bug fixes

source/train/XWorkerCollect.cpp

@@ -58,6 +58,9 @@ void XWorkerCollect::CollectData(XList * sourceList, XModel * target, long sleep
 {
     TensorList & tp = target->params;
     int finished = 0;
+    
+    for (int j = 0; j < tp.count; j++)
+        target->flags[j] = PARAM_STATE_NOT_READY;
 
     /* check */
     for (int i = 0; i < sourceList->count; i++) {
@@ -73,9 +76,9 @@ void XWorkerCollect::CollectData(XList * sourceList, XModel * target, long sleep
         
         if (collectMode == DATA_COLLECT_P2P) {
             for (int j = 0; j < tp.count; j++) {
-                /* target->flags[j] is ready only if model finishes the computation
+                /* tp[j]->isGradFinished is true only if the model finishes the computation
                 (in another process) */
-                if (target->flags[j] != PARAM_STATE_READY)
+                if (target->flags[j] == PARAM_STATE_COLLECTED || !tp[j]->isGradFinished)
                     continue;
 
                 /* check if all the models (or part of them) are ready */
@@ -83,9 +86,9 @@ void XWorkerCollect::CollectData(XList * sourceList, XModel * target, long sleep
                     XModel * source = (XModel*)sourceList->GetItem(i);
                     TensorList & sp = source->params;
 
-                    /* source->flags[j] is ready only if model finishes the computation
+                    /* sp[j]->isGradFinished is true only if the model finishes the computation
                     (in another process) */
-                    if (source->flags[j] == PARAM_STATE_READY) {
+                    if (source->flags[j] != PARAM_STATE_COLLECTED && sp[j]->isGradFinished) {
 
                         /* data transmit */
                         CollectP2P(sp.GetItem(j), tp.GetItem(j));
@@ -103,9 +106,9 @@ void XWorkerCollect::CollectData(XList * sourceList, XModel * target, long sleep
 
                 bool ready = true;
 
-                /* target->flags[j] is ready only if model finishes the computation
-                   (in another process) */
-                if (target->flags[j] != PARAM_STATE_READY)
+                /* tp[j]->isGradFinished is true only if the model finishes the computation
+                 (in another process) */
+                if (target->flags[j] == PARAM_STATE_COLLECTED || !tp[j]->isGradFinished)
                     continue;
 
                 /* check if all the models (or part of them) are ready */
@@ -113,12 +116,16 @@ void XWorkerCollect::CollectData(XList * sourceList, XModel * target, long sleep
                     XModel * source = (XModel*)sourceList->GetItem(i);
                     TensorList & sp = source->params;
                     
-                    /* source->flags[j] is ready only if model finishes the computation
-                      (in another process) */
-                    if (source->flags[j] != PARAM_STATE_READY) {
+                    /* sp[j]->isGradFinished is true only if the model finishes the computation
+                     (in another process) */
+                    if (source->flags[j] == PARAM_STATE_COLLECTED || !sp[j]->isGradFinished) {
                         ready = false;
                         break;
                     }
+                    else if(source->flags[j] == PARAM_STATE_NOT_READY){
+                        source->flags[j] = PARAM_STATE_READY;
+                    }
+                        
                 }
 
                 if (ready) {
@@ -152,14 +159,13 @@ void XWorkerCollect::CollectData(XList * sourceList, XModel * target, long sleep
             break;
 
         /* reset the flags */
-        for (int i = 0; i < tp.count; i++) {
-            target->flags[i] = PARAM_STATE_COLLECTED;
-        }
+        for (int j = 0; j < tp.count; j++)
+            target->flags[j] = PARAM_STATE_COLLECTED;
 
 #ifdef _WIN32
         Sleep((DWORD)sleepTime);
 #else
-        sleep(sleepTime / 1000);
+        sleep((unsigned)sleepTime / 1000);
 #endif
     }
 }
@@ -230,4 +236,4 @@ void XWorkerCollect::CollectAllReduce(XList * all)
     ShowNTErrors("TODO!");
 }
 
-}
\ No newline at end of file
+}

source/train/XWorkerJob.cpp

@@ -89,7 +89,7 @@ bool XWorkerJob::AddJobRefresh(XModel * myModel)
     XList args(1);
     args.Add(myModel);
 
-    queue.EnqueueJob(XModel::Refresh, &args);
+    queue.EnqueueJob((void*)(char*)XModel::Refresh, &args);
 
     return true;
 }
@@ -112,7 +112,7 @@ bool XWorkerJob::AddJobNeuralNet(XModel * myModel, XList * inputs, XList * outpu
     args.AddList(inputs);
     args.AddList(outputs);
 
-    queue.EnqueueJob(XModel::Run, &args);
+    queue.EnqueueJob((void*)(char*)XModel::Run, &args);
 
     return true;
 }