use hardware path instead of automatic for cache

offloading-cacher/cache.hpp

@@ -455,12 +455,12 @@ inline void dsacache::Cache::SubmitTask(CacheData* task, const int dst_node, con
 inline dml::handler<dml::mem_copy_operation, std::allocator<uint8_t>> dsacache::Cache::ExecuteCopy(
         const uint8_t* src, uint8_t* dst, const size_t size, const int node
 ) const {
+    numa_run_on_node(node);
+
     dml::const_data_view srcv = dml::make_view(src, size);
     dml::data_view dstv = dml::make_view(dst, size);
 
-    numa_run_on_node(node);
-
-    return dml::submit<dml::automatic>(dml::mem_copy.block_on_fault(), srcv, dstv);
+    return dml::submit<dml::hardware>(dml::mem_copy.block_on_fault(), srcv, dstv);
 }
 
 inline void dsacache::Cache::GetCacheNode(uint8_t* src, const size_t size, int* OUT_DST_NODE, int* OUT_SRC_NODE) const {