--- ray/src/rt/raypcalls.c	2008/02/20 05:21:29	2.19
+++ ray/src/rt/raypcalls.c	2009/12/16 01:06:50	2.26
@@ -1,5 +1,5 @@
 #ifndef lint
-static const char	RCSid[] = "$Id: raypcalls.c,v 2.19 2008/02/20 05:21:29 greg Exp $";
+static const char	RCSid[] = "$Id: raypcalls.c,v 2.26 2009/12/16 01:06:50 greg Exp $";
 #endif
 /*
  *  raypcalls.c - interface for parallel rendering using Radiance
@@ -13,11 +13,11 @@ static const char	RCSid[] = "$Id: raypcalls.c,v 2.19 2
  *  These calls are designed similarly to the ones in raycalls.c,
  *  but allow for multiple rendering processes on the same host
  *  machine.  There is no sense in specifying more child processes
- *  than you have processors, but one child may help by allowing
+ *  than you have processor cores, but one child may help by allowing
  *  asynchronous ray computation in an interactive program, and
  *  will protect the caller from fatal rendering errors.
  *
- *  You should first read and undrstand the header in raycalls.c,
+ *  You should first read and understand the header in raycalls.c,
  *  as some things are explained there that are not repated here.
  *
  *  The first step is opening one or more rendering processes
@@ -25,13 +25,15 @@ static const char	RCSid[] = "$Id: raypcalls.c,v 2.19 2
  *  ray_pinit() loads the octree and data structures into the
  *  caller's memory, and ray_popen() synchronizes the ambient
  *  file, if any.  Shared memory permits all sorts of queries
- *  that wouldn't be possible otherwise, without causing any real
+ *  that wouldn't be possible otherwise without causing any real
  *  memory overhead, since all the static data are shared
- *  between processes.  Rays are then traced using a simple
+ *  between processes.  Rays are traced using a simple
  *  queuing mechanism, explained below.
  *
  *  The ray queue buffers RAYQLEN rays before sending to
- *  children, each of which may internally buffer RAYQLEN rays.
+ *  children, each of which may internally buffer RAYQLEN rays
+ *  during evaluation.  Rays are not returned in the order
+ *  they are sent when multiple processes are open.
  *
  *  Rays are queued and returned by a single
  *  ray_pqueue() call.  A ray_pqueue() return
@@ -71,9 +73,11 @@ static const char	RCSid[] = "$Id: raypcalls.c,v 2.19 2
  *  results aren't ready, but will immediately return 0.
  *  If the second argument is 0, the call will block
  *  until a value is available, returning 0 only if the
- *  queue is completely empty.  A negative return value
+ *  queue is completely empty.  Setting the second argument
+ *  to -1 returns 0 unless a ray is ready in the queue and
+ *  no system calls are needed.  A negative return value
  *  indicates that a rendering process died.  If this
- *  happens, ray_close(0) is automatically called to close
+ *  happens, ray_pclose(0) is automatically called to close
  *  all child processes, and ray_pnprocs is set to zero.
  *
  *  If you just want to fill the ray queue without checking for
@@ -84,17 +88,21 @@ static const char	RCSid[] = "$Id: raypcalls.c,v 2.19 2
  *		ray_psend(&myRay);
  *	}
  *
- *  Note that it is a fatal error to call ra_psend() when
- *  ray_pnidle is zero.  The ray_presult() and/or ray_pqueue()
- *  functions may be called subsequently to read back the results.
+ *  Note that it is a mistake to call ra_psend() when
+ *  ray_pnidle is zero, and nothing will be sent in
+ *  this case.  Otherwise, the ray_presult() and/or ray_pqueue()
+ *  functions may be called subsequently to read back the results
+ *  of rays queued by ray_psend().
  *
  *  When you are done, you may call ray_pdone(1) to close
  *  all child processes and clean up memory used by Radiance.
  *  Any queued ray calculations will be awaited and discarded.
  *  As with ray_done(), ray_pdone(0) hangs onto data files
  *  and fonts that are likely to be used in subsequent renderings.
- *  Whether you want to bother cleaning up memory or not, you
- *  should at least call ray_pclose(0) to clean the child processes.
+ *  Whether you need to clean up memory or not, you should
+ *  at least call ray_pclose(0) to await the child processes.
+ *  The caller should define a quit() function that calls
+ *  ray_pclose(0) if ray_pnprocs > 0.
  *
  *  Warning:  You cannot affect any of the rendering processes
  *  by changing global parameter values onece ray_pinit() has
@@ -124,7 +132,7 @@ static const char	RCSid[] = "$Id: raypcalls.c,v 2.19 2
  *  returning a negative value from ray_pqueue() or
  *  ray_presult().  If you get a negative value from either
  *  of these calls, you can assume that the processes have
- *  been cleaned up with a call to ray_close(), though you
+ *  been cleaned up with a call to ray_pclose(), though you
  *  will have to call ray_pdone() yourself if you want to
  *  free memory.  Obviously, you cannot continue rendering
  *  without risking further errors, but otherwise your
@@ -160,13 +168,13 @@ static struct child_proc {
 	int	fd_send;			/* write to child here */
 	int	fd_recv;			/* read from child here */
 	int	npending;			/* # rays in process */
-	unsigned long  rno[RAYQLEN];		/* working on these rays */
+	RNUMBER	rno[RAYQLEN];			/* working on these rays */
 } r_proc[MAX_NPROCS];			/* our child processes */
 
