready, but will immediately return 0.
74   *  If the second argument is 0, the call will block
75   *  until a value is available, returning 0 only if the
76 < *  queue is completely empty.  A negative return value
76 > *  queue is completely empty.  Setting the second argument
77 > *  to -1 returns 0 unless a ray is ready in the queue and
78 > *  no system calls are needed.  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
# Line 81 | Line 88 | static const char      RCSid[] = "$Id$";
88   *              ray_psend(&myRay);
89   *      }
90   *
91 < *  The ray_presult() and/or ray_pqueue() functions may then be
92 < *  called to read back the results.
91 > *  Note that it is a mistake to call ra_psend() when
92 > *  ray_pnidle is zero, and nothing will be sent in
93 > *  this case.  Otherwise, the ray_presult() and/or ray_pqueue()
94 > *  functions may be called subsequently to read back the results
95 > *  of rays queued by ray_psend().
96   *
97   *  When you are done, you may call ray_pdone(1) to close
98   *  all child processes and clean up memory used by Radiance.
99   *  Any queued ray calculations will be awaited and discarded.
100   *  As with ray_done(), ray_pdone(0) hangs onto data files
101   *  and fonts that are likely to be used in subsequent renderings.
102 < *  Whether you want to bother cleaning up memory or not, you
103 < *  should at least call ray_pclose(0) to clean the child processes.
102 > *  Whether you need to clean up memory or not, you should
103 > *  at least call ray_pclose(0) to await the child processes.
104 > *  The caller should define a quit() function that calls
105 > *  ray_pclose(0) if ray_pnprocs > 0.
106   *
107   *  Warning:  You cannot affect any of the rendering processes
108   *  by changing global parameter values onece ray_pinit() has
# Line 99 | Line 111 | static const char      RCSid[] = "$Id$";
111   *  If you just want to reap children so that you can alter the
112   *  rendering parameters without reloading the scene, use the
113   *  ray_pclose(0) and ray_popen(nproc) calls to close
114 < *  then restart the child processes.
114 > *  then restart the child processes after the changes are made.
115   *
116   *  Note:  These routines are written to coordinate with the
117   *  definitions in raycalls.c, and in fact depend on them.
118   *  If you want to trace a ray and get a result synchronously,
119   *  use the ray_trace() call to compute it in the parent process.
120 + *  This will not interfere with any subprocess calculations,
121 + *  but beware that a fatal error may end with a call to quit().
122   *
123   *  Note:  One of the advantages of using separate processes
124   *  is that it gives the calling program some immunity from
125   *  fatal rendering errors.  As discussed in raycalls.c,
126   *  Radiance tends to throw up its hands and exit at the
127   *  first sign of trouble, calling quit() to return control
128 < *  to the system.  Although you can avoid exit() with
128 > *  to the top level.  Although you can avoid exit() with
129   *  your own longjmp() in quit(), the cleanup afterwards
130   *  is always suspect.  Through the use of subprocesses,
131   *  we avoid this pitfall by closing the processes and
132   *  returning a negative value from ray_pqueue() or
133   *  ray_presult().  If you get a negative value from either
134   *  of these calls, you can assume that the processes have
135 < *  been cleaned up with a call to ray_close(), though you
135 > *  been cleaned up with a call to ray_pclose(), though you
136   *  will have to call ray_pdone() yourself if you want to
137 < *  free memory.  Obviously, you cannot continue rendering,
138 < *  but otherwise your process should not be compromised.
137 > *  free memory.  Obviously, you cannot continue rendering
138 > *  without risking further errors, but otherwise your
139 > *  process should not be compromised.
