[Router] Fix intra-cluster pin handling for tiles with heterogeneous subtiles

.gitignore

@@ -35,7 +35,7 @@ titan_release*.tar.gz
 vtr_flow/arch/titan/*.xml
 vtr_flow/benchmarks/titan_blif/other_benchmarks
 vtr_flow/benchmarks/titan_blif/titan23
-vtr_flow/benchmarks/titan_blif/titan_new
+vtr_flow/benchmarks/titan_blif/titanium
 
 
 #

vpr/src/route/rr_graph_generation/rr_graph_intra_cluster.cpp

@@ -12,6 +12,16 @@
 #include "rr_graph_switch_utils.h"
 #include "check_rr_graph.h"
 
+/**
+ * @brief Returns the list of pins (both cluster-level and intra-cluster-level) of the given cluster block.
+ * @param physical_tile The physical tile type that the cluster block is mapped to.
+ * @param cluster_blk_id The cluster block ID.
+ * @param sub_tile_index The sub-tile absolute index (in comparison to relative index which is the index among sub-tiles of the same type) in the physical tile type.
+ */
+static std::vector<int> get_cluster_block_pins(t_physical_tile_type_ptr physical_tile,
+                                               ClusterBlockId cluster_blk_id,
+                                               int sub_tile_index);
+
 static void set_clusters_pin_chains(const ClusteredNetlist& clb_nlist,
                                     vtr::vector<ClusterBlockId, t_cluster_pin_chain>& pin_chains,
                                     bool is_flat);
@@ -179,6 +189,53 @@ static void add_pin_chain(const std::vector<int>& pin_chain,
                           std::vector<std::vector<t_pin_chain_node>>& all_chains,
                           bool is_new_chain);
 
+static std::vector<int> get_cluster_block_pins(t_physical_tile_type_ptr physical_tile,
+                                               ClusterBlockId cluster_blk_id,
+                                               int sub_tile_index) {
+    // To get the list of pins, we first add the pins on the tile-level, then add the pins on the intra-tile-level.
+
+    // A counter to keep track of number of tile-level pins for the sub-tiles before the current sub-tile.
+    int seen_sub_tiles_num_cluser_pins = 0;
+    // The number of tile-level pins for the sub-tile instance that the cluster block is mapped to.
+    int cluster_sub_tile_inst_num_pins;
+    // A flag to check if the sub-tile instance that the cluster block is mapped to has been found.
+    bool found_sub_tile = false;
+
+    // Iterate over all the sub-tiles to find the sub-tile instance that the cluster block is mapped to.
+    for (const t_sub_tile& sub_tile: physical_tile->sub_tiles) {
+        if (sub_tile.capacity.is_in_range(sub_tile_index)) {
+            // This sub-tile type is the one that the cluster block is mapped to.
+            found_sub_tile = true;
+            // The number of tile-level pins for all isntances of the same sub-tile type is the same. Thus,
+            // we can the the number of tile-level pins for the sub-tile instance by dividing the total number of pins
+            // for the given sub-tile type by the number of instances.
+            cluster_sub_tile_inst_num_pins = (sub_tile.num_phy_pins / sub_tile.capacity.total());
+            int rel_cap = sub_tile_index - sub_tile.capacity.low;
+            // Add the number of tile-level pins for the instances before the current sub-tile instance to the counter.
+            seen_sub_tiles_num_cluser_pins += rel_cap * cluster_sub_tile_inst_num_pins;
+            break;
+        } else {
+            // This sub-tile type is not the one that the cluster block is mapped to.
+            // Add the number of tile-level pins for this sub-tile type to the counter
+            // and continue to the next sub-tile type.
+            seen_sub_tiles_num_cluser_pins += sub_tile.num_phy_pins;
+        }
+    }
+
+    VTR_ASSERT(found_sub_tile);
+    std::vector<int> pin_num_vec(cluster_sub_tile_inst_num_pins);
+    // Pin numbers are assigned such that each instance’s tile-level pins
+    // occupy a continuous range equal to the total number of tile-level pins for that instance.
+    // Since we know the starting index, we use std::iota to generate that range.
+    std::iota(pin_num_vec.begin(), pin_num_vec.end(), seen_sub_tiles_num_cluser_pins);
+
+    // Add the intra-cluster-level pins to the list.
+    std::vector<int> internal_pins = get_cluster_internal_pins(cluster_blk_id);
+    pin_num_vec.insert(pin_num_vec.end(), internal_pins.begin(), internal_pins.end());
+
+    return pin_num_vec;
+}
+
 static void set_clusters_pin_chains(const ClusteredNetlist& clb_nlist,
                                     vtr::vector<ClusterBlockId, t_cluster_pin_chain>& pin_chains,
                                     bool is_flat) {

vpr/src/util/vpr_utils.cpp

@@ -1906,23 +1906,6 @@ std::vector<int> get_cluster_netlist_intra_tile_pins_at_loc(const t_physical_til
     return pin_num_vec;
 }
 
-std::vector<int> get_cluster_block_pins(t_physical_tile_type_ptr physical_tile,
-                                        ClusterBlockId cluster_blk_id,
-                                        int abs_cap) {
-    int max_num_pin = get_tile_total_num_pin(physical_tile) / physical_tile->capacity;
-    int num_tile_pin_per_inst = physical_tile->num_pins / physical_tile->capacity;
-    std::vector<int> pin_num_vec(num_tile_pin_per_inst);
-    std::iota(pin_num_vec.begin(), pin_num_vec.end(), abs_cap * num_tile_pin_per_inst);
-
-    pin_num_vec.reserve(max_num_pin);
-
-    auto internal_pins = get_cluster_internal_pins(cluster_blk_id);
-    pin_num_vec.insert(pin_num_vec.end(), internal_pins.begin(), internal_pins.end());
-
-    pin_num_vec.shrink_to_fit();
-    return pin_num_vec;
-}
-
 t_arch_switch_inf create_internal_arch_sw(float delay) {
     t_arch_switch_inf arch_switch_inf;
     arch_switch_inf.set_type(e_switch_type::MUX);

vpr/src/util/vpr_utils.h

@@ -329,10 +329,6 @@ std::vector<int> get_cluster_netlist_intra_tile_pins_at_loc(const t_physical_til
                                                             const vtr::vector<ClusterBlockId, std::unordered_set<int>>& pin_chains_num,
                                                             t_physical_tile_type_ptr physical_type);
 
-std::vector<int> get_cluster_block_pins(t_physical_tile_type_ptr physical_tile,
-                                        ClusterBlockId cluster_blk_id,
-                                        int abs_cap);
-
 t_arch_switch_inf create_internal_arch_sw(float delay);
 
 void add_pb_child_to_list(std::list<const t_pb*>& pb_list, const t_pb* parent_pb);