include reduce into duration measurement

benchmarks/babelstream/src/babelStreamMainTest.cpp

@@ -414,7 +414,7 @@ void testKernels()
     std::vector<double> bytesReadWriteMB;
     if(kernelsToBeExecuted == KernelsToRun::All)
     {
-        // Verify the results for all 5 babelstream kernels
+        // Verify the results for init and 4 babelstream kernels
         // Find sum of the errors as sum of the differences from expected values
         DataType initVal{static_cast<DataType>(0.0)};
         DataType sumErrC{initVal}, sumErrB{initVal}, sumErrA{initVal};
@@ -453,6 +453,7 @@ void testKernels()
             // A specific work-division is used for dotKernel
             auto const workDivDot = WorkDiv{Vec{gridBlockExtent}, Vec{blockThreadExtent}, Vec::all(1)};
 
+            float result{0.0f};
             measureKernelExec(
                 [&]()
                 {
@@ -464,15 +465,13 @@ void testKernels()
                         bufAccInputBPtr,
                         alpaka::getPtrNative(bufAccSumPerBlock),
                         static_cast<alpaka::Idx<Acc>>(arraySize));
+                    alpaka::memcpy(queue, bufHostSumPerBlock, bufAccSumPerBlock, gridBlockExtent);
+                    alpaka::wait(queue);
+                    DataType const* sumPtr = std::data(bufHostSumPerBlock);
+                    result = std::reduce(sumPtr, sumPtr + gridBlockExtent, 0.0f);
                 },
                 "DotKernel");
 
-            alpaka::memcpy(queue, bufHostSumPerBlock, bufAccSumPerBlock, gridBlockExtent);
-            alpaka::wait(queue);
-
-            DataType const* sumPtr = std::data(bufHostSumPerBlock);
-            float const result = std::reduce(sumPtr, sumPtr + gridBlockExtent, 0.0f);
-
             float const expectedSum = static_cast<float>(static_cast<DataType>(arraySize) * expectedA * expectedB);
 
             //  Dot product should be identical to arraySize*valA*valB