140   */
141  
142 + #include  "rtprocess.h"
143   #include  "ray.h"
144 <
144 > #include  "ambient.h"
145 > #include  <sys/types.h>
146 > #include  <sys/wait.h>
147   #include  "selcall.h"
148  
149   #ifndef RAYQLEN
150 < #define RAYQLEN         16              /* # rays to send at once */
150 > #define RAYQLEN         24              /* # rays to send at once */
151   #endif
152  
153   #ifndef MAX_RPROCS
# Line 146 | Line 164 | int            ray_pnprocs = 0;        /* number of child processes */
164   int             ray_pnidle = 0;         /* number of idle children */
165  
166   static struct child_proc {
167 <        int     pid;                            /* child process id */
167 >        RT_PID  pid;                            /* child process id */
168          int     fd_send;                        /* write to child here */
169          int     fd_recv;                        /* read from child here */
170          int     npending;                       /* # rays in process */
171 <        unsigned long  rno[RAYQLEN];            /* working on these rays */
171 >        RNUMBER rno[RAYQLEN];                   /* working on these rays */
172   } r_proc[MAX_NPROCS];                   /* our child processes */
173  
174   static RAY      r_queue[2*RAYQLEN];     /* ray i/o buffer */
175 < static int      r_send_next;            /* next send ray placement */
176 < static int      r_recv_first;           /* position of first unreported ray */
177 < static int      r_recv_next;            /* next receive ray placement */
175 > static int      r_send_next = 0;        /* next send ray placement */
176 > static int      r_recv_first = RAYQLEN; /* position of first unreported ray */
177 > static int      r_recv_next = RAYQLEN;  /* next received ray placement */
178  
179 + static int      samplestep = 1;         /* sample step size */
180 +
181   #define sendq_full()    (r_send_next >= RAYQLEN)
182  
183 + static int ray_pflush(void);
184 + static void ray_pchild(int fd_in, int fd_out);
185  
186 +
187   void
188 < ray_pinit(otnm, nproc)          /* initialize ray-tracing processes */
189 < char    *otnm;
190 < int     nproc;
188 > ray_pinit(              /* initialize ray-tracing processes */
189 >        char    *otnm,
190 >        int     nproc
191 > )
192   {
193          if (nobjects > 0)               /* close old calculation */
194                  ray_pdone(0);
195  
196          ray_init(otnm);                 /* load the shared scene */
197  
174 –        preload_objs();                 /* preload auxiliary data */
175 –
176 –                                        /* set shared memory boundary */
177 –        shm_boundary = (char *)malloc(16);
178 –        strcpy(shm_boundary, "SHM_BOUNDARY");
179 –
180 –        r_send_next = 0;                /* set up queue */
181 –        r_recv_first = r_recv_next = RAYQLEN;
182 –
198          ray_popen(nproc);               /* fork children */
199   }
200  
201  
202   static int
203 < ray_pflush()                    /* send queued rays to idle children */
203 > ray_pflush(void)                        /* send queued rays to idle children */
204   {
205          int     nc, n, nw, i, sfirst;
206  
# Line 197 | Line 212 | ray_pflush()                   /* send queued rays to idle children */
212          for (i = ray_pnprocs; nc && i--; ) {
213                  if (r_proc[i].npending > 0)
214                          continue;       /* child looks busy */
215 <                n = (r_send_next - sfirst)/nc--;
215 >                n = (r_send_next - sfirst) / nc--;
216                  if (!n)
217                          continue;
218                                          /* smuggle set size in crtype */
# Line 213 | Line 228 | ray_pflush()                   /* send queued rays to idle children */
228                  ray_pnidle--;           /* now she's busy */
229          }
230          if (sfirst != r_send_next)
231 <                error(CONSISTENCY, "code screwup in ray_pflush");
231 >                error(CONSISTENCY, "code screwup in ray_pflush()");
232          r_send_next = 0;
233          return(sfirst);                 /* return total # sent */
234   }
235  
236  
237 < void
238 < ray_psend(r)                    /* add a ray to our send queue */
239 < RAY     *r;
237 > int
238 > ray_psend(                      /* add a ray to our send queue */
239 >        RAY     *r
240 > )
241   {
242 <        if (r == NULL)
243 <                return;
242 >        int     rv;
243 >
244 >        if ((r == NULL) | (ray_pnidle <= 0))
245 >                return(0);
246                                          /* flush output if necessary */
247 <        if (sendq_full() && ray_pflush() <= 0)
248 <                error(INTERNAL, "ray_pflush failed in ray_psend");
247 >        if (sendq_full() && (rv = ray_pflush()) <= 0)
248 >                return(rv);
249  
250 <        r_queue[r_send_next] = *r;
251 <        r_send_next++;
250 >        r_queue[r_send_next++] = *r;
251 >        return(1);
252   }
253  
254  
255   int
256 < ray_pqueue(r)                   /* queue a ray for computation */
257 < RAY     *r;
256 > ray_pqueue(                     /* queue a ray for computation */
257 >        RAY     *r
258 > )
259   {
260          if (r == NULL)
261                  return(0);
