resize source and destination pointer holders properly before use and use path from template and not dml::software for cache flush in benchmark loop

benchmarks/benchmark.cpp

@@ -71,6 +71,9 @@ void execute_dml_memcpy(std::vector<TaskData>& args, const uint64_t iterations)
 
     // initialize data fields for use
 
+    SOURCE_.resize(args.size());
+    DESTINATION_.resize(args.size());
+
     for (uint32_t tid = 0; tid < args.size(); tid++) {
         SOURCE_[tid] = numa_alloc_onnode(args[tid].size, args[tid].nnode_src);
         DESTINATION_[tid] = numa_alloc_onnode(args[tid].size, args[tid].nnode_dst);
@@ -90,8 +93,8 @@ void execute_dml_memcpy(std::vector<TaskData>& args, const uint64_t iterations)
 
             dml::data_view srcv = dml::make_view(reinterpret_cast<uint8_t*>(SOURCE_[tid]), args[tid].size);
             dml::data_view dstv = dml::make_view(reinterpret_cast<uint8_t*>(DESTINATION_[tid]), args[tid].size);
-            auto rsrc = dml::execute<dml::software>(dml::cache_flush, srcv);
-            auto rdst = dml::execute<dml::software>(dml::cache_flush, dstv);
+            auto rsrc = dml::execute<path>(dml::cache_flush, srcv);
+            auto rdst = dml::execute<path>(dml::cache_flush, dstv);
             TaskData* task = &args[tid];
             CHECK_STATUS(rsrc.status, "Flushing Cache for Source failed!");
             CHECK_STATUS(rdst.status, "Flushing Cache for Destination failed!");