[packer] Changing List of Feasible Candidates to Priority Queue

vpr/src/pack/greedy_candidate_selector.h

@@ -23,6 +23,7 @@
 #include "vtr_vector.h"
 #include "vtr_random.h"
 #include "vtr_vector_map.h"
+#include "lazy_pop_unique_priority_queue.h"
 
 // Forward declarations
 class AtomNetlist;
@@ -97,13 +98,6 @@ struct ClusterGainStats {
     ///        with the cluster.
     AttractGroupId attraction_grp_id;
 
-    /// @brief Array of feasible blocks to select from [0..max_array_size-1]
-    ///
-    /// Sorted in ascending gain order so that the last cluster_ctx.blocks is
-    /// the most desirable (this makes it easy to pop blocks off the list.
-    std::vector<PackMoleculeId> feasible_blocks;
-    int num_feasible_blocks;
-
     /// @brief The flat placement location of this cluster.
     ///
     /// This is some function of the positions of the molecules which have been
@@ -126,6 +120,25 @@ struct ClusterGainStats {
     ///        set when the stats are created based on the primitive pb type
     ///        of the seed.
     bool is_memory = false;
+
+    /// @brief List of feasible block and its gain pairs.
+    ///        The list is maintained in heap structure with the highest gain block
+    ///        at the front.
+    LazyPopUniquePriorityQueue<PackMoleculeId, float> feasible_blocks;
+
+    /// @brief Indicator for the initial search for feasible blocks.
+    bool initial_search_for_feasible_blocks;
+
+    /// @brief Limit for the number of candiate proposed at each stage.
+    unsigned candidates_propose_limit;
+
+    /// @brief Counter for the number of candiate proposed at each stage.
+    unsigned num_candidates_proposed;
+
+    /// @brief Check if the current stage candidates proposed limit is reached.
+    bool current_stage_candidates_proposed_limit_reached() {
+        return num_candidates_proposed >= candidates_propose_limit;
+    }
 };
 
 /**
@@ -444,7 +457,7 @@ class GreedyCandidateSelector {
     //                      Cluster Candidate Selection
     // ===================================================================== //
 
-    /*
+    /**
      * @brief Add molecules with strong connectedness to the current cluster to
      *        the list of feasible blocks.
      */
@@ -471,7 +484,7 @@ class GreedyCandidateSelector {
         LegalizationClusterId legalization_cluster_id,
         const ClusterLegalizer& cluster_legalizer);
 
-    /*
+    /**
      * @brief Add molecules based on transitive connections (eg. 2 hops away)
      *        with current cluster.
      */
@@ -481,7 +494,7 @@ class GreedyCandidateSelector {
         const ClusterLegalizer& cluster_legalizer,
         AttractionInfo& attraction_groups);
 
-    /*
+    /**
      * @brief Add molecules based on weak connectedness (connected by high
      *        fanout nets) with current cluster.
      */
@@ -491,7 +504,7 @@ class GreedyCandidateSelector {
         const ClusterLegalizer& cluster_legalizer,
         AttractionInfo& attraction_groups);
 
-    /*
+    /**
      * @brief If the current cluster being packed has an attraction group
      *        associated with it (i.e. there are atoms in it that belong to an
      *        attraction group), this routine adds molecules from the associated