262                                          /* check for full send queue */
263          if (sendq_full()) {
264 <                RAY     mySend;
246 <                int     rval;
247 <                mySend = *r;
264 >                RAY     mySend = *r;
265                                          /* wait for a result */
266 <                rval = ray_presult(r, 0);
266 >                if (ray_presult(r, 0) <= 0)
267 >                        return(-1);
268                                          /* put new ray in queue */
269 <                r_queue[r_send_next] = mySend;
270 <                r_send_next++;
271 <                return(rval);           /* done */
269 >                r_queue[r_send_next++] = mySend;
270 >
271 >                return(1);
272          }
273 <                                        /* add ray to send queue */
274 <        r_queue[r_send_next] = *r;
257 <        r_send_next++;
273 >                                        /* else add ray to send queue */
274 >        r_queue[r_send_next++] = *r;
275                                          /* check for returned ray... */
276          if (r_recv_first >= r_recv_next)
277                  return(0);
278                                          /* ...one is sitting in queue */
279 <        *r = r_queue[r_recv_first];
263 <        r_recv_first++;
279 >        *r = r_queue[r_recv_first++];
280          return(1);
281   }
282  
283  
284   int
285 < ray_presult(r, poll)            /* check for a completed ray */
286 < RAY     *r;
287 < int     poll;
285 > ray_presult(            /* check for a completed ray */
286 >        RAY     *r,
287 >        int     poll
288 > )
289   {
290          static struct timeval   tpoll;  /* zero timeval struct */
291          static fd_set   readset, errset;
292          int     n, ok;
293 <        register int    pn;
293 >        int     pn;
294  
295          if (r == NULL)
296                  return(0);
297                                          /* check queued results first */
298          if (r_recv_first < r_recv_next) {
299 <                *r = r_queue[r_recv_first];
283 <                r_recv_first++;
299 >                *r = r_queue[r_recv_first++];
300                  return(1);
301          }
302 +        if (poll < 0)                   /* immediate polling mode? */
303 +                return(0);
304 +
305          n = ray_pnprocs - ray_pnidle;   /* pending before flush? */
306  
307          if (ray_pflush() < 0)           /* send new rays to process */
# Line 294 | Line 313 | int    poll;
313                  n = ray_pnprocs - ray_pnidle;
314          if (n <= 0)                     /* return if nothing to await */
315                  return(0);
316 +        if (!poll && ray_pnprocs == 1)  /* one process -> skip select() */
317 +                FD_SET(r_proc[0].fd_recv, &readset);
318 +
319   getready:                               /* any children waiting for us? */
320          for (pn = ray_pnprocs; pn--; )
321                  if (FD_ISSET(r_proc[pn].fd_recv, &readset) ||
322                                  FD_ISSET(r_proc[pn].fd_recv, &errset))
323                          break;
324 <                                        /* call select if we must */
324 >                                        /* call select() if we must */
325          if (pn < 0) {
326                  FD_ZERO(&readset); FD_ZERO(&errset); n = 0;
327                  for (pn = ray_pnprocs; pn--; ) {
# Line 314 | Line 336 | getready:                              /* any children waiting for us? */
336                                  poll ? &tpoll : (struct timeval *)NULL)) < 0)
337                          if (errno != EINTR) {
338                                  error(WARNING,
339 <                                        "select call failed in ray_presult");
339 >                                        "select call failed in ray_presult()");
340                                  ray_pclose(0);
341                                  return(-1);
342                          }
# Line 346 | Line 368 | getready:                              /* any children waiting for us? */
368          }
369                                          /* preen returned rays */
370          for (n = r_recv_next - r_recv_first; n--; ) {
371 <                register RAY    *rp = &r_queue[r_recv_first + n];
371 >                RAY     *rp = &r_queue[r_recv_first + n];
372                  rp->rno = r_proc[pn].rno[n];
373                  rp->parent = NULL;
374                  rp->newcset = rp->clipset = NULL;
# Line 354 | Line 376 | getready:                              /* any children waiting for us? */
376                  rp->slights = NULL;
377          }
378                                          /* return first ray received */
379 <        *r = r_queue[r_recv_first];
358 <        r_recv_first++;
379 >        *r = r_queue[r_recv_first++];
380          return(1);
381   }
382  
383  
384   void
385 < ray_pdone(freall)               /* reap children and free data */
386 < int     freall;
385 > ray_pdone(              /* reap children and free data */
386 >        int     freall
387 > )
388   {
389          ray_pclose(0);                  /* close child processes */
390  
# Line 370 | Line 392 | int    freall;
392                  free((void *)shm_boundary);
