Merge pull request #3031 from verilog-to-routing/temp_lookahead_chan_cong

Congestion-Aware Initial Accumulated Cost

Author: soheilshahrouz

doc/src/vpr/command_line_usage.rst

@@ -1824,6 +1824,20 @@ The following options are only valid when the router is in timing-driven mode (t
 
      **Default:** ``map``
 
+.. option:: --router_initial_acc_cost_chan_congestion_threshold <float>
+
+    Utilization threshold above which initial accumulated routing cost (acc_cost) is increased to penalize congested channels.
+    Used to bias routing away from highly utilized regions during early routing iterations.
+
+    **Default:** ``0.5``
+
+.. option:: --router_initial_acc_cost_chan_congestion_weight <float>
+    Weight applied to the excess channel utilization (above threshold) when computing the initial accumulated cost (acc_cost)of routing resources.
+
+    Higher values make the router more sensitive to early congestion.
+
+    **Default:** ``0.5``
+
 .. option:: --router_max_convergence_count <float>
 
     Controls how many times the router is allowed to converge to a legal routing before halting.

libs/librrgraph/src/base/check_rr_graph.cpp

@@ -53,9 +53,9 @@ void check_rr_graph(const RRGraphView& rr_graph,
                     const t_chan_width& chan_width,
                     const e_graph_type graph_type,
                     bool is_flat) {
-    e_route_type route_type = DETAILED;
+    e_route_type route_type = e_route_type::DETAILED;
     if (graph_type == e_graph_type::GLOBAL) {
-        route_type = GLOBAL;
+        route_type = e_route_type::GLOBAL;
     }
 
     auto total_edges_to_node = std::vector<int>(rr_graph.num_nodes());
@@ -420,7 +420,7 @@ void check_rr_node(const RRGraphView& rr_graph,
                 VPR_FATAL_ERROR(VPR_ERROR_ROUTE,
                                 "in check_rr_node: CHANX out of range for endpoints (%d,%d) and (%d,%d)\n", xlow, ylow, xhigh, yhigh);
             }
-            if (route_type == GLOBAL && xlow != xhigh) {
+            if (route_type == e_route_type::GLOBAL && xlow != xhigh) {
                 VPR_ERROR(VPR_ERROR_ROUTE,
                           "in check_rr_node: node %d spans multiple channel segments (not allowed for global routing).\n", inode);
             }
@@ -431,7 +431,7 @@ void check_rr_node(const RRGraphView& rr_graph,
                 VPR_FATAL_ERROR(VPR_ERROR_ROUTE,
                                 "Error in check_rr_node: CHANY out of range for endpoints (%d,%d) and (%d,%d)\n", xlow, ylow, xhigh, yhigh);
             }
-            if (route_type == GLOBAL && ylow != yhigh) {
+            if (route_type == e_route_type::GLOBAL && ylow != yhigh) {
                 VPR_ERROR(VPR_ERROR_ROUTE,
                           "in check_rr_node: node %d spans multiple channel segments (not allowed for global routing).\n", inode);
             }
@@ -480,7 +480,7 @@ void check_rr_node(const RRGraphView& rr_graph,
 
         case e_rr_type::CHANX:
         case e_rr_type::CHANY:
-            if (route_type == DETAILED) {
+            if (route_type == e_route_type::DETAILED) {
                 nodes_per_chan = chan_width.max;
                 tracks_per_node = 1;
             } else {

libs/librrgraph/src/base/rr_graph_type.h

@@ -13,7 +13,8 @@ struct t_chan_width {
     std::vector<int> y_list;
 };
 
-enum e_route_type {
+/// @brief Specifies whether global routing or combined global and detailed routing is performed.
+enum class e_route_type {
     GLOBAL,
     DETAILED
 };

libs/librrgraph/src/base/rr_node_types.h

@@ -103,8 +103,8 @@ typedef vtr::Range<edge_idx_iterator> edge_idx_range;
 
 typedef std::vector<std::map<int, int>> t_arch_switch_fanin;
 
-/*
- * Resistance/Capacitance data for an RR Nodes
+/**
+ * @brief Resistance/Capacitance data for an RR Node.
  *
  * In practice many RR nodes have the same values, so they are fly-weighted
  * to keep t_rr_node small. Each RR node holds an rc_index which allows
@@ -121,8 +121,8 @@ typedef std::vector<std::map<int, int>> t_arch_switch_fanin;
 struct t_rr_rc_data {
     t_rr_rc_data(float Rval, float Cval) noexcept;
 
-    float R;
-    float C;
+    float R;    ///< Resistance to go through an RR node
+    float C;    ///<  Total capacitance of an RR node.
 };
 