vpr/src/util/lazy_pop_unique_priority_queue.h

@@ -0,0 +1,216 @@
+/**
+ * @file
+ * @author  Rongbo Zhang
+ * @date    2025-04-23
+ * @brief   This file contains the definition of the LazyPopUniquePriorityQueue class.
+ *
+ *  The class LazyPopUniquePriorityQueue is a priority queue that allows for lazy deletion of elements.
+ *  The elements are pair of key and sort-value. The key is a unique value to identify the item, and the sort-value is used to sort the item.
+ *  It is implemented using a vector and 2 sets, one set keeps track of the elements in the queue, and the other set keeps track of the elements that are pending deletion, 
+ *  so that they can be removed from the queue when they are popped.
+ * 
+ *  Currently, the class supports the following functions:
+ *      LazyPopUniquePriorityQueue::push(): Pushes a key-sort-value (K-SV) pair into the priority queue and adds the key to the tracking set.
+ *      LazyPopUniquePriorityQueue::pop(): Returns the K-SV pair with the highest SV whose key is not pending deletion.
+ *      LazyPopUniquePriorityQueue::remove(): Removes an element from the priority queue immediately.
+ *      LazyPopUniquePriorityQueue::remove_at_pop_time(): Removes an element from the priority queue when it is popped.
+ *      LazyPopUniquePriorityQueue::empty(): Returns whether the queue is empty.
+ *      LazyPopUniquePriorityQueue::clear(): Clears the priority queue vector and the tracking sets.
+ *      LazyPopUniquePriorityQueue::size(): Returns the number of elements in the queue.
+ *      LazyPopUniquePriorityQueue::contains(): Returns true if the key is in the queue, false otherwise.
+ */
+
+#pragma once
+
+#include <unordered_set>
+#include <vector>
+#include <algorithm>
+
+/**
+ * @brief Lazy Pop Unique Priority Queue
+ *
+ * This is a priority queue that is used to sort items which are identified by the key  
+ * and sorted by the sort value.
+ * 
+ * It uses a vector to store the key and sort value pair. 
+ * It uses a set to store the keys that are in the vector for uniqueness checking
+ * and a set to store the delete pending keys which will be removed at pop time.
+ */
+
+template<typename T_key, typename T_sort>
+class LazyPopUniquePriorityQueue {
+  public:
+    /**  @brief The custom comparsion struct for sorting the items in the priority queue.
+     *          A less than comparison will put the item with the highest sort value to the front of the queue.
+     *          A greater than comparison will put the item with the lowest sort value to the front of the queue.
+     */
+    struct LazyPopUniquePriorityQueueCompare {
+        bool operator()(const std::pair<T_key, T_sort>& a,
+                        const std::pair<T_key, T_sort>& b) const {
+            return a.second < b.second;
+        }
+    };
+
+    /// @brief The vector maintained as heap to store the key and sort value pair.
+    std::vector<std::pair<T_key, T_sort>> heap;
+
+    /// @brief The set to store the keys that are in the queue. This is used to ensure uniqueness
+    std::unordered_set<T_key> content_set;
+
+    /// @brief The set to store the delete pending item from the queue refered by the key.
+    std::unordered_set<T_key> delete_pending_set;
+
+    /**
+     * @brief Push the key and the sort value as a pair into the priority queue.
+     *
+     *  @param key
+     *              The unique key for the item that will be pushed onto the queue.
+     *  @param value
+     *              The sort value used for sorting the item.
+     */
+    void push(T_key key, T_sort value) {
+        // Insert the key and sort value pair into the queue if it is not already present
+        if (content_set.find(key) != content_set.end()) {
+            // If the key is already in the queue, do nothing
+            return;
+        }
+        // Insert the key and sort value pair into the heap and track the key
+        // The new item is added to the end of the vector and then the push_heap function is call
+        // to push the item to the correct position in the heap structure.
+        heap.emplace_back(key, value);
+        std::push_heap(heap.begin(), heap.end(), LazyPopUniquePriorityQueueCompare());
+        content_set.insert(key);
+    }
+
+    /**
+     * @brief Pop the top item from the priority queue.
+     *
+     *  @return The key and sort value pair.
+     */
+    std::pair<T_key, T_sort> pop() {
+        std::pair<T_key, T_sort> top_pair;
+        while (heap.size() > 0) {
+            top_pair = heap.front();
+            // Remove the key from the heap and the tracking set.