393                  shm_boundary = NULL;
394          }
395 +
396          ray_done(freall);               /* free rendering data */
397   }
398  
399  
400   static void
401 < ray_pchild(fd_in, fd_out)       /* process rays (never returns) */
402 < int     fd_in;
403 < int     fd_out;
401 > ray_pchild(     /* process rays (never returns) */
402 >        int     fd_in,
403 >        int     fd_out
404 > )
405   {
406          int     n;
407 <        register int    i;
407 >        int     i;
408 >                                        /* flag child process for quit() */
409 >        ray_pnprocs = -1;
410                                          /* read each ray request set */
411          while ((n = read(fd_in, (char *)r_queue, sizeof(r_queue))) > 0) {
412                  int     n2;
413 <                if (n % sizeof(RAY))
413 >                if (n < sizeof(RAY))
414                          break;
389 –                n /= sizeof(RAY);
415                                          /* get smuggled set length */
416 <                n2 = r_queue[0].crtype - n;
416 >                n2 = sizeof(RAY)*r_queue[0].crtype - n;
417                  if (n2 < 0)
418 <                        error(INTERNAL, "buffer over-read in ray_pchild");
418 >                        error(INTERNAL, "buffer over-read in ray_pchild()");
419                  if (n2 > 0) {           /* read the rest of the set */
420 <                        i = readbuf(fd_in, (char *)(r_queue+n),
421 <                                        sizeof(RAY)*n2);
397 <                        if (i != sizeof(RAY)*n2)
420 >                        i = readbuf(fd_in, (char *)r_queue + n, n2);
421 >                        if (i != n2)
422                                  break;
423                          n += n2;
424                  }
425 +                n /= sizeof(RAY);
426                                          /* evaluate rays */
427                  for (i = 0; i < n; i++) {
428                          r_queue[i].crtype = r_queue[i].rtype;
429                          r_queue[i].parent = NULL;
430                          r_queue[i].clipset = NULL;
431                          r_queue[i].slights = NULL;
432 <                        r_queue[i].revf = raytrace;
433 <                        samplendx++;
432 >                        r_queue[i].rlvl = 0;
433 >                        samplendx += samplestep;
434                          rayclear(&r_queue[i]);
435                          rayvalue(&r_queue[i]);
436                  }
437                                          /* write back our results */
438                  i = writebuf(fd_out, (char *)r_queue, sizeof(RAY)*n);
439                  if (i != sizeof(RAY)*n)
440 <                        error(SYSTEM, "write error in ray_pchild");
440 >                        error(SYSTEM, "write error in ray_pchild()");
441          }
442          if (n)
443 <                error(SYSTEM, "read error in ray_pchild");
443 >                error(SYSTEM, "read error in ray_pchild()");
444          ambsync();
445          quit(0);                        /* normal exit */
446   }
447  
448  
449   void
450 < ray_popen(nadd)                 /* open the specified # processes */
451 < int     nadd;
450 > ray_popen(                      /* open the specified # processes */
451 >        int     nadd
452 > )
453   {
454                                          /* check if our table has room */
455          if (ray_pnprocs + nadd > MAX_NPROCS)
456                  nadd = MAX_NPROCS - ray_pnprocs;
457          if (nadd <= 0)
458                  return;
459 <        fflush(stderr);                 /* clear pending output */
460 <        fflush(stdout);
459 >        ambsync();                      /* load any new ambient values */
460 >        if (shm_boundary == NULL) {     /* first child process? */
461 >                preload_objs();         /* preload auxiliary data */
462 >                                        /* set shared memory boundary */
463 >                shm_boundary = (char *)malloc(16);
464 >                strcpy(shm_boundary, "SHM_BOUNDARY");
465 >        }
466 >        fflush(NULL);                   /* clear pending output */
467 >        samplestep = ray_pnprocs + nadd;
468          while (nadd--) {                /* fork each new process */
469                  int     p0[2], p1[2];
470                  if (pipe(p0) < 0 || pipe(p1) < 0)
# Line 443 | Line 476 | int    nadd;
476                                  close(r_proc[pn].fd_recv);
477                          }
478                          close(p0[0]); close(p1[1]);
479 +                        close(0);       /* don't share stdin */
480                                          /* following call never returns */
481                          ray_pchild(p1[0], p0[1]);
482                  }
483                  if (r_proc[ray_pnprocs].pid < 0)
484                          error(SYSTEM, "cannot fork child process");
485                  close(p1[0]); close(p0[1]);
486 +                if (rand_samp)          /* decorrelate random sequence */
487 +                        srandom(random());
488 +                else
489 +                        samplendx++;
490 +                /*
491 +                 * Close write stream on exec to avoid multiprocessing deadlock.