 static RAY	r_queue[2*RAYQLEN];	/* ray i/o buffer */
-static int	r_send_next;		/* next send ray placement */
-static int	r_recv_first;		/* position of first unreported ray */
-static int	r_recv_next;		/* next receive ray placement */
+static int	r_send_next = 0;	/* next send ray placement */
+static int	r_recv_first = RAYQLEN;	/* position of first unreported ray */
+static int	r_recv_next = RAYQLEN;	/* next received ray placement */
 
 #define sendq_full()	(r_send_next >= RAYQLEN)
 
@@ -174,7 +182,7 @@ static int ray_pflush(void);
 static void ray_pchild(int fd_in, int fd_out);
 
 
-extern void
+void
 ray_pinit(		/* initialize ray-tracing processes */
 	char	*otnm,
 	int	nproc
@@ -185,15 +193,6 @@ ray_pinit(		/* initialize ray-tracing processes */
 
 	ray_init(otnm);			/* load the shared scene */
 
-	preload_objs();			/* preload auxiliary data */
-
-					/* set shared memory boundary */
-	shm_boundary = (char *)malloc(16);
-	strcpy(shm_boundary, "SHM_BOUNDARY");
-
-	r_send_next = 0;		/* set up queue */
-	r_recv_first = r_recv_next = RAYQLEN;
-
 	ray_popen(nproc);		/* fork children */
 }
 
@@ -227,28 +226,31 @@ ray_pflush(void)			/* send queued rays to idle childre
 		ray_pnidle--;		/* now she's busy */
 	}
 	if (sfirst != r_send_next)
-		error(CONSISTENCY, "code screwup in ray_pflush");
+		error(CONSISTENCY, "code screwup in ray_pflush()");
 	r_send_next = 0;
 	return(sfirst);			/* return total # sent */
 }
 
 
-extern void
+int
 ray_psend(			/* add a ray to our send queue */
 	RAY	*r
 )
 {
-	if (r == NULL)
-		return;
+	int	rv;
+
+	if ((r == NULL) | (ray_pnidle <= 0))
+		return(0);
 					/* flush output if necessary */
-	if (sendq_full() && ray_pflush() <= 0)
-		error(INTERNAL, "ray_pflush failed in ray_psend");
+	if (sendq_full() && (rv = ray_pflush()) <= 0)
+		return(rv);
 
 	r_queue[r_send_next++] = *r;
+	return(1);
 }
 
 
-extern int
+int
 ray_pqueue(			/* queue a ray for computation */
 	RAY	*r
 )
@@ -263,7 +265,7 @@ ray_pqueue(			/* queue a ray for computation */
 			return(-1);
 					/* put new ray in queue */
 		r_queue[r_send_next++] = mySend;
-				/* XXX r_send_next may now be > RAYQLEN */
+
 		return(1);
 	}
 					/* else add ray to send queue */
@@ -277,7 +279,7 @@ ray_pqueue(			/* queue a ray for computation */
 }
 
 
-extern int
+int
 ray_presult(		/* check for a completed ray */
 	RAY	*r,
 	int	poll
@@ -295,6 +297,9 @@ ray_presult(		/* check for a completed ray */
 		*r = r_queue[r_recv_first++];
 		return(1);
 	}
+	if (poll < 0)			/* immediate polling mode? */
+		return(0);
+
 	n = ray_pnprocs - ray_pnidle;	/* pending before flush? */
 
 	if (ray_pflush() < 0)		/* send new rays to process */
@@ -314,7 +319,7 @@ getready:				/* any children waiting for us? */
 		if (FD_ISSET(r_proc[pn].fd_recv, &readset) ||
 				FD_ISSET(r_proc[pn].fd_recv, &errset))
 			break;
-					/* call select if we must */
+					/* call select() if we must */
 	if (pn < 0) {
 		FD_ZERO(&readset); FD_ZERO(&errset); n = 0;
 		for (pn = ray_pnprocs; pn--; ) {
@@ -329,7 +334,7 @@ getready:				/* any children waiting for us? */
 				poll ? &tpoll : (struct timeval *)NULL)) < 0)
 			if (errno != EINTR) {
 				error(WARNING,
-					"select call failed in ray_presult");
+					"select call failed in ray_presult()");
 				ray_pclose(0);
 				return(-1);
 			}
@@ -374,7 +379,7 @@ getready:				/* any children waiting for us? */
 }
 