 // This is the data type of fast lookups of an rr-node given an (rr_type, layer, x, y, and the side)

libs/librrgraph/src/base/rr_rc_data.h

@@ -2,8 +2,8 @@
 
 #include "rr_node_types.h"
 
-/*
- * Returns the index to a t_rr_rc_data matching the specified values.
+/**
+ * @brief Returns the index to a t_rr_rc_data matching the specified values.
  *
  * If an existing t_rr_rc_data matches the specified R/C it's index
  * is returned, otherwise the t_rr_rc_data is created.

libs/libvtrutil/src/vtr_assert.h

@@ -101,7 +101,7 @@
  *
  * Note that to avoid 'unused' variable warnings when assertions are
  * disabled, we pass the expr and msg to sizeof(). We use sizeof specifically
- * since it accepts expressions, and the C++ standard gaurentees sizeof's arguments
+ * since it accepts expressions, and the C++ standard guarantees sizeof's arguments
  * are never evaluated (ensuring any expensive expressions are not evaluated when
  * assertions are disabled). To avoid warnings about the unused result of sizeof()
  * we cast it to void.
@@ -140,7 +140,7 @@ namespace assert {
  * function will never return. This should ensure the
  * compiler won't warn about detected conditions such as
  * dead-code or potential null pointer dereferences
- * which are gaurded against by assertions.
+ * which are guarded against by assertions.
  */
 [[noreturn]] void handle_assert(const char* expr, const char* file, unsigned int line, const char* function, const char* msg);
 } // namespace assert

libs/libvtrutil/src/vtr_prefix_sum.h

@@ -123,6 +123,11 @@ class PrefixSum1D {
         return prefix_sum_[upper_x + 1] - prefix_sum_[lower_x];
     }
 
+    /// @brief Checks if the prefix sum is initialized or it is empty.
+    bool empty() const {
+        return prefix_sum_.empty();
+    }
+
   private:
     /**
      * @brief The 1D prefix sum of the original array of values.
@@ -268,6 +273,11 @@ class PrefixSum2D {
                + prefix_sum_[lower_x][lower_y];
     }
 
+    /// @brief Checks if the prefix sum is initialized or it is empty.
+    bool empty() const {
+        return prefix_sum_.empty();
+    }
+
   private:
     /**
      * @brief The 2D prefix sum of the original grid of values.

vpr/src/analytical_place/detailed_placer.cpp

@@ -101,7 +101,7 @@ void AnnealerDetailedPlacer::optimize_placement() {
     placer_->place();
 
     // Copy the placement solution into the global placement solution.
-    placer_->copy_locs_to_global_state(g_vpr_ctx.mutable_placement());
+    placer_->update_global_state();
 
     // Since the placement was modified, need to resynchronize the pins in the
     // clusters.

vpr/src/base/CheckSetup.cpp

@@ -34,8 +34,8 @@ void CheckSetup(const t_packer_opts& packer_opts,
         }
     }
 
-    if ((GLOBAL == router_opts.route_type)
-        && (placer_opts.place_algorithm.is_timing_driven())) {
+    if (e_route_type::GLOBAL == router_opts.route_type
+        && placer_opts.place_algorithm.is_timing_driven()) {
         /* Works, but very weird.  Can't optimize timing well, since you're
          * not doing proper architecture delay modelling. */
         VTR_LOG_WARN(
@@ -106,7 +106,7 @@ void CheckSetup(const t_packer_opts& packer_opts,
         }
     }
 
-    if (DETAILED == router_opts.route_type) {
+    if (e_route_type::DETAILED == router_opts.route_type) {
         if ((chans.chan_x_dist.type != UNIFORM)
             || (chans.chan_y_dist.type != UNIFORM)) {
             VPR_FATAL_ERROR(VPR_ERROR_OTHER,

vpr/src/base/SetupVPR.cpp

@@ -488,6 +488,9 @@ static void SetupRouterOpts(const t_options& Options, t_router_opts* RouterOpts)
     RouterOpts->router_debug_sink_rr = Options.router_debug_sink_rr;
     RouterOpts->router_debug_iteration = Options.router_debug_iteration;
     RouterOpts->lookahead_type = Options.router_lookahead_type;
+    RouterOpts->initial_acc_cost_chan_congestion_threshold = Options.router_initial_acc_cost_chan_congestion_threshold;
+    RouterOpts->initial_acc_cost_chan_congestion_weight = Options.router_initial_acc_cost_chan_congestion_weight;
+
     RouterOpts->max_convergence_count = Options.router_max_convergence_count;
     RouterOpts->reconvergence_cpd_threshold = Options.router_reconvergence_cpd_threshold;
     RouterOpts->initial_timing = Options.router_initial_timing;