Merge pull request verilog-to-routing#2493 from verilog-to-routing/pe…

…r_iteration

Parallelize cost updates & wirelength computation

vpr/src/route/route_common.cpp

@@ -190,6 +190,28 @@ void pathfinder_update_acc_cost_and_overuse_info(float acc_fac, OveruseInfo& ove
     auto& device_ctx = g_vpr_ctx.device();
     const auto& rr_graph = device_ctx.rr_graph;
     auto& route_ctx = g_vpr_ctx.mutable_routing();
+
+#ifdef VPR_USE_TBB
+    tbb::combinable<size_t> overused_nodes(0), total_overuse(0), worst_overuse(0);
+    tbb::parallel_for_each(rr_graph.nodes().begin(), rr_graph.nodes().end(), [&](RRNodeId rr_id) {
+        int overuse = route_ctx.rr_node_route_inf[rr_id].occ() - rr_graph.node_capacity(rr_id);
+
+        // If overused, update the acc_cost and add this node to the overuse info
+        // If not, do nothing
+        if (overuse > 0) {
+            route_ctx.rr_node_route_inf[rr_id].acc_cost += overuse * acc_fac;
+
+            ++overused_nodes.local();
+            total_overuse.local() += overuse;
+            worst_overuse.local() = std::max(worst_overuse.local(), size_t(overuse));
+        }
+    });
+
+    // Update overuse info
+    overuse_info.overused_nodes = overused_nodes.combine(std::plus<size_t>());
+    overuse_info.total_overuse = total_overuse.combine(std::plus<size_t>());
+    overuse_info.worst_overuse = worst_overuse.combine([](size_t a, size_t b) { return std::max(a, b); });
+#else
     size_t overused_nodes = 0, total_overuse = 0, worst_overuse = 0;
 
     for (const RRNodeId& rr_id : rr_graph.nodes()) {
@@ -210,6 +232,7 @@ void pathfinder_update_acc_cost_and_overuse_info(float acc_fac, OveruseInfo& ove
     overuse_info.overused_nodes = overused_nodes;
     overuse_info.total_overuse = total_overuse;
     overuse_info.worst_overuse = worst_overuse;
+#endif
 }
 
 /** Update pathfinder cost of all nodes rooted at rt_node, including rt_node itself */

vpr/src/route/route_net.cpp

@@ -251,6 +251,23 @@ WirelengthInfo calculate_wirelength_info(const Netlist<>& net_list, size_t avail
 
     auto& route_ctx = g_vpr_ctx.routing();
 
+#ifdef VPR_USE_TBB
+    tbb::combinable<size_t> thread_used_wirelength(0);
+
+    tbb::parallel_for_each(net_list.nets().begin(), net_list.nets().end(), [&](ParentNetId net_id) {
+        if (!net_list.net_is_ignored(net_id)
+            && net_list.net_sinks(net_id).size() != 0 /* Globals don't count. */
+            && route_ctx.route_trees[net_id]) {
+            int bends, wirelength, segments;
+            bool is_absorbed;
+            get_num_bends_and_length(net_id, &bends, &wirelength, &segments, &is_absorbed);
+
+            thread_used_wirelength.local() += wirelength;
+        }
+    });
+
+    used_wirelength = thread_used_wirelength.combine(std::plus<size_t>());
+#else
     for (auto net_id : net_list.nets()) {
         if (!net_list.net_is_ignored(net_id)
             && net_list.net_sinks(net_id).size() != 0 /* Globals don't count. */
@@ -262,6 +279,7 @@ WirelengthInfo calculate_wirelength_info(const Netlist<>& net_list, size_t avail
             used_wirelength += wirelength;
         }
     }
+#endif
 
     return WirelengthInfo(available_wirelength, used_wirelength);
 }