+            // The pop_heap function will move the top item in the heap structure to the end of the vector container.
+            // Then the pop_back function will remove the last item.
+            std::pop_heap(heap.begin(), heap.end(), LazyPopUniquePriorityQueueCompare());
+            heap.pop_back();
+            content_set.erase(top_pair.first);
+
+            // Checking if the key with the highest sort value is in the delete pending set.
+            // If it is, ignore the current top item and remove the key from the delete pending set. Then get the next top item.
+            // Otherwise, the top item found, break the loop.
+            if (delete_pending_set.find(top_pair.first) != delete_pending_set.end()) {
+                delete_pending_set.erase(top_pair.first);
+                top_pair = std::pair<T_key, T_sort>();
+            } else {
+                break;
+            }
+        }
+
+        // If there is zero non-pending-delete item, clear the queue.
+        if (empty()) {
+            clear();
+        }
+
+        return top_pair;
+    }
+
+    /**
+     * @brief Remove the item with matching key value from the priority queue
+     *        This will immediately remove the item and re-heapify the queue.
+     *        
+     *        This function is expensive, as it requires a full re-heapify of the queue.
+     *        The time complexity is O(n log n) for the re-heapify, where n is the size of the queue.
+     *        It is recommended to use remove_at_pop_time() instead.
+     *  @param key
+     *             The key of the item to be delected from the queue.
+     */
+    void remove(T_key key) {
+        // If the key is in the priority queue, remove it from the heap and reheapify.
+        // Otherwise, do nothing.
+        if (content_set.find(key) != content_set.end()) {
+            content_set.erase(key);
+            delete_pending_set.erase(key);
+            for (int i = 0; i < heap.size(); i++) {
+                if (heap[i].first == key) {
+                    heap.erase(heap.begin() + i);
+                    break;
+                }
+            }
+
+            // If this delete caused the queue to have zero non-pending-delete item, clear the queue.
+            if (empty()) {
+                clear();
+                // Otherwise re-heapify the queue
+            } else {
+                std::make_heap(heap.begin(), heap.end(), LazyPopUniquePriorityQueueCompare());
+            }
+        }
+    }
+
+    /**
+     * @brief Remove the item with matching key value from the priority queue at pop time.
+     *        Add the key to the delete pending set for tracking, 
+     *        and it will be deleted when it is popped.
+     *      
+     *        This function will not immediately delete the key from the
+     *        priority queue. It will be deleted when it is popped. Thus do not
+     *        expect a size reduction in the priority queue immediately.
+     *  @param key
+     *             The key of the item to be delected from the queue at pop time.
+     */
+    void remove_at_pop_time(T_key key) {
+        // If the key is in the list, start tracking it in the delete pending list.
+        // Otherwise, do nothing.
+        if (content_set.find(key) != content_set.end()) {
+            delete_pending_set.insert(key);
+
+            // If this marks the last non-pending-delete item as to-be-deleted, clear the queue
+            if (empty()) {
+                clear();
+            }
+        }
+    }
+
+    /**
+     * @brief Check if the priority queue is empty, i.e. there is zero non-pending-delete item.
+     *
+     *  @return True if the priority queue is empty, false otherwise.
+     */
+    bool empty() {
+        return size() == 0;
+    }
+
+    /**
+     * @brief Clears the priority queue and the tracking sets.
+     *
+     *  @return None
+     */
+    void clear() {
+        heap.clear();
+        content_set.clear();
+        delete_pending_set.clear();
+    }
+
+    /**
+     * @brief Get the number of non-pending-delete items in the priority queue.
+     *
+     *  @return The number of non-pending-delete items in the priority queue.
+     */
+    size_t size() {
+        return heap.size() - delete_pending_set.size();
+    }
+
+    /**
+     * @brief Check if the item referred to the key is in the priority queue.
+     *
+     *  @param key
+     *              The key of the item.
+     *  @return True if the key is in the priority queue, false otherwise.
+     */
+    bool contains(T_key key) {
+        return content_set.find(key) != content_set.end();
+    }
+};