492 +                 * No use in read stream without it, so set flag there as well.
493 +                 */
494 +                fcntl(p1[1], F_SETFD, FD_CLOEXEC);
495 +                fcntl(p0[0], F_SETFD, FD_CLOEXEC);
496                  r_proc[ray_pnprocs].fd_send = p1[1];
497                  r_proc[ray_pnprocs].fd_recv = p0[0];
498                  r_proc[ray_pnprocs].npending = 0;
# Line 459 | Line 503 | int    nadd;
503  
504  
505   void
506 < ray_pclose(nsub)                /* close one or more child processes */
507 < int     nsub;
506 > ray_pclose(             /* close one or more child processes */
507 >        int     nsub
508 > )
509   {
510          static int      inclose = 0;
511 <        RAY     res;
511 >        RAY             res;
512 >        int             i, status = 0;
513                                          /* check recursion */
514          if (inclose)
515                  return;
516          inclose++;
517 +                                        /* check no child / in child */
518 +        if (ray_pnprocs <= 0)
519 +                return;
520                                          /* check argument */
521          if ((nsub <= 0) | (nsub > ray_pnprocs))
522                  nsub = ray_pnprocs;
523                                          /* clear our ray queue */
524          while (ray_presult(&res,0) > 0)
525                  ;
526 <                                        /* clean up children */
527 <        while (nsub--) {
528 <                int     status;
529 <                ray_pnprocs--;
530 <                close(r_proc[ray_pnprocs].fd_recv);
531 <                close(r_proc[ray_pnprocs].fd_send);
532 <                while (wait(&status) != r_proc[ray_pnprocs].pid)
533 <                        ;
534 <                if (status) {
535 <                        sprintf(errmsg,
536 <                                "rendering process %d exited with code %d",
537 <                                        r_proc[ray_pnprocs].pid, status>>8);
538 <                        error(WARNING, errmsg);
526 >        r_send_next = 0;                /* hard reset in case of error */
527 >        r_recv_first = r_recv_next = RAYQLEN;
528 >                                        /* close send pipes */
529 >        for (i = ray_pnprocs-nsub; i < ray_pnprocs; i++)
530 >                close(r_proc[i].fd_send);
531 >
532 >        if (nsub == 1) {                /* awaiting single process? */
533 >                if (waitpid(r_proc[ray_pnprocs-1].pid, &status, 0) < 0)
534 >                        status = 127<<8;
535 >                close(r_proc[ray_pnprocs-1].fd_recv);
536 >        } else                          /* else unordered wait */
537 >                for (i = 0; i < nsub; ) {
538 >                        int     j, mystatus;
539 >                        RT_PID  pid = wait(&mystatus);
540 >                        for (j = ray_pnprocs-nsub; j < ray_pnprocs; j++)
541 >                                if (r_proc[j].pid == pid) {
542 >                                        if (mystatus)
543 >                                                status = mystatus;
544 >                                        close(r_proc[j].fd_recv);
545 >                                        ++i;
546 >                                }
547                  }
548 <                ray_pnidle--;
548 >        ray_pnprocs -= nsub;
549 >        ray_pnidle -= nsub;
550 >        if (status) {
551 >                sprintf(errmsg, "rendering process exited with code %d", status>>8);
552 >                error(WARNING, errmsg);
553          }
554          inclose--;
494 – }
495 –
496 –
497 – void
498 – quit(ec)                        /* make sure exit is called */
499 – int     ec;
500 – {
501 –        exit(ec);
555   }

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines