Added alltoallv test and optional args variable on collective args (#514)

* Added alltoallv test and optional args variable on collective args

test/AllReduce_GroupCall.cpp

@@ -16,11 +16,11 @@ namespace RcclUnitTesting
     std::vector<ncclDataType_t> const& dataTypes       = {ncclFloat};
     std::vector<ncclRedOp_t>    const& redOps          = {ncclSum};
     std::vector<int>            const  numElements     = {1048576, 53327, 1024};
-    int                         const  root            = 0;
     bool                        const  inPlace         = false;
     bool                        const  useManagedMem   = false;
     int                         const  numCollPerGroup = numElements.size();
 
+    OptionalColArgs options;
     // This tests runs 3 collectives in the same group call
     bool isCorrect = true;
     for (int totalRanks = testBed.ev.minGpus; totalRanks <= testBed.ev.maxGpus && isCorrect; ++totalRanks)
@@ -33,30 +33,32 @@ namespace RcclUnitTesting
       testBed.InitComms(TestBed::GetDeviceIdsList(numProcesses, totalRanks), numCollPerGroup);
 
       for (int redOpIdx = 0; redOpIdx < redOps.size() && isCorrect; ++redOpIdx)
-      for (int dataIdx = 0; dataIdx < dataTypes.size() && isCorrect; ++dataIdx)
       {
-        if (testBed.ev.showNames)
-          INFO("%s %d-ranks AllReduce %d Grouped Calls (%s-%s)\n",
-               isMultiProcess ? "MP" : "SP",
-               totalRanks, numCollPerGroup,
-               ncclRedOpNames[redOps[redOpIdx]], ncclDataTypeNames[dataTypes[dataIdx]]);
-
-        // Run all element sizes in parallel as single group
-        for (int collIdx = 0; collIdx < numCollPerGroup; ++collIdx)
+        options.redOp = redOps[redOpIdx];
+        for (int dataIdx = 0; dataIdx < dataTypes.size() && isCorrect; ++dataIdx)
         {
-          testBed.SetCollectiveArgs(funcType,
-                                    dataTypes[dataIdx],
-                                    redOps[redOpIdx],
-                                    root,
-                                    numElements[collIdx],
-                                    numElements[collIdx],
-                                    collIdx);
+          if (testBed.ev.showNames)
+            INFO("%s %d-ranks AllReduce %d Grouped Calls (%s-%s)\n",
+                 isMultiProcess ? "MP" : "SP",
+                 totalRanks, numCollPerGroup,
+                 ncclRedOpNames[redOps[redOpIdx]], ncclDataTypeNames[dataTypes[dataIdx]]);
+
+          // Run all element sizes in parallel as single group
+          for (int collIdx = 0; collIdx < numCollPerGroup; ++collIdx)
+          {
+            testBed.SetCollectiveArgs(funcType,
+                                      dataTypes[dataIdx],
+                                      numElements[collIdx],
+                                      numElements[collIdx],
+                                      options,
+                                      collIdx);
+          }
+          testBed.AllocateMem(inPlace, useManagedMem);
+          testBed.PrepareData();
+          testBed.ExecuteCollectives();
+          testBed.ValidateResults(isCorrect);
+          testBed.DeallocateMem();
         }
-        testBed.AllocateMem(inPlace, useManagedMem);
-        testBed.PrepareData();
-        testBed.ExecuteCollectives();
-        testBed.ValidateResults(isCorrect);
-        testBed.DeallocateMem();
       }
       testBed.DestroyComms();
     }