Index: src/CUDAVolCPotential.cu
===================================================================
RCS file: /vmd/cvsroot/vmd/src/CUDAVolCPotential.cu,v
retrieving revision 1.26
diff -u -r1.26 CUDAVolCPotential.cu
--- src/CUDAVolCPotential.cu	27 Mar 2008 19:37:42 -0000	1.26
+++ src/CUDAVolCPotential.cu	28 Jan 2009 12:24:27 -0000
@@ -29,7 +29,6 @@
 
 typedef struct {
   int threadid;
-  int threadcount;
   float* atoms;
   float* grideners;
   long int numplane;
@@ -37,6 +36,8 @@
   long int numpt;
   long int natoms;
   float gridspacing;
+  vmd_mutex_t *loopmutex;
+  int *loopindex;
 } enthrparms;
 
 /* thread prototype */
@@ -226,6 +227,8 @@
   int i;
   enthrparms *parms;
   vmd_thread_t * threads;
+  int loopindex = 0; // global loop index for all threads
+  vmd_mutex_t loopmutex; // mutex protecting global loop index
   vmd_timerhandle globaltimer;
   double totalruntime;
 
@@ -252,11 +255,12 @@
   /* allocate array of threads */
   threads = (vmd_thread_t *) calloc(numprocs * sizeof(vmd_thread_t), 1);
 
+  vmd_mutex_init(&loopmutex);
+
   /* allocate and initialize array of thread parameters */
   parms = (enthrparms *) malloc(numprocs * sizeof(enthrparms));
   for (i=0; i<numprocs; i++) {
     parms[i].threadid = i;
-    parms[i].threadcount = numprocs;
     parms[i].atoms = atoms;
     parms[i].grideners = grideners;
     parms[i].numplane = numplane;
@@ -264,6 +268,8 @@
     parms[i].numpt = numpt;
     parms[i].natoms = natoms;
     parms[i].gridspacing = gridspacing;
+    parms[i].loopmutex = &loopmutex;
+    parms[i].loopindex = &loopindex;
   }
 
   printf("GPU padded grid size: %d x %d x %d\n", 
@@ -291,6 +297,8 @@
   cudaenergythread((void *) &parms[0]);
 #endif
 
+  vmd_mutex_destroy(&loopmutex);
+
   // Measure GFLOPS
   vmd_timer_stop(globaltimer);
   totalruntime = vmd_timer_time(globaltimer);
@@ -330,7 +338,8 @@
   const long int natoms = parms->natoms;
   const float gridspacing = parms->gridspacing;
   const int threadid = parms->threadid;
-  const int threadcount = parms->threadcount;
+  vmd_mutex_t *loopmutex = parms->loopmutex;
+  int *i = parms->loopindex;
   double lasttime, totaltime;
 
   cudaSetDevice(threadid);
@@ -389,8 +398,12 @@
 
   // For each point in the cube...
   int iterations=0;
-  int k;
-  for (k=threadid; k<numplane; k+= threadcount) {
+  int computedplanes=0;
+  vmd_mutex_lock(loopmutex);
+  for (; *i<numplane; (*i)++) {
+    vmd_mutex_unlock(loopmutex);
+    computedplanes++;
+    int k = *i;
     int y;
     int atomstart;
     float zplane = k * (float) gridspacing;
@@ -436,13 +449,18 @@
  
     totaltime = vmd_timer_timenow(timer);
     if (msg_timer_timeout(msgt)) {
+       vmd_mutex_lock(loopmutex);
+       k = *i;
+       vmd_mutex_unlock(loopmutex);
       // XXX: we have to use printf here as msgInfo is not thread-safe yet.
-      printf("thread[%d] plane %d/%ld time %.2f, elapsed %.1f, est. total: %.1f\n",
-             threadid, k, numplane,
+      printf("thread[%d] plane %d/%ld (%d computed) time %.2f, elapsed %.1f, est. total: %.1f\n",
+             threadid, k, numplane, computedplanes,
              totaltime - lasttime, totaltime,
              totaltime * numplane / (k+1));
     }
+    vmd_mutex_lock(loopmutex);
   }
+  vmd_mutex_unlock(loopmutex);
 
   vmd_timer_destroy(timer); // free timer
   msg_timer_destroy(msgt); // free timer