Change GreedySeedSelector to work with molecules instead of atoms

Author: AmirhosseinPoolad

vpr/src/pack/greedy_clusterer.cpp

@@ -144,8 +144,7 @@ GreedyClusterer::do_clustering(ClusterLegalizer& cluster_legalizer,
                                      pre_cluster_timing_manager_);
 
     // Pick the first seed molecule.
-    PackMoleculeId seed_mol_id = seed_selector.get_next_seed(prepacker,
-                                                             cluster_legalizer);
+    PackMoleculeId seed_mol_id = seed_selector.get_next_seed(cluster_legalizer);
 
     /****************************************************************
      * Clustering
@@ -213,8 +212,7 @@ GreedyClusterer::do_clustering(ClusterLegalizer& cluster_legalizer,
                           cluster_legalizer);
 
         // Pick new seed.
-        seed_mol_id = seed_selector.get_next_seed(prepacker,
-                                                  cluster_legalizer);
+        seed_mol_id = seed_selector.get_next_seed(cluster_legalizer);
     }
 
     // If this architecture has LE physical block, report its usage.

vpr/src/pack/greedy_seed_selector.cpp

@@ -9,6 +9,8 @@
 
 #include <algorithm>
 #include <cmath>
+#include <limits>
+#include <vector>
 #include "PreClusterTimingManager.h"
 #include "atom_netlist.h"
 #include "cluster_legalizer.h"
@@ -129,11 +131,12 @@ static inline float get_seed_gain(AtomBlockId blk_id,
  *        criticalities.
  */
 static inline void print_seed_gains(const char* fname,
-                                    const std::vector<AtomBlockId>& seed_atoms,
-                                    const vtr::vector<AtomBlockId, float>& atom_gain,
+                                    const std::vector<PackMoleculeId>& seed_mols,
+                                    const vtr::vector<PackMoleculeId, float>& molecule_gain,
                                     const vtr::vector<AtomBlockId, float>& atom_criticality,
                                     const AtomNetlist& atom_netlist,
-                                    const LogicalModels& models) {
+                                    const LogicalModels& models,
+                                    const Prepacker& prepacker) {
     FILE* fp = vtr::fopen(fname, "w");
 
     // For pretty formatting determine the maximum name length
@@ -148,16 +151,18 @@ static inline void print_seed_gains(const char* fname,
 
     fprintf(fp, "%-*s %-*s %8s %8s\n", max_name_len, "atom_block_name", max_type_len, "atom_block_type", "gain", "criticality");
     fprintf(fp, "\n");
-    for (auto blk_id : seed_atoms) {
-        std::string name = atom_netlist.block_name(blk_id);
-        fprintf(fp, "%-*s ", max_name_len, name.c_str());
+    for (auto mol_id : seed_mols) {
+        for (AtomBlockId blk_id : prepacker.get_molecule(mol_id).atom_block_ids) {
+            std::string name = atom_netlist.block_name(blk_id);
+            fprintf(fp, "%-*s ", max_name_len, name.c_str());
 
-        std::string model_name = models.model_name(atom_netlist.block_model(blk_id));
-        fprintf(fp, "%-*s ", max_type_len, model_name.c_str());
+            std::string model_name = models.model_name(atom_netlist.block_model(blk_id));
+            fprintf(fp, "%-*s ", max_type_len, model_name.c_str());
 
-        fprintf(fp, "%*f ", std::max((int)strlen("gain"), 8), atom_gain[blk_id]);
-        fprintf(fp, "%*f ", std::max((int)strlen("criticality"), 8), atom_criticality[blk_id]);
-        fprintf(fp, "\n");
+            fprintf(fp, "%*f ", std::max((int)strlen("gain"), 8), molecule_gain[mol_id]);
+            fprintf(fp, "%*f ", std::max((int)strlen("criticality"), 8), atom_criticality[blk_id]);
+            fprintf(fp, "\n");
+        }
     }
 
     fclose(fp);
@@ -169,8 +174,8 @@ GreedySeedSelector::GreedySeedSelector(const AtomNetlist& atom_netlist,
                                        const t_molecule_stats& max_molecule_stats,
                                        const LogicalModels& models,
                                        const PreClusterTimingManager& pre_cluster_timing_manager)
-    : seed_atoms_(atom_netlist.blocks().begin(), atom_netlist.blocks().end()) {
-    // Seed atoms list is initialized with all atoms in the atom netlist.
+    : seed_mols_(prepacker.molecules().begin(), prepacker.molecules().end()) {
+    // Seed molecule list is initialized with all molecule in the netlist.
 