 
-extern void
+void
 ray_pdone(		/* reap children and free data */
 	int	freall
 )
@@ -385,6 +390,7 @@ ray_pdone(		/* reap children and free data */
 		free((void *)shm_boundary);
 		shm_boundary = NULL;
 	}
+
 	ray_done(freall);		/* free rendering data */
 }
 
@@ -407,7 +413,7 @@ ray_pchild(	/* process rays (never returns) */
 					/* get smuggled set length */
 		n2 = sizeof(RAY)*r_queue[0].crtype - n;
 		if (n2 < 0)
-			error(INTERNAL, "buffer over-read in ray_pchild");
+			error(INTERNAL, "buffer over-read in ray_pchild()");
 		if (n2 > 0) {		/* read the rest of the set */
 			i = readbuf(fd_in, (char *)r_queue + n, n2);
 			if (i != n2)
@@ -421,6 +427,7 @@ ray_pchild(	/* process rays (never returns) */
 			r_queue[i].parent = NULL;
 			r_queue[i].clipset = NULL;
 			r_queue[i].slights = NULL;
+			r_queue[i].rlvl = 0;
 			samplendx++;
 			rayclear(&r_queue[i]);
 			rayvalue(&r_queue[i]);
@@ -428,16 +435,16 @@ ray_pchild(	/* process rays (never returns) */
 					/* write back our results */
 		i = writebuf(fd_out, (char *)r_queue, sizeof(RAY)*n);
 		if (i != sizeof(RAY)*n)
-			error(SYSTEM, "write error in ray_pchild");
+			error(SYSTEM, "write error in ray_pchild()");
 	}
 	if (n)
-		error(SYSTEM, "read error in ray_pchild");
+		error(SYSTEM, "read error in ray_pchild()");
 	ambsync();
 	quit(0);			/* normal exit */
 }
 
 
-extern void
+void
 ray_popen(			/* open the specified # processes */
 	int	nadd
 )
@@ -448,6 +455,12 @@ ray_popen(			/* open the specified # processes */
 	if (nadd <= 0)
 		return;
 	ambsync();			/* load any new ambient values */
+	if (shm_boundary == NULL) {	/* first child process? */
+		preload_objs();		/* preload auxiliary data */
+					/* set shared memory boundary */
+		shm_boundary = (char *)malloc(16);
+		strcpy(shm_boundary, "SHM_BOUNDARY");
+	}
 	fflush(NULL);			/* clear pending output */
 	while (nadd--) {		/* fork each new process */
 		int	p0[2], p1[2];
@@ -460,6 +473,7 @@ ray_popen(			/* open the specified # processes */
 				close(r_proc[pn].fd_recv);
 			}
 			close(p0[0]); close(p1[1]);
+			close(0);	/* don't share stdin */
 					/* following call never returns */
 			ray_pchild(p1[0], p0[1]);
 		}
@@ -481,7 +495,7 @@ ray_popen(			/* open the specified # processes */
 }
 
 
-extern void
+void
 ray_pclose(		/* close one or more child processes */
 	int	nsub
 )
@@ -492,20 +506,25 @@ ray_pclose(		/* close one or more child processes */
 	if (inclose)
 		return;
 	inclose++;
+					/* check no child / in child */
+	if (ray_pnprocs <= 0)
+		return;
 					/* check argument */
 	if ((nsub <= 0) | (nsub > ray_pnprocs))
 		nsub = ray_pnprocs;
 					/* clear our ray queue */
 	while (ray_presult(&res,0) > 0)
 		;
+	r_send_next = 0;		/* hard reset in case of error */
+	r_recv_first = r_recv_next = RAYQLEN;
 					/* clean up children */
 	while (nsub--) {
 		int	status;
 		ray_pnprocs--;
-		close(r_proc[ray_pnprocs].fd_recv);
 		close(r_proc[ray_pnprocs].fd_send);
 		if (waitpid(r_proc[ray_pnprocs].pid, &status, 0) < 0)
 			status = 127<<8;
+		close(r_proc[ray_pnprocs].fd_recv);
 		if (status) {
 			sprintf(errmsg,
 				"rendering process %d exited with code %d",
@@ -515,14 +534,4 @@ ray_pclose(		/* close one or more child processes */
 		ray_pnidle--;
 	}
 	inclose--;
-}
-
-
-void
-quit(ec)			/* make sure exit is called */
-int	ec;
-{
-	if (ray_pnprocs > 0)	/* close children if any */
-		ray_pclose(0);		
-	exit(ec);
 }