vtr_flow/tasks/regression_tests/vtr_reg_basic/basic_no_timing/config/golden_results.txt

@@ -1,5 +1,5 @@
- 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	  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	 
- k4_N10_memSize16384_memData64.xml	  ch_intrinsics.v	  common	  1.71	  vpr	  62.29 MiB	  	  -1	  -1	  0.45	  18372	  3	  0.09	  -1	  -1	  33140	  -1	  -1	  71	  99	  1	  0	  success	  v8.0.0-11920-g63becbef4-dirty	  release IPO VTR_ASSERT_LEVEL=2	  GNU 9.4.0 on Linux-4.15.0-213-generic x86_64	  2024-12-04T15:29:41	  betzgrp-wintermute.eecg.utoronto.ca	  /home/elgamma8/research/release/vtr-verilog-to-routing	  63780	  99	  130	  353	  483	  1	  222	  301	  13	  13	  169	  clb	  auto	  22.7 MiB	  0.06	  730	  30541	  5185	  13290	  12066	  62.3 MiB	  0.05	  0.00	  28	  1583	  11	  3.33e+06	  2.25e+06	  384474.	  2275.00	  0.18	 
- k4_N10_memSize16384_memData64.xml	  diffeq1.v	  common	  3.90	  vpr	  66.30 MiB	  	  -1	  -1	  0.72	  23492	  23	  0.30	  -1	  -1	  34028	  -1	  -1	  77	  162	  0	  5	  success	  v8.0.0-11920-g63becbef4-dirty	  release IPO VTR_ASSERT_LEVEL=2	  GNU 9.4.0 on Linux-4.15.0-213-generic x86_64	  2024-12-04T15:29:41	  betzgrp-wintermute.eecg.utoronto.ca	  /home/elgamma8/research/release/vtr-verilog-to-routing	  67888	  162	  96	  1200	  1141	  1	  675	  340	  13	  13	  169	  clb	  auto	  25.9 MiB	  0.18	  5120	  92848	  24971	  61178	  6699	  66.3 MiB	  0.19	  0.00	  52	  9637	  13	  3.33e+06	  2.76e+06	  671819.	  3975.26	  1.14	 
- k4_N10_memSize16384_memData64.xml	  single_wire.v	  common	  2.10	  vpr	  59.81 MiB	  	  -1	  -1	  0.16	  16372	  1	  0.17	  -1	  -1	  29680	  -1	  -1	  0	  1	  0	  0	  success	  v8.0.0-11920-g63becbef4-dirty	  release IPO VTR_ASSERT_LEVEL=2	  GNU 9.4.0 on Linux-4.15.0-213-generic x86_64	  2024-12-04T15:29:41	  betzgrp-wintermute.eecg.utoronto.ca	  /home/elgamma8/research/release/vtr-verilog-to-routing	  61244	  1	  1	  1	  2	  0	  1	  2	  3	  3	  9	  -1	  auto	  21.3 MiB	  0.00	  2	  3	  0	  3	  0	  59.8 MiB	  0.01	  0.00	  2	  1	  1	  30000	  0	  1489.46	  165.495	  0.01	 
- k4_N10_memSize16384_memData64.xml	  single_ff.v	  common	  2.13	  vpr	  59.62 MiB	  	  -1	  -1	  0.15	  16244	  1	  0.17	  -1	  -1	  29552	  -1	  -1	  1	  2	  0	  0	  success	  v8.0.0-11920-g63becbef4-dirty	  release IPO VTR_ASSERT_LEVEL=2	  GNU 9.4.0 on Linux-4.15.0-213-generic x86_64	  2024-12-04T15:29:41	  betzgrp-wintermute.eecg.utoronto.ca	  /home/elgamma8/research/release/vtr-verilog-to-routing	  61048	  2	  1	  3	  4	  1	  3	  4	  3	  3	  9	  -1	  auto	  21.2 MiB	  0.00	  6	  9	  6	  0	  3	  59.6 MiB	  0.01	  0.00	  16	  5	  1	  30000	  30000	  2550.78	  283.420	  0.01	 
+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	initial_placed_wirelength_est	placed_wirelength_est	total_swap	accepted_swap	rejected_swap	aborted_swap	place_mem	place_time	place_quench_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	
+k4_N10_memSize16384_memData64.xml	ch_intrinsics.v	common	1.17	vpr	63.18 MiB		-1	-1	0.21	18728	3	0.06	-1	-1	32704	-1	-1	72	99	1	0	success	v8.0.0-12603-g716d96fe0-dirty	release IPO VTR_ASSERT_LEVEL=2	Clang 18.1.3 on Linux-6.8.0-58-generic x86_64	2025-05-01T22:32:41	betzgrp-wintermute	/home/zhan6738/VTR/vtr-verilog-to-routing/vtr_flow/tasks	64692	99	130	353	483	1	220	302	13	13	169	clb	auto	23.4 MiB	0.03	1748	641	31674	5814	13912	11948	63.2 MiB	0.03	0.00	36	1209	9	3.33e+06	2.28e+06	481319.	2848.04	0.18	
+k4_N10_memSize16384_memData64.xml	diffeq1.v	common	2.84	vpr	66.43 MiB		-1	-1	0.32	23332	23	0.28	-1	-1	33440	-1	-1	78	162	0	5	success	v8.0.0-12603-g716d96fe0-dirty	release IPO VTR_ASSERT_LEVEL=2	Clang 18.1.3 on Linux-6.8.0-58-generic x86_64	2025-05-01T22:32:41	betzgrp-wintermute	/home/zhan6738/VTR/vtr-verilog-to-routing/vtr_flow/tasks	68020	162	96	1200	1141	1	690	341	14	14	196	clb	auto	26.8 MiB	0.11	8696	5304	81261	22686	53433	5142	66.4 MiB	0.09	0.00	46	10726	18	4.32e+06	2.79e+06	735717.	3753.66	1.10	
+k4_N10_memSize16384_memData64.xml	single_wire.v	common	0.50	vpr	61.17 MiB		-1	-1	0.06	17192	1	0.02	-1	-1	29568	-1	-1	0	1	0	0	success	v8.0.0-12603-g716d96fe0-dirty	release IPO VTR_ASSERT_LEVEL=2	Clang 18.1.3 on Linux-6.8.0-58-generic x86_64	2025-05-01T22:32:41	betzgrp-wintermute	/home/zhan6738/VTR/vtr-verilog-to-routing/vtr_flow/tasks	62636	1	1	1	2	0	1	2	3	3	9	-1	auto	22.6 MiB	0.00	2	2	3	0	3	0	61.2 MiB	0.00	0.00	2	1	1	30000	0	1489.46	165.495	0.00	
+k4_N10_memSize16384_memData64.xml	single_ff.v	common	0.51	vpr	61.02 MiB		-1	-1	0.05	17192	1	0.02	-1	-1	29212	-1	-1	1	2	0	0	success	v8.0.0-12603-g716d96fe0-dirty	release IPO VTR_ASSERT_LEVEL=2	Clang 18.1.3 on Linux-6.8.0-58-generic x86_64	2025-05-01T22:32:41	betzgrp-wintermute	/home/zhan6738/VTR/vtr-verilog-to-routing/vtr_flow/tasks	62484	2	1	3	4	1	3	4	3	3	9	-1	auto	22.4 MiB	0.00	6	6	9	6	0	3	61.0 MiB	0.00	0.00	16	5	1	30000	30000	2550.78	283.420	0.00