     // Pre-compute the criticality of each atom
     // Default criticalities set to zero (e.g. if not timing driven)
@@ -183,20 +188,31 @@ GreedySeedSelector::GreedySeedSelector(const AtomNetlist& atom_netlist,
         }
     }
 
-    // Maintain a lookup table of the seed gain for each atom. This will be
-    // used to sort the seed atoms.
+    // Maintain a lookup table of the seed gain for each molecule. This will be
+    // used to sort the seed molecules.
     // Initially all gains are zero.
-    vtr::vector<AtomBlockId, float> atom_gains(atom_netlist.blocks().size(), 0.f);
-
-    // Get the seed gain of each atom.
-    for (AtomBlockId blk_id : atom_netlist.blocks()) {
-        atom_gains[blk_id] = get_seed_gain(blk_id,
-                                           atom_netlist,
-                                           prepacker,
-                                           models,
-                                           seed_type,
-                                           max_molecule_stats,
-                                           atom_criticality);
+    vtr::vector<PackMoleculeId, float> molecule_gains(seed_mols_.size(), 0.f);
+
+    // Get the seed gain of each molecule.
+    for (PackMoleculeId mol_id : seed_mols_) {
+        // Gain of each molecule is the maximum gain of its atoms
+        float mol_gain = std::numeric_limits<float>::lowest();
+        const std::vector<AtomBlockId>& molecule_atoms = prepacker.get_molecule(mol_id).atom_block_ids;
+        for (AtomBlockId blk_id : molecule_atoms) {
+            // If the molecule does not fit the entire pack pattern, it's possible to have invalid block ids in the molecule_atoms vector
+            if (blk_id == AtomBlockId::INVALID()) {
+                continue;
+            }
+            float atom_gain = get_seed_gain(blk_id,
+                                            atom_netlist,
+                                            prepacker,
+                                            models,
+                                            seed_type,
+                                            max_molecule_stats,
+                                            atom_criticality);
+            mol_gain = std::max(mol_gain, atom_gain);
+        }
+        molecule_gains[mol_id] = mol_gain;
     }
 
     // Sort seeds in descending order of seed gain (i.e. highest seed gain first)
@@ -207,47 +223,42 @@ GreedySeedSelector::GreedySeedSelector(const AtomNetlist& atom_netlist,
     // std::sort which does not specify how equal values are handled). Using a stable
     // sort ensures that regardless of the underlying sorting algorithm the same seed
     // order is produced regardless of compiler.
-    auto by_descending_gain = [&](const AtomBlockId lhs, const AtomBlockId rhs) {
-        return atom_gains[lhs] > atom_gains[rhs];
+    auto by_descending_gain = [&](const PackMoleculeId lhs, const PackMoleculeId rhs) {
+        return molecule_gains[lhs] > molecule_gains[rhs];
     };
-    std::stable_sort(seed_atoms_.begin(), seed_atoms_.end(), by_descending_gain);
+    std::stable_sort(seed_mols_.begin(), seed_mols_.end(), by_descending_gain);
 
     // Print the seed gains if requested.
     if (getEchoEnabled() && isEchoFileEnabled(E_ECHO_CLUSTERING_BLOCK_CRITICALITIES)) {
         print_seed_gains(getEchoFileName(E_ECHO_CLUSTERING_BLOCK_CRITICALITIES),
-                         seed_atoms_, atom_gains, atom_criticality, atom_netlist, models);
+                         seed_mols_, molecule_gains, atom_criticality, atom_netlist, models, prepacker);
     }
 
     // Set the starting seed index (the index of the first molecule to propose).
     // The index of the first seed to propose is the first molecule in the
-    // seed atoms vector (i.e. the one with the highest seed gain).
+    // seed molecules vector (i.e. the one with the highest seed gain).
     seed_index_ = 0;
 }
 
-PackMoleculeId GreedySeedSelector::get_next_seed(const Prepacker& prepacker,
-                                                 const ClusterLegalizer& cluster_legalizer) {
-    while (seed_index_ < seed_atoms_.size()) {
-        // Get the current seed atom at the seed index and increment the
+PackMoleculeId GreedySeedSelector::get_next_seed(const ClusterLegalizer& cluster_legalizer) {
+    while (seed_index_ < seed_mols_.size()) {
+        // Get the current seed molecule at the seed index and increment the
         // seed index.
         // All previous seed indices have been either proposed already or
         // are already clustered. This process assumes that once an atom
         // is clustered it will never become unclustered.
-        AtomBlockId seed_blk_id = seed_atoms_[seed_index_++];
+        PackMoleculeId seed_molecule_id = seed_mols_[seed_index_++];
 
-        // If this atom has been clustered, it cannot be proposed as a seed.
+        // If this molecule has been clustered, it cannot be proposed as a seed.
         // Skip to the next seed.
-        if (cluster_legalizer.is_atom_clustered(seed_blk_id))
+        if (cluster_legalizer.is_mol_clustered(seed_molecule_id)) {
             continue;
-
-        // Get the molecule that contains this atom and return it as the
-        // next seed.
-        PackMoleculeId seed_molecule_id = prepacker.get_atom_molecule(seed_blk_id);
-        VTR_ASSERT(!cluster_legalizer.is_mol_clustered(seed_molecule_id));
+        }
         return seed_molecule_id;
     }
 
     // If the previous loop does not return a molecule, it implies that all
-    // atoms have been clustered or have already been proposed as a seed.
+    // molecule have been clustered or have already been proposed as a seed.
     // Return nullptr to signify that there are no further seeds.
     return PackMoleculeId::INVALID();
 }

vpr/src/pack/greedy_seed_selector.h

@@ -68,32 +68,24 @@ class GreedySeedSelector {
      * This method assumes that once a molecule is clustered, it will never be
      * unclustered.
      *
-     *  @param prepacker
-     *              The prepacker object that stores the molecules.
      *  @param cluster_legalizer
      *              The cluster legalizer object that is used to create the
      *              clusters. This is used to check if a molecule has already
      *              been clustered or not.
      */
-    PackMoleculeId get_next_seed(const Prepacker& prepacker,
-                                 const ClusterLegalizer& cluster_legalizer);
+    PackMoleculeId get_next_seed(const ClusterLegalizer& cluster_legalizer);
 
-    // TODO: Maybe create an update_seed_gains method to update the seed atoms
+    // TODO: Maybe create an update_seed_gains method to update the seed molecules
     //       list using current clustering information.
 
   private:
-    /// @brief The index of the next seed to propose in the seed_atoms vector.
+    /// @brief The index of the next seed to propose in the seed_mols_ vector.
     ///        This is set to 0 in the constructor and incremented as more seeds
     ///        are proposed.
     size_t seed_index_;
 
-    /// @brief A list of seed atoms, sorted in decreasing order of gain. This
+    /// @brief A list of seed molecules, sorted in decreasing order of gain. This
     ///        is computed in the constructor and is traversed when a new seed
     ///        is being proposed.
-    // FIXME: This should really be seed molecules. It looks like the only
-    //        reason it isn't is because of the atom criticality. May want to
-    //        create the concept of molecule criticality. Currently, the max
-    //        criticality of any block in the molecule is technically being
-    //        used.
-    std::vector<AtomBlockId> seed_atoms_;
+    std::vector<PackMoleculeId> seed_mols_;
 };

vtr_flow/tasks/regression_tests/vtr_reg_strong_odin/strong_global_routing/config/golden_results.txt

@@ -1,4 +1,4 @@
  arch	  circuit	  script_params	  vtr_flow_elapsed_time	  vtr_max_mem_stage	  vtr_max_mem	  error	  odin_synth_time	  max_odin_mem	  parmys_synth_time	  max_parmys_mem	  abc_depth	  abc_synth_time	  abc_cec_time	  abc_sec_time	  max_abc_mem	  ace_time	  max_ace_mem	  num_clb	  num_io	  num_memories	  num_mult	  vpr_status	  vpr_revision	  vpr_build_info	  vpr_compiler	  vpr_compiled	  hostname	  rundir	  max_vpr_mem	  num_primary_inputs	  num_primary_outputs	  num_pre_packed_nets	  num_pre_packed_blocks	  num_netlist_clocks	  num_post_packed_nets	  num_post_packed_blocks	  device_width	  device_height	  device_grid_tiles	  device_limiting_resources	  device_name	  pack_mem	  pack_time	  placed_wirelength_est	  total_swap	  accepted_swap	  rejected_swap	  aborted_swap	  place_mem	  place_time	  place_quench_time	  placed_CPD_est	  placed_setup_TNS_est	  placed_setup_WNS_est	  placed_geomean_nonvirtual_intradomain_critical_path_delay_est	  place_delay_matrix_lookup_time	  place_quench_timing_analysis_time	  place_quench_sta_time	  place_total_timing_analysis_time	  place_total_sta_time	  ap_mem	  ap_time	  ap_full_legalizer_mem	  ap_full_legalizer_time	  min_chan_width	  routed_wirelength	  min_chan_width_route_success_iteration	  logic_block_area_total	  logic_block_area_used	  min_chan_width_routing_area_total	  min_chan_width_routing_area_per_tile	  min_chan_width_route_time	  min_chan_width_total_timing_analysis_time	  min_chan_width_total_sta_time	  crit_path_num_rr_graph_nodes	  crit_path_num_rr_graph_edges	  crit_path_collapsed_nodes	  crit_path_routed_wirelength	  crit_path_route_success_iteration	  crit_path_total_nets_routed	  crit_path_total_connections_routed	  crit_path_total_heap_pushes	  crit_path_total_heap_pops	  critical_path_delay	  geomean_nonvirtual_intradomain_critical_path_delay	  setup_TNS	  setup_WNS	  hold_TNS	  hold_WNS	  crit_path_routing_area_total	  crit_path_routing_area_per_tile	  router_lookahead_computation_time	  crit_path_route_time	  crit_path_create_rr_graph_time	  crit_path_create_intra_cluster_rr_graph_time	  crit_path_tile_lookahead_computation_time	  crit_path_router_lookahead_computation_time	  crit_path_total_timing_analysis_time	  crit_path_total_sta_time	 
  timing/k6_N10_mem32K_40nm.xml	  stereovision3.v	  common	  2.02	  vpr	  67.00 MiB	  	  0.08	  10496	  -1	  -1	  4	  0.20	  -1	  -1	  36452	  -1	  -1	  19	  11	  0	  0	  success	  v8.0.0-12163-g0dba7016b-dirty	  Release VTR_ASSERT_LEVEL=2	  GNU 11.4.0 on Linux-6.8.0-51-generic x86_64	  2025-02-19T17:54:19	  haydar-Precision-5820-Tower	  /home/haydar/vtr-verilog-to-routing	  68608	  11	  30	  262	  292	  2	  99	  60	  7	  7	  49	  clb	  auto	  27.4 MiB	  0.11	  419	  1815	  318	  1436	  61	  67.0 MiB	  0.04	  0.00	  1.93141	  -140.772	  -1.93141	  1.88461	  0.01	  0.000706488	  0.000606458	  0.0228093	  0.0209659	  -1	  -1	  -1	  -1	  8	  283	  18	  1.07788e+06	  1.02399e+06	  -1	  -1	  0.38	  0.186498	  0.171178	  2100	  3116	  -1	  280	  18	  572	  1139	  59841	  29637	  1.93141	  1.88461	  -140.772	  -1.93141	  0	  0	  -1	  -1	  0.00	  0.07	  0.00	  -1	  -1	  0.00	  0.0346595	  0.0310733	 
- nonuniform_chan_width/k6_N10_mem32K_40nm_nonuniform.xml	  stereovision3.v	  common	  2.04	  vpr	  66.32 MiB	  	  0.07	  10496	  -1	  -1	  4	  0.26	  -1	  -1	  36580	  -1	  -1	  19	  11	  0	  0	  success	  v8.0.0-12163-g0dba7016b-dirty	  Release VTR_ASSERT_LEVEL=2	  GNU 11.4.0 on Linux-6.8.0-51-generic x86_64	  2025-02-19T17:54:19	  haydar-Precision-5820-Tower	  /home/haydar/vtr-verilog-to-routing	  67916	  11	  30	  262	  292	  2	  99	  60	  7	  7	  49	  clb	  auto	  27.3 MiB	  0.08	  428	  1698	  248	  1401	  49	  66.3 MiB	  0.04	  0.00	  1.93141	  -140.772	  -1.93141	  1.88461	  0.01	  0.000764131	  0.000648712	  0.0183265	  0.0161998	  -1	  -1	  -1	  -1	  10	  297	  21	  1.07788e+06	  1.02399e+06	  -1	  -1	  0.45	  0.176554	  0.15406	  2100	  3116	  -1	  286	  18	  539	  1058	  53794	  27022	  1.93141	  1.88461	  -140.772	  -1.93141	  0	  0	  -1	  -1	  0.00	  0.08	  0.00	  -1	  -1	  0.00	  0.0402826	  0.0356284	 
+ nonuniform_chan_width/k6_N10_mem32K_40nm_nonuniform.xml	  stereovision3.v	  common	  2.04	  vpr	  66.32 MiB	  	  0.07	  10496	  -1	  -1	  4	  0.26	  -1	  -1	  36580	  -1	  -1	  19	  11	  0	  0	  success	  v8.0.0-12163-g0dba7016b-dirty	  Release VTR_ASSERT_LEVEL=2	  GNU 11.4.0 on Linux-6.8.0-51-generic x86_64	  2025-02-19T17:54:19	  haydar-Precision-5820-Tower	  /home/haydar/vtr-verilog-to-routing	  67916	  11	  30	  262	  292	  2	  99	  60	  7	  7	  49	  clb	  auto	  27.3 MiB	  0.08	  428	  1698	  248	  1401	  49	  66.3 MiB	  0.04	  0.00	  1.93141	  -140.772	  -1.93141	  1.88461	  0.01	  0.000764131	  0.000648712	  0.0183265	  0.0161998	  -1	  -1	  -1	  -1	  14	  297	  21	  1.07788e+06	  1.02399e+06	  -1	  -1	  0.45	  0.176554	  0.15406	  2100	  3116	  -1	  286	  18	  539	  1058	  53794	  27022	  1.93141	  1.88461	  -140.772	  -1.93141	  0	  0	  -1	  -1	  0.00	  0.08	  0.00	  -1	  -1	  0.00	  0.0402826	  0.0356284	 
  nonuniform_chan_width/k6_N10_mem32K_40nm_pulse.xml	  stereovision3.v	  common	  1.99	  vpr	  66.21 MiB	  	  0.09	  10368	  -1	  -1	  4	  0.24	  -1	  -1	  36668	  -1	  -1	  19	  11	  0	  0	  success	  v8.0.0-12163-g0dba7016b-dirty	  Release VTR_ASSERT_LEVEL=2	  GNU 11.4.0 on Linux-6.8.0-51-generic x86_64	  2025-02-19T17:54:19	  haydar-Precision-5820-Tower	  /home/haydar/vtr-verilog-to-routing	  67800	  11	  30	  262	  292	  2	  99	  60	  7	  7	  49	  clb	  auto	  27.1 MiB	  0.09	  447	  1815	  292	  1481	  42	  66.2 MiB	  0.07	  0.00	  1.93141	  -140.772	  -1.93141	  1.88461	  0.01	  0.00090328	  0.000782565	  0.0181809	  0.0161296	  -1	  -1	  -1	  -1	  16	  296	  17	  1.07788e+06	  1.02399e+06	  -1	  -1	  0.25	  0.115099	  0.100427	  2100	  3116	  -1	  300	  17	  545	  1102	  57605	  27890	  1.93141	  1.88461	  -140.772	  -1.93141	  0	  0	  -1	  -1	  0.00	  0.12	  0.01	  -1	  -1	  0.00	  0.0397962	  0.0356125