re-implemented the free method

Author: pletzer

src/DistDataCollector.h

@@ -157,11 +157,11 @@ class DistDataCollector {
     /**
      * @brief Free the MPI window and empty the collected data
      */
-    void free() { // This should be removed
-        // if (this->win != MPI_WIN_NULL) {
-        //     MPI_Win_free(&this->win);
-        // }
-        // No need to free the data, MPI_Win_free will free the pointer
+    void free() { // Should this be removed and just implemented in the destructor?
+        if (this->win != MPI_WIN_NULL) {
+            MPI_Win_free(&this->win);
+        }
+        //No need to free the data, MPI_Win_free will free the pointer
         //MPI_Free_mem(this->collectedData);
     }
 

tests/testAsyncPutGet.cxx

@@ -141,14 +141,20 @@ void testPutGet(int num_chunks, int num_size, int ms) {
     // make sure data has been initialized before remote memory accesses are issued
     MPI_Barrier(MPI_COMM_WORLD);
 
+    std::vector<double> localData(num_size);
+    dataCollector1.fence();
     if (rank > 0) {
         int chunk_id = rank;
-        std::vector<double> localData(num_size);
-
-        dataCollector1.get(chunk_id, localData.data());
+        dataCollector1.getAsync(chunk_id, localData.data());
+    }
+    dataCollector1.fence();
 
-        dataCollector2.put(chunk_id, localData.data());
+    dataCollector2.fence();
+    if (rank > 0) {
+        int chunk_id = rank;
+        dataCollector2.putAsync(chunk_id, localData.data());
     }
+    dataCollector2.fence();
 
     MPI_Barrier(MPI_COMM_WORLD);
 
@@ -161,7 +167,7 @@ void testPutGet(int num_chunks, int num_size, int ms) {
         for (auto i = 0; i < num_chunks; ++i) {
 
             // fetch a chunk asynchronously
-            dataCollector2.get(i, localData.data());
+            dataCollector2.getAsync(i, localData.data());
 
             // ... do some work ...
             std::this_thread::sleep_for(std::chrono::milliseconds(ms));

tests/testDistDataCollector.cxx

@@ -33,7 +33,7 @@ void test(int numSize, int numChunksPerRank) {
     }
 
     // Make sure all the chunks have been received. All the above "put" operations
-    // must have completed. Now the data cane be read.
+    // must have completed. Now the data can be read.
     ddc.fence();
 
     double toc = MPI_Wtime();
@@ -49,19 +49,17 @@ void test(int numSize, int numChunksPerRank) {
             std::vector<double> localData = ddc.get(chunkId);
 
             for (auto val : localData) {
-                //assert(val == chunkId); // this sometimes fails
                 if (val != chunkId) {
                     std::cout << "rank = " << rank << " val = " << val << " should have been " << chunkId << '\n';
                 }
+                assert(val == chunkId);
             }
         }
 
         double toc = MPI_Wtime();
         timeGet += toc - tic;
     }
 
-    MPI_Barrier(MPI_COMM_WORLD);
-
     if (rank == 0) {
         int numChunk = ddc.getNumChunks();
         int numSize = ddc.getNumSize();
@@ -98,7 +96,7 @@ int main(int argc, char* argv[]) {
 
     // Parse the command line arguments
     CmdLineArgParser cmdLine;
-    cmdLine.set("-numSize", 15000, "Size of the data to put/get");
+    cmdLine.set("-numSize", 100000, "Size of the data to put/get");
     cmdLine.set("-numChunksPerRank", 1, "Number of chunks per rank");
     bool success = cmdLine.parse(argc, argv);
     bool help = cmdLine.get<bool>("-help") || cmdLine.get<bool>